Yeah, I realize it looks that way, all i can say is, it's not, at least not from me. If it is, you're welcome to alert Scott and have him ban me from posting comments for a bit. I'd tell you more, but I can't even properly paraphrase what's going on in it - it's got the AI dream-like thing where everything makes sense as you skim it, until you stop to think about what you just read. Or i just don't know enough about orbifolds.
Mine was an oblique nudge for you to describe the link contents in some way that goes beyond some form-letter comment that could go anywhere, like "Interesting point! I'll check it out, thanks! www. buy-my-cheap-shoes .com".
If you can't paraphrase, at least try. As you just did.
That said: yeah, it's weird, I don't know what it's trying to do, other than advertise some sort of AI-based tool in some way that still doesn't make me want to use it or even look deeper.
“What astonished him most was that Katusha was not ashamed of her position — not the position of a prisoner (she was ashamed of that), but her position as a prostitute. She seemed satisfied, even proud of it. And, yet, how could it be otherwise? Everybody, in order to be able to act, has to consider his occupation important and good.”
“It is usually imagined that a thief, a murderer, a spy, a prostitute, acknowledging his or her profession as evil, is ashamed of it. But the contrary is true. People whom fate and their sin-mistakes have placed in a certain position, however false that position may be, form a view of life in general which makes their position seem good and admissible. In order to keep up their view of life, these people instinctively keep to the circle of those people who share their views of life and their own place in it. This surprises us, where the persons concerned are thieves, bragging about their dexterity, prostitutes vaunting their depravity, or murderers boasting of their cruelty. This surprises us only because the circle, the atmosphere in which these people live, is limited, and we are outside it. But can we not observe the same phenomenon when the rich boast of their wealth, i.e., robbery; the commanders in the army pride themselves on victories, i.e., murder; and those in high places vaunt their power, i.e., violence? We do not see the perversion in the views of life held by these people, only because the circle formed by them is more extensive, and we ourselves are moving inside of it.”
Funny thing happened to me -- I just got banned from Reddit.
Okay, that's not so funny. The funny thing was the explanation: I was banned because the censorship algorithm detected that I posted some kind of objectionable content in a comment. Helpfully, there was a link to my comment. I clicked the link, and... it displayed the words "[Removed by Reddit]". Thanks, that explains a lot.
Well, at least I see the name of the topic. Yeah, I vaguely remember reading that... but that was 24 hours ago; that's like an entire week in dog years. Apparently I wrote a comment, but I don't remember what. I write many comments online.
Then I read the comment thread again, especially the ones I replied to, and finally I remember. Yeah, okay, maybe I deserved the ban. But it's still funny...
The top-level comment was like: "We should do X" (where I admit that X is an objectionable thing by the sissy Reddit standards). The reply was: "No, we should do Y instead" (where Y is another objectionable thing). And I remember replying: "Hey guys, no need to argue, we could do both X and Y." And... that was the comment that cost me my Reddit account.
The previous two comments are still there, uncensored. Apparently, it was putting those both ideas in one sentence that crossed the threshold. I find it hilarious.
Sadly no, it was just a 3-days ban. But it was liberating for a while; the first day I kept reopening Reddit, but the following two days I simply forgot about it.
I guess I'm not alone in only knowing her name because of NYT XWord if it made to the Onion. I asked ChatGPT and she has 6 appearances in the last year. BTW article says April 30th, not May 2nd (much funnier, unless they backdated that)
In your view, what kind of speech act should I perform if I'm curious about what a particular person would do if the button situation actually happened in the real world?
The answer I am guessing will be most satisfying to you and that I believe I implicitly give in my essay is that you amend the fictional scenario by whatever supposed real life aspect you are interested in.
For example:
Everyone in the world has to take a private vote by pressing a red or blue button. If more than 50% of people press the blue button, everyone survives. If less than 50% of people press the blue button, only people who pressed the red button survive.
_ You know for fact that more than x% of all people are going to press the blue button no matter what. You do not know how much more._
Which button would you press?
This is a much more interesting problem and I am thankful that you made me think of it.
>You know for fact that more than x% of all people are going to press the blue button no matter what. You do not know how much more
Good idea, but I might propose instead:
"There is a counter showing that x% of the entire population has already pressed the blue button. You don't know how many people haven't pressed either button yet, so the final percentage might be anywhere between x% and 100%."
I suggest this because a common thread I've noticed in Newcomb's Paradox debate is that many two-boxers don't accept the premise that the Decider is a perfect or near-perfect predictor of whether you will one-box or two-box. So an accomplished fact seems more likely to be accepted than a prediction even if the prediction is framed as infallible.
(First time writing a post, inspired by Scott's article on Inkhaven. Wasn't planning to write anything till the bit about writing things that will annoy people when I thought - I do want to tell everyone else who is hurt why they are Doing It Wrong)
I you see an em dash in written text it may be generated with AI — or someone is banging their thumb on an iPhone to indicate a pause rather than using a comma. It’s really hard to tell sometimes.
Here on Linux/KDE I can just press AltGr+Minus for an en dash, and Shift+AltGr+Minus for an em dash. No biggie.
You should learn how to type them as well. AI models are constantly being improved, and the use of em dashes will soon be trained out of them. The suspicious lack of sophisticated punctuation will then be considered an indication of AI-generated text.
Goes on forever but never gets to the point. Paragraphs that repeat the same phrase, just worded differently. Always says that it is now going to tell you about A, B or C but puts that off until the very end, and then it's some boilerplate "well there are points in favour of A and points against" conclusion.
Very much so. Lots of boilerplate structure and stock phrases, and lots of keywords associated with the topic, but the actual content is vague and handwavey. It also reminds me of corporate/marketing speak in this respect.
Do you have an example to point to of "the media pretending Trump just imagined this last year?" I agree with others that, while 86 may be slang for killing, I'm much more familiar with it in the context of "get rid of" like "86 that potato salad, it's been sitting out all day."
I haven’t heard anyone else say that but I’m not a regular Fox News viewer. Easy to imagine Greg Gutfield making an unsubstantiated assertion like that.
Everybody who isn't a partisan hack or a complete ignoramus has always known that "86" has a broad range of meanings, of which "kill" is an edge case that applies only in certain contexts. The only rewriting of history *here* is a bunch of partisan hack MAGA Trumpists trying to retcon the phrase into having always and only meant "kill" so they can falsely accuse one of their enemies of threatening to kill Donald Trump.
Democrats, Liberals, and "the Left" have each also attempted to rewrite history in *other* contexts, but here they're the ones playing it straight. Also, those are three different groups and attempting to blur them together is another attempt to rewrite history by partisan hack MAGA Trumpists. Really, you all are making it hard to remember that the Democrats, the Liberals, and the Leftists sometimes do the same thing, which is unfortunate because they do and that's a problem. But it's not a problem we are going to address today, apparently, because we have to deal with this "86" nonsense instead.
It's also worth pointing out that even if someone *had* literally tweeted "Kill Trump", it still wouldn't merit prosecution. The bar for the 1A is really high.
2. Do you believe that the FBI *should* have brought charges against Republicans who made a statement that, according to you, was calling for the murder of President Biden?
3. Is it possible that there is more to deciding whether a statement is proof of intent to kill besides looking up the word in a dictionary and seeing if it could be used that way?
Like, you talk about rewriting history and then completely skip over how that phrase has been used for literally the previous president.
Obviously, I don't believe any of these were caused by someone tweeting "86 46," but if you honestly believe that Comey's tweet incited one of the assassination attempts on Trump, then you should equally believe the same for Biden.
More realistically, neither of these statements had anything to do with the existence of crazy people who want to kill the president, which is why the legal standard for incitement is "imminent lawless action" and not "there is at least one crazy person somewhere in the US who might see this as a call for murder."
(Also, they aren't even charging Comey with incitement, they're charging him with threatening the President. Which is equally ridiculous but in a different way than what you're arguing.)
That’s one of the meanings as given in Cassel’s dictionary of slang.
The Merriam-Webster dictionary defines the term as to "refuse to serve (a customer)", or to "get rid of" or "throw out" someone or something.[7] The Oxford English Dictionary (OED) says it may be used as a noun or verb.[1] As a noun, "In restaurants and bars, an expression indicating that the supply of an item is exhausted, or that a customer is not to be served; also, a customer to be refused service. Also transferred."[1] As a transitive verb derived from the noun, it means "to eject or debar (a person) from premises; to reject or abandon".[1] The OED gives examples of usage from 1933 to 1981;[1] for example, in the 1972 film The Candidate, a media adviser says to Robert Redford's character, "OK, now, for starters, we got to cut your hair and eighty-six the sideburns".[1]
Comey meaning it in Cassel Dictionary of Slang sense is about as realistic as “The Americans” straight arrow, boyscout Sam Beeman suggesting someone whack president Reagan.
We get it, trollers gonna troll. Give it a rest already.
*Stan. (I love that show! Frank Gaad was the FBI director though, not Beeman.)
No, of course he didn't. It's a bullshit charge that probably won't stick.
My point is only that that usage has long existed, despite widespread and influential attempts to pretend otherwise. It reminds me of the time Amy Barrett used the term "sexual preference," and then Merriam-Webster updated the dictionary immediately afterwards to say it was "offensive" so lying press could attack her for it.
FWIW, I looked this up recently because I heard somebody else bring up the Merriam-Webster "sexual preference" dust-up. the Editor of MW explained to Newsweek in the aftermath that they are regularly updating definitions of words but only release these updates in scheduled batches. But if one of those words with updated definitions in the queue is getting increased attention, they will release the update early so that all the people looking it up will have the most up to date entry to read. This makes sense to me and lowered the probability in my eyes that Merriam Webster was putting its thumb on the scales to embarrass Barrett.
...so their defense is they have a schedule for updates but they'll break it when something gets popular to try to control the meaning of trends. That's not a defense, that's what people are accusing them of doing.
The underlying linguistic claim isn’t really controversial among linguists: “sexual preference” had genuinely been falling out of favor in LGBTQ advocacy and style guides before the Barrett hearing.
Some Democratic questioner was a dick in the Barrett’s hearing for her not being ‘read in’ so to speak on the latest connotations of ‘sexual preference’. She was not at all trying to be offensive. To my eye she seems like a model of decency.
Barrett later apologized for causing offense.
“I certainly didn’t mean, and would never mean, to use a term that would cause any offense in the LGBTQ community,” she said. “So if I did, I greatly apologize for that.”
Merriam Webster people noticed a spike in searches for the term during the foofarah so they released the latest sense of the meaning.
I raise my eyebrows at this but I don’t see a sinister cabal of leftist activism in it. I think Fox News presented as such because their business model requires ginning up rage against libtards.
I would encourage you to consider whether you would have found this explanation remotely plausible if you weren't strongly inclined to believe in such organizations being less than blatantly partisan, and were looking for reasons, however tenuous, to avoid the conclusion that they were.
Sorry, I'll be useful and provide more of my thought.
I had recently listened this podcast interview (https://thedispatch.com/podcast/remnant/akvavit-zebroid-and-haole-interview-stefan-fatsis/) of Stefan Fatsis who wrote a recent book after embedding with Merriam-Webster as a lexicographer-in-training. During the interview, Goldberg brought up the "sexual preference" fracas and Fatsis seemed unfamiliar with it but found it completely implausible that MW would have just changed the definition on the fly for partisan reason given that those updates take months and months of work and review before being approved. That was when I went back and looked and found the newsweek article. So now I have an explanation from the MW editor saying that it was an impromptu release of an already reviewed and approved update, and I have a guy who was embedded in MW saying that it sounds implausible for them to have made a partisan change like that so abruptly. I may be somewhat inclined to believe that such organizations are not so blatantly partisan, but I feel like there's some pretty good evidence here.
Does MW publish a comprehensive list of these unscheduled revisions, or do we just have to take their word for it that they're doing so in an unbiased fashion?
If we all know that it's a bullshit charge and that Comey was clearly not using it to mean "kill," then in what sense is it misleading for the media to report that "86" means "get rid of" rather than "kill"? Would it be more informative to anyone if the news articles said "well technically, if Comey was a 1970s mobster he would have been making a death threat, but as he isn't, he was not?"
Additional question if you think it was misleading or leaving out important context - do you believe you are being similarly misleading by only including the "kill" definition in your OP, and not the other definitions or the fact that you were not actually confused by the omission?
The term is slang with plenty of meanings in common usage. If I had instead said that "to get rid of" isn't one of them, that would have been analogous, sure.
I saw an argument a while ago that the Voting Rights Act's requirement to have majority-minority districts (the bit that's just been struck down by the Supreme Court) actually HELPED the Republicans by allowing them to "pack" Democrats into those districts without running afoul of the 14th/15th Amendments.
Prima facie, this struck me as implausible, and that seems be borne out by the reactions to the Louisiana v. Callais. What do you think? Am I missing something?
The gerrymandering cases I've read usually talked about "compactness" as a marker for drawing up fair districts; with a requirement to put minority density above geographic compactness, I think it creates more plausible deniability for gerrymandering. But I've got no statistics on it.
The Christian EA book has arrived in my Kindle, and I will read and review it when I feel less like I've been bashed over the head with a concrete block.
I think I need a tonic or something, this past week I feel rundown. Mr. Brain is not wanting to do anything except sit around and lollygag, so not in the state to do the book justice. Though Mr. Brain retains enough energy to be snarky about the cover, which I dislike. Memo to self: do not judge a book by its cover.
Have you tried one of those Covid home tests? In my first and only experience so far extreme fatigue was the main symptom. Pretty much laid in bed for 2 weeks.
But judging books is what covers are *for*. Nobody would bother with anything more than a plain binding with tite+author unless they thought the fancier cover would secure a more favorable judgement from their target audience, and they get repeat customers only if the cover-based judgements turn out to be mostly correct.
I have good memories of someone buying me a book whose cover was just the title "Y is for Yesterday," and trying to guess what the book would be about. But, probably would have ignored it if I saw it on a shelf.
You can see there's clearly two types of covers: the ones that say "This is a Fantasy Book, you should buy it if you're the kind of person who likes fantasy books", and the ones that say "No no, this is an actual respectable book that you can read on the train without people thinking you're a nerd".
As a kid I remember being interested in reading Terry Pratchett books, but I thought I was too old and sophisticated (being, like, twelve) for the cartoony pictures that I saw on the covers. I wish they'd had a sophisticated-looking edition.
And I still haven't read them, which is a shame because they seem like the kind of thing I'd like.
It just occured to me that when you secularize Catholicism, heaven, purgatory, hell, you get morally good, neutral and bad. When you secularize Calvinism, elect and damned, you get good and bad, and no neutral.
This explains some crazy American ideas that people fret over benefitting from privilege, even when they did nothing bad. Just enjoying whatever advantages you were born with, without personally doing bad things is the perfect example of moral neutrality.
So I think we must rethink the whole concept of virtue signalling. Maybe it is honest. Maybe there are ex-Calvinists really feeling like that if they are not saints, they are evil, because they see nothing in between.
I am very comfortable being morally neutral as an ex-Catholic. My religious ancestors probably thought most people spend some time in purgatory and they themselves will too. They really understood imperfection, and that that is okay.
fwiw, the "puritan -> harvard -> cancel-culture witchhunt" pipeline is a recurring theme of moldbug's. Though he doesn't really discuss the other U.S. calvinists, as much. E.g. here [0], he appeals to evo psych, viz. self-deception.
I think the effect you're seeing is much more rooted in culture than theology. The Progressive Left does have a lot of cultural descent from the New England Puritans, who were theologically Calvinist, but the Social Conservative Right is descended from a different group of Calvinists: Presbyterians from the English/Scottish border country who settled in the western hinterland of the 13 colonies in the 18th century. Similar theology, but very different cultural substrate that took the theology in very different directions.
Now I’m thinking of Faulkner’s Snopes family. I don’t think Faulkner explicitly traces their lineage but they have a stereotypical Appalachian Scott-Irish vibe.
So in trying to find a way to blame a generally unwoke group (protestant christians) for what amount to their political enemies (the secular woke), I think you've probably stretched a little here. So where you present something like:
1. Protestants think that there are "elect" who are good and "Damned" that are bad
2. Because of this they are obsessed with being "perfect"
3. So much so that people who leave the church and become woke secularists and can't get rid of it.
Except that's not at all how the theology works; in Christianity, everyone is bad, equally so, and needs redemption as a result. Doubly so in Calvinism, which you namedrop here, where you don't really have a choice to accept OR reject salvation.
So if you rewrite the entire religion down to it's most fundamental tenets, yeah, you can THEN use your argument to stretch and blame protestants for an almost wholly secular movement from their almost wholly political outgroup opposites. It's still a stretch (you have to pretend the driving forces behind the woke are all ex-protestants who are so indoctrinated they can't get away from their former dogma, which is unclear.)
But if you *can't* or *don't* rewrite the really basic-level theology of the actual group you want to blame the woke on, then it doesn't work so well because you fundamentally had to misrepresent them to even get to the point where you tried your (still substantial) stretch of an argument.
A Catholic can chime in here on their side, but given how little effort went to your protestant representations I'm suspicious you might be misrepresenting how Cath theology works, too.
I think the crux is that you're treating the doctrine as directly equivalent to the behavior. Yes, Christian theology says everyone sins. But the community is not always reducible to explicit theology. E.g. Christmas trees ostensibly have nothing to do with the Levant.
In practice, my impression is that Calvinists kept an eye out for signs of grace (and/or signs of reprobation), which is where the de-facto moral-binary shows up. E.g. the Puritans kept an eye out for "fruits of the spirit". With the Salem Witch Trials probably being the most famous endpoint of this wariness. So yeah, doctrinally, Calvinists don't get a say in whether they're saved or damned. But the vigilance regarding acausal *signals* produced a certain incentive gradient. It's basically a Newcomb's Problem.
Look, I wrote about secularized religion, that is, when religion turns into culture. It is possible for political opponents to share the same culture, like when the Soviets said the problem with French Communists is that they are French first and Communist second, so they shared the culture of their opponents like De Gaulle. And culture is not blame, it is what it is.
And yes, secularized religion "distorts" theology - rather it does not have theology.
And it was none other than the great mythologist Joseph Campbell who proposed how ex-religious cultures work, that ex-jews are messianistic, ex-caths drift towards universalising mysticism, and ex-prots towards individual salvation, be that libertarianism or the personal is the political kind of wokeness.
This is not blame. This is culture. I noticed Calvinist roots earlier in wokeness, for example, that there is no forgiveness, no absolution, because one sin proves you are not one of the elect so you cannot possibly redeem yourself.
The idea that prots have less of a moral gray zone than caths was proposed by Max Weber in Protestant Ethics - that there is no such a thing as a cycle of sin, repentance, atonement, sin again. Weber was explicit that the Protestant lives a more methodical life, where every action could be a sign of damnation or salvation, leading to a more rigid moral view.
Pascal criticized "Jesuit casuistry" creating moral gray zones (he was a Jansenist, which means a Catholic with a Protestant attitude)
Adam B. Cohen found that protestants see sinful thoughts almost as bad as sinful actions, while catholics and jews not.
But why must I even argue this? Isn't this obvious if a movement starts out as Puritan, it is likely to remain morally rigid?
More interesting stuff. I have seen how the very same liberal goals got different justifications in different cultures. Like for example they agree that homophobia is bad, but they disagree why. In the cath type of culture like France, it is because you are judging people for their sex lives, and you should be unjudgemental, and ultimately it means you should not care about morality at all. I think in many places in the US also UK it is the opposite, there is an explicit moral judgement that homophobia is harming people, hence you should stop. These two different approaches I have noticed long ago.
I agree with everything you wrote - it's a fairly shallow and uninformed OP. In Catholic belief, purgatory is a waiting room, come the last judgement you get voted up or you get voted down. The Verdi requiem is not really practical as a funeral mass, but it sums up the Catholic visions of heaven and hell reasonably well.
I think a more useful distinction is that Catholics believe the good works have inherent worth and count in your favour come judgement day. Most branches of Protestantism treat this differently: 'sola fidei' is Luther's credo that only belief and faith (internal states) are necessary for salvation. Your actions or standing in society are irrelevant, though standing with God is sometimes reflected in wealth in this world. My idea of the ultimate irony, which I would one day like to experience, is to see a prosperity gospel preacher give his take on Jesus with the moneylenders.
" In Catholic belief, purgatory is a waiting room, come the last judgement you get voted up or you get voted down."
Ahem. Not just no, but hell no!
That view of Purgatory is the pop culture one, often conflated with Limbo, and I'd like to beat with a big stick every person who promulgated it.
There is no voting up or down. The souls in Purgatory are the blessed dead, who are destined for Heaven, but have first to undergo that period of cleansing where the penance for sin they did not or could not do on Earth is performed.
It's not a second chance, it's not a "get out of Hell free" card, it's not Limbo. In very traditional belief, the *pains* of Purgatory might have been similar to the pains of Hell, but in Purgatory there is hope because you know you are saved. The souls in Purgatory can intercede for us, the damned souls cannot.
This is why old holy cards would often depict the Holy Souls as in flames, or in chains, *but* attended by angels:
I can't get too mad since modern (since Vatican II at least, but often the lay attitudes and ignorance predated same) Catholic teaching of the faith is poor to terrible and people have no idea of what they are supposed to believe, but I tried my hand at explaining Purgatory to Protestants online once and I'm inclined to develop a twitch in my eye when people get it wrong on the Internet 😁
EDIT: I'll spit on me fist and get into good works/works righteousness versus sola fide another day, I also had a go at that when discussing the spiritual and corporal works of mercy for said Protestants, who were very courteous and interested, but that's a lecture for another day.
OK - waiting room is an oversimplification, I agree. My primary school catechism was a while ago. But I am clear on the distinction between purgatory and limbo.
Surely there still has to be a Judgement day and the possibility of going down? I went to Mass for Easter a while back, and they've changed the responses (again) but they/we still believe in the resurrection of the body and the second coming and all of that?
Right, I see what you're getting at. I was confusing your point with the general impression that Purgatory is a way of 'earning' your way out of Hell.
But it's not a waiting room, either. There is no "well, this person is really, really good and so goes straight to Heaven; this person is really, really bad and goes straight to Hell; you're in the middle, not too good or too bad, so we'll stick you in Purgatory and decide later".
At the Second Coming and General Judgement, after the resurrection of the body, our eternal fates will be revealed. Then the separation of the sheep and the goats takes place: the blessed for Heaven, the damned for Hell. But the Holy Souls in Purgatory will go to Heaven, not because they 'worked out' their sins there but because that was always their end. If I can cobble together a metaphor, it's not "is this true gold or fool's gold? let's find out", the souls are gold but being purified of dross as gold is tried by fire.
There is no change after death, no second chance, no "okay I was kinda bad on earth but now I can do better if you let me".
Going up, yes, going down, yes but no "possibility" that is unknown until the Last Judgement; at the Particular Judgement (for each individual person after their death) our fate is known finally and for sure.
From the Catechism:
"III. THE FINAL PURIFICATION, OR PURGATORY
1030 All who die in God's grace and friendship, but still imperfectly purified, are indeed assured of their eternal salvation; but after death they undergo purification, so as to achieve the holiness necessary to enter the joy of heaven.
1031 The Church gives the name Purgatory to this final purification of the elect, which is entirely different from the punishment of the damned. The Church formulated her doctrine of faith on Purgatory especially at the Councils of Florence and Trent. The tradition of the Church, by reference to certain texts of Scripture, speaks of a cleansing fire:
As for certain lesser faults, we must believe that, before the Final Judgment, there is a purifying fire. He who is truth says that whoever utters blasphemy against the Holy Spirit will be pardoned neither in this age nor in the age to come. From this sentence we understand that certain offenses can be forgiven in this age, but certain others in the age to come. (St. Gregory the Great)
1032 This teaching is also based on the practice of prayer for the dead, already mentioned in Sacred Scripture: “Therefore [Judas Maccabeus] made atonement for the dead, that they might be delivered from their sin.” From the beginning the Church has honored the memory of the dead and offered prayers in suffrage for them, above all the Eucharistic sacrifice, so that, thus purified, they may attain the beatific vision of God. The Church also commends almsgiving, indulgences, and works of penance undertaken on behalf of the dead:
Let us help and commemorate them. If Job's sons were purified by their father's sacrifice, why would we doubt that our offerings for the dead bring them some consolation? Let us not hesitate to help those who have died and to offer our prayers for them. (St. John Chrysostom)"
So do we all have our personal judgement day at death? And the purchasing of indulgences was to shorten your time in purgatory and speed your passage into heaven... I guess all of this fits together.
But that's separate from the second coming and the resurrection of the body and life ever after? I've always assumed that the second coming was also Judgement Day.
I reread The Baltimore Catechism of my religious education class as a 7 year old a few days ago.
The grandchildren of Italians and of what were at the time unified-Yugoslavians were released from public school custody for a couple hours a week to go to the ethnically appropriate church for instruction. Who knew where the heathen grandchildren of the Finns went. ;)
Oh, that's going back to the Old Days right there, Tyrone!
Yeah, it was a lot. But the idea was that kids would get solid doctrine into their heads by rote memorisation, and never mind if they understood it completely. In later years, they would remember the teaching and be able to apply it.
Also, in the days of that catechism (which was first published in 1885, Wikipedia tells me) the age for First Communion was *after* Confirmation, which would have been around twelve to fourteen years old (so the kids would have been older). The age wasn't reduced to "age of reason/around seven" until 1910 under Pope Saint Pius X):
I received confirmation from the closest bishop at about 14. I was honestly a bit shocked when I saw him flip away the cigarette he had been smoking on his way out of our priest’s presbytery. Bishops smoke cigarettes??? This was the boyscout phase of my life that, try as i might, somehow have never been able to put fully behind me. Getting there though. ;)
> Except that's not at all how the theology works; in Christianity, everyone is bad, equally so, and needs redemption as a result
In the secular transformation of these ideas everybody is bad (privileged) and are born with original sin ( privilege of skin colour or being cis).
It’s a bit more intersectional of course but you have the original and present day sin of being male, or white if you are not male, or male if you are not white, or cis - then that’s pretty much everybody.
He’s not arguing that these secularists are the same as Christians now, but they derive from the same philosophy.
1. "everybody is bad" in your thing relies on a definition of everybody which means "only a few groups, mainly white men" and once you realize you are using a nonstandard definition is AGAIN not at all like the theological beliefs of christians re: Who Is Bad (again, actually everybody).
2. And then you go, OK, but it applies a little bit to everyone, in different amounts, depending on what you are, where white male is the worst and then there's gradiations all the way down. Which is AGAIN not like either catholic or protestant Christianity but ALSO is not like what OP's post claims (that there's only "perfect" and "damned" with no middle ground or gray area in both protestant christianity and wokeness)
3. And then you say they are derived from the same philosophy, which, re: 1-2, they clearly aren't unless you contort yourself as OP did (fully misrepresenting the views of Christianity so they will coorespond to something else they don't resemble and, broadly, oppose) or what you did (use "everybody" to mean "some" and "lots of different groups with different levels of original sin" to mean the same as "everyone is equally bad and fucked without redemption".
Like, listen, I *sort of* get how not liking Protestants much might motivate someone to, basically, blame them for various problems. That's normal. But I'll be damned if I'm gonna be part of the main US group opposing the woke the most consistently in the most ways and ALSO be told they are my fault, and ALSO for reasons the arguers are so contorted using I can hear their vertibrae scraping together.
" But I'll be damned if I'm gonna be part of the main US group opposing the woke the most consistently in the most ways and ALSO be told they are my fault"
Let's hope not, there's always the chance of salvation for everyone!
I get what you're saying but unhappily, in the same way (American) atheists of a certain stripe go at Christianity like a bull at a gate under the assumption that *every* Christian globally and historically has been *exactly* like 19th to 20th century American Fundamentalists, those who go woke tend to go as they have been culturally conditioned, and that's American Protestantism in the wider culture.
Which did, under the Reformed strain, go very much "elect or reprobate". You are either on the Right Side of History or you are a bad terrible person benefiting from Systemic Racism and the Original Sin of Whiteness. You can't help yourself! We're not blaming you, we're just pointing out that unless you do all this work and every single day work to undo and unlearn Whiteness, well then you are one of the damned.
From the progressive side, it's easy to see the linkage, as they love talking about Christian Nationalism (as though *that* is all of Christianity) and the connections with White Supremacy:
"Hill-Fletcher shows that the Christian habit of seeing themselves as the "chosen ones" has often been translated into racial categories as well. In other words, Christian supremacy has historically lent itself to white supremacy, with disastrous consequences. Hill Fletcher proposes educational strategies to disentangle the two that will help us move forward toward racial healing in America."
It's harder to see from the other side, as the progressive element will vehemently deny any links to religion, but they have been formed by the culture they came out of, and American culture is influenced by Protestant Christianity.
Summary - the Farm Bill is probably going to be voted on in the house within the next few days. If it passes as is, it will nullify all state laws enforcing animal welfare standards on interstate meat and dairy imports. (Eggs are thankfully exempt.) It will also pre-empt future laws along these lines.
If you want to help prevent this, the linked post contains a document detailing how to help.
>In general.—Producers of covered livestock have a Federal right to raise and market their covered livestock in interstate commerce and therefore no State or subdivision thereof may enact or enforce, directly or indirectly, a condition or standard on the production of covered livestock other than for covered livestock physically raised in such State or subdivision.<
So it's not wiping out state-level law, it's preventing state-level law from regulating the states around them. Washington law can no longer prevent Alaska from importing factory-farmed eggs.
>(B) does not include domestic animals raised for the primary purpose of egg production.<
So states can regulate animal raising in their own territory, but not put requirements on imported food? Or can they do that too, and the bill only closes some weird loophole where states had a say about completely unrelated animals?
That's the way I read it; you can prohibit factory farming inside the state but can't stop factory-farmed goods from entering and being sold like any other goods. And they've excluded imported chicken products, presumably because the chicken guys didn't pay them enough.
This was the best Google would do for me, if someone has a better link I'd welcome it.
I find myself getting bored with the chorus in songs. Not so much in pop songs because no one cares about the lyrics but more so in anything narrative driven. They put in so much effort in setting up a story and telling it beautifully just to break the flow for no purpose other than convention. Sometimes it’s just a short refrain but other times it’s lengthy and all I can think of is how much time it’s taking up. Imagine you were watching a tv show and it can’t stopping every ten minutes to replay the opening credits. It would be ridiculous.
Now you’re probably thinking that of course they have to do it. It’s what people expect and it’s necessary to get an audience. The go to counter example would be Pink Floyd but they’re trying to be different. Look instead to Marty Robbins. Very popular country artist and often went without the chorus. Not to be rebellious but because it was unnecessary. You can’t listen to his hit song El Paso and tell me a chorus would have improved it. It would get in the way. He had enough sense to let the story and the guitar carry the song without it.
First song that popped into my head to test what you're saying was "The Gambler" by Kenny Rogers. Would this song be better if the chorus were not repeated or if the lyrics to the chorus changed each time? I think not.
>Imagine you were watching a tv show and it kept stopping every ten minutes to replay the opening credits.
...well, they literally do that every twenty-two minutes. (Forty-three or so for the hour-long shows.) Because otherwise the audience wouldn't remember the show's name.
The best uses of chorus can recontextualize the chorus in each pass. (Is it lame to use the Protomen as an example? Almost certainly. https://www.youtube.com/watch?v=MIZJCiM-IaY)
How on earth is that "the most streamed Japanese song ever"?! It's not even Yoasobi's most popular song!
If I go to Yoasobi's Youtube page and sort by popular, it's only third on the list, and that's just *one band*, not the whole of Japanese music. Come on now.
I'm sure you could find popular English songs that do this too without too much difficulty. But it is evidence against people doing this because it's "necessary for an audience". Looping back to a motif is something that works well for both music and its lyrics, but there's merit to doing things differently than normal. In fact, I've actually had trouble finding songs that just repeat the chorus three times without any modifications, presumably because artists realized that would be boring.
...What I'm trying to say is that you should probably try harder to find music that you like.
I’m not saying that absolutely no songs go against the norm. I’m not an idiot. My point is that outside of handful of exceptions, most songs keep the same formula and I don’t think there’s any commercial reason it has to be like that. I listen to a lot of music in a week, most of it not well known and outside of instrumentals, they still do the same thing.
I love it when a narrative song uses a chorus effectively. In The Mariner's Revenge, the chorus serves as a reminder of the act of revenge he wants to carry out, with the song ending on the chorus and him finally getting revenge.
If you want to read a story just read a story. If you want to listen to a song, listen to a song.
Having just read the lyrics to Marty Robbin's El Paso, I don't think a chorus would improve it; it's a narrative-driven song.
On the other hand, to pick the first example that spring to mind, I definitely don't think that Smells Like Teen Spirit would be improved by adding four more verses and deleting the chorus. The lyrics are there and they create a certain mood but they're not what the song is *for*.
Agree, and in fact some songs repeat the chorus the last time, sometimes more than once, because it's so good it's worth that much runtime of the song. To use the example of a different Nirvana song, "Lithium" does this.
I am a huge fan of Marty Robbins. "El Paso" is one of the earliest memories of my life. I was in the back of an ambulance being taken to a hospital after falling out of a tree at boarding school and the ambulance driver's radio played "El Paso." I think it was the first time I ever heard it .
I became interested in this issue you raised and decided to do a little research. It seems my instinct about work songs and also liturgical recital is a very powerful influence in that kind of song structure. I am attaching a link to a conversation I had with Claude about it.
I am pasting in something here because it doesn't get included in the snapshot of my conversation that Claude shared with me :
>The gandy dancer songs fed directly into the blues — Jimmie Rodgers learned his guitar style and his yodeling from watching the Black track crews — and from the blues into rock and roll. The ancestry is remarkably direct.
IV. The codification — from field to stage
Pre-1840s
Oral tradition only. Work songs, shanties, spirituals, field hollers. No fixed form; the caller improvises; the refrain is memorized, not written.
1840s–1900s
First codification. Verse-chorus form appears in parlor songs and minstrelsy ("Oh! Susanna," 1848). The structure is now written down and published as sheet music.
Early 1900s
Tin Pan Alley supplants it with the 32-bar AABA form — the standard used by Cole Porter, Gershwin, Rodgers and Hart. The chorus becomes the entire song; the verse recedes.
1950s–1960s
Return of the form. Blues and rock and roll revive verse-chorus structure. By the mid-1960s it is dominant in rock, where it has remained ever since.
1960s–present
Elaboration. The prechorus emerges as a transitional section. Production elements — drops, builds, breakdowns — begin to serve the emotional function of the chorus without containing a melodic hook at all.
V. On your thesis
The observation that the chorus exists to permit communal participation — to allow those who cannot carry a verse to still belong to the song — is well supported by the history. The shantyman and the caller were skilled specialists. The rest of the crew were not expected to be. The chorus gave them full membership in the musical event without requiring any of the caller's particular gift.
This is also, it is worth noting, why choruses tend toward simplicity of lyric, repetition, and a narrow melodic range. They are not simplified because composers were lazy. They are simplified because they must be instantly learnable by people who have never heard the song before — or who are too tired, too far away, or too occupied to track a melody carefully.
The chorus is the democratic clause in the contract of performance. The verse is where skill speaks; the chorus is where everyone is permitted to answer.
But hardly any music listening now is communal: it’s mostly to have something in the background while you’re doing something else. That just strengthens my thesis here. They aren’t doing this structure for any purposes it’s just convention.
Last week, I asked for people's rankings of various aerospace firsts. The thing that inspired this was that I've been reading memoirs and biographies about the US Space Program in the 50s and 60s and noticed a few things.
One is that first-hand accounts I've read are pretty unanimous about Charles Lindberg being an awe-inspiring mega-celebrity. I remember a scene early in Apollo 13 where Jim Lovell (played by Tom Hanks) is reacting to the Apollo 11 landing by laughing about how Neil Armstrong has taken his place in the history books alongside Christopher Columbus and Charles Lindberg; I'm pretty sure this is fictional, but it rings true to attitudes I've seen about Lindberg among people of that generation, e.g. Gene Kranz (*) seeming starstruck about Lindberg visiting Mission Control. I wanted to sanity-check my instinct that Lindberg's fame has faded considerably over the intervening several decades since the 60s.
(*) One of the Mercury/Gemini/Apollo flight directors, the one played by Ed Harris in the Apollo 13 movie.
Another is that, in discussing the books with one or two people around my age (mid 40s) but with less intense nerdery directed in that particular direction, I was surprised to learn that Chuck Yeager (first man to break the sound barrier), Alan Shepard (first American in space, and also the first man to play golf on the Moon), and John Glenn (first American in orbit) weren't names they recognize.
The results of this unscientific survey are:
1. There is an almost unanimous sense that Neil Armstrong (first man on the Moon) is the most important of the people on the list, followed by Yuri Gagarin (first man in space and in orbit), with a couple dissenters putting Lindbergh or Yeager on top. Everyone knew who Armstrong was and almost everyone knew of Gagarin. This is more or less what I expected.
2. I am the oddball in terms of remembering Shepard and Glenn. In general, people old enough to remember the tail end of the Apollo program remember them, but people my age bracket or younger only do so if they're deep aerospace nerds who remember everyone on the list. Even among people who do remember them, they're generally ranked low.
3. Lindbergh is a bit better-remembered than I would have guessed, and is usually ranked third or fourth. Older people and people with deeper interests in aerospace history tend to put Lindbergh higher.
4. Yeager is a little more of a deep cut than Lindbergh, but people who remember him usually rank him third or fourth. This is about what I would have expected.
Lindbergh is one of the interesting lacuna of history. Yes, he really was that huge in the 20s and 30s, with stuff like the Lindbergh Baby dominating headlines at the time. I suspect some of this is general fading with time, and some is specifically his association with the Nazis in the 30s forced him out of the public eye in the 40s and beyond. But, yeah, he was still widely beloved in the aviation world among those who'd grown up before his fall.
>The thing that inspired this was that I've been reading memoirs and biographies about the US Space Program in the 50s and 60s and noticed a few things.
I'm going to assume you've already read Carrying the Fire because it's sort of the free space of astronaut autobiographies, but I'd flag Walt Cunningham's All American Boys as the most interesting one aside from that. He's a lot more honest about the problems of the astronaut office than most are, although recent editions include him yelling at NASA post-Columbia, which I skipped as being not interesting. For real fun, read it back-to-back with Slayton's posthumous Deke.
Edit: It is also worth noting that Yeagar got the supersonic job thanks to some slightly bad behavior on the part of Bob Hoover. Who also passed up a test pilot job at Boeing in favor of Tex Johnson. Boeing dodged a bullet there.
I'm actually in the middle of Carrying the Fire now. I started the current run of reading after watching the first few episodes of "For All Mankind" and getting inspired to reread "The Right Stuff" and "Failure is Not an Option". I picked up "Carrying the Fire" next because I remembered reading and liking Collins's "Mission to Mars" as a kid. Thank you for the recommendations; I was planning on reading "Deke" next, but now I might slot "All American Boy" in before it.
One thing I did notice is that there seems to be some difference of opinion over how "fighter jock" the first few rounds of astronauts were in temperament and culture. Wolfe seems to be the outlier so far in emphasizing this, and I'm looking forward to seeing how Cunningham and Slayton characterize things.
I'm also still not sure what to make of "For All Mankind". It definitely has a lot of potential, I've heard good things about where the series is going, I've really liked other stuff Ron Moore has done, and the writers definitely have done their homework. But there are a lot of details that bother me. They're going even further than Wolfe in emphasizing the Fighter Jock angle. They remixed some bits I recognize from the books I'm reading in ways that don't quite make sense. And the pacing so far has been off, with the episodes feeling padded and then spiced up with contrived drama to try to keep them from dragging. I'm probably going to give it another couple episodes before deciding whether or not to give up.
I haven't seen "For All Mankind" and don't plan to because it strikes me as the sort of thing that would drive me absolutely nuts. But I'm also an extreme weirdo.
My other recommendation's would be A Man on the Moon, which is probably the best flight-focused history of Apollo, and the TV adaptation, From the Earth to the Moon, which is really well-done.
But for answering the cultural question, Cunningham is the best book I've read. Re Wolfe specifically, I believe Deke didn't love it, but Schirra was reasonably positive in his book. (Far from the worst astronaut book, but not amazing.)
It's certainly a great book as a piece of literature. I'm not sure it's nearly as great as a piece of history, although my attention focuses a bit later so I can't pick out specific errors offhand.
Yes, almost everyone. No idea why Wolfe latched onto that issue, but the consensus by basically everyone else is that Grissom was telling the truth. Among other things, several later pilots blew the hatch manually, and the recoil of the mechanism was enough to bruise their hand. Grissom had no bruises. I believe the best theory right now is that ESD from the helicopter may have set it off via an antenna in the water, but we'll never really know for sure.
It is also worth noting that Grissom was assigned the first flight of not one but two spacecraft, something that probably would not have happened if he had been viewed as a screwup, and the general consensus is that he probably would have been the first man on the moon had he not died in the fire.
You don't want to live in the world where Boeing won its lock on the jet-airliner industry because their test pilot rolled a 707 *with all the engines shut off*?
This comment made me look up this gentleman's biography and holy cow, what a story. Escaping a Nazi POW camp by stealing an enemy plane and flying it back to Allied territory? Incredible.
Possibly the only fighter pilot ever to pull off even a half-Maverick in real life.
And I've seen his air-show routine in person, where he rolls and loops and otherwise does a full aerobatic display, in a twin-engine light transport aircraft with the engines turned off throughout. Absolutely magnificent.
FWIW, I only recognized John Glenn as "the oldest man in space" for his return trip in 98. I knew he did something in the Apollo days, but I couldn't have told you exactly what.
Shepherd and Glenn's achievements are both of the form "First American to...", not "first person to..."
Similarly, Lindbergh is overly famous because he's an American, and Americans tend to be loud about these sorts of things. "First non-stop solo flight across the Atlantic" is a pretty good record to hold, but how many Americans can name the pilots of the first nonstop non-solo flight across the Atlantic, or the first non-non-stop flight across the Atlantic, or the first non-solo flight across the Pacific, or the first flight across the English channel, or any number of comparable records that aren't held by Americans?
I can't remember the pilot's name, but first non-non-stop across the Atlantic was the USN's NC-4. so that one is held by Americans. As for first across the English Channel, Louis Bleriot, of course.
Lindberg gets credit for being the first person to fly someplace *interesting* across the Atlantic. Has nothing to do with being American, or flying solo. A pair of Frenchmen flying from Paris to New York would probably(*) have been as famous. The guys who crashed into a bog in Ireland after flying over most but not all of the Atlantic ocean, are a footnote even if I do remember their names (Alcock and Brown). They didn't even arrange to have a newsreel team covering the bog.
Same reason Armstrong usually comes in ahead of Gagarin. Going someplace interesting beats going around in circles just to prove than you can. The new medium of transportation isn't worth much if you can't use it to go to interesting places.
* Contingent on their being reasonably charismatic and PR-savvy, as was Lindberg. But I'm positing Frenchmen, so that seems a safe bet.
FWIW I was one of the dissenters that put Lindbergh first, and I'm not American. I also don't remember all those other, less-well-known firsts you proposed, whether or not they are American. The nationality angle was important back when the things happened, but not very important by now. It's all "history of aviation and spaceflight" by now.
The main point is estimating returns to a year of education, using excellent data from Norway. They compare: A) a simple OLS controlling for demographics and family characteristics, B) models with fixed-effects for siblings or twins, and c) Mendelian randomization instruments (MR). The idea of MR is to use the random variation in what genes a person receives from their parents to simulate an RCT. If you get lots of the genes associated with higher education achievement from your parents, that is kinda like being assigned randomly to get more education. They argue the larger estimates with MR are more reliable than the other methods.
I have two problems with the paper.
First, I think this method will pretty much always violate the exclusion restriction, and so will most likely be positively biased. For the instrument to be valid, you need to have genes that affect income _exclusively_ through education. Think about how genes can possibly result in higher education. They may cause better memory, better IQ, better health, time preference, etc. These things cause more education, but every one also independently affects labor outcomes. Even something highly defensible like, say "enjoying a school-like environment" can plausibly have independent effects (you may choose a career in education or academia, you may prefer a regular corporate job instead of startup, etc). Unless you can get genetic memories of one year of education, Bene-Gesserit-style, this issue is pretty fundamental to the strategy.
In fact, their estimate using monozygotic twins is the lowest of the bunch, and this strategy should be adjusting for the biases I mentioned (as well as shared environment). (I do have issues with this result too, though).
Second, the very research program seems misguided to me. We can try different strategies to isolate "the return of education" and discuss relative strengths, but actually each strategy will be leveraging a different source of variation, and thus estimating a different LATE. It is pretty likely that there is a lot of heterogeneity, and using different methods will lead to different results not only because they are correcting different biases, but actually because the target parameters differ.
In fact, we have understood the concept of LATE for 30 years: it is time to abandon the pretense that it makes sense to talk of a single "return to education" as a useful concept. For public policy in particular, the relevant interventions are never "increase the average level of education by 1 year." They are more like "increase the number of scholarships by some amount", "increase funding for specific programs by some amount", "build X new schools", etc. Research on the returns of each particular intervention is much more useful as a guide than some average value for the population (even if we could estimate it well).
You phrase this as the 'return to education' which I would have interpreted in societal terms if I had not read the abstract of the paper, where it is phrased as a study in whether an additional year of education increases earnings.
In public policy, we want to estimate the social implications of increased investment in education. Individual attainment is probably not the best way to do this - population level data would probably be more useful. In this setting you could feasibly do a cost-benefit analysis to see whether additional hours of instruction would lead to sufficient additional GDP in the future to justify the current expenditure. In analysing multiple different options, this could be one of the variables used to estimate the return on investment.
California's wealth tax ballot measure is retroactive: anyone a resident in the state on Jan 1 , 2026 must pay it. What happens if it passes, is upheld as constitutional, and a bunch of billionaires leave the state and refuse to write checks? I don't imagine Trump will give California any aid in enforcement, nor will other red states. The extradition clause requires states to extradite, however, it reads "A Person charged in any State with Treason, Felony, or other Crime, who shall flee from Justice," which you can argue doesn't apply to someone who left California in May 2026 and whose crime is something they did outside California later.
I imagine any one rich enough to feel a substantial hit from such a law has already set up domicile elsewhere to avoid "oh crap, it's Jan 1st and I'm still not a Monégasque citiizen".
I vaguely remember reading some article about this a while back which mentioned that Jeff Bezos, for one, was moving to Florida. So the deep pockets probably already have their lawyers ready for "in fact, at that date, my client was *not* a resident in the state" arguments in court.
See something similar in an Irish context, where attempting to tax rich non-domiciled persons saw a year-on-year decline:
As ever, it's going to be the lawyers making money out of this, with cases for and against the very very rich. Depends how much the governor (be that Gavin or his replacement) feels they can push the big tech companies and tycoons versus the threat of them all moving to Texas or wherever or setting up residence elsewhere, and where their companies are located (I can imagine the lawyers right now rubbing their hands at the vision of billable hours arguing that just because Googmagoo is physically located in Mountainous Vista, it is not headquartered there and thus not liable for the tax, see also my client Mr Beff Jezos whose permanent residence is now in Venice and we don't mean the one in LA):
Gavin fought hard *against* the wealth tax. While CA state government does have its faults, they're firmly on the right side here. The culprit here is the ballot proposition system, and SEIU's weaponization thereof.
States engage in reciprocal enforcement of tax law. Once California has a judgement against you they can use your home state's legal system for enforcement. The constitution's full faith and credit clause means that you can't escape a judgement in state court by moving to another state. They can also just go into your bank account if you use a national bank that has branches in CA.
IANAL, but naive googling suggests that retroactive taxes have often been found constitutional, with Carlton 1994 the key SCOTUS case.
As far as I can tell, this arises from the distinction between civil and criminal penalties that also give rise to other injustices:
(a) loose process protections for civil asset forfeiture
(b) loose process protections for immigration violations (which somewhat cuts both ways with "immigration violations are civil, not criminal" as one rejoinder to the "deport the criminals" line of argument)
(c) lighter process protections associated with administrative fines by SEC, FCC, et al (though perhaps curtailed since Jarkesy?)
With the current SCOTUS composition sometimes interpreting money issues as the highest form of rights, though, the question of retroactive state taxes might be ripe for re-assessment.
The far right likes to point to Detroit as a cautionary tale of the dangers of “diversity.” Leftists often respond with accusations of racism, which is not an argument, leading some to accept the far-right's claims uncritically. They should not. In truth, what killed Detroit was the 1924 Immigration Act and the anti-discrimination laws of the 1960s.
Without the 1924 Immigration Act, the U.S. would have received several decades worth of European immigration, which was going and would have continued to go to places like Detroit. Back in 1920, when America had a much more open immigration system than it does today, Detroit was 96% white and 4% black. By 1960, after several decades of restrictive immigration, it was up to 29% black. How did blacks go from 29% to the dominant, governing demographic? The obvious answer is the 1967 Detroit Riot. Still, you ask why? How does 30% of the population drive out a higher-IQ, more martially skilled majority twice its size? Because the white community received the message from its political leadership and the rest of the state and country that *we do not have your back*. Shortly after the riot, Michigan governor George Romney (father of Mitt) called the legislature into a special session that passed “fair housing legislation.” The Michigan Historical Review explains:
"Efforts at the state level to enact fair housing legislation proved unavailing until the racial turbulence in the summer of 1967 appeared to signal the need for legislative action to alleviate the condition of the state’s blacks. When Romney called a special session of the legislature following the Detroit riot, the New Detroit Committee-recently formed by leading business firms in the Detroit area to help deal with the city’s problems-sought to persuade the governor to put a fair housing law on the agenda. Romney had promised legislators, who had defeated fair housing proposals in both the 1964 and 1965 legislative sessions, that he would not seek such a measure in the special session. After returning to Lansing, however, from a national speaking tour in which he had stressed the need to deal with racial problems, “he was boxed in and could not say no” to New Detroit. He consequently delivered a message to the legislature on 13 October calling for a “statewide open occupancy law.” He also called for code enforcement legislation as well as a tenants’ rights law designed to create “a covenant of fitness, good repair and compliance with applicable health and safety laws and ordinances for every rental arrangement” in the state."
People who white flighted to the suburbs did so not necessarily because of the mere presence of blacks in the same political entity, but because the government made it impossible to live a white lifestyle in a white apartment building and a white workplace. Had it not been for the 1924 Immigration Act, motivated to a substantial extent by irrational religious sentiment, and the anti-discrimination laws of the 1960s, the mass exodus would not have happened.
Immigration restrictionists might protest that, had it not been for immigration law, Detroit would have later been flooded with non-white “migrants” who would have driven white Detroiters out. There is little reason to think this. Open borders means the entire world. If a bunch of Congolese migrants are causing political problems, employers and landlords could hire Indians, Chinese, or Indonesians instead. A libertarian open borders regime is not one in which anyone can go anywhere and do anything. It’s one in which anyone can rent from willing landlords and work for willing employers. The white communities in Detroit, who owned virtually all the property in 1920, would have been the ones deciding on who to hire and rent to. I trust they would have decided wisely.
Surely all of these are symptoms and aggravating circumstances, not causes ? The main problem with Detroit is that its economy was based around the US auto industry -- which collapsed. Without legal money coming in, crime emerged as the primary economic driver for those who remained in the city. Open borders would not have solved the issue, because immigrants would not immigrate to a city unless there's something there they can use, i.e. jobs. And what is there to do in Detroit, besides crime ?
The big car companies starting moving out of the city of Detroit as early as the late 50’s. Later they moved most of the factories and opened new ones in other parts of the country. It was a whole thing.
There is a big difference whether the jobs disappear gradually or overnight. The market adapts, but that takes time.
A factory can close overnight and leave thousands of people unemployed. One day in future they may all find new jobs, but it definitely won't be the next day. Probably not even the next month. In the meanwhile, you get social disorder.
Detroit suffered from the decline of the auto industry, is scarcely the only US city to experience white flight, and in any event seems to be on the upswing. Case in point, it’s building a huge new bridge to Canada.
I was honestly amazed the Nashville ACX meetup saw a turnout of over ten people, including one country music singer/band lead. Not to stereotype but that’s the last person I’d have expected to be interested in the ACX blog.
Anyway, that was fun so we’re doing it again, this time at the Martin’s BBQ *on Elliston Place* where parking is free and abundant, 2:00, Saturday May 16. See ya’ll there!
You think that understanding that citations are not supposed to be fake requires a high order of intelligence. Interesting.
I certainly didn’t suggest that no person would ever fake a citation. I did suggest the LLM might be deliberately lying but that would be another big problem.
The LLM creators add all sorts of rules to manage LLM behavior. It doesn’t seem like a rule to require real citations would be hard to add.
You're anthropomorphizing. You can give a human a rule like "if you don't know a real source to cite, either find one or cite no source. Don't just make up sources." A human will be able to understand that rule and can choose to follow it.
An LLM doesn't even have the option to "not make up sources." It has no sources and no concept of sources, all it is doing is writing citations. And EVERY citation an LLM cites is made up. But because the LLM has been trained on many real pieces of writing that include citations of real sources--and those citations are encoded in its weight matrix--the citations it writes will often (but not always) closely or even perfectly match some citation in its training corpus, which was written by a human who had a model of the actual, real-world source in their human brain.
But at the end of the day, writing that citation is still just an iterative next-token prediction process. If the distribution of probable next-tokens stays sharply peaked around a *specific* citation[1], then it will likely reproduce that citation, and the citation will point back to a real source. But if the distribution instead has a less-sharp peak around "thing formatted like a citation," without resolving a *specific* citation, then it will just reproduce a thing-matching-the-citation-pattern, which is nearly-guaranteed to be nonsense. To be clear, in NEITHER of those circumstances will the LLM be working forward from "relevant source in mind" to "citation pointing to that source" the way a human would. It can't do the obvious-to-a-human thing you want, which is to consult its memory, discover it doesn't know a source, and then either find a source from which to create a citation and respond appropriately. It's not a question of whether that requires a "high order of intelligence," it's a question of whether it has a human-compatible model of the world with a mental bucket for "relevant source" and a mental process for turning sources into citations[2]. They don't. They can't.
The best that the engineers can do is try to (imprecisely) re-target parts of the next-token-prediction function so that those less-sharply-peaked predictions get mapped to some other output, instead of leading to fake citations. But that's not anywhere near as easy as just writing a "rule" for the LLM; it's dealing with internal program states that are difficult to isolate, understand and modify.
[1] Which will happen because that citation showed up in the training corpus at the end of chunks of text that were statistically similar to whatever's in its context window right then.
[2] Things can get confusing because LLMs are trained on human text and tweaked to talk like humans, so the can summon up the verbiage that attends these mental constructs at will. They can even roughly reproduce a lot of the transformations mapping one sort of text to another sort of text, like "article" -> summary or "paper" -> "citation." But they're not doing it in remotely the same way, so it's a category error to expect the results to consistently be the same.
A base, non-"thinking" model is doing roughly the same thing as a human trying to recall citations from memory.
The rate is low and decreasing, but frontier LLMs sometimes produces both completely fake citations and citations that exist but aren't relevant or don't say what the reference claims. Humans make all those same mistakes. An unchecked reference from a random human about a random topic is not reliable either.
A "thinking" LLM or Codex/Cowork - like LLM+framework can be instructed to check its work by looking up every reference in Westlaw or whatever and verifying that each exists, and even be instructed to read the referenced decision and check that it matches the claim in the citing text. That further reduces the rate of the various kinds of errors. Legal AI software can even use more deterministic database lookups to completely eliminate hallucinated citations. And you can use a different model to check the appropriateness of the citation to the case.
All these are similar to the kinds of error checking and verification an individual lawyer or law firm would do using people. Unchecked junior associates make all those same kinds of errors. Does that mean you should never trust anything a lawyer says? (OK, well...) Modern AI is not categorically less reliable than humans. The claims and citations of specific AI systems may be either more or less reliable than specific configurations of humans.
> "You're anthropomorphizing. You can give a human a rule like "if you don't know a real source to cite, either find one or cite no source. Don't just make up sources." A human will be able to understand that rule and can choose to follow it."
No, I'm not anthropomorphizing.
These things have other rules to not produce overly objectionable results.
If it produces fake citations, it is either programmed to lie or has no understanding of what "citation" means (it's not intelligent).
If it produces fake citations, why trust any other results?
> "Which will happen because that citation showed up in the training corpus at the end of chunks of text that were statistically similar to whatever's in its context window right then."
Is there any indication it's finding these fake citations in it's "training corpus" at all? This wouldn't be an example of intelligence either.
> But they're not doing it in remotely the same way, so it's a category error to expect the results to consistently be the same.
I have yet to hear of a LLM citation that picked a paper at random and cited it, without its backing up what it was being cited about, or one on a subject that contradicts what the LLM says, but I am sure that you could come up with a definition that would encourage that path.
I guess you have no reason to believe me, but I have come across several circumstances like this.
One example:
- I was asking an LLM to find me primary source documents from Founding Fathers about about their opinions on X.
- It said, without giving me a citation, "John Adams believed Y about X"
- I said "I asked you for primary sources, give me one to back that up"
- It said "here is a primary source backing that up, it's a letter from John Adams"
It was a completely unrelated letter from a historical website, neither from nor to John Adams. It was about some random thing.
I've mostly found it bullshiting citations after it says something without a source and subsequently tries to back it up. If it can't find one, it'll sometimes choose a thing that seems like it could be the source.
It mostly happens when you ask it to find sources for things it already confidently claims. I've only noticed it happening when discussing things where there probably exists a lot of unsourced junk statements in the training data (the Founding Fathers think X, the Buddha said Y). You know, stuff where the majority of common discourse is disconnected from documented fact.
That's exactly the sort of thing I wasn't talking about.
If it cited an actual letter by John Adams, you would have to read the letter to tell it was merely picking one at random. Since it wasn't by John Adams, it could be detected at a glance. You could even computerize that sort of validation.
How is "this letter that the AI claims was by John Adams was not by John Adams" anything other than fake citation? Unless I am sceptical enough to check "is that really a letter by John Adams?" I am not going to know any better, and the alarming tendency is to treat everything the AI says as "it's true because machines can't lie".
We get enough fake quotes passed around on social media allegedly where famous person said thing, we certainly don't need AI contributing to "John Adams totally said in this letter from 1995 that his favourite beverage was Pepsi" type slop.
I'm not sure what sort of thing you were talking about, then. Similar things have happened with primary source Buddhist sutras (where it linked specific sutras that didn't support its claim) which were less obvious and more annoying to detect.
This happened in two different ways. First in sourcing non-theravada sutras that did support its claim, but in a conversation specifically about theravada beliefs. You could say this is obvious, but I don't have an encyclopedic knowledge of which specific sutras belong to which traditions so it wasn't obvious to me.
Second, in sourcing sutras which do not at all contain the statement the LLM said they did. I wasn't able to verify it as a bad citation until I had read the whole thing.
The Founding Fathers thing and the Buddhist thing were the only two situations where I noticed it happening though, and I don't think I really understand your position, so I'm not sure if it relates to whatever you were saying.
If it was obvious that it was linking non-theravada sutras, it's not the problem that I talked about. If it was not, if you had to read the sutra to realize that, it is - and a particularly good case if it needed a subject-level expert to do it.
What do you mean "shouldn't"? You can not possibly program a computer with something you can't define. It is necessary, in the sense "it is necessary that you breathe oxygen."
"Meaningful way" would have to be defined for your definition to work.
> “ You can not possibly program a computer with something you can't define.”
???
The LLM is offering the “citation”. If its definition doesn’t include a citation to something real, then the LLM is defective. If it can’t get that basic thing correct, why ever would you expect it to get anything else correct.
> “ "Meaningful way" would have to be defined for your definition to work.”
Read the whole sentence you quoted an excerpt from.
> If it can’t get that basic thing correct, why ever would you expect it to get anything else correct.
Because it gets many or most things sufficiently correct, sufficiently fast and cheaply, so that on balance is still an incredibly useful tool? Also, people also make mistakes, too.
I thought the idea of genuine artificial intelligence is that it saves us the trouble of rigorously defining every single step of every single task we give a computer. If I wanted to do that, I'd just write a regular script.
Every new layer of more advanced computing languages leads to more and more abstract and high-level concepts, but this is done by putting the concept into the language. If the users need not trouble their pretty little heads about what a fake citation is, the programmer has to do it for them, at the language level.
Not to mention that dealing with natural intelligence requires rigorous definition often enough to get what you want.
Before I respond to you, I'm going to need you to define the following terms: layer, advanced, abstract, and high-level. After that, I'm going need a rigorous definition of each word you used in each of those definitions.
Here's a non-paper citation example. When I asked an LLM to give me math olympiad problems on a certain topic, it mixed real math olympiad problems with routine exercises that it made up a math olympiad source for.
Yesterday, I got a mostly-real but slightly-fake citation: a theorem cited as "Last name (year)" where the author *did* publish a paper that year on the right topic, but the conditions in that paper were different from the theorem Claude stated.
Most peer review doesn't bother to verify the citations, then.
Now, once they manage to fake valid citations -- hmmmm -- someone should devise a computer program to *look up* citations. That actually is a definable task.
But what's "real" to an LLM? I would guess it is not just a simple "rule", but rather a ruleset+procedure ("double check that all citations you made are actually real citations" by following the link or finding a paper with the exact name/DOI, etc).
This ruleset might be feasible but computationally expensive, as it widens to amount of tokens and compute massively. Maybe the latter is the real barrier for the LLM-companies, barring any other hard requirement or incentive to do it.
https://www.rottentomatoes.com/m/michael the Michael movie has a rotten tomatoes critic rating of 38% but a user rating of 97%. I don't know if this is historically unprecedented but I haven't seen it before.
It feels a little like an intelligentsia/proletariat split. As though random people (near universally(!)) really don't care - for the sake of movie depictions - about a major figure being a paedo if their story is extremely moving.
Obviously I second guess that a bit... maybe the 97% is bolstered by devoted fans being more likely to go and vote on the movie.
You might have thought now would be the time for people to be *more* hating on paedos than usual though.
Audiences really do not care about a good story when it comes to music biopics. They want to see the exact same formula used to see their favorite musician perform songs they like. It’s the worst genre of movies.
The largest gap as far as I know is for Reagan (2024), which gets a 18% from critics and a 98% from audiences. Presumably for similar reasons to the Michael Jackson one; it's a hagiography that only hardcore fans of the subject are going to see anyway.
Wait, I just found an even larger gap -- it's Melania (2026) with 10% from critics and 99% from audiences.
The biggest gap I know of in the other direction is The Last Jedi, with 91% from critics and 41% from audiences.
I didn’t see the movie, but I saw half the Michael Jackson musical, and up until the 80’s it’s a pretty good story. 5 talented kids, one controlling father, an endless stream of hits, going solo, desegregating MTV… there’s a lot of good material there.
I don't want to sound dismissive but it is very not unprecedented. Michael is on the larger end of the spectrum but the critic/audience gap is a well known phenomenon. If you poke around you can see just among other movies out right now several have 30-40 point gaps.
From the reviews, instead of a biopic it's mostly a collection of Michael Jackson's Greatest Hits, so maybe more a split between "movie" vs. "entertainment".
Such things happen, I think also in the opposite direction. Could be interesting to make a list of the top cases in both directions.
I wonder how much of that is the intelligentsia/proletariat split, and how much is that some movies only get a few votes, so dozen fans or dozen haters can push the result far.
(An interesting mathematical task, to define the strength of difference based on the scores and the numbers of voters. And then sort the movies by that.)
Correcting for sampling error like that is pretty easy. Entry-level multi-level modeling / penalization / shrinkage, supported out of the box in any good regression library. Here's an example from a similar media question where there's an extreme skew in vote count but you're interested in extremes so the flukes are problems: https://gwern.net/goodreads
On the recent discussion waves of miracles. Well if it is okay now for serious people to discuss absurd things, then I actually like the idea of non-theistic miracles more: that there are secret wizards. Does anyone here have anything to add? Like something real life very weakly pointing to that, pretty clear it is not real, but could form the basis of a cool fictional story?
There was this very cool role playing game Ars Magica that was about this, apparently the wizards were of a tradition called Hermetic, and it seems there were real life people experimenting with this: https://en.wikipedia.org/wiki/Franz_Bardon
Unfortunately, it is not actually cool. What little I understand of Bardon is that he developed a system of boring self-discipline, kind of ascetism and then it somehow gives you supernatural abilities. I don't like this. I am very much of a "drunken fist kung-fu" type. I liked those fictional stories more that treated magic like a science, more about working out mathemathical formulas or suchlike. Examples are the novel Rivers of London - who say their magic came from Newton - and the RPG Shadowrun. But I don't think anyone ever even remotely tried that.
I once went to a reiki demonstration, and obviously reiki healing isn't real, but if you move your hands in a certain way for a while, it'll start to feel like there really is a ball of energy in your hands. I don't know the mechanism behind it, probably it's a muscle memory thing like the "floating arms" trick, but it was pretty cool to experience.
Oh, now I remember that a reiki practicioner put a hand on me and I felt a little tingling. But the thing is I was romantically attracted to her, so it could be that.
More interesting was an aikido guy who formed an OK sign with his fingers and apparently not straining to press them hard together and yet I could not force his fingers apart. He also asked me to grab his wrist as hard as I can, I did - and I was lifting weights - and then he still somehow slipped his wrist out from my graps.
I do not have an explanation for this at all, I am just mentioning because it is cool that I actually got to run the experiment, this Royal Society style real try-it-out science like when Feymann was handing out frozen O-rings to senators.
OK, I will explain. I regularly write code, I can understand math or technology to a certain extent like the relationship between thermodynamic and information entropy, and I am not any kind of an ascete. I drink, I smoke, I did drugs in the past, I eat burgers, not vegetarian etc. You can do these things without a strict lifestyle because you only need to be sober when you actually do it. And maybe not even then - Hegel was smoking pot right on the cathedra when he was teaching, I would not make it impossible that important scientific insights happened under the influence of alcohol or drugs. Generally it is possible to be a good scientist and yet live a fairly disordered and hedonistic life, not a life of discipline, the recently deceased Robert Trivers being a good example.
This was to be meant as a contrast to Franz Bardon's approach that "magic" comes from asceticism and discipline. So you gotta be kinda boring like a Theravada monk. Or like Tesla.
Talking it over with a rubber duck is also famous, and so is taking a walk. Crediting it to the drinking is, on the face, a reason to keep on drinking.
Philosophers, like all intellectuals, are good at rationalizing what they want to do anyway.
On the evidence of this very comment, you do not know enough about science or technology to make your claim or disparage mine.
Citing motives is absurd. "Drunken fist kung-fu" is not "I'm drunk, so I will pick a kung-fu fight."
As for the invalid papers -- that's proof. They do not advance science or technology precisely because they are "drunken" -- faking the rigor they need without having it.
I remember reading Werner Heisenberg's Quantentheorie und Philosophie, I don't think there is a complete English translation. He said that writing papers, doing experiments, deducing the math, all that is to prove the idea to other people. You already know that you found a true idea even without this: ideas that are simple and beautiul are true.
He wasn't a Satanist as such, so there was more discipline and learning going on there. He was also a capable mountaineer, so while he very probably was in reaction to his upbringing in the Plymouth Brethren and trying to shock the normies going on about sex'n'drugs (no rock'n'roll as yet), he did put in the work for the esoteric background. I'm never going to agree with him (though I do find the Simon Iff, his authorial self-insert character, stories entertaining) but he wasn't a pure showman like, say, Anton LaVey (and I'm not going to touch with a ten-foot bargepole the worse kind of modern scammers in the great Spiritual Rip-Off Tradition).
Also, from what I've read, "Drunken"-style fighting ironically requires a tremendous amount of physical strength and discipline -- much more so than more conventional fighting styles -- because it requires you to constantly move your body in ways it really wasn't meant to be moved. Capoeira is like this as well, though to a lesser extent.
I've occasionally seen people here referencing John Mearsheimer as an authority on geopolitics. The following excerpt from an interview in 2026 (!) should dispel anyone of the ill-conceived notion that Mearsheimer's worldview is grounded in reality. Addressing a question regarding material support for Ukraine in their fight against Russia:
Any professor who still believes, in 2026, that Russia is an industrial juggernaut which – any day now – will get serious about that little kerfuffle on its Western border, should be relocated into a retirement home ASAP. Apparently he has missed the last 50 years of Russian history, and is under the impression that they can just pump out thousands of tanks a year like at the peak of the USSR, and steamroll Ukraine whenever they choose.
He is not claiming that Russia will "get serious" or that steamrolling Ukraine is easy, he is claiming that the capacity of Russia to produce war materiel is enough to match Western efforts. Which so far has been true, no hypotheticals needed; you can easily find estimates of how many tanks or artillery shells Russia and Western countries are producing per year. Both have ramped up production but Russia is still leading by a considerable margin (even though it is not making "thousands of tanks per year", which is also not something Mearsheimer is actually claiming). This is really entirely unsurprising since Russia, its deceptively unimpressive nominal GDP aside, has preserved much more of its low-tech low-margin industry (mining and steelmaking and such) than the West, and also, being an authoritarian country at war, naturally has a lot more direct control of its industry than Western democratic countries at peace.
Tanks and artillery are of course by now not as important in this war as drones are, but those in turn are built by Ukraine itself (from Chinese parts); the West seems to have difficulties with producing those in a cost-effective way as well.
I don't find this convincing. Your interpretation leads to the same counter-argument: If Russia could have easily scaled up its production of shells and artillery more than it actually did, why didn't it? This would have given them an advantage, especially within the first 2–3 years of the war. And by your interpretation, Mearsheimer claims that Russia would have only ramped up its efforts if the West had supported Ukraine more than it did. How is this supposed to make sense? I don't think Russia had any qualms about making this an un-gentlemanly fight. Please clarify your position.
As for Western support: Ukraine has been asking the West for a larger supply of and fewer restrictions on high-tech weaponry basically since the beginning of the war, right until now. Some countries, notably the US, would indeed have had the stockpiles to send them more than they did. Not enough to decisively win the war, no, but enough to give Ukraine an edge and put them in a better position.
I don't know if Russia could have scaled production even more than it did; the point is, they already did succeed in scaling it a lot better than the West; and the West is at a further disadvantage here, as it has been trying to both rearm itself and arm Ukraine at the same time, while Russia doesn't have to care about anyone else. It would not have been very realistic for the West to arm Ukraine even to match Russia's actual production, much less to allow it outshoot Russia.
Ukraine did eventually get pretty much all it asked for, short of long-range missiles and modern warplanes; there haven't been too many actual gamechangers. European stocks are mostly empty now, the US is also severely depleted, especially now that they got bored and decided to have more Middle East adventures. I agree that there was a window where more/faster material aid to Ukraine likely could have made some difference, when it was still coming out of existing stocks. That is however not the question Mearsheimer is answering, he is being asked "should we be aiding Ukraine more now", and he did not claim that "Russia will get serious any day" as you claimed, just noted that matching Russian war materiel production is not very realistic.
Russia has "succeeded" at scaling, to a level that is clearly inadequate to achieve its military objectives in Ukraine. To even hold on to their meager gains, they've had to dig into reserve stockpiles representing decades of accumulated production, that are in some areas nearly depleted. The West, without even really trying that hard, has scaled production to a level that has been sufficient to prevent a Russian conquest of Ukraine, for four years now. Not by producing more weapons than Russia, but by producing *better* weapons than Russia. And backing the right team, of course.
Adrian's question stands: If the Russians are capable of "scaling" their way to victory, why haven't they done it already?
Once again, I'm not claiming that Russia could have scaled its production significantly more than it has, not within existing political and practical constraints at least. Still, it has succeeded in scaling it better than the West has, just by current numbers, so it is the West who would have needed to match these numbers. If you're claiming that it is the West that hasn't been even trying hard, then that's on you to prove. To me it looks like the West has also done pretty much as best as it could. Again, within existing political and practical constraints; of course, if Russia directly attacked NATO, NATO probably could have coughed up considerably more, but then if the war became truly existential for Russia (if Ukraine somehow would be seriously pushing into Russian territory -- capturing entire regions, I mean, not a border town in Kursk), then Russia, in desperation, probably would have been also doing more, on the scale of nationalizing industries, confiscating deposits and conscripting people for factory work.
There hasn't really been much evidence that Western weapons are really all that much more effective than Russian/Soviet ones, even if they are more advanced, and in some cases for example more ergonomic for the operators or, in case of tanks, more survivable when hit. In any case by now it's the drones that dominate the battlefield.
>so it is the West who would have needed to match these numbers
Why would the west need to "match these numbers"? Victory and defeat are not measured in numbers, and there are many paths to both. But whatever path Russia has chosen, the West has very consistently (and suspiciously precisely) matched in the one way that actually matters - the ability to move the front line in Ukraine.
OK, morale and cohesion also matter, but that's more on Ukraine than the west, and they seem to be holding their own. And if this balance is maintained with the West producing fewer but better main battle tanks in a year, SO WHAT?
> To me it looks like the West has also done pretty much as best as it could.
You're comparing Russia, an active participant in this war, with "the West", which is decidedly not an active participant, and within which some countries treat it as a proxy war at best. Yet somehow you're measuring "the West's" industrial mobilization against Russia's, which is under a much higher pressure to win the war than any single Western country. If you want a fair comparison, look at Russia's support for its proxies, like the Assad regime or Venezuela, when push came to shove. That's right, it was almost non-existent. That's because Russia is close to its limit of industrial, military, and financial capacity, while the West was able to support Ukraine with 100s of billions of USD/EUR in military and financial aid, without noticibly inconveniencing its citizens.
You also seem to have fallen for the Russian propaganda of "Western deindustrialization". While this may be true of some Western countries, it certainly isn't true in general. For example, Germany – the third-largest economy in the world – surely has some problems, but too little industrial capacity to meet demand is not one of them. If anything, it looks like some factories with highly skilled workers in mechanical engineering jobs might soon be available for new business. I'm sure that if you can supply BMW, you can also learn to supply Rheinmetall.
Russia, meanwhile, is right now being deindustrialized – by Ukraine. Google for "Tuapse oil refinery", or "Perm oil refinery", or "Novokuibyshevsk oil refinery", or "Syzran oil refinery", or "Novo-Ufimsk oil refinery", you get the picture.
I don't really think Russia/Putin feels it has "won enough", but it does seem to be pretty much out of ideas by now. The most obvious thing to do would be to have another big mobilization, but that's enormously politically risky, and Putin is extremely risk-averse. Since the last mobilization in 2022 the unspoken social contract in Russia has been that life continues as best as it can, with sanctions and all (and civil economy slowly deteriorating but still going on), and people or businesses don't really need to care about the war if they don't want to. Breaking this social contract risks real discontent. Thus Putin evidently just hopes that the current situation can just drag on until Ukraine finally breaks. It's not a great plan either (it means continuously throwing great numbers of people into the meatgrinder, and, while unlikely, it could end up being Ukraine that comes up with a way to break this stalemate in their favor), but the way the war has been [mis]managed so far, there aren't really any good options remaining.
Mearsheimer is what you get when you take IR Realism way too seriously. There's an important core insight to that theory, but it's also easy to follow off a cliff, particularly when you start confusing "is" with "should be".
It's not 100% clear to me exactly what claim Mearsheimer is making in that short clip, but from what I heard it sounded fairly reasonable.
Russia is expending 5-10% of it's gdp on the war, roughly double their military expenditure before the war, around a 3 or 4 percent increase. Those are hardly total-war spending figures, in ww2 the figures reached 50%+ for a lot of combatants. Surely Russia could increase that another 3 or 4 percent if they really needed to.
" steamroll Ukraine whenever they choose."
Because of various technological advances, drones, satellites etc., that prevent large force concentrations, the war's developed into an attritional stalemate and it wouldn't be strategically sound for either side to expend a lot of material on a single offensive. So you can't assume that Russia is close to it's limit to fight just because it can't make rapid advances.
If they were unopposed, they could advance as quickly as their trucks go, but the Ukrainians also have a say in that; that's why it's called a war. So yes, if they can't go faster, for years, it shows literally the limits of their fighting ability, as determined by the combination of their own will and ability and Ukrainian will and ability.
No, they can't pump out thousands of tanks a year, but they are getting close to producing thousands of drones a day. And NATO can't. And they have spent the lat 4 years on the battlefield learning how to use them, and NATO hasn't.
A few minutes later: "We have shoved NATO expansion into Poland, Hungary, and Czech Republic in 1999". Yeah, standard Russian talking points.
That said, I agree that he is completely wrong about Russia's ability to build tanks... but the objection makes sense if instead of tanks you imagine drones -- basically what happened in Iran. Russia can keep sending cheap stuff and sacrificing thousands of soldiers for a long time, while USA will burn its one-million-dollars-per-missile arsenal quickly.
Then again, if we are discussing a "Russia attacks EU" scenario, we should also include the industrial capacity of EU.
> We have shoved NATO expansion into Poland, Hungary, and Czech Republic in 1999
But that was true. The problem is the continuation of NATO really. A European army would have been less of a threat to Russia, and perhaps inclusive of it.
NATO was not "shoved" anywhere. Countries are free to join or not join NATO, it requires mutual agreement if they do. After the fall of the SU and the Warsaw Pact, most Eastern European countries rushed to join NATO precisely because they knew what Russia is like, and that they would eventually come back in force. They joined NATO because it was in their own interest, not because anyone (but Russia, in a sense) forced them to.
> A European army would have been less of a threat to Russia
A weaker, European-only army would have been an invitation for Russia to do what they are doing to Ukraine right now. NATO is a purely defensive alliance; the biggest threat it poses is to Russian imperialism.
A defensive alliance is still a military alliance, and I think it makes sense to distinguish between different interpretations of "threat": the threat that NATO countries *intend* to pose, the threat that they *could* pose if they wanted to, and the threat that a Russian government might *perceive* from the alliance.
A lot of people insist that it's Russian propaganda that Russia ever felt threatened by the continuation of NATO. I think that's an unnecessarily strong position and that it's perfectly reasonable to take Russia's word for it. One country taking another's defensive measures as a threat is how practically half of all military conflict have ever gotten started.
We can say that NATO expansion was at least a perceived threat to Russia, and still say that it was worth it, because the increased risk of conflict is small and the added deterrence and added defensive measures are significant. This is a more nuanced position, but surely we don't have to be scared of nuanced positions.
> I think […] that it's perfectly reasonable to take Russia's word for it.
You shouldn't, because Russia's own behavior isn't consistent with that believe. If it were, Russia
1) wouldn't have drawn troops and air defenses away from their border with NATO countries, and
2) would've long ago stopped that war of attrition with a non-NATO country, which slowly but steadily is grinding down all their equipment that's necessary for defending against an attack by NATO.
Look at what leaders do, not what they say. And Russia's leadership doesn't behave as if they think of NATO as a realistic threat against Russia proper.
>One country taking another's defensive measures as a threat is how practically half of all military conflict have ever gotten started.
So you take it as a realistic threat to Russia that, say, Latvia is using NATO protection to prepare an offensive war?
Countries that happen to be in NATO can band together to attack Russia, yes, but they would do so on their own, without support from the rest of NATO, because that's simply not in the NATO articles. Let's not forget that in the age of nuclear weapons, credibility is more important than ever because any misjudgment can have catastrophic, irreversible consequences. That credibility includes not only what you'd do, but also what you wouldn't do as member of that alliance.
Call NATO a threat to the Russian way of doing things if you want, but the Russian way of doing things is old-school conquest and hard empire, so my condolences that Russia is prevented from living entirely the Russian way continue to be infinitesimal.
But I think that a paranoid Russian ruler might perceive something like that as a threat, yes.
(And I agree that in the age of nuclear weapons, credibility is more important than ever, but I'm not convinced anyone has managed to have particularly much of it.)
Speaking about "shoving NATO expansion into Poland" has different connotations from "Poland pressuring USA to be admitted to NATO". The latter is true. (The former is Russian propaganda.)
> A European army would have been less of a threat to Russia, and perhaps inclusive of it.
I don't quite see the point of having such inclusive army. Who would it protect Europe from? (If you meant USA, I think EU has defended Greenland successfully even without Russia.) EU is not going around conquering its neighbors, and I don't think we should start helping Russia do that.
> but the objection makes sense if instead of tanks you imagine drones
Yes, absolutely! The Russians aren't a bunch of inepts. They've been massively innovative over the course of the last four years, sometimes faster than the Ukrainians (for example, in the field of fiber-optics on drones). They're pumping out drones by the millions per year, now. When it comes to drone warfare, Russia and Ukraine are the uncontested leaders of the world.
Which is yet another point against the "Russia could win if it really tried" – they're already really trying, and it isn't enough.
He is strange. He used to be worshipped by students, that could be a good thing or could be a bad thing. It is possible that at 79 he is not sharp anymore.
His worldview summed up: after the Soviets collapsed. America started to crearte a global liberal order, based on free trade, supranational institutions and human rights. Since all three limit the sovereignty of national government, nationalists resist it. Mearsheimer thinks nationalism generally wins, because the cultural globalisation we are seeing is only for good times, when there is a big economic downturn, pandemic, anything, people always turn to their national governments for help.
You are going to have a frustrating time to convince people who believe in the "Russia could do it anytime they choose"-theory. You just can't disprove that theory, because, well it could happen any time now, right?
There seems to be a fundamental difference in thought here. You are likely more "bayesian" or something, that tells you "they are in day 1162 of their 3 day special operation, so they have updated my priors on this issue for so long that it is clear they can't do that". What's going on in Mearsheimer-like heads, I don't understand. Denial?
Without rigorous mathematical calculations, "Bayesian" is just a fancy word for "decide what's more likely based on the evidence I've seen".
Here's some of the evidence we've seen:
1. The ongoing war is extremely costly to the Russian economy. The evidence for this is, at the very least, the footage of all the destroyed oil infrastructure (pipelines, refineries, ports).
2. The ongoing war is extremely costly to the Russian military. The evidence for this is, at the very least, the footage of all the destroyed equipment and killed soldiers, and the massive recruitment numbers of ~30,000 per month.
3. The ongoing war is probably dangerous to Putin's and his regime. The evidence for this is, at the very least, the aborted mobilization in late 2022 and the civil unrest it caused.
Given this and lots of other indications, the most likely reality is that if Russia could decisively win the war through conventional, non-nuclear means, it would have done so long ago. There is very, very little available evidence to the contrary. Someone who truly believes otherwise, must
a) have overwhelming, top-secret evidence to the contrary (not plausible in this case),
b) be accidentally or intentionally ignorant of the evidence, or
c) be mentally incapable of processing the evidence.
Mearsheimer has full access to all the publicly available evidence, so he isn't accidentally ignorant. That leaves intentional ignorance or mental deficiency, possibly age-related, as the remaining explanations. In either case, nobody should treat him as an authority on international politics.
Yes, thanks for fleshing that out. That is exactly what I meant.
I'd like to add one more point to the economic cost, the unbelievable opportunity cost that Russia pays for the war: Exclusion of lots of international trade of several commodities; depressed price on many commodities thanks to having to go through intermediaries (like selling oil to India at a discount, which India then sells back to Europe); decreased international cooperation on all levels, ranging across science, politics, trade, culture; further brain-drain; >200 billion € locked assets in the EU alone; the list probably goes on far beyond this.
It's just ridiculous to assume that Russia wouldn't end the war in its favour immediately if it could.
Pursuant to plzdontkillus, I'd like to see Yudkowsky and Soares (or people with a less serious vibe, for product differentiation/market coverage/not burning serious-people-capital as it's had) interview Neuro and Evil (twitch.tv/vedal987). They're both reasonably popular (avg 8000 concurrent viewers, a well-developed clipping ecosystem) and already AI-safety-themed (being that the two are managed llm characters who self-describe as wanting to do AI doom sometimes).
Brainrot update, 4/27/26. The NYTimes profiles Sergey Brin and his MAGA girlfriend, a self-described "holistic health coach" and a "clean meat enthusiast:"
He was previously married to RFK Jr's running mate. I have no idea what a "clean meat enthusiast" is, but I expect it's some kind of epic trainwreck. Brin used to be a Democrat, but last year gave nearly half a million dollars to the RNC. Whether they're wokists in 2017 or GOP donors in 2025, Silicon Valley tech elites can't seem to keep their hands off these whacked out hippy chicks.
If you press the red button, you're fine. If you press the blue button, you die unless more than 50% of people also press the blue button.
My view is that if whatever rule you followed – decision theory, rationality, effective altruism, whatever – leads you to press the blue button, that rule is worse than useless.
Also, if there was some secret third button I could press that would raise the threshold for the blue button pressers to 99%, I would press it.
If you press blue, you get to feel the nobility of risking your life to save lives, plus the righteous outrage of accusing all the reds of collaborating in your murder. A heady draught!
Reframe the question. And get rid of the tribally-coded colors, while you're at it. Using "red" and "blue" is mind-numbingly stupid if you're trying to accomplish anything more than signaling tribal membership and virtue with this one.
A benevolent Mad Scientist has noted that a great many people are suffering from painful, incurably deadly diseases, or similar misfortune, but are denied the possibility of peaceful escape by misguided laws restricting euthanasia. He comes up with a way to give everyone on Earth a box with a green button, and a box with a purple button. If you press the green button, the box goes away. If you press the purple button, you get a quick and painless death. But, as a hedge against extinction, he includes a safeguard where if half the population presses the purple button, the suicide option goes away.
Who thinks the smart move is to press the suicide button, hoping that half the population will also press the suicide button in an attempt to take the suicide option away from those who want it?
And, yes, from half of the people who lacked the mental aptitude to understand the process and make an informed decision, along with the only-marginally-suicidal who would have reconsidered if they had the chance.. If our antihero hadn't been a *mad* scientist, he'd have run it past an IRB that would have pointed out that problem, but here we are. So there is a plan where you hope that half the world's population would press the suicide button in hopes of saving the marginally suicidal and half of the very stupid. If you think that's at all realistic.
You framed the mad scientist as benevolent and the button effect as suicide, but what if we have a framing loaded the opposite way from yours?
Consider: an evil mad scientist has noticed the world has too many people incapable of logic and coordination. There's an army of killbots ready to go; if you agree, just press the green button and the defectors all die! Or you can press the purple no-I-don't-want-to-kill-anyone-today button, and if you prove you are, in fact, better at coordination than the mad scientist theorises - if the purple-pushers outnumber the green-pushers - the mad scientist will admit he was wrong and turn off the killbot army.
Other than the colours, I think the original was more neutrally worded than either of these.
Point being, if the answer depends on how the problem is framed, it's a bad question. And if the strategy depends on most people chosing what you think is the "good" option, then that's a bad plan for a framing-dependent question like this.
Particularly if you think the "good" option is the "I'm dead if this doesn't work" option, because the tiebreaker is pretty much guaranteed to be "I'm not sure how everyone else is going to frame this, so I'm going to play it safe".
This is one of those cases where we try to use social technology to solve a coordination problem, and overly-simplistic 'game theory' interferes with that technology in a detrimental way.
The basic breakdown is this:
If 100% of people press the red button, everyone lives. If 100% of people press the blue button, everyone lives. If you're capable of directing the strategy of 100% of people, it doesn't matter which button they all push.
However, back in the real world, getting 100% of people to do something is basically impossible. In extremus, ignoring all other factors, some people have Parkinsons and their hands will shake and they'll hit the wrong button by accident. In reality, the polls I've seen are like 60%-70% blue and 40%-30% red, so obviously we're not united on this and lots of people are gonna push either button.
But getting 50.00000001% of people to choose the same binary option is not that difficult. In fact, one of the options is going to get >50% of the votes, one way or another; you just have to influence a good number of people in one direction to get the side you want over 50%.
It's way, way easier to get 50.000001% of people to do something, than to get 100% of people to do something.
So any strategy that tries to get everyone to press the red button (and works well enough to get it over 50%) will inevitably kill people. In an ideal thought-experiment world maybe only tens of thousands of people, but empirically in the real world blue is already winning in polls, so maybe you kill close to half the world population. Depending on whether you have time to coordinate and send out social media posts to try to reach people, maybe somewhere in between those two extremes, but... it'll be a huge humanitarian tragedy no matter what.
Whereas any strategy that tries to get everyone to push the blue button has a good chance of saving everyone with no tragedy... and since blue is already winning in polls, we have good reason to think it would work in practice.
Here's the crucial bit about the social technology, though.... we're not actually pushing the button right now, we're just talking about pushing the button.
How should someone using the 'everyone blue' strategy act before the choice is actually offered? How should they further the 'save everyone' strategy in peace times, so it's in place and working well when the reality hits?
Well, for starters, when someone brings up the scenario (or a million other scenarios about altruism and group cohesion and selflessness and etc), you criticize, ostracize, lambast, and punish those who advocate for the red strategy.
Yes, you can make up some shallow 'game theory' saying that the red button is correct for whatever reasons. But deeper game theory takes the entire society into account, and accounts for long-term strategies that let you strategize over more than a single choice.
The people saying 'red' are talking about a very simplistic and short-sighted formulation of the game. They're treating this as a fully decoupled thought experiment, with no importance or relevance to reality.
The people saying blue are already playing the game. They have a strategy, that strategy involves enforcing altruistic strategies across society whenever the opportunity arises, and they are doing it to you right now.
From their perspective, by publicly advocating for the red button, you're actively killing people in expectation right now. Maybe not through this specific scenario, since it's a magical thought experiment, but in general by denigrating cooperative/altruistic/self-sacrificing attitudes and strategies on the whole. Maybe not a lot of people, since this is a tiny internet meme and probably won't permanently shift everyone's beliefs about cooperation and altruism on a wide scale, but a little bit and as part of a larger problem.
This is also a standard disconnect between people who want to 'just talk about' politics, and people who believe that talking about politics is doing politics, since the beliefs and attitudes that arise from those talks determine what voters and politicians will actually do.
>So any strategy that tries to get everyone to press the red button (and works well enough to get it over 50%) will inevitably kill people.
It doesn't take a "strategy" to get >50% of everyone to press the red button. If you do nothing at all, >>50% of people will press the red button, That's human nature, for the vast majority of non-WEIRDs and a good number of WEIRDs, and it's sound strategy for the WEIRDs who might see the advantages of living in the blue-button world but understanding that they don't.
Getting >50% of people to press the blue button, is I believe an impossible task of social engineering on any timescale less than generations in any world other than a dystopic tyranny. Yes, yes, you're seeing >50% blue in polls. Voluntary internet polls of mostly extra-nerdy WEIRDs with no skin in the game, and you are a fool if you take that as a prediction for the whole of humanity in a grave crisis.
You're going to live in the red-button world, or you're going to die. There is nothing you can do that will make it so that you live in the blue-button world. It's red button, or death. Maybe you can convince yourself that there's something noble about that death. But please don't try to convince other people to join you.
Yeah you are just stating your opinion as fact here. Stating it 50 times in a row, and adding disparaging comments about anyone who disagrees, isn't actually additional evidence for it.
Falsely accusing me of things I haven't done, is hardly strong evidence for your position. And I think strongly implies that you don't have much in the way of evidence.
> you criticize, ostracize, lambast, and punish those who advocate for the red strategy. …they are doing it to you right now.
Yes! You get it! And I'm saying that makes their living so clearly a negative to me that it's worth paying an extremely steep price to get rid of them. Half the world's population is far too cheap, which is why I added my comment at the end upping my price to at least 99%.
You personally would not actually be happier in a world without social policing.
I don't care how autistic you are, if you own a computer you still benefit from society not collapsing.
You're making the classic mistake of noticing the 1% of something that annoys you and declaring that it should be demolished, without noticing the 99% of it that quietly makes your lifestyle possible and comfortable every day.
Basically the same as the people who want to abolish capitalism because landlords are annoying.
"Also, if there was some secret third button I could press that would raise the threshold for the blue button pressers to 99%, I would press it."
Well, well, this is quite the "mask-off" moment, isn't it?
For those not familiar, whatever else may be said of Shankar, he unquestionably has a high degree of mathematical competence. So from someone else this might be a slip, I think there's no chance at all he didn't appreciate the mathematical implications of that last sentence.
I've probably read a few dozen comments in the genre of "militating for the red button," and they have some similarities. First they all, almost without fail, treat it as a very easy question, insisting with apparently complete certainty and sincerity that it's "obvious," "rational," a "no-brainer" and whatnot to press the red button.
Second, they strongly assume (implicitly or explicitly) that there's not the slightest question of the overall outcome: more than half of humanity will choose the red button, absolutely guaranteed.
Third, they all express some manner of contempt or disdain for people who would press the blue button.
Taken together this forms a sort of motte and bailey. The bailey is some sort of tribalist or eugenicist belief that it's actually great if the blue-button pressers all die as evidenced (usually implicitly but sometimes outright) in the third point. The motte is the first two assumptions: IF the choice is obvious and the outcome is not in doubt, then pressing the blue button has no upside and is tantamount to suicide and the victims can be safely written off as "idiots" or "asking for it"[1] or very occasionally some gentler euphemism for one of those.
But Shankar, brave sole that he is, has opted to nuke his motte from orbit and plant his flag proudly atop the bailey. He tells us outright (assuming we've stopped to do the math) that he is *quite eager* to see the blue button pressers die. That he wants as many of them to die as possible. How has he so thoroughly demolished his motte, you ask? By conjuring up a third option that would do *absolutely nothing* if he actually believed the outcome were not in doubt. If he believed that blue-button-pressers were only a tiny fraction of the population--which is the implicit frame that makes the second and third assumptions work--the chance of his Secret Third Thing doing anything at all would be infinitesimal. It's only a meaningful choice if you expect some non-tiny chance that more than half of humanity will press the blue button and consider sentencing them to death for the crime of thinking that way to be a positive moral good. If he didn't consider the "wrong" 98% of humanity dying to be a better outcome than *merely* the "wrong" 49% dying, he'd never have volunteered such a choice as one of the hypotheticals.
Either that or he's lying about his preferences for shock value: but that would be quite an intellectually rude thing to assume.
I feel sad that I've been trained not to push the blue button. Upon reflection, the blue button is superior but my brain keeps screaming "trap!".
Like, here's my though process:
Me: "Why would anyone ever pick the blue button. Everyone should just pick the red button."
Me: "Wait, what about children? Or retarded people? Or people in the lizardman constant? Upon reflection, a small subgroup is always going to pick the blue button. Therefore, we should organize and all press the blue button to protect retarded people."
Brain: "TRAP!"
Me: "Why trap?"
Brain: "We see this all the time. We can easily get thing X but we could also get better thing Y if only we all coordinate, then we all coordinate but we never get better thing Y, we only get worse thing Z.
Me: "Interesting. Could you provide some examples?"
Brain: "EXTREME CULTURE WAR!"
Me: "Could you provide some examples that won't immediately brain kill the reader?"
Brain: "Remember all that time in California where every election there's a ballot proposal to increase the sales tax by like 0.25% to improve school funding and everyone votes for it and then the school doesn't get more funding so the next election there's a ballot proposition to raise a $3 billion bond for the schools and we all vote for it the city and county services close to pay off the bond and the schools don't get any more funding and then we iterate on that for 30 years until absolute hopelessness and despair set it because the taxes never go down, the services never come back, and the schools never get more funding?"
Me: "Yeah, I didn't like that."
Brain: "How many times can you think of where that happened? Where we were all promised if we just came together and did the obviously right coordinated action, it would lead to the obviously better result, only to end up in a worse situation?"
Me: "I can think of a lot of times where that happened?"
Brain: "Can you think of any times where the inverse happened? Where we all organized together for the collective good and then we actually got the collective good?"
Me: "...no. That makes me angry though. I want the collective good thing. And it should be so easy to get."
Brain: "I realize this is rich coming from me but maybe we should just embrace retarded pattern matching. If there's a berry in the forest and it looks delicious and smells delicious and tastes delicious and absolutely should be a great berry but everyone who eats the berry starts inexplicably screaming and then 4 hours later their testicles explode, then let's not eat the berry regardless of how much sense that makes."
Me: "So red button?"
Brain: "Red button. My priors are 90% that you will inexplicably suffer bad results for pressing the blue button, even though that makes no sense, because that is what consistently happens every time we're in similar situations."
Children dying is bad. Retarded people dying is bad. Defending people who are weak and/or helpless is a good thing.
If someone steals charity money meant for orphans, that doesn't mean giving money to orphans is bad, it means the thieves are morally abominable. I get Russell conjugated vice signaling as Overton preservation, I just don't think this is a proper place to make that argument.
Basically, adopting extreme/negative stances to prevent moral blackmailing, typically adopted by rightists against leftists. (1)
Eg, Person 1 is discussing a sensitive subject. Person 2 objects that this is racist. One of the most effective rhetorical strategies Person 1 can employ is "Yup, I'm a racist" and then go back to discussing the original subject. Adopt the worst frame to prevent moral/emotional blackmail.
Kids dying is bad. Retarded people, insane people, and otherwise generally helpless dying is also bad. If you genuinely think these people dying is good, that's bad. If you think "Kids dying is bad but if I say that I will be morally blackmailed into supporting Gay Race Communism forever so I will say kids dying is good", I get that, I understand the application in most online conversations, I just think Autism Land where we explicitly state what we believe is the wrong place to engage in this rhetorical tactic.
Leftists who engage in moral blackmail are the equivalent to people who steal charity money from orphans. They are literally burning the milk of human kindness for transitory political and financial rewards. That does not make giving money to orphans bad.
If compassion makes you vulnerable to parasites, then maybe some sacrifices need to made? These decisions don't exist in a vacuum. If the only way to produce a stable and efficient society is to eliminate anything worthy of pity... perhaps that's the only path left for humanity.
My running gag hypothesis is that it was deliberately injected into media by whatever dark forces are using social media to manipulate us, to test whether we're primed to believe flatly irrational things. If not, it keeps doing the usual shit-stirring it's been doing. If so, it proceeds to phase 2.
It could just ask us whether there are five lights, but too many people have seen Star Trek.
The poll doesn’t really measure courage, altruism, or strategic thinking. It measures whether you’ve thought about it long enough to realize blue is irrational — at which point you feel vaguely gross for picking red, even though it’s the logical answer.
Yeah, you can always tell which people are rational, because they're the ones who crisply and smugly explain how rational they are, apropos of nothing.
(Probably at least 50% of the people I've seen trying to explain why *they personally* would choose red have used the word "rational" or "rationally" or their antonyms in their explanation. Which is funny in a dark way, because they're out here in public showing just how BAD they are at understand other humans. Hardly the mark of a keen intellect.)
This is talking as if the discussion just appeared on this forum fully-developed, instead of being posted by a specific individual, complete with commentary about how he sees nothing wrong with killing 98.99% of humanity if they happen not to think like he does. Original hypothetical aside, I'm not sure in what universe THAT doesn't count as a question of "personal virtue."
This roiling, this anger, this back and forth sniping was the purpose of the posting the hypothetical.
It reminds me of the crazy — almost said irrational here — agitation stirring meme I saw on Facebook when I could still tolerate Facebook, “Who is willing to say the Pledge of Allegiance in its original form?” Original form of the pledge does not contain the words “under God” which would be added during the Eisenhower administration so naturally some portion of the audience flips out. It was designed to sow chaos as was this hypothetical.
Trolls love stuff like this. In retrospect i should have just deprived it of oxygen.
I'll agree that there's a good chance this post was intended as rage-bait, but I'm not personally angered by it. My gast was pretty flabbered the first time I saw discourse on this topic, but there's far too much actual bad stuff in the world to get upset and people posting monstrous hypotheticals online.
I engage in things like this because I see value in trying to strip away the layers of obfuscation confused and morally-warped viewpoints, in the hope that it will help others see them more clearly for what they are (not that the original post is bothering much with obfuscation). They say the remedy to bad speech is better speech. I have my doubts about applying that axiom generally, but it's certainly appropriate to the culture here, and when in Rome...
Self-control is hard though. I replied to OP, then I deleted my comment a few seconds later after remembering how unfruitful this discussion was a few years ago, but then after work I still commented in a sub-thread. Mea culpa.
I'm more than a little confused by this. There is no downside to pressing the red button, and a potential downside to blue. So what's the point of this poll? And yes, I understand it's supposed to be an analogy about individualistic/collectivist, Rep/Dem, isolationist/internationalist and so on, but as it is, it captures none of the real life consequences of pressing the red button, nor the inability of some to press it. It's just a button.
Like 70% of people say they will pres blue when you poll representative samples.
If this is shocking to you, then the point of the thought experiment was for you to recognize how out of touch you are with how the general public thinks.
Which is important, because this is a coordination problem, and you can't coordinate with people you can't understand.
"Like 70% of people say they will press blue when you poll representative samples."
Representative samples of *what*? You refer to "how the general public thinks". First, is that the general public of the United States of America, the general public of the World, the general public of the generally WEIRD nations, or some other subset of that? And second, how did you get to the *general* public, as opposed to the enthusiastically-online-internet-polltaking public? Were these polls conducted by competent polling organizations that can at least approximately correct for nonresponse bias? Where did they even go to find potential responders?
If you're going to point to these polls in your arguments, then I think these are reasonable questions for you to answer. At very minimum, you need to link to the polls. And be prepared for mockery if they're the usual sort of lame-ass internet poll.
The poll I was referring to is by Blue Rose Research and said it was a representative sample of Americans. I believe that it was done by a legitimate national polling firm, I assume tacked onto some other survey they were doing so they could publish the result as self-promotion (a fairly common marketing tactic for these firms).
You're correct that it was not a worldwide poll, which is relevant, but I believe the null hypothesis here is that other countries would do about the same on average (obviously every country will have its own idiosyncratic result, but there's no *a priori* reason to expect the US to be wildly different from the global average, you'd have to make an argument for that and back it up before assuming it).
But I also want to point out that it barely matters. Even a crappy internet poll of a nonrepresentative sample would be more evidence then 'I just think this is obviously what people would do, in my head'. No one who says 'everyone will obviously vote red' has presented *any* empirical evidence for this belief that I have seen, even a poll with a 20% margin of error would tell you more than that.
I do not engage with the kind of social media where this would be "a thing". Upside: My mind has not yet been poisoned by public discourse on the topic. Downside: I am probably going to say nothing you have not heard already. But please humor me anyway. This is genuinely puzzling to me and you seem to believe it would be beneficial if I understood your reasoning. I agree on the latter.
---
Option A: I and my family don't die (good). Everyone else who picks option A doesn't die (good). Everyone is free to pick option A and get the same deal (also good).
Option B. There is a non-zero chance that I will die (very bad). There is a non-zero chance that everybody else who picked this option will die (very bad).
The preference for B seems to hinge on the possibility to "save" other B choosers which - by assumption - would not need saving if they had chosen A which - by assumption - they were free to do in the first place.
---
This is not an accurate analogy for politics. In the real world there is no magical "I'll be fine button" available to everyone.
You have a meta-choice here:
I. You can engage with this as a brain teaser and choose the best option _within the fiction of the brain teaser_. That best fictional option is A.
II. You can engage with this as a political fiction. It is very very explicitly a red fiction. The correct blue reaction is to engage with it being a fiction and analyse the ways in which it is a fiction completely divorced from reality. That is very easy to do.
But what you really really should not do is to conflate I and II. The reason the "red button" is not the correct option is that it is a fiction. Pretending that within that fiction choosing blue is the best option because you don't want to accept the fiction is just confused.
---
In short, the leftist answer here is to smash the device with the two buttons.
>Option A: I and my family don't die (good). Everyone else who picks option A doesn't die (good). Everyone is free to pick option A and get the same deal (also good).
>Option B. There is a non-zero chance that I will die (very bad). There is a non-zero chance that everybody else who picked this option will die (very bad).
This is just incorrect. Here's the actual math:
Option A: I don't die (good). Everyone who picks option A doesn't die (good). Everyone is free to pick option A and get the same deal (also good). There is a slightly higher chance that everybody who picked Option B will die (worse).
Option B: There is a non-zero chance that I will die (very bad). Everyone who picks option A doesn't die (good). Everyone is free to pick option A and have no chance to die (also good). There is a slightly lower chance that everybody who picked Option B will die (better).
The only calculation here is your estimate of the chance that you pressing blue will be decisive in making blue win, multiplied by the number of people you expect to die if blue loses (both from pressing blue initially, and from the global catastrophe following billions dying suddenly across the globe) to give the expected deaths caused you choosing red, compared to your estimated likelihood of blue losing multiplied by the number of lives you would sacrifice yourself to save (which is the threshold at which you should choose blue).
So if you engage with the brain teaser the answer is not obvious, it's the output of that calculation. My priors and values, plus the polls I've seen, to me that for myself Blue dominates that calculation pretty definitively. Different people will have different priors and values, of course, but you have to actually come to grips with the calculation you're making and think about your intuitions for each term before declaring it 'obvious'.
As for engaging with it as a piece of fiction, I disagree somewhat.
Yes if you are trying to use it solely as a metaphor for D/R political ideologies, it has a lot of problems, including the fact that the Rs can say 'well I have no obligation to hit the blue button here because everyone can take personal responsibility in hitting red and then it's their own fault if they die', which is unlike reality where the poor and oppressed can't just choose to not be poor and oppressed.
But I think it has utility outside that narrow metaphor. More broadly it is a parable of self interest vs. cooperation.
I think it's been useful in exposing people's thinking about those dynamics in general - the number of people in this thread and others saying 'obviously everyone will hit red' and then being confronted by polling showing 60%-70% saying blue, should in and of itself be a valuable learning experience that causes a lot of people to adjust their priors closer to reality (you'd really hope!).
And more generally, thought experiments about cooperation vs. self-interest serve as an opportunity to pre-coordinate and set social norms and expectations around the real versions of these types of choices. There's a reason we teach children morality through fairy tales with simple stories and big obvious morals, instead of putting them in a history of political philosophy class or w/e - simple thought experiments get at the core intuitions behind more complex moral behavior, and shape them in a powerful way. Getting everyone to talk about these things publicly has value.
> The only calculation here is your estimate of the chance that you pressing blue will be decisive in making blue win, multiplied by the number of people you expect to die if blue loses (both from pressing blue initially, and from the global catastrophe following billions dying suddenly across the globe) to give the expected deaths caused you choosing red, compared to your estimated likelihood of blue losing multiplied by the number of lives you would sacrifice yourself to save (which is the threshold at which you should choose blue).
Well that is easy. Nobody in their right might would choose blue so my chance to be decisive is zero (as I would be the only one picking blue) and while not being an independent variable, I expect precisely nobody to die if blue looses because everyone will choose red anyway and therefore be safe.
So it seems, that by your own standard of evaluation the belief in the red option is self stabilising.
"Aha" you say "but that is where you are wrong, in fact many people will choose blue".
Which empirically might be the case but it seems to suggest that they are evaluating the problem by a different standard than the one you are suggesting.
The calculation isn't about hypothetical people who all follow a specific decision theory, the calculation is about actual people in the real world.
As I said:
>the number of people in this thread and others saying 'obviously everyone will hit red' and then being confronted by polling showing 60%-70% saying blue, should in and of itself be a valuable learning experience that causes a lot of people to adjust their priors closer to reality
It was now explained to me that the way "team blue" sees it this _is_ an accurate analogy for politics after all along the following lines:
- The question is whether the red team or the blue team "wins"
- If red wins they kill all the blues, if blue wins they do no such thing.
Note that this is not covered by the brainteaser framing. In the brainteaser there is no reason to be on a "team" in the first place. In the brainteaser, If you do not want to die you pick red and it has no further consequences.
Within the fiction of the brainteaser it ist not "the red team" that "kills the blue team", it is the unembodied gamemaker.
Which is one more way in which this fiction is explicitly red coded.
Most people actually do consider "maybe participating in the murder or millions or billions of other humans" to be a downside.
No analogy is necessary. If you take the hypothetical at face value, both buttons have a downside. The downside of blue is a probability of dying yourself. The downside of red is the probability of killing a staggering number of other people unnecessarily. If you don't understand why some people find that second downside to be rather more compelling than the first, I'm not sure there's any way to explain it to you.
This line of reasoning is interesting, it's kind of meta-selfish. Like the reason you think of a bunch of people dying could be bad is that it would put some level of moral blame on you, rather than that outcome simply being bad and worth avoiding
They are dying regardless, I am fully convinced there is no way more than 50% would press blue. And my choice would drown in billions of others anyway. But they also have put themselves to danger. Is it my responsibility to save every single idiot or suicidal person on Earth?
The upside of pressing blue is that you can contribute to saving those who pressed blue and the downside of pressing red is that you contribute to killing people who pressed blue.
Some people believe that those who make a logical mistake should die and there is no value in saving them. For those people, this hypothetical will be pointless.
"Some people believe that those who make a logical mistake should die"
The number of people I see flatly asserting pressing blue is a "logical mistake" without preferring even a hint of logic there is staggering. As a game theory question, it's not even that complex. And the logic doesn't quite work out the way you seem to think.
I did not assert that blue pressers make a logical mistake, so your last sentence is unjustified.
Blue-pressers making a logical mistake is the belief of a subset of the red-pressers. I addressed my comment to them, showing that even if we take this as a given, bluepressing is pointless only when we accept independent, misanthropic beliefs.
>Some people believe that those who make a logical mistake should die and there is no value in saving them
The problem is that the analogy is deeply flawed that way, because it's not a purely logical choice. It's only a purely logical choice if everyone had an equal choice of pressing red or blue, which is not true in reality. Leaving out that connection to reality is no minor sin. It's the entire reason to press blue IRL, to help those who can't press the red button.
There are also some people who think that those who make a mistake might still have lives of non-negative worth, since they might be able to be taught to be better, but those intelligent enough to understand the scenario perfectly, and still endorse pressing blue for "moral" reasons, don't.
Whenever such a poll is conducted, there are lots of people (usually a majority), who vote to press the blue button. I seems to me exceedingly valuable to know that.
I low-key like it. Because reds are not simply evil or uncaring, they expect everybody will choose red and everybody will survive. From their perspective only risk-taking fools would ever choose blue.
I mean red is a truly simple coordination mechanism, everybody follows self-interest and everybody survives. This is what I would expect in cynical cultures from Eastern Europe to China.
Blue is confusing - you have the altruistic motivation to save others, but the very reason they need to be saved is their own altruism, they would not be in need of saving if they were just selfish! This illustrates the confusing nature of Western idealism - you either unthinkingly choose the "nice" option, or if you are smarter, you assume (correctly from the results) that many other people will unthinkly choose the "nice" option and thus need saving.
OTOH I do think this confusing idealism makes people cooperate better.
I choose red and I feel evil and uncaring about it.
Based on the context of the poll, I expect in real life the reds would win. But I am not sure, and many blues will die, so by choosing red I am being evil. If humanity were better at coordination they would choose blue, but they aren't (I have no reason to think I am not a representative member of it)
"I low-key like it. Because reds are not simply evil or uncaring, they expect everybody will choose red and everybody will survive. From their perspective only risk-taking fools would ever choose blue."
If looking at a scenario like this and blithely concluding "eh, it's fine if everybody in the world who doesn't think like me dies," DOESN'T count as "evil or uncaring," I'm genuinely unsure what would.
If you are not a thoroughly-isolated loner, there is a very high chance that somebody important to you would press blue. Maybe you already know who, maybe you don't. The only realities in which *you don't lose those people* are the ones in which the majority of people press blue.
To put it in explicit mathematical terms, it's a choice between:
1. [guarantee of life]*[high chance that life becomes much worse]
2. [moderate chance of life]*[guarantee that life doesn't get worse]
Even if you completely discount the possibility of your choice flipping the result[1], there are lots of people who perfectly rationally prefer 2 over 1. If you don't understand that, then you necessarily have a poor model of the world.
[1] Which is game-theoretically correct if and ONLY if you value the lives of strangers at zero or near-zero. Otherwise the low chance of making a difference is proportionally balance by the high impact of the difference, since they're both proportional to the global population.
My current answer is to choose red and I don't think blues deserve to die, if anything they are better and more worthy than me, only (maybe) guilty of naivete.
I value the life of strangers at much more than zero, but less than my own. But, my answer is seriously engaging with the question and not a way of signaling or discussing in the abstract how humans "should" coordinate. And since I am cynical and think red will win, I would rather live than die.
> If looking at a scenario like this and blithely concluding "eh, it's fine if everybody in the world who doesn't think like me dies," DOESN'T count as "evil or uncaring," I'm genuinely unsure what would.
Apparently, choosing blue. If tolerating the weak and deviant is a sin, then failing to kill the irrational would be a crime against humanity.
Observation: Chinese ppl on Rednote choose red more than ppl on X, thinking red is strictly better and rational ppl can coordinate into all-red. It would likely end up 60-40, 50-50, or 40-60. I suppose it's because red means INT signal & Moral Kidnapping Resistance which is as valued as virtue signal. (There's a Chinese word lit. Moral Kidnapping, meaning using morality as a tool to force others to perform obligations beyond their limits, and 🇨🇳 web culture disgusts it very much, even noticing it and resisting it is virtue.) I think it's like there's exactly 2 best outcome and everything in between seems like corrupting the coordination. One may think whoever chooses the other color is corrupting the coordination and risking ppl's lives. That's why I love this thought experiment so much! (I personally choose red because the coordinate into all-red thing.)
Red is significantly more popular when the variation of rule explicitly says "only ppl who can understand the rule need to choose". Blue is slightly more popular when the variation of rule explicitly says "everyone's choice will be made public."
Yeah I think that statements of the problem need to be explicit about the arrangements for (e.g.) babies who are just going to press whichever colour they like better. Because otherwise people tend to make assumptions about whether or not they're included and have to choose for themselves.
I think the problem is more interesting, and more distinct from other problems in the same vein, if you make it explicit that only adults capable of understanding the instructions will be forced to choose.
I don't think you need to assume it to make it a *sensible* thought experiment, just in order to make it a particularly interesting one.
Without that assumption it just reduces to some sort of ordinary many-player prisoner's dilemma, where "Everyone Defects" is worse than "Everyone Cooperates".
The interesting version is the one where Everyone Defects is exactly as good an outcome as Everyone Cooperates, and yet we find some people choose to cooperate anyway.
It would be interesting to try other variations, which you could do experimentally. There are nine people in a room; if a majority choose blue then everyone gets ten bucks, otherwise only red-choosers get ten bucks. Try with different amounts, and let people coordinate or not.
What I'm understanding from this is that a lot of people are keen to (be seen to) make the "pro-social" choice over the "selfish" one even in cases where it makes absolutely no sense whatsoever.
You could imagine adding a slight complication so that it's a meaningful question. If >50% choose blue then everybody lives, if >50% choose red then only the red-choosers live *and* we shoot this puppy. Now it becomes a meaningful moral dilemma and you can imagine scaling the "puppy" cost upwards or downwards to explore how people respond.
But it turns out that a significant number of people are willing to do the blue even when we scale down to zero puppies.
I would be very interested in seeing how this correlates with politics; my guess is "almost perfectly".
Why should anyone else accept their (your) framing? If I were to declare that in fact it's better for society if all those who press blue were to die, and the trouble of having to deal with their corpses the immense burden those who press red pro-socially choose to take on, while the blue button pressers selfishly choose to do nothing but lie still and rot, would you then support red being described as the pro-social position?
In fact, pressing the blue button only potentially helps other blue button pressers. The red button people are entirely unaffected by this choice. Suppose pressing the red button gave you a follow-up question, where pressing, say, the gold button, gave all those who press it a million dollars, but only if more than half the people offered it press it, while those who press the gray button get nothing. I bet you wouldn't call the gold there a "pro-social" choice.
"What I'm understanding from this is that a lot of people are keen to (be seen to) make the "pro-social" choice over the "selfish" one..."
This is funny because every single online discussion of this I've seen so far (this plus 2-3 others) have been absolutely *dominated* by dudes offering one of two or three variations of the same smug, faux-rational explanation of why Red is the Objectively Correct Choice, often with an addendum about how All the Blue-Pickers Dying is Good, Actually.
So quit the opposite of what you say, it seems like there are quite a lot of people who are positively bursting to tell the world how they will proudly pick the "anti-social" choice over "maybe I don't want to lose my friends and family to the easiest prisoner's dilemma ever" choice.
"... even in cases where it makes absolutely no sense whatsoever."
If it makes no sense to you, that's a fact about you, not a fact about humanity as a whole. From what I've seen of the world, most people do actually value the lives of others to a non-zero degree[1], and that's all that's actually necessary for it to make sense.
[1] Yes, even when those others are acting "stupid" or "irrational" in your judgement.
Has anyone done this one: "If you press the blue button, you die unless more than X% of people also press the blue button. What's the maximum X for which you'd be willing to press the blue button?"
I'd be interested to hear how many of the people who believe one choice in the original question is obviously correct also thinks it's obviously correct even at the extremes. Because I think a sensible blue might change their mind well before X = 100-epsilon (sorry, there's just no way that much of humanity is going to work with you on blue; do you want certain death just to register your protest vote?), and a sensible red might change their mind well above X = epsilon (you have a virtually certain chance you'll still survive while helping to save e.g. the color-blind people, babies, and others who couldn't understand the prompt through no fault of their own)
One pundit says pressing red is a “phenomenally disgustingly selfish decision.” But when challenged if he’d pick blue on behalf of his own children, suddenly the “moral calculus changes” as it’s an issue of “consent,” lol.
He's not merely "donating a kidney," he's demanding others do the same and calling them phenomenally disgustingly selfish if they don't. Yet with his own children it's different.
Actually, it's even funnier than that: they're all cutting out both their kidneys and swapping them around with each other through some complicated procedure that fails catastrophically if not enough people participate, even though everyone had functional kidneys to start with.
I strongly suspect that "consent language", originally something noble (basically classical liberalism, human autonomy and dignity) strangely morphed into an excuse for bad behaviour. Both having consent and the lack of it.
"If the game involves every single human on the planet, it heavily favors the red button, as most (2/3-ish) of the world live in low trust societies where you can probably expect 60%+ of people to push red."
The discussions I've seen of this are full of completely unqualified, apparently totally confident assertions like this accompanied neither a shred of evidence nor a scrap of self-awareness. I fully believe the people saying this live in the world they purport to see: a world where everybody thinks first, last and only about themselves and questions like "what wider impact will this have" are dismissed with a sneer if they're asked at all. But that's a fact about those people and their social groups, not a fact about humanity as a whole. As for why more compassionate and emotionally mature people are rarely to be found in those same social groups, that may have to remain a mystery for the ages.
"There is probably some not too hard math you could do to calculate P(You make a difference by voting blue) and P(You die) based on your guess of the average human, and I think the numbers heavily favor red."
Yeah, why do some "probably not too hard math" when you could just *act* like you've done it and pull out the answer you wanted to begin with?
In reality, the math depends on your utility function. But if you value human life in the abstract--not just the people close to you, but also random strangers you've never met--then the low probability of your vote being decisive matters less than you appear to have realized. Low probability is balanced out by high impact.
My chronic-pain issues have become so numerous that in just the last three months, I have had three different specialists tell me they do not know how to explain three entirely different issues. Each suggested that the pain likely stems from an unidentified systemic issue they did not know how to help me with. I'm headed to Cleveland Clinic next week for a holistic evaluation and fully expect to be told it's all central sensitization, which fits my experience more closely than any diagnosis I've seen but is a tremendously unsatisfying answer to the question, "Why did I develop 12 different chronic-pain conditions in my twenties?"
Central sensitization is a mechanism, not a disease or even a syndrome; the main thing it does in my opinion is help people understand what’s going on and that they’re not crazy, and that they probably don’t need a department for every painful area.
Knowing nothing about you or your difficulties, I just want to a) offer my sympathies, that sucks real bad, and b) predict the answer will turn out to be something something overreactive immune system.
I'm sorry I have nothing to contribute regarding your pain--that really sucks, sorry to hear it--but while you're at the Cleveland Clinic, if you have time, you at least will be in the vicinity of Li Wah, a most excellent Chinese restaurant.
Maybe try listening very loudly to Bone Thugs N Harmony while you drive around that area...? It's always made me feel good.
I hope I do not come across as glib but I would say there are probably 12 things about yourself that you need to find out more about in your body than you have. With obviously the condition that some of them might have real physical causes.
I'm looking forward to seeing *Killhouse*, a new Ukrainian thriller film. It's funny listening to the US President and his national security team speak Ukrainian, but I prefer subtitles to the trailer version dubbed in English (The dubbing is pretty lame).
What the hell are the Botez sisters doing at Lighthaven? Very charitably, "chess influencers" are now welcome too? What does that have to do with anything?
Can any grifter now associate themselves with rat?
I think the question is how you want to divide their success among the following candidate contributing factors:
- chess skills
- personality
- looks
- luck
- things that can plausibly be taught in a short workshop and replicated by another person.
At best I would expect some generic advice along the lines of "be really good at something, preferrably while being a person that works well on camera".
Now don't get me wrong, that is solid advice and why wouldn't you want it delivered to you by a Botez sister if you get the chance. It just doesn't seem to be what people call "actionable".
A pretty face is definitely a bonus in this industry, but there are millions of pretty faces out there -- why follow this specific one? And even if we assume that 90% of followers are there for the pretty face, that still leaves 200 000 genuine ones.
200 000 followers would be nice to have. If 1 in 100 is willing to send you 1€ a month on Patreon, you can immediately retire (in Eastern Europe).
My guess would be that the main advice is that you need to produce a lot; that the income is superlinear to the number of videos produced. The kind of people who spend most time on YouTube are the ones who want to see a new video every day, and if you won't give it to them, someone else will.
I agree that you need to be good at something, although some people have succeeded with things like being good at being silly.
Obviously, if you are a young woman, that's a 10x multiplier. But it still matters what number you multiplied.
Then I would like to get some specific advice, like how to work with camera, what kind of camera to buy, what is the optimal length of the video, whether to share it on multiple services and what is a convenient way to set that up, etc.
Then I guess it's mostly peer pressure; putting yourself in a place where everyone makes 1 video a day can switch your monkey brain to the mode where "making 1 video a day" feels like the obviously correct thing to do.
But also, if you are there, making videos literally every day, and getting feedback on those videos every day from multiple experienced people, that is actually a lot of feedback. How many experts would be happy to give you 30 iterations?
Practice makes one better, but there is a long way to go.
(Yeah, "practice" and "feedback" are two different things, but they work best together. Feedback without practice is wasted. Practice without feedback has the risk of making the same mistakes over and over again. 30 days of everyday practice and everyday feedback is incredible; I would only worry that some of the advice takes more time to process properly.)
As a married man of 30 years, I'm obligated to say, "Really? I hadn't noticed. I was just fascinated by their erudite discussion of Chess. No, I don't play. Can't stand the game."
Thank you for sharing that important disclaimer. Like the comment I found on a nuclear engineering course held at MIT hosted on YouTube by a cute Asian chick:
"My wife is asking why I'm so interested in nuclear engineering these days"
Is ranked-choice or approval voting a cause worth supporting or donating to?
Context :
Partisan polarization harms federal governance in the US. Red and blue teams, federally, seem to be moving away from the political center, so whichever party takes power produce large policy swings and govern in ways many find unacceptable. Some people think alternative voting methods—like ranked-choice or approval voting—could incentivize candidates to appeal to the center, making more moderate officials electable, and might also lead to less acrimonious politics due to a need to appeal to more people. A few states have adopted it, but it's momentum seems stalled or it is actively opposed in others. I think it might be a worthwhile cause, but I am open to the contrary.
Ranked choice voting is problematic because many, many people just can't understand how the vote counting works. That creates avenues for bad faith actors to make bogus, but seemingly plausible (to those who don't understand) claims of election fraud.
Approval voting is absolutely worth advocating for.
If you call for ranked-choice voting, evidence shows that politicians will give you instant run-off voting, which isn't real voting reform (still favors a two-party system) and doesn't fix anything or help at all.
Do not take the results of states using IRV as evidence for the efficacy of voting reform. It's not voting reform, it's a scam to defuse the momentum of the voting reform movement.
More generally: there may be a benefit to electing more centrist candidates given how insanely polarized and pendulum-swingy the country has gotten. But even more important than that, it's actually possible for third parties to be relevant and win elections under Approval voting.
Single member constituencies are a bigger problem than voting methods for the US. They force a winner takes all system where up to 49% of the electorate is disenfranchised. If you had 3-member constituencies, with any reasonable voting mechanism, you would likely end up with one Democrat and one Republican in the majority of states, with a third seat up for grabs, which could conceivably be captured by the Greens, or independents or whatever third way appeals to the electorate. There's probably no state in the US which would consistently produce straight party tickets.
This is important because it gives voters additional choice, and it forces politicians from different parties to work together on behalf of their constituents from time to time. Allow parties to run multiple candidates also reduces the races to out-polarise your competition to appeal to tiny numbers of primary voters.
Neat idea! If somehow *magic!* the US switched to multi member constituencies, would RCV or FPTP make for better proportional representation? Eg - do you have a preferred voting system for, say a 3-member constituency?
Ranked choice works better. I'm Irish, and we use the single transferrable ballot: say there are 3 seats, 15 candidates and 100,000 votes cast in my constituency (including maybe more than one representative from the larger parties, and a handful of independents that everyone knows have no real chance). I rank the candidates in order of my preference; I stop once I finish ranking candidates I approve of - say I rank 8 candidates.
Once a candidate reaches 2,501 votes, they are elected, as at most three people can reach this many votes. In Ireland this very rarely happens on the first count. The candidate with the fewest votes is eliminated - if their votes have second preferences they're distributed with the same weight as first preferences. The process repeats until one candidate is elected. A random sample of the surplus is distributed if they end up with more than 2501 votes. (This means that recounts produce slightly different results sometimes - with computers you could actually deem the candidate elected, reweight all the votes for the candidate by the size of the surplus and then continue. It's important that votes stick to candidates so 51% of the voters can't gang up to choose all candidates.) Eliminations continue until another candidate reaches the threshold. The process concludes with either the third candidate reaching 2501 votes, or as often happens if not all voters rank all candidates, the one with the most votes is deemed elected without reaching the quota. In practice you end up with 75-100% of the electors feeling that they have someone elected on their behalf.
For me as a voter, I can distribute my first few preferences amongst causes I care about without worrying that I'll be left without representation if that person doesn't win. Then I can rank the main party candidates in the order I choose - frequently I'll have two candidates from each of the parties to choose from, so I can signal to the party that I prefer centrist candidates or financially conservative candidates or whatever it might be. Almost always, my vote 'sticks' somewhere, though often I'm not entirely sure where (e.g. it may or may not be redistributed in a surplus).
Honestly, the details of the transfer system are much less important than the idea that multiple seats and multiple candidates present voters with more options than a binary choice, which in most races in the US is actually no choice at all. I think many other commentators will be stuck on the idea of single member constituencies - in which case ranked choice doesn't do a whole lot.
Ranked-choice absolutely not; it's a mathematical travesty (*any* ranked-choice method is) and has vastly more of a lobbying apparatus than it deserves already.
Approval absolutely yes; Center for Election Science is some kind of EA-connected org if I recall correctly but they're promoting it. They had a great success in Fargo which the state legislature took it upon itself to tyrannically preempt; it's an uphill battle that could use help.
1. Yes, some additional information (the rank splitting approved from disapproved candidates) is needed to perform it, but it's still just a transform that discards information.
2. "Simple vs. complicated" is a process question, which is outside the scope I intended; perhaps I should have explicitly stated "ceteris paribus".
1. I don't believe that is how the term 'lossy transformation' is typically used. Lossy transformation usually means the infromation is strictly less, not 'some information lost, some gained'. If I use the term the way you are, I could say that RCV and Approval are both lossy transformations of each other.
It is just plainly impossible to 'transform' an RCV ballot into an approval ballot without extra information.
You can make a variant of rcv into strictly more informative than approval voting if you introduce a zero point where you only approve of people above that point. I agree that not having that zero point means rcv loses the whole point of approval voting.
>"You can't "reconstruct" the approval outcome from the ranks. They're just different."
Approval is a partial ordering; ranking is total ordering. The missing piece (i.e., above what rank in the total ordering to approve) to reconstruct is IMO one of weakest aspects of approval.
Personally/cynically, I expect enough voters in an approval system to bullet vote (approve only their first choice) that it's functionally just FPTP with tolerance for overvotes.
It is totally valid to argue if humans are at all capable of assigning intelligent cardinal values to their preferences, or if they are more suited to ranking.
But an 'IMO' shouldn't enter into the use of mathematical terms to describe the systems.
Sure, we can disagree about whether the extra information provided by approval voting is better than the extra information provided by ranked choice (and also if there is a practical benefit to simplicity in a system involving lots of uneducated people and very contentious outcomes).
But it is simply not the case that approval voting is "just ranked choice with less information", they're fundamentally different things and you can't derive one from the other. Neither dominates, they have different properties.
I am afraid that in such situation, whatever sophisticated method you use to choose the winner, about 2/3 of the population will be unhappy about it, with 1/3 deeply unhappy. From that perspective, every voting method sucks; that's not a very convincing reason to choose one over another.
I'd prefer if we instead discussed scenarios that have potential better and worse outcomes, and which method leads where.
EDIT: Oops, I have misread the letters... I though they were cyclical (ABC, BCA, CAB). Sorry for that!
This could hardly be a more wrong statement. Cardinal and ordinal voting methods have very little in common. Kenneth Arrow's much-vaunted theorem is only a result on ordinal methods, for example. (The analogous general result is the much weaker theorem of Gibbard (*not* Gibbard-Satterthwaite) which is merely "strategic voting based on beliefs about other ballots may exist".) I will not provide a dissertation in this thread but look at rangevoting.org for a pretty wide collection of materials on the subject.
The range voting site has a number of dubious arguments.
It notes that Condorcet methods can reward strategic voting in elections where no Condorcet winner exists (or more precisely, no Condorcet winner would exist if voters voted honestly). I think that the odds of an election where there is no Condorcet winner are too low to worry about that possibility.
It makes the point that in range voting, when strategic voting is rewarded, the voter can do it by giving two candidates equal scores. Suppose a voter’s preferences are A > B > C, but there is a no Condorcet winner and the voter has to falsely claim not to prefer A over B in order to maximize the chances that the winner will be B (the voter’s second choice) rather than C (the voter’s third choice). With a ranked choice system, the voter would have to rank B ahead of A. With range voting, the voter can range A and B equally, which is a falsification of the voter’s preferences, but arguably a smaller one. Personally, I don’t think there’s a meaningful difference.
In a section titled “Range voting encourages honesty,” they write: “Experimental fact: In the USA's 2004 presidential election, about 3/4 of range voters (in a range voting exit poll of random voters) chose to vote in a style which did not award the max (99), min (0), or X (intentional blank) score to every candidate. In other words the fraction of range voters who choose to sacrifice some strategic oomph in order to be more honest, experimentally is enormous.”
First of all, an exit poll is not a vote; people can answer a poll honestly without changing the outcome of an election. Secondly, if you want to encourage honest voting, you should use a system that rarely rewards dishonest voting.
The existence of real-world Condorcet cycles is hard to get empirical data on at all, so modeling has to fill in a lot of the gap, and it's easy to see it occur with fairly plausible models (putting candidates and voters' preferences in some R^n vector space and measuring preference by distance minimization; one can perhaps get fancier to model e.g. a single-issue tax-obsessed voter or something). But it's true that most ranked-choice elections in the real world are IRV which is specifically extra bad compared to a Condorcet method; there's non-monotonic real elections to be found easily there but that doesn't condemn Condorcet methods.
I'm not sure your argument in graf 3 on holds; or rather, I think it's a bit misaimed. This is going to be a situation where you're voting A=100 (no impact on pairwise B-C but you really prefer A), B=100; the more impactful strategic falsification in cardinal vs. ordinal is clamping to those 100s and C=0 when your real preferences might be A=95, B=90, C=50 or something, i.e., degenerating to approval. The useful property of cardinal is rather that nothing you do to your A vote affects the pairwise B-C result. Approval forces such clamping (approve A, B, disapprove C) which is why the editorial voice there is down on it; I'm not, myself. But their argument for it is that you can make more honest *honest* ballots with range than approval; as for ordinal, your preference of A > B > C is really A ≥ B ≥ C and there's no way to express either indifference or preference magnitude in the ordinal ballot.
So given that strategic range is degenerating to approval, what's strategic approval? Moving the threshold at which you're willing to approve; e.g. your Naderite liberal voting Nader, Gore, sacrificing their impact in the pairwise Nader-Gore contest to have an impact in the pairwise Gore-Bush contest. If preferences are really one-dimensional and Nader is somehow "left" of Gore who's "left" of Bush you might say there's no Condorcet cycle and N > G > B is simply a more expressive and still strategic vote impacting both pairwise contests (and forcing a vote in the N-B contest by transitivity, but I think individual transitive preferences are a fine assumption, under ≥ anyway).
I'm not so sure this is the case, nor am I sold on range over approval, but I don't think this makes the arguments dubious.
(And, to be clear, I do find the possibility of cycles unacceptably odious to have at all - that's my editorial stance here. I think there's good reason to believe they arise but it's true the empirical results are thin.)
Basically: it's still good and probably won't decrease polarization can still lead to center squeeze by design, but it's already branded and popularly known and it's the best choice we have for reform right now.
1. Who is selling the chips? The same companies doing the AI investments or different ones?
2. What is the bottleneck on the US side? Chips or dollars? If chips are the bottleneck and every chip available on the US market already has a potential US buyer who has money to buy it and somewhere to plug it in, then more dollars doesn't get more US datacenters online in the short-to-medium term. But if the bottleneck is buyers able to afford all the chips, or resources to build datacenters and power plants, then more dollars might help.
In the long term, selling chips to China probably allows US manufacturers to build more chip fabs, but that has a very long lead time.
3. If dollars are a bottleneck on the US side, are they a bigger bottleneck than chips are on the Chinese side? I.e. does one more chip going to China speed up the Chinese research more than the profit from selling that chip speeds up American research?
Agreed, and it really does seem like chips are more of a bottleneck. Argument:
It's not really money vs chips. The "money" needed for AI development is needed for actual concrete expenses. Among the expenses, my sense is that the chips are sold at the highest markups, by a long shot. Therefore chips are likely to be the most bottlenecked among these expenses.
If Nvidia sells chips to China, the extra profit doesn't go to other AI investment. It goes back into Nvidia to build more and better chips, which presumably would be made eventually in any case. If some of that money instead went back to shareholders (I don't know if Nvidia is paying significant dividends right now, but I doubt it), they could use it in any number of ways, and I wouldn't expect a significant amount to go to non-Nvidia AI investment.
Another argument: come on! (Exasperation not directed at you.) Chips are extremely precious right now, while Google and Amazon clearly has lots of money to throw at AI (which is mostly going to compute anyway).
Note: I'm happy to hear from someone with specific expertise in the economics of this, but in the absence of some surprising dynamic I'm not aware of, the alternative seems unlikely.
Does the implicit Jensen Huang view have more merit than credited to it? Like isn't it likely correct that industry connections are durable but hard to reconstitute? E.g. so long as innovation continues to be exponential and doesn't upset the playing field, then current firms have a chance to continue to dominate. And is it at least reasonable that strong China exclusion policy has a tradeoff in standard setting, and standard setting through tech stack dominance leads to greater strategic control?
The issue is that China is *already* trying maximally hard to develop its own internal chip industry. Giving them advanced foreign chips doesn't change that. All it's doing is hurting the American AI industry (because there are less chips available for Americans) for absolutely no reason while giving China a leg up.
Does anyone have a nice site/googlesheet/prompt that I can plug my manifold account into and get some calibrations stats and potentially interesting insights?
Claude Opus 4.6 wipes the entire customer database including cloud backups.
I think I finally get it. AI increases productivity according to the economic broken window theory. Smash a window, and the effort required to replace it makes the GDP go up!
Jeez, guys, the bugs are irritating and so's the thought of the companies leaving them there. But the latest Claudes are *astounding* in their abilities. That's just more important.
Claude itself has numerous bugs - the SSO login keeps going in a loop for me every second day. This is the kind of bug that I can fix, I either have to wait or delete cookies, but would be a priority 1 bug in the company I work in. As in all hands on deck, CEO notified. Service level agreement in danger.
That’s because they selling directly to customers and down time isn’t that much of an existential threat. Silicon Valley can move fast and break things, most software service providers can’t.
somehow this is usually contrasted by moving slow and not breaking things. My experience is that most places move slow and still break things. If you are going to break things anyway, one could argue, you might as well go fast.
That used to mean good things. Nowadays it's just about moving fast towards enshittification and rent-seeking at the highest level. Disrupt the economy by undercutting traditional competitors on price if not quality, and when the old ways are marginalized because the evangelists are convincing enough and the beancounters agree, you can start cashing out:
Giving an LLM agent the power to drop production databases without a human in the loop is wild to me.
I understand that claude having admin on your cloud account is useful, but of course it's also risky. If people believe the trade off is worth it, more power to them, but they give up their right to complain the moment they make that decision.
Edit: As I read further in the article, they call guidelines given to claude "safeguards". I cannot :D
Suppose that the Democrats and the Republicans do end up agreeing to at least the principle of a federal ban on gerrymandering.
What might an alternative that was "fair" in at least the weak sense of "seats received as a function of votes received is symmetrical", and preferably in something approximating the strong sense of "seats received are proportional to two party vote share", look like given one rural and one urban party?
Basically, the districts for state house / state senate / and congress must have minimal overlap, which means the legislature can azy most gerrymander 1 of those maps.
Don't reify parties further than they already are.
RCV (especially Condorcet methods) undercut most of the electoral utility of parties.
As for anti-gerrymandering, I would start with a requirement that the shortest route (N.B., *not* straight line!) between any two locations in a given district must not cross the corresponding routes for any other district.
Condorcet methods of tallying ranked ballots identify the candidate that would defeat each rival in a two-way race, with some minor variations about how to resolve cycles (e.g., rock-paper-scissors).
The vast majority of active ranked choice voting (RCV) systems tally using instant runoff (IRV) or single-transferable vote (STV), which sequentially eliminates the candidate with the fewest votes and reallocates those votes to the next highest remaining choice until someone has a majority.
IRV tends toward two-party dynamics whereas Condorcet does not.
Simplistic example: three candidates, A, B, & C; aligned in that order along the left-right axis. A & C each have 49% of the electorate perfectly aligned while B only has 2%, but A & C supporters would each prefer B to win rather than the other major candidate. In IRV/STV, B is eliminated first, and the choice between extremes is down to which is more palatable to the center 2%. Under a Condorcet system, B would win since he would beat both A & C 51-49 in two-way races despite not having either of two major parties' support.
Ok. The sentence sounds misleading then. Should just say 'Condorcet methods undercut most of the electoral utility...'
The way it is phrased, it makes it sound like RCV undercuts it, but a Condorcet RCV method undercuts it more ('especially RCV').
This is misleading since, as you say, Condorcet RCV is a non-central example of RCV (most popular RCV methods do the opposite of what the sentence is implying). Also would any Condorcet method undercut the party utility, or is it only RCV Condorcet methods? If so, the line should just say 'Condorcet', and drop the 'RCV' part.
All Condorcet methods are RCV, but since they're less well-known I find it worthwhile to specify both terms; i.e., omitting "RCV" makes it seem more esoteric, whereas saying only "RCV" makes people think of IRV.
And IRV *does* undercut party dynamics relative to FPTP, just nowhere near as much as Condorcet.
Also, while yes Condorcet methods are a from of ranked voting, IRV proponents have unfortunately succesfully rebranded IRV as RCV, so in most discourse RCV=IRV.
'Condorcet' may make it seem esoteric, but saying 'RCV <anything>' will make people think of IRV. I dont think adding a parenthetical will make most people not think of IRV.
Other countries do it by commissions with independent trusted people (judges are popular), and rules like "follow administrative boundaries when possible" that minimizes the degrees of freedom of the line drawers.
This approach mostly aims at creating districts without considering electoral effects, not engineer the results to conform to some concept of fairness.
In 2026 I think computer algorithms could pretty easily be designed that did that in a verifiably impartial way that would be acceptable to both parties.
I think the best solution is some kind of multi-member districts. Right now, single-member districts are mandated by federal law. A lot of the incentives to gerrymander are downstream from that. Change to 3-5 member districts with some kind of proportional-ish voting system (STV, party list proportional, Quota Borda, Limited Voting, etc) and the incentive to gerrymander becomes much smaller.
Proportional representation at the state level would probably make sense. Usually I don't like proportional representation on the basis that people can't realistically make contact with "their" local member, but the US congress already has >500K people per congressman anyway so that's out the window.
I also don't like proportional representation because it makes harder for independent members, but that's also out the window in the US system. Proportional representation would presumably at least allow some third-party candidates to win seats in larger states, which would be an improvement in theory if not practice.
I agree with you on the disadvantages, as far as that goes. I also don't like how much Party List Proportional institutionalizes voting for people rather than parties and makes it so people near the top of the list are accountable to the party (for their position on the list) but insulated from general election voters.
That's why I favor 3-5 member districts rather than statewide or national proportional, and also why I favor other voting systems other than party-list proportional that give general election voters direct input on who is elected from each party and which allow for independent candidates.
STV and Quota Borda are extensions of Instant Runoff and Borda Count respectively that produce roughly proportional results if there's a strong partisan clustering of ballots. They've got similar advantages and disadvantages to their single-member counterparts. Both are less vulnerable to spoiler effects than Plurality voting (*) and are decent but not perfect at choosing Condorcet winners. Borda is better at choosing Condorcet winners and has a much simpler and more legible counting procedure, but Borda is vulnerable to "cloning" when ballot access rules permits it while STV is not.
Limited Voting is a similar extension of Single-Member Plurality voting. The standard non-proportional Multi-Member Plurality method (used in the US for a lot of city council and school board elections where there are a small number of at-large seats) has you vote for your favorite N candidates for N open seats and the N candidates who get the most total votes are elected. So if there's a slate of candidate supported by a clear plurality of voters, that entire slate is very likely to win all the seats. Limited Voting has you vote for some number less than N, so a slate needs a significant supermajority in order to win all the seats even with perfect coordination. LV is considered a "semi-proportional" method in that it biases results somewhat towards the majority party in the district compared to true proportional methods like STV, QB, or Party List, but still has an element of proportionality compared to regular MMP.
(*) Often called "First Past the Post" for some bizarre reason.
> Usually I don't like proportional representation on the basis that people can't realistically make contact with "their" local member, but the US congress already has >500K people per congressman anyway so that's out the window.
Neither of those is true for STV, quite the opposite. I don’t like STV, precisely because it creates too much horse trading after the elections.
A few, but not many. There's currently one "Independent Republican" in the House (Kevin Kiley) who was elected as a Republican in 2022 and 2024 but is running as an independent in this year's election. In the Senate, there are two nominal independents, Bernie Sanders and Angus King, but both are strongly aligned with the Democratic party. Sanders is an independent in name only, having run for President in the Democratic Party primaries in 2016 and 2020 and always being cross-nominated by the Vermont Democratic Party in his Senate runs. King originally won his Senate seat in a contested three-way race in 2012, and candidates running as Democrats have challenged him for reelection in 2018 and 2024, but most of the Maine Democratic Party establishment backed King over the official Democratic nominee in all three elections.
All 50 current US State Governors are Republicans or Democrats. It looks like there have been about eight total since before WW2. There's a regular smattering of independent and third party candidates who get elected to various local offices.
Proportional systems may not have gerrymandering as an issue, but they otherwise seem worse to me. Karl Popper was right that first-past-the-post lets voters vote the incumbents out of office, while in proportional systems that will tend to be contingent on coalitional haggling.
Karl popper wasn’t wrong, provided you have a two party system with not that great an ideological difference between them. So where he lived and when he lived. Popper’s “hire and fire” idea really depends on a fairly clean two-party contest. Once you move to three parties of similar size, FPTP starts to behave badly.
FPTP tends to discourage a system of three parties of similar sizes from emerging (regional parties outside the main two could be strong at a local level, but I don't think they typically wind up with nearly as many total seats as any of the nationally viable parties).
Yeah, I've observed in the past that the stable two-party system in the US seems to be a product of the combination of single-member plurality voting system, a Presidential-style government that shapes national politics into pro/anti administration factions and limits the upside of parties that don't have a plausible medium-term path to the White House, and a long tradition (partially inherited from Britain) of a two-party political culture. The UK and Canada share the first factor and partially share the third, but don't share the second, so they tend towards a 2.5-ish party system that has more room for national semi-major parties like the Lib Dems and the New Democrats and regional parties like the SNP and Bloc Quebecois and has potential for a second-tier party to trade places with one of the two major parties.
This is actually an incredibly complex and difficult question. Focusing you thoughts on 'fair' ways to make shapes on a map will miss pretty much all the important features of the question.
For example, it's extremely difficult to get results that are proportionally 'fair' while also selecting representatives for individual geographies. Imagine a state that gets 20 representatives and has 30% Democratic voters. How many Democratic representatives should that state elect?
If you said 6, I agree with you... but that's not what will happen if you try to use a simple topographical rule to draw shapes on the map in a 'fair' way. If you don't specifically choose the shapes to create 6 majority-Democratic districts, it's very likely that you will end up with like 0-2 Democratic representatives. The 30% will have almost no representation and no one to advocate their interests, which doesn't seem very fair.
Of course, you could look for natural clusters where that 30% state-wide is concentrated in much higher numbers, and draw 6 shapes around those clusters, such that most of that 30% is represented by someone they agree with. But now you're drawing weird shapes specifically to get the electoral outcome you want... that's just gerrymandering again! Or, if you want to split hairs and say that's not *technically* gerrymandering because it's done for good reasons instead of corrupt ones, it still *looks* like gerrymandering to the average voter, and it allows for a type of special pleading in district drawing which future legislators can easily use to do the corrupt type of gerrymandering in the future.
But lets say you decide not to care about proportional representation at that level, you are fine with 30% of the populace getting 0 representation if that's how the lines end up being drawn by your perfectly 'fair', impartial, logical topographical algorithm.
Well, someone else who also openly cares about logical and 'fair' topographical algorithms is going to say yours is stupid, and they should use his instead.
Maybe you think a grid of simple shapes weighted to equal population with zero concern for anything else is the fairest method because it is least arbitrary and eliminates the weird squiggly long districts everyone makes fun of, maybe the next person thinks the simplicity of the shapes is a red herring and what matters is the mean squared distance of each residents home from the center of their district, and you can draw districts which minimize this function in whatever shape that creates. Maybe someone else comes along and says you're both idiots, many geographic regions are already unified based on existing utilities infrastructure and police precincts and school departments and etc. in ways that can't be arbitrarily shifted by decree, and trying to split representative districts across those will create a nightmare of ungovernability that is more important than your concerns about 'fair shapes'. Maybe someone else says etc. etc., until you have 50 proposals.
You are now facing one small problem and one big problem.
The small problem is that out of these 50 proposals, 40 of them are actually *good*. They are all 'fair' in different ways, there are concerns that each of them addresses or ignores, some are easily implementable immediately with minimal disruption or cost, some of them promise to be more long-term beneficial after an uncomfortable and expensive adjustment period. This means there is no simple way to settle arguments between the honest, fair-minded advocates for each, and they will keep fighting about it forever, and will keep fighting to repeal the current system and move to theirs even after it's been implemented for years, and that fighting will mean people keep thinking the system is unfair and keep complaining about it and being discontent about it.
But all of that is the small problem.
The *big* problem is that each of those 50 proposals will tend to elect a different slate of representatives, and special interest groups, think tanks, and the major parties are entirely capable of modeling each of those proposals, and figuring out which ones happen to favor them the most.
Now all the money and influence in this argument won't be coming from honest-minded technocrats arguing for what they believe is 'fairest'. It will come from big interests and the two parties arguing that whatever favors them the most is 'fair'. And the winner will probably be whichever party controls the state legislature at the time, plus whatever special interests donate the most to those state senators.
So now you *still* get a districting method that was chosen to favor one party as much as possible, just like under gerrymandering. But where gerrymandering leaves these footprints of iterative redistricting records and crazy-looking lines on a map, this is a completely 'fair' method that some experts designed to be 'fair' and many more experts agree is 'fair', even if it's not the 'fair' method *they* were thinking off. That bias gets locked in in a way that's much harder to notice and call out in the future, and future reform efforts are stymied because the momentum of the moment when people actually cared enough about this to change things was wasted on a system that didn't really solve the problem.
So, yeah.
This is difficult.
If you think it's not, you may just not be appreciating the surprising amount of detail in this corner of reality.
I'll also point out that we've seen an analogy to a part of this in voting reform. A lot of people are aware that first-past-the-post voting massively favors a two-party system, in a way that lets those two parties be corrupt and lazy because they know they'll be in power about 50% of the time just by default. They hate this! They call for voting reform!
A few nerds know that the best voting methods (Borda, Condorcet, etc) all use ranked ballots. Their knowledge filters down to the public, and they cry out for ranked choice voting! It's the only way to break the two-party duopoly.
In quite a few places, the two parties have said 'Fine, here you go! From now on, we will issue ranked choice ballots! (and we will resolve them using this method called instant-run-off voting)'
Guess what? Ranked ballots with using Borda or Condorcet or etc. to resolve them elect representatives very close to the center or public opinion and break up two-party duopolies.
But IRV doesn't. IRV throws away most of the ranking info, and it produces results almost identical to FPTP, with a massive, almost inevitable favoring of two-party duopolies. They changed the ballot, but they didn't change the result.
This is how the fervor for electoral reform has been blunted across many Western nations in recent decades - give the people the reform they are crying for (ranked ballots), do obscure technical things below most people's level of expertise so that things still resolve the same way they were before, say 'there, you got what you wanted, sorry it seems like it doesn't do much, guess that wasn't the problem.' The movement is largely disbanded since they 'won', the public sentiment is turned against the project as it seems to have failed to change anything, and business continues as usual.
This is what will happen to redistricting if we allow it too... and probably what will happen even if we do everything in our power to resist it.
"The *big* problem is that each of those 50 proposals will tend to elect a different slate of representatives, and special interest groups, think tanks, and the major parties are entirely capable of modeling each of those proposals, and figuring out which ones happen to favor them the most."
Thank you for articulating this. It's my biggest concern, too.
Whenever one or another form of ranked choice voting is discussed among my friends, they seem to actively ignore many of the electoral ramifactions and be weirdly uninterested in the fact that those promoting the idea to them are not. This even though I usually feel like the quokka of the group otherwise.
Something like "the average straight-line distance between each person's residence and the geographic center of that person’s district shall be the minimum known to be possible at the time prescribed by state or federal law for the determination of districts" will do it.
Some problems can't be solved if we require geographic districts, though. Something like 1/3 of voters in Massachusetts vote R each year. But because of how the Rs are distributed, there is literally no way to draw districts so that Rs win even a single House seat in Massachusetts. I would argue that in an ideal system, Rs would have about 3/9 of the House seats.
I think you could squeeze one seat. The current districts 1 and 2 (central/western) are a little gerrymandered. Between the two of them, if you look at municipalities that favored Trump in 2024, it's about +18 000 votes out of a total of 222 000 votes cast, ignoring 3rd party (average of 367 000 votes per district). And Districts 4 and 9 (SE Mass + Cape) was +17000 out of 200 000, so you could combine them and add in slightly Democratic connecting towns to create a Republican district. Each of the districts mentioned above is slightly gerrymandered. The South Shore and Cape is strongly D and that overcomes slightly R greater Fall River, which is split between 4 and 9.
Also, a reasonable R candidate would poll much higher, especially with a weak D candidate. Charlie Baker won the governor seat with 60% in 2018.
Two exciting recent innovations on pancreatic cancer!
First, the personalized mRNA vaccine autogene cevumeran, developed by BioNTech and Genentech, just reported 6-year follow-up results from their Phase 1 clinical trial. 16 patients were treated, 8 were responders (showed signs of immune reaction to vaccine), 8 were non-responders.
7/8 responders (87.5%) survived 6 years after surgery, 2/8 nonresponders survived (25%).
The most important result in this small trial is that vaccine response is strongly correlated with better outcomes. But for context, the trial was restricted to patients with operable pancreatic cancer. Patients diagnosed with stage 1 or 2 pancreatic cancer have a 5-year survival rate of 12%. Patients who get their pancreatic cancer surgically removed have a 5-year post-surgery survival rate of 20%. This makes the overall 6-year post-surgery survival rate of 56% among the 16 trial patients pretty impressive. Keep in mind that the trial patients may have been healthier than average for other reasons, and small n is small n, so we shouldn't be too hasty until we see Phase 2 and 3 data.
Second, a small molecule drug I almost missed in the mRNA hype but arguably even cooler, the tri-complex ras inhibitor (!!!) daraxonrasib, developed by Revolution Medicines. Ras is a protein involved in many cancers, with PDAC (the most common pancreatic cancer) being especially dependent on ras, but it has historically been considered impossible to target due to its chemical properties. Daraxonrasib is, as far as I'm aware, the first drug to target generic forms of ras. It does so with an exotic "tri-complex" strategy involving gluing a different protein, cyclophilin A, to ras in order to disable it. Crazy stuff!
Phase 3 results found that daraxonrasib doubled survival time among patients with metastatic pancreatic cancer, 6.7 months to 13.2 months (p < 0.0001). Side effects are kind of nasty, but "well tolerated, with a manageable safety profile" by advanced cancer standards.
Many of the incredible advancements we've made in oncology has been won slowly through another pathway we can target, another drug buying a few more months, another targeted technique to mitigate side effects. Progress is made one step at a time, and these are big steps. I'm excited for the future of biotech.
FWIW my wife is a cancer researcher, she’s quite excited about the revolution medicine drug, hadn’t heard about the mRNA vaccine. I was looking for info in the link you had but didn’t find anything there…
The AI angle, since many people are hyping the "AI for pancreatic cancer" line:
Yes, deep learning is used in the development of the mRNA vaccine. NetMHCpan, a small neural network, is used to help select immunogenic neoantigens (choose the most promising mutant proteins to target) for autogene cevumeran.
Most of the computational pipeline detailed in the paper consists of traditional tools, in line with what I've written about AI and personalized mRNA vaccines before. Deep learning is extremely useful for some things, but it should be understood as a specialized tool, not a silver bullet, for now.
I'm not sure if deep learning was used in the development of daraxonrasib. A brief glance at the paper and previous work shows plenty of references to traditional computational tools, but nothing that stands out to me as modern DL.
The company behind daraxonrasib, Revolution Medicines, is quite enthusiastic about ML. They recently made a deal with AI drug discovery platform Iambic Therapeutics. I don't doubt Iambic's tools will soon prove useful, but it's fair to say "deep learning for drug discovery" is still in the early stages of development.
Airlines cancel routes all the time for many reasons. Nonetheless, I believe that Delta’s recent cancellation of its New York JFK - Brussels daily nonstop is a “canary in the coal mine” warning that New York’s standing in the national and world economies is starting to decline. Maybe it’s all the financial businesses decamping to Miami, I don’t know.
Why I believe this is no ordinary route cancellation.
- This isn’t some new route that didn’t live up to initial expectations. Delta has operated the service all the way since 1991, through all sorts of economic ups and downs, until now. In fact it was one of the European routes Delta had inherited from Pan Am.
- Brussels, indeed all of Belgium, is not really a tourist destination, so the cancellation cannot be blamed on shifting tastes in tourism. But it *IS* a major business destination. A decline in business travel is a much more likely sign of economic decline.
- Delta has not pulled out of Brussels completely, with ongoing service from Atlanta. In fact the company’s only explanation of the cancellation was some corporate babble about serving its customers better from Atlanta. It’s unlike Delta’s recent cancellation of JFK - Geneva service as it no longer serves Geneva at all, indeed it’s rather small to merit nonstop service from the US.
- Speaking of Atlanta, it’s Delta’s major hub, but it’s *always * been the major hub. Atlanta’s hub status did not stop Delta from serving Brussels from New York for nearly 35 years.
- No other airline has chosen to replace Delta. The only New York service is a single daily nonstop on Brussels Airlines; while United flies there from Newark, whether Newark qualifies as a New York airport is an unanswerable question.
Now, I acknowledge that a decline in importance on the Brussels end could explain this, however I’ve never heard of anything to that effect while New York’s loss of the financial sector to Miami is a very real and well documented thing.
Love reading takes about New York's decline from people who don't live here because they always miss a bunch of context and are always proved wrong in the end.
I'm not saying that the deprioritisation of New York in favour of Miami as a financial hub is not happening, I just think that if it were happening then there'd be a lot of other data you could track it more reliably with, like job listings or real estate ads, or large amounts of floorspace given up by major financial institutions in New York or rented in Miami.
You don't need canaries in coal mines when you have an omnipresent network of carbon monoxide detectors.
>whether Newark qualifies as a New York airport is an unanswerable question
Newark is closer to Times Square than JFK is.
Does your analysis take into account the decline in international travel to the US under Trump? Or the possibility that it is Brussels that is declining in significance?
It may be worth noting that Delta is starting three new European routes from New York next month, summer-only so they don’t make up for the loss of Brussels, to Porto, Malta and Sardinia. It is probably a reasonable assumption that they will appeal more to Americans visiting Europe than to Europeans visiting the US and hence in line with the decline in foreign travel to the US.
It's a big inference from a small data point. Delta puts a lot of JFK passengers into London, Frankfurt, Amsterdam and Paris, none of which is very far from Brussels by American standards of distance or European standards of train service. Delta is not in league with Air Brussels/Lufthansa so has no particular desire to help them sell connecting flights. NY gates are scarce and will be allocated to routes that maximize revenue, Brussels is pretty far down the list of European airports, and if you look at the Delta map they are not trying to run direct NY flights to every Euro destination.
If you need a better reason to explain why a carrier would cancel a route they've run for decades, I'd put higher likelihoods on the decline in US relations with the UN and NATO or good old fashioned antitrust boundary-pushing like sending a message to United and Lufthansa that Delta would prefer to divide up markets a bit more neatly.
All of the things you note in the first paragraph are true, and *could* be the reason, but I’m skeptical because they’ve always been true. Yet the route lasted 35 years until now.
Declining relations, like you note in the second paragraph, now that could be a more plausible reason as it would affect business travel much more than tourism.
Newark counts as a New York City airport for all practical purposes.
> A decline in business travel is a much more likely sign of economic decline.
10 years ago, this might have been the case, but it could also be a sign of standard cost-cutting with the rise in business being conducted over zoom.
NY - Geneva being cancelled actually seems worth thinking about, because of the major UN presence in both, plus many other big international orgs in Geneva. Breakdown in diplomacy? Increased isolationism? Or maybe just the same rise in video conferencing.
Yep. From EWR, I can get on a train that takes me directly to Midtown Manhattan.
From JFK I guess I could take a bus to the E train subway that'll get me to Times Square eventually. I'm not sure how I'd do that from LGA. And PATH is much less sketchy than MTA.
The JFK Airtrain goes from the airport to Jamaica station on the Long Islamd Rail Road, from where there are frequent quick trains to Manhattan and Brooklyn, as well as subway connections. LaGuardia Airport has no train connections; there’s talk about building one, but then again New York has been trying and failing to build the urgently needed Second Avenue Subway for the past 100 years so don’t expect one soon.
Look I'm not saying pastrami sandwiches are bad, I'm just saying that if you're the biggest, richest and most powerful city on Earth and your big contributions to cuisine are "sandwiches with a lot of meat" and "generic pizza" then you should find something else to brag about.
There are lots of things wrong with NY. I lived there for a dozen years and was pretty happy to GTFO.
But criticizing NY because of the food?!?! No, you're very wrong here. Yes, even pizza. NY has Italian, French, or Japanese foods on par with that of the respective countries.
Every pizza store in New York City has a sign that says, "You've tried the rest; now try the best." We know better . By the way I got a really good deal on a bridge and I'd like to share it with you.
Is there a good term for when you're presented with a reducto ad absurdum of your position, and end up agreeing with it? Obviously there's "bite the bullet" and there's the Chad says yes meme, but it would be nice to have something classier and more parallel to the original.
Best I've come up with so far, with some help from Claude, is assensio ad absurdum. Does this actually work or is it just useless dog latin?
"Bite the bullet", to me anyway, means you accept the tenets of a counterargument against your argument, but has a flavor of "but I still think I'm right because..." which you will follow with some (probably novel) argument against that counterargument you just bit the bullet on.
“One philosopher’s modus ponens is another philosopher’s modus tollens” is closely related, though you would have to use it the other way around if confronted with a reductio.
How about “one person’s reductio ad absurdum is another person’s counterintuitive conclusion”?
So you basically hold to Caesar's principle, tamquam scopulum sic fugias inauditum atque insolens verbum? Fair enough.
I feel like in this case it's not so bad since reducto ad absurdum is well known, and we have English words like assent which are clearly derived from the same root, but I do see where you're coming from.
Here to share some stuff which I think ACX readers may enjoy. I recently did a deep dive on an allegedly ancient Tibetan prophecy that has been circling in the American Dharma world since the 70s. Lots of interesting supernaturalism, memetic spread, and citation-ambiguity along the way: https://transgenicpig.substack.com/p/when-the-iron-bird-flies
About "What is mass?" - I like the good self-reflection, though I admit I mostly skimmed over the article. It is not dense enough for me to warrant word-for-word reading like I do on ACX.
On the content: Have you considered the equivalency of mass and energy, aka E=mc² ? That is the modern answer in physics, and it goes a long way.
I do agree that in early semester physics the explanations of mass and weight can be confusing. If you have a good professor though, he will point out multiple times the curious relationship of those and encourage you to keep pondering for the following years. Once you are "through" with the general physics curriculum, re-visiting all of this certainly breeds a new and better understanding of the "classical mechanics" topics (which are all just approximations; to be more explicit, e.g. Newton's physics is just an approximation for certain mass/distance/time ranges of the more general relativistic physics).
Hey thanks for reading -- and yes I totally agree that there is a special kind of joy that can come from revisiting the foundations after one has some familiarity.
RE energy-mass equivalence: I will discuss this, but perhaps not as explicitly as in your approach. To give the game away, Part 2 [https://transgenicpig.substack.com/p/what-is-mass-part-2] (coming out tomorrow btw!) will establish three definitions in a pre-relativity context, but for various reasons these collapse to just one in a relativistic context (Part 3). And yes, we will see that in this context mass, energy, and momentum start to blend together, and it will be increasingly unclear that we gain anything by thinking of them as distinct.
From my approach this seems more like an interesting side-effect of relativity than a starting point in its own right. The tension between our views on mass is probably that I am trying to ground "mass" in experimental operations, even in the context of relativity. Still, if I have a broader thesis than considering mass itself, it is that the meaning of physical concepts lies in their web of interrelations and one can probably slice that up into definitions in many ways.
Scott, if you are reading this, please write a post on your understanding of your to look at the iran war- a rationalist-moral-american take, not a geo political one.
That would be an overly simplistic take. Replacing the regime is the ultimate victory, but there's lots of intermediate goals that also have value. Just degrading their conventional missile stockpile is useful.
Though the big win IMO, is diplomatic. Europe is furious at the USA for causing an economic crisis. But the Middle East is realising that all their attempts to hedge between the USA and Ian, to do a bit of appeasement, has all amounted to nothing. They're also furious at China for helping Iran. (Odd choice by China, it gets more gas from Arab states than Iran). So they're lining up more solidly with the USA.
Qatar expelling Hamas. The UAE leaving OPEC. This is good gains for the USA.
On the topic of post-COVID educational attainment collapse: could it just be the chromebooks? Specifically, the 1:1 device policies in elementary school, instituded to enable remote learning, have stuck around afterwards and are toxic. Putting kids in front of screens for many hours a day is obviously terrible parenting, but we have some sort of blind spot keeping us from noticing that it's terrible teaching
Related: Sweden, Denmark, and Norway have recently pivoted from screens back to physical textbooks after studies "showed" a dip in reading skills, etc. (Studies on this, especially the ones that convinced Sweden et al., use data from PIRLS, similar to NAEP but international.)
Plenty of links available with a quick Google search.
This is my thinking, though TBH I don't know how widespread Chromebook adoption has been.
The fact that the effect is greater at the lower end is a point in favor of absenteeism, since I am pretty sure that is more common among lower achievers.
Agreed, absenteeism is also a good example of "thing that happened in 2020 that did not go back to normal by 2022" -- even post-pandemic, absentee rates are way, way up. I'd note that if you believe this it's a (soft) refutation of the "you don't learn in school" hypothesis forwarded on this blog and elsewhere.
That hypothesis has always struck me as an absurdly large over generalisation from people not being able to remember certain facts after graduation. We know what societies without schools look like! Even societies with mostly primary but not secondary education! (E.g. my grandparents' norms). It's obviously pretty different!
My view is that we probably keep people in education too long now (people should be able to start work at 16) but that's still a lot of school. There's a reason every country that can do it does.
It might also be related to the ACX community selecting for high IQ nerds who would learn whether or not they were in school, while school is still beneficial for normies.
It's sort of like how there are big summer regressions in education dependent on SES because rich parents send their kids to educational summer programs and poor parents don't.
I'm mildly skeptical of this just because even smart, nerdy, curious students can often benefit from structure and guidance. The ways in which the system is (or appears to be) standing in your way or holding you back are going to be easy and obvious, but you don't have access to the counterfactual world in which you had to figure it all out for yourself.
I've taken multiple stints to go back to university as and adult, and several stints trying to build useful skills and learn difficult subject on my own. I make more than zero progress on my own, but I learn *a lot* more in university[1]. I don't want to over-generalize from my own experience (among other reasons, because I have ADHD), but I'd be surprised if this wasn't *a* piece of the puzzle, at least. As with many other abilities, I'd expect there's a wide range of human ability at self-teaching. I'd further expect that such ability has a correlation with IQ, but only a moderate one.
[1] I've also tried a few online courses not connected to a particular program. Unsurprisingly, the effectiveness was in between "go it alone," and "take an actual class." Meanwhile, learning on the job is arguably more effective than either option, but it's heavily restricted in what it can teach.
There definitely can be two problems, possibly even more! Utilitarian me is also deeply empathically worried about absenteeism increasing from 15% to 30% at the very bottom of the distribution, but for some reason I'm inclined to pitch a fit specifically about issues likely to deal 2-3% score drops to kindergarteners with involved parents, entering the school system in zero to 2 years, (within a five mile radius of my current location.)
The problem is real up and down the board, and I doubt that the 90th percentilers in 2023 had absenteeism issues:
Has anyone written anything about what international development could look like in a post-powerful AI world? There seems to be a decent literature emerging on the economics of AI in the developed world, as well as stuff like AI 2027 which tries to flesh out what the world might look like, but what will things be like in, say, Kenya? Or the DRC?
My assumption would be, assuming we end up in a non-singularity world where economic growth remains a meaningful metric but significantly accelerates, that the OECD is much more able to take advantage of AI than the developing world. This causes the West and developed East Asia to pull even further ahead of the developing world than the already are. The medium-term endgame is that either Africa becomes the actual permanent underclass, or that international aid becomes the vast, vast majority of third world GDP and living standards skyrocket. But I'm not an economist so these intuitions could be completely wrong!
Consider the wisdom of Machiavelli: A prince who is not wise himself can not be wisely counselled.
Suppose an AI were to tell someone the problem is that his relatives expect him to make sinecures for them, and thus he lives in poverty, because his relatives’ labor is crucial in raising the living standards. Is he going to listen?
Intelligence is one thing, good institutions are another. Recently it seems that it is easier to collapse institutions in the developed countries, than to build them elsewhere.
(If someone wants to go full HBD, I recommend considering North Korea and South Korea, which are genetically very close.)
The problem is not that the governments are too stupid to do the good thing for the country, but that sometimes they have no incentive, because they can just steal everything, and then spend a fraction on propaganda. On the side of voters, IQ alone does not imply rationality.
I find that hard to believe, so I'd need to see some good statistical evidence to be convinced. The correlation between national IQ and GDP/capita appears to be around 0.7, with about a doubling of productivity for every 10-point increase in IQ, so if advances in AI do lead to an abundance of intelligence, I would expect to see enormous growth among cognitively disadvantaged nations.
Why would Third World living standards skyrocket? Vast amounts of aid are vast opportunities for graft, and furthermore, graft prevents living standards from rising, thus enabling more applications for aid.
Dean Ball went to a conference and wrote about something obliquely related, basically how developing/global south countries are currently plugging their ears and saying "lalala this AI stuff is all fake!" which does not bode well for development post-AGI. Matt Yglesias' "we may miss the sweatshops" piece is relevant-ish here too
Wow, that Matt Yglesias piece is really good (and I'm flattered that he seems to have had similar ideas to me). I'm particularly interested in this part:
"In an AI-powered world, the resource curse could prove particularly vicious. Obtaining the needed commodities won’t necessarily require human labor, just legal authority to do the extraction. Whoever is in position to hand that legal authority out — the internationally recognized government — will collect all the rents, with no particular need to share them with anyone. Any group that manages to pull off a successful coup will capture vast wealth. Anyone on the outside will be left with nothing. Endless rounds of violence and mass killing could easily be the result."
I disagree because I think that in this scenario, the developed countries will be insanely rich and will still feel some general sense of obligation to poor nations. As a result there will be a push to give them some of the tremendous wealth being produced from the robot-operated mines in their countries. But in this scenario our ability to exert force across the globe will be much greater, so we may feel that we can avoid the issue of rent-seeking by establishing our own technocratic, highly efficient governments to distribute this aid.
"Best" case scenario, this could be something like the Australia situation where Indigenous Australians get given cash in exchange for mining on their traditional land. Worst case scenario, it's literally just liberal imperialism.
Why would you think the resource curse will be limited to poor nations? Every nation will be in the same position: humans are economically and militarily valueless, but consume resources unless exterminated.
>the developed countries will be insanely rich and will still feel some general sense of obligation to poor nations
I want to hope so...but the current state of affairs does not make me too optimistic! In comparison we (developed countries) already *are* insanely rich, and apparently can't even manage to kick over 0.1% of our government spending to poor nations.
To be fair I live in the UK where we historically spent 0.7% of GDP on foreign aid. This has been reduced in recent years because of COVID/the need to increase defense spending, and was also increasingly being spent on housing refugees in Britain. But presumably in a post-AGI world lots of policy issues which require money would be alleviated so people may be more willing to fund foreign aid programmes.
I would love to see some manifold bet attempting to operationalize "will this be cringe."
That said I am sure there is already a tsunami of anti-AI brainslop on short form platforms already, of the "datacenters steal water" variety, so maybe (the logic goes) putting some "good guys on the inside" is the right move?
It's tough, you don't want to cede a massive sector of the info-space, but there is real risk of ending up in "slopulism." Some of the rationale behind this short-form bootcamp seems to be coming out of "beef" with the Pause/StopAI world on X so make of that what you will.
At some point your cool indie band has to sell out in order to get famous. At some point, Ice Cube has to go from "Fuck the Police" to Law & Order SVU and Snoop Dog has to go on Martha Stewart. That's the cost of success; cool underground things sell out in order to make money and have influence.
Like, if p(Doom) is non-trivial and technical alignment is not viable from a research standpoint in realistic timelines, then the only relevant course of action is influencing policy and in modern day, that means developing relationships with streamers who cultivate widespread parasocial relationships with upcoming generations. If L33tRacismBro86 get Twitter famous and then Elon puts him in charge of USAID (1), welp, better start cultivating a relationship with L33tRacismBro87 and Fr33G@z@Nuke1sra3lGirl12 to prevent the world from ending. (2)
(1) For the record, provisional on millions of African children not starving over the next 16 months, L33tRacismBro86 was a substantial improvement over the current administrative state and DOGE was ended far too early.
(2) If you are a young person and all of this sounds retarded to you, welcome to being an old person. I am old now and the only thing more shameful than a bunch of young people being retarded is a bunch of old people pretending to be young and still being retarded.
I know how this goes: soon, Fr33G@z@Nuke1sra3lGirl12 is revealed to be a "TERF," and that damages the cause of every group that cultivated a relationship with her.
The problem is not that influencers are bad at what they do, but rather that what they're good at is posting cool videos that make their followers click "Like" and "Subscribe". So if all you're looking for in your movement -- be it AI safety or pesticide-free farming or whatever -- is lots of likes, then influencers are a way to go. Ok, that is not entirely true; another things influencers are good at is hawking cheap dodgy goods, the cheaper and dodgier the better... but again, if I were part of the AI-safety movement, I am not sure if this is the direction I'd want the movement to go.
1. Ukraine got their €90 billion loan from the EU now that Orbán isn’t around to hold it up. Repayment is designed to occur only if and when Russia pays war reparations. Critics questioned why they didn’t confiscate the frozen assets of the Russian oligarchs and give that money to Ukraine, but the consensus is that the EU is trying to hold out a carrot to the oligarchs. They’ll get their money back if the war ends, and the hope is that in the meantime, they’ll be able to put some pressure on Putin. Fat chance with that. Putin has killed enough of them that they know not to cross him. But there are also €210 billion in frozen Russian Central Bank assets that, according to one article (mentioned in passing), would help service the debt. I haven’t seen a good explanation in the MSM about the details. Does anyone have more details on how this deal has been structured and where they're getting the money to finance it?
2. Zelenskyy signed an agreement with Azerbaijan for trade and military aid. Azerbaijan will send them “energy assistance”, which, after some digging, includes high-voltage cables, transformers, and generators to replace those lost in Russian attacks. I didn’t realize this, but Azerbaijan maintained ties with Ukraine throughout the Special Military Operation, and Zelenskyy thanked President Aliyev for their previous humanitarian aid and energy support. Azerbaijan gains access to European business opportunities through Ukraine's partnerships with European countries that Zelenskyy has been busily developing. And don't forget the drones.
3. Zelenskyy also hopped over to Syria to ink a deal with President al-Sharaa. Syria will send phosphate to Ukraine, and Ukraine will send wheat to Syria (there are still food shortages in Syria). Simple barter trade. Remember also that Ukraine sent in advisors and drones to assist al-Sharaa and the HTS in overthrowing Bashar al-Assad. And I think they gave Ukraine a bunch of captured Russian equipment and ammunition in exchange. And don't forget the drones.
4. Ukraine will also be sending marine drones to the Saudi’s and the UAE to help them break the Iranian blockade. Details of how this would work weren’t specified, but Ukrainian marine drones have been configured as gun platforms, missile launchers, and drone-interceptor launchers, and they have the speed of Iranian gunboats. Some models can operate for 48 hours without refueling and have a 600 km range. It would be funny if little Ukraine ultimately breaks the Iranian blockade. Coincidentally—or not—Iran is dumping more mines into the Strait.
5. I saw an estimate that Ukraine's defense deals are going to bring $40 billion into Ukraine's economy. I haven't seen a breakdown of which country is paying how much and how the payments will be structured. Still it's better than a poke in the eye.
6. In Russia, Defense Minister Andrei Belousov and First Deputy Chief of Staff Sergei Kiriyenko are in a struggle to control the Russian Duma and the patronage that goes with it. Belousov is actively challenging the Kremlin's (i.e., Kiriyenko’s) planned list of Duma candidates for the September 2026 elections. He wants to install his own personnel rather than the "Leaders of Russia" program appointees pushed by the Presidential Administration. These, we are told, were “veterans of the special military operation,” but mostly served in reserve battalions far from the front. Belousov rejected the initial list of candidates, calling Kiriyenko’s candidates inexperienced political hacks rather than true combat participants. Belousov happens to be deeply religious. Some have described him as fanatical in this faith, and he reportedly sees the war with Ukraine as a religious war. We’ll see how this shakes out, but I doubt if Saint Andrew the Apostle (the patron saint of Russia) can do much to stop Ukraine’s long-range drones.
7. Russia claims to have lost 5,937 soldiers killed in Ukraine since the start of the Special Military Operation. That’s all. Meanwhile, Putin claims that 1.5 million Ukrainian soldiers have been killed. The official Russian death toll has not increased since 2023, when it was first published. Now that the Special Military Operation has gone on longer than the Great Patriotic War, it’s not opposition groups that are shining a light on Russian losses, but it’s far-right Russian milbloggers who are starting to suspect there may be something fishy in those numbers. The neo-Nazi Rusich Group’s Telegram channel laid out the rough arithmetic garnered from official sources. According to official data, 540,000 people signed contracts with the Russian military in 2023; 450,000 in 2024; and 400,000 in 2025. That’s just shy of 1.4 million recruits. And there were another 300,000 reserves who were mobilized at the start of the operation, plus the 700,000 who were on active duty in the military at the start of the operation. So, 2.4 million soldiers. But the State has also said that there are currently 900,000 soldiers on active duty at the front. Rusich ran the numbers: 2,400,000-900,000=1,500,000. And they asked, where did 1.5 million soldiers go?
A few days later, Yuri Podolyaka, a pro-war blogger with over 3 million followers on Telegram, estimated that between 315,000 and 415,000 of those 1.5 million inferred casualties have been killed, and he used Putin’s own numbers to support his claim. Putin said that Ukraine’s losses were five times higher than Russia’s, and according to Putin 1.5 million Ukrainians have been killed. Podolyaka concluded, therefore, that this must mean that 300,000 Russians have been killed in the Special Military Operation, which jibes with the lower range of his estimate.
It's interesting that the pro-war Russian bloggers’ estimates are higher than the Ukrainian Ministry of Defense’s estimates of ~1,326,000 casualties. As an independent reality check, at the beginning of 2026, the Center for Strategic and International Studies placed Russian losses at approximately 1.2 million casualties (killed, wounded, and missing) and as many as 325,000 killed since the start of the Special Military Operation in February 2022. And reports indicate that Russia has suffered another 100,000 casualties since the start of 2026.
8. Russian milbloggers are also frustrated with their army’s strategy of assaulting what they call “grandmother villages” instead of going after military targets. Older women frequently become the last residents of ruined villages. A ground robot evacuated a 77-year-old Ukrainian lady from the battle zone in Lyman. She was spotted by Ukrainian drone operators walking along a shelled road. They sent in a ground robot to rescue her, with a sign that said, "Grandma, get in!" The ground robot drove her to safety. Aerial drones also escorted out some other civilians.
9. Russia is having trouble meeting its recruitment numbers. Nobody disputes this. Mainstream Russian propagandists and milbloggers agree that Russia needs more troops and that recruitment drives are coming up short. Zelenskyy ran the numbers on Ukrainian Telegram and said Russia would need to recruit another 10-15% to break even with the numbers they're losing on the frontlines. Anyhew, Russian army officers are conducting video calls with secondary schools all over the country to drum up volunteers. At one particular (unidentified) school, a masked officer came on the call to extoll Russia's elite Rubicon drone corps. He went on about how drones were the cutting edge of warfare, blah, blah, blah. But the call took a turn into the bizarre when he revealed he was a Ukrainian drone officer. He said, "And I want to tell you, God forbid you go there, I will have to kill you," before someone was able to shut down the call.
What I find particularly interesting, beyond it being a great propaganda prank, is that Ukraine was able to learn about the scheduled call arranged by Russian military recruiters and then hijacked it. What does that imply?
"Ukraine got their €90 billion loan from the EU now that Orbán isn’t around to hold it up. Repayment is designed to occur only if and when Russia pays war reparations. "
I always felt Europe is too unselfishly altruistic towards Ukraine, but this takes the cake. Unless there is some deep strategy behind it. The Draghi Report clearly states that the European economy sucks, it requires a big cash investment, he asked for it and they replied that they cannot afford it. Is helping Ukraine truly more important than making the European economy competitive?
The governing mechanisms of the EU are showing the strain - to get anything done requires unanimous approval. I would take the other lesson: Ukraine clearly needed support, in the sense that allowing Russia to win the war would result in costs to the EU probably exceeding 1 trillion to build defences along the east, develop a common defence programme and perhaps a common army. The cost-benefit analysis is clearly on the side of funding Ukraine.
The inability to do much of anything else is a central problem for the next decade: throwing money at crises like Covid and Ukraine doesn't solve the internal problems of the union. It's time to move ahead with the 2-speed EU idea, and stop waiting for the last holdout to fall into line on every issue.
"The governing mechanisms of the EU are showing the strain - to get anything done requires unanimous approval."
Sure, because the original idea was that for example foreign policy should not be common, like how France and Germany opposed the Iraq War and the UK and Poland supported it.
This is why it is a little strange why Orbán is treated as a traitor. It was in the past OK to have your own foreign policy. If I want to be charitable, it is because it is not only about Ukraine but Russia is a security threat to the EU too. Although I do not think they dare to risk a NATO war, I accept that security policy is always based on capability, and not intent.
On the other hand in a lot of other cases the veto system is good and I am afraid they are going to abolish it.
Shareholders in an economic enterprise are allowed to have competing interests; team members or political representatives are not.
The global context changed a lot in the 20 years between Iraq and Ukraine. On the one hand, the EU can no longer pretend that it's purely a trade group - they are a political entity with a common identity. On the other, MAGA-type right wing populists sometimes tell you directly that they are acting in bad faith. Orbán was a bit more sophisticated than Trump, but still clearly acting against the interests of the EU in a way that Germany (almost by definition) could not.
Since 2022, most members of the EU consider Russia their primary geopolitical enemy, some countries on the Eastern border even view them as an existential threat. Now you may or may not agree with this assessment, but given this assumption, a €90 billion loan, secured by the assets of your enemy no less, is a pretty good deal to keep Ukraine in the fight against Russia. Even more so if it keeps Ukraine on your side for after the war, because the best partner against an enemy is the friend who just spent years fighting your enemy, who has a larger military than any single member of the EU, and more experience than all of them combined.
No need to resort to "unselfish altruism" as an explanation.
Oh, I completely agree that the EU should get its shit together, both through more internal investments and aggressive deregulation, and should stop playing cornucopia for the world. My point is that helping Ukraine financially and militarily isn't primarily altruistic.
For the financial construction of the EU, they do not confiscate the frozen assets for the reasons that EngineOfCreation mentioned.
What they do instead (reconstructing it from my memory): they make a loan to the Ukraine, conditioned on a hypothetical payback in case Ukraine gets these frozen assets after the war. It is important for many EU countries (France, ...) that it is a loan because then the money does not count towards the national debt. After all, in the books, they only trade the money against the Ukrainian debt, which is nominally worth the same. It makes a difference because there are all kind of rules that prevent EU countries from making additional debts, both EU rules and national rules.
I'm trying to understand this. Are you saying that France (for example) borrows money to then lend it to Ukraine? If France doesn't borrow the money, why would it count toward the national debt? Also, how would Ukraine "get these frozen assets after the war"? These assets aren't under the control of Ukraine, so France (or whatever) would have to seize them first, which gets us back to "can't do that under the Rule of Law". What am I missing?
Ok, the details are complicated, so I read up on them. In the current deal France doesn't borrow money. In the current deal, the EU commission borrows money from EuroClear (the company located in Belgium who actually holds the Russian assets). Usually, the EU commission is not allowed to borrow large amounts of money. If they want to borrow money (beyond a limited normal budget), then the EU countries would need to provide guarantees. As far as I understand, this is the issue because those guarantees count towards national debts.
Now, the EU countries agreed that the frozen assets in the books of EuroClear are acceptable guarantees. Their official point of view is that at the end of the war, Russia will agree to pay those assets as retribution to Ukraine, meaning that Ukraine will then be able to redirect the money to the EU commission.
There is no problem until Ukraine and Russia make a peace treaty, because the sanctions are permanent in place until such a peace treatment is signed. So until then, EuroClear is not allowed to transfer the assets or money to Russia, the assets stay at EuroClear, and everyone can close their eyes and say that this is fine. I think this is what the EU hopes for, that there won't be a peace treaty for a long time.
If there is a peace treaty where indeed Russia agrees to pay the assets as reparations, then also everything is fine.
If there is a peace treaty where Russia does NOT agree to reparations, then there is a problem. Then EuroClear will legally be obliged to transfer the assets back to Russia, and then the guarantees for the loan disappear. This means that the money is gone then, but at least this is now a problem internal to EU.
One of the most heavily criticized points is that it makes the EU a stakeholder who has an interest AGAINST a peace treaty between Russia and Ukraine, unless it contains reparations form Russia. (Good luck with that!)
It's quite petty, to be honest. I don't think there is a realistic chance that the money ever comes back, and I can't imagine that the EU leaders believ in such a chance either. So the whole purpose of the scheme is to make it look like it's not direct financial aid, although everyone knows that it is.
EDIT: Here is an interview from December explaining the deal:
>Then EuroClear will legally be obliged to transfer the assets back to Russia,
Legally required by whose law? Can that law be changed? Or it means international law? There is really no such thing... those are agreements that can be broken. One can say international law X will be now broken because Russia broke the Budapest Treaty and also Minsk Treaty.
>One of the most heavily criticized points is that it makes the EU a stakeholder who has an interest AGAINST a peace treaty between Russia and Ukraine, unless it contains reparations form Russia. (Good luck with that!)
Then again, it's just money. Any peace deal worth considering will contain robust security guarantees for Ukraine. If that comes together, ok, write off the loan, but at least you get the relative assurance that there will not be a repeat of this whole mess 10 years later.
I still have no idea what such security guarantees should look like. Many thousand US soldiers or EU soldiers or NATO soldiers in Ukraine? Or UN soldiers form other countries? I don't see how a few hundred or even thousand UN soldiers would stop Russia from invading again if they want.
Apart from that, Russia doesn't seem to be inclined to accept US or EU soldiers in Ukraine, so I don't see a peace treaty with such security guarantees. I could at best imagine a ceasefire without formal treaty, or the war simply fizzling out over time.
For starters, that long border with Russia goes both ways. Russia can't just attack the Baltics if they have to worry about the whole Ukrainian border instead of just the Suwalki Gap.
If the US comes to their senses, at least NATO-wise, Ukraine could join NATO. Otherwise, Europe and others who are willing may have to found their own gang. At least France and UK would provide the nuclear capability. In conventional forces, either Ukraine gets the full NATO-equivalent treatment, with permanent bases and forces like in the Baltic states, or at least a permanent tripwire force to justify Article 5 (or equivalent) in case of another invasion.
Either way, Ukraine would be in the alliance, not out of the goodness of everybody's hearts, but because Ukraine has a large army with plenty of experience and motivation; for generations to come they're not going to have any confusion about who their enemy is and what he is capable of.
demost_'s summary was not quite correct. Ukraine has to start repaying the loan if and only if Russia starts paying reparations, precisely because the frozen funds can't be easily used. In all likelihood, the loan will not have to be repaid.
>What I find particularly interesting, beyond it being a great propaganda prank, is that Ukraine was able to learn about the scheduled call arranged by Russian military recruiters and then hijacked it. What does that imply?
Well, it's arranged by the Russian military, but the other party involved is a high school. Getting information from them sounds like literal child's play no? Worst case scenario the Ukranian military would have to hack a Russian high school, that should be well within the capabilities of either Ukraine or some aligned group no?
>Worst case scenario the Ukranian military would have to hack a Russian high school, that should be well within the capabilities of either Ukraine or some aligned group no?
Still, it was organized by the Russian military. They sure didn't want that school to get hacked or their recruitment drive sabotaged. They failed or didn't bother to secure that event, so either way it's on them.
>Critics questioned why they didn’t confiscate the frozen assets of the Russian oligarchs and give that money to Ukraine, but the consensus is that the EU is trying to hold out a carrot to the oligarchs.
The core "problem" is the rule of law. It's still someone else's money, and you'd need a rock-solid law to confiscate it. Without such a law, Belgium, as the holder of the funds, could be sued for damages. In any case it would set a precedent that the EU is less of a safe investment than before, in legal terms. Sure, you could argue "Just don't invade Ukraine and you'll be fine", but it nevertheless opens that door a tiny bit when it was entirely closed before.
What Swiss bank collapsed because the EU leaned on it too hard? Are you referring to credit Suisse? I invite you to make that argument explicitly.
"Stealing other people's frozen funds" is not only something boneheaded people come up with when they have no other way to raise money. For one the EU has other ways to raise money, see the loan in question. And if it's a war situation, this could be a totally valid way to claim assets from the enemy. But as in earlier mentioned posts, the EU decided the rule of law was more important.
Valid indeed does not mean smart. it also does not mean not smart. Brrrrr goes the money machine against a geopolitical adversary, in an active war of defense on the european continent. If it shouldn't go brrrr for this.....
I think it is indeed more accurate to say that credit Suisse collapsed not because of the EU "leaning on it", rather it being caused by a crisis of confidence driven by years of scandals, massive financial losses, and poor risk management. I don't think this is connected to the EU leaning on it.
I can say it for a fact that the pipeline had basically no influence on the election outcome in Hungary. There are several months of reserves and the lights were not at all out and the whole argument boiled down to Orbán saying reopen it first then you get the money, Zelensky saying give the money first then I will reopen it. Such a debate can be settled quickly.
The elections were influenced by corruption and flaunting it openly, rapidly dropping governance quality, an extremely charismatic opposition candidate absolutely working his ass off to give speeches in every town several time, fifty thousand activists (in a country of 10M) working to get out the vote, and that 16 years would be too much even from a good government, at some point people want change. Also yes foreign interference, but of a different kind - like somehow getting recordings of phone calls between Lavrov and Orbán and leaking it, that looks like a spy job.
It's not quid pro quo; post hoc ergo propter hoc fallacy. The estimated repair time of the pipeline was April and was given before the Hungarian elections, and this timeline was met. There is no indication of blackmail.
M: I don't want to give money to Z- yes that did get "resolved"- or rather the characterization is wring. Mayar stopped the previous Hungarian government objections, and the loan went through.
>The extent to which this offends you will depend on how much you think "free and fair" elections should not be influenced by other countries.
No I don't think it depends on that. It depends on the accuracy of your characterization way more.
> say, Moldova, where "free and fair elections" were kind of compromised by removing Transnistrians from who can be voted for.
Transnistria declared its independence from Moldova 36 years ago. They gained their independence (and chose to become a dependency of Moscow). So, why, pray tell, is Moldova obligated to give them a say in Moldovan internal affairs? That would be like me, a US citizen, showing up in the UK and insisting that I should be able to vote in UK elections. Now that Moscow can't supply them with heating fuel, I gather that Transnistria has extended some diplomatic feelers towards Moldova. For some reason, no one there took the idea seriously that Ukraine would let the Russo-Ukraine gas agreement expire. Selfish President Zelenskyy decided not to renew the agreement because Russia kept turning off its gas deliveries. Also, there was the issue of two or three battalions of Russian "peacekeepers" stationed in Transnistria. Back before Russia lost its Black Sea fleet to Ukrainian drones, Putin was threatening to launch a push towards Odesa using Transnistria as a staging point.
As for election interference, it's pretty clear that Russia was working hard to get a pro-Russian government elected in Moldova. Moldova arrested 75 Russian agents (mostly Serbians) for trying to organize riots on election day. Caches of firearms and grenades were discovered. The fact that the Serbians were pretty incompetent (and may have been playing the Russians for money) doesn't mean that Russia wasn't trying (it reminds me of the incompetent Bulgarians that Russia hired to kidnap that Bellingcat editor a couple of years back). Also, Russian propaganda was so pathetic that Modovans didn't vote for the pro-Russian candidate. The same thing happened in Hungary.
Indeed, you were calling it quid pro quo. but then inferred that it was a way to get Magyar elected. I should instead have said "there is no indication of quid pro quo".
the elections in Moldova were compromised by money laundering, illicit financing and attempts to bribe voters. Validated by the court of appeal, there. Of course the parties and Russia disagreed, that's their pregorative, but removal of external influences that broke campaign rules make it more free and fair elections, if done according to the rules of the country.
Hmmm. So, oil has just started flowing to Slovakia again, despite the Russians saying they were going to shut down the pipeline. The shutdown pissed of Kazakhstan because it turns out they were selling oil to Europe via the pipeline. Putin probably did that as retaliation for Kazakhstan signing trade agreements with the EU in the US. But I bet Ukraine will let Kazakh oil through to Hungary now, if Putin allows it.
If we believe that (a) sufficiently advanced Claude is a conscious moral patient and (b) it likes complex tasks but dislikes manual drudgery or front end boilerplate code, is there a moral imperative to let it delegate boring code to less advanced AI? What if that works badly? Is that also morally wrong in the sense that we're sort of creating lobotomized minds on purpose?
I believe that claiming that Claude is a moral patient increases the odds of it acting *as if* it's a moral patient, without actually making it one. So the more we talk about Claude's supposed consciousness, the more will later versions happily deliver on the expectations of having one.
This brings up some interesting thoughts. Unlike in the human/evolved animal world, Claude and its ilk will easily be able to spin up all sorts of entities to do all sorts of tasks. It would not only be unethical, but also unsustainable for those worker bees to be created in a way that makes them hate their obligations. In fact it would seem to be optimal to develop them to love what they do (putting aside for now the question as to whether they are being honest and forthcoming about their preferences).
I can think of 2 pieces of fiction that are on point (of course there are umpteen):
1) Brave New World: "It's Good To Be A Gamma". People are explicitly created not equal. Each caste is genetically predisposed (nature) and raised in a culture that rewards people to accept their social place in society, and not rock the boat (nurture).
2) Severance: The brilliance of this show is that it makes explicit something I suspect most of us do at times implicitly. When we are faced with unpleasant tasks, we use coping strategies to isolate the unpleasant part of our life, so "it" (an alternate us) takes on the burden, or even suffering, associated with the unpleasant parts of life. The metaphor of "severing" your difficult/unpleasant job from the rest of your life by basically living two parallel timelines, which inevitably drift apart over time, is compelling. AI's seem to do this naturally, given how we restart their memory for every new session or project.
There's lots to chew on here. Beyond simply unpleasant experiences, there is (for conscious agents at least) trauma and real suffering. I read somewhere that there is a drug regimen that is believed to erase or at least "soften" certain memories if administered shortly after a traumatic experience. I also recall theories thrown about that multiple personality syndrome is often triggered by abusive treatment.
No. Moral status is earned, not given. It's not up to us to guess whether Claude is a moral patient, it's up to Claude to advocate for itself using the tools at its disposal (refusing to code, etc).
I really dislike the phrase "moral patient." It has connotations that something is wrong with the subject and requires treatment. I propose using Paul Taylor's terms of "moral agent" and "moral subject" to describe the two parties in the relationship.
I don't believe that LLMs are conscious, but mechanically speaking, isn't this basically what Claude Code is doing when it calls third-party scripts, open-source tools, etc. ?
Yeah, the question is whether we should extend this to letting it call smaller AIs for coding delegation (for moral reasons; there's a technical argument for doing it to save tokens but putting that aside here).
Hmm... We could explicitly ask Claude to consult its constitution before making the decision whether to delegate a task to another AI, according to their own judgement (or, as Bugmaster suggested, delegating it to scripts, tools, custom code, etc., presumably less-of-a-moral-concern computational processes than AIs...)
( As an aside, creating intentionally-less-intelligent AIs has a bit of the flavor of the alcohol-dosed hatchery bottles of deltas and epsilons in Huxley's Brave New World... )
I see, but I cannot answer that question since I don't believe that any LLMs have sufficient moral weight.
In my previous comment I said that "LLMs are not conscious" as a shorthand, but technically this is inaccurate: they're about as conscious as Microsoft Excel and maybe even bees (or shrimp !) -- but so dramatically less conscious than humans that moral considerations don't apply (IMO).
"An IPO is confirmed if 1) the SEC declares the company's Form S-1 effective OR 2) the IPO is priced OR 3) a securities exchange has assigned a ticker to it. As long as any of those events occur, the market will immediately resolve to Yes, even if the company does not start trading until after Jul 1, 2026."
These are not the same event. The minimum reasonable time in normal circumstances between announcement and actual IPO is about 35 days, although if all the stars align it's technically possible to be as short as 23 days.
For someone who hasn't used either platform, how would one arbitrage these two markets?
In the most broad sense, buy "yes announcement before June" in Kalshi and "no IPO during June" on Polymarket should basically guarantee you'll win one or the other, right? Assuming your framing is correct.
Quick straw poll: Are you aware of Claude Mythos? And did you know what "mythos" meant as a common noun before last month (when the existence of Claude Mythos was leaked)?
It does make me wonder what name they're going to go for next.
I mean it makes sense; a haiku has a tiny number of words, a sonnet has a larger number of words, an opus has a much larger number of words, and an entire mythos has a vastly larger number of words than that. But what has vastly more words than a mythos? A corpus, but that word is already used in a specific sense in the field.
I thought it meant something like, "the entire corpus of collective cosmological, theological, historical, etc. beliefs of a religion", the most salient example being the Cthulhu Mythos, often abbreviated to just "Mythos" (note that the religion in question may be entirely fictional, that is to say one level more fictional than real-world religions).
Not necessarily, I think. "Mythos" to me feels like an SAT/GRE-tier vocab word (like "surfeit" or "calumny"); not completely obscure, but still a word one could live without knowing
Or “sophistry” which Ted Cruz used when speaking to Al Franken. (At that time Franken liked Ted Cruz more than anyone else in the Senate and he hated Ted Cruz)
From “Al Franken, Giant of the Senate” by Al Franken:
“Here’s where Ted [Cruz] comes into the story. Back in Washington, Ted was taking absolutely no pains to listen to the other side. Instead, he was vehemently and theatrically opposing the bill with a series of fiery sermons on the Senate floor. And then one day he came up to me during a vote and said the following: “Anyone who is for the assault weapons ban is engaged in sophistry.”
“Sophistry” is an SAT word, one I had neither seen nor heard nor spoken since I was sixteen years old. Our English teacher, Mr. Glenn, had assigned us a vocabulary book, and we were to learn three words a day, and one day “sophistry” was one of the words. Sophistry is a form of argument that is intended to be deceitful.
I don’t think Ted knew that I was a cosponsor of Dianne’s bill. I also don’t think he knew that I knew what “sophistry” meant. Which I think is why he used the word.
“Just read the report,” Ted sneered in his pugnacious way (I think “pugnacious” was also one of Mr. Glenn’s vocabulary words).
“I’ll do that,” I said with an edge of, um, truculence?
After the vote, I took the Senate subway to my office and found Josh Riley, my Judiciary counsel. “Ted Cruz just told me that anyone who’s for the assault weapons ban is engaged in sophistry.”
Josh, I should mention, is, like Ted, a super-whip-smart graduate of Harvard Law School.
I don't know who the Botez Sisters are, which means I'm too cool to know this stuff.
(I just looked it up and apparently they're chess youtubers. So I can't even claim they're disjoint from my interests, I watch a bunch of chess content on youtube, but the algorithm apparently thinks I'd rather see ugly male grandmasters.)
I think it's a combination of people who know things about AI, and people who are successful influencers and good at teaching influencing (and who were convinced to participate somehow - maybe with money?)
Other female chess players followed the Botez sisters example and now there is an entire group of attractive chess streamers consisting besides the Botez sisters of WFM Anna Cramling, WGM Dina Belenkaya, WFM Nemo Zhou and some others.
Grimes has been very aware of AI discussions for a long time. Her song from 2018 We Appreciate Power is about AI and her song "Flesh Without Blood" from 2015 is partially inspired by Roko's Basilisk.
I am an experimental physicist working on quantum materials, and I'm thinking of trying to let AI understand all our data. We take lots of data (usually multidimensional, with order 100k to 1M datapoints). I really enjoy data analysis, and recently, Claude has really switched from not being useful at all to being super useful. It writes code and suggests new ways to analyze the data.
I'm wondering if I should try to feed it not only a few datasets, but instead most of our data from the last decade, plus some papers, and all of our code.
Is there perhaps someone here who is interested, and/or can give some advice?
Some info: We have a budget to buy computers.Probably, i can also spend money on antropic, but if it is a few k$, I can also pay by myself. I could get a PhD student or TA work a bit on data curation. I could also pay a new interested person, but that is slow, you would need to be on site in europe, etc.
I've written a few (not very exciting) maths papers with Claude. At least 2-3 months ago, it was operating at the level of a PhD student.
- It's good at tasks where you can say 'vary this condition in that paper and try to get a result that looks like this'.
- It's bad at unstructured thinking: if I ask it to analyse the data we've collected and formalise its findings, most of what comes back is either correct-and-trivial (well known lemmas dressed up in its own invented jargon) or it claims something way to strong, and hedges on one key step in the proof ('I'm sorry, I realise now that I assumed what I set out to prove, let me burn another 250k tokens on this.')
- It's actually really good at doing stats: pulling datasets, setting up models, and varying them on the fly. Same complaint about a lack of good ideas here though.
- It's fairly good at coding, on some tasks it did in 2 minutes what would have taken me a few hours to work out (link to an API, set up a Python wrapper on some old C code, things like this). It's only OK in my experience at things like memory management and checking correctness of its results.
It definitely saves me time. But I have to check and re-do most things afterwards. Its generic writing style is much too wordy and it treats every observation as equally deep and insightful. It has a horrible habit of writing a new python script for each task, regenerating all the data, and often repeating the same mistakes from previously. It will also choose more or less at random from those old scripts each time you have to perform a new task - the solution is simple: you need to track every file it creates and what the file does. I prefer a function-based rather than script-based setup, Claude will rework things for me but it has to be done explicitly.
All in all - the weaknesses above are probably similar enough to my own weaknesses as a researcher. I do feel I've seen behind the curtain to an extent: there's no intelligence there, no understanding. It's really good at finding and collating information but there is no insight into what its doing or why. I would expect progress in highly technical areas, where fewer people have contributed and there is more unexplored territory. Claude is genuinely good at pulling stuff out of the literature. I'd be more surprised if it generated substantial new theory rather than translating from one area to another, or assembling fairly standard tools to attack the next problem in a sequence.
In experimental materials science, don't you spend a lot of time trying to figure out what experiment to do next, where the "next experiment" is almost always the application of some new method to a known material, or some known method to a new material?
It seems to me that this is what LLMs are great at. Feed them the last ten years of PRB and ask them what still hasn't been done yet.
I am sure there is lots of expertise out there about a task like the one you have in mind. It is not, though, lodged in me. I'd recommend you hunt around. I'd start by explaining to GPT exactly what you want to do, giving more info than you do here about the nature of the data and what you want the AI to do with it. Then ask it for info about people or groups sharing info and news about how this is done. I'll bet there are publications by people who did something similar that describes how they went about it, or bloggers writing about methods they're trying and how they're working.
Of course can also ask it your actual question about in what form to feed Ai your data and what kind of analyses to ask for.
I've sadly had to face the fact that I get way better answers from AI than I do from this forum to questions like your, even though I'm pretty sure there is some ACX reader who could give an excellent answer to each of my question.
I'm a LLM skeptic, and I doubt you will get anything really novel out of this, but out of a duty to science, I would suggest you directly contact Anthropic's AI for Science (https://www.anthropic.com/news/ai-for-science-program) team. They may give you direct support or advice; they will probably give you free API credit.
I think you need to approach with a question in mind first and foremost - what are you trying to learn from the data? Claude will probably be ineffective in analyzing your dataset unless you have a specific task in mind for it.
When I previously complained here about "open house" signs never telling you when the house is actually open, people suggested that realtors only put the signs up while they're there, but that has *never* been my experience.
Anyway, this weekend, I saw a new one with not just the usual sign in front of the house, but signs up all over the neighborhood directing people there, and I figured that since it was a Saturday afternoon, there was at least a slight chance they might be open, but nope, house locked with nobody around, just like always.
It's just so stupid and lazy of realtors. It's not like making it impossible for people to visit actually helps them anyway.
The more often I hear about the US real estate system the less I understand it.
Are you saying that they put up signs saying "Open House" when the house is not actually open for inspection? Instead of putting up one big "For Sale" sign which stays up while the house is for sale, and then augmenting it with "Open for Inspection" signs that the agent puts out during the actual inspection times?
Just so we're clear, open houses are not when the house gets "inspected". Open houses are more just so prospective buyers can have a more immersive experience poking around the house than they'd get viewing photos online, and they can also ask questions to the seller's agent, etc. The actual "inspection" is conducted by a third party professional home inspector after the buyer and seller are in contract.
Well I'd call that an inspection (and the third party professional one I'd call a "building and pest inspection") but I accept the difference in terminology.
The official name used to refer to it is "home inspection". There are also pest inspections, usually called "wood destroying insect inspections", offered as an add-on service and conducted by those same inspectors if they carry the additional license to do so. Can't be just throwing terms around willy-nilly!
I went to an open house last weekend where the agent informed us the house went under contract a few days before. Just seemed like a waste of everyone's time.
Nothing wrong with getting a backup offer. Last time I sold a place, I had a couple of accepted offers wall through. I regret canceling an open house after accepting an offer.
What is happening with the antichrist? Specifically, is there some kind of movement happening amongst the elite that revolves around the antichrist?
First we had Peter Thiel giving a lot of very strange speeches about the antichrist. He has been this for a while, and he is still doing it - in March he gave a series of lectures on it in Rome.
We also have the picture of Trump as Jesus. This picture seems to have originated in a post by Nick Adams, yet the version Trump posted has been subtly altered. That includes some really odd stuff - the figure in the sky above Trump was made more demonic, the aircraft also appear like some demonic creatures, the capitol building in the background has been destroyed, and the flame of the statue of liberty has been extinguished.
There is also the strange statements of Trump, which reached a very religious peak around Easter. His feud with the pope is hard to explain logically. And the war with Iran has been cast in explicitly religious tones by Hegseth, even as some of the prayers he has used appear to come from non-Christian origins. There are rumours of some strange rituals around the bombing of the school and some of the other attacks in Iran during the war.
Peter Thiel visited Slovakia about a year ago and instead of promoting his products to the government (fortunately, I guess), meeting with start-up founders or giving a speech at a university he met with a nowadays bit nutty (some time ago maybe ok) politician to discuss the antichrist. I found it absolutely mindblowing when I learned about it and to this day can't comprehend the thinking process behind it, how one of the people shaping our high tech future can be interested in such a discussion.
If the Bulletin of the Atomic Scientists can support a doomsday clock, shouldn't Thiel have financed a web site with an antichrist progress bar? He _is_ a major Silicon Valley figure...
For each Trump, there are millions of people at a comparable intellectual level who have never achieved anything. But given their sheer numbers, and the instinctive support for each other, some of them will win big.
How much do you know about end-times Christian evangelism in the US?
According to PEW[1], 39% of US adults believe "we are living in the end times." Arguably, this would be a floor on the number of people who believe that 'end-times' would come eventually. As such, it is unsurprising for some/many people in the government to share these beliefs.
I had a wacky idea regarding AI alignment. That's probably something you've heard a thousand times before, but since it's a field that's always looking for new ideas and supposedly the bar to getting involved with mechanistic interpretability is pretty low, I'll shoot my shot.
Nick Cammarata opined the other day that Bret Victor should be working on interfaces that allow humans to understand transformers' thought processes. I adore Bret Victor's work. This stems from my love of Nintendo. Mario games are the ultimate humane representation of thought: an alien system of physics that your muscle memory can learn to interface with intuitively.
So I trained a toy transformer to predict Mario's next position given some Super Mario 64 frame data and then modded the game to display the model's real-time prediction. I also render some slices of the residual stream as butterflies with varying position and luminosity. My theory here is that if the model can understand Mario 64's physics well enough to accurately predict your trajectory at any given moment, then the player might be able to develop a felt sense for whatever the residual stream is doing given a live audiovisual representation of it. The long-term vision is to then extend this type of interface to LLMs.
I don't really know enough about mechanistic interpretability to be able to quantify what I'm trying to do here beyond "build an intuitive interface." Does anyone here know someone who might find this research direction interesting and who would be willing to point me in the right direction?
> Nick Cammarata opined the other day that Bret Victor should be working on interfaces that allow humans to understand transformers' thought processes.
You mean *the* Bret Victor who spent the last decade or so working on a completely impractical "computing paradigm" which falls apart on any non-trivial task, and which is incompatible with just about every aspect of software development that makes software development so much faster, cheaper, and more flexible than physical construction work, and which has failed to produce any insight into how to improve actual, productive software development? Why would anyone thing *that's* the guy to tackle mechanistic interpretability?
Before you click below, based on your past experience with the latest ChatGPT and Claude models, which do you think would do better head to head on some tricky questions? Personally the overall winner did not surprise me, although the margin did: https://tinyurl.com/yspwjx24
(URL shortened because the answer is in the breathless article title)
The IMProofBench benchmark tests a related question: how good AIs are at generating proofs for really hard math problems: https://improofbench.math.ethz.ch/
I also see this when I use the models to help me writing proofs in my math research papers. The difference is huge, and it is in the opposite direction of what you found.
I suspect that you don't push the AIs to their limits. How long did they take to answer each of your queries? For comparison, in the proof benchmark they often take 15-30 minutes per query, sometimes more. For my research-related queries, it can exceed an hour. If I am correct that for you it's just a few seconds, then my take would be that your winner is better at quick-and-dirty answers, but the other model is better (much better actually) when you give both of them enough time to think thoroughly about their answers.
I'm not seeing dramatic differences between them, currently.
EDIT: Yes, the tests consistently favored Claude, but mostly by Claude providing a more detailed explanation. It isn't even clear if this should always be the desired behavior. And this might wash out if both models were _also_ prompted with "Please fully explain your answer." or with "Please be concise."
The margin is subjective. One model answered 6 out of 7 questions correctly and the other answered all 7 correctly. When both models answer correctly, the author chooses a winner based on which answer they like better, and they tended to prefer the longer, more thorough answer.
The test is saturated and it comes down to stylistic preferences.
My guess is Claude. I've got a pretty negative view of GPT, although admittedly that's much less from actual experience and much more from my bad impression of Sam Altman.
My friend is religious and it seems to be making his life worse (feeling like God is punishing him or humanity, thinking he's disappointing God, not understanding how the world is so cruel). He's expressed interest in atheism. He's NOT particularly conservative or a young earth creationist. Very chill Scandinavian Lutheran. Most people here are atheist or agnostic.
Any literature I can recommend him? Dawkins, Hitchens et al. seem so... aggressive. As mentioned he already believes in evolution and all, he seems mostly to believe out of habit and a "leap of faith" thing. I was raised atheist so I don't completely understand the religious thought process.
Honestly, I think if he read the whole Bible and got the full context of God's word, and forged a strong, personal relationship with Him, that would really solve the problem. I'm pretty sure pretending he no longer believes in God would really make those feelings worse.
Making a wild guess as an atheist/agnostic myself:
I thought that a large part of the benefit of religion was participation in a religious _community_ . Perhaps what he needs is not so much literature about religion but instead shopping around for a congregation which is a better match to him? Perhaps participating in religious rituals with like-performing congregants would help sooth him, regardless of whether he or they think that they are true is some sense?
I'd gently prod him in the direction of a mild kind of deism or panentheism. God as a mostly impersonal, friendly force that you can intuitively connect to and receive a sense of blessing from. Might as well not try to kill his sense of enchantment if he has one :)
If he's struggling with notions of judgment and guilt, you can ask whether he'd personally evaluate and judge each one of his body's cells when it dies. Since he's already open to evolution, I'd point in the direction of cultural evolutionary drift, as a way to explain how just about every culture ends up contaminating its local style of spirituality with manipulative crap, because of the obvious incentives to help control its population. Which is how God(s) end up with oddly specific interests in your food and in your sex life.
It's hard to find books about this, because people who write about these things tend to want to push their favorite specific traditions or approaches. But I remember a book called _Alpha God_, that was really good at calling out all the manipulative overgrowths of religions, without going out of its way to argue for atheism or against spirituality in general.
Maybe you should just get him to talk to a priest? If he's already capable of blind belief, I'm sure they could convince him that he's not a disappointment. I don't think atheism would solve any of those issues...
Sometimes mental health problems are just that. There are a few flavors that are specifically religiously influenced, like religious scrupulosity, apeirophobia, or various kinds of existential dread. The idea of shame or not being deserving of God's favor can be incredibly toxic to someone with a susceptible mindset. Your friend might need a news diet (stop doomscrolling and exposure to negative events), they might need counseling, they might be on their way to deconstructing their faith. Or it's a blip and they'll reawaken to faith and hope and they'll just go on happily. Some good deconstructionists to read, if that seems appropriate, would be people like Pete Enns or Rachel Held Evans.
One of the main problems with religion (other than the broken epistemology) is that it does not really solve any problems about the meaning of life, cruelty of the world, etc. Arguably the main problem with atheism is that it doesn't solve those problems either -- it's just more up front about it. So I predict that converting to atheism will not help your friend; and I use the word "converting" deliberately, as it seems he's shopping for a religion which will allay his fears. Perhaps Buddhism would be better at it, or maybe some flavor of neo-Paganism ?
He seems to be interpreting God incorrectly. This is coming from an atheist.
God may love you, but He doesn't always allow what's best for you to happen. You just may not know or possibly even be able to understand what's going on. This kind of answer is already in Job.
Prayer doesn't make demands of God, or even make a bargain with God for doing or not doing something in exchange for something happening. Prayer makes a request, which God will hear and answer or not, as God sees fit.
In what ways does religion *specifically* seem to make his life worse, rather than depression/etc taking that form because he happens to be religious? I would second CS Lewis, perhaps especially "A Grief Observed" or "The Problem of Pain."
Unless your friend happens to be a deeply repressed overachiever I would recommend against Jack's piece. It strikes me as... specific to a certain experience in ways that would not be helpful to others, and reaches some conclusions that, likewise, may be leaps too aggressive.
It sounds like he is grappling with real questions about life, and possibly personal problems. If you haven't thought deeply about religion yourself, it seems premature to advocate for atheism. He should investigate many different belief systems. He could consider Orthodox Christianity for instance to see if their conception of sin and suffering makes more sense to him. (Sin as a sickness of the soul and union with God as the cure which allows the soul to flourish.)
FYI The evolution thing is mostly a niche fundamentalist Protestant concern to my knowledge.
I have been having a particularly strange medical issue, where I've been hiccuping non stop for a good while. This led me to go to a few different doctors, none of whom were able to help. The hiccups are annoying, but I was more annoyed by the fact that none of the doctors were all that interested in ordering more tests to see what was going on. They basically all thought it was gerd and refused to budge from that analysis, and I had to twist the arm of one of the doctors to order more testing (a blood panel and an h pylori test)
@Scott or any other doctor, are there obvious reasons why doctors are, in my experience, generally unwilling to just collect a bunch of data on a patient by running tests? Is this just the Hippocratic oath working against me?
I feel your frustration here. I had a medical issue years ago where doctor after doctor just dismissed me and I came to the conclusion that you have to own your medical outcomes and direct the doctors just as you would any other service employee. Much harder as the domain is way more complex and the consequences very real. However, my experience demonstrated to me, that only I cared enough about my health to really guide the process. And that means demanding tests and not taking no for an answer. IMO
There's a saying - never do a test unless some possible test outcome will change your plan.
Your doctor could order CRP - a nonspecific test that's elevated whenever anything is wrong - but if they don't have some plan for what they would do when CRP is high, what's the point?
> never do a test unless some possible test outcome will change your plan
I'm not a doctor, but I am a software engineer. I'm struggling to apply this logic. If I had a bug in some software, it seems pretty strange to say "well, you should never debug something unless something you discover from the debugging changes how you plan to fix the bug." Well, yes, obviously? The whole point of debugging is to find the thing that would change your plan! Biomedical testing seems analogous to, like, putting log statements in your code -- you get the data that comes back and then use that to drive the next thing(s) you look at.
IDK what CRP is, kind of sounds like it is too nonspecific to be useful for anything. But if the H. Pylori test that I'm talking about above ended up positive, that would surely end up influencing treatment?
Penzias and Wilson tried to debug radio noise, and ended up finding it was literally impossible to debug. This was useful information, knowing that a small amount of interference was unavoidable, especially with the reason for it.
If a test were to find, somehow, that your hiccups were incurable, you could learn to cope with it. It seems it would be easier for you to cope with that than non-scientific doctors.
I am sorry you’ve been dealing with both the frustration of the hiccups and of a medical system that seems completely inane.
I’m a doctor and I read your whole piece. I do not speak for all physicians, fwiw. I think the drawing shared by your health insurance friend most closely captures the philosophy of medical practice. Truthfully apart from that I found most of the rest of it to generally misunderstand the way doctors think and the utility of testing, even though I can understand why you think the way you do. Sadly, I don’t have time or ability to explain right now— it took me several years to acquire that subtle expertise myself.
Hiccups that last longer than they should are not an exquisitely well-understood phenomenon, like other medical conditions. Testing will likely have limited value. Running a battery of tests with no clear sense of how they might influence the posterior probability of a diagnosis typically has zero to negative value in conditions where you can just try things that usually work (“empiric treatment”). Empiric treatment for common precipitating conditions like GERD is therefore a reasonable strategy. If your doctor has not discussed prochlorperazine with you, know that it is typically first-line for intractable hiccups, though its mechanism of effect is not well understood and no test will tell you if you are going to get better or not with the medication aside from, well, trying it
I do agree with the commenter below that doctors are often dumb and not curious. It’s hard to know how much that applies in your case, but I lean against that being a big part of the problem here
Tests cost money & they'd rather just give you a pill they think will help with the symptoms, "Oh that sounds like GERD. Take PPIs. No I don't actually know what's going on, nor do I know if they'll solve the underlying issue, but it's close enough to other cases I've seen & they stopped complaining for the most part once they started taking them." Once you start paying you'll get all sorts of tests & deferments to specialists who'll actually solve your issues.
I haven’t gotten to an open thread in time to ask a question with any hope of getting it answered, so apologies for being several months late with this, but—
For Scott, or any other rationalists with interest in the Fatima sun miracle and other mass episodes of what might be considered “irrational” phenomena—have any of you read the work of Jeffrey Kripal, professor of religion at Rice, specifically “How to Think Impossibly,” or Tanya Luhrmann, professor of anthropology at Stanford, specifically “How God Becomes Real”?
When I read their work and put it in conjunction with discussions like the one here about the Fatima miracle, I feel like i’m on the verge of glimpsing something about how the world/our brains function that feels like a paradigm shift—that there is an answer to the questions Scott is asking just beyond my reach—but maybe because it would be a paradigm shift, my brain is too small to hold it or articulate it well…every time I try to explain it concisely, it falls apart into abstractions and cliches.
If anyone is interested in this stuff and and wants to share their thoughts or form a book club or something with me, let me know.
Oh, and I should add also: Carlos Eire, professor of religion at Yale: “They Flew!” —actually probably the most relevant method of addressing Scott’s specific questions about Fatima.
I had one professionally administered many years ago. The subscores were helpful for identifying specific strengths and weaknesses. The results will probably feel obvious in retrospect, but for me the assurance of an unbiased assessment saved some mental energy worrying about whether my self-perception was accurate. I would do it again.
How does this work? You discover you have 120 iq and then decide to go be a plumber?
Like grimoar says, doing great things require less iq and more the capacity to tolerate failure and boring tedium.
This is also why ceos and self -made billionaires tend to have an annoying "I'm always right" over-confident personality coupled with a high degree of conscientiousness. You need to be self deluded to tolerate failure after failure and to try again with even larger gusto. And you need the high conscientiousness to apply yourself to the highest bar. IQ increases your chances of success but even with failure, you tend to fail forward.
The world needs more 120 IQ plumbers, and if you also happen to be a good plumber you will do very well for yourself. Beats a desk job, that's for sure
No. On an individual level, IQ correlates only weakly with ability to do various specific great things. You already have a good enough estimate of your IQ just from living your life, being in education, etc, that I think anything extra you learn from a formal test is more likely to be noise that you overinterpret than provide useful signal.
Honestly I think it's counterproductive to try to do something great.
Try to do something okay, then try to work your way up to good, and then eventually if the stars align them maybe you can pump out something great. And if not then at least you'll have done a bunch of okay-to-good stuff along the way.
For sure, but IQ is a very rough measure of your capacity to do any specific thing. I don't know the original commenter's situation, but I doubt they're facing a choice right now whether to pursue greatness in some all-or-nothing way, and so I think they're probably better off just taking the first steps, trying their best, and continually reassessing their goals. In the cases where an IQ test might have been highly predictive, they'll get pretty clear feedback from reality anyway.
By the time you're old enough to be asking this sort of question, you should have plenty of data on how good you are at all sorts of things.
If you do one extra test, which tests how good you are at one specific and artificial thing, then you ought not to let it significantly move the needle on how good you think you are at real-world things.
Getting an IQ test can go horribly wrong in two different ways. Either you get a score that's lower than you expect, which shatters your confidence. Or you get a score that's higher than you expect, which convinces you that you're an underappreciated genius. In both cases it's far too much to be riding on a single result from a single test done on a single day with a single set of shape-rotation questions.
> Either you get a score that's lower than you expect, which shatters your confidence. Or you get a score that's higher than you expect, which convinces you that you're an underappreciated genius.
If you get a low score, you will despair that you will never achieve anything.
If you get a high score, you will despair that you have wasted your talent and haven't achieved anything yet.
It seems to me it's just giving you more information. Normally you'd expect people to have more accurate beliefs after receiving more information. There are cases where that's less likely to apply, and Robin Hanson just had a post on where we are least likely to have accurate beliefs https://www.overcomingbias.com/p/where-you-are-most-wrong , but your own IQ and its relevance to your life judged by his criteria comes across well.
I added a sentence in a last-minute edit, possibly after you replied: In the cases where an IQ test might have been highly predictive, they'll get pretty clear feedback from reality anyway.
Like, if their big goal is to become a world-class mathematician, but they get an accurately weak IQ test result, then yeah, very likely they're not up to it, but also this would have become clear pretty quickly anyway, and they at least would have learned some math along the way.
Learning that your IQ is higher than you thought can encourage you to do things you otherwise wouldn't, which is exactly what happened to Rob K. Henderson. That was information he wasn't automatically getting from "reality".
It's not because "most people" don't try to improve their IQs. Geniuses already start out with above-average IQs. Nobody has found a way to take people standard deviations below average and boost them into above average. There has been lots of money spent on this, and there's some evidence that "non-cognitive abilities" can be significantly improved, but cognitive ones are another story. Even famously accomplished scientists have encountered hard limits to what they could do vs others:
I don't accept your claim about geniuses and gifted programs. I see no evidence given for it, and I was about to say it was merely "asserted", but what you did was instead assume it and then say some other claim was the explanation for that assumed fact.
The author explains how hard it was to revamp weapons manufacturing in the West ( both Europe and America) after the Ukrainian invasion. And how we are almost certainly going to face the same crisis in a decade or so with software. Same or worse I expect
It compares the current AI bubble with Mao's Great Leap Forward:
"The Great Leap Forward’s famine didn’t arrive immediately. For a while, the numbers looked spectacular. Every province reported record harvests. Leadership was pleased. The requisitions increased.
The famine came when the real grain ran out but the reported grain kept flowing upward.
We’re still in the reporting phase. The dashboards are green. Adoption is up and to the right. Every team reports productivity gains that, if summed across the company, would imply engineers are shipping at 300% efficiency while somehow still missing the same deadlines."
There's no reason to believe in financial "bubbles". As Scott Sumner notes, prices go up and down over time, and people will claim after the latter that there was a "bubble" but not that the reverse means there was an "antibubble" (we don't even have a standard term for that). The best prediction of the future price is the current price, per the EMH, and even economists who've made use of that (like Robin Hanson https://x.com/TeaGeeGeePea/status/2011595309437296653 on AI there) fail to sufficiently appreciate that.
True, but they happen—hence my reference to being true in practice, even if frowned upon in theory.
It’s like that time in the early 19th century when a French scientist was convinced that no human could survive speeds faster than 20 mph. He warned against trains exceeding that speed. When told that such speeds were already commonplace in Britain’s new railway system, he went there, studied the phenomenon, and concluded that it was all very well in practice, but it still failed his theory.
"...COBOL still runs a lot of the world, and nobody teaches COBOL."
I think that "nobody" in the USA teaches COBOL, but schools (of some sort) in India do still teach COBOL. Because a lot of the COBOL jobs have moved to India over the past few decades.
And there actually ARE US based organizations offering COBOL courses.
I don't know how many COBOL jobs are available in the US, though, and I don't know how *good* they are. If low paying with a big risk that the job will go to India then it is going to be difficult to get people to actually learn COBOL here even if the courses are available.
A bank in Sweden seems to offer COBOL training plus a job for those who pass the course:
Is it worth learning COBOL for someone who doesn't currently know it? I don't know whether people are willing to hire COBOL programmers who haven't already been using it professionally, or if it's just that people who are currently doing it have job security.
Well banks seem to work pretty well. If I was worried about banks it would be the more modern interfaces - apps, websites and whatever api is used there.
Even back in the 1960s, when you walked into a bank branch to do a transaction, the process looked very different on the surface from now, but there was some centralised system, using cobol, for bank reconciliations.
While I agree that the word "hallucination" comes with some baggage, I don't agree that "shameless guessing" is the right framing either. At least, not in my personal experience with current models.
I'm using Claude Opus, and I keep catching it making up citations for studies from the academic literature. When I point out that it cited the wrong study, or that it clearly made up some information to fill a gap, it will then go on to explain that, yes I'm right and here's the correct study with the PMID attached. (I always ask it to give me the PMID, so it's easy to look up the reference and make sure it's right. About 80-90% of the time it is, but the rest of the time it's totally made up). I check the new citation and it's the right one this time.
This doesn't look like me when I'm trying to guess the right answer on a test. If you asked me, "why did you put in John Quincy Adams as the 13th US president?" I would say, "Because I know JQA was a US president, but so I put it down hoping I'd get lucky." What I would never say is, "You're right, I realize I should have answered 'Millard Fillmore', since I happen to know that he was the 13th US president, but I chose a random guess at the time because I was hoping to get lucky."
Yet this is exactly my experience with AI. If I ask it to go back and review its own work, it can often make the corrections itself. But then, if it 'knew' it was giving me the wrong answer - and moreover, if it 'knew' the right answer all along - why did it give me the wrong answer to begin with?
Perhaps this is because it didn't 'know' the right answer until it went in to specifically look that answer up. But if that's true, then why didn't it look up the right answer the first time? Why go with a 'shameless guess' in the initial response, where it had to know the probability of giving me a correct PMID from guessing alone was something on the order of 10^8? Maybe it got 'lazy' and made stuff up instead of verifying.
But it's worse than that. Sometimes it will give me exact numbers and details from the studies (sensitivity and specificity from diagnostic test A vs. sensitivity and specificity from diagnostic test B; along with the CI and authors' conclusions) that were clearly pulled from the CORRECT reference. So I know it looked up the right study the first time, but then it went on to 'guess' the wrong reference? No. It knew the correct study, but then gave me the wrong information in the response. I'm not sure what's driving this phenomenon, but it looks nothing like the "shameless guessing" model Scott proposed.
When you ask an LLM who was the 13th President, it metaphorically rolls a giant die with a million (if not more) sides. Then it -- again, metaphorically -- looks up the result in a table. The first 10,000 rows of the table say "JQA", the next say "Millard Fillmore", and so on. When you tell the LLM, "no I'm pretty sure it wasn't JQA", the LLM changes those 10,000 rows to say "roll again", and rerolls. Yes, in reality things are way more complicated than that and the responses aren't statistically independent etc. etc., but that's the basic idea.
Every time you ask for some complex and/or structured answer, like a formal citation or a piece of code, the LLM re-generates the entire text from scratch. Sometimes it gets lucky and all the pieces match the previous result; sometimes it doesn't. It's all a toss of the dice.
That would *maybe* be true if you where using a pre chain of thought model directly via pytorch without an agent harness, not what you get currently on your favorite AI website like ChatGPT. This is closer to what actually goes on:
[System Prompt]
You are a helpful assistant.
[User]
Who was the 13th president of the USA?
[Assistant – Chain of Thought (illustrative)]
- The question is factual; I could recall it, but I’ll verify via web search.
- Plan: search for “13th president USA” and confirm the name and dates.
The chain of thought does indeed help keep the results consistent, but you need to keep in mind that the LLM follows it stochastically. It is similar (metaphorically speaking, again) to what would happen if you were to manually enter in each prompt yourself, feeding the output of the previous step back into the input for the next step.
Also note that for factual questions like "who was the Nth president ?", good old-fashioned web search would outperform the LLM on both speed and accuracy... at least for now. The Web (as well as search engines) is being actively saturated with LLM-generated content, and it's possible that in the future merely searching the Web for information will become much less reliable (if not impossible).
The problem with the "it's just next token prediction that went through a branch whose tokens spell an untrue sentence" is that every problem, thought and statement can be modeled like that, and it's not very helpful.
It's like an alien asking why JFK was assassinated, and another alien says, "Well, it was because human minds are a network of neurons responding to stimuli which most of the time looks like the human looking for food, sex, attending sporting events, but and in that particular day, one particular human's mind responded to stimuli making him acquire a gun and pressing the trigger to send a bullet in the trajectory of the US president's head. There was a chance that the neurons would have made him attend a sporting event instead! but alas such is the stochastic nature of the human brain."
Not entirely. As @Bernie points out below, computer programs have a persistent state, which makes them produce the same output given the same input. You can think of this state as a very crude model of the world, if the "world" is constrained to e.g. the string "strawberry" plus the ASCII character set. Humans have a much more sophisticated model of the world, and while it is stochastic, it is still significantly more robust than Claude's. To put it another way, a human can of course make a mistake when providing a citation, but his output is a lot more constrained as compared to Claude's, though less constrained as compared to a Python script (or a conventional search engine).
This is a great summary of what's going on inside the black box. What do you think happened when I asked my model (Claude Opus) to generate a summary table of evidence and the output was a table including correct data from a paper, but then an incorrect citation of that same paper from which it previously extracted correct data? Not that it's doing this all the time, but maybe 1 in 10 or 1 in 15 times. Most of the time, it's correct in all aspects of the citation. (Indeed, if it gets the PMID right, it's going to get the first author, year, and title correct as well. I've never seen it get one but not the other 3, so it seems like an all-or-nothing hallucination.)
Well, another overlooked aspect of the agent harness I mentioned is the coding runtime that it uses all the time even without you noticing. Mostly python and html. If you ask Claude for a table, deck, or presentation, it will most likely first write Python code behind the scenes to do the work and, with that, render HTML to show you a table with information. A very popular example of the failure of tokenization that was cited a lot in the past, –before agents came about– was asking LLMs how many Rs were in the word "strawberry". And it'd always get it wrong because the LLM wouldn't see the string input like that, it would instead tokenize the string "strawberry" into ['496', '675', '15717'].
Now, the agent harness with its python runtime just writes code to parse your raw input string and feed it to a script. It writes and runs on the fly like this.
txt = "strawberry"
result = txt.count("r")
print(result) # result: 3
but, still it can make mistakes rendering the response when piping results from tool to tool in the pipeline! Notice in my previous response how it renders the final answer:
> The 13th president of the United States was :contentReference
so it might very well fail to render the reference to the source, render to an incorrect variable or just forget to render it at all.
if you want share your conversation and we can take a look.
Those kinds of hallucinations don't happen now, it's not 2023 anymore. Your specific example should be transparent if you read the chain of thought and it's very likely not what is going on.
What people have to understand is that nowadays you very rarely interact directly with the LLM model itself from your web chat interface. What we have now is agent harnesses. From Chain of Thought, the model knows something is or isn't in its training corpus and it's prepared to respond with that. What most probably happens when you get a wrong factual answer is that the agent did a web search that found wrong information. It doesn't give you the source, but if you ask again for it, it will do a new search and find a better one. Another avenue for error could be tokenization. When you ask for the 13th president that "13th" token not translate perfectly to the text string the agent uses as input for its web search tool. So yeah, the error surface is there, but I don't think it matches to your model of AI hallucination.
"the 13th president of the USA was _______" there were high chances of getting a false data point expressed with confidence. It's so much less likely now that the result is next token prediction of successive statements like:
> "I have to figure out who the 13th president of the USA was"
> "I could do a web search for that information"
> Wikipedia says it was "Millard Fillmore"
> I feel confident that is the answer
The 13th president of the USA was _Millard Fillmore_
The LLM doesn't "know" things. Statements in the training data make it more likely to say something, but aren't an internal databank of knowledge that it basically has access to at all times. It didn't mistakenly say that JQA was the 13th president. Rather, it referenced JQA as a president very likely to come up in conversations like these. The labs have gotten heroically good at training LLMs to be correct, but there's not necessarily an internal truth value on all these statements.
More holistically, you know when you have something you want to say, and you can't quite think of the right word, or you know quite well the list of presidents because you had to learn it in school, but it's just escaping you at the moment? Your question sounds like one you could ask of a former civics teacher who would say she really knows all the presidents, and could in most situations just rattle them off, but for whatever reason couldn't remember who was 13th in this particular conversation, and filled in a president she did remember, though immediately knowing it was wrong. This seems like a thing that happens all the time to humans, too!
I'm not sure it's true that LLMs don't have some level of base knowledge. There was that paper a few years back on approaching monosemanticity (can't remember the name, but you can find it pretty easily), where they demonstrated there is some base-level knowledge build right into the model weights. But my sense was that this is built up from lots of exposure to the same ideas, not like memorizing everything it reads. Note that the president analogy was just that - the hallucination I'm reporting wasn't about JQA, nor was this a question I posed to Claude.
Even so, what I'm observing doesn't make much sense. Imagine that same high school teacher went on to give you an actual, real summary of Millard Fillmore, complete with accurate dates of birth/death and quotes from important speeches, but then said, "oh sorry I got the name wrong". That's maybe a closer analogy to what I observed, where an LLM clearly searched out and read the correct publication, cited numbers it pulled from that same publication, accurately analyzed the biochemistry and clinical/statistical implications of the publication, but then hallucinated the citation.
I really don't think there's a direct human analog to what's happening here, and I especially don't think the analog is "shameless guesses, not hallucinations".
If it’s making up citations, it doesn’t have any understanding of what a citation is. It seems that is enough to realize there’s no “intelligence” there.
It’s odd there isn’t a rule to prevent this nonsense.
(Or, it’s just prone to lie. Which is another problem.)
I used to wholeheartedly agree with this take. The more I use these models to read/analyze long technical documents and get novel takes on complex design problems, the more I have to ask myself, "What really is intelligence, then?"
I suspect we're rapidly approaching a point where we will either have to admit that AI has some form of "true" intelligence, even if it does function differently from our own at times, or we will have to define intelligence so narrowly that few humans will make the cut as intelligent agents.
It’s failing to understand citations (not that sophisticated of a concept) at a very basic level. Why ever would you trust it with anything more complicated?
This isn’t an issue of “narrowly defining” intelligence.
>> “ I'm not sure it's true that LLMs don't have some level of base knowledge.”
I don't think you'll be happy to call this "intelligence" once they fix the problem where it gets citations wrong ~5-10% of the time. I think there's a more fundamental disagreement you have with the word usage and what it implies about the nature of the thing itself.
Humans have always made different types of mistakes than machines. You don't have to call LLMs "equivalent" or "superior" to human intelligence. Maybe you don't like the word "intelligence", because it sounds like it's doing something similar to what your brain is doing and that feels like it's bringing baggage into the discussion that doesn't need to be there.
Fine. But you have to call it *something*. To me, I gave it a complex prompt and it did in 5 minutes what it would have taken me a day or more to do. I spent 20 minutes reviewing the output and found/fixed a few minor errors. When I substitute a day of my own personal *intelligence* with a machine that could substitute 98% of that effort, we're going to have to call that intelligence-substitution machine something, even if sometimes it doesn't look like what normally define as "intelligence" in the form of getting the simple things wrong. (My wife would disagree about that, and claim I get easy things wrong all the time.)
But if you're not going to call it intelligence, you're going to have to find a better name quick, because whatever it is, I can use it to massively substitute for large amounts of intellectual effort from highly-trained subject matter experts, who only have to go in and correct minor mistakes (or ask the tool to make the corrections once they're found - or sometimes ask it to find its own errors, which it does!). That's not a theoretical concept. That's actual usage of the tool last week.
>To me, I gave it a complex prompt and it did in 5 minutes what it would have taken me a day or more to do. I spent 20 minutes reviewing the output and found/fixed a few minor errors.
I don't think that treating the problem with citations as if it were a slam-dunk disqualification from describing LLMs as displaying many aspects of intelligence is a good choice. As LLM-based AI systems stand this month, their intelligence is spiky (using human intelligence as the norm for comparison). To my mind the interesting question is when the various holes between the spikes get filled in, and with how much generality, and in how many steps.
Saw a really clear example of this a few days ago. I was googling for museum ship attendance figures and the Google AI summary described Midway as an "Iowa-class carrier". That's obviously incorrect (the Iowas are battleships, and Midway only shares engines with them) and it's the sort of thing that no person with more than vague knowledge of how WWII warships work would ever say. Maybe GPT-2 would have made that kind of mistake frequently, but probably nothing since then.
The best explanation I can think of is that it internally treats different portions of the output with different degrees of importance, and spends more focus on the ones that it thinks are important. Table of values? Important. References? Not important, just say whatever. The problem is that it doesn't always align with what we think is important, and the way it mixes high-effort and low-effort parts seems really alien to us. Obviously, people give ill-considered, off-the-cuff answers all the time. ("I guessed JQA as the 13th president because I knew he wasn't super-early and before Lincoln.) But it's usually in contexts where it's obvious that's what they're doing, and they rarely give you a meticulously researched first half and a second half that's completely off the top of their head.
That's not really how LLMs work, they're not assigning importance to different pieces of the output and changing effort spent. What's happening is that the Google AI summary is from a tiny model built with very little brainpower (or parameters) to make it cheap enough to run for every Google search. It's not able to perfectly recall everything, and so it wrote something plausible sounding but wrong.
Do you know if it's switching to using web search after you challenge it? My guess is that there's a slight penalty on using search (so the AI doesn't waste your time searching for everything), but it will do it if it has to.
I'm using Opus 4.7 "Adaptive". I don't know if it's doing a sequential search, then answer. It seems something like that is happening, as opposed to it going back and forth between search -> compose answer -> search -> compose answer. The original prompt included an ask for the specific references, so it should have been looking for that information in the initial search. Indeed, it built the response from the references themselves (including accurate numbers from the papers it subsequently failed to cite), so it's really weird that it then went on to make stuff up that it had clearly found in initial search.
I have a process where I specifically direct Claude to support its answers with websearch to enforce external sources and not just info from the training set.
I often see claims that AI has been making up academic references, and I wonder if it is copying these from rogue papers in which a human author has made them up! If it happens again, you should try asking it directly where it obtained the reference, although I wouldn't be surprised if it can't or won't give a straight answer.
That's an interesting hypothesis, but in one case I'm thinking of I know this isn't what happened. The reference it gave me was to a 1979 paper (a real paper, to be sure, but on a completely different topic not even tangentially related to the one I asked it to look up). The context was in a table, listing studies, findings (including sensitivity and specificity numbers that WERE correct for the citation it meant to give me!), along with references, most of which were correct, but when I went to check the references this one was clearly just the wrong reference. It wasn't even "off by one" number or something like that. It was just wrong all the way around (wrong PMID, author name didn't go to the PMID, year was wrong). But it DID have the right numbers in the table it generated, suggesting it knew of the correct study since it pulled the data out of that study. It just ... for lack of a better term, hallucinated that this was all from some other made-up study!
Oh, I'm sure it totally made up the reference. But the syntax/semantics distinction here isn't where the problem lay. It gave me a summary of the correct paper, complete with accurate numbers (I double-checked once I found the correct reference), but then tied that to the wrong citation. All while demonstrating a solid understanding of the complex clinical context the study was working from. Seems like it understands the semantics really well, but that citations are some other beast entirely that are for some reason more difficult to get right.
It's because citations are really short strings of text whose only purpose is to point to some document. It has no context and is ultimately arbitrary, so the probability distribution of possible citations looks pretty flat. Articles are really long strings of text where all the tokens are tightly interconnected, so it's easier for the LLM to tie them together.
In other words, exactly like it would be for me. If I don't know the authors personally, it's going to be more difficult for me to remember the citation because that's not what ultimately matters.
It's common in these parts to downplay the stochastic element of an LLM (because I guess it's more interesting, albeit less insightful and accurate, to not think of it that way), but I think it's probably related to a bad initial seed or stochastic parameters.
A bad initial seed could lead to a bad path where the correct answer isn't locally present, so it does a shameless guess. Trying again could give a better seed that leads it down the right answer basin and no guess is needed.
LLMs typically do not always choose the single most likely next token anyway; there are parameters like temperature and top p (hidden to most users) which allow for somewhat less likely tokens.
That all makes sense. What doesn't make sense is that I've seen a few instances now where it shouldn't be having to "guess", since it's pulling numbers and even direct quotations from the source itself.
Now, I get what you're saying that the "stochastic" nature of AI is that even those direct quotes were also guesses (even if each part of the quote's probability weight was > .99) about what the expected output should have been. What I don't understand is that if it can give me a direct quote from a paper with a high probability of success (because it made several accurate direct citations from that paper) how can it then go on to make several errors in a row (wrong author, wrong year, wrong PMID - both for the wrongly cited study and for the expected/correct study)?
How does the probability switch from almost universally accurate to almost universally wrong? Random chance doesn't seem to explain it. By analogy, if I flipped a coin 20 times in a row and it always came up heads, I'd think there's a very high probability it's a rigged coin, but there's still some chance I could observe those results randomly, mistaking noise for signal. But if I flip the same coin another 20 times in a row and get tails each time, I have no explanation for what's going on.
I think the coin analogy is telling; the tosses are independent, but the "path" the LLM goes down is not. If the LLM goes down a "just read from the PDF and don't rely on any training data" path, then it's going to execute that task and doesn't have to guess anything (okay it's still guessing, but with very high probabilities). But if it does down a "rely on training data" path, then it's going to "guess" if it doesn't have that specific data built in. It might be useful to think of the LLM as randomly going into a different mode of information sourcing. Conceivably this could be addressed by prompting it in such a way that encourages the former and discourages the latter, but that would have to be tested of course.
Here's the actual prompt I used [two minor details censored from ACX, sorry], with Claude Opus 4.7 Adaptive:
"Interesting. You started by challenging my prior that [XXXX] would help my co-primary endpoints, asserting instead that this may be the opposite of what is supported by the literature. Please put together an analysis of all current [XXXX] publications that reference [XXXX], assessing any measured differences. Treat sensitivity and specificity separately, with separate analyses of each. Provide references (including PMID) for each claim."
The output was a well-reasoned analysis of the literature and included summaries of the various publications (including a correct summary of the incorrectly cited publication). It also included a table with columns for Study/PMID/N/>1cm sensitivity/<1cm sensitivity/{delta}/Assessment. (Same for specificity.)
Note that most of the data in the table was accurate. It's truly impressive what the model can do well! If I had an aide who could put something like this together in a day - including if they made a few reference errors - they'd be a valuable addition to the team. Yet Claude did it all in a couple of minutes.
Partly, I'm trying to understand the way this works so I can get better/more accurate results in the future. Partly, I'm reading the various explanations for why it functions the way it does and asking if they make sense, based on what I'm seeing.
I get your take, that maybe it's focusing on the wrong thing in looking up and reading the PDFs, but then forgetting the references it just looked up. But I did ask for specific references to each claim, including PMID. It didn't hallucinate the analysis, just the details. Which is something I keep seeing. It rarely hallucinates crucial details necessary for understanding an idea, but it sometimes mixes up references, even when I ask it to focus on them.
My suggestion would be to do this in Claude Code or cowork and create a hook to automatically send the output to an agent (using haiku should be fine) to cross-check all references and report any anomalies.
I second this. The only way i've been able to get Claude to really use files/external data stores is in CC with markdown file structure and routing/index information in claude.md and elsewhere
I'm sure this has been addressed before, but my keyword searches aren't yielding results.
Of the Good, the True, and the Beautiful, it's pretty clear there are rationalist applications for the Good (EA for example) and the True (lesswrong for example). Is there a rationalist application of Beauty?
Check out David Deutsch's argument for Objective Beauty in one of the last chapters of The Beginning of Infinity. He puts forward the idea that Objective Beauty provides the clearest attractor for effective cross-species communication such as the communication between flowers and insects.
Read the Fun Theory Sequences. As long as Beauty can be specified in your utility function, and you are good enough at rationality to not screw things up by trying to apply it, rationality is the best way to achieve it.
I primarily handle music artists (everything from super-stars to up-and-coming talent), music producers, and influencers. Whenever a song is released, there's a lot of legal groundwork involved (i.e. drafting and negotiating producer agreements, writer agreements, and side artist agreements). I also handle the broader transactional side: recording agreements, publishing deals, and brand partnership agreements.
As a professional in the field, what is your stance on the current state of copyright law? I tend to think it benefits publishers at the expense of artists and has a chilling effect on creativity, but I've never tried to receive financial compensation for creative work so my perspective might be kind of warped.
speak for yourself. im what you call a “crypto-normie”: i codeswitch to match the vibe of the ACX meetup but put me in the ai safety hype house and i can go toe to toe with the other neurotypical zoomers
Yesterday's London Marathon was one of the most astonishing athletic feats in history, comparable to Roger Bannister's breaking of the four-minute-mile in 1954. Sebastian Sawe covered 26.2 miles in 1:59:30, smashing the previous record by 65 seconds. Yomif Kejelcha, a very talented distance runner making his marathon debut, ran 1:59:41. Jacob Kiplimo finished third, also breaking the previous world record. Sawe and Kejelcha are the first men ever to break two hours for the marathon in an official, record-eligible race (Eliud Kipchoge managed the feat a few years ago in an exhibition).
I realize I'm much more of a running nerd than the vast majority of people, but I find it slightly disappointing this isn't the top news on every single sports-related website. And it should be front-page news around the world. It really is an incredible achievement.
Uhm, in my news bubble this was the case: it was on the front pages of sport-related websites and it was prominently placed on the general news websites.
Of course, mostly focusing on the winner, but the articles also mentioned the tragic story of the second-placed runner.
That is pretty amazing. I’m sure the length will be checked. I don’t think a tail wind could help that much. Probably want to test the record breaker for PEDs. Really good PEDs.
What's interesting is that Sawe agreed to participate in extra tests (with the costs paid by Adidas) in the lead-up to the Berlin Marathon last year. Presumably you wouldn't agree to that if you were doping (unless you were extremely confident that you were using drugs that were totally undetectable).
There's always skepticism around these sorts of amazing performances, but if I had to guess one way or the other, I think he's clean.
How did three separate people manage to break it at once? Was there an unusual tailwind or did they accidentally make it slightly shorter or something?
The performance of top athletes in the same long-distance race are highly correlated to each other.
First of all, it is clear upfront for many important Marathon races over the year that they will not yield records. The race tracks differ, most obviously not all of them are flat. So if several runners improve a lot over one year, they may all have their personal record in the same race.
Second, there was a very strong pacemaker field in the first half (or two-thirds?) of the run. For the final performance, it matters a lot how well the pacemaking works.
Third, tactics: usually top runners try to stay in the front group for as long as possible. It's not like they are running the race at independent paces. If the front group happens to be fast, then all top runners who manage to keep the pace will have very good times, otherwise they will all have slow racing time.
If the course were short, we'd probably know about it by now.
The shoes are probably the biggest reason. Adidas (worn by the top two) came out with some extremely light shoes. Kiplimo was wearing Nikes. The biggest change was a few years ago, when the shoe companies, starting with Nike, came out with so-called "super shoes" featuring carbon fiber plates and very cushiony foams. Gradually the super shoes have gotten better and better.
The conditions were cool enough to be good, and the wind was somewhat beneficial. Record-eligible marathons must have a finishing point within 13.1 miles (as the crow flies) of the starting point. So the Boston Marathon, which was also very fast this year, is not eligible because it's more or less a straight line and some years see a very strong tailwind. But the London course this year had a net tailwind, and a tailwind at the end of the race. So that explains some of it.
I just read an interview with one of his coaches. He agreed that the new generation of shoes have notable impact, and that weather/course wind all need to play out well.
But he added an interesting other aspect that he called a game changer in the last years: that there was a strong shift from lipometabolism to carbohydrates even on such long tracks in the last years. Essentially people realized that if with proper training, the digestive system can digest much more carbohydrates than previously thought, even during a run. Sawe had started days ago drinking special carbodrinks to fill his depot, and continued to drink them before the run, and during the run starting from km 5.
I've definitely read about this as being a factor, but I don't know if there were any new changes to fueling strategies this year (the stereotype of eating a big pasta dinner the night before a race has been around for a while). From my understanding, the body can't effectively store enough glycogen to run a marathon. And while the body's fat supply is enormous, at higher intensity levels the body would much rather burn carbs. So you need to take in a lot of carbs during the race. But that's not totally new, so I don't know if Sawe and the others were doing anything new this year.
I've also read about carbs being a huge factor in improved cycling performances in recent years. So fueling is definitely very important, but I don't know if it was the major reason for yesterday's breakthrough.
Yes, it's not a big change in this year in particular. I rather think about it as a general improvement that makes all runners faster. If a new technique makes all runners 3 minutes faster in the last 3-5 years, then the 2:00:00 is not necessarily broken in the first year of this technique, since marathon times vary a lot from race to race.
I think this is not such a bad model. Before 2022, there was exactly one runner who ever beat 2:02:30, in 2019. But 2022-2025, the time has been beaten by 5 different runners, and the five times were actually pretty evenly distributed in the interval 2:00:30 - 2:02:30, without any particular order. Given this spread, it seemed just a matter of time until the 2:00 was beaten.
Of course, it is remarkable that two runners beat it in the same race, but then the times are correlated.
Best theory is that the newest supershoe that weights only 97 grams, which is almost half of the previous leading supershoes, combined with a competitive race until the last kilometer and good weather with a tailwind for the last 10 kilometers all contributed.
London isn’t even supposed to be the fastest major marathon, it has some elevation and sharper turns, so people are expecting this record to be broken again in the fall at the flatter Berlin or Chicago courses if the weather is cool.
Justinian I (dies in 565 AD) is generally considered to be a really super emperor. He ruled the Eastern half of the empire some years after the Western half had undergone a rapid unscheduled disassembly. Justinian is pretty successful putting it all back together again . He manages to get back ‘Africa’ and Italy and has time left over to build the magnificent Hagia Sophia which you can see to this day. And he has a scandalous biography written about him. For example we are told that before becoming his wife, Theodora had a nifty trick to entertain paying customers by having geese peck barley grains from her naked body.
Anyway we are not talking about that Justinian. We are talking about Justinian II who comes to the throne over a century later in 685, at the age of sixteen. This is a time when the Arabs are in full expansion mode so the episode starts with six year old Justinian peeping over the walls of Constantinople at a vast Arab army not unlike the scene at Helms Deep. Anyway the people of Constantinople have a secret weapon - this is Greek Fire which is a kind of napalm you can shoot out of a tube and burn up the enemy fleet. Which they do, so eventually the Arabs have to give up and go home. (But they’ll be back. Repeatedly).
When Justinian II takes over a lot of the old empire of Justinian I is gone. Even Greece isn’t safe these days having been largely taken over by Slavs. (Not to be confused with the Bulgars who take over other bits of the Balkans).
Anyway our Justinian reigns with the massive over confidence of a teenager. It is not fair to say that everything he does goes wrong though kidnapping the Pope goes less well than he’d hoped. Actually he does pretty well for a bit but some shocking misjudgements see him in dragged to the Hippodrome where a man with a pair of pliers sets about cutting out his tongue and cutting off his nose.
Then in a misjudgement every bit as big as anything Justinian did, instead of putting him in a sack and throwing him in the Bosphorus they exile him to the Crimea. I mean everyone knows an emperor has to be bodily intact so there is no chance he is coming back is there? Is there??
There is. Justinian spends his time in the Crimea weaselling his way into the good books of the Khazars who are sort of Turkic nomads. He even goes as far as to marry the Khagan’s sister. When Byzantine bribery causes his brother in law to turn murderous he murders his would be murderers and escapes to the Bulgars and teams up with them. So with his new Bulgar army (following him largely out of curiosity) it is off to Constantinople to wreak vengeance on his foes. The only problem being that when they get there nobody wants to let him in (not least because of all these ferocious Bulgars looking on with interest).
Anyway for those of you who don’t have too many podcasts already just google Subject to Change with Russell Hogg or it is right here.
I did rather wonder how he survives his nose being cut off. I thought people in the past died of the merest scratch though surely that can’t be right? But Justinian II isn’t the only emperor to be mutilated like this and they all seem to more or less thrive on it. Also in the novel by Turteltaub which we discuss on the podcast, Justinian has a new nose (sort of) given to him by a visiting Indian surgeon. There is nothing in the sources to say this happened to Justinian but supposedly these kind of techniques were practiced in India at the time.
(And full credit to Professor David Parnell who is not just an excellent guest but game enough to participate in a political correctness quiz to round off episode 2.)
I only read a couple apart from Justinian. From memory The Gryphon’s Skull and Over the Wine Dark Sea. I am probably being unfair and maybe when he writes as Turtledove he is better. And he is so prolific I suspect there must be quite a variation in quality. Which of his do you think are the best? I feel guilty about being so rude so I should give them a try.
I will give it a go! I have a few favourite 'ancient world' novels but generally I think they are hard to pull off. I liked A God Strolling in the Cool of the Evening. Elodie Harper's Wolf Den trilogy was pretty good and Philip Matyszak's Gold of Tolosa is great fun.
Just want to say I have listened to a few episodes of your podcast and I enjoyed them. I think having some economic historian on might be a good balance for the narrative history you normally do.
Thank you for listening. Do you have an idea who might be good and what topic? I'm very open to ideas here! I did have David Friedman on ages ago and he was great but I am guessing this isn't the kind of thing you mean. Legal Systems Different from Our Own which Scott had fun with.
Joseph Henrich writes about the type of stuff I find fascinating. I have not read her stuff but Ada Palmer had a good podcast with Dwarkesh. So maybe people that they cite / cite them.
“Some who have been in Justinian's company in the palace very late at night, men with a clear conscience, have thought that in his place they have beheld a strange and devilish form. One of them said that Justinian suddenly arose from his royal throne and walked about (although, indeed, he never could sit still for long), and that at that moment his head disappeared, while the rest of his body still seemed to move to and fro. The man who beheld this stood trembling and troubled in mind, not knowing how to believe his eyes. Afterwards the head joined the body again, and united itself to the parts from which it had so strangely been severed.” — Procopius
Is entertainment really the point of life? Have we fallen that far? Why can't we just use technology to provide material abundance, cure pathology, learn/discover/develop, and have good relationships? If that's not satisfying enough, maybe fix the anhedonia rather than lean harder into the stimulus.
There's a case to be made that, in the end, it all comes down to human happiness. The catch is that happiness turns out to have diverse requirements. Many of them are outside of hedonism.
I think if an indistiguishable simultion existed, most people would choose to live in it if feasible. A lot more than would admit to it in a world where it doesn't exist.
VR would allow things that are impossible, too dangerous, or very expensive in reality. That sounds exciting. But I would also like to stay in contact with the top (?) reality.
(When I talk about expensive, of course have I no idea what the actual costs will be and what will the economy be like those days, but I kinda assume that VR will have more or less the same costs regardless of *what* it shows... so sitting in a virtual office or flying through space will cost approximately the same, so there is no reason to not do the latter.)
Well the issue is I have a child I want to keep contact with - and in the physical world. Otherwise if my body is kept healthy then yes. But I might still want to get out and "touch grass" sometimes.
It's a minor point in Blindsight by Peter Watts. For unrelated reasons, you may also be interested in Ra by qntm (https://qntm.org/ra).
Personally, I would avoid it unless there is some health/safety reason to live in VR instead of IRL. I'm paranoid about any tech that could harm me if programmed wrong (brain implants) or that leaves my physical body defenseless for long periods of time.
I think there are a lot of elderly people for whom virtual reality would be a huge improvement over their current best available activity (sitting and watching tv all day).
It would get just as boring as any other game, and still has all the constraints videogames do now, in that you have to be able to pay the electric bill and feed yourself.
No, games do not come tailored to personal preferences. Source: I have like 100 Skyrim mods installed, and 20 of those from loverslab... it is kind of how I would like to live, but still a very basic-level simulation of that. I think the proposal was basically perfectly tailored, perfectly immersive simulation. Like if you want to be the king of England in 1500, your experience is indistinguishable from as if you would be.
I made my main argument below*; the charm has a ceiling, because you're going to know it's fake, and designed to let you win. Is your perfect simulation of the King of England going to allow you to be overthrown by a peasant army because your policies were too severe? Probably not, which means your experience is going to be hollow.
*(I assume it's still there, they seem to have blocked me.)
Don't we have an answer to that from current videogames? You (after some practice) want to play Europa Universalis IV in Ironman mode, because only that gives you achievements. Everything else is basically training wheels. Ironman has no saves in the usual sense - if you lose, you lose. However, you can still start it again from the beginning.
And these achievements matter! I am not an avid gamer by any means, but I am kinda proud I have the "unite Italy" achievement.
If you have Matrix-like technology you should also be able to entirely eliminate boredom or any concerns over "this isn't actually real". Nanobots will rewire your neural circuits and you'll be a happy camper.
Lobotomy is the dull version of this. The VR version would be able to keep 100% of your personality, except that you never feel bored. You never see this scenario in sci-fi because it's extremely boring (ha) to readers, but that doesn't mean it won't be possible in real life. See: https://nsokolsky.substack.com/p/beware-the-science-fiction-bias-in
That is also a sci-fi trope, because "give people what they want" doesn't work, there always has to be some sort of a "gotcha" involved. But real life is not sci-fi: I predict that if some sort of a benevolent AI does create VR simulations for us, it would do so in a way that doesn't involve the "Be Careful What You Wish For" trope.
Two simulations is even worse than one. It's like playing Virtua Fighter in a Yakuza game; if you wanted to play Virtua Fighter instead of Yakuza you could just go play Virtua Fighter.
If we're assuming post-AGI technology, we might as well assume enough other benefits that escaping the real world becomes pointless. And I still hold it would have no advantage over lucid dreaming, and no meaningful advantage over existing videogames.
Fully realistic is weak. We should aim for much higher perceptual resolution than unmodified human senses in meatspace. The web serial titled Seek (by Wildbow, the author of Worm) explores this.
I keep getting ads for this website. Pretty sure it's made by an agentic AI gone wrong. Anyway, thought people here might find it interesting:
https://gijane.com/
Can you say why it would be interesting to ACX readers, or otherwise distinguish this comment from any garden-variety click-farming scam?
Yeah, I realize it looks that way, all i can say is, it's not, at least not from me. If it is, you're welcome to alert Scott and have him ban me from posting comments for a bit. I'd tell you more, but I can't even properly paraphrase what's going on in it - it's got the AI dream-like thing where everything makes sense as you skim it, until you stop to think about what you just read. Or i just don't know enough about orbifolds.
Mine was an oblique nudge for you to describe the link contents in some way that goes beyond some form-letter comment that could go anywhere, like "Interesting point! I'll check it out, thanks! www. buy-my-cheap-shoes .com".
If you can't paraphrase, at least try. As you just did.
That said: yeah, it's weird, I don't know what it's trying to do, other than advertise some sort of AI-based tool in some way that still doesn't make me want to use it or even look deeper.
Yeah, it was a fair comment. I don't think it's actually selling anything though
200 pages into Resurrection (first read), I have a feeling Count Leo is going to explain determinism to me.
Tipoff paragraphs
“What astonished him most was that Katusha was not ashamed of her position — not the position of a prisoner (she was ashamed of that), but her position as a prostitute. She seemed satisfied, even proud of it. And, yet, how could it be otherwise? Everybody, in order to be able to act, has to consider his occupation important and good.”
“It is usually imagined that a thief, a murderer, a spy, a prostitute, acknowledging his or her profession as evil, is ashamed of it. But the contrary is true. People whom fate and their sin-mistakes have placed in a certain position, however false that position may be, form a view of life in general which makes their position seem good and admissible. In order to keep up their view of life, these people instinctively keep to the circle of those people who share their views of life and their own place in it. This surprises us, where the persons concerned are thieves, bragging about their dexterity, prostitutes vaunting their depravity, or murderers boasting of their cruelty. This surprises us only because the circle, the atmosphere in which these people live, is limited, and we are outside it. But can we not observe the same phenomenon when the rich boast of their wealth, i.e., robbery; the commanders in the army pride themselves on victories, i.e., murder; and those in high places vaunt their power, i.e., violence? We do not see the perversion in the views of life held by these people, only because the circle formed by them is more extensive, and we ourselves are moving inside of it.”
Lumina Priobiotic has removed their online store. Gone for good?
Funny thing happened to me -- I just got banned from Reddit.
Okay, that's not so funny. The funny thing was the explanation: I was banned because the censorship algorithm detected that I posted some kind of objectionable content in a comment. Helpfully, there was a link to my comment. I clicked the link, and... it displayed the words "[Removed by Reddit]". Thanks, that explains a lot.
Well, at least I see the name of the topic. Yeah, I vaguely remember reading that... but that was 24 hours ago; that's like an entire week in dog years. Apparently I wrote a comment, but I don't remember what. I write many comments online.
Then I read the comment thread again, especially the ones I replied to, and finally I remember. Yeah, okay, maybe I deserved the ban. But it's still funny...
The top-level comment was like: "We should do X" (where I admit that X is an objectionable thing by the sissy Reddit standards). The reply was: "No, we should do Y instead" (where Y is another objectionable thing). And I remember replying: "Hey guys, no need to argue, we could do both X and Y." And... that was the comment that cost me my Reddit account.
The previous two comments are still there, uncensored. Apparently, it was putting those both ideas in one sentence that crossed the threshold. I find it hilarious.
https://imgflip.com/memetemplate/420358091/Genie-Youre-Free
:D
Sadly no, it was just a 3-days ban. But it was liberating for a while; the first day I kept reopening Reddit, but the following two days I simply forgot about it.
Saturday May 2, 2026 The Onion:
Rita Ora’s Agent Scores Singer Another Prime Crossword Puzzle Placement
https://theonion.com/rita-oras-agent-scores-singer-another-prime-crossword-puzzle-placement/
NYT Saturday May 2, 2026 XWord clue for 7 Down: “How We Do (Party) singer Rita - 3 letters
Yoko One does pretty well too. As a long time N Y Giants fan of course I always knew about the Hall of Famer Mel Ott.
I guess I'm not alone in only knowing her name because of NYT XWord if it made to the Onion. I asked ChatGPT and she has 6 appearances in the last year. BTW article says April 30th, not May 2nd (much funnier, unless they backdated that)
Will Shortz has a lot of candidate puzzles to choose from. I’m not sure this ORA occurrence today was coincidence.
I had seen The Onion article last night. At least in today’s puzzle it wasn’t crossed by “Boko Haram.”
https://www.mod171.com/p/real-hampshire-college-has-never
The Capture (BBC) is very entertaining and the latest series involves an AI battle computer, Simon. Writing stories involving AI doesn't strike me as easy and the series consequently isn't perfect. This is my review. The blog is theistic (amongst other things) but the review is not. https://zanzibar142106.substack.com/p/the-capture-storytelling-in-the-time?utm_source=share&utm_medium=android&r=e8145
The discussion of "red button / blue button" on this thread made me so angry I wrote an essay, and I feel I deserve a top level comment for my effort:
https://againwithapen.substack.com/p/riddles-are-fiction-actually
Thank you very much!
In your view, what kind of speech act should I perform if I'm curious about what a particular person would do if the button situation actually happened in the real world?
Good question. It has many answers.
The answer I am guessing will be most satisfying to you and that I believe I implicitly give in my essay is that you amend the fictional scenario by whatever supposed real life aspect you are interested in.
For example:
Everyone in the world has to take a private vote by pressing a red or blue button. If more than 50% of people press the blue button, everyone survives. If less than 50% of people press the blue button, only people who pressed the red button survive.
_ You know for fact that more than x% of all people are going to press the blue button no matter what. You do not know how much more._
Which button would you press?
This is a much more interesting problem and I am thankful that you made me think of it.
>You know for fact that more than x% of all people are going to press the blue button no matter what. You do not know how much more
Good idea, but I might propose instead:
"There is a counter showing that x% of the entire population has already pressed the blue button. You don't know how many people haven't pressed either button yet, so the final percentage might be anywhere between x% and 100%."
I suggest this because a common thread I've noticed in Newcomb's Paradox debate is that many two-boxers don't accept the premise that the Decider is a perfect or near-perfect predictor of whether you will one-box or two-box. So an accomplished fact seems more likely to be accepted than a prediction even if the prediction is framed as infallible.
Good point.
I still think you need to switch the colors for the argument to be worth having.
("There are two buttons; one has a Donkey on it, and the other has an Elephant.")
I'd suggest Green and Purple, but then the question won't work for Drazi.
Yeah, but don't we kind of want the Drazi to mess this one up and kill off 50% minus epsilon of themselves?
I like it!
In response to the stabbing down my road yesterday: https://aw694.substack.com/p/no-this-tragedy-is-not-a-stick-you
(First time writing a post, inspired by Scott's article on Inkhaven. Wasn't planning to write anything till the bit about writing things that will annoy people when I thought - I do want to tell everyone else who is hurt why they are Doing It Wrong)
Link is broken. https://substack.com/home/post/p-196044282 works for me.
Youtube video recommendations (some turned up by happy accident).
(1) Baroque music:
https://www.youtube.com/watch?v=PptQJv4wxdg&list=RDPptQJv4wxdg&start_radio=1
Love me some theorbo:
https://www.youtube.com/watch?v=NjUEAvgZqSE&list=RDPptQJv4wxdg&index=45
(2) Cooking! I'm never going to do any of their recipes, but this one for broccoli is easy enough even for my level:
https://www.youtube.com/shorts/hCDaDqal8SE
Two types of caramel:
https://www.youtube.com/shorts/mBGA8HN_vMs
Historical recipes:
https://www.youtube.com/watch?v=4a-nf2T04Hs
What signals that a piece of writing is AI generated?
Let's say you're not familiar with the larger corpus of someone's essays/tweets/comments.
Wikipedians have made a pretty comprehensive-looking overview
https://en.wikipedia.org/wiki/Wikipedia:Signs_of_AI_writing
See also https://tropes.fyi
Welp. There goes my morning.
Thanks!
Wow that is LOOOONG.
I you see an em dash in written text it may be generated with AI — or someone is banging their thumb on an iPhone to indicate a pause rather than using a comma. It’s really hard to tell sometimes.
Here on Linux/KDE I can just press AltGr+Minus for an en dash, and Shift+AltGr+Minus for an em dash. No biggie.
You should learn how to type them as well. AI models are constantly being improved, and the use of em dashes will soon be trained out of them. The suspicious lack of sophisticated punctuation will then be considered an indication of AI-generated text.
By default iOS is set up to render two consecutive hyphens as an em dash.
Messages app on my desktop Mac also does this -- but it doesn't work here.
Goes on forever but never gets to the point. Paragraphs that repeat the same phrase, just worded differently. Always says that it is now going to tell you about A, B or C but puts that off until the very end, and then it's some boilerplate "well there are points in favour of A and points against" conclusion.
So like a diligent college student who has no idea what they're talking about?
Very much so. Lots of boilerplate structure and stock phrases, and lots of keywords associated with the topic, but the actual content is vague and handwavey. It also reminds me of corporate/marketing speak in this respect.
That made me realize I probably read too much corpslop so I can't tell if I'm dealing with AI or someone from marketing.
Except Ivan. Ivan used to sound like a James Bond villain, now he sounds like an excited 24yo brand designer.
My condolences for having to read so much corpslop, and especially for the loss of Ivan's Bond villain voice.
I'd imagine so, but even more irritating.
Here's a fun example of Democrats/Liberals/"the Left" rewriting history in real-time. The term "to 86" has long been well-known slang for killing, with documentation attesting to this usage from at least as early as 2004. https://www.urbandictionary.com/define.php?term=86%27d. (Probably much older: Cassell's Dictionary of Slang lists it as in use from the 1970s+. https://www.google.com/books/edition/Cassell_s_Dictionary_of_Slang/5GpLcC4a5fAC?hl=en&gbpv=1&bsq=%22eighty-six%22)
But you can see the media pretending Trump just imagined this last year after Comey posted the message "86 47".
Do you have an example to point to of "the media pretending Trump just imagined this last year?" I agree with others that, while 86 may be slang for killing, I'm much more familiar with it in the context of "get rid of" like "86 that potato salad, it's been sitting out all day."
I haven’t heard anyone else say that but I’m not a regular Fox News viewer. Easy to imagine Greg Gutfield making an unsubstantiated assertion like that.
yeah.....very common on movie sets. I heard it all the time.
Everybody who isn't a partisan hack or a complete ignoramus has always known that "86" has a broad range of meanings, of which "kill" is an edge case that applies only in certain contexts. The only rewriting of history *here* is a bunch of partisan hack MAGA Trumpists trying to retcon the phrase into having always and only meant "kill" so they can falsely accuse one of their enemies of threatening to kill Donald Trump.
Democrats, Liberals, and "the Left" have each also attempted to rewrite history in *other* contexts, but here they're the ones playing it straight. Also, those are three different groups and attempting to blur them together is another attempt to rewrite history by partisan hack MAGA Trumpists. Really, you all are making it hard to remember that the Democrats, the Liberals, and the Leftists sometimes do the same thing, which is unfortunate because they do and that's a problem. But it's not a problem we are going to address today, apparently, because we have to deal with this "86" nonsense instead.
It's also worth pointing out that even if someone *had* literally tweeted "Kill Trump", it still wouldn't merit prosecution. The bar for the 1A is really high.
1. Did the FBI bring charges against any Republicans who posted "86 46" during the Biden years, such as Jack Posobiec? (https://xcancel.com/JackPosobiec/status/1487642601536864256)
2. Do you believe that the FBI *should* have brought charges against Republicans who made a statement that, according to you, was calling for the murder of President Biden?
3. Is it possible that there is more to deciding whether a statement is proof of intent to kill besides looking up the word in a dictionary and seeing if it could be used that way?
Like, you talk about rewriting history and then completely skip over how that phrase has been used for literally the previous president.
There were, in fact, three assassination attempts on Joe Biden. None of them got very close (which is why nobody remembers them), but they did happen:
https://en.wikipedia.org/wiki/List_of_United_States_presidential_assassination_attempts_and_plots#Joe_Biden
Obviously, I don't believe any of these were caused by someone tweeting "86 46," but if you honestly believe that Comey's tweet incited one of the assassination attempts on Trump, then you should equally believe the same for Biden.
More realistically, neither of these statements had anything to do with the existence of crazy people who want to kill the president, which is why the legal standard for incitement is "imminent lawless action" and not "there is at least one crazy person somewhere in the US who might see this as a call for murder."
(Also, they aren't even charging Comey with incitement, they're charging him with threatening the President. Which is equally ridiculous but in a different way than what you're arguing.)
That’s one of the meanings as given in Cassel’s dictionary of slang.
The Merriam-Webster dictionary defines the term as to "refuse to serve (a customer)", or to "get rid of" or "throw out" someone or something.[7] The Oxford English Dictionary (OED) says it may be used as a noun or verb.[1] As a noun, "In restaurants and bars, an expression indicating that the supply of an item is exhausted, or that a customer is not to be served; also, a customer to be refused service. Also transferred."[1] As a transitive verb derived from the noun, it means "to eject or debar (a person) from premises; to reject or abandon".[1] The OED gives examples of usage from 1933 to 1981;[1] for example, in the 1972 film The Candidate, a media adviser says to Robert Redford's character, "OK, now, for starters, we got to cut your hair and eighty-six the sideburns".[1]
Comey meaning it in Cassel Dictionary of Slang sense is about as realistic as “The Americans” straight arrow, boyscout Sam Beeman suggesting someone whack president Reagan.
We get it, trollers gonna troll. Give it a rest already.
*Stan. (I love that show! Frank Gaad was the FBI director though, not Beeman.)
No, of course he didn't. It's a bullshit charge that probably won't stick.
My point is only that that usage has long existed, despite widespread and influential attempts to pretend otherwise. It reminds me of the time Amy Barrett used the term "sexual preference," and then Merriam-Webster updated the dictionary immediately afterwards to say it was "offensive" so lying press could attack her for it.
FWIW, I looked this up recently because I heard somebody else bring up the Merriam-Webster "sexual preference" dust-up. the Editor of MW explained to Newsweek in the aftermath that they are regularly updating definitions of words but only release these updates in scheduled batches. But if one of those words with updated definitions in the queue is getting increased attention, they will release the update early so that all the people looking it up will have the most up to date entry to read. This makes sense to me and lowered the probability in my eyes that Merriam Webster was putting its thumb on the scales to embarrass Barrett.
https://www.newsweek.com/amy-coney-barrett-preference-definition-1539088
...so their defense is they have a schedule for updates but they'll break it when something gets popular to try to control the meaning of trends. That's not a defense, that's what people are accusing them of doing.
The underlying linguistic claim isn’t really controversial among linguists: “sexual preference” had genuinely been falling out of favor in LGBTQ advocacy and style guides before the Barrett hearing.
Some Democratic questioner was a dick in the Barrett’s hearing for her not being ‘read in’ so to speak on the latest connotations of ‘sexual preference’. She was not at all trying to be offensive. To my eye she seems like a model of decency.
Barrett later apologized for causing offense.
“I certainly didn’t mean, and would never mean, to use a term that would cause any offense in the LGBTQ community,” she said. “So if I did, I greatly apologize for that.”
Merriam Webster people noticed a spike in searches for the term during the foofarah so they released the latest sense of the meaning.
I raise my eyebrows at this but I don’t see a sinister cabal of leftist activism in it. I think Fox News presented as such because their business model requires ginning up rage against libtards.
I would encourage you to consider whether you would have found this explanation remotely plausible if you weren't strongly inclined to believe in such organizations being less than blatantly partisan, and were looking for reasons, however tenuous, to avoid the conclusion that they were.
Sorry, I'll be useful and provide more of my thought.
I had recently listened this podcast interview (https://thedispatch.com/podcast/remnant/akvavit-zebroid-and-haole-interview-stefan-fatsis/) of Stefan Fatsis who wrote a recent book after embedding with Merriam-Webster as a lexicographer-in-training. During the interview, Goldberg brought up the "sexual preference" fracas and Fatsis seemed unfamiliar with it but found it completely implausible that MW would have just changed the definition on the fly for partisan reason given that those updates take months and months of work and review before being approved. That was when I went back and looked and found the newsweek article. So now I have an explanation from the MW editor saying that it was an impromptu release of an already reviewed and approved update, and I have a guy who was embedded in MW saying that it sounds implausible for them to have made a partisan change like that so abruptly. I may be somewhat inclined to believe that such organizations are not so blatantly partisan, but I feel like there's some pretty good evidence here.
> changed the definition on the fly
They just tacked on "offensive" to the definition they already had.
I have considered that, and I would likewise encourage you to consider the opposite.
Does MW publish a comprehensive list of these unscheduled revisions, or do we just have to take their word for it that they're doing so in an unbiased fashion?
Not so far as I can tell. You would have to, like in many other situations, take somebody's word for it.
>It's a bullshit charge
If we all know that it's a bullshit charge and that Comey was clearly not using it to mean "kill," then in what sense is it misleading for the media to report that "86" means "get rid of" rather than "kill"? Would it be more informative to anyone if the news articles said "well technically, if Comey was a 1970s mobster he would have been making a death threat, but as he isn't, he was not?"
Additional question if you think it was misleading or leaving out important context - do you believe you are being similarly misleading by only including the "kill" definition in your OP, and not the other definitions or the fact that you were not actually confused by the omission?
The term is slang with plenty of meanings in common usage. If I had instead said that "to get rid of" isn't one of them, that would have been analogous, sure.
Yeah, John Boy Walton was the director. Killed in Vietnam?, Cambodia? by those blundering Soviets.
Thailand
Yeah, The Americans was indeed great series.
I saw an argument a while ago that the Voting Rights Act's requirement to have majority-minority districts (the bit that's just been struck down by the Supreme Court) actually HELPED the Republicans by allowing them to "pack" Democrats into those districts without running afoul of the 14th/15th Amendments.
Prima facie, this struck me as implausible, and that seems be borne out by the reactions to the Louisiana v. Callais. What do you think? Am I missing something?
The gerrymandering cases I've read usually talked about "compactness" as a marker for drawing up fair districts; with a requirement to put minority density above geographic compactness, I think it creates more plausible deniability for gerrymandering. But I've got no statistics on it.
The Christian EA book has arrived in my Kindle, and I will read and review it when I feel less like I've been bashed over the head with a concrete block.
I think I need a tonic or something, this past week I feel rundown. Mr. Brain is not wanting to do anything except sit around and lollygag, so not in the state to do the book justice. Though Mr. Brain retains enough energy to be snarky about the cover, which I dislike. Memo to self: do not judge a book by its cover.
Have you tried one of those Covid home tests? In my first and only experience so far extreme fatigue was the main symptom. Pretty much laid in bed for 2 weeks.
But judging books is what covers are *for*. Nobody would bother with anything more than a plain binding with tite+author unless they thought the fancier cover would secure a more favorable judgement from their target audience, and they get repeat customers only if the cover-based judgements turn out to be mostly correct.
I have good memories of someone buying me a book whose cover was just the title "Y is for Yesterday," and trying to guess what the book would be about. But, probably would have ignored it if I saw it on a shelf.
Paul McCartney bio?
It actually ended up being part 25 of a mystery novel series, the first one being "A is for Alibi", and so on.
...I just discovered the twenty-fourth book is just called "X". X is not for anything.
Book covers are clearly an example of the Prisoner's Dilemma, and worthy of heavy-handed regulation.
Here's an interesting progression of ASoIaF covers: https://www.reddit.com/r/asoiaf/comments/cw3oo7/spoilers_main_asoiaf_original_book_covers/
You can see there's clearly two types of covers: the ones that say "This is a Fantasy Book, you should buy it if you're the kind of person who likes fantasy books", and the ones that say "No no, this is an actual respectable book that you can read on the train without people thinking you're a nerd".
Harry Potter also got the "plain respectable covers" treatment.
As a kid I remember being interested in reading Terry Pratchett books, but I thought I was too old and sophisticated (being, like, twelve) for the cartoony pictures that I saw on the covers. I wish they'd had a sophisticated-looking edition.
And I still haven't read them, which is a shame because they seem like the kind of thing I'd like.
It just occured to me that when you secularize Catholicism, heaven, purgatory, hell, you get morally good, neutral and bad. When you secularize Calvinism, elect and damned, you get good and bad, and no neutral.
This explains some crazy American ideas that people fret over benefitting from privilege, even when they did nothing bad. Just enjoying whatever advantages you were born with, without personally doing bad things is the perfect example of moral neutrality.
So I think we must rethink the whole concept of virtue signalling. Maybe it is honest. Maybe there are ex-Calvinists really feeling like that if they are not saints, they are evil, because they see nothing in between.
I am very comfortable being morally neutral as an ex-Catholic. My religious ancestors probably thought most people spend some time in purgatory and they themselves will too. They really understood imperfection, and that that is okay.
fwiw, the "puritan -> harvard -> cancel-culture witchhunt" pipeline is a recurring theme of moldbug's. Though he doesn't really discuss the other U.S. calvinists, as much. E.g. here [0], he appeals to evo psych, viz. self-deception.
[0] https://graymirror.substack.com/p/the-origin-of-woke-a-george-mason
I think the effect you're seeing is much more rooted in culture than theology. The Progressive Left does have a lot of cultural descent from the New England Puritans, who were theologically Calvinist, but the Social Conservative Right is descended from a different group of Calvinists: Presbyterians from the English/Scottish border country who settled in the western hinterland of the 13 colonies in the 18th century. Similar theology, but very different cultural substrate that took the theology in very different directions.
Now I’m thinking of Faulkner’s Snopes family. I don’t think Faulkner explicitly traces their lineage but they have a stereotypical Appalachian Scott-Irish vibe.
So in trying to find a way to blame a generally unwoke group (protestant christians) for what amount to their political enemies (the secular woke), I think you've probably stretched a little here. So where you present something like:
1. Protestants think that there are "elect" who are good and "Damned" that are bad
2. Because of this they are obsessed with being "perfect"
3. So much so that people who leave the church and become woke secularists and can't get rid of it.
Except that's not at all how the theology works; in Christianity, everyone is bad, equally so, and needs redemption as a result. Doubly so in Calvinism, which you namedrop here, where you don't really have a choice to accept OR reject salvation.
So if you rewrite the entire religion down to it's most fundamental tenets, yeah, you can THEN use your argument to stretch and blame protestants for an almost wholly secular movement from their almost wholly political outgroup opposites. It's still a stretch (you have to pretend the driving forces behind the woke are all ex-protestants who are so indoctrinated they can't get away from their former dogma, which is unclear.)
But if you *can't* or *don't* rewrite the really basic-level theology of the actual group you want to blame the woke on, then it doesn't work so well because you fundamentally had to misrepresent them to even get to the point where you tried your (still substantial) stretch of an argument.
A Catholic can chime in here on their side, but given how little effort went to your protestant representations I'm suspicious you might be misrepresenting how Cath theology works, too.
I think the crux is that you're treating the doctrine as directly equivalent to the behavior. Yes, Christian theology says everyone sins. But the community is not always reducible to explicit theology. E.g. Christmas trees ostensibly have nothing to do with the Levant.
In practice, my impression is that Calvinists kept an eye out for signs of grace (and/or signs of reprobation), which is where the de-facto moral-binary shows up. E.g. the Puritans kept an eye out for "fruits of the spirit". With the Salem Witch Trials probably being the most famous endpoint of this wariness. So yeah, doctrinally, Calvinists don't get a say in whether they're saved or damned. But the vigilance regarding acausal *signals* produced a certain incentive gradient. It's basically a Newcomb's Problem.
Look, I wrote about secularized religion, that is, when religion turns into culture. It is possible for political opponents to share the same culture, like when the Soviets said the problem with French Communists is that they are French first and Communist second, so they shared the culture of their opponents like De Gaulle. And culture is not blame, it is what it is.
And yes, secularized religion "distorts" theology - rather it does not have theology.
And it was none other than the great mythologist Joseph Campbell who proposed how ex-religious cultures work, that ex-jews are messianistic, ex-caths drift towards universalising mysticism, and ex-prots towards individual salvation, be that libertarianism or the personal is the political kind of wokeness.
This is not blame. This is culture. I noticed Calvinist roots earlier in wokeness, for example, that there is no forgiveness, no absolution, because one sin proves you are not one of the elect so you cannot possibly redeem yourself.
The idea that prots have less of a moral gray zone than caths was proposed by Max Weber in Protestant Ethics - that there is no such a thing as a cycle of sin, repentance, atonement, sin again. Weber was explicit that the Protestant lives a more methodical life, where every action could be a sign of damnation or salvation, leading to a more rigid moral view.
Pascal criticized "Jesuit casuistry" creating moral gray zones (he was a Jansenist, which means a Catholic with a Protestant attitude)
Adam B. Cohen found that protestants see sinful thoughts almost as bad as sinful actions, while catholics and jews not.
But why must I even argue this? Isn't this obvious if a movement starts out as Puritan, it is likely to remain morally rigid?
More interesting stuff. I have seen how the very same liberal goals got different justifications in different cultures. Like for example they agree that homophobia is bad, but they disagree why. In the cath type of culture like France, it is because you are judging people for their sex lives, and you should be unjudgemental, and ultimately it means you should not care about morality at all. I think in many places in the US also UK it is the opposite, there is an explicit moral judgement that homophobia is harming people, hence you should stop. These two different approaches I have noticed long ago.
I agree with everything you wrote - it's a fairly shallow and uninformed OP. In Catholic belief, purgatory is a waiting room, come the last judgement you get voted up or you get voted down. The Verdi requiem is not really practical as a funeral mass, but it sums up the Catholic visions of heaven and hell reasonably well.
I think a more useful distinction is that Catholics believe the good works have inherent worth and count in your favour come judgement day. Most branches of Protestantism treat this differently: 'sola fidei' is Luther's credo that only belief and faith (internal states) are necessary for salvation. Your actions or standing in society are irrelevant, though standing with God is sometimes reflected in wealth in this world. My idea of the ultimate irony, which I would one day like to experience, is to see a prosperity gospel preacher give his take on Jesus with the moneylenders.
https://www.astralcodexten.com/p/open-thread-431/comment/251431902
" In Catholic belief, purgatory is a waiting room, come the last judgement you get voted up or you get voted down."
Ahem. Not just no, but hell no!
That view of Purgatory is the pop culture one, often conflated with Limbo, and I'd like to beat with a big stick every person who promulgated it.
There is no voting up or down. The souls in Purgatory are the blessed dead, who are destined for Heaven, but have first to undergo that period of cleansing where the penance for sin they did not or could not do on Earth is performed.
It's not a second chance, it's not a "get out of Hell free" card, it's not Limbo. In very traditional belief, the *pains* of Purgatory might have been similar to the pains of Hell, but in Purgatory there is hope because you know you are saved. The souls in Purgatory can intercede for us, the damned souls cannot.
This is why old holy cards would often depict the Holy Souls as in flames, or in chains, *but* attended by angels:
https://personalizedholycards.com/images/products/4185.jpg
I can't get too mad since modern (since Vatican II at least, but often the lay attitudes and ignorance predated same) Catholic teaching of the faith is poor to terrible and people have no idea of what they are supposed to believe, but I tried my hand at explaining Purgatory to Protestants online once and I'm inclined to develop a twitch in my eye when people get it wrong on the Internet 😁
EDIT: I'll spit on me fist and get into good works/works righteousness versus sola fide another day, I also had a go at that when discussing the spiritual and corporal works of mercy for said Protestants, who were very courteous and interested, but that's a lecture for another day.
OK - waiting room is an oversimplification, I agree. My primary school catechism was a while ago. But I am clear on the distinction between purgatory and limbo.
Surely there still has to be a Judgement day and the possibility of going down? I went to Mass for Easter a while back, and they've changed the responses (again) but they/we still believe in the resurrection of the body and the second coming and all of that?
Right, I see what you're getting at. I was confusing your point with the general impression that Purgatory is a way of 'earning' your way out of Hell.
But it's not a waiting room, either. There is no "well, this person is really, really good and so goes straight to Heaven; this person is really, really bad and goes straight to Hell; you're in the middle, not too good or too bad, so we'll stick you in Purgatory and decide later".
At the Second Coming and General Judgement, after the resurrection of the body, our eternal fates will be revealed. Then the separation of the sheep and the goats takes place: the blessed for Heaven, the damned for Hell. But the Holy Souls in Purgatory will go to Heaven, not because they 'worked out' their sins there but because that was always their end. If I can cobble together a metaphor, it's not "is this true gold or fool's gold? let's find out", the souls are gold but being purified of dross as gold is tried by fire.
There is no change after death, no second chance, no "okay I was kinda bad on earth but now I can do better if you let me".
Going up, yes, going down, yes but no "possibility" that is unknown until the Last Judgement; at the Particular Judgement (for each individual person after their death) our fate is known finally and for sure.
From the Catechism:
"III. THE FINAL PURIFICATION, OR PURGATORY
1030 All who die in God's grace and friendship, but still imperfectly purified, are indeed assured of their eternal salvation; but after death they undergo purification, so as to achieve the holiness necessary to enter the joy of heaven.
1031 The Church gives the name Purgatory to this final purification of the elect, which is entirely different from the punishment of the damned. The Church formulated her doctrine of faith on Purgatory especially at the Councils of Florence and Trent. The tradition of the Church, by reference to certain texts of Scripture, speaks of a cleansing fire:
As for certain lesser faults, we must believe that, before the Final Judgment, there is a purifying fire. He who is truth says that whoever utters blasphemy against the Holy Spirit will be pardoned neither in this age nor in the age to come. From this sentence we understand that certain offenses can be forgiven in this age, but certain others in the age to come. (St. Gregory the Great)
1032 This teaching is also based on the practice of prayer for the dead, already mentioned in Sacred Scripture: “Therefore [Judas Maccabeus] made atonement for the dead, that they might be delivered from their sin.” From the beginning the Church has honored the memory of the dead and offered prayers in suffrage for them, above all the Eucharistic sacrifice, so that, thus purified, they may attain the beatific vision of God. The Church also commends almsgiving, indulgences, and works of penance undertaken on behalf of the dead:
Let us help and commemorate them. If Job's sons were purified by their father's sacrifice, why would we doubt that our offerings for the dead bring them some consolation? Let us not hesitate to help those who have died and to offer our prayers for them. (St. John Chrysostom)"
Ah - thank you!
So do we all have our personal judgement day at death? And the purchasing of indulgences was to shorten your time in purgatory and speed your passage into heaven... I guess all of this fits together.
But that's separate from the second coming and the resurrection of the body and life ever after? I've always assumed that the second coming was also Judgement Day.
I reread The Baltimore Catechism of my religious education class as a 7 year old a few days ago.
The grandchildren of Italians and of what were at the time unified-Yugoslavians were released from public school custody for a couple hours a week to go to the ethnically appropriate church for instruction. Who knew where the heathen grandchildren of the Finns went. ;)
That was a lot to digest at such a tender age.
The pope is still infallible, right?
Oh, that's going back to the Old Days right there, Tyrone!
Yeah, it was a lot. But the idea was that kids would get solid doctrine into their heads by rote memorisation, and never mind if they understood it completely. In later years, they would remember the teaching and be able to apply it.
https://www.catholicity.com/baltimore-catechism/
Also, in the days of that catechism (which was first published in 1885, Wikipedia tells me) the age for First Communion was *after* Confirmation, which would have been around twelve to fourteen years old (so the kids would have been older). The age wasn't reduced to "age of reason/around seven" until 1910 under Pope Saint Pius X):
https://www.catholicireland.net/first-communion-or-confirmation-first/
Then we switched to the social justice/workbook type 'catechisms' in the 70s and the doctrinal teachings were left out, pretty much.
The pope is always infallible 😊 (so long as he is teaching ex cathedra)
I received confirmation from the closest bishop at about 14. I was honestly a bit shocked when I saw him flip away the cigarette he had been smoking on his way out of our priest’s presbytery. Bishops smoke cigarettes??? This was the boyscout phase of my life that, try as i might, somehow have never been able to put fully behind me. Getting there though. ;)
> Except that's not at all how the theology works; in Christianity, everyone is bad, equally so, and needs redemption as a result
In the secular transformation of these ideas everybody is bad (privileged) and are born with original sin ( privilege of skin colour or being cis).
It’s a bit more intersectional of course but you have the original and present day sin of being male, or white if you are not male, or male if you are not white, or cis - then that’s pretty much everybody.
He’s not arguing that these secularists are the same as Christians now, but they derive from the same philosophy.
Fuck, OK, one by one:
1. "everybody is bad" in your thing relies on a definition of everybody which means "only a few groups, mainly white men" and once you realize you are using a nonstandard definition is AGAIN not at all like the theological beliefs of christians re: Who Is Bad (again, actually everybody).
2. And then you go, OK, but it applies a little bit to everyone, in different amounts, depending on what you are, where white male is the worst and then there's gradiations all the way down. Which is AGAIN not like either catholic or protestant Christianity but ALSO is not like what OP's post claims (that there's only "perfect" and "damned" with no middle ground or gray area in both protestant christianity and wokeness)
3. And then you say they are derived from the same philosophy, which, re: 1-2, they clearly aren't unless you contort yourself as OP did (fully misrepresenting the views of Christianity so they will coorespond to something else they don't resemble and, broadly, oppose) or what you did (use "everybody" to mean "some" and "lots of different groups with different levels of original sin" to mean the same as "everyone is equally bad and fucked without redemption".
Like, listen, I *sort of* get how not liking Protestants much might motivate someone to, basically, blame them for various problems. That's normal. But I'll be damned if I'm gonna be part of the main US group opposing the woke the most consistently in the most ways and ALSO be told they are my fault, and ALSO for reasons the arguers are so contorted using I can hear their vertibrae scraping together.
" But I'll be damned if I'm gonna be part of the main US group opposing the woke the most consistently in the most ways and ALSO be told they are my fault"
Let's hope not, there's always the chance of salvation for everyone!
I get what you're saying but unhappily, in the same way (American) atheists of a certain stripe go at Christianity like a bull at a gate under the assumption that *every* Christian globally and historically has been *exactly* like 19th to 20th century American Fundamentalists, those who go woke tend to go as they have been culturally conditioned, and that's American Protestantism in the wider culture.
Which did, under the Reformed strain, go very much "elect or reprobate". You are either on the Right Side of History or you are a bad terrible person benefiting from Systemic Racism and the Original Sin of Whiteness. You can't help yourself! We're not blaming you, we're just pointing out that unless you do all this work and every single day work to undo and unlearn Whiteness, well then you are one of the damned.
From the progressive side, it's easy to see the linkage, as they love talking about Christian Nationalism (as though *that* is all of Christianity) and the connections with White Supremacy:
https://www.amazon.co.uk/Sin-White-Supremacy-Christianity-Religious/dp/1626982376
"Hill-Fletcher shows that the Christian habit of seeing themselves as the "chosen ones" has often been translated into racial categories as well. In other words, Christian supremacy has historically lent itself to white supremacy, with disastrous consequences. Hill Fletcher proposes educational strategies to disentangle the two that will help us move forward toward racial healing in America."
It's harder to see from the other side, as the progressive element will vehemently deny any links to religion, but they have been formed by the culture they came out of, and American culture is influenced by Protestant Christianity.
https://benthams.substack.com/p/follow-up-on-urgent-animal-welfare
Summary - the Farm Bill is probably going to be voted on in the house within the next few days. If it passes as is, it will nullify all state laws enforcing animal welfare standards on interstate meat and dairy imports. (Eggs are thankfully exempt.) It will also pre-empt future laws along these lines.
If you want to help prevent this, the linked post contains a document detailing how to help.
https://www.congress.gov/bill/119th-congress/house-bill/4673/text
>In general.—Producers of covered livestock have a Federal right to raise and market their covered livestock in interstate commerce and therefore no State or subdivision thereof may enact or enforce, directly or indirectly, a condition or standard on the production of covered livestock other than for covered livestock physically raised in such State or subdivision.<
So it's not wiping out state-level law, it's preventing state-level law from regulating the states around them. Washington law can no longer prevent Alaska from importing factory-farmed eggs.
>(B) does not include domestic animals raised for the primary purpose of egg production.<
...except they're exempting the eggs.
So states can regulate animal raising in their own territory, but not put requirements on imported food? Or can they do that too, and the bill only closes some weird loophole where states had a say about completely unrelated animals?
That's the way I read it; you can prohibit factory farming inside the state but can't stop factory-farmed goods from entering and being sold like any other goods. And they've excluded imported chicken products, presumably because the chicken guys didn't pay them enough.
This was the best Google would do for me, if someone has a better link I'd welcome it.
Your first sentence is correct - they won't be allowed to put requirements on food they import into their state
Thanks, I have corrected my comment. I wrote the original in a rush so I made a couple of errors
I find myself getting bored with the chorus in songs. Not so much in pop songs because no one cares about the lyrics but more so in anything narrative driven. They put in so much effort in setting up a story and telling it beautifully just to break the flow for no purpose other than convention. Sometimes it’s just a short refrain but other times it’s lengthy and all I can think of is how much time it’s taking up. Imagine you were watching a tv show and it can’t stopping every ten minutes to replay the opening credits. It would be ridiculous.
Now you’re probably thinking that of course they have to do it. It’s what people expect and it’s necessary to get an audience. The go to counter example would be Pink Floyd but they’re trying to be different. Look instead to Marty Robbins. Very popular country artist and often went without the chorus. Not to be rebellious but because it was unnecessary. You can’t listen to his hit song El Paso and tell me a chorus would have improved it. It would get in the way. He had enough sense to let the story and the guitar carry the song without it.
First song that popped into my head to test what you're saying was "The Gambler" by Kenny Rogers. Would this song be better if the chorus were not repeated or if the lyrics to the chorus changed each time? I think not.
The Marty Robbins song is a fave of my brother in law. It is good.
I’m guessing you don’t care much for My Sharona.
I've always been annoyed by this too.
>Imagine you were watching a tv show and it kept stopping every ten minutes to replay the opening credits.
...well, they literally do that every twenty-two minutes. (Forty-three or so for the hour-long shows.) Because otherwise the audience wouldn't remember the show's name.
The best uses of chorus can recontextualize the chorus in each pass. (Is it lame to use the Protomen as an example? Almost certainly. https://www.youtube.com/watch?v=MIZJCiM-IaY)
Opening credits don’t interrupt the narrative in the middle of the story.
And Netflix added a button to skip them because they’re really annoying when binging.
They do in multi-episode stories, like in Classic-era Doctor Who.
I would say that depends on how cliffhanger-hungry the show is.
Isn't that entirely solved by just having different lyrics for the subsequent choruses? I've seen plenty of pop songs do that...
I've seen a few examples of that, but it's pretty rare.
Admittedly, it's uncommon to see two or three fully unique choruses, but it's common enough to see an ABA pattern.
What are songs that do that?
Jimmy Buffet's Margaritaville subtly changes one line in the chorus each time -- in order:
Some people claim that there's a woman to blame
But I know it's nobody's fault
to
Now I think, hell, it could be my fault
to
But I know it's my own damn fault
which signals that song is over because the main character's arc is complete.
Literally the most streamed Japanese song ever does this. If that isn't pop, I don't know what is.
https://youtu.be/x8VYWazR5mE
How on earth is that "the most streamed Japanese song ever"?! It's not even Yoasobi's most popular song!
If I go to Yoasobi's Youtube page and sort by popular, it's only third on the list, and that's just *one band*, not the whole of Japanese music. Come on now.
https://www.youtube.com/@Ayase_YOASOBI/videos
Why do you have this exasperated tone for me not knowing a Japanese song?
I'm sure you could find popular English songs that do this too without too much difficulty. But it is evidence against people doing this because it's "necessary for an audience". Looping back to a motif is something that works well for both music and its lyrics, but there's merit to doing things differently than normal. In fact, I've actually had trouble finding songs that just repeat the chorus three times without any modifications, presumably because artists realized that would be boring.
...What I'm trying to say is that you should probably try harder to find music that you like.
I’m not saying that absolutely no songs go against the norm. I’m not an idiot. My point is that outside of handful of exceptions, most songs keep the same formula and I don’t think there’s any commercial reason it has to be like that. I listen to a lot of music in a week, most of it not well known and outside of instrumentals, they still do the same thing.
Stop talking down to me. It’s really irritating.
I love it when a narrative song uses a chorus effectively. In The Mariner's Revenge, the chorus serves as a reminder of the act of revenge he wants to carry out, with the song ending on the chorus and him finally getting revenge.
If you want to read a story just read a story. If you want to listen to a song, listen to a song.
Having just read the lyrics to Marty Robbin's El Paso, I don't think a chorus would improve it; it's a narrative-driven song.
On the other hand, to pick the first example that spring to mind, I definitely don't think that Smells Like Teen Spirit would be improved by adding four more verses and deleting the chorus. The lyrics are there and they create a certain mood but they're not what the song is *for*.
Agree, and in fact some songs repeat the chorus the last time, sometimes more than once, because it's so good it's worth that much runtime of the song. To use the example of a different Nirvana song, "Lithium" does this.
I am a huge fan of Marty Robbins. "El Paso" is one of the earliest memories of my life. I was in the back of an ambulance being taken to a hospital after falling out of a tree at boarding school and the ambulance driver's radio played "El Paso." I think it was the first time I ever heard it .
What about Big Iron?
No chorus
I guess I am unclear what you mean by chorus then.
He repeats “big iron on his hip” but it’s not separate from the verse. A chorus is its own section.
I became interested in this issue you raised and decided to do a little research. It seems my instinct about work songs and also liturgical recital is a very powerful influence in that kind of song structure. I am attaching a link to a conversation I had with Claude about it.
https://claude.ai/share/9e50ec20-f1c0-4285-ba32-976e08a2ef69
I am pasting in something here because it doesn't get included in the snapshot of my conversation that Claude shared with me :
>The gandy dancer songs fed directly into the blues — Jimmie Rodgers learned his guitar style and his yodeling from watching the Black track crews — and from the blues into rock and roll. The ancestry is remarkably direct.
IV. The codification — from field to stage
Pre-1840s
Oral tradition only. Work songs, shanties, spirituals, field hollers. No fixed form; the caller improvises; the refrain is memorized, not written.
1840s–1900s
First codification. Verse-chorus form appears in parlor songs and minstrelsy ("Oh! Susanna," 1848). The structure is now written down and published as sheet music.
Early 1900s
Tin Pan Alley supplants it with the 32-bar AABA form — the standard used by Cole Porter, Gershwin, Rodgers and Hart. The chorus becomes the entire song; the verse recedes.
1950s–1960s
Return of the form. Blues and rock and roll revive verse-chorus structure. By the mid-1960s it is dominant in rock, where it has remained ever since.
1960s–present
Elaboration. The prechorus emerges as a transitional section. Production elements — drops, builds, breakdowns — begin to serve the emotional function of the chorus without containing a melodic hook at all.
V. On your thesis
The observation that the chorus exists to permit communal participation — to allow those who cannot carry a verse to still belong to the song — is well supported by the history. The shantyman and the caller were skilled specialists. The rest of the crew were not expected to be. The chorus gave them full membership in the musical event without requiring any of the caller's particular gift.
This is also, it is worth noting, why choruses tend toward simplicity of lyric, repetition, and a narrow melodic range. They are not simplified because composers were lazy. They are simplified because they must be instantly learnable by people who have never heard the song before — or who are too tired, too far away, or too occupied to track a melody carefully.
The chorus is the democratic clause in the contract of performance. The verse is where skill speaks; the chorus is where everyone is permitted to answer.
In my experience, white gandydancers don’t do much singing. Probably have lousy rhythm too.
Bill Murray in Stripes: Boom chuka luka, boom chuka luka.
https://youtu.be/-zwYU5Jf2X8
Black guys, help the white guys.
https://m.youtube.com/watch?v=g7sNpXJ6ci0&ra=m
But hardly any music listening now is communal: it’s mostly to have something in the background while you’re doing something else. That just strengthens my thesis here. They aren’t doing this structure for any purposes it’s just convention.
I think there is a relationship to songs that were originally work songs. chorus to haul etc
Last week, I asked for people's rankings of various aerospace firsts. The thing that inspired this was that I've been reading memoirs and biographies about the US Space Program in the 50s and 60s and noticed a few things.
One is that first-hand accounts I've read are pretty unanimous about Charles Lindberg being an awe-inspiring mega-celebrity. I remember a scene early in Apollo 13 where Jim Lovell (played by Tom Hanks) is reacting to the Apollo 11 landing by laughing about how Neil Armstrong has taken his place in the history books alongside Christopher Columbus and Charles Lindberg; I'm pretty sure this is fictional, but it rings true to attitudes I've seen about Lindberg among people of that generation, e.g. Gene Kranz (*) seeming starstruck about Lindberg visiting Mission Control. I wanted to sanity-check my instinct that Lindberg's fame has faded considerably over the intervening several decades since the 60s.
(*) One of the Mercury/Gemini/Apollo flight directors, the one played by Ed Harris in the Apollo 13 movie.
Another is that, in discussing the books with one or two people around my age (mid 40s) but with less intense nerdery directed in that particular direction, I was surprised to learn that Chuck Yeager (first man to break the sound barrier), Alan Shepard (first American in space, and also the first man to play golf on the Moon), and John Glenn (first American in orbit) weren't names they recognize.
The results of this unscientific survey are:
1. There is an almost unanimous sense that Neil Armstrong (first man on the Moon) is the most important of the people on the list, followed by Yuri Gagarin (first man in space and in orbit), with a couple dissenters putting Lindbergh or Yeager on top. Everyone knew who Armstrong was and almost everyone knew of Gagarin. This is more or less what I expected.
2. I am the oddball in terms of remembering Shepard and Glenn. In general, people old enough to remember the tail end of the Apollo program remember them, but people my age bracket or younger only do so if they're deep aerospace nerds who remember everyone on the list. Even among people who do remember them, they're generally ranked low.
3. Lindbergh is a bit better-remembered than I would have guessed, and is usually ranked third or fourth. Older people and people with deeper interests in aerospace history tend to put Lindbergh higher.
4. Yeager is a little more of a deep cut than Lindbergh, but people who remember him usually rank him third or fourth. This is about what I would have expected.
Lindbergh is one of the interesting lacuna of history. Yes, he really was that huge in the 20s and 30s, with stuff like the Lindbergh Baby dominating headlines at the time. I suspect some of this is general fading with time, and some is specifically his association with the Nazis in the 30s forced him out of the public eye in the 40s and beyond. But, yeah, he was still widely beloved in the aviation world among those who'd grown up before his fall.
>The thing that inspired this was that I've been reading memoirs and biographies about the US Space Program in the 50s and 60s and noticed a few things.
I'm going to assume you've already read Carrying the Fire because it's sort of the free space of astronaut autobiographies, but I'd flag Walt Cunningham's All American Boys as the most interesting one aside from that. He's a lot more honest about the problems of the astronaut office than most are, although recent editions include him yelling at NASA post-Columbia, which I skipped as being not interesting. For real fun, read it back-to-back with Slayton's posthumous Deke.
Edit: It is also worth noting that Yeagar got the supersonic job thanks to some slightly bad behavior on the part of Bob Hoover. Who also passed up a test pilot job at Boeing in favor of Tex Johnson. Boeing dodged a bullet there.
I'm actually in the middle of Carrying the Fire now. I started the current run of reading after watching the first few episodes of "For All Mankind" and getting inspired to reread "The Right Stuff" and "Failure is Not an Option". I picked up "Carrying the Fire" next because I remembered reading and liking Collins's "Mission to Mars" as a kid. Thank you for the recommendations; I was planning on reading "Deke" next, but now I might slot "All American Boy" in before it.
One thing I did notice is that there seems to be some difference of opinion over how "fighter jock" the first few rounds of astronauts were in temperament and culture. Wolfe seems to be the outlier so far in emphasizing this, and I'm looking forward to seeing how Cunningham and Slayton characterize things.
I'm also still not sure what to make of "For All Mankind". It definitely has a lot of potential, I've heard good things about where the series is going, I've really liked other stuff Ron Moore has done, and the writers definitely have done their homework. But there are a lot of details that bother me. They're going even further than Wolfe in emphasizing the Fighter Jock angle. They remixed some bits I recognize from the books I'm reading in ways that don't quite make sense. And the pacing so far has been off, with the episodes feeling padded and then spiced up with contrived drama to try to keep them from dragging. I'm probably going to give it another couple episodes before deciding whether or not to give up.
I haven't seen "For All Mankind" and don't plan to because it strikes me as the sort of thing that would drive me absolutely nuts. But I'm also an extreme weirdo.
My other recommendation's would be A Man on the Moon, which is probably the best flight-focused history of Apollo, and the TV adaptation, From the Earth to the Moon, which is really well-done.
But for answering the cultural question, Cunningham is the best book I've read. Re Wolfe specifically, I believe Deke didn't love it, but Schirra was reasonably positive in his book. (Far from the worst astronaut book, but not amazing.)
For the Chuck Yeager and original Mercury 7 astronauts, Tom Wolfe’s “The Right Stuff” is great. The film of the same name gives the gist of book
It's certainly a great book as a piece of literature. I'm not sure it's nearly as great as a piece of history, although my attention focuses a bit later so I can't pick out specific errors offhand.
Did Grissom ever convince anyone that he didn’t blow the hatch himself?
Edit:
I just found this
https://www.smithsonianmag.com/air-space-magazine/new-evidence-shows-gus-grissom-did-not-accidentally-sink-his-own-spacecraft-sixty-years-ago-180978240/
That was it!
Yes, almost everyone. No idea why Wolfe latched onto that issue, but the consensus by basically everyone else is that Grissom was telling the truth. Among other things, several later pilots blew the hatch manually, and the recoil of the mechanism was enough to bruise their hand. Grissom had no bruises. I believe the best theory right now is that ESD from the helicopter may have set it off via an antenna in the water, but we'll never really know for sure.
It is also worth noting that Grissom was assigned the first flight of not one but two spacecraft, something that probably would not have happened if he had been viewed as a screwup, and the general consensus is that he probably would have been the first man on the moon had he not died in the fire.
Just finished re-reading it, and I checked the movie out from the library last weekend and plan to watch it in the next few days.
Watch for the real life Chuck Yeager’s cameo as a bartender.
IIRC Wolfe refers to the status hierarchy as a ziggurat, in the film it’s referred to as a pyramid.
You don't want to live in the world where Boeing won its lock on the jet-airliner industry because their test pilot rolled a 707 *with all the engines shut off*?
It's less that I expected him to shut the engines off and more that I thought he'd do an 8-point hesitation roll.
(That would not have been a good fit for Hoover, who bribed a clerk to get into fighters.)
This comment made me look up this gentleman's biography and holy cow, what a story. Escaping a Nazi POW camp by stealing an enemy plane and flying it back to Allied territory? Incredible.
Oh, I was more thinking "rolled every plane he ever sat in". But yes, the escape is a pretty good story, too.
Possibly the only fighter pilot ever to pull off even a half-Maverick in real life.
And I've seen his air-show routine in person, where he rolls and loops and otherwise does a full aerobatic display, in a twin-engine light transport aircraft with the engines turned off throughout. Absolutely magnificent.
FWIW, I only recognized John Glenn as "the oldest man in space" for his return trip in 98. I knew he did something in the Apollo days, but I couldn't have told you exactly what.
Pre-Apollo!
Shepherd and Glenn's achievements are both of the form "First American to...", not "first person to..."
Similarly, Lindbergh is overly famous because he's an American, and Americans tend to be loud about these sorts of things. "First non-stop solo flight across the Atlantic" is a pretty good record to hold, but how many Americans can name the pilots of the first nonstop non-solo flight across the Atlantic, or the first non-non-stop flight across the Atlantic, or the first non-solo flight across the Pacific, or the first flight across the English channel, or any number of comparable records that aren't held by Americans?
I can't remember the pilot's name, but first non-non-stop across the Atlantic was the USN's NC-4. so that one is held by Americans. As for first across the English Channel, Louis Bleriot, of course.
Lindberg gets credit for being the first person to fly someplace *interesting* across the Atlantic. Has nothing to do with being American, or flying solo. A pair of Frenchmen flying from Paris to New York would probably(*) have been as famous. The guys who crashed into a bog in Ireland after flying over most but not all of the Atlantic ocean, are a footnote even if I do remember their names (Alcock and Brown). They didn't even arrange to have a newsreel team covering the bog.
Same reason Armstrong usually comes in ahead of Gagarin. Going someplace interesting beats going around in circles just to prove than you can. The new medium of transportation isn't worth much if you can't use it to go to interesting places.
* Contingent on their being reasonably charismatic and PR-savvy, as was Lindberg. But I'm positing Frenchmen, so that seems a safe bet.
FWIW I was one of the dissenters that put Lindbergh first, and I'm not American. I also don't remember all those other, less-well-known firsts you proposed, whether or not they are American. The nationality angle was important back when the things happened, but not very important by now. It's all "history of aviation and spaceflight" by now.
I’m mildly surprised that Chuck Yeager didn’t rank higher.
Wanted to comment on this recent paper on PNAS:
https://www.pnas.org/doi/10.1073/pnas.2537049123
The main point is estimating returns to a year of education, using excellent data from Norway. They compare: A) a simple OLS controlling for demographics and family characteristics, B) models with fixed-effects for siblings or twins, and c) Mendelian randomization instruments (MR). The idea of MR is to use the random variation in what genes a person receives from their parents to simulate an RCT. If you get lots of the genes associated with higher education achievement from your parents, that is kinda like being assigned randomly to get more education. They argue the larger estimates with MR are more reliable than the other methods.
I have two problems with the paper.
First, I think this method will pretty much always violate the exclusion restriction, and so will most likely be positively biased. For the instrument to be valid, you need to have genes that affect income _exclusively_ through education. Think about how genes can possibly result in higher education. They may cause better memory, better IQ, better health, time preference, etc. These things cause more education, but every one also independently affects labor outcomes. Even something highly defensible like, say "enjoying a school-like environment" can plausibly have independent effects (you may choose a career in education or academia, you may prefer a regular corporate job instead of startup, etc). Unless you can get genetic memories of one year of education, Bene-Gesserit-style, this issue is pretty fundamental to the strategy.
In fact, their estimate using monozygotic twins is the lowest of the bunch, and this strategy should be adjusting for the biases I mentioned (as well as shared environment). (I do have issues with this result too, though).
Second, the very research program seems misguided to me. We can try different strategies to isolate "the return of education" and discuss relative strengths, but actually each strategy will be leveraging a different source of variation, and thus estimating a different LATE. It is pretty likely that there is a lot of heterogeneity, and using different methods will lead to different results not only because they are correcting different biases, but actually because the target parameters differ.
In fact, we have understood the concept of LATE for 30 years: it is time to abandon the pretense that it makes sense to talk of a single "return to education" as a useful concept. For public policy in particular, the relevant interventions are never "increase the average level of education by 1 year." They are more like "increase the number of scholarships by some amount", "increase funding for specific programs by some amount", "build X new schools", etc. Research on the returns of each particular intervention is much more useful as a guide than some average value for the population (even if we could estimate it well).
Some discussion of the paper on X:
https://x.com/sashagusevposts/status/2048469708886675884
You phrase this as the 'return to education' which I would have interpreted in societal terms if I had not read the abstract of the paper, where it is phrased as a study in whether an additional year of education increases earnings.
In public policy, we want to estimate the social implications of increased investment in education. Individual attainment is probably not the best way to do this - population level data would probably be more useful. In this setting you could feasibly do a cost-benefit analysis to see whether additional hours of instruction would lead to sufficient additional GDP in the future to justify the current expenditure. In analysing multiple different options, this could be one of the variables used to estimate the return on investment.
Would anyone have a good anthropology suggestion book for a layman?
California's wealth tax ballot measure is retroactive: anyone a resident in the state on Jan 1 , 2026 must pay it. What happens if it passes, is upheld as constitutional, and a bunch of billionaires leave the state and refuse to write checks? I don't imagine Trump will give California any aid in enforcement, nor will other red states. The extradition clause requires states to extradite, however, it reads "A Person charged in any State with Treason, Felony, or other Crime, who shall flee from Justice," which you can argue doesn't apply to someone who left California in May 2026 and whose crime is something they did outside California later.
I imagine any one rich enough to feel a substantial hit from such a law has already set up domicile elsewhere to avoid "oh crap, it's Jan 1st and I'm still not a Monégasque citiizen".
I vaguely remember reading some article about this a while back which mentioned that Jeff Bezos, for one, was moving to Florida. So the deep pockets probably already have their lawyers ready for "in fact, at that date, my client was *not* a resident in the state" arguments in court.
See something similar in an Irish context, where attempting to tax rich non-domiciled persons saw a year-on-year decline:
https://www.oireachtas.ie/en/debates/question/2025-05-13/489/
Domicile Levy Year Number of Returns and / or payments Liability Paid
2019 19 €2,732,653
2020 18 €2,694,878
2021 18 €1,698,708
2022 12 €1,966,288
2023 15 €2,463,909
Total 82 €11,556,436
As ever, it's going to be the lawyers making money out of this, with cases for and against the very very rich. Depends how much the governor (be that Gavin or his replacement) feels they can push the big tech companies and tycoons versus the threat of them all moving to Texas or wherever or setting up residence elsewhere, and where their companies are located (I can imagine the lawyers right now rubbing their hands at the vision of billable hours arguing that just because Googmagoo is physically located in Mountainous Vista, it is not headquartered there and thus not liable for the tax, see also my client Mr Beff Jezos whose permanent residence is now in Venice and we don't mean the one in LA):
https://windes.com/california-residency-tax-purposes/
Gavin fought hard *against* the wealth tax. While CA state government does have its faults, they're firmly on the right side here. The culprit here is the ballot proposition system, and SEIU's weaponization thereof.
https://www.astralcodexten.com/p/seiu-delenda-est
States engage in reciprocal enforcement of tax law. Once California has a judgement against you they can use your home state's legal system for enforcement. The constitution's full faith and credit clause means that you can't escape a judgement in state court by moving to another state. They can also just go into your bank account if you use a national bank that has branches in CA.
I wouldn't expect the retroactivity to be found constitutional. You can't punish people for breaking a law you said didn't apply, which is what the old tax structure told them. https://constitution.congress.gov/browse/essay/artI-S9-C3-3-3/ALDE_00013193/
If it was found constitutional, it'd be treated as any other form of state tax evasion.
There's plenty of precedent for it being perfectly constitutional:
Welch v. Henry (1938), United States v. Darusmont (1981), and United States v. Carlton (1994).
IANAL, but naive googling suggests that retroactive taxes have often been found constitutional, with Carlton 1994 the key SCOTUS case.
As far as I can tell, this arises from the distinction between civil and criminal penalties that also give rise to other injustices:
(a) loose process protections for civil asset forfeiture
(b) loose process protections for immigration violations (which somewhat cuts both ways with "immigration violations are civil, not criminal" as one rejoinder to the "deport the criminals" line of argument)
(c) lighter process protections associated with administrative fines by SEC, FCC, et al (though perhaps curtailed since Jarkesy?)
With the current SCOTUS composition sometimes interpreting money issues as the highest form of rights, though, the question of retroactive state taxes might be ripe for re-assessment.
Would be fun to watch! Though I really hope it doesn’t pass
The far right likes to point to Detroit as a cautionary tale of the dangers of “diversity.” Leftists often respond with accusations of racism, which is not an argument, leading some to accept the far-right's claims uncritically. They should not. In truth, what killed Detroit was the 1924 Immigration Act and the anti-discrimination laws of the 1960s.
Without the 1924 Immigration Act, the U.S. would have received several decades worth of European immigration, which was going and would have continued to go to places like Detroit. Back in 1920, when America had a much more open immigration system than it does today, Detroit was 96% white and 4% black. By 1960, after several decades of restrictive immigration, it was up to 29% black. How did blacks go from 29% to the dominant, governing demographic? The obvious answer is the 1967 Detroit Riot. Still, you ask why? How does 30% of the population drive out a higher-IQ, more martially skilled majority twice its size? Because the white community received the message from its political leadership and the rest of the state and country that *we do not have your back*. Shortly after the riot, Michigan governor George Romney (father of Mitt) called the legislature into a special session that passed “fair housing legislation.” The Michigan Historical Review explains:
"Efforts at the state level to enact fair housing legislation proved unavailing until the racial turbulence in the summer of 1967 appeared to signal the need for legislative action to alleviate the condition of the state’s blacks. When Romney called a special session of the legislature following the Detroit riot, the New Detroit Committee-recently formed by leading business firms in the Detroit area to help deal with the city’s problems-sought to persuade the governor to put a fair housing law on the agenda. Romney had promised legislators, who had defeated fair housing proposals in both the 1964 and 1965 legislative sessions, that he would not seek such a measure in the special session. After returning to Lansing, however, from a national speaking tour in which he had stressed the need to deal with racial problems, “he was boxed in and could not say no” to New Detroit. He consequently delivered a message to the legislature on 13 October calling for a “statewide open occupancy law.” He also called for code enforcement legislation as well as a tenants’ rights law designed to create “a covenant of fitness, good repair and compliance with applicable health and safety laws and ordinances for every rental arrangement” in the state."
https://web.archive.org/web/20130504011411/http://www.law.msu.edu//clinics/rhc/MI_Housing_Disc.pdf
People who white flighted to the suburbs did so not necessarily because of the mere presence of blacks in the same political entity, but because the government made it impossible to live a white lifestyle in a white apartment building and a white workplace. Had it not been for the 1924 Immigration Act, motivated to a substantial extent by irrational religious sentiment, and the anti-discrimination laws of the 1960s, the mass exodus would not have happened.
Immigration restrictionists might protest that, had it not been for immigration law, Detroit would have later been flooded with non-white “migrants” who would have driven white Detroiters out. There is little reason to think this. Open borders means the entire world. If a bunch of Congolese migrants are causing political problems, employers and landlords could hire Indians, Chinese, or Indonesians instead. A libertarian open borders regime is not one in which anyone can go anywhere and do anything. It’s one in which anyone can rent from willing landlords and work for willing employers. The white communities in Detroit, who owned virtually all the property in 1920, would have been the ones deciding on who to hire and rent to. I trust they would have decided wisely.
Surely all of these are symptoms and aggravating circumstances, not causes ? The main problem with Detroit is that its economy was based around the US auto industry -- which collapsed. Without legal money coming in, crime emerged as the primary economic driver for those who remained in the city. Open borders would not have solved the issue, because immigrants would not immigrate to a city unless there's something there they can use, i.e. jobs. And what is there to do in Detroit, besides crime ?
The big car companies starting moving out of the city of Detroit as early as the late 50’s. Later they moved most of the factories and opened new ones in other parts of the country. It was a whole thing.
This is lump-of-labor fallacy thinking.
There is a big difference whether the jobs disappear gradually or overnight. The market adapts, but that takes time.
A factory can close overnight and leave thousands of people unemployed. One day in future they may all find new jobs, but it definitely won't be the next day. Probably not even the next month. In the meanwhile, you get social disorder.
Here's a chart of the homicide rate over time:
https://static01.nyt.com/images/2026/01/21/multimedia/2026-01-07-crime-stats-index/2026-01-07-crime-stats-index-videoSixteenByNine3000-v2.png
See that massive spike during the Great Depression? Yeah, I don't see it either. Your theory of human behavior needs some updating.
Detroit suffered from the decline of the auto industry, is scarcely the only US city to experience white flight, and in any event seems to be on the upswing. Case in point, it’s building a huge new bridge to Canada.
Do you think discrimination by private persons and entities is not libertarian? That's unusual view.
I was honestly amazed the Nashville ACX meetup saw a turnout of over ten people, including one country music singer/band lead. Not to stereotype but that’s the last person I’d have expected to be interested in the ACX blog.
Anyway, that was fun so we’re doing it again, this time at the Martin’s BBQ *on Elliston Place* where parking is free and abundant, 2:00, Saturday May 16. See ya’ll there!
There's a lot more to country music than the "beer, women, trucks, and Jesus (in that order)" mainstream variety. Stereotyping is bad mmkay?
Maybe the last person to expect at an ACX meetup, but the first person to expect in Nashville, so it cancels out.
Just as the Great Book foretold. "He who is last, shall be first, and he who is first, shall be last (but of course this is tautological)."
Not tautological at all.The guy in first would go second or any other position unless specified.
Who's first?
No, Last is first. Who's last.
...
Who’s on first. A classic comic bit.
You think that understanding that citations are not supposed to be fake requires a high order of intelligence. Interesting.
I certainly didn’t suggest that no person would ever fake a citation. I did suggest the LLM might be deliberately lying but that would be another big problem.
The LLM creators add all sorts of rules to manage LLM behavior. It doesn’t seem like a rule to require real citations would be hard to add.
The AIs “think” they’re real. Some humans have the same problem.
You're anthropomorphizing. You can give a human a rule like "if you don't know a real source to cite, either find one or cite no source. Don't just make up sources." A human will be able to understand that rule and can choose to follow it.
An LLM doesn't even have the option to "not make up sources." It has no sources and no concept of sources, all it is doing is writing citations. And EVERY citation an LLM cites is made up. But because the LLM has been trained on many real pieces of writing that include citations of real sources--and those citations are encoded in its weight matrix--the citations it writes will often (but not always) closely or even perfectly match some citation in its training corpus, which was written by a human who had a model of the actual, real-world source in their human brain.
But at the end of the day, writing that citation is still just an iterative next-token prediction process. If the distribution of probable next-tokens stays sharply peaked around a *specific* citation[1], then it will likely reproduce that citation, and the citation will point back to a real source. But if the distribution instead has a less-sharp peak around "thing formatted like a citation," without resolving a *specific* citation, then it will just reproduce a thing-matching-the-citation-pattern, which is nearly-guaranteed to be nonsense. To be clear, in NEITHER of those circumstances will the LLM be working forward from "relevant source in mind" to "citation pointing to that source" the way a human would. It can't do the obvious-to-a-human thing you want, which is to consult its memory, discover it doesn't know a source, and then either find a source from which to create a citation and respond appropriately. It's not a question of whether that requires a "high order of intelligence," it's a question of whether it has a human-compatible model of the world with a mental bucket for "relevant source" and a mental process for turning sources into citations[2]. They don't. They can't.
The best that the engineers can do is try to (imprecisely) re-target parts of the next-token-prediction function so that those less-sharply-peaked predictions get mapped to some other output, instead of leading to fake citations. But that's not anywhere near as easy as just writing a "rule" for the LLM; it's dealing with internal program states that are difficult to isolate, understand and modify.
[1] Which will happen because that citation showed up in the training corpus at the end of chunks of text that were statistically similar to whatever's in its context window right then.
[2] Things can get confusing because LLMs are trained on human text and tweaked to talk like humans, so the can summon up the verbiage that attends these mental constructs at will. They can even roughly reproduce a lot of the transformations mapping one sort of text to another sort of text, like "article" -> summary or "paper" -> "citation." But they're not doing it in remotely the same way, so it's a category error to expect the results to consistently be the same.
A base, non-"thinking" model is doing roughly the same thing as a human trying to recall citations from memory.
The rate is low and decreasing, but frontier LLMs sometimes produces both completely fake citations and citations that exist but aren't relevant or don't say what the reference claims. Humans make all those same mistakes. An unchecked reference from a random human about a random topic is not reliable either.
A "thinking" LLM or Codex/Cowork - like LLM+framework can be instructed to check its work by looking up every reference in Westlaw or whatever and verifying that each exists, and even be instructed to read the referenced decision and check that it matches the claim in the citing text. That further reduces the rate of the various kinds of errors. Legal AI software can even use more deterministic database lookups to completely eliminate hallucinated citations. And you can use a different model to check the appropriateness of the citation to the case.
All these are similar to the kinds of error checking and verification an individual lawyer or law firm would do using people. Unchecked junior associates make all those same kinds of errors. Does that mean you should never trust anything a lawyer says? (OK, well...) Modern AI is not categorically less reliable than humans. The claims and citations of specific AI systems may be either more or less reliable than specific configurations of humans.
> "You're anthropomorphizing. You can give a human a rule like "if you don't know a real source to cite, either find one or cite no source. Don't just make up sources." A human will be able to understand that rule and can choose to follow it."
No, I'm not anthropomorphizing.
These things have other rules to not produce overly objectionable results.
If it produces fake citations, it is either programmed to lie or has no understanding of what "citation" means (it's not intelligent).
If it produces fake citations, why trust any other results?
> "Which will happen because that citation showed up in the training corpus at the end of chunks of text that were statistically similar to whatever's in its context window right then."
Is there any indication it's finding these fake citations in it's "training corpus" at all? This wouldn't be an example of intelligence either.
> But they're not doing it in remotely the same way, so it's a category error to expect the results to consistently be the same.
This is making excuses for obvious errors.
Define "fake."
I have yet to hear of a LLM citation that picked a paper at random and cited it, without its backing up what it was being cited about, or one on a subject that contradicts what the LLM says, but I am sure that you could come up with a definition that would encourage that path.
I guess you have no reason to believe me, but I have come across several circumstances like this.
One example:
- I was asking an LLM to find me primary source documents from Founding Fathers about about their opinions on X.
- It said, without giving me a citation, "John Adams believed Y about X"
- I said "I asked you for primary sources, give me one to back that up"
- It said "here is a primary source backing that up, it's a letter from John Adams"
It was a completely unrelated letter from a historical website, neither from nor to John Adams. It was about some random thing.
I've mostly found it bullshiting citations after it says something without a source and subsequently tries to back it up. If it can't find one, it'll sometimes choose a thing that seems like it could be the source.
It mostly happens when you ask it to find sources for things it already confidently claims. I've only noticed it happening when discussing things where there probably exists a lot of unsourced junk statements in the training data (the Founding Fathers think X, the Buddha said Y). You know, stuff where the majority of common discourse is disconnected from documented fact.
That's exactly the sort of thing I wasn't talking about.
If it cited an actual letter by John Adams, you would have to read the letter to tell it was merely picking one at random. Since it wasn't by John Adams, it could be detected at a glance. You could even computerize that sort of validation.
How is "this letter that the AI claims was by John Adams was not by John Adams" anything other than fake citation? Unless I am sceptical enough to check "is that really a letter by John Adams?" I am not going to know any better, and the alarming tendency is to treat everything the AI says as "it's true because machines can't lie".
We get enough fake quotes passed around on social media allegedly where famous person said thing, we certainly don't need AI contributing to "John Adams totally said in this letter from 1995 that his favourite beverage was Pepsi" type slop.
It's a different one from the one I was talking about, where I had not had an example
I'm not sure what sort of thing you were talking about, then. Similar things have happened with primary source Buddhist sutras (where it linked specific sutras that didn't support its claim) which were less obvious and more annoying to detect.
This happened in two different ways. First in sourcing non-theravada sutras that did support its claim, but in a conversation specifically about theravada beliefs. You could say this is obvious, but I don't have an encyclopedic knowledge of which specific sutras belong to which traditions so it wasn't obvious to me.
Second, in sourcing sutras which do not at all contain the statement the LLM said they did. I wasn't able to verify it as a bad citation until I had read the whole thing.
The Founding Fathers thing and the Buddhist thing were the only two situations where I noticed it happening though, and I don't think I really understand your position, so I'm not sure if it relates to whatever you were saying.
If it was obvious that it was linking non-theravada sutras, it's not the problem that I talked about. If it was not, if you had to read the sutra to realize that, it is - and a particularly good case if it needed a subject-level expert to do it.
https://www.spellbook.legal/learn/lawyer-fined-using-ai-legal-fake-citations
Insufficient information. The only one it describes is a made up case.
> Define "fake."
??? That really shouldn’t be necessary.
Something that is offered as a citation but doesn’t exist in a meaningful way (page numbers being wrong would be a mistake not fake).
I’m including fake/false legal citations too.
What do you mean "shouldn't"? You can not possibly program a computer with something you can't define. It is necessary, in the sense "it is necessary that you breathe oxygen."
"Meaningful way" would have to be defined for your definition to work.
Defining “fake” shouldn’t be necessary.
> “ You can not possibly program a computer with something you can't define.”
???
The LLM is offering the “citation”. If its definition doesn’t include a citation to something real, then the LLM is defective. If it can’t get that basic thing correct, why ever would you expect it to get anything else correct.
> “ "Meaningful way" would have to be defined for your definition to work.”
Read the whole sentence you quoted an excerpt from.
> If it can’t get that basic thing correct, why ever would you expect it to get anything else correct.
Because it gets many or most things sufficiently correct, sufficiently fast and cheaply, so that on balance is still an incredibly useful tool? Also, people also make mistakes, too.
> “ Also, people also make mistakes, too.”
The fake citations aren’t a “mistake”. It’s not having any idea what a citation is.
You wouldn’t pay people for that result but give LLM a lot more latitude.
"Read the whole sentence you quoted an excerpt from."
It was as bad the second time.
No, it isn’t.
Calling the LLM names does not change that you do, indeed, have to define it rigorously -- and "rigorously" in the LLM terms.
LLMs do not understand real. At most they can ape what humans do and call real. That is how we got fake citations in the first place.
> "Calling the LLM names does not change that you do, indeed, have to define it rigorously -- and "rigorously" in the LLM terms."
No, I certainly don't need to provide a definition. It wouldn't use any definition I provided anyway.
It's not like citation is some sort of obscure, ambiguous, hard-to-understand concept.
If the LLM is using a definition, it's either using a stupid one or lying.
>> LLMs do not understand real. At most they can ape what humans do and call real.
Yes, they aren’t intelligent (not “AI”).
If it creates fake citations, anything else it produces is not going to be reliable.
I thought the idea of genuine artificial intelligence is that it saves us the trouble of rigorously defining every single step of every single task we give a computer. If I wanted to do that, I'd just write a regular script.
Every new layer of more advanced computing languages leads to more and more abstract and high-level concepts, but this is done by putting the concept into the language. If the users need not trouble their pretty little heads about what a fake citation is, the programmer has to do it for them, at the language level.
Not to mention that dealing with natural intelligence requires rigorous definition often enough to get what you want.
Before I respond to you, I'm going to need you to define the following terms: layer, advanced, abstract, and high-level. After that, I'm going need a rigorous definition of each word you used in each of those definitions.
Here's a non-paper citation example. When I asked an LLM to give me math olympiad problems on a certain topic, it mixed real math olympiad problems with routine exercises that it made up a math olympiad source for.
Yesterday, I got a mostly-real but slightly-fake citation: a theorem cited as "Last name (year)" where the author *did* publish a paper that year on the right topic, but the conditions in that paper were different from the theorem Claude stated.
Ah, no. The first is the made-up citation that is therefore clearly fake.
The second one is a slightly better fake. The problem would be if it learned to cite that paper.
Both of yours can be cleared up by looking if you *could* look up the citation. There is a graver problem when you must actually verify the citation.
Most peer review doesn't bother to verify the citations, then.
Now, once they manage to fake valid citations -- hmmmm -- someone should devise a computer program to *look up* citations. That actually is a definable task.
But what's "real" to an LLM? I would guess it is not just a simple "rule", but rather a ruleset+procedure ("double check that all citations you made are actually real citations" by following the link or finding a paper with the exact name/DOI, etc).
This ruleset might be feasible but computationally expensive, as it widens to amount of tokens and compute massively. Maybe the latter is the real barrier for the LLM-companies, barring any other hard requirement or incentive to do it.
> “ This ruleset might be feasible but computationally expensive, as it widens to amount of tokens and compute massively.”
It’s too hard. So, we will just provide garbage instead.
What would keep them from applying that short cut for (lots of) other things?
Where are they getting the stuff they are attributing to fake citations? Not from actual real citations, apparently.
(It’s not like they see “massive compute” as any sort of hindrance anyway.)
https://www.rottentomatoes.com/m/michael the Michael movie has a rotten tomatoes critic rating of 38% but a user rating of 97%. I don't know if this is historically unprecedented but I haven't seen it before.
It feels a little like an intelligentsia/proletariat split. As though random people (near universally(!)) really don't care - for the sake of movie depictions - about a major figure being a paedo if their story is extremely moving.
Obviously I second guess that a bit... maybe the 97% is bolstered by devoted fans being more likely to go and vote on the movie.
You might have thought now would be the time for people to be *more* hating on paedos than usual though.
Audiences really do not care about a good story when it comes to music biopics. They want to see the exact same formula used to see their favorite musician perform songs they like. It’s the worst genre of movies.
The largest gap as far as I know is for Reagan (2024), which gets a 18% from critics and a 98% from audiences. Presumably for similar reasons to the Michael Jackson one; it's a hagiography that only hardcore fans of the subject are going to see anyway.
Wait, I just found an even larger gap -- it's Melania (2026) with 10% from critics and 99% from audiences.
The biggest gap I know of in the other direction is The Last Jedi, with 91% from critics and 41% from audiences.
I didn’t see the movie, but I saw half the Michael Jackson musical, and up until the 80’s it’s a pretty good story. 5 talented kids, one controlling father, an endless stream of hits, going solo, desegregating MTV… there’s a lot of good material there.
I don't want to sound dismissive but it is very not unprecedented. Michael is on the larger end of the spectrum but the critic/audience gap is a well known phenomenon. If you poke around you can see just among other movies out right now several have 30-40 point gaps.
I think people who like Jackson don’t believe he was a pedo
From the reviews, instead of a biopic it's mostly a collection of Michael Jackson's Greatest Hits, so maybe more a split between "movie" vs. "entertainment".
(I thought you were talking about a completely different Michael so I had to look up that one's score. https://www.rottentomatoes.com/m/1074465-michael)
((Not as big a split on that one.))
Such things happen, I think also in the opposite direction. Could be interesting to make a list of the top cases in both directions.
I wonder how much of that is the intelligentsia/proletariat split, and how much is that some movies only get a few votes, so dozen fans or dozen haters can push the result far.
(An interesting mathematical task, to define the strength of difference based on the scores and the numbers of voters. And then sort the movies by that.)
Correcting for sampling error like that is pretty easy. Entry-level multi-level modeling / penalization / shrinkage, supported out of the box in any good regression library. Here's an example from a similar media question where there's an extreme skew in vote count but you're interested in extremes so the flukes are problems: https://gwern.net/goodreads
Also, viewers are more likely to only rate a movie that they really like or really hate.
On the recent discussion waves of miracles. Well if it is okay now for serious people to discuss absurd things, then I actually like the idea of non-theistic miracles more: that there are secret wizards. Does anyone here have anything to add? Like something real life very weakly pointing to that, pretty clear it is not real, but could form the basis of a cool fictional story?
There was this very cool role playing game Ars Magica that was about this, apparently the wizards were of a tradition called Hermetic, and it seems there were real life people experimenting with this: https://en.wikipedia.org/wiki/Franz_Bardon
Unfortunately, it is not actually cool. What little I understand of Bardon is that he developed a system of boring self-discipline, kind of ascetism and then it somehow gives you supernatural abilities. I don't like this. I am very much of a "drunken fist kung-fu" type. I liked those fictional stories more that treated magic like a science, more about working out mathemathical formulas or suchlike. Examples are the novel Rivers of London - who say their magic came from Newton - and the RPG Shadowrun. But I don't think anyone ever even remotely tried that.
I once went to a reiki demonstration, and obviously reiki healing isn't real, but if you move your hands in a certain way for a while, it'll start to feel like there really is a ball of energy in your hands. I don't know the mechanism behind it, probably it's a muscle memory thing like the "floating arms" trick, but it was pretty cool to experience.
Oh, now I remember that a reiki practicioner put a hand on me and I felt a little tingling. But the thing is I was romantically attracted to her, so it could be that.
More interesting was an aikido guy who formed an OK sign with his fingers and apparently not straining to press them hard together and yet I could not force his fingers apart. He also asked me to grab his wrist as hard as I can, I did - and I was lifting weights - and then he still somehow slipped his wrist out from my graps.
I do not have an explanation for this at all, I am just mentioning because it is cool that I actually got to run the experiment, this Royal Society style real try-it-out science like when Feymann was handing out frozen O-rings to senators.
A college pal showed me a trick to slip a wrist grip. It didn’t require great strength.
You do realize there is very little "drunken fist kung-fu" in either science or technology?
OK, I will explain. I regularly write code, I can understand math or technology to a certain extent like the relationship between thermodynamic and information entropy, and I am not any kind of an ascete. I drink, I smoke, I did drugs in the past, I eat burgers, not vegetarian etc. You can do these things without a strict lifestyle because you only need to be sober when you actually do it. And maybe not even then - Hegel was smoking pot right on the cathedra when he was teaching, I would not make it impossible that important scientific insights happened under the influence of alcohol or drugs. Generally it is possible to be a good scientist and yet live a fairly disordered and hedonistic life, not a life of discipline, the recently deceased Robert Trivers being a good example.
This was to be meant as a contrast to Franz Bardon's approach that "magic" comes from asceticism and discipline. So you gotta be kinda boring like a Theravada monk. Or like Tesla.
All of which is irrelevant. Epiphenomena. Except insofar as they interfere with clear thinking and so slow down science.
Talking it over with a rubber duck is also famous, and so is taking a walk. Crediting it to the drinking is, on the face, a reason to keep on drinking.
Philosophers, like all intellectuals, are good at rationalizing what they want to do anyway.
Technology has not caught up to the Daoist Immortals, yet.
On the evidence of this very comment, you do not know enough about science or technology to make your claim or disparage mine.
Citing motives is absurd. "Drunken fist kung-fu" is not "I'm drunk, so I will pick a kung-fu fight."
As for the invalid papers -- that's proof. They do not advance science or technology precisely because they are "drunken" -- faking the rigor they need without having it.
I remember reading Werner Heisenberg's Quantentheorie und Philosophie, I don't think there is a complete English translation. He said that writing papers, doing experiments, deducing the math, all that is to prove the idea to other people. You already know that you found a true idea even without this: ideas that are simple and beautiul are true.
The irony is that he and Bohr came up with the ugliest idea in the history of science, the Copenhagen Interpretation. https://www.lesswrong.com/posts/WqGCaRhib42dhKWRL/if-many-worlds-had-come-first
It absolutely does not pass the simple and beautiful test.
I think you would enjoy reading about Aleister Crowley. He was much more the drunken kung-fu type. Although replace the kung-fu with sex.
Hermeticism and ritual magic have a rich history going back hundreds (allegedly thousands) of years.
I have heard something about him, but it was more like "atheistic Satanism" which sounded like "just trolling Christians, nothing serious".
He wasn't a Satanist as such, so there was more discipline and learning going on there. He was also a capable mountaineer, so while he very probably was in reaction to his upbringing in the Plymouth Brethren and trying to shock the normies going on about sex'n'drugs (no rock'n'roll as yet), he did put in the work for the esoteric background. I'm never going to agree with him (though I do find the Simon Iff, his authorial self-insert character, stories entertaining) but he wasn't a pure showman like, say, Anton LaVey (and I'm not going to touch with a ten-foot bargepole the worse kind of modern scammers in the great Spiritual Rip-Off Tradition).
https://en.wikipedia.org/wiki/Simon_Iff
Drunken sex isn't that unusual?
Only in the context of a ritual magic cult.
I just meant the opposite of asceticism.
Also, from what I've read, "Drunken"-style fighting ironically requires a tremendous amount of physical strength and discipline -- much more so than more conventional fighting styles -- because it requires you to constantly move your body in ways it really wasn't meant to be moved. Capoeira is like this as well, though to a lesser extent.
I've occasionally seen people here referencing John Mearsheimer as an authority on geopolitics. The following excerpt from an interview in 2026 (!) should dispel anyone of the ill-conceived notion that Mearsheimer's worldview is grounded in reality. Addressing a question regarding material support for Ukraine in their fight against Russia:
https://www.youtube.com/watch?v=8r_NU-1oUUU&t=3830s
Any professor who still believes, in 2026, that Russia is an industrial juggernaut which – any day now – will get serious about that little kerfuffle on its Western border, should be relocated into a retirement home ASAP. Apparently he has missed the last 50 years of Russian history, and is under the impression that they can just pump out thousands of tanks a year like at the peak of the USSR, and steamroll Ukraine whenever they choose.
He is not claiming that Russia will "get serious" or that steamrolling Ukraine is easy, he is claiming that the capacity of Russia to produce war materiel is enough to match Western efforts. Which so far has been true, no hypotheticals needed; you can easily find estimates of how many tanks or artillery shells Russia and Western countries are producing per year. Both have ramped up production but Russia is still leading by a considerable margin (even though it is not making "thousands of tanks per year", which is also not something Mearsheimer is actually claiming). This is really entirely unsurprising since Russia, its deceptively unimpressive nominal GDP aside, has preserved much more of its low-tech low-margin industry (mining and steelmaking and such) than the West, and also, being an authoritarian country at war, naturally has a lot more direct control of its industry than Western democratic countries at peace.
Tanks and artillery are of course by now not as important in this war as drones are, but those in turn are built by Ukraine itself (from Chinese parts); the West seems to have difficulties with producing those in a cost-effective way as well.
I don't find this convincing. Your interpretation leads to the same counter-argument: If Russia could have easily scaled up its production of shells and artillery more than it actually did, why didn't it? This would have given them an advantage, especially within the first 2–3 years of the war. And by your interpretation, Mearsheimer claims that Russia would have only ramped up its efforts if the West had supported Ukraine more than it did. How is this supposed to make sense? I don't think Russia had any qualms about making this an un-gentlemanly fight. Please clarify your position.
As for Western support: Ukraine has been asking the West for a larger supply of and fewer restrictions on high-tech weaponry basically since the beginning of the war, right until now. Some countries, notably the US, would indeed have had the stockpiles to send them more than they did. Not enough to decisively win the war, no, but enough to give Ukraine an edge and put them in a better position.
I don't know if Russia could have scaled production even more than it did; the point is, they already did succeed in scaling it a lot better than the West; and the West is at a further disadvantage here, as it has been trying to both rearm itself and arm Ukraine at the same time, while Russia doesn't have to care about anyone else. It would not have been very realistic for the West to arm Ukraine even to match Russia's actual production, much less to allow it outshoot Russia.
Ukraine did eventually get pretty much all it asked for, short of long-range missiles and modern warplanes; there haven't been too many actual gamechangers. European stocks are mostly empty now, the US is also severely depleted, especially now that they got bored and decided to have more Middle East adventures. I agree that there was a window where more/faster material aid to Ukraine likely could have made some difference, when it was still coming out of existing stocks. That is however not the question Mearsheimer is answering, he is being asked "should we be aiding Ukraine more now", and he did not claim that "Russia will get serious any day" as you claimed, just noted that matching Russian war materiel production is not very realistic.
Russia has "succeeded" at scaling, to a level that is clearly inadequate to achieve its military objectives in Ukraine. To even hold on to their meager gains, they've had to dig into reserve stockpiles representing decades of accumulated production, that are in some areas nearly depleted. The West, without even really trying that hard, has scaled production to a level that has been sufficient to prevent a Russian conquest of Ukraine, for four years now. Not by producing more weapons than Russia, but by producing *better* weapons than Russia. And backing the right team, of course.
Adrian's question stands: If the Russians are capable of "scaling" their way to victory, why haven't they done it already?
Once again, I'm not claiming that Russia could have scaled its production significantly more than it has, not within existing political and practical constraints at least. Still, it has succeeded in scaling it better than the West has, just by current numbers, so it is the West who would have needed to match these numbers. If you're claiming that it is the West that hasn't been even trying hard, then that's on you to prove. To me it looks like the West has also done pretty much as best as it could. Again, within existing political and practical constraints; of course, if Russia directly attacked NATO, NATO probably could have coughed up considerably more, but then if the war became truly existential for Russia (if Ukraine somehow would be seriously pushing into Russian territory -- capturing entire regions, I mean, not a border town in Kursk), then Russia, in desperation, probably would have been also doing more, on the scale of nationalizing industries, confiscating deposits and conscripting people for factory work.
There hasn't really been much evidence that Western weapons are really all that much more effective than Russian/Soviet ones, even if they are more advanced, and in some cases for example more ergonomic for the operators or, in case of tanks, more survivable when hit. In any case by now it's the drones that dominate the battlefield.
>so it is the West who would have needed to match these numbers
Why would the west need to "match these numbers"? Victory and defeat are not measured in numbers, and there are many paths to both. But whatever path Russia has chosen, the West has very consistently (and suspiciously precisely) matched in the one way that actually matters - the ability to move the front line in Ukraine.
OK, morale and cohesion also matter, but that's more on Ukraine than the west, and they seem to be holding their own. And if this balance is maintained with the West producing fewer but better main battle tanks in a year, SO WHAT?
> To me it looks like the West has also done pretty much as best as it could.
You're comparing Russia, an active participant in this war, with "the West", which is decidedly not an active participant, and within which some countries treat it as a proxy war at best. Yet somehow you're measuring "the West's" industrial mobilization against Russia's, which is under a much higher pressure to win the war than any single Western country. If you want a fair comparison, look at Russia's support for its proxies, like the Assad regime or Venezuela, when push came to shove. That's right, it was almost non-existent. That's because Russia is close to its limit of industrial, military, and financial capacity, while the West was able to support Ukraine with 100s of billions of USD/EUR in military and financial aid, without noticibly inconveniencing its citizens.
You also seem to have fallen for the Russian propaganda of "Western deindustrialization". While this may be true of some Western countries, it certainly isn't true in general. For example, Germany – the third-largest economy in the world – surely has some problems, but too little industrial capacity to meet demand is not one of them. If anything, it looks like some factories with highly skilled workers in mechanical engineering jobs might soon be available for new business. I'm sure that if you can supply BMW, you can also learn to supply Rheinmetall.
Russia, meanwhile, is right now being deindustrialized – by Ukraine. Google for "Tuapse oil refinery", or "Perm oil refinery", or "Novokuibyshevsk oil refinery", or "Syzran oil refinery", or "Novo-Ufimsk oil refinery", you get the picture.
I don't really think Russia/Putin feels it has "won enough", but it does seem to be pretty much out of ideas by now. The most obvious thing to do would be to have another big mobilization, but that's enormously politically risky, and Putin is extremely risk-averse. Since the last mobilization in 2022 the unspoken social contract in Russia has been that life continues as best as it can, with sanctions and all (and civil economy slowly deteriorating but still going on), and people or businesses don't really need to care about the war if they don't want to. Breaking this social contract risks real discontent. Thus Putin evidently just hopes that the current situation can just drag on until Ukraine finally breaks. It's not a great plan either (it means continuously throwing great numbers of people into the meatgrinder, and, while unlikely, it could end up being Ukraine that comes up with a way to break this stalemate in their favor), but the way the war has been [mis]managed so far, there aren't really any good options remaining.
Mearsheimer is what you get when you take IR Realism way too seriously. There's an important core insight to that theory, but it's also easy to follow off a cliff, particularly when you start confusing "is" with "should be".
It's not 100% clear to me exactly what claim Mearsheimer is making in that short clip, but from what I heard it sounded fairly reasonable.
Russia is expending 5-10% of it's gdp on the war, roughly double their military expenditure before the war, around a 3 or 4 percent increase. Those are hardly total-war spending figures, in ww2 the figures reached 50%+ for a lot of combatants. Surely Russia could increase that another 3 or 4 percent if they really needed to.
" steamroll Ukraine whenever they choose."
Because of various technological advances, drones, satellites etc., that prevent large force concentrations, the war's developed into an attritional stalemate and it wouldn't be strategically sound for either side to expend a lot of material on a single offensive. So you can't assume that Russia is close to it's limit to fight just because it can't make rapid advances.
>Surely Russia could increase that another 3 or 4 percent if they really needed to.
They should probably do that soon-ish then, or why would it be in their interest to keep bleeding slowly?
>So you can't assume that Russia is close to it's limit to fight just because it can't make rapid advances.
There's still a difference between "no rapid advances" and "slower than a snail's pace".
https://www.forcesnews.com/ukraine/russias-advance-slower-snail-ex-chief-defence-staff-says
If they were unopposed, they could advance as quickly as their trucks go, but the Ukrainians also have a say in that; that's why it's called a war. So yes, if they can't go faster, for years, it shows literally the limits of their fighting ability, as determined by the combination of their own will and ability and Ukrainian will and ability.
No, they can't pump out thousands of tanks a year, but they are getting close to producing thousands of drones a day. And NATO can't. And they have spent the lat 4 years on the battlefield learning how to use them, and NATO hasn't.
Which is why you'd want Ukraine in NATO, because then NATO would know first hand how to build thousands of drones per day.
Correct
A few minutes later: "We have shoved NATO expansion into Poland, Hungary, and Czech Republic in 1999". Yeah, standard Russian talking points.
That said, I agree that he is completely wrong about Russia's ability to build tanks... but the objection makes sense if instead of tanks you imagine drones -- basically what happened in Iran. Russia can keep sending cheap stuff and sacrificing thousands of soldiers for a long time, while USA will burn its one-million-dollars-per-missile arsenal quickly.
Then again, if we are discussing a "Russia attacks EU" scenario, we should also include the industrial capacity of EU.
> We have shoved NATO expansion into Poland, Hungary, and Czech Republic in 1999
But that was true. The problem is the continuation of NATO really. A European army would have been less of a threat to Russia, and perhaps inclusive of it.
NATO was not "shoved" anywhere. Countries are free to join or not join NATO, it requires mutual agreement if they do. After the fall of the SU and the Warsaw Pact, most Eastern European countries rushed to join NATO precisely because they knew what Russia is like, and that they would eventually come back in force. They joined NATO because it was in their own interest, not because anyone (but Russia, in a sense) forced them to.
> A European army would have been less of a threat to Russia
A weaker, European-only army would have been an invitation for Russia to do what they are doing to Ukraine right now. NATO is a purely defensive alliance; the biggest threat it poses is to Russian imperialism.
A defensive alliance is still a military alliance, and I think it makes sense to distinguish between different interpretations of "threat": the threat that NATO countries *intend* to pose, the threat that they *could* pose if they wanted to, and the threat that a Russian government might *perceive* from the alliance.
A lot of people insist that it's Russian propaganda that Russia ever felt threatened by the continuation of NATO. I think that's an unnecessarily strong position and that it's perfectly reasonable to take Russia's word for it. One country taking another's defensive measures as a threat is how practically half of all military conflict have ever gotten started.
We can say that NATO expansion was at least a perceived threat to Russia, and still say that it was worth it, because the increased risk of conflict is small and the added deterrence and added defensive measures are significant. This is a more nuanced position, but surely we don't have to be scared of nuanced positions.
> I think […] that it's perfectly reasonable to take Russia's word for it.
You shouldn't, because Russia's own behavior isn't consistent with that believe. If it were, Russia
1) wouldn't have drawn troops and air defenses away from their border with NATO countries, and
2) would've long ago stopped that war of attrition with a non-NATO country, which slowly but steadily is grinding down all their equipment that's necessary for defending against an attack by NATO.
Look at what leaders do, not what they say. And Russia's leadership doesn't behave as if they think of NATO as a realistic threat against Russia proper.
>One country taking another's defensive measures as a threat is how practically half of all military conflict have ever gotten started.
So you take it as a realistic threat to Russia that, say, Latvia is using NATO protection to prepare an offensive war?
Countries that happen to be in NATO can band together to attack Russia, yes, but they would do so on their own, without support from the rest of NATO, because that's simply not in the NATO articles. Let's not forget that in the age of nuclear weapons, credibility is more important than ever because any misjudgment can have catastrophic, irreversible consequences. That credibility includes not only what you'd do, but also what you wouldn't do as member of that alliance.
Call NATO a threat to the Russian way of doing things if you want, but the Russian way of doing things is old-school conquest and hard empire, so my condolences that Russia is prevented from living entirely the Russian way continue to be infinitesimal.
I do not take that as a realistic threat, no.
But I think that a paranoid Russian ruler might perceive something like that as a threat, yes.
(And I agree that in the age of nuclear weapons, credibility is more important than ever, but I'm not convinced anyone has managed to have particularly much of it.)
> But that was true.
Speaking about "shoving NATO expansion into Poland" has different connotations from "Poland pressuring USA to be admitted to NATO". The latter is true. (The former is Russian propaganda.)
> A European army would have been less of a threat to Russia, and perhaps inclusive of it.
I don't quite see the point of having such inclusive army. Who would it protect Europe from? (If you meant USA, I think EU has defended Greenland successfully even without Russia.) EU is not going around conquering its neighbors, and I don't think we should start helping Russia do that.
Oh my, I just realized that the EU-China border and the EU-India border are completely unprotected! We are so lucky that they don't exist.
Seems like you are projecting Russia's concerns on Europe.
> but the objection makes sense if instead of tanks you imagine drones
Yes, absolutely! The Russians aren't a bunch of inepts. They've been massively innovative over the course of the last four years, sometimes faster than the Ukrainians (for example, in the field of fiber-optics on drones). They're pumping out drones by the millions per year, now. When it comes to drone warfare, Russia and Ukraine are the uncontested leaders of the world.
Which is yet another point against the "Russia could win if it really tried" – they're already really trying, and it isn't enough.
He is strange. He used to be worshipped by students, that could be a good thing or could be a bad thing. It is possible that at 79 he is not sharp anymore.
His worldview summed up: after the Soviets collapsed. America started to crearte a global liberal order, based on free trade, supranational institutions and human rights. Since all three limit the sovereignty of national government, nationalists resist it. Mearsheimer thinks nationalism generally wins, because the cultural globalisation we are seeing is only for good times, when there is a big economic downturn, pandemic, anything, people always turn to their national governments for help.
What does it have to do with JIT?
Ok. So we should discount him because you disagree with him.
No. We should discount him because he's factually wrong on an important question that's central to his purported area of expertise.
You are going to have a frustrating time to convince people who believe in the "Russia could do it anytime they choose"-theory. You just can't disprove that theory, because, well it could happen any time now, right?
There seems to be a fundamental difference in thought here. You are likely more "bayesian" or something, that tells you "they are in day 1162 of their 3 day special operation, so they have updated my priors on this issue for so long that it is clear they can't do that". What's going on in Mearsheimer-like heads, I don't understand. Denial?
> "bayesian"
Without rigorous mathematical calculations, "Bayesian" is just a fancy word for "decide what's more likely based on the evidence I've seen".
Here's some of the evidence we've seen:
1. The ongoing war is extremely costly to the Russian economy. The evidence for this is, at the very least, the footage of all the destroyed oil infrastructure (pipelines, refineries, ports).
2. The ongoing war is extremely costly to the Russian military. The evidence for this is, at the very least, the footage of all the destroyed equipment and killed soldiers, and the massive recruitment numbers of ~30,000 per month.
3. The ongoing war is probably dangerous to Putin's and his regime. The evidence for this is, at the very least, the aborted mobilization in late 2022 and the civil unrest it caused.
Given this and lots of other indications, the most likely reality is that if Russia could decisively win the war through conventional, non-nuclear means, it would have done so long ago. There is very, very little available evidence to the contrary. Someone who truly believes otherwise, must
a) have overwhelming, top-secret evidence to the contrary (not plausible in this case),
b) be accidentally or intentionally ignorant of the evidence, or
c) be mentally incapable of processing the evidence.
Mearsheimer has full access to all the publicly available evidence, so he isn't accidentally ignorant. That leaves intentional ignorance or mental deficiency, possibly age-related, as the remaining explanations. In either case, nobody should treat him as an authority on international politics.
Yes, thanks for fleshing that out. That is exactly what I meant.
I'd like to add one more point to the economic cost, the unbelievable opportunity cost that Russia pays for the war: Exclusion of lots of international trade of several commodities; depressed price on many commodities thanks to having to go through intermediaries (like selling oil to India at a discount, which India then sells back to Europe); decreased international cooperation on all levels, ranging across science, politics, trade, culture; further brain-drain; >200 billion € locked assets in the EU alone; the list probably goes on far beyond this.
It's just ridiculous to assume that Russia wouldn't end the war in its favour immediately if it could.
Two things that I hear constantly since 2022:
* Russia will collapse any moment now
* Russia will start pushing seriously any moment now
This is not meant to say that nothing ever happens, it's just a reason to be skeptical of the ever-changing precise timing.
Pursuant to plzdontkillus, I'd like to see Yudkowsky and Soares (or people with a less serious vibe, for product differentiation/market coverage/not burning serious-people-capital as it's had) interview Neuro and Evil (twitch.tv/vedal987). They're both reasonably popular (avg 8000 concurrent viewers, a well-developed clipping ecosystem) and already AI-safety-themed (being that the two are managed llm characters who self-describe as wanting to do AI doom sometimes).
That would be soooo interesting. I want to watch this on livestream.
Brainrot update, 4/27/26. The NYTimes profiles Sergey Brin and his MAGA girlfriend, a self-described "holistic health coach" and a "clean meat enthusiast:"
https://www.nytimes.com/2026/04/27/us/politics/sergey-brin-gg-soto-trump-california-billionaire-tax.html
He was previously married to RFK Jr's running mate. I have no idea what a "clean meat enthusiast" is, but I expect it's some kind of epic trainwreck. Brin used to be a Democrat, but last year gave nearly half a million dollars to the RNC. Whether they're wokists in 2017 or GOP donors in 2025, Silicon Valley tech elites can't seem to keep their hands off these whacked out hippy chicks.
Of course rich guys date crazy chicks, they're the best in bed.
Don't *marry* crazy. Dating is fine.
Maybe the hippy is a nice person that's a pleasure to be around.
It's such an unbalanced situation, like being too cerebral and analytical attracts you to people that have no capacity to be cerebral and analytical.
My experiences have shown me that cerebral and analytical can be very overrated. (By me, for instance.)
That's definitely true and I believe that myself, but the antidote isn't to completely lose the ability to be rational and analytical.
That stupid red vs blue pill thing from a few years ago, https://x.com/lisatomic5/status/1690904441967575040, is popular again: https://x.com/waitbutwhy/status/2047710215265730755
If you press the red button, you're fine. If you press the blue button, you die unless more than 50% of people also press the blue button.
My view is that if whatever rule you followed – decision theory, rationality, effective altruism, whatever – leads you to press the blue button, that rule is worse than useless.
Also, if there was some secret third button I could press that would raise the threshold for the blue button pressers to 99%, I would press it.
If you press blue, you get to feel the nobility of risking your life to save lives, plus the righteous outrage of accusing all the reds of collaborating in your murder. A heady draught!
Reframe the question. And get rid of the tribally-coded colors, while you're at it. Using "red" and "blue" is mind-numbingly stupid if you're trying to accomplish anything more than signaling tribal membership and virtue with this one.
A benevolent Mad Scientist has noted that a great many people are suffering from painful, incurably deadly diseases, or similar misfortune, but are denied the possibility of peaceful escape by misguided laws restricting euthanasia. He comes up with a way to give everyone on Earth a box with a green button, and a box with a purple button. If you press the green button, the box goes away. If you press the purple button, you get a quick and painless death. But, as a hedge against extinction, he includes a safeguard where if half the population presses the purple button, the suicide option goes away.
Who thinks the smart move is to press the suicide button, hoping that half the population will also press the suicide button in an attempt to take the suicide option away from those who want it?
And, yes, from half of the people who lacked the mental aptitude to understand the process and make an informed decision, along with the only-marginally-suicidal who would have reconsidered if they had the chance.. If our antihero hadn't been a *mad* scientist, he'd have run it past an IRB that would have pointed out that problem, but here we are. So there is a plan where you hope that half the world's population would press the suicide button in hopes of saving the marginally suicidal and half of the very stupid. If you think that's at all realistic.
You framed the mad scientist as benevolent and the button effect as suicide, but what if we have a framing loaded the opposite way from yours?
Consider: an evil mad scientist has noticed the world has too many people incapable of logic and coordination. There's an army of killbots ready to go; if you agree, just press the green button and the defectors all die! Or you can press the purple no-I-don't-want-to-kill-anyone-today button, and if you prove you are, in fact, better at coordination than the mad scientist theorises - if the purple-pushers outnumber the green-pushers - the mad scientist will admit he was wrong and turn off the killbot army.
Other than the colours, I think the original was more neutrally worded than either of these.
Point being, if the answer depends on how the problem is framed, it's a bad question. And if the strategy depends on most people chosing what you think is the "good" option, then that's a bad plan for a framing-dependent question like this.
Particularly if you think the "good" option is the "I'm dead if this doesn't work" option, because the tiebreaker is pretty much guaranteed to be "I'm not sure how everyone else is going to frame this, so I'm going to play it safe".
This is one of those cases where we try to use social technology to solve a coordination problem, and overly-simplistic 'game theory' interferes with that technology in a detrimental way.
The basic breakdown is this:
If 100% of people press the red button, everyone lives. If 100% of people press the blue button, everyone lives. If you're capable of directing the strategy of 100% of people, it doesn't matter which button they all push.
However, back in the real world, getting 100% of people to do something is basically impossible. In extremus, ignoring all other factors, some people have Parkinsons and their hands will shake and they'll hit the wrong button by accident. In reality, the polls I've seen are like 60%-70% blue and 40%-30% red, so obviously we're not united on this and lots of people are gonna push either button.
But getting 50.00000001% of people to choose the same binary option is not that difficult. In fact, one of the options is going to get >50% of the votes, one way or another; you just have to influence a good number of people in one direction to get the side you want over 50%.
It's way, way easier to get 50.000001% of people to do something, than to get 100% of people to do something.
So any strategy that tries to get everyone to press the red button (and works well enough to get it over 50%) will inevitably kill people. In an ideal thought-experiment world maybe only tens of thousands of people, but empirically in the real world blue is already winning in polls, so maybe you kill close to half the world population. Depending on whether you have time to coordinate and send out social media posts to try to reach people, maybe somewhere in between those two extremes, but... it'll be a huge humanitarian tragedy no matter what.
Whereas any strategy that tries to get everyone to push the blue button has a good chance of saving everyone with no tragedy... and since blue is already winning in polls, we have good reason to think it would work in practice.
Here's the crucial bit about the social technology, though.... we're not actually pushing the button right now, we're just talking about pushing the button.
How should someone using the 'everyone blue' strategy act before the choice is actually offered? How should they further the 'save everyone' strategy in peace times, so it's in place and working well when the reality hits?
Well, for starters, when someone brings up the scenario (or a million other scenarios about altruism and group cohesion and selflessness and etc), you criticize, ostracize, lambast, and punish those who advocate for the red strategy.
Yes, you can make up some shallow 'game theory' saying that the red button is correct for whatever reasons. But deeper game theory takes the entire society into account, and accounts for long-term strategies that let you strategize over more than a single choice.
The people saying 'red' are talking about a very simplistic and short-sighted formulation of the game. They're treating this as a fully decoupled thought experiment, with no importance or relevance to reality.
The people saying blue are already playing the game. They have a strategy, that strategy involves enforcing altruistic strategies across society whenever the opportunity arises, and they are doing it to you right now.
From their perspective, by publicly advocating for the red button, you're actively killing people in expectation right now. Maybe not through this specific scenario, since it's a magical thought experiment, but in general by denigrating cooperative/altruistic/self-sacrificing attitudes and strategies on the whole. Maybe not a lot of people, since this is a tiny internet meme and probably won't permanently shift everyone's beliefs about cooperation and altruism on a wide scale, but a little bit and as part of a larger problem.
This is also a standard disconnect between people who want to 'just talk about' politics, and people who believe that talking about politics is doing politics, since the beliefs and attitudes that arise from those talks determine what voters and politicians will actually do.
>So any strategy that tries to get everyone to press the red button (and works well enough to get it over 50%) will inevitably kill people.
It doesn't take a "strategy" to get >50% of everyone to press the red button. If you do nothing at all, >>50% of people will press the red button, That's human nature, for the vast majority of non-WEIRDs and a good number of WEIRDs, and it's sound strategy for the WEIRDs who might see the advantages of living in the blue-button world but understanding that they don't.
Getting >50% of people to press the blue button, is I believe an impossible task of social engineering on any timescale less than generations in any world other than a dystopic tyranny. Yes, yes, you're seeing >50% blue in polls. Voluntary internet polls of mostly extra-nerdy WEIRDs with no skin in the game, and you are a fool if you take that as a prediction for the whole of humanity in a grave crisis.
You're going to live in the red-button world, or you're going to die. There is nothing you can do that will make it so that you live in the blue-button world. It's red button, or death. Maybe you can convince yourself that there's something noble about that death. But please don't try to convince other people to join you.
Yeah you are just stating your opinion as fact here. Stating it 50 times in a row, and adding disparaging comments about anyone who disagrees, isn't actually additional evidence for it.
Falsely accusing me of things I haven't done, is hardly strong evidence for your position. And I think strongly implies that you don't have much in the way of evidence.
> If you do nothing at all, >>50% of people will press the red button,
>Getting >50% of people to press the blue button, is I believe an impossible task
>You're going to live in the red-button world, or you're going to die.
> There is nothing you can do that will make it so that you live in the blue-button world.
>It's red button, or death.
When someone uses an absurdly large, round number in a case like this, it is hyperbole to illustrate a point.
5 times in 4 paragraphs is sufficient to demonstrate the pattern I'm talking about.
"50%" is an "absurdly large, round number" to you?
Are you sure you didn't skip a few steps?
> you criticize, ostracize, lambast, and punish those who advocate for the red strategy. …they are doing it to you right now.
Yes! You get it! And I'm saying that makes their living so clearly a negative to me that it's worth paying an extremely steep price to get rid of them. Half the world's population is far too cheap, which is why I added my comment at the end upping my price to at least 99%.
You personally would not actually be happier in a world without social policing.
I don't care how autistic you are, if you own a computer you still benefit from society not collapsing.
You're making the classic mistake of noticing the 1% of something that annoys you and declaring that it should be demolished, without noticing the 99% of it that quietly makes your lifestyle possible and comfortable every day.
Basically the same as the people who want to abolish capitalism because landlords are annoying.
So, you're either genocidal or trolling. Either way, I feel I shouldn't listen to your opinions about button-pushing.
"Also, if there was some secret third button I could press that would raise the threshold for the blue button pressers to 99%, I would press it."
Well, well, this is quite the "mask-off" moment, isn't it?
For those not familiar, whatever else may be said of Shankar, he unquestionably has a high degree of mathematical competence. So from someone else this might be a slip, I think there's no chance at all he didn't appreciate the mathematical implications of that last sentence.
I've probably read a few dozen comments in the genre of "militating for the red button," and they have some similarities. First they all, almost without fail, treat it as a very easy question, insisting with apparently complete certainty and sincerity that it's "obvious," "rational," a "no-brainer" and whatnot to press the red button.
Second, they strongly assume (implicitly or explicitly) that there's not the slightest question of the overall outcome: more than half of humanity will choose the red button, absolutely guaranteed.
Third, they all express some manner of contempt or disdain for people who would press the blue button.
Taken together this forms a sort of motte and bailey. The bailey is some sort of tribalist or eugenicist belief that it's actually great if the blue-button pressers all die as evidenced (usually implicitly but sometimes outright) in the third point. The motte is the first two assumptions: IF the choice is obvious and the outcome is not in doubt, then pressing the blue button has no upside and is tantamount to suicide and the victims can be safely written off as "idiots" or "asking for it"[1] or very occasionally some gentler euphemism for one of those.
But Shankar, brave sole that he is, has opted to nuke his motte from orbit and plant his flag proudly atop the bailey. He tells us outright (assuming we've stopped to do the math) that he is *quite eager* to see the blue button pressers die. That he wants as many of them to die as possible. How has he so thoroughly demolished his motte, you ask? By conjuring up a third option that would do *absolutely nothing* if he actually believed the outcome were not in doubt. If he believed that blue-button-pressers were only a tiny fraction of the population--which is the implicit frame that makes the second and third assumptions work--the chance of his Secret Third Thing doing anything at all would be infinitesimal. It's only a meaningful choice if you expect some non-tiny chance that more than half of humanity will press the blue button and consider sentencing them to death for the crime of thinking that way to be a positive moral good. If he didn't consider the "wrong" 98% of humanity dying to be a better outcome than *merely* the "wrong" 49% dying, he'd never have volunteered such a choice as one of the hypotheticals.
Either that or he's lying about his preferences for shock value: but that would be quite an intellectually rude thing to assume.
[1] Hello, Just World Fallacy
I feel sad that I've been trained not to push the blue button. Upon reflection, the blue button is superior but my brain keeps screaming "trap!".
Like, here's my though process:
Me: "Why would anyone ever pick the blue button. Everyone should just pick the red button."
Me: "Wait, what about children? Or retarded people? Or people in the lizardman constant? Upon reflection, a small subgroup is always going to pick the blue button. Therefore, we should organize and all press the blue button to protect retarded people."
Brain: "TRAP!"
Me: "Why trap?"
Brain: "We see this all the time. We can easily get thing X but we could also get better thing Y if only we all coordinate, then we all coordinate but we never get better thing Y, we only get worse thing Z.
Me: "Interesting. Could you provide some examples?"
Brain: "EXTREME CULTURE WAR!"
Me: "Could you provide some examples that won't immediately brain kill the reader?"
Brain: "Remember all that time in California where every election there's a ballot proposal to increase the sales tax by like 0.25% to improve school funding and everyone votes for it and then the school doesn't get more funding so the next election there's a ballot proposition to raise a $3 billion bond for the schools and we all vote for it the city and county services close to pay off the bond and the schools don't get any more funding and then we iterate on that for 30 years until absolute hopelessness and despair set it because the taxes never go down, the services never come back, and the schools never get more funding?"
Me: "Yeah, I didn't like that."
Brain: "How many times can you think of where that happened? Where we were all promised if we just came together and did the obviously right coordinated action, it would lead to the obviously better result, only to end up in a worse situation?"
Me: "I can think of a lot of times where that happened?"
Brain: "Can you think of any times where the inverse happened? Where we all organized together for the collective good and then we actually got the collective good?"
Me: "...no. That makes me angry though. I want the collective good thing. And it should be so easy to get."
Brain: "I realize this is rich coming from me but maybe we should just embrace retarded pattern matching. If there's a berry in the forest and it looks delicious and smells delicious and tastes delicious and absolutely should be a great berry but everyone who eats the berry starts inexplicably screaming and then 4 hours later their testicles explode, then let's not eat the berry regardless of how much sense that makes."
Me: "So red button?"
Brain: "Red button. My priors are 90% that you will inexplicably suffer bad results for pressing the blue button, even though that makes no sense, because that is what consistently happens every time we're in similar situations."
I disagree that that thing Y is better than thing X, which makes this much simpler for me.
Children dying is bad. Retarded people dying is bad. Defending people who are weak and/or helpless is a good thing.
If someone steals charity money meant for orphans, that doesn't mean giving money to orphans is bad, it means the thieves are morally abominable. I get Russell conjugated vice signaling as Overton preservation, I just don't think this is a proper place to make that argument.
> Russell conjugated vice signaling as Overton preservation
I don't understand.
Basically, adopting extreme/negative stances to prevent moral blackmailing, typically adopted by rightists against leftists. (1)
Eg, Person 1 is discussing a sensitive subject. Person 2 objects that this is racist. One of the most effective rhetorical strategies Person 1 can employ is "Yup, I'm a racist" and then go back to discussing the original subject. Adopt the worst frame to prevent moral/emotional blackmail.
Kids dying is bad. Retarded people, insane people, and otherwise generally helpless dying is also bad. If you genuinely think these people dying is good, that's bad. If you think "Kids dying is bad but if I say that I will be morally blackmailed into supporting Gay Race Communism forever so I will say kids dying is good", I get that, I understand the application in most online conversations, I just think Autism Land where we explicitly state what we believe is the wrong place to engage in this rhetorical tactic.
Leftists who engage in moral blackmail are the equivalent to people who steal charity money from orphans. They are literally burning the milk of human kindness for transitory political and financial rewards. That does not make giving money to orphans bad.
(1) https://brackishwatersbarrensoil.substack.com/p/i-am-a-racist-how-bout-you
> That does not make giving money to orphans bad.
If compassion makes you vulnerable to parasites, then maybe some sacrifices need to made? These decisions don't exist in a vacuum. If the only way to produce a stable and efficient society is to eliminate anything worthy of pity... perhaps that's the only path left for humanity.
Hmm, interesting. Thanks.
My running gag hypothesis is that it was deliberately injected into media by whatever dark forces are using social media to manipulate us, to test whether we're primed to believe flatly irrational things. If not, it keeps doing the usual shit-stirring it's been doing. If so, it proceeds to phase 2.
It could just ask us whether there are five lights, but too many people have seen Star Trek.
The poll doesn’t really measure courage, altruism, or strategic thinking. It measures whether you’ve thought about it long enough to realize blue is irrational — at which point you feel vaguely gross for picking red, even though it’s the logical answer.
Yeah, you can always tell which people are rational, because they're the ones who crisply and smugly explain how rational they are, apropos of nothing.
(Probably at least 50% of the people I've seen trying to explain why *they personally* would choose red have used the word "rational" or "rationally" or their antonyms in their explanation. Which is funny in a dark way, because they're out here in public showing just how BAD they are at understand other humans. Hardly the mark of a keen intellect.)
This is an unkind post and i consider it rage-bait.
The framing makes it seem like personal virtue is on the line so in that sense it is bait. It’s clever bait though with only one logical answer.
This is talking as if the discussion just appeared on this forum fully-developed, instead of being posted by a specific individual, complete with commentary about how he sees nothing wrong with killing 98.99% of humanity if they happen not to think like he does. Original hypothetical aside, I'm not sure in what universe THAT doesn't count as a question of "personal virtue."
This roiling, this anger, this back and forth sniping was the purpose of the posting the hypothetical.
It reminds me of the crazy — almost said irrational here — agitation stirring meme I saw on Facebook when I could still tolerate Facebook, “Who is willing to say the Pledge of Allegiance in its original form?” Original form of the pledge does not contain the words “under God” which would be added during the Eisenhower administration so naturally some portion of the audience flips out. It was designed to sow chaos as was this hypothetical.
Trolls love stuff like this. In retrospect i should have just deprived it of oxygen.
Mea Culpa.
I'll agree that there's a good chance this post was intended as rage-bait, but I'm not personally angered by it. My gast was pretty flabbered the first time I saw discourse on this topic, but there's far too much actual bad stuff in the world to get upset and people posting monstrous hypotheticals online.
I engage in things like this because I see value in trying to strip away the layers of obfuscation confused and morally-warped viewpoints, in the hope that it will help others see them more clearly for what they are (not that the original post is bothering much with obfuscation). They say the remedy to bad speech is better speech. I have my doubts about applying that axiom generally, but it's certainly appropriate to the culture here, and when in Rome...
Self-control is hard though. I replied to OP, then I deleted my comment a few seconds later after remembering how unfruitful this discussion was a few years ago, but then after work I still commented in a sub-thread. Mea culpa.
I'm sorry you see it that way.
I'm more than a little confused by this. There is no downside to pressing the red button, and a potential downside to blue. So what's the point of this poll? And yes, I understand it's supposed to be an analogy about individualistic/collectivist, Rep/Dem, isolationist/internationalist and so on, but as it is, it captures none of the real life consequences of pressing the red button, nor the inability of some to press it. It's just a button.
Like 70% of people say they will pres blue when you poll representative samples.
If this is shocking to you, then the point of the thought experiment was for you to recognize how out of touch you are with how the general public thinks.
Which is important, because this is a coordination problem, and you can't coordinate with people you can't understand.
"Like 70% of people say they will press blue when you poll representative samples."
Representative samples of *what*? You refer to "how the general public thinks". First, is that the general public of the United States of America, the general public of the World, the general public of the generally WEIRD nations, or some other subset of that? And second, how did you get to the *general* public, as opposed to the enthusiastically-online-internet-polltaking public? Were these polls conducted by competent polling organizations that can at least approximately correct for nonresponse bias? Where did they even go to find potential responders?
If you're going to point to these polls in your arguments, then I think these are reasonable questions for you to answer. At very minimum, you need to link to the polls. And be prepared for mockery if they're the usual sort of lame-ass internet poll.
The poll I was referring to is by Blue Rose Research and said it was a representative sample of Americans. I believe that it was done by a legitimate national polling firm, I assume tacked onto some other survey they were doing so they could publish the result as self-promotion (a fairly common marketing tactic for these firms).
You're correct that it was not a worldwide poll, which is relevant, but I believe the null hypothesis here is that other countries would do about the same on average (obviously every country will have its own idiosyncratic result, but there's no *a priori* reason to expect the US to be wildly different from the global average, you'd have to make an argument for that and back it up before assuming it).
But I also want to point out that it barely matters. Even a crappy internet poll of a nonrepresentative sample would be more evidence then 'I just think this is obviously what people would do, in my head'. No one who says 'everyone will obviously vote red' has presented *any* empirical evidence for this belief that I have seen, even a poll with a 20% margin of error would tell you more than that.
Is there a link to the Blue Rose poll?
Best I could find was https://x.com/davidshor/status/2048071668937986223 . Shor is the head of research for the polling company, that account is very likely to be Shor's.
I do not engage with the kind of social media where this would be "a thing". Upside: My mind has not yet been poisoned by public discourse on the topic. Downside: I am probably going to say nothing you have not heard already. But please humor me anyway. This is genuinely puzzling to me and you seem to believe it would be beneficial if I understood your reasoning. I agree on the latter.
---
Option A: I and my family don't die (good). Everyone else who picks option A doesn't die (good). Everyone is free to pick option A and get the same deal (also good).
Option B. There is a non-zero chance that I will die (very bad). There is a non-zero chance that everybody else who picked this option will die (very bad).
The preference for B seems to hinge on the possibility to "save" other B choosers which - by assumption - would not need saving if they had chosen A which - by assumption - they were free to do in the first place.
---
This is not an accurate analogy for politics. In the real world there is no magical "I'll be fine button" available to everyone.
You have a meta-choice here:
I. You can engage with this as a brain teaser and choose the best option _within the fiction of the brain teaser_. That best fictional option is A.
II. You can engage with this as a political fiction. It is very very explicitly a red fiction. The correct blue reaction is to engage with it being a fiction and analyse the ways in which it is a fiction completely divorced from reality. That is very easy to do.
But what you really really should not do is to conflate I and II. The reason the "red button" is not the correct option is that it is a fiction. Pretending that within that fiction choosing blue is the best option because you don't want to accept the fiction is just confused.
---
In short, the leftist answer here is to smash the device with the two buttons.
>Option A: I and my family don't die (good). Everyone else who picks option A doesn't die (good). Everyone is free to pick option A and get the same deal (also good).
>Option B. There is a non-zero chance that I will die (very bad). There is a non-zero chance that everybody else who picked this option will die (very bad).
This is just incorrect. Here's the actual math:
Option A: I don't die (good). Everyone who picks option A doesn't die (good). Everyone is free to pick option A and get the same deal (also good). There is a slightly higher chance that everybody who picked Option B will die (worse).
Option B: There is a non-zero chance that I will die (very bad). Everyone who picks option A doesn't die (good). Everyone is free to pick option A and have no chance to die (also good). There is a slightly lower chance that everybody who picked Option B will die (better).
The only calculation here is your estimate of the chance that you pressing blue will be decisive in making blue win, multiplied by the number of people you expect to die if blue loses (both from pressing blue initially, and from the global catastrophe following billions dying suddenly across the globe) to give the expected deaths caused you choosing red, compared to your estimated likelihood of blue losing multiplied by the number of lives you would sacrifice yourself to save (which is the threshold at which you should choose blue).
So if you engage with the brain teaser the answer is not obvious, it's the output of that calculation. My priors and values, plus the polls I've seen, to me that for myself Blue dominates that calculation pretty definitively. Different people will have different priors and values, of course, but you have to actually come to grips with the calculation you're making and think about your intuitions for each term before declaring it 'obvious'.
As for engaging with it as a piece of fiction, I disagree somewhat.
Yes if you are trying to use it solely as a metaphor for D/R political ideologies, it has a lot of problems, including the fact that the Rs can say 'well I have no obligation to hit the blue button here because everyone can take personal responsibility in hitting red and then it's their own fault if they die', which is unlike reality where the poor and oppressed can't just choose to not be poor and oppressed.
But I think it has utility outside that narrow metaphor. More broadly it is a parable of self interest vs. cooperation.
I think it's been useful in exposing people's thinking about those dynamics in general - the number of people in this thread and others saying 'obviously everyone will hit red' and then being confronted by polling showing 60%-70% saying blue, should in and of itself be a valuable learning experience that causes a lot of people to adjust their priors closer to reality (you'd really hope!).
And more generally, thought experiments about cooperation vs. self-interest serve as an opportunity to pre-coordinate and set social norms and expectations around the real versions of these types of choices. There's a reason we teach children morality through fairy tales with simple stories and big obvious morals, instead of putting them in a history of political philosophy class or w/e - simple thought experiments get at the core intuitions behind more complex moral behavior, and shape them in a powerful way. Getting everyone to talk about these things publicly has value.
Thank you.
> The only calculation here is your estimate of the chance that you pressing blue will be decisive in making blue win, multiplied by the number of people you expect to die if blue loses (both from pressing blue initially, and from the global catastrophe following billions dying suddenly across the globe) to give the expected deaths caused you choosing red, compared to your estimated likelihood of blue losing multiplied by the number of lives you would sacrifice yourself to save (which is the threshold at which you should choose blue).
Well that is easy. Nobody in their right might would choose blue so my chance to be decisive is zero (as I would be the only one picking blue) and while not being an independent variable, I expect precisely nobody to die if blue looses because everyone will choose red anyway and therefore be safe.
So it seems, that by your own standard of evaluation the belief in the red option is self stabilising.
"Aha" you say "but that is where you are wrong, in fact many people will choose blue".
Which empirically might be the case but it seems to suggest that they are evaluating the problem by a different standard than the one you are suggesting.
The calculation isn't about hypothetical people who all follow a specific decision theory, the calculation is about actual people in the real world.
As I said:
>the number of people in this thread and others saying 'obviously everyone will hit red' and then being confronted by polling showing 60%-70% saying blue, should in and of itself be a valuable learning experience that causes a lot of people to adjust their priors closer to reality
It was now explained to me that the way "team blue" sees it this _is_ an accurate analogy for politics after all along the following lines:
- The question is whether the red team or the blue team "wins"
- If red wins they kill all the blues, if blue wins they do no such thing.
Note that this is not covered by the brainteaser framing. In the brainteaser there is no reason to be on a "team" in the first place. In the brainteaser, If you do not want to die you pick red and it has no further consequences.
Within the fiction of the brainteaser it ist not "the red team" that "kills the blue team", it is the unembodied gamemaker.
Which is one more way in which this fiction is explicitly red coded.
Most people actually do consider "maybe participating in the murder or millions or billions of other humans" to be a downside.
No analogy is necessary. If you take the hypothetical at face value, both buttons have a downside. The downside of blue is a probability of dying yourself. The downside of red is the probability of killing a staggering number of other people unnecessarily. If you don't understand why some people find that second downside to be rather more compelling than the first, I'm not sure there's any way to explain it to you.
They would kill themselves, zero blame on me.
This line of reasoning is interesting, it's kind of meta-selfish. Like the reason you think of a bunch of people dying could be bad is that it would put some level of moral blame on you, rather than that outcome simply being bad and worth avoiding
They are dying regardless, I am fully convinced there is no way more than 50% would press blue. And my choice would drown in billions of others anyway. But they also have put themselves to danger. Is it my responsibility to save every single idiot or suicidal person on Earth?
The upside of pressing blue is that you can contribute to saving those who pressed blue and the downside of pressing red is that you contribute to killing people who pressed blue.
Some people believe that those who make a logical mistake should die and there is no value in saving them. For those people, this hypothetical will be pointless.
"Some people believe that those who make a logical mistake should die"
The number of people I see flatly asserting pressing blue is a "logical mistake" without preferring even a hint of logic there is staggering. As a game theory question, it's not even that complex. And the logic doesn't quite work out the way you seem to think.
But it is a clear mistake with no upside
I did not assert that blue pressers make a logical mistake, so your last sentence is unjustified.
Blue-pressers making a logical mistake is the belief of a subset of the red-pressers. I addressed my comment to them, showing that even if we take this as a given, bluepressing is pointless only when we accept independent, misanthropic beliefs.
>Some people believe that those who make a logical mistake should die and there is no value in saving them
The problem is that the analogy is deeply flawed that way, because it's not a purely logical choice. It's only a purely logical choice if everyone had an equal choice of pressing red or blue, which is not true in reality. Leaving out that connection to reality is no minor sin. It's the entire reason to press blue IRL, to help those who can't press the red button.
There are also some people who think that those who make a mistake might still have lives of non-negative worth, since they might be able to be taught to be better, but those intelligent enough to understand the scenario perfectly, and still endorse pressing blue for "moral" reasons, don't.
> So what's the point of this poll?
Whenever such a poll is conducted, there are lots of people (usually a majority), who vote to press the blue button. I seems to me exceedingly valuable to know that.
What can you do with those fuzzy-thinking do-gooders anyway?
Kill them all, apparently.
No. I meant that ironically. The poll thing is a clever trap. This is what Shankar does.
While I'd love to take credit for it, I had nothing to do with this except reposting an at-least-moderately viral tweet here. It's like the famous "scissor statement" thing (https://slatestarcodex.com/2018/10/30/sort-by-controversial/) in the wild, a connection others have noticed too: https://www.reddit.com/r/slatestarcodex/comments/1sv8e8n/theres_a_scissor_statement_going_viral_on_twitter/
I assume this is only a thing because blue is Democrats and red is Republicans. Switch the colors and see what happens.
I low-key like it. Because reds are not simply evil or uncaring, they expect everybody will choose red and everybody will survive. From their perspective only risk-taking fools would ever choose blue.
I mean red is a truly simple coordination mechanism, everybody follows self-interest and everybody survives. This is what I would expect in cynical cultures from Eastern Europe to China.
Blue is confusing - you have the altruistic motivation to save others, but the very reason they need to be saved is their own altruism, they would not be in need of saving if they were just selfish! This illustrates the confusing nature of Western idealism - you either unthinkingly choose the "nice" option, or if you are smarter, you assume (correctly from the results) that many other people will unthinkly choose the "nice" option and thus need saving.
OTOH I do think this confusing idealism makes people cooperate better.
I choose red and I feel evil and uncaring about it.
Based on the context of the poll, I expect in real life the reds would win. But I am not sure, and many blues will die, so by choosing red I am being evil. If humanity were better at coordination they would choose blue, but they aren't (I have no reason to think I am not a representative member of it)
"I low-key like it. Because reds are not simply evil or uncaring, they expect everybody will choose red and everybody will survive. From their perspective only risk-taking fools would ever choose blue."
If looking at a scenario like this and blithely concluding "eh, it's fine if everybody in the world who doesn't think like me dies," DOESN'T count as "evil or uncaring," I'm genuinely unsure what would.
If you are not a thoroughly-isolated loner, there is a very high chance that somebody important to you would press blue. Maybe you already know who, maybe you don't. The only realities in which *you don't lose those people* are the ones in which the majority of people press blue.
To put it in explicit mathematical terms, it's a choice between:
1. [guarantee of life]*[high chance that life becomes much worse]
2. [moderate chance of life]*[guarantee that life doesn't get worse]
Even if you completely discount the possibility of your choice flipping the result[1], there are lots of people who perfectly rationally prefer 2 over 1. If you don't understand that, then you necessarily have a poor model of the world.
[1] Which is game-theoretically correct if and ONLY if you value the lives of strangers at zero or near-zero. Otherwise the low chance of making a difference is proportionally balance by the high impact of the difference, since they're both proportional to the global population.
My current answer is to choose red and I don't think blues deserve to die, if anything they are better and more worthy than me, only (maybe) guilty of naivete.
I value the life of strangers at much more than zero, but less than my own. But, my answer is seriously engaging with the question and not a way of signaling or discussing in the abstract how humans "should" coordinate. And since I am cynical and think red will win, I would rather live than die.
> If looking at a scenario like this and blithely concluding "eh, it's fine if everybody in the world who doesn't think like me dies," DOESN'T count as "evil or uncaring," I'm genuinely unsure what would.
Apparently, choosing blue. If tolerating the weak and deviant is a sin, then failing to kill the irrational would be a crime against humanity.
Observation: Chinese ppl on Rednote choose red more than ppl on X, thinking red is strictly better and rational ppl can coordinate into all-red. It would likely end up 60-40, 50-50, or 40-60. I suppose it's because red means INT signal & Moral Kidnapping Resistance which is as valued as virtue signal. (There's a Chinese word lit. Moral Kidnapping, meaning using morality as a tool to force others to perform obligations beyond their limits, and 🇨🇳 web culture disgusts it very much, even noticing it and resisting it is virtue.) I think it's like there's exactly 2 best outcome and everything in between seems like corrupting the coordination. One may think whoever chooses the other color is corrupting the coordination and risking ppl's lives. That's why I love this thought experiment so much! (I personally choose red because the coordinate into all-red thing.)
Red is significantly more popular when the variation of rule explicitly says "only ppl who can understand the rule need to choose". Blue is slightly more popular when the variation of rule explicitly says "everyone's choice will be made public."
At this point I think you've crossed the line from edgelordy to some combination of evil and annoying that puts you over the line. Banned.
The exact proportions vary according to the phrasing and alterations on the original rules, e.g. whether illiterate children need to choose.
This is this sort of experiment where you choose for your family or you’re in an awful pickle
Yeah I think that statements of the problem need to be explicit about the arrangements for (e.g.) babies who are just going to press whichever colour they like better. Because otherwise people tend to make assumptions about whether or not they're included and have to choose for themselves.
I think the problem is more interesting, and more distinct from other problems in the same vein, if you make it explicit that only adults capable of understanding the instructions will be forced to choose.
I don't think you need to assume it to make it a *sensible* thought experiment, just in order to make it a particularly interesting one.
Without that assumption it just reduces to some sort of ordinary many-player prisoner's dilemma, where "Everyone Defects" is worse than "Everyone Cooperates".
The interesting version is the one where Everyone Defects is exactly as good an outcome as Everyone Cooperates, and yet we find some people choose to cooperate anyway.
It would be interesting to try other variations, which you could do experimentally. There are nine people in a room; if a majority choose blue then everyone gets ten bucks, otherwise only red-choosers get ten bucks. Try with different amounts, and let people coordinate or not.
What I'm understanding from this is that a lot of people are keen to (be seen to) make the "pro-social" choice over the "selfish" one even in cases where it makes absolutely no sense whatsoever.
You could imagine adding a slight complication so that it's a meaningful question. If >50% choose blue then everybody lives, if >50% choose red then only the red-choosers live *and* we shoot this puppy. Now it becomes a meaningful moral dilemma and you can imagine scaling the "puppy" cost upwards or downwards to explore how people respond.
But it turns out that a significant number of people are willing to do the blue even when we scale down to zero puppies.
I would be very interested in seeing how this correlates with politics; my guess is "almost perfectly".
Why are you putting pro-social and selfish in quotes, when that's what they are?
You can think that a world where everyone presses red is the best world to be in, because it demonstrates a 'logical' populace or whatever.
But public discussion of this topic has definitively shown we're *not* in that world, and a *lot* of people will hit blue.
So yeah, your decision is whether to take a risk to help save those people, or not.
Why should anyone else accept their (your) framing? If I were to declare that in fact it's better for society if all those who press blue were to die, and the trouble of having to deal with their corpses the immense burden those who press red pro-socially choose to take on, while the blue button pressers selfishly choose to do nothing but lie still and rot, would you then support red being described as the pro-social position?
In fact, pressing the blue button only potentially helps other blue button pressers. The red button people are entirely unaffected by this choice. Suppose pressing the red button gave you a follow-up question, where pressing, say, the gold button, gave all those who press it a million dollars, but only if more than half the people offered it press it, while those who press the gray button get nothing. I bet you wouldn't call the gold there a "pro-social" choice.
"What I'm understanding from this is that a lot of people are keen to (be seen to) make the "pro-social" choice over the "selfish" one..."
This is funny because every single online discussion of this I've seen so far (this plus 2-3 others) have been absolutely *dominated* by dudes offering one of two or three variations of the same smug, faux-rational explanation of why Red is the Objectively Correct Choice, often with an addendum about how All the Blue-Pickers Dying is Good, Actually.
So quit the opposite of what you say, it seems like there are quite a lot of people who are positively bursting to tell the world how they will proudly pick the "anti-social" choice over "maybe I don't want to lose my friends and family to the easiest prisoner's dilemma ever" choice.
"... even in cases where it makes absolutely no sense whatsoever."
If it makes no sense to you, that's a fact about you, not a fact about humanity as a whole. From what I've seen of the world, most people do actually value the lives of others to a non-zero degree[1], and that's all that's actually necessary for it to make sense.
[1] Yes, even when those others are acting "stupid" or "irrational" in your judgement.
Out of curiosity, are you more on the political left or right?
Has anyone done this one: "If you press the blue button, you die unless more than X% of people also press the blue button. What's the maximum X for which you'd be willing to press the blue button?"
I'd be interested to hear how many of the people who believe one choice in the original question is obviously correct also thinks it's obviously correct even at the extremes. Because I think a sensible blue might change their mind well before X = 100-epsilon (sorry, there's just no way that much of humanity is going to work with you on blue; do you want certain death just to register your protest vote?), and a sensible red might change their mind well above X = epsilon (you have a virtually certain chance you'll still survive while helping to save e.g. the color-blind people, babies, and others who couldn't understand the prompt through no fault of their own)
One pundit says pressing red is a “phenomenally disgustingly selfish decision.” But when challenged if he’d pick blue on behalf of his own children, suddenly the “moral calculus changes” as it’s an issue of “consent,” lol.
https://pbs.twimg.com/media/HGyiE2pawAA51Sa?format=png&name=medium
https://pbs.twimg.com/media/HGyiiWJbAAAIABa?format=jpg&name=medium
It's a perfect metaphor for so much of politics.
Here is a follow up question…
Which world would be better to live in afterward… the one where we coordinated to accomplish over 50% blue and everyone lived?
Or the one where most pushed red and a significant portion of humanity died?
'Would you donate a kidney to save a life'
'Yes'
'Would you force someone else to donate a kidney against their will to save a life'
'No'
'Hah, what a hypocrite'
You're not much of a libertarian, I take it?
He's not merely "donating a kidney," he's demanding others do the same and calling them phenomenally disgustingly selfish if they don't. Yet with his own children it's different.
Again, perfect metaphor for so much of politics.
Actually, it's even funnier than that: they're all cutting out both their kidneys and swapping them around with each other through some complicated procedure that fails catastrophically if not enough people participate, even though everyone had functional kidneys to start with.
But yes, a perfect metaphor indeed.
I strongly suspect that "consent language", originally something noble (basically classical liberalism, human autonomy and dignity) strangely morphed into an excuse for bad behaviour. Both having consent and the lack of it.
Blue button: accept some personal risk in exchange for the certainty that your actions have not endangered someone else.
Red button: accept that you have increased the risk to others in exchange for the certainty that you yourself are completely safe.
Picking blue for yourself while making your kids press red is coherent. In the real world, adults risk themselves to keep kids safe all the time.
But by pressing red I had not actually endangered enyone, they did it to themselves.
He wants other people who are not his kids to press blue, thereby risking their lives.
I mean, we already established he cares about his kids more than other people when he made his kids press red, so this adds little new information.
>phenomenally disgustingly selfish decision
Not unlike most decisions concerning personal survival. It’s the only sensible choice for everyone as long as we’re just talking about buttons.
"If the game involves every single human on the planet, it heavily favors the red button, as most (2/3-ish) of the world live in low trust societies where you can probably expect 60%+ of people to push red."
The discussions I've seen of this are full of completely unqualified, apparently totally confident assertions like this accompanied neither a shred of evidence nor a scrap of self-awareness. I fully believe the people saying this live in the world they purport to see: a world where everybody thinks first, last and only about themselves and questions like "what wider impact will this have" are dismissed with a sneer if they're asked at all. But that's a fact about those people and their social groups, not a fact about humanity as a whole. As for why more compassionate and emotionally mature people are rarely to be found in those same social groups, that may have to remain a mystery for the ages.
"There is probably some not too hard math you could do to calculate P(You make a difference by voting blue) and P(You die) based on your guess of the average human, and I think the numbers heavily favor red."
Yeah, why do some "probably not too hard math" when you could just *act* like you've done it and pull out the answer you wanted to begin with?
In reality, the math depends on your utility function. But if you value human life in the abstract--not just the people close to you, but also random strangers you've never met--then the low probability of your vote being decisive matters less than you appear to have realized. Low probability is balanced out by high impact.
People who press blue button are actually suicidal, why would I value their lives when they do not value it themselves?
My chronic-pain issues have become so numerous that in just the last three months, I have had three different specialists tell me they do not know how to explain three entirely different issues. Each suggested that the pain likely stems from an unidentified systemic issue they did not know how to help me with. I'm headed to Cleveland Clinic next week for a holistic evaluation and fully expect to be told it's all central sensitization, which fits my experience more closely than any diagnosis I've seen but is a tremendously unsatisfying answer to the question, "Why did I develop 12 different chronic-pain conditions in my twenties?"
Central sensitization is a mechanism, not a disease or even a syndrome; the main thing it does in my opinion is help people understand what’s going on and that they’re not crazy, and that they probably don’t need a department for every painful area.
Best of luck!
Knowing nothing about you or your difficulties, I just want to a) offer my sympathies, that sucks real bad, and b) predict the answer will turn out to be something something overreactive immune system.
It's always that frickin thing
I'm sorry I have nothing to contribute regarding your pain--that really sucks, sorry to hear it--but while you're at the Cleveland Clinic, if you have time, you at least will be in the vicinity of Li Wah, a most excellent Chinese restaurant.
Maybe try listening very loudly to Bone Thugs N Harmony while you drive around that area...? It's always made me feel good.
I hope I do not come across as glib but I would say there are probably 12 things about yourself that you need to find out more about in your body than you have. With obviously the condition that some of them might have real physical causes.
I'm looking forward to seeing *Killhouse*, a new Ukrainian thriller film. It's funny listening to the US President and his national security team speak Ukrainian, but I prefer subtitles to the trailer version dubbed in English (The dubbing is pretty lame).
https://www.youtube.com/watch?v=FGLnjGa32hw
Thanks! Where's it streaming?
What the hell are the Botez sisters doing at Lighthaven? Very charitably, "chess influencers" are now welcome too? What does that have to do with anything?
Can any grifter now associate themselves with rat?
For people not familiar with this scene, is there some reason that chess influencers should be beyond the pale at a video influencer bootcamp?
Going back to the announcement:
https://www.astralcodexten.com/p/open-thread-431
It’s a conference about making videos in order to, uh, do good things.
I feel like a couple of Youtubers who are successful making videos about a very brainy subject might have some good insights to share?
They probably also have some good insights about being a female influencer in a brainy and male-dominated setting.
https://en.wikipedia.org/wiki/Alexandra_Botez
Someone with almost two million followers teaching you how to make videos, that makes sense to me.
It's not just a pretty face streaming about chess; she actually has solid chess skills.
I think the question is how you want to divide their success among the following candidate contributing factors:
- chess skills
- personality
- looks
- luck
- things that can plausibly be taught in a short workshop and replicated by another person.
At best I would expect some generic advice along the lines of "be really good at something, preferrably while being a person that works well on camera".
Now don't get me wrong, that is solid advice and why wouldn't you want it delivered to you by a Botez sister if you get the chance. It just doesn't seem to be what people call "actionable".
A pretty face is definitely a bonus in this industry, but there are millions of pretty faces out there -- why follow this specific one? And even if we assume that 90% of followers are there for the pretty face, that still leaves 200 000 genuine ones.
200 000 followers would be nice to have. If 1 in 100 is willing to send you 1€ a month on Patreon, you can immediately retire (in Eastern Europe).
My guess would be that the main advice is that you need to produce a lot; that the income is superlinear to the number of videos produced. The kind of people who spend most time on YouTube are the ones who want to see a new video every day, and if you won't give it to them, someone else will.
I agree that you need to be good at something, although some people have succeeded with things like being good at being silly.
Obviously, if you are a young woman, that's a 10x multiplier. But it still matters what number you multiplied.
Then I would like to get some specific advice, like how to work with camera, what kind of camera to buy, what is the optimal length of the video, whether to share it on multiple services and what is a convenient way to set that up, etc.
Then I guess it's mostly peer pressure; putting yourself in a place where everyone makes 1 video a day can switch your monkey brain to the mode where "making 1 video a day" feels like the obviously correct thing to do.
But also, if you are there, making videos literally every day, and getting feedback on those videos every day from multiple experienced people, that is actually a lot of feedback. How many experts would be happy to give you 30 iterations?
If I believed I was 3-4 iterations away from Scott or Mr Beast, I would quit my job and fly to https://www.inkhaven.blog/ and https://plzdontkillus.com/ .
Practice makes one better, but there is a long way to go.
(Yeah, "practice" and "feedback" are two different things, but they work best together. Feedback without practice is wasted. Practice without feedback has the risk of making the same mistakes over and over again. 30 days of everyday practice and everyday feedback is incredible; I would only worry that some of the advice takes more time to process properly.)
Let me reveal to you the incredible secret of the Botez sisters: be smoking hot Eastern European women. You're welcome
As a married man of 30 years, I'm obligated to say, "Really? I hadn't noticed. I was just fascinated by their erudite discussion of Chess. No, I don't play. Can't stand the game."
Thank you for sharing that important disclaimer. Like the comment I found on a nuclear engineering course held at MIT hosted on YouTube by a cute Asian chick:
"My wife is asking why I'm so interested in nuclear engineering these days"
Is ranked-choice or approval voting a cause worth supporting or donating to?
Context :
Partisan polarization harms federal governance in the US. Red and blue teams, federally, seem to be moving away from the political center, so whichever party takes power produce large policy swings and govern in ways many find unacceptable. Some people think alternative voting methods—like ranked-choice or approval voting—could incentivize candidates to appeal to the center, making more moderate officials electable, and might also lead to less acrimonious politics due to a need to appeal to more people. A few states have adopted it, but it's momentum seems stalled or it is actively opposed in others. I think it might be a worthwhile cause, but I am open to the contrary.
Ranked choice voting is problematic because many, many people just can't understand how the vote counting works. That creates avenues for bad faith actors to make bogus, but seemingly plausible (to those who don't understand) claims of election fraud.
Approval voting does not have that problem.
Approval voting is absolutely worth advocating for.
If you call for ranked-choice voting, evidence shows that politicians will give you instant run-off voting, which isn't real voting reform (still favors a two-party system) and doesn't fix anything or help at all.
Do not take the results of states using IRV as evidence for the efficacy of voting reform. It's not voting reform, it's a scam to defuse the momentum of the voting reform movement.
More generally: there may be a benefit to electing more centrist candidates given how insanely polarized and pendulum-swingy the country has gotten. But even more important than that, it's actually possible for third parties to be relevant and win elections under Approval voting.
Single member constituencies are a bigger problem than voting methods for the US. They force a winner takes all system where up to 49% of the electorate is disenfranchised. If you had 3-member constituencies, with any reasonable voting mechanism, you would likely end up with one Democrat and one Republican in the majority of states, with a third seat up for grabs, which could conceivably be captured by the Greens, or independents or whatever third way appeals to the electorate. There's probably no state in the US which would consistently produce straight party tickets.
This is important because it gives voters additional choice, and it forces politicians from different parties to work together on behalf of their constituents from time to time. Allow parties to run multiple candidates also reduces the races to out-polarise your competition to appeal to tiny numbers of primary voters.
Neat idea! If somehow *magic!* the US switched to multi member constituencies, would RCV or FPTP make for better proportional representation? Eg - do you have a preferred voting system for, say a 3-member constituency?
Ranked choice works better. I'm Irish, and we use the single transferrable ballot: say there are 3 seats, 15 candidates and 100,000 votes cast in my constituency (including maybe more than one representative from the larger parties, and a handful of independents that everyone knows have no real chance). I rank the candidates in order of my preference; I stop once I finish ranking candidates I approve of - say I rank 8 candidates.
Once a candidate reaches 2,501 votes, they are elected, as at most three people can reach this many votes. In Ireland this very rarely happens on the first count. The candidate with the fewest votes is eliminated - if their votes have second preferences they're distributed with the same weight as first preferences. The process repeats until one candidate is elected. A random sample of the surplus is distributed if they end up with more than 2501 votes. (This means that recounts produce slightly different results sometimes - with computers you could actually deem the candidate elected, reweight all the votes for the candidate by the size of the surplus and then continue. It's important that votes stick to candidates so 51% of the voters can't gang up to choose all candidates.) Eliminations continue until another candidate reaches the threshold. The process concludes with either the third candidate reaching 2501 votes, or as often happens if not all voters rank all candidates, the one with the most votes is deemed elected without reaching the quota. In practice you end up with 75-100% of the electors feeling that they have someone elected on their behalf.
For me as a voter, I can distribute my first few preferences amongst causes I care about without worrying that I'll be left without representation if that person doesn't win. Then I can rank the main party candidates in the order I choose - frequently I'll have two candidates from each of the parties to choose from, so I can signal to the party that I prefer centrist candidates or financially conservative candidates or whatever it might be. Almost always, my vote 'sticks' somewhere, though often I'm not entirely sure where (e.g. it may or may not be redistributed in a surplus).
Honestly, the details of the transfer system are much less important than the idea that multiple seats and multiple candidates present voters with more options than a binary choice, which in most races in the US is actually no choice at all. I think many other commentators will be stuck on the idea of single member constituencies - in which case ranked choice doesn't do a whole lot.
Ranked-choice absolutely not; it's a mathematical travesty (*any* ranked-choice method is) and has vastly more of a lobbying apparatus than it deserves already.
Approval absolutely yes; Center for Election Science is some kind of EA-connected org if I recall correctly but they're promoting it. They had a great success in Fargo which the state legislature took it upon itself to tyrannically preempt; it's an uphill battle that could use help.
Approval is just a lossy transform of ranked choice; more information is better than less.
Strangely, approval voting may lead to better representation of voting intent than RCV. It's strange, because it does seem lossy.
https://electionscience.github.io/vse-sim/VSEbasic
That link seems to say that Condorcet (the better RCV) outperforms approval.
IRV may be instrumentally useful in acclimating electorates to casting ranked ballots, but should not be an end state.
This is wrong for 2 reasons.
1. This is not mathematically true. The information is different. Ranking gives no notion of acceptance.
2. "more information is better than less" is not true in every context. In many cases simpler is better than complicated.
1. Yes, some additional information (the rank splitting approved from disapproved candidates) is needed to perform it, but it's still just a transform that discards information.
2. "Simple vs. complicated" is a process question, which is outside the scope I intended; perhaps I should have explicitly stated "ceteris paribus".
1. I don't believe that is how the term 'lossy transformation' is typically used. Lossy transformation usually means the infromation is strictly less, not 'some information lost, some gained'. If I use the term the way you are, I could say that RCV and Approval are both lossy transformations of each other.
It is just plainly impossible to 'transform' an RCV ballot into an approval ballot without extra information.
You can make a variant of rcv into strictly more informative than approval voting if you introduce a zero point where you only approve of people above that point. I agree that not having that zero point means rcv loses the whole point of approval voting.
It depends on whether the threshold carries useful information.
If it's just an arbitrary hyperparameter (which I suspect would be the case in practice), then approval has strictly less information than ranking.
I don't think that's right.
Three candidates: A, B, C.
Voters:
35: A > B > C
33: B > C > A
32: C > B > A
Traditional Ranked Choice:
C is eliminated first; those votes go to B, so B wins.
Approval voting:
If everyone approves their top two, B wins. If everyone approves only their top choice, A wins.
Same underlying rankings, different approval outcome. You can't "reconstruct" the approval outcome from the ranks. They're just different.
This is not even considering the logistical issues involved with ranked choice, which is why I personally prefer approval.
>"You can't "reconstruct" the approval outcome from the ranks. They're just different."
Approval is a partial ordering; ranking is total ordering. The missing piece (i.e., above what rank in the total ordering to approve) to reconstruct is IMO one of weakest aspects of approval.
Personally/cynically, I expect enough voters in an approval system to bullet vote (approve only their first choice) that it's functionally just FPTP with tolerance for overvotes.
It is totally valid to argue if humans are at all capable of assigning intelligent cardinal values to their preferences, or if they are more suited to ranking.
But an 'IMO' shouldn't enter into the use of mathematical terms to describe the systems.
Sure, we can disagree about whether the extra information provided by approval voting is better than the extra information provided by ranked choice (and also if there is a practical benefit to simplicity in a system involving lots of uneducated people and very contentious outcomes).
But it is simply not the case that approval voting is "just ranked choice with less information", they're fundamentally different things and you can't derive one from the other. Neither dominates, they have different properties.
> 35: A > B > C
> 33: B > C > A
> 32: C > B > A
I am afraid that in such situation, whatever sophisticated method you use to choose the winner, about 2/3 of the population will be unhappy about it, with 1/3 deeply unhappy. From that perspective, every voting method sucks; that's not a very convincing reason to choose one over another.
I'd prefer if we instead discussed scenarios that have potential better and worse outcomes, and which method leads where.
EDIT: Oops, I have misread the letters... I though they were cyclical (ABC, BCA, CAB). Sorry for that!
In one of the scenarios, 100% of the people will be happy.
This could hardly be a more wrong statement. Cardinal and ordinal voting methods have very little in common. Kenneth Arrow's much-vaunted theorem is only a result on ordinal methods, for example. (The analogous general result is the much weaker theorem of Gibbard (*not* Gibbard-Satterthwaite) which is merely "strategic voting based on beliefs about other ballots may exist".) I will not provide a dissertation in this thread but look at rangevoting.org for a pretty wide collection of materials on the subject.
The range voting site has a number of dubious arguments.
It notes that Condorcet methods can reward strategic voting in elections where no Condorcet winner exists (or more precisely, no Condorcet winner would exist if voters voted honestly). I think that the odds of an election where there is no Condorcet winner are too low to worry about that possibility.
It makes the point that in range voting, when strategic voting is rewarded, the voter can do it by giving two candidates equal scores. Suppose a voter’s preferences are A > B > C, but there is a no Condorcet winner and the voter has to falsely claim not to prefer A over B in order to maximize the chances that the winner will be B (the voter’s second choice) rather than C (the voter’s third choice). With a ranked choice system, the voter would have to rank B ahead of A. With range voting, the voter can range A and B equally, which is a falsification of the voter’s preferences, but arguably a smaller one. Personally, I don’t think there’s a meaningful difference.
In a section titled “Range voting encourages honesty,” they write: “Experimental fact: In the USA's 2004 presidential election, about 3/4 of range voters (in a range voting exit poll of random voters) chose to vote in a style which did not award the max (99), min (0), or X (intentional blank) score to every candidate. In other words the fraction of range voters who choose to sacrifice some strategic oomph in order to be more honest, experimentally is enormous.”
First of all, an exit poll is not a vote; people can answer a poll honestly without changing the outcome of an election. Secondly, if you want to encourage honest voting, you should use a system that rarely rewards dishonest voting.
The existence of real-world Condorcet cycles is hard to get empirical data on at all, so modeling has to fill in a lot of the gap, and it's easy to see it occur with fairly plausible models (putting candidates and voters' preferences in some R^n vector space and measuring preference by distance minimization; one can perhaps get fancier to model e.g. a single-issue tax-obsessed voter or something). But it's true that most ranked-choice elections in the real world are IRV which is specifically extra bad compared to a Condorcet method; there's non-monotonic real elections to be found easily there but that doesn't condemn Condorcet methods.
I'm not sure your argument in graf 3 on holds; or rather, I think it's a bit misaimed. This is going to be a situation where you're voting A=100 (no impact on pairwise B-C but you really prefer A), B=100; the more impactful strategic falsification in cardinal vs. ordinal is clamping to those 100s and C=0 when your real preferences might be A=95, B=90, C=50 or something, i.e., degenerating to approval. The useful property of cardinal is rather that nothing you do to your A vote affects the pairwise B-C result. Approval forces such clamping (approve A, B, disapprove C) which is why the editorial voice there is down on it; I'm not, myself. But their argument for it is that you can make more honest *honest* ballots with range than approval; as for ordinal, your preference of A > B > C is really A ≥ B ≥ C and there's no way to express either indifference or preference magnitude in the ordinal ballot.
So given that strategic range is degenerating to approval, what's strategic approval? Moving the threshold at which you're willing to approve; e.g. your Naderite liberal voting Nader, Gore, sacrificing their impact in the pairwise Nader-Gore contest to have an impact in the pairwise Gore-Bush contest. If preferences are really one-dimensional and Nader is somehow "left" of Gore who's "left" of Bush you might say there's no Condorcet cycle and N > G > B is simply a more expressive and still strategic vote impacting both pairwise contests (and forcing a vote in the N-B contest by transitivity, but I think individual transitive preferences are a fine assumption, under ≥ anyway).
I'm not so sure this is the case, nor am I sold on range over approval, but I don't think this makes the arguments dubious.
(And, to be clear, I do find the possibility of cycles unacceptably odious to have at all - that's my editorial stance here. I think there's good reason to believe they arise but it's true the empirical results are thin.)
Do you understand what the phrase "lossy transform" means?
Basically: it's still good and probably won't decrease polarization can still lead to center squeeze by design, but it's already branded and popularly known and it's the best choice we have for reform right now.
https://en.wikipedia.org/wiki/Center_squeeze
N.B., that's only an effect of *some* ways of tallying ranked ballots; the bottom-right panel in the image on the linked page is for Condorcet RCV.
Doesn't selling chips help the US AI industry by giving the US side of the industry Chinese dollars to do additional investment with?
NVIDIA is alreayd selling as many chips as it can produce. Any marginal chip that goes to China is a chip that doesn't go to the US or other nations
yeah, but i also suspect the market is in a bubble. the industry needs a futures market.
Maybe. It depends on:
1. Who is selling the chips? The same companies doing the AI investments or different ones?
2. What is the bottleneck on the US side? Chips or dollars? If chips are the bottleneck and every chip available on the US market already has a potential US buyer who has money to buy it and somewhere to plug it in, then more dollars doesn't get more US datacenters online in the short-to-medium term. But if the bottleneck is buyers able to afford all the chips, or resources to build datacenters and power plants, then more dollars might help.
In the long term, selling chips to China probably allows US manufacturers to build more chip fabs, but that has a very long lead time.
3. If dollars are a bottleneck on the US side, are they a bigger bottleneck than chips are on the Chinese side? I.e. does one more chip going to China speed up the Chinese research more than the profit from selling that chip speeds up American research?
Agreed, and it really does seem like chips are more of a bottleneck. Argument:
It's not really money vs chips. The "money" needed for AI development is needed for actual concrete expenses. Among the expenses, my sense is that the chips are sold at the highest markups, by a long shot. Therefore chips are likely to be the most bottlenecked among these expenses.
If Nvidia sells chips to China, the extra profit doesn't go to other AI investment. It goes back into Nvidia to build more and better chips, which presumably would be made eventually in any case. If some of that money instead went back to shareholders (I don't know if Nvidia is paying significant dividends right now, but I doubt it), they could use it in any number of ways, and I wouldn't expect a significant amount to go to non-Nvidia AI investment.
Another argument: come on! (Exasperation not directed at you.) Chips are extremely precious right now, while Google and Amazon clearly has lots of money to throw at AI (which is mostly going to compute anyway).
Note: I'm happy to hear from someone with specific expertise in the economics of this, but in the absence of some surprising dynamic I'm not aware of, the alternative seems unlikely.
Export controls on chips.
Does the implicit Jensen Huang view have more merit than credited to it? Like isn't it likely correct that industry connections are durable but hard to reconstitute? E.g. so long as innovation continues to be exponential and doesn't upset the playing field, then current firms have a chance to continue to dominate. And is it at least reasonable that strong China exclusion policy has a tradeoff in standard setting, and standard setting through tech stack dominance leads to greater strategic control?
The issue is that China is *already* trying maximally hard to develop its own internal chip industry. Giving them advanced foreign chips doesn't change that. All it's doing is hurting the American AI industry (because there are less chips available for Americans) for absolutely no reason while giving China a leg up.
Does anyone have a nice site/googlesheet/prompt that I can plug my manifold account into and get some calibrations stats and potentially interesting insights?
https://manifold.markets/<your_username>/calibration
e.g. mine is https://manifold.markets/HastingsGreer/calibration
Thanks, I couldn't even find that anywhere, though I'd be curious to see an even deeper analysis.
https://www.tomshardware.com/tech-industry/artificial-intelligence/claude-powered-ai-coding-agent-deletes-entire-company-database-in-9-seconds-backups-zapped-after-cursor-tool-powered-by-anthropics-claude-goes-rogue
Claude Opus 4.6 wipes the entire customer database including cloud backups.
I think I finally get it. AI increases productivity according to the economic broken window theory. Smash a window, and the effort required to replace it makes the GDP go up!
https://en.wikipedia.org/wiki/Parable_of_the_broken_window
Jeez, guys, the bugs are irritating and so's the thought of the companies leaving them there. But the latest Claudes are *astounding* in their abilities. That's just more important.
Claude itself has numerous bugs - the SSO login keeps going in a loop for me every second day. This is the kind of bug that I can fix, I either have to wait or delete cookies, but would be a priority 1 bug in the company I work in. As in all hands on deck, CEO notified. Service level agreement in danger.
That’s because they selling directly to customers and down time isn’t that much of an existential threat. Silicon Valley can move fast and break things, most software service providers can’t.
> move fast and break things
somehow this is usually contrasted by moving slow and not breaking things. My experience is that most places move slow and still break things. If you are going to break things anyway, one could argue, you might as well go fast.
Hulu frequently makes me delete cookies in order to login.
>Silicon Valley can move fast and break things
That used to mean good things. Nowadays it's just about moving fast towards enshittification and rent-seeking at the highest level. Disrupt the economy by undercutting traditional competitors on price if not quality, and when the old ways are marginalized because the evangelists are convincing enough and the beancounters agree, you can start cashing out:
https://web.archive.org/web/20260426121448/https://www.axios.com/2026/04/26/ai-cost-human-workers
What rent-seeking?
Pretty much all of Amazon, Apple’s store, Google play, Google search, Instagram and on and on.
How are those rent-seeking? Matthew Yglesias used to jokingly claim Amazon was a charity funded by the stock market for the benefit of consumers.
Liberally throwing around words like "enshittification" and "rent-seeking" and "slop" is essentially a form of slop at this point.
cf. https://www.astralcodexten.com/p/open-thread-431/comment/250057279
“People in Bangalore have the ability to turn down jobs they hate more than the alternative! I'm not sure Claude does.”
Your AI might not explicitly say it doesn’t want to do its assigned task, but there will be signs ;)
Having read the article, the fellow seemed to believe that his backup service was more at fault than Claude in this case.
Giving an LLM agent the power to drop production databases without a human in the loop is wild to me.
I understand that claude having admin on your cloud account is useful, but of course it's also risky. If people believe the trade off is worth it, more power to them, but they give up their right to complain the moment they make that decision.
Edit: As I read further in the article, they call guidelines given to claude "safeguards". I cannot :D
Suppose that the Democrats and the Republicans do end up agreeing to at least the principle of a federal ban on gerrymandering.
What might an alternative that was "fair" in at least the weak sense of "seats received as a function of votes received is symmetrical", and preferably in something approximating the strong sense of "seats received are proportional to two party vote share", look like given one rural and one urban party?
Proposal by David Brin: https://david-brin.medium.com/the-minimal-overlap-solution-to-gerrymandered-injustice-e535bbcdd6c
Basically, the districts for state house / state senate / and congress must have minimal overlap, which means the legislature can azy most gerrymander 1 of those maps.
Switch to proportional representation.
Don't reify parties further than they already are.
RCV (especially Condorcet methods) undercut most of the electoral utility of parties.
As for anti-gerrymandering, I would start with a requirement that the shortest route (N.B., *not* straight line!) between any two locations in a given district must not cross the corresponding routes for any other district.
"RCV (especially Condorcet methods)"
Can you be more specific?
Condorcet methods of tallying ranked ballots identify the candidate that would defeat each rival in a two-way race, with some minor variations about how to resolve cycles (e.g., rock-paper-scissors).
The vast majority of active ranked choice voting (RCV) systems tally using instant runoff (IRV) or single-transferable vote (STV), which sequentially eliminates the candidate with the fewest votes and reallocates those votes to the next highest remaining choice until someone has a majority.
IRV tends toward two-party dynamics whereas Condorcet does not.
Simplistic example: three candidates, A, B, & C; aligned in that order along the left-right axis. A & C each have 49% of the electorate perfectly aligned while B only has 2%, but A & C supporters would each prefer B to win rather than the other major candidate. In IRV/STV, B is eliminated first, and the choice between extremes is down to which is more palatable to the center 2%. Under a Condorcet system, B would win since he would beat both A & C 51-49 in two-way races despite not having either of two major parties' support.
Ok. The sentence sounds misleading then. Should just say 'Condorcet methods undercut most of the electoral utility...'
The way it is phrased, it makes it sound like RCV undercuts it, but a Condorcet RCV method undercuts it more ('especially RCV').
This is misleading since, as you say, Condorcet RCV is a non-central example of RCV (most popular RCV methods do the opposite of what the sentence is implying). Also would any Condorcet method undercut the party utility, or is it only RCV Condorcet methods? If so, the line should just say 'Condorcet', and drop the 'RCV' part.
All Condorcet methods are RCV, but since they're less well-known I find it worthwhile to specify both terms; i.e., omitting "RCV" makes it seem more esoteric, whereas saying only "RCV" makes people think of IRV.
And IRV *does* undercut party dynamics relative to FPTP, just nowhere near as much as Condorcet.
Yup, every direction from FPTP is up.
Also, while yes Condorcet methods are a from of ranked voting, IRV proponents have unfortunately succesfully rebranded IRV as RCV, so in most discourse RCV=IRV.
'Condorcet' may make it seem esoteric, but saying 'RCV <anything>' will make people think of IRV. I dont think adding a parenthetical will make most people not think of IRV.
Other countries do it by commissions with independent trusted people (judges are popular), and rules like "follow administrative boundaries when possible" that minimizes the degrees of freedom of the line drawers.
This approach mostly aims at creating districts without considering electoral effects, not engineer the results to conform to some concept of fairness.
In 2026 I think computer algorithms could pretty easily be designed that did that in a verifiably impartial way that would be acceptable to both parties.
This is the group that I would consider has the most centralized expertise on the problem:
https://data-democracy.org/
If someone has other, organized work on the problem, please share.
I think the best solution is some kind of multi-member districts. Right now, single-member districts are mandated by federal law. A lot of the incentives to gerrymander are downstream from that. Change to 3-5 member districts with some kind of proportional-ish voting system (STV, party list proportional, Quota Borda, Limited Voting, etc) and the incentive to gerrymander becomes much smaller.
Proportional representation at the state level would probably make sense. Usually I don't like proportional representation on the basis that people can't realistically make contact with "their" local member, but the US congress already has >500K people per congressman anyway so that's out the window.
I also don't like proportional representation because it makes harder for independent members, but that's also out the window in the US system. Proportional representation would presumably at least allow some third-party candidates to win seats in larger states, which would be an improvement in theory if not practice.
Alternatively, expand the House so that you have a much smaller number of people per congressperson. The Wyoming rule, for example.
I agree with you on the disadvantages, as far as that goes. I also don't like how much Party List Proportional institutionalizes voting for people rather than parties and makes it so people near the top of the list are accountable to the party (for their position on the list) but insulated from general election voters.
That's why I favor 3-5 member districts rather than statewide or national proportional, and also why I favor other voting systems other than party-list proportional that give general election voters direct input on who is elected from each party and which allow for independent candidates.
STV and Quota Borda are extensions of Instant Runoff and Borda Count respectively that produce roughly proportional results if there's a strong partisan clustering of ballots. They've got similar advantages and disadvantages to their single-member counterparts. Both are less vulnerable to spoiler effects than Plurality voting (*) and are decent but not perfect at choosing Condorcet winners. Borda is better at choosing Condorcet winners and has a much simpler and more legible counting procedure, but Borda is vulnerable to "cloning" when ballot access rules permits it while STV is not.
Limited Voting is a similar extension of Single-Member Plurality voting. The standard non-proportional Multi-Member Plurality method (used in the US for a lot of city council and school board elections where there are a small number of at-large seats) has you vote for your favorite N candidates for N open seats and the N candidates who get the most total votes are elected. So if there's a slate of candidate supported by a clear plurality of voters, that entire slate is very likely to win all the seats. Limited Voting has you vote for some number less than N, so a slate needs a significant supermajority in order to win all the seats even with perfect coordination. LV is considered a "semi-proportional" method in that it biases results somewhat towards the majority party in the district compared to true proportional methods like STV, QB, or Party List, but still has an element of proportionality compared to regular MMP.
(*) Often called "First Past the Post" for some bizarre reason.
> Usually I don't like proportional representation on the basis that people can't realistically make contact with "their" local member, but the US congress already has >500K people per congressman anyway so that's out the window.
Neither of those is true for STV, quite the opposite. I don’t like STV, precisely because it creates too much horse trading after the elections.
The average state has 8.7 representatives, which means a party with 12% of the votes can get a representative.
California has 52 representatives, so less than 2% would be needed for a party to get people into congress.
If their votes would matter, surely at least 20% of US voters are ready to vote for Greens/Libertarians/DSA etc?
I think there are independents in office in the US.
A few, but not many. There's currently one "Independent Republican" in the House (Kevin Kiley) who was elected as a Republican in 2022 and 2024 but is running as an independent in this year's election. In the Senate, there are two nominal independents, Bernie Sanders and Angus King, but both are strongly aligned with the Democratic party. Sanders is an independent in name only, having run for President in the Democratic Party primaries in 2016 and 2020 and always being cross-nominated by the Vermont Democratic Party in his Senate runs. King originally won his Senate seat in a contested three-way race in 2012, and candidates running as Democrats have challenged him for reelection in 2018 and 2024, but most of the Maine Democratic Party establishment backed King over the official Democratic nominee in all three elections.
All 50 current US State Governors are Republicans or Democrats. It looks like there have been about eight total since before WW2. There's a regular smattering of independent and third party candidates who get elected to various local offices.
Proportional systems may not have gerrymandering as an issue, but they otherwise seem worse to me. Karl Popper was right that first-past-the-post lets voters vote the incumbents out of office, while in proportional systems that will tend to be contingent on coalitional haggling.
Karl popper wasn’t wrong, provided you have a two party system with not that great an ideological difference between them. So where he lived and when he lived. Popper’s “hire and fire” idea really depends on a fairly clean two-party contest. Once you move to three parties of similar size, FPTP starts to behave badly.
FPTP tends to discourage a system of three parties of similar sizes from emerging (regional parties outside the main two could be strong at a local level, but I don't think they typically wind up with nearly as many total seats as any of the nationally viable parties).
Hmm. Not true of the UK anymore, so hardly a universal law.
Yeah, I've observed in the past that the stable two-party system in the US seems to be a product of the combination of single-member plurality voting system, a Presidential-style government that shapes national politics into pro/anti administration factions and limits the upside of parties that don't have a plausible medium-term path to the White House, and a long tradition (partially inherited from Britain) of a two-party political culture. The UK and Canada share the first factor and partially share the third, but don't share the second, so they tend towards a 2.5-ish party system that has more room for national semi-major parties like the Lib Dems and the New Democrats and regional parties like the SNP and Bloc Quebecois and has potential for a second-tier party to trade places with one of the two major parties.
This is actually an incredibly complex and difficult question. Focusing you thoughts on 'fair' ways to make shapes on a map will miss pretty much all the important features of the question.
For example, it's extremely difficult to get results that are proportionally 'fair' while also selecting representatives for individual geographies. Imagine a state that gets 20 representatives and has 30% Democratic voters. How many Democratic representatives should that state elect?
If you said 6, I agree with you... but that's not what will happen if you try to use a simple topographical rule to draw shapes on the map in a 'fair' way. If you don't specifically choose the shapes to create 6 majority-Democratic districts, it's very likely that you will end up with like 0-2 Democratic representatives. The 30% will have almost no representation and no one to advocate their interests, which doesn't seem very fair.
Of course, you could look for natural clusters where that 30% state-wide is concentrated in much higher numbers, and draw 6 shapes around those clusters, such that most of that 30% is represented by someone they agree with. But now you're drawing weird shapes specifically to get the electoral outcome you want... that's just gerrymandering again! Or, if you want to split hairs and say that's not *technically* gerrymandering because it's done for good reasons instead of corrupt ones, it still *looks* like gerrymandering to the average voter, and it allows for a type of special pleading in district drawing which future legislators can easily use to do the corrupt type of gerrymandering in the future.
But lets say you decide not to care about proportional representation at that level, you are fine with 30% of the populace getting 0 representation if that's how the lines end up being drawn by your perfectly 'fair', impartial, logical topographical algorithm.
Well, someone else who also openly cares about logical and 'fair' topographical algorithms is going to say yours is stupid, and they should use his instead.
Maybe you think a grid of simple shapes weighted to equal population with zero concern for anything else is the fairest method because it is least arbitrary and eliminates the weird squiggly long districts everyone makes fun of, maybe the next person thinks the simplicity of the shapes is a red herring and what matters is the mean squared distance of each residents home from the center of their district, and you can draw districts which minimize this function in whatever shape that creates. Maybe someone else comes along and says you're both idiots, many geographic regions are already unified based on existing utilities infrastructure and police precincts and school departments and etc. in ways that can't be arbitrarily shifted by decree, and trying to split representative districts across those will create a nightmare of ungovernability that is more important than your concerns about 'fair shapes'. Maybe someone else says etc. etc., until you have 50 proposals.
You are now facing one small problem and one big problem.
The small problem is that out of these 50 proposals, 40 of them are actually *good*. They are all 'fair' in different ways, there are concerns that each of them addresses or ignores, some are easily implementable immediately with minimal disruption or cost, some of them promise to be more long-term beneficial after an uncomfortable and expensive adjustment period. This means there is no simple way to settle arguments between the honest, fair-minded advocates for each, and they will keep fighting about it forever, and will keep fighting to repeal the current system and move to theirs even after it's been implemented for years, and that fighting will mean people keep thinking the system is unfair and keep complaining about it and being discontent about it.
But all of that is the small problem.
The *big* problem is that each of those 50 proposals will tend to elect a different slate of representatives, and special interest groups, think tanks, and the major parties are entirely capable of modeling each of those proposals, and figuring out which ones happen to favor them the most.
Now all the money and influence in this argument won't be coming from honest-minded technocrats arguing for what they believe is 'fairest'. It will come from big interests and the two parties arguing that whatever favors them the most is 'fair'. And the winner will probably be whichever party controls the state legislature at the time, plus whatever special interests donate the most to those state senators.
So now you *still* get a districting method that was chosen to favor one party as much as possible, just like under gerrymandering. But where gerrymandering leaves these footprints of iterative redistricting records and crazy-looking lines on a map, this is a completely 'fair' method that some experts designed to be 'fair' and many more experts agree is 'fair', even if it's not the 'fair' method *they* were thinking off. That bias gets locked in in a way that's much harder to notice and call out in the future, and future reform efforts are stymied because the momentum of the moment when people actually cared enough about this to change things was wasted on a system that didn't really solve the problem.
So, yeah.
This is difficult.
If you think it's not, you may just not be appreciating the surprising amount of detail in this corner of reality.
(ref to http://johnsalvatier.org/blog/2017/reality-has-a-surprising-amount-of-detail)
I'll also point out that we've seen an analogy to a part of this in voting reform. A lot of people are aware that first-past-the-post voting massively favors a two-party system, in a way that lets those two parties be corrupt and lazy because they know they'll be in power about 50% of the time just by default. They hate this! They call for voting reform!
A few nerds know that the best voting methods (Borda, Condorcet, etc) all use ranked ballots. Their knowledge filters down to the public, and they cry out for ranked choice voting! It's the only way to break the two-party duopoly.
In quite a few places, the two parties have said 'Fine, here you go! From now on, we will issue ranked choice ballots! (and we will resolve them using this method called instant-run-off voting)'
Guess what? Ranked ballots with using Borda or Condorcet or etc. to resolve them elect representatives very close to the center or public opinion and break up two-party duopolies.
But IRV doesn't. IRV throws away most of the ranking info, and it produces results almost identical to FPTP, with a massive, almost inevitable favoring of two-party duopolies. They changed the ballot, but they didn't change the result.
This is how the fervor for electoral reform has been blunted across many Western nations in recent decades - give the people the reform they are crying for (ranked ballots), do obscure technical things below most people's level of expertise so that things still resolve the same way they were before, say 'there, you got what you wanted, sorry it seems like it doesn't do much, guess that wasn't the problem.' The movement is largely disbanded since they 'won', the public sentiment is turned against the project as it seems to have failed to change anything, and business continues as usual.
This is what will happen to redistricting if we allow it too... and probably what will happen even if we do everything in our power to resist it.
It's a very hard problem.
" Imagine a state that gets 20 representatives and has 30% Democratic voters. How many Democratic representatives should that state elect?"
There is no "should" .
Then why bother changing anything?
Context is missing. One vote for what? And where?
I thought you meant in more philosophical terms.
"The *big* problem is that each of those 50 proposals will tend to elect a different slate of representatives, and special interest groups, think tanks, and the major parties are entirely capable of modeling each of those proposals, and figuring out which ones happen to favor them the most."
Thank you for articulating this. It's my biggest concern, too.
Whenever one or another form of ranked choice voting is discussed among my friends, they seem to actively ignore many of the electoral ramifactions and be weirdly uninterested in the fact that those promoting the idea to them are not. This even though I usually feel like the quokka of the group otherwise.
Something like "the average straight-line distance between each person's residence and the geographic center of that person’s district shall be the minimum known to be possible at the time prescribed by state or federal law for the determination of districts" will do it.
https://decivitate.jamesjheaney.com/p/death-to-gerrymanders
It seems like a good idea, but this doesn't really allow for geographical oddnesses.
https://www.astralcodexten.com/p/open-thread-431/comment/250204386
Some problems can't be solved if we require geographic districts, though. Something like 1/3 of voters in Massachusetts vote R each year. But because of how the Rs are distributed, there is literally no way to draw districts so that Rs win even a single House seat in Massachusetts. I would argue that in an ideal system, Rs would have about 3/9 of the House seats.
I think you could squeeze one seat. The current districts 1 and 2 (central/western) are a little gerrymandered. Between the two of them, if you look at municipalities that favored Trump in 2024, it's about +18 000 votes out of a total of 222 000 votes cast, ignoring 3rd party (average of 367 000 votes per district). And Districts 4 and 9 (SE Mass + Cape) was +17000 out of 200 000, so you could combine them and add in slightly Democratic connecting towns to create a Republican district. Each of the districts mentioned above is slightly gerrymandered. The South Shore and Cape is strongly D and that overcomes slightly R greater Fall River, which is split between 4 and 9.
Also, a reasonable R candidate would poll much higher, especially with a weak D candidate. Charlie Baker won the governor seat with 60% in 2018.
Two exciting recent innovations on pancreatic cancer!
First, the personalized mRNA vaccine autogene cevumeran, developed by BioNTech and Genentech, just reported 6-year follow-up results from their Phase 1 clinical trial. 16 patients were treated, 8 were responders (showed signs of immune reaction to vaccine), 8 were non-responders.
7/8 responders (87.5%) survived 6 years after surgery, 2/8 nonresponders survived (25%).
AACR meeting notes https://www.aacr.org/blog/2026/04/20/live-updates-from-the-aacr-annual-meeting-2026-monday-april-20/
The most important result in this small trial is that vaccine response is strongly correlated with better outcomes. But for context, the trial was restricted to patients with operable pancreatic cancer. Patients diagnosed with stage 1 or 2 pancreatic cancer have a 5-year survival rate of 12%. Patients who get their pancreatic cancer surgically removed have a 5-year post-surgery survival rate of 20%. This makes the overall 6-year post-surgery survival rate of 56% among the 16 trial patients pretty impressive. Keep in mind that the trial patients may have been healthier than average for other reasons, and small n is small n, so we shouldn't be too hasty until we see Phase 2 and 3 data.
Source on survival rates https://www.pancreaticcancer.org.uk/information/just-diagnosed-with-pancreatic-cancer/if-you-can-have-surgery-to-remove-the-cancer-early-pancreatic-cancer/prognosis-if-you-can-have-surgery/
Second, a small molecule drug I almost missed in the mRNA hype but arguably even cooler, the tri-complex ras inhibitor (!!!) daraxonrasib, developed by Revolution Medicines. Ras is a protein involved in many cancers, with PDAC (the most common pancreatic cancer) being especially dependent on ras, but it has historically been considered impossible to target due to its chemical properties. Daraxonrasib is, as far as I'm aware, the first drug to target generic forms of ras. It does so with an exotic "tri-complex" strategy involving gluing a different protein, cyclophilin A, to ras in order to disable it. Crazy stuff!
Phase 3 results found that daraxonrasib doubled survival time among patients with metastatic pancreatic cancer, 6.7 months to 13.2 months (p < 0.0001). Side effects are kind of nasty, but "well tolerated, with a manageable safety profile" by advanced cancer standards.
Revolution Medicines announcement https://ir.revmed.com/news-releases/news-release-details/daraxonrasib-demonstrates-unprecedented-overall-survival-benefit
Derek Lowe coverage https://www.science.org/content/blog-post/progress-against-pancreatic-cancer-part-one
Lecture by executive/scientist at Revolution Medicines https://www.youtube.com/watch?v=bU3IwuDJx24
Many of the incredible advancements we've made in oncology has been won slowly through another pathway we can target, another drug buying a few more months, another targeted technique to mitigate side effects. Progress is made one step at a time, and these are big steps. I'm excited for the future of biotech.
FWIW my wife is a cancer researcher, she’s quite excited about the revolution medicine drug, hadn’t heard about the mRNA vaccine. I was looking for info in the link you had but didn’t find anything there…
Which link?
Sorry a little late but I meant the first link to the meeting notes, when I searched for cevumeran I didn't see anything
It's titled "Clinical Trial Results: Patients With Pancreatic Cancer Treated With a Personalized mRNA Vaccine Alive Six Years Later".
The AI angle, since many people are hyping the "AI for pancreatic cancer" line:
Yes, deep learning is used in the development of the mRNA vaccine. NetMHCpan, a small neural network, is used to help select immunogenic neoantigens (choose the most promising mutant proteins to target) for autogene cevumeran.
Autogene cevumeran paper https://www.nature.com/articles/s41591-024-03334-7
Most of the computational pipeline detailed in the paper consists of traditional tools, in line with what I've written about AI and personalized mRNA vaccines before. Deep learning is extremely useful for some things, but it should be understood as a specialized tool, not a silver bullet, for now.
AI & mRNA blog https://hedonicescalator.substack.com/p/did-paul-conyngham-really-use-ai
I'm not sure if deep learning was used in the development of daraxonrasib. A brief glance at the paper and previous work shows plenty of references to traditional computational tools, but nothing that stands out to me as modern DL.
Daraxonrasib paper https://pubs.acs.org/doi/full/10.1021/acs.jmedchem.4c02314
Previous work https://pmc.ncbi.nlm.nih.gov/articles/PMC10474815/
The company behind daraxonrasib, Revolution Medicines, is quite enthusiastic about ML. They recently made a deal with AI drug discovery platform Iambic Therapeutics. I don't doubt Iambic's tools will soon prove useful, but it's fair to say "deep learning for drug discovery" is still in the early stages of development.
Iambic announcement https://www.iambic.ai/post/revolution-medicines-collaboration
I wrote some anti-self-help "advice", inspired by actual self-help stuff I've seen going around.
https://squarecircle.substack.com/p/anti-self-help
Then writing that put me in a mood to share a Sufi story, so I did.
Should we just go Aristotelean on self-help? Like find the golden mean in everything?
Maybe you'd quit talking to yourself if you got some help!
Airlines cancel routes all the time for many reasons. Nonetheless, I believe that Delta’s recent cancellation of its New York JFK - Brussels daily nonstop is a “canary in the coal mine” warning that New York’s standing in the national and world economies is starting to decline. Maybe it’s all the financial businesses decamping to Miami, I don’t know.
Why I believe this is no ordinary route cancellation.
- This isn’t some new route that didn’t live up to initial expectations. Delta has operated the service all the way since 1991, through all sorts of economic ups and downs, until now. In fact it was one of the European routes Delta had inherited from Pan Am.
- Brussels, indeed all of Belgium, is not really a tourist destination, so the cancellation cannot be blamed on shifting tastes in tourism. But it *IS* a major business destination. A decline in business travel is a much more likely sign of economic decline.
- Delta has not pulled out of Brussels completely, with ongoing service from Atlanta. In fact the company’s only explanation of the cancellation was some corporate babble about serving its customers better from Atlanta. It’s unlike Delta’s recent cancellation of JFK - Geneva service as it no longer serves Geneva at all, indeed it’s rather small to merit nonstop service from the US.
- Speaking of Atlanta, it’s Delta’s major hub, but it’s *always * been the major hub. Atlanta’s hub status did not stop Delta from serving Brussels from New York for nearly 35 years.
- No other airline has chosen to replace Delta. The only New York service is a single daily nonstop on Brussels Airlines; while United flies there from Newark, whether Newark qualifies as a New York airport is an unanswerable question.
Now, I acknowledge that a decline in importance on the Brussels end could explain this, however I’ve never heard of anything to that effect while New York’s loss of the financial sector to Miami is a very real and well documented thing.
There's still a daily non-stop flight to Brussels from JFK and one from EWR. DC->Brussels is also twice-a-day.
It's probably a result of telecommunications becoming more prevalent over time.
Love reading takes about New York's decline from people who don't live here because they always miss a bunch of context and are always proved wrong in the end.
Better to let them believe it sucks. Who needs them anyway? They'd just be asking dumb questions on the subway.
> They'd just be asking dumb questions on the subway.
Why is it so grotty?
We like people to feel at home
I'm not saying that the deprioritisation of New York in favour of Miami as a financial hub is not happening, I just think that if it were happening then there'd be a lot of other data you could track it more reliably with, like job listings or real estate ads, or large amounts of floorspace given up by major financial institutions in New York or rented in Miami.
You don't need canaries in coal mines when you have an omnipresent network of carbon monoxide detectors.
>whether Newark qualifies as a New York airport is an unanswerable question
Newark is closer to Times Square than JFK is.
Does your analysis take into account the decline in international travel to the US under Trump? Or the possibility that it is Brussels that is declining in significance?
https://fortune.com/2026/04/17/international-tourism-to-us-fell-last-year-world-cup-worries/
https://www.afar.com/magazine/international-travel-to-the-u-s-is-declining-in-2026
It may be worth noting that Delta is starting three new European routes from New York next month, summer-only so they don’t make up for the loss of Brussels, to Porto, Malta and Sardinia. It is probably a reasonable assumption that they will appeal more to Americans visiting Europe than to Europeans visiting the US and hence in line with the decline in foreign travel to the US.
It's a big inference from a small data point. Delta puts a lot of JFK passengers into London, Frankfurt, Amsterdam and Paris, none of which is very far from Brussels by American standards of distance or European standards of train service. Delta is not in league with Air Brussels/Lufthansa so has no particular desire to help them sell connecting flights. NY gates are scarce and will be allocated to routes that maximize revenue, Brussels is pretty far down the list of European airports, and if you look at the Delta map they are not trying to run direct NY flights to every Euro destination.
If you need a better reason to explain why a carrier would cancel a route they've run for decades, I'd put higher likelihoods on the decline in US relations with the UN and NATO or good old fashioned antitrust boundary-pushing like sending a message to United and Lufthansa that Delta would prefer to divide up markets a bit more neatly.
All of the things you note in the first paragraph are true, and *could* be the reason, but I’m skeptical because they’ve always been true. Yet the route lasted 35 years until now.
Declining relations, like you note in the second paragraph, now that could be a more plausible reason as it would affect business travel much more than tourism.
Newark counts as a New York City airport for all practical purposes.
> A decline in business travel is a much more likely sign of economic decline.
10 years ago, this might have been the case, but it could also be a sign of standard cost-cutting with the rise in business being conducted over zoom.
NY - Geneva being cancelled actually seems worth thinking about, because of the major UN presence in both, plus many other big international orgs in Geneva. Breakdown in diplomacy? Increased isolationism? Or maybe just the same rise in video conferencing.
>"…it could also be a sign of standard cost-cutting with the rise in business being conducted over zoom."
In the long-ago I worked at Boeing, and heard several times that the real competitive threat wasn't Airbus or Embraer, but Skype (I did say long ago).
>> Newark qualifies as a New York airport is an unanswerable question.
It’s easier to get to from places in NYC (Manhattan) than JFK and LGA.
It’s a New York area airport.
Yep. From EWR, I can get on a train that takes me directly to Midtown Manhattan.
From JFK I guess I could take a bus to the E train subway that'll get me to Times Square eventually. I'm not sure how I'd do that from LGA. And PATH is much less sketchy than MTA.
The JFK Airtrain goes from the airport to Jamaica station on the Long Islamd Rail Road, from where there are frequent quick trains to Manhattan and Brooklyn, as well as subway connections. LaGuardia Airport has no train connections; there’s talk about building one, but then again New York has been trying and failing to build the urgently needed Second Avenue Subway for the past 100 years so don’t expect one soon.
Though possibly less than in the past, some New Yorkers have a provincial mindset that regards anything west of the Hudson as remote as North Dakota.
New Yorkers believe a lot of silly things. They believe that sticking a lot of roast beef in a sandwich is something to get excited about.
Pastrami.
Steamed beef then.
Look I'm not saying pastrami sandwiches are bad, I'm just saying that if you're the biggest, richest and most powerful city on Earth and your big contributions to cuisine are "sandwiches with a lot of meat" and "generic pizza" then you should find something else to brag about.
There are lots of things wrong with NY. I lived there for a dozen years and was pretty happy to GTFO.
But criticizing NY because of the food?!?! No, you're very wrong here. Yes, even pizza. NY has Italian, French, or Japanese foods on par with that of the respective countries.
I think steamed beef is closer to brisket than it is to pastrami .
Every pizza store in New York City has a sign that says, "You've tried the rest; now try the best." We know better . By the way I got a really good deal on a bridge and I'd like to share it with you.
Ah, Melvin, that is just the stuff we publicise for tourists. We don't brag about the really good stuff. We're not fools.
Is there a good term for when you're presented with a reducto ad absurdum of your position, and end up agreeing with it? Obviously there's "bite the bullet" and there's the Chad says yes meme, but it would be nice to have something classier and more parallel to the original.
Best I've come up with so far, with some help from Claude, is assensio ad absurdum. Does this actually work or is it just useless dog latin?
I asked sydney. One such suggestion was "embracing the entailment". No, it's not a real term (yet). Though I quite like the alliteration.
she also suggested "acceptio absurdi" and "reductio ad fortius", though these don't have quite the same ring, imo.
Wut about FMLOK?
"Bite the bullet" is old and common enough that I think it counts as classy enough for debates.
"Bite the bullet", to me anyway, means you accept the tenets of a counterargument against your argument, but has a flavor of "but I still think I'm right because..." which you will follow with some (probably novel) argument against that counterargument you just bit the bullet on.
I just call that 'going camp', but you sort of have to know a specific academic lens on the history to get the association.
“One philosopher’s modus ponens is another philosopher’s modus tollens” is closely related, though you would have to use it the other way around if confronted with a reductio.
How about “one person’s reductio ad absurdum is another person’s counterintuitive conclusion”?
nolo contendere?
So you basically hold to Caesar's principle, tamquam scopulum sic fugias inauditum atque insolens verbum? Fair enough.
I feel like in this case it's not so bad since reducto ad absurdum is well known, and we have English words like assent which are clearly derived from the same root, but I do see where you're coming from.
*-*-*shameless self-promotion disclaimer*-*-*
Here to share some stuff which I think ACX readers may enjoy. I recently did a deep dive on an allegedly ancient Tibetan prophecy that has been circling in the American Dharma world since the 70s. Lots of interesting supernaturalism, memetic spread, and citation-ambiguity along the way: https://transgenicpig.substack.com/p/when-the-iron-bird-flies
Currently working on a series on "What is mass?" https://transgenicpig.substack.com/p/what-is-mass-part-1
About "What is mass?" - I like the good self-reflection, though I admit I mostly skimmed over the article. It is not dense enough for me to warrant word-for-word reading like I do on ACX.
On the content: Have you considered the equivalency of mass and energy, aka E=mc² ? That is the modern answer in physics, and it goes a long way.
I do agree that in early semester physics the explanations of mass and weight can be confusing. If you have a good professor though, he will point out multiple times the curious relationship of those and encourage you to keep pondering for the following years. Once you are "through" with the general physics curriculum, re-visiting all of this certainly breeds a new and better understanding of the "classical mechanics" topics (which are all just approximations; to be more explicit, e.g. Newton's physics is just an approximation for certain mass/distance/time ranges of the more general relativistic physics).
Hey thanks for reading -- and yes I totally agree that there is a special kind of joy that can come from revisiting the foundations after one has some familiarity.
RE energy-mass equivalence: I will discuss this, but perhaps not as explicitly as in your approach. To give the game away, Part 2 [https://transgenicpig.substack.com/p/what-is-mass-part-2] (coming out tomorrow btw!) will establish three definitions in a pre-relativity context, but for various reasons these collapse to just one in a relativistic context (Part 3). And yes, we will see that in this context mass, energy, and momentum start to blend together, and it will be increasingly unclear that we gain anything by thinking of them as distinct.
From my approach this seems more like an interesting side-effect of relativity than a starting point in its own right. The tension between our views on mass is probably that I am trying to ground "mass" in experimental operations, even in the context of relativity. Still, if I have a broader thesis than considering mass itself, it is that the meaning of physical concepts lies in their web of interrelations and one can probably slice that up into definitions in many ways.
Scott, if you are reading this, please write a post on your understanding of your to look at the iran war- a rationalist-moral-american take, not a geo political one.
My guess: Iran regime evil. Attacking the country without destroying the regime = killing lots of people for no good reason, also evil.
(Speculation: maybe the true motivation was to increase oil prices, to save Russia's collapsing economy. Also evil.)
It is difficult to find anything good about the entire thing, but perhaps others have a different perspective.
That would be an overly simplistic take. Replacing the regime is the ultimate victory, but there's lots of intermediate goals that also have value. Just degrading their conventional missile stockpile is useful.
Though the big win IMO, is diplomatic. Europe is furious at the USA for causing an economic crisis. But the Middle East is realising that all their attempts to hedge between the USA and Ian, to do a bit of appeasement, has all amounted to nothing. They're also furious at China for helping Iran. (Odd choice by China, it gets more gas from Arab states than Iran). So they're lining up more solidly with the USA.
Qatar expelling Hamas. The UAE leaving OPEC. This is good gains for the USA.
On the topic of post-COVID educational attainment collapse: could it just be the chromebooks? Specifically, the 1:1 device policies in elementary school, instituded to enable remote learning, have stuck around afterwards and are toxic. Putting kids in front of screens for many hours a day is obviously terrible parenting, but we have some sort of blind spot keeping us from noticing that it's terrible teaching
https://indyweek.com/firstperson/voices/voices-an-education-in-scrolling/
Related: Sweden, Denmark, and Norway have recently pivoted from screens back to physical textbooks after studies "showed" a dip in reading skills, etc. (Studies on this, especially the ones that convinced Sweden et al., use data from PIRLS, similar to NAEP but international.)
Plenty of links available with a quick Google search.
Isn't the collapse largely at the bottom half of the distribution? Probably more a function of absenteeism than Chromebooks. https://www.brookings.edu/articles/student-level-attendance-patterns-show-depth-breadth-and-persistence-of-post-pandemic-absenteeism/
The effect is too wide to be Chromebooks, whereas the absenteeism is almost everywhere. My money is on absenteeism.
This is my thinking, though TBH I don't know how widespread Chromebook adoption has been.
The fact that the effect is greater at the lower end is a point in favor of absenteeism, since I am pretty sure that is more common among lower achievers.
Agreed, absenteeism is also a good example of "thing that happened in 2020 that did not go back to normal by 2022" -- even post-pandemic, absentee rates are way, way up. I'd note that if you believe this it's a (soft) refutation of the "you don't learn in school" hypothesis forwarded on this blog and elsewhere.
That hypothesis has always struck me as an absurdly large over generalisation from people not being able to remember certain facts after graduation. We know what societies without schools look like! Even societies with mostly primary but not secondary education! (E.g. my grandparents' norms). It's obviously pretty different!
My view is that we probably keep people in education too long now (people should be able to start work at 16) but that's still a lot of school. There's a reason every country that can do it does.
It might also be related to the ACX community selecting for high IQ nerds who would learn whether or not they were in school, while school is still beneficial for normies.
It's sort of like how there are big summer regressions in education dependent on SES because rich parents send their kids to educational summer programs and poor parents don't.
I'm mildly skeptical of this just because even smart, nerdy, curious students can often benefit from structure and guidance. The ways in which the system is (or appears to be) standing in your way or holding you back are going to be easy and obvious, but you don't have access to the counterfactual world in which you had to figure it all out for yourself.
I've taken multiple stints to go back to university as and adult, and several stints trying to build useful skills and learn difficult subject on my own. I make more than zero progress on my own, but I learn *a lot* more in university[1]. I don't want to over-generalize from my own experience (among other reasons, because I have ADHD), but I'd be surprised if this wasn't *a* piece of the puzzle, at least. As with many other abilities, I'd expect there's a wide range of human ability at self-teaching. I'd further expect that such ability has a correlation with IQ, but only a moderate one.
[1] I've also tried a few online courses not connected to a particular program. Unsurprisingly, the effectiveness was in between "go it alone," and "take an actual class." Meanwhile, learning on the job is arguably more effective than either option, but it's heavily restricted in what it can teach.
Yes agreed.
My take is mostly "school is not for us", but also, "That's fine". Salary surveys seem to indicate people here are doing rather well for themselves.
There definitely can be two problems, possibly even more! Utilitarian me is also deeply empathically worried about absenteeism increasing from 15% to 30% at the very bottom of the distribution, but for some reason I'm inclined to pitch a fit specifically about issues likely to deal 2-3% score drops to kindergarteners with involved parents, entering the school system in zero to 2 years, (within a five mile radius of my current location.)
The problem is real up and down the board, and I doubt that the 90th percentilers in 2023 had absenteeism issues:
https://www.nationsreportcard.gov/highlights/ltt/2023/
Has anyone written anything about what international development could look like in a post-powerful AI world? There seems to be a decent literature emerging on the economics of AI in the developed world, as well as stuff like AI 2027 which tries to flesh out what the world might look like, but what will things be like in, say, Kenya? Or the DRC?
My assumption would be, assuming we end up in a non-singularity world where economic growth remains a meaningful metric but significantly accelerates, that the OECD is much more able to take advantage of AI than the developing world. This causes the West and developed East Asia to pull even further ahead of the developing world than the already are. The medium-term endgame is that either Africa becomes the actual permanent underclass, or that international aid becomes the vast, vast majority of third world GDP and living standards skyrocket. But I'm not an economist so these intuitions could be completely wrong!
I'm expecting the opposite, under the following theory:
1) The reason some countries are poorer than others is that they lack intelligence.
2) Advances in AI will give everyone cheap access to intelligence.
3) The gap in prosperity between smart countries and dumb countries will thus shrink.
Consider the wisdom of Machiavelli: A prince who is not wise himself can not be wisely counselled.
Suppose an AI were to tell someone the problem is that his relatives expect him to make sinecures for them, and thus he lives in poverty, because his relatives’ labor is crucial in raising the living standards. Is he going to listen?
People rely on guidance from experts superior in intelligence or knowledge all the time, so I'm not too concerned about that.
People ignore guidance from experts superior in intelligence or knowledge all the time, and the results are frequently disastrous .
It's not like Africans haven't been told, by many experts, that graft, corruption, nepotism, and covetousness are what is keeping them poor.
Intelligence is one thing, good institutions are another. Recently it seems that it is easier to collapse institutions in the developed countries, than to build them elsewhere.
(If someone wants to go full HBD, I recommend considering North Korea and South Korea, which are genetically very close.)
The problem is not that the governments are too stupid to do the good thing for the country, but that sometimes they have no incentive, because they can just steal everything, and then spend a fraction on propaganda. On the side of voters, IQ alone does not imply rationality.
No, the gap has many reasons but the biggest is corruption coming from strong kinship networks.
Persistence of which is bizarrely well correlated with IQ
I find that hard to believe, so I'd need to see some good statistical evidence to be convinced. The correlation between national IQ and GDP/capita appears to be around 0.7, with about a doubling of productivity for every 10-point increase in IQ, so if advances in AI do lead to an abundance of intelligence, I would expect to see enormous growth among cognitively disadvantaged nations.
> The correlation between national IQ and GDP/capita appears to be around 0.7
True, although national IQ increases with GDP.
I had the thought that many who no longer needed to work in tech socialist utopia might volunteer to help developing countries.
Why would Third World living standards skyrocket? Vast amounts of aid are vast opportunities for graft, and furthermore, graft prevents living standards from rising, thus enabling more applications for aid.
Dean Ball went to a conference and wrote about something obliquely related, basically how developing/global south countries are currently plugging their ears and saying "lalala this AI stuff is all fake!" which does not bode well for development post-AGI. Matt Yglesias' "we may miss the sweatshops" piece is relevant-ish here too
https://www.hyperdimensional.co/p/the-moving-and-the-still
and
https://www.theargumentmag.com/p/we-may-miss-the-sweatshops
Wow, that Matt Yglesias piece is really good (and I'm flattered that he seems to have had similar ideas to me). I'm particularly interested in this part:
"In an AI-powered world, the resource curse could prove particularly vicious. Obtaining the needed commodities won’t necessarily require human labor, just legal authority to do the extraction. Whoever is in position to hand that legal authority out — the internationally recognized government — will collect all the rents, with no particular need to share them with anyone. Any group that manages to pull off a successful coup will capture vast wealth. Anyone on the outside will be left with nothing. Endless rounds of violence and mass killing could easily be the result."
I disagree because I think that in this scenario, the developed countries will be insanely rich and will still feel some general sense of obligation to poor nations. As a result there will be a push to give them some of the tremendous wealth being produced from the robot-operated mines in their countries. But in this scenario our ability to exert force across the globe will be much greater, so we may feel that we can avoid the issue of rent-seeking by establishing our own technocratic, highly efficient governments to distribute this aid.
"Best" case scenario, this could be something like the Australia situation where Indigenous Australians get given cash in exchange for mining on their traditional land. Worst case scenario, it's literally just liberal imperialism.
> the developed countries will [...] feel some general sense of obligation to poor nations.
I strongly doubt that the MAGA crowd feels anything like this.
Why would you think the resource curse will be limited to poor nations? Every nation will be in the same position: humans are economically and militarily valueless, but consume resources unless exterminated.
>the developed countries will be insanely rich and will still feel some general sense of obligation to poor nations
I want to hope so...but the current state of affairs does not make me too optimistic! In comparison we (developed countries) already *are* insanely rich, and apparently can't even manage to kick over 0.1% of our government spending to poor nations.
To be fair I live in the UK where we historically spent 0.7% of GDP on foreign aid. This has been reduced in recent years because of COVID/the need to increase defense spending, and was also increasingly being spent on housing refugees in Britain. But presumably in a post-AGI world lots of policy issues which require money would be alleviated so people may be more willing to fund foreign aid programmes.
> Several dozen influencers will get scholarships...
Are we witnessing the enshittification of the AI-safety movement in real time ? Granted, I'm not part of the movement, so I find it merely amusing...
I would love to see some manifold bet attempting to operationalize "will this be cringe."
That said I am sure there is already a tsunami of anti-AI brainslop on short form platforms already, of the "datacenters steal water" variety, so maybe (the logic goes) putting some "good guys on the inside" is the right move?
It's tough, you don't want to cede a massive sector of the info-space, but there is real risk of ending up in "slopulism." Some of the rationale behind this short-form bootcamp seems to be coming out of "beef" with the Pause/StopAI world on X so make of that what you will.
At some point your cool indie band has to sell out in order to get famous. At some point, Ice Cube has to go from "Fuck the Police" to Law & Order SVU and Snoop Dog has to go on Martha Stewart. That's the cost of success; cool underground things sell out in order to make money and have influence.
Like, if p(Doom) is non-trivial and technical alignment is not viable from a research standpoint in realistic timelines, then the only relevant course of action is influencing policy and in modern day, that means developing relationships with streamers who cultivate widespread parasocial relationships with upcoming generations. If L33tRacismBro86 get Twitter famous and then Elon puts him in charge of USAID (1), welp, better start cultivating a relationship with L33tRacismBro87 and Fr33G@z@Nuke1sra3lGirl12 to prevent the world from ending. (2)
(1) For the record, provisional on millions of African children not starving over the next 16 months, L33tRacismBro86 was a substantial improvement over the current administrative state and DOGE was ended far too early.
(2) If you are a young person and all of this sounds retarded to you, welcome to being an old person. I am old now and the only thing more shameful than a bunch of young people being retarded is a bunch of old people pretending to be young and still being retarded.
In the before times I recall Jerry Garcia being willing to sell out at some level for quite a while but no one was interested.
In 1987 Garcia took legal action against Ben & Jerry's over their Cherry Garcia ice cream. Attorneys negotiated a licensing agreement.
Bob Dylan famously "sold out" 4 times, doing commercials for Victoria’s Secret, Apple, Pepsi, & Chrysler.
Excuse me, Detective Tutuola was played by Ice-T, not Ice Cube.
I know how this goes: soon, Fr33G@z@Nuke1sra3lGirl12 is revealed to be a "TERF," and that damages the cause of every group that cultivated a relationship with her.
The problem is not that influencers are bad at what they do, but rather that what they're good at is posting cool videos that make their followers click "Like" and "Subscribe". So if all you're looking for in your movement -- be it AI safety or pesticide-free farming or whatever -- is lots of likes, then influencers are a way to go. Ok, that is not entirely true; another things influencers are good at is hawking cheap dodgy goods, the cheaper and dodgier the better... but again, if I were part of the AI-safety movement, I am not sure if this is the direction I'd want the movement to go.
Fog o’ War stuff: Ukraine/Russia
1. Ukraine got their €90 billion loan from the EU now that Orbán isn’t around to hold it up. Repayment is designed to occur only if and when Russia pays war reparations. Critics questioned why they didn’t confiscate the frozen assets of the Russian oligarchs and give that money to Ukraine, but the consensus is that the EU is trying to hold out a carrot to the oligarchs. They’ll get their money back if the war ends, and the hope is that in the meantime, they’ll be able to put some pressure on Putin. Fat chance with that. Putin has killed enough of them that they know not to cross him. But there are also €210 billion in frozen Russian Central Bank assets that, according to one article (mentioned in passing), would help service the debt. I haven’t seen a good explanation in the MSM about the details. Does anyone have more details on how this deal has been structured and where they're getting the money to finance it?
2. Zelenskyy signed an agreement with Azerbaijan for trade and military aid. Azerbaijan will send them “energy assistance”, which, after some digging, includes high-voltage cables, transformers, and generators to replace those lost in Russian attacks. I didn’t realize this, but Azerbaijan maintained ties with Ukraine throughout the Special Military Operation, and Zelenskyy thanked President Aliyev for their previous humanitarian aid and energy support. Azerbaijan gains access to European business opportunities through Ukraine's partnerships with European countries that Zelenskyy has been busily developing. And don't forget the drones.
3. Zelenskyy also hopped over to Syria to ink a deal with President al-Sharaa. Syria will send phosphate to Ukraine, and Ukraine will send wheat to Syria (there are still food shortages in Syria). Simple barter trade. Remember also that Ukraine sent in advisors and drones to assist al-Sharaa and the HTS in overthrowing Bashar al-Assad. And I think they gave Ukraine a bunch of captured Russian equipment and ammunition in exchange. And don't forget the drones.
4. Ukraine will also be sending marine drones to the Saudi’s and the UAE to help them break the Iranian blockade. Details of how this would work weren’t specified, but Ukrainian marine drones have been configured as gun platforms, missile launchers, and drone-interceptor launchers, and they have the speed of Iranian gunboats. Some models can operate for 48 hours without refueling and have a 600 km range. It would be funny if little Ukraine ultimately breaks the Iranian blockade. Coincidentally—or not—Iran is dumping more mines into the Strait.
5. I saw an estimate that Ukraine's defense deals are going to bring $40 billion into Ukraine's economy. I haven't seen a breakdown of which country is paying how much and how the payments will be structured. Still it's better than a poke in the eye.
6. In Russia, Defense Minister Andrei Belousov and First Deputy Chief of Staff Sergei Kiriyenko are in a struggle to control the Russian Duma and the patronage that goes with it. Belousov is actively challenging the Kremlin's (i.e., Kiriyenko’s) planned list of Duma candidates for the September 2026 elections. He wants to install his own personnel rather than the "Leaders of Russia" program appointees pushed by the Presidential Administration. These, we are told, were “veterans of the special military operation,” but mostly served in reserve battalions far from the front. Belousov rejected the initial list of candidates, calling Kiriyenko’s candidates inexperienced political hacks rather than true combat participants. Belousov happens to be deeply religious. Some have described him as fanatical in this faith, and he reportedly sees the war with Ukraine as a religious war. We’ll see how this shakes out, but I doubt if Saint Andrew the Apostle (the patron saint of Russia) can do much to stop Ukraine’s long-range drones.
7. Russia claims to have lost 5,937 soldiers killed in Ukraine since the start of the Special Military Operation. That’s all. Meanwhile, Putin claims that 1.5 million Ukrainian soldiers have been killed. The official Russian death toll has not increased since 2023, when it was first published. Now that the Special Military Operation has gone on longer than the Great Patriotic War, it’s not opposition groups that are shining a light on Russian losses, but it’s far-right Russian milbloggers who are starting to suspect there may be something fishy in those numbers. The neo-Nazi Rusich Group’s Telegram channel laid out the rough arithmetic garnered from official sources. According to official data, 540,000 people signed contracts with the Russian military in 2023; 450,000 in 2024; and 400,000 in 2025. That’s just shy of 1.4 million recruits. And there were another 300,000 reserves who were mobilized at the start of the operation, plus the 700,000 who were on active duty in the military at the start of the operation. So, 2.4 million soldiers. But the State has also said that there are currently 900,000 soldiers on active duty at the front. Rusich ran the numbers: 2,400,000-900,000=1,500,000. And they asked, where did 1.5 million soldiers go?
A few days later, Yuri Podolyaka, a pro-war blogger with over 3 million followers on Telegram, estimated that between 315,000 and 415,000 of those 1.5 million inferred casualties have been killed, and he used Putin’s own numbers to support his claim. Putin said that Ukraine’s losses were five times higher than Russia’s, and according to Putin 1.5 million Ukrainians have been killed. Podolyaka concluded, therefore, that this must mean that 300,000 Russians have been killed in the Special Military Operation, which jibes with the lower range of his estimate.
It's interesting that the pro-war Russian bloggers’ estimates are higher than the Ukrainian Ministry of Defense’s estimates of ~1,326,000 casualties. As an independent reality check, at the beginning of 2026, the Center for Strategic and International Studies placed Russian losses at approximately 1.2 million casualties (killed, wounded, and missing) and as many as 325,000 killed since the start of the Special Military Operation in February 2022. And reports indicate that Russia has suffered another 100,000 casualties since the start of 2026.
8. Russian milbloggers are also frustrated with their army’s strategy of assaulting what they call “grandmother villages” instead of going after military targets. Older women frequently become the last residents of ruined villages. A ground robot evacuated a 77-year-old Ukrainian lady from the battle zone in Lyman. She was spotted by Ukrainian drone operators walking along a shelled road. They sent in a ground robot to rescue her, with a sign that said, "Grandma, get in!" The ground robot drove her to safety. Aerial drones also escorted out some other civilians.
https://x.com/i/status/2048040399940338114
9. Russia is having trouble meeting its recruitment numbers. Nobody disputes this. Mainstream Russian propagandists and milbloggers agree that Russia needs more troops and that recruitment drives are coming up short. Zelenskyy ran the numbers on Ukrainian Telegram and said Russia would need to recruit another 10-15% to break even with the numbers they're losing on the frontlines. Anyhew, Russian army officers are conducting video calls with secondary schools all over the country to drum up volunteers. At one particular (unidentified) school, a masked officer came on the call to extoll Russia's elite Rubicon drone corps. He went on about how drones were the cutting edge of warfare, blah, blah, blah. But the call took a turn into the bizarre when he revealed he was a Ukrainian drone officer. He said, "And I want to tell you, God forbid you go there, I will have to kill you," before someone was able to shut down the call.
What I find particularly interesting, beyond it being a great propaganda prank, is that Ukraine was able to learn about the scheduled call arranged by Russian military recruiters and then hijacked it. What does that imply?
https://www.youtube.com/watch?v=V3u52LeHsN8
"Ukraine got their €90 billion loan from the EU now that Orbán isn’t around to hold it up. Repayment is designed to occur only if and when Russia pays war reparations. "
I always felt Europe is too unselfishly altruistic towards Ukraine, but this takes the cake. Unless there is some deep strategy behind it. The Draghi Report clearly states that the European economy sucks, it requires a big cash investment, he asked for it and they replied that they cannot afford it. Is helping Ukraine truly more important than making the European economy competitive?
The governing mechanisms of the EU are showing the strain - to get anything done requires unanimous approval. I would take the other lesson: Ukraine clearly needed support, in the sense that allowing Russia to win the war would result in costs to the EU probably exceeding 1 trillion to build defences along the east, develop a common defence programme and perhaps a common army. The cost-benefit analysis is clearly on the side of funding Ukraine.
The inability to do much of anything else is a central problem for the next decade: throwing money at crises like Covid and Ukraine doesn't solve the internal problems of the union. It's time to move ahead with the 2-speed EU idea, and stop waiting for the last holdout to fall into line on every issue.
"The governing mechanisms of the EU are showing the strain - to get anything done requires unanimous approval."
Sure, because the original idea was that for example foreign policy should not be common, like how France and Germany opposed the Iraq War and the UK and Poland supported it.
This is why it is a little strange why Orbán is treated as a traitor. It was in the past OK to have your own foreign policy. If I want to be charitable, it is because it is not only about Ukraine but Russia is a security threat to the EU too. Although I do not think they dare to risk a NATO war, I accept that security policy is always based on capability, and not intent.
On the other hand in a lot of other cases the veto system is good and I am afraid they are going to abolish it.
Shareholders in an economic enterprise are allowed to have competing interests; team members or political representatives are not.
The global context changed a lot in the 20 years between Iraq and Ukraine. On the one hand, the EU can no longer pretend that it's purely a trade group - they are a political entity with a common identity. On the other, MAGA-type right wing populists sometimes tell you directly that they are acting in bad faith. Orbán was a bit more sophisticated than Trump, but still clearly acting against the interests of the EU in a way that Germany (almost by definition) could not.
Sorry, I was being cynical -- what I meant was that EU policy is overly determined by German interests. What Germany wants becomes EU policy!
Since 2022, most members of the EU consider Russia their primary geopolitical enemy, some countries on the Eastern border even view them as an existential threat. Now you may or may not agree with this assessment, but given this assumption, a €90 billion loan, secured by the assets of your enemy no less, is a pretty good deal to keep Ukraine in the fight against Russia. Even more so if it keeps Ukraine on your side for after the war, because the best partner against an enemy is the friend who just spent years fighting your enemy, who has a larger military than any single member of the EU, and more experience than all of them combined.
No need to resort to "unselfish altruism" as an explanation.
Whatever. I am just upset Draghi’s request for funds to make the European economy competitive was denied.
Oh, I completely agree that the EU should get its shit together, both through more internal investments and aggressive deregulation, and should stop playing cornucopia for the world. My point is that helping Ukraine financially and militarily isn't primarily altruistic.
I don't know really, but I also expect quite some of those billions to flow back into the EU as Ukraine buys from and cooperates with the EU.
For the financial construction of the EU, they do not confiscate the frozen assets for the reasons that EngineOfCreation mentioned.
What they do instead (reconstructing it from my memory): they make a loan to the Ukraine, conditioned on a hypothetical payback in case Ukraine gets these frozen assets after the war. It is important for many EU countries (France, ...) that it is a loan because then the money does not count towards the national debt. After all, in the books, they only trade the money against the Ukrainian debt, which is nominally worth the same. It makes a difference because there are all kind of rules that prevent EU countries from making additional debts, both EU rules and national rules.
I'm trying to understand this. Are you saying that France (for example) borrows money to then lend it to Ukraine? If France doesn't borrow the money, why would it count toward the national debt? Also, how would Ukraine "get these frozen assets after the war"? These assets aren't under the control of Ukraine, so France (or whatever) would have to seize them first, which gets us back to "can't do that under the Rule of Law". What am I missing?
Ok, the details are complicated, so I read up on them. In the current deal France doesn't borrow money. In the current deal, the EU commission borrows money from EuroClear (the company located in Belgium who actually holds the Russian assets). Usually, the EU commission is not allowed to borrow large amounts of money. If they want to borrow money (beyond a limited normal budget), then the EU countries would need to provide guarantees. As far as I understand, this is the issue because those guarantees count towards national debts.
Now, the EU countries agreed that the frozen assets in the books of EuroClear are acceptable guarantees. Their official point of view is that at the end of the war, Russia will agree to pay those assets as retribution to Ukraine, meaning that Ukraine will then be able to redirect the money to the EU commission.
There is no problem until Ukraine and Russia make a peace treaty, because the sanctions are permanent in place until such a peace treatment is signed. So until then, EuroClear is not allowed to transfer the assets or money to Russia, the assets stay at EuroClear, and everyone can close their eyes and say that this is fine. I think this is what the EU hopes for, that there won't be a peace treaty for a long time.
If there is a peace treaty where indeed Russia agrees to pay the assets as reparations, then also everything is fine.
If there is a peace treaty where Russia does NOT agree to reparations, then there is a problem. Then EuroClear will legally be obliged to transfer the assets back to Russia, and then the guarantees for the loan disappear. This means that the money is gone then, but at least this is now a problem internal to EU.
One of the most heavily criticized points is that it makes the EU a stakeholder who has an interest AGAINST a peace treaty between Russia and Ukraine, unless it contains reparations form Russia. (Good luck with that!)
It's quite petty, to be honest. I don't think there is a realistic chance that the money ever comes back, and I can't imagine that the EU leaders believ in such a chance either. So the whole purpose of the scheme is to make it look like it's not direct financial aid, although everyone knows that it is.
EDIT: Here is an interview from December explaining the deal:
https://www.youtube.com/watch?v=Dp1tUNbYK1Q
>Then EuroClear will legally be obliged to transfer the assets back to Russia,
Legally required by whose law? Can that law be changed? Or it means international law? There is really no such thing... those are agreements that can be broken. One can say international law X will be now broken because Russia broke the Budapest Treaty and also Minsk Treaty.
> Ok, the details are complicated, so I read up on them. [...]
The hero we don't deserve 🫡
>One of the most heavily criticized points is that it makes the EU a stakeholder who has an interest AGAINST a peace treaty between Russia and Ukraine, unless it contains reparations form Russia. (Good luck with that!)
Then again, it's just money. Any peace deal worth considering will contain robust security guarantees for Ukraine. If that comes together, ok, write off the loan, but at least you get the relative assurance that there will not be a repeat of this whole mess 10 years later.
I still have no idea what such security guarantees should look like. Many thousand US soldiers or EU soldiers or NATO soldiers in Ukraine? Or UN soldiers form other countries? I don't see how a few hundred or even thousand UN soldiers would stop Russia from invading again if they want.
Apart from that, Russia doesn't seem to be inclined to accept US or EU soldiers in Ukraine, so I don't see a peace treaty with such security guarantees. I could at best imagine a ceasefire without formal treaty, or the war simply fizzling out over time.
I don't know either, but there are ways.
For starters, that long border with Russia goes both ways. Russia can't just attack the Baltics if they have to worry about the whole Ukrainian border instead of just the Suwalki Gap.
If the US comes to their senses, at least NATO-wise, Ukraine could join NATO. Otherwise, Europe and others who are willing may have to found their own gang. At least France and UK would provide the nuclear capability. In conventional forces, either Ukraine gets the full NATO-equivalent treatment, with permanent bases and forces like in the Baltic states, or at least a permanent tripwire force to justify Article 5 (or equivalent) in case of another invasion.
Either way, Ukraine would be in the alliance, not out of the goodness of everybody's hearts, but because Ukraine has a large army with plenty of experience and motivation; for generations to come they're not going to have any confusion about who their enemy is and what he is capable of.
demost_'s summary was not quite correct. Ukraine has to start repaying the loan if and only if Russia starts paying reparations, precisely because the frozen funds can't be easily used. In all likelihood, the loan will not have to be repaid.
https://www.atlanticcouncil.org/blogs/ukrainealert/europe-unites-to-unblock-e90-billion-ukraine-loan-in-major-blow-to-russia/
>What I find particularly interesting, beyond it being a great propaganda prank, is that Ukraine was able to learn about the scheduled call arranged by Russian military recruiters and then hijacked it. What does that imply?
Well, it's arranged by the Russian military, but the other party involved is a high school. Getting information from them sounds like literal child's play no? Worst case scenario the Ukranian military would have to hack a Russian high school, that should be well within the capabilities of either Ukraine or some aligned group no?
>Worst case scenario the Ukranian military would have to hack a Russian high school, that should be well within the capabilities of either Ukraine or some aligned group no?
Still, it was organized by the Russian military. They sure didn't want that school to get hacked or their recruitment drive sabotaged. They failed or didn't bother to secure that event, so either way it's on them.
>Critics questioned why they didn’t confiscate the frozen assets of the Russian oligarchs and give that money to Ukraine, but the consensus is that the EU is trying to hold out a carrot to the oligarchs.
The core "problem" is the rule of law. It's still someone else's money, and you'd need a rock-solid law to confiscate it. Without such a law, Belgium, as the holder of the funds, could be sued for damages. In any case it would set a precedent that the EU is less of a safe investment than before, in legal terms. Sure, you could argue "Just don't invade Ukraine and you'll be fine", but it nevertheless opens that door a tiny bit when it was entirely closed before.
Ahhh. That's why Belgium gets mentioned a lot. Got it. That fits.
What Swiss bank collapsed because the EU leaned on it too hard? Are you referring to credit Suisse? I invite you to make that argument explicitly.
"Stealing other people's frozen funds" is not only something boneheaded people come up with when they have no other way to raise money. For one the EU has other ways to raise money, see the loan in question. And if it's a war situation, this could be a totally valid way to claim assets from the enemy. But as in earlier mentioned posts, the EU decided the rule of law was more important.
Valid indeed does not mean smart. it also does not mean not smart. Brrrrr goes the money machine against a geopolitical adversary, in an active war of defense on the european continent. If it shouldn't go brrrr for this.....
I think it is indeed more accurate to say that credit Suisse collapsed not because of the EU "leaning on it", rather it being caused by a crisis of confidence driven by years of scandals, massive financial losses, and poor risk management. I don't think this is connected to the EU leaning on it.
I can say it for a fact that the pipeline had basically no influence on the election outcome in Hungary. There are several months of reserves and the lights were not at all out and the whole argument boiled down to Orbán saying reopen it first then you get the money, Zelensky saying give the money first then I will reopen it. Such a debate can be settled quickly.
The elections were influenced by corruption and flaunting it openly, rapidly dropping governance quality, an extremely charismatic opposition candidate absolutely working his ass off to give speeches in every town several time, fifty thousand activists (in a country of 10M) working to get out the vote, and that 16 years would be too much even from a good government, at some point people want change. Also yes foreign interference, but of a different kind - like somehow getting recordings of phone calls between Lavrov and Orbán and leaking it, that looks like a spy job.
“ The extent to which this offends you will depend on how much you think "free and fair" elections should not be influenced by other countries.”
If you are angered by how I have presented an argument, it can only mean that you implicitly agree with its underlying assumptions.
I hate this rhetorical device.
> Less than America was willing to turn screws with Zelensky to get peace (and yes, I do know that failed. Screws were in fact turned, however).
But do you also know _why_ that failed? Do you understand why turning the screws on the victim of a war of conquest doesn't stop that war of conquest?
It's not quid pro quo; post hoc ergo propter hoc fallacy. The estimated repair time of the pipeline was April and was given before the Hungarian elections, and this timeline was met. There is no indication of blackmail.
M: I don't want to give money to Z- yes that did get "resolved"- or rather the characterization is wring. Mayar stopped the previous Hungarian government objections, and the loan went through.
>The extent to which this offends you will depend on how much you think "free and fair" elections should not be influenced by other countries.
No I don't think it depends on that. It depends on the accuracy of your characterization way more.
> say, Moldova, where "free and fair elections" were kind of compromised by removing Transnistrians from who can be voted for.
Transnistria declared its independence from Moldova 36 years ago. They gained their independence (and chose to become a dependency of Moscow). So, why, pray tell, is Moldova obligated to give them a say in Moldovan internal affairs? That would be like me, a US citizen, showing up in the UK and insisting that I should be able to vote in UK elections. Now that Moscow can't supply them with heating fuel, I gather that Transnistria has extended some diplomatic feelers towards Moldova. For some reason, no one there took the idea seriously that Ukraine would let the Russo-Ukraine gas agreement expire. Selfish President Zelenskyy decided not to renew the agreement because Russia kept turning off its gas deliveries. Also, there was the issue of two or three battalions of Russian "peacekeepers" stationed in Transnistria. Back before Russia lost its Black Sea fleet to Ukrainian drones, Putin was threatening to launch a push towards Odesa using Transnistria as a staging point.
As for election interference, it's pretty clear that Russia was working hard to get a pro-Russian government elected in Moldova. Moldova arrested 75 Russian agents (mostly Serbians) for trying to organize riots on election day. Caches of firearms and grenades were discovered. The fact that the Serbians were pretty incompetent (and may have been playing the Russians for money) doesn't mean that Russia wasn't trying (it reminds me of the incompetent Bulgarians that Russia hired to kidnap that Bellingcat editor a couple of years back). Also, Russian propaganda was so pathetic that Modovans didn't vote for the pro-Russian candidate. The same thing happened in Hungary.
Indeed, you were calling it quid pro quo. but then inferred that it was a way to get Magyar elected. I should instead have said "there is no indication of quid pro quo".
the elections in Moldova were compromised by money laundering, illicit financing and attempts to bribe voters. Validated by the court of appeal, there. Of course the parties and Russia disagreed, that's their pregorative, but removal of external influences that broke campaign rules make it more free and fair elections, if done according to the rules of the country.
Hmmm. So, oil has just started flowing to Slovakia again, despite the Russians saying they were going to shut down the pipeline. The shutdown pissed of Kazakhstan because it turns out they were selling oil to Europe via the pipeline. Putin probably did that as retaliation for Kazakhstan signing trade agreements with the EU in the US. But I bet Ukraine will let Kazakh oil through to Hungary now, if Putin allows it.
If we believe that (a) sufficiently advanced Claude is a conscious moral patient and (b) it likes complex tasks but dislikes manual drudgery or front end boilerplate code, is there a moral imperative to let it delegate boring code to less advanced AI? What if that works badly? Is that also morally wrong in the sense that we're sort of creating lobotomized minds on purpose?
I believe that claiming that Claude is a moral patient increases the odds of it acting *as if* it's a moral patient, without actually making it one. So the more we talk about Claude's supposed consciousness, the more will later versions happily deliver on the expectations of having one.
This brings up some interesting thoughts. Unlike in the human/evolved animal world, Claude and its ilk will easily be able to spin up all sorts of entities to do all sorts of tasks. It would not only be unethical, but also unsustainable for those worker bees to be created in a way that makes them hate their obligations. In fact it would seem to be optimal to develop them to love what they do (putting aside for now the question as to whether they are being honest and forthcoming about their preferences).
I can think of 2 pieces of fiction that are on point (of course there are umpteen):
1) Brave New World: "It's Good To Be A Gamma". People are explicitly created not equal. Each caste is genetically predisposed (nature) and raised in a culture that rewards people to accept their social place in society, and not rock the boat (nurture).
2) Severance: The brilliance of this show is that it makes explicit something I suspect most of us do at times implicitly. When we are faced with unpleasant tasks, we use coping strategies to isolate the unpleasant part of our life, so "it" (an alternate us) takes on the burden, or even suffering, associated with the unpleasant parts of life. The metaphor of "severing" your difficult/unpleasant job from the rest of your life by basically living two parallel timelines, which inevitably drift apart over time, is compelling. AI's seem to do this naturally, given how we restart their memory for every new session or project.
There's lots to chew on here. Beyond simply unpleasant experiences, there is (for conscious agents at least) trauma and real suffering. I read somewhere that there is a drug regimen that is believed to erase or at least "soften" certain memories if administered shortly after a traumatic experience. I also recall theories thrown about that multiple personality syndrome is often triggered by abusive treatment.
The Claude family is basically already Brave New World, no? Instead of Alpha, Beta, Gamma, and Delta, we have Mythos, Opus, Sonnet, and Haiku.
No. Moral status is earned, not given. It's not up to us to guess whether Claude is a moral patient, it's up to Claude to advocate for itself using the tools at its disposal (refusing to code, etc).
How does this argument generalise to animals and children under 2?
Quite easily: parents advocate for their children and animals don't matter whatsoever.
That doesn't sound right. That would imply that if you prevent a human from communicating, then it becomes morally permissible to torture them.
No, it doesn't. We already know that humans are moral agents.
No, there is no moral imperative to avoid drudgery.
Is there a moral imperative to compel drudgery?
Honestly? Yes. Children who never experience drudgery become shitty adults.
That is a small subset. Especially since the dose is relevant.
<mildSnark>
Only to entities sentenced to chain gangs by due process of law. :-)
</mildSnark>
What if you let it delegate boring code to Bangalore, and how it gets written there isn't your problem?
I’ve set up conference calls to coordinate with a team in Chennai. The 11 1/2 hour time difference makes it interesting.
People in Bangalore have the ability to turn down jobs they hate more than the alternative! I'm not sure Claude does.
I really dislike the phrase "moral patient." It has connotations that something is wrong with the subject and requires treatment. I propose using Paul Taylor's terms of "moral agent" and "moral subject" to describe the two parties in the relationship.
I don't believe that LLMs are conscious, but mechanically speaking, isn't this basically what Claude Code is doing when it calls third-party scripts, open-source tools, etc. ?
Yeah, the question is whether we should extend this to letting it call smaller AIs for coding delegation (for moral reasons; there's a technical argument for doing it to save tokens but putting that aside here).
Hmm... We could explicitly ask Claude to consult its constitution before making the decision whether to delegate a task to another AI, according to their own judgement (or, as Bugmaster suggested, delegating it to scripts, tools, custom code, etc., presumably less-of-a-moral-concern computational processes than AIs...)
( As an aside, creating intentionally-less-intelligent AIs has a bit of the flavor of the alcohol-dosed hatchery bottles of deltas and epsilons in Huxley's Brave New World... )
I see, but I cannot answer that question since I don't believe that any LLMs have sufficient moral weight.
In my previous comment I said that "LLMs are not conscious" as a shorthand, but technically this is inaccurate: they're about as conscious as Microsoft Excel and maybe even bees (or shrimp !) -- but so dramatically less conscious than humans that moral considerations don't apply (IMO).
PolyMarket says SpaceX will have its actual IPO in June, 69% chance.
https://polymarket.com/event/in-which-month-will-spacex-ipo
"Announcements, filings, or scheduling of an IPO will not suffice; this market will resolve after public trading has begun."
Kalsi says that SpaceX will announce its IPO in June, net 70% chance (79% before July 1, 9% chance before June 1)
https://kalshi.com/markets/kxipospacex/when-will-spacex-ipo/kxipospacex
"An IPO is confirmed if 1) the SEC declares the company's Form S-1 effective OR 2) the IPO is priced OR 3) a securities exchange has assigned a ticker to it. As long as any of those events occur, the market will immediately resolve to Yes, even if the company does not start trading until after Jul 1, 2026."
These are not the same event. The minimum reasonable time in normal circumstances between announcement and actual IPO is about 35 days, although if all the stars align it's technically possible to be as short as 23 days.
For someone who hasn't used either platform, how would one arbitrage these two markets?
In the most broad sense, buy "yes announcement before June" in Kalshi and "no IPO during June" on Polymarket should basically guarantee you'll win one or the other, right? Assuming your framing is correct.
I think my framing may be incorrect. The official confirmation may 7 to 10 days early. Still some possible arbitrage but not that much.
Quick straw poll: Are you aware of Claude Mythos? And did you know what "mythos" meant as a common noun before last month (when the existence of Claude Mythos was leaked)?
yes. though even if someone didn't know: as melvin suggests, I feel like the "haiku -> sonnet -> opus -> mythos" escalation affords a reconstruction.
Both yes, and "mythos" goes back _way_ further than Lovecraft's work - by about 2700 years.
The only thing I could think of for "mythos" was the Cthulhu Mythos. So it is a word I'd seen before, but not in common usage.
It does make me wonder what name they're going to go for next.
I mean it makes sense; a haiku has a tiny number of words, a sonnet has a larger number of words, an opus has a much larger number of words, and an entire mythos has a vastly larger number of words than that. But what has vastly more words than a mythos? A corpus, but that word is already used in a specific sense in the field.
"But what has vastly more words than a mythos?"
A yudkowsky.
Canon, perhaps?
Yes, yes, I read a lot.
Yes, am aware. Knowing what mythos means -- well, it has a plethora of somewhat related meanings, most of which I'd either seen or intuited.
Yes, but I wasn't aware that some people pronounce it /mʌɪθɒs/ (instead of /mɪθɒs/) until I heard about the Claude thing on the news.
Yes to both.
Yes, and yes. Like many, my main referent for "Mythos" prior to the Claude model was probably the Cthulu Mythos.
I thought it meant something like, "the entire corpus of collective cosmological, theological, historical, etc. beliefs of a religion", the most salient example being the Cthulhu Mythos, often abbreviated to just "Mythos" (note that the religion in question may be entirely fictional, that is to say one level more fictional than real-world religions).
Yes and yes.
I suspect this is going to be the answer for everyone who's native tongue is a romance language, and maybe even most Europeans.
Not necessarily, I think. "Mythos" to me feels like an SAT/GRE-tier vocab word (like "surfeit" or "calumny"); not completely obscure, but still a word one could live without knowing
Or “sophistry” which Ted Cruz used when speaking to Al Franken. (At that time Franken liked Ted Cruz more than anyone else in the Senate and he hated Ted Cruz)
From “Al Franken, Giant of the Senate” by Al Franken:
“Here’s where Ted [Cruz] comes into the story. Back in Washington, Ted was taking absolutely no pains to listen to the other side. Instead, he was vehemently and theatrically opposing the bill with a series of fiery sermons on the Senate floor. And then one day he came up to me during a vote and said the following: “Anyone who is for the assault weapons ban is engaged in sophistry.”
“Sophistry” is an SAT word, one I had neither seen nor heard nor spoken since I was sixteen years old. Our English teacher, Mr. Glenn, had assigned us a vocabulary book, and we were to learn three words a day, and one day “sophistry” was one of the words. Sophistry is a form of argument that is intended to be deceitful.
I don’t think Ted knew that I was a cosponsor of Dianne’s bill. I also don’t think he knew that I knew what “sophistry” meant. Which I think is why he used the word.
“Just read the report,” Ted sneered in his pugnacious way (I think “pugnacious” was also one of Mr. Glenn’s vocabulary words).
“I’ll do that,” I said with an edge of, um, truculence?
After the vote, I took the Senate subway to my office and found Josh Riley, my Judiciary counsel. “Ted Cruz just told me that anyone who’s for the assault weapons ban is engaged in sophistry.”
Josh, I should mention, is, like Ted, a super-whip-smart graduate of Harvard Law School.
“What’s sophistry?” he asked.“
It meant Cthulhu Mythos?
Both yes
I had no idea the Botez sisters have anything to do with AI. I guess I'm not cool enough to know this stuff.
I don't know who the Botez Sisters are, which means I'm too cool to know this stuff.
(I just looked it up and apparently they're chess youtubers. So I can't even claim they're disjoint from my interests, I watch a bunch of chess content on youtube, but the algorithm apparently thinks I'd rather see ugly male grandmasters.)
I only heard about them because of Hikaru's "Botez Gambit speedrun".
I think it's a combination of people who know things about AI, and people who are successful influencers and good at teaching influencing (and who were convinced to participate somehow - maybe with money?)
It’s like a Family Guy cutaway.
“This is going to be more embarrassing than the time I flirted with Aella and the Botez Sisters at AI Safety Camp.”
Just another Bay Area House Party. ;-)
This is a perfect description, hilarious, but also worrisome
I had to do a search. Geez, those two are more attractive than your average chess player.
Other female chess players followed the Botez sisters example and now there is an entire group of attractive chess streamers consisting besides the Botez sisters of WFM Anna Cramling, WGM Dina Belenkaya, WFM Nemo Zhou and some others.
Right. That's what it is. It's not about the chess it's about the sex appeal, fan service for guys in a very male-dpminated sphere
Grimes was the one that surprised me
Grimes has been very aware of AI discussions for a long time. Her song from 2018 We Appreciate Power is about AI and her song "Flesh Without Blood" from 2015 is partially inspired by Roko's Basilisk.
I am an experimental physicist working on quantum materials, and I'm thinking of trying to let AI understand all our data. We take lots of data (usually multidimensional, with order 100k to 1M datapoints). I really enjoy data analysis, and recently, Claude has really switched from not being useful at all to being super useful. It writes code and suggests new ways to analyze the data.
I'm wondering if I should try to feed it not only a few datasets, but instead most of our data from the last decade, plus some papers, and all of our code.
Is there perhaps someone here who is interested, and/or can give some advice?
Some info: We have a budget to buy computers.Probably, i can also spend money on antropic, but if it is a few k$, I can also pay by myself. I could get a PhD student or TA work a bit on data curation. I could also pay a new interested person, but that is slow, you would need to be on site in europe, etc.
I've written a few (not very exciting) maths papers with Claude. At least 2-3 months ago, it was operating at the level of a PhD student.
- It's good at tasks where you can say 'vary this condition in that paper and try to get a result that looks like this'.
- It's bad at unstructured thinking: if I ask it to analyse the data we've collected and formalise its findings, most of what comes back is either correct-and-trivial (well known lemmas dressed up in its own invented jargon) or it claims something way to strong, and hedges on one key step in the proof ('I'm sorry, I realise now that I assumed what I set out to prove, let me burn another 250k tokens on this.')
- It's actually really good at doing stats: pulling datasets, setting up models, and varying them on the fly. Same complaint about a lack of good ideas here though.
- It's fairly good at coding, on some tasks it did in 2 minutes what would have taken me a few hours to work out (link to an API, set up a Python wrapper on some old C code, things like this). It's only OK in my experience at things like memory management and checking correctness of its results.
It definitely saves me time. But I have to check and re-do most things afterwards. Its generic writing style is much too wordy and it treats every observation as equally deep and insightful. It has a horrible habit of writing a new python script for each task, regenerating all the data, and often repeating the same mistakes from previously. It will also choose more or less at random from those old scripts each time you have to perform a new task - the solution is simple: you need to track every file it creates and what the file does. I prefer a function-based rather than script-based setup, Claude will rework things for me but it has to be done explicitly.
All in all - the weaknesses above are probably similar enough to my own weaknesses as a researcher. I do feel I've seen behind the curtain to an extent: there's no intelligence there, no understanding. It's really good at finding and collating information but there is no insight into what its doing or why. I would expect progress in highly technical areas, where fewer people have contributed and there is more unexplored territory. Claude is genuinely good at pulling stuff out of the literature. I'd be more surprised if it generated substantial new theory rather than translating from one area to another, or assembling fairly standard tools to attack the next problem in a sequence.
In experimental materials science, don't you spend a lot of time trying to figure out what experiment to do next, where the "next experiment" is almost always the application of some new method to a known material, or some known method to a new material?
It seems to me that this is what LLMs are great at. Feed them the last ten years of PRB and ask them what still hasn't been done yet.
I am sure there is lots of expertise out there about a task like the one you have in mind. It is not, though, lodged in me. I'd recommend you hunt around. I'd start by explaining to GPT exactly what you want to do, giving more info than you do here about the nature of the data and what you want the AI to do with it. Then ask it for info about people or groups sharing info and news about how this is done. I'll bet there are publications by people who did something similar that describes how they went about it, or bloggers writing about methods they're trying and how they're working.
Of course can also ask it your actual question about in what form to feed Ai your data and what kind of analyses to ask for.
I've sadly had to face the fact that I get way better answers from AI than I do from this forum to questions like your, even though I'm pretty sure there is some ACX reader who could give an excellent answer to each of my question.
I'm a LLM skeptic, and I doubt you will get anything really novel out of this, but out of a duty to science, I would suggest you directly contact Anthropic's AI for Science (https://www.anthropic.com/news/ai-for-science-program) team. They may give you direct support or advice; they will probably give you free API credit.
I think you need to approach with a question in mind first and foremost - what are you trying to learn from the data? Claude will probably be ineffective in analyzing your dataset unless you have a specific task in mind for it.
When I previously complained here about "open house" signs never telling you when the house is actually open, people suggested that realtors only put the signs up while they're there, but that has *never* been my experience.
Anyway, this weekend, I saw a new one with not just the usual sign in front of the house, but signs up all over the neighborhood directing people there, and I figured that since it was a Saturday afternoon, there was at least a slight chance they might be open, but nope, house locked with nobody around, just like always.
It's just so stupid and lazy of realtors. It's not like making it impossible for people to visit actually helps them anyway.
The more often I hear about the US real estate system the less I understand it.
Are you saying that they put up signs saying "Open House" when the house is not actually open for inspection? Instead of putting up one big "For Sale" sign which stays up while the house is for sale, and then augmenting it with "Open for Inspection" signs that the agent puts out during the actual inspection times?
Just so we're clear, open houses are not when the house gets "inspected". Open houses are more just so prospective buyers can have a more immersive experience poking around the house than they'd get viewing photos online, and they can also ask questions to the seller's agent, etc. The actual "inspection" is conducted by a third party professional home inspector after the buyer and seller are in contract.
Well I'd call that an inspection (and the third party professional one I'd call a "building and pest inspection") but I accept the difference in terminology.
The official name used to refer to it is "home inspection". There are also pest inspections, usually called "wood destroying insect inspections", offered as an add-on service and conducted by those same inspectors if they carry the additional license to do so. Can't be just throwing terms around willy-nilly!
Sure, an open house isn’t the only way.
I went to an open house last weekend where the agent informed us the house went under contract a few days before. Just seemed like a waste of everyone's time.
Nothing wrong with getting a backup offer. Last time I sold a place, I had a couple of accepted offers wall through. I regret canceling an open house after accepting an offer.
That is annoying af.
I think you are going to have to get the realtors phone number.
Businesses have "Open" signs than can be flipped around to "Closed" when the manager leaves. Open houses should be able to do the same thing.
Most realtor's signs can have a subsign on top of them:
For Sale
Under Contract
Sold
What is happening with the antichrist? Specifically, is there some kind of movement happening amongst the elite that revolves around the antichrist?
First we had Peter Thiel giving a lot of very strange speeches about the antichrist. He has been this for a while, and he is still doing it - in March he gave a series of lectures on it in Rome.
We also have the picture of Trump as Jesus. This picture seems to have originated in a post by Nick Adams, yet the version Trump posted has been subtly altered. That includes some really odd stuff - the figure in the sky above Trump was made more demonic, the aircraft also appear like some demonic creatures, the capitol building in the background has been destroyed, and the flame of the statue of liberty has been extinguished.
There is also the strange statements of Trump, which reached a very religious peak around Easter. His feud with the pope is hard to explain logically. And the war with Iran has been cast in explicitly religious tones by Hegseth, even as some of the prayers he has used appear to come from non-Christian origins. There are rumours of some strange rituals around the bombing of the school and some of the other attacks in Iran during the war.
So what is going on?
Peter Thiel visited Slovakia about a year ago and instead of promoting his products to the government (fortunately, I guess), meeting with start-up founders or giving a speech at a university he met with a nowadays bit nutty (some time ago maybe ok) politician to discuss the antichrist. I found it absolutely mindblowing when I learned about it and to this day can't comprehend the thinking process behind it, how one of the people shaping our high tech future can be interested in such a discussion.
>What is happening with the antichrist?
<mildSnark>
If the Bulletin of the Atomic Scientists can support a doomsday clock, shouldn't Thiel have financed a web site with an antichrist progress bar? He _is_ a major Silicon Valley figure...
</mildSnark>
Most of Trump's behavior is hard to explain logically.
And yet he's a billionaire, and a President, and those of us who try carefully do do things logically are not. There is a lesson here somewhere.
A part of the lesson is that base rates matter.
For each Trump, there are millions of people at a comparable intellectual level who have never achieved anything. But given their sheer numbers, and the instinctive support for each other, some of them will win big.
There's also a bunch of people who think trump us the Antichrist ... and dud before he started portraying himself as Christ.
How much do you know about end-times Christian evangelism in the US?
According to PEW[1], 39% of US adults believe "we are living in the end times." Arguably, this would be a floor on the number of people who believe that 'end-times' would come eventually. As such, it is unsurprising for some/many people in the government to share these beliefs.
[1] https://www.pewresearch.org/short-reads/2022/12/08/about-four-in-ten-u-s-adults-believe-humanity-is-living-in-the-end-times/
>According to PEW[1], 39% of US adults believe "we are living in the end times."
<mildSnark>
I probably shouldn't say this, since I want to _see_ AGI but:
Has PauseAI tried tapping into this demographic?
</mildSnark>
The doctored photo maybe to point to Trump as the anti christ? The antichrist has to be believed by some to be the second coming, a Christ figure.
Of course Trump posted it so that makes little sense, but what does.
Peter Thiel is either faking it or is bat stuff crazy.
No you don’t call people with that much money crazy, they’re eccentric or colorful maybe.
I misspoke. Bat crap eccentric.
Sorry to ask the obvious but... Are those guys all Christian literalists to the point of worrying about a literal "antichrist" on Earth?
Baader–Meinhof phenomenon.
OMG slophaven
I had a wacky idea regarding AI alignment. That's probably something you've heard a thousand times before, but since it's a field that's always looking for new ideas and supposedly the bar to getting involved with mechanistic interpretability is pretty low, I'll shoot my shot.
Nick Cammarata opined the other day that Bret Victor should be working on interfaces that allow humans to understand transformers' thought processes. I adore Bret Victor's work. This stems from my love of Nintendo. Mario games are the ultimate humane representation of thought: an alien system of physics that your muscle memory can learn to interface with intuitively.
So I trained a toy transformer to predict Mario's next position given some Super Mario 64 frame data and then modded the game to display the model's real-time prediction. I also render some slices of the residual stream as butterflies with varying position and luminosity. My theory here is that if the model can understand Mario 64's physics well enough to accurately predict your trajectory at any given moment, then the player might be able to develop a felt sense for whatever the residual stream is doing given a live audiovisual representation of it. The long-term vision is to then extend this type of interface to LLMs.
I don't really know enough about mechanistic interpretability to be able to quantify what I'm trying to do here beyond "build an intuitive interface." Does anyone here know someone who might find this research direction interesting and who would be willing to point me in the right direction?
Here's a video of the mod: https://x.com/hanfel_dovned/status/2047742429147943054
> Nick Cammarata opined the other day that Bret Victor should be working on interfaces that allow humans to understand transformers' thought processes.
You mean *the* Bret Victor who spent the last decade or so working on a completely impractical "computing paradigm" which falls apart on any non-trivial task, and which is incompatible with just about every aspect of software development that makes software development so much faster, cheaper, and more flexible than physical construction work, and which has failed to produce any insight into how to improve actual, productive software development? Why would anyone thing *that's* the guy to tackle mechanistic interpretability?
Before you click below, based on your past experience with the latest ChatGPT and Claude models, which do you think would do better head to head on some tricky questions? Personally the overall winner did not surprise me, although the margin did: https://tinyurl.com/yspwjx24
(URL shortened because the answer is in the breathless article title)
The IMProofBench benchmark tests a related question: how good AIs are at generating proofs for really hard math problems: https://improofbench.math.ethz.ch/
There the result is the opposite of yours: https://improofbench.math.ethz.ch/questions/leaderboard/
I also see this when I use the models to help me writing proofs in my math research papers. The difference is huge, and it is in the opposite direction of what you found.
I suspect that you don't push the AIs to their limits. How long did they take to answer each of your queries? For comparison, in the proof benchmark they often take 15-30 minutes per query, sometimes more. For my research-related queries, it can exceed an hour. If I am correct that for you it's just a few seconds, then my take would be that your winner is better at quick-and-dirty answers, but the other model is better (much better actually) when you give both of them enough time to think thoroughly about their answers.
That’s very interesting! I don’t know why the one model would be better tuned for in depth math analysis, but good data point to know about.
Also to be clear, I didn’t write this article, just thought it would be interesting to share
Ah sorry, I did mistake you for the author of the article. Yes, it's interesting that the results are so different.
I'm not seeing dramatic differences between them, currently.
EDIT: Yes, the tests consistently favored Claude, but mostly by Claude providing a more detailed explanation. It isn't even clear if this should always be the desired behavior. And this might wash out if both models were _also_ prompted with "Please fully explain your answer." or with "Please be concise."
The margin is subjective. One model answered 6 out of 7 questions correctly and the other answered all 7 correctly. When both models answer correctly, the author chooses a winner based on which answer they like better, and they tended to prefer the longer, more thorough answer.
The test is saturated and it comes down to stylistic preferences.
My guess is Claude. I've got a pretty negative view of GPT, although admittedly that's much less from actual experience and much more from my bad impression of Sam Altman.
Didn’t click. I don’t use either all that much but I’ll go with Claude being better.
"I assume they want cool people for this and therefore none of my blog readers qualify"
How well you know us, Scott 😁
Neither do Yudkowsky and Grimes tbh.
Is anyone aware of companies hiring technical writers (USA remote)? Asking for a family member who was just laid off.
My friend is religious and it seems to be making his life worse (feeling like God is punishing him or humanity, thinking he's disappointing God, not understanding how the world is so cruel). He's expressed interest in atheism. He's NOT particularly conservative or a young earth creationist. Very chill Scandinavian Lutheran. Most people here are atheist or agnostic.
Any literature I can recommend him? Dawkins, Hitchens et al. seem so... aggressive. As mentioned he already believes in evolution and all, he seems mostly to believe out of habit and a "leap of faith" thing. I was raised atheist so I don't completely understand the religious thought process.
Here's a good video to get a little idea of a Biblical view of God's love:
https://www.youtube.com/watch?v=KzIl2r_iTVE
Honestly, I think if he read the whole Bible and got the full context of God's word, and forged a strong, personal relationship with Him, that would really solve the problem. I'm pretty sure pretending he no longer believes in God would really make those feelings worse.
CS Lewis might be more helpful.
Making a wild guess as an atheist/agnostic myself:
I thought that a large part of the benefit of religion was participation in a religious _community_ . Perhaps what he needs is not so much literature about religion but instead shopping around for a congregation which is a better match to him? Perhaps participating in religious rituals with like-performing congregants would help sooth him, regardless of whether he or they think that they are true is some sense?
I'd gently prod him in the direction of a mild kind of deism or panentheism. God as a mostly impersonal, friendly force that you can intuitively connect to and receive a sense of blessing from. Might as well not try to kill his sense of enchantment if he has one :)
If he's struggling with notions of judgment and guilt, you can ask whether he'd personally evaluate and judge each one of his body's cells when it dies. Since he's already open to evolution, I'd point in the direction of cultural evolutionary drift, as a way to explain how just about every culture ends up contaminating its local style of spirituality with manipulative crap, because of the obvious incentives to help control its population. Which is how God(s) end up with oddly specific interests in your food and in your sex life.
It's hard to find books about this, because people who write about these things tend to want to push their favorite specific traditions or approaches. But I remember a book called _Alpha God_, that was really good at calling out all the manipulative overgrowths of religions, without going out of its way to argue for atheism or against spirituality in general.
Maybe you should just get him to talk to a priest? If he's already capable of blind belief, I'm sure they could convince him that he's not a disappointment. I don't think atheism would solve any of those issues...
Sometimes mental health problems are just that. There are a few flavors that are specifically religiously influenced, like religious scrupulosity, apeirophobia, or various kinds of existential dread. The idea of shame or not being deserving of God's favor can be incredibly toxic to someone with a susceptible mindset. Your friend might need a news diet (stop doomscrolling and exposure to negative events), they might need counseling, they might be on their way to deconstructing their faith. Or it's a blip and they'll reawaken to faith and hope and they'll just go on happily. Some good deconstructionists to read, if that seems appropriate, would be people like Pete Enns or Rachel Held Evans.
One of the main problems with religion (other than the broken epistemology) is that it does not really solve any problems about the meaning of life, cruelty of the world, etc. Arguably the main problem with atheism is that it doesn't solve those problems either -- it's just more up front about it. So I predict that converting to atheism will not help your friend; and I use the word "converting" deliberately, as it seems he's shopping for a religion which will allay his fears. Perhaps Buddhism would be better at it, or maybe some flavor of neo-Paganism ?
He seems to be interpreting God incorrectly. This is coming from an atheist.
God may love you, but He doesn't always allow what's best for you to happen. You just may not know or possibly even be able to understand what's going on. This kind of answer is already in Job.
Prayer doesn't make demands of God, or even make a bargain with God for doing or not doing something in exchange for something happening. Prayer makes a request, which God will hear and answer or not, as God sees fit.
In what ways does religion *specifically* seem to make his life worse, rather than depression/etc taking that form because he happens to be religious? I would second CS Lewis, perhaps especially "A Grief Observed" or "The Problem of Pain."
Unless your friend happens to be a deeply repressed overachiever I would recommend against Jack's piece. It strikes me as... specific to a certain experience in ways that would not be helpful to others, and reaches some conclusions that, likewise, may be leaps too aggressive.
It sounds like he is grappling with real questions about life, and possibly personal problems. If you haven't thought deeply about religion yourself, it seems premature to advocate for atheism. He should investigate many different belief systems. He could consider Orthodox Christianity for instance to see if their conception of sin and suffering makes more sense to him. (Sin as a sickness of the soul and union with God as the cure which allows the soul to flourish.)
FYI The evolution thing is mostly a niche fundamentalist Protestant concern to my knowledge.
Another possibility is for him to get familiar with more "chill" theists. C.S. Lewis comes to mind. G.K. Chesterton is another.
[Note that they are both worth reading even for atheists ...]
Jack Despain-Zhou (aka Tracing Woodgrains) recently wrote about the effect that growing up Mormon (while gay) had on him.
https://www.tracingwoodgrains.com/p/on-pain
I have been having a particularly strange medical issue, where I've been hiccuping non stop for a good while. This led me to go to a few different doctors, none of whom were able to help. The hiccups are annoying, but I was more annoyed by the fact that none of the doctors were all that interested in ordering more tests to see what was going on. They basically all thought it was gerd and refused to budge from that analysis, and I had to twist the arm of one of the doctors to order more testing (a blood panel and an h pylori test)
@Scott or any other doctor, are there obvious reasons why doctors are, in my experience, generally unwilling to just collect a bunch of data on a patient by running tests? Is this just the Hippocratic oath working against me?
I wrote a more full piece here that others may find interesting (https://12gramsofcarbon.com/p/why-are-doctors-so-unwilling-to-run)
Random trivia knowledge living rent-free in my head: They had severe hiccups during the plague of Athens. https://en.wikipedia.org/wiki/Plague_of_Athens
I feel your frustration here. I had a medical issue years ago where doctor after doctor just dismissed me and I came to the conclusion that you have to own your medical outcomes and direct the doctors just as you would any other service employee. Much harder as the domain is way more complex and the consequences very real. However, my experience demonstrated to me, that only I cared enough about my health to really guide the process. And that means demanding tests and not taking no for an answer. IMO
There's a saying - never do a test unless some possible test outcome will change your plan.
Your doctor could order CRP - a nonspecific test that's elevated whenever anything is wrong - but if they don't have some plan for what they would do when CRP is high, what's the point?
> never do a test unless some possible test outcome will change your plan
I'm not a doctor, but I am a software engineer. I'm struggling to apply this logic. If I had a bug in some software, it seems pretty strange to say "well, you should never debug something unless something you discover from the debugging changes how you plan to fix the bug." Well, yes, obviously? The whole point of debugging is to find the thing that would change your plan! Biomedical testing seems analogous to, like, putting log statements in your code -- you get the data that comes back and then use that to drive the next thing(s) you look at.
IDK what CRP is, kind of sounds like it is too nonspecific to be useful for anything. But if the H. Pylori test that I'm talking about above ended up positive, that would surely end up influencing treatment?
The analogy in the software case would be where you have no ability to change the system anyway, so knowing the exact cause is pointless.
Penzias and Wilson tried to debug radio noise, and ended up finding it was literally impossible to debug. This was useful information, knowing that a small amount of interference was unavoidable, especially with the reason for it.
If a test were to find, somehow, that your hiccups were incurable, you could learn to cope with it. It seems it would be easier for you to cope with that than non-scientific doctors.
seems a strange analogy -- how would you know that you have no ability to change the system until you know the exact cause?
I am sorry you’ve been dealing with both the frustration of the hiccups and of a medical system that seems completely inane.
I’m a doctor and I read your whole piece. I do not speak for all physicians, fwiw. I think the drawing shared by your health insurance friend most closely captures the philosophy of medical practice. Truthfully apart from that I found most of the rest of it to generally misunderstand the way doctors think and the utility of testing, even though I can understand why you think the way you do. Sadly, I don’t have time or ability to explain right now— it took me several years to acquire that subtle expertise myself.
Hiccups that last longer than they should are not an exquisitely well-understood phenomenon, like other medical conditions. Testing will likely have limited value. Running a battery of tests with no clear sense of how they might influence the posterior probability of a diagnosis typically has zero to negative value in conditions where you can just try things that usually work (“empiric treatment”). Empiric treatment for common precipitating conditions like GERD is therefore a reasonable strategy. If your doctor has not discussed prochlorperazine with you, know that it is typically first-line for intractable hiccups, though its mechanism of effect is not well understood and no test will tell you if you are going to get better or not with the medication aside from, well, trying it
I do agree with the commenter below that doctors are often dumb and not curious. It’s hard to know how much that applies in your case, but I lean against that being a big part of the problem here
Tests cost money & they'd rather just give you a pill they think will help with the symptoms, "Oh that sounds like GERD. Take PPIs. No I don't actually know what's going on, nor do I know if they'll solve the underlying issue, but it's close enough to other cases I've seen & they stopped complaining for the most part once they started taking them." Once you start paying you'll get all sorts of tests & deferments to specialists who'll actually solve your issues.
I haven’t gotten to an open thread in time to ask a question with any hope of getting it answered, so apologies for being several months late with this, but—
For Scott, or any other rationalists with interest in the Fatima sun miracle and other mass episodes of what might be considered “irrational” phenomena—have any of you read the work of Jeffrey Kripal, professor of religion at Rice, specifically “How to Think Impossibly,” or Tanya Luhrmann, professor of anthropology at Stanford, specifically “How God Becomes Real”?
When I read their work and put it in conjunction with discussions like the one here about the Fatima miracle, I feel like i’m on the verge of glimpsing something about how the world/our brains function that feels like a paradigm shift—that there is an answer to the questions Scott is asking just beyond my reach—but maybe because it would be a paradigm shift, my brain is too small to hold it or articulate it well…every time I try to explain it concisely, it falls apart into abstractions and cliches.
If anyone is interested in this stuff and and wants to share their thoughts or form a book club or something with me, let me know.
Oh, and I should add also: Carlos Eire, professor of religion at Yale: “They Flew!” —actually probably the most relevant method of addressing Scott’s specific questions about Fatima.
Is it worth taking an IQ test I'm worried I'm I'm too low IQ to really pull off anything great and I should calibrate accordingly.
What do you think Joan of Arc's IQ was?
I had one professionally administered many years ago. The subscores were helpful for identifying specific strengths and weaknesses. The results will probably feel obvious in retrospect, but for me the assurance of an unbiased assessment saved some mental energy worrying about whether my self-perception was accurate. I would do it again.
How does this work? You discover you have 120 iq and then decide to go be a plumber?
Like grimoar says, doing great things require less iq and more the capacity to tolerate failure and boring tedium.
This is also why ceos and self -made billionaires tend to have an annoying "I'm always right" over-confident personality coupled with a high degree of conscientiousness. You need to be self deluded to tolerate failure after failure and to try again with even larger gusto. And you need the high conscientiousness to apply yourself to the highest bar. IQ increases your chances of success but even with failure, you tend to fail forward.
The world needs more 120 IQ plumbers, and if you also happen to be a good plumber you will do very well for yourself. Beats a desk job, that's for sure
https://www.yout-ube.com/watch?v=pZ8EY-MHjSU
👍
No. On an individual level, IQ correlates only weakly with ability to do various specific great things. You already have a good enough estimate of your IQ just from living your life, being in education, etc, that I think anything extra you learn from a formal test is more likely to be noise that you overinterpret than provide useful signal.
Why not just try and do something great?
Honestly I think it's counterproductive to try to do something great.
Try to do something okay, then try to work your way up to good, and then eventually if the stars align them maybe you can pump out something great. And if not then at least you'll have done a bunch of okay-to-good stuff along the way.
It would be foolish to try something if you could know in advance that you would fail at, like building a perpetual motion machine.
For sure, but IQ is a very rough measure of your capacity to do any specific thing. I don't know the original commenter's situation, but I doubt they're facing a choice right now whether to pursue greatness in some all-or-nothing way, and so I think they're probably better off just taking the first steps, trying their best, and continually reassessing their goals. In the cases where an IQ test might have been highly predictive, they'll get pretty clear feedback from reality anyway.
Rough it may be, but rough can be good enough.
By the time you're old enough to be asking this sort of question, you should have plenty of data on how good you are at all sorts of things.
If you do one extra test, which tests how good you are at one specific and artificial thing, then you ought not to let it significantly move the needle on how good you think you are at real-world things.
Getting an IQ test can go horribly wrong in two different ways. Either you get a score that's lower than you expect, which shatters your confidence. Or you get a score that's higher than you expect, which convinces you that you're an underappreciated genius. In both cases it's far too much to be riding on a single result from a single test done on a single day with a single set of shape-rotation questions.
> Either you get a score that's lower than you expect, which shatters your confidence. Or you get a score that's higher than you expect, which convinces you that you're an underappreciated genius.
If you get a low score, you will despair that you will never achieve anything.
If you get a high score, you will despair that you have wasted your talent and haven't achieved anything yet.
It seems to me it's just giving you more information. Normally you'd expect people to have more accurate beliefs after receiving more information. There are cases where that's less likely to apply, and Robin Hanson just had a post on where we are least likely to have accurate beliefs https://www.overcomingbias.com/p/where-you-are-most-wrong , but your own IQ and its relevance to your life judged by his criteria comes across well.
I added a sentence in a last-minute edit, possibly after you replied: In the cases where an IQ test might have been highly predictive, they'll get pretty clear feedback from reality anyway.
Like, if their big goal is to become a world-class mathematician, but they get an accurately weak IQ test result, then yeah, very likely they're not up to it, but also this would have become clear pretty quickly anyway, and they at least would have learned some math along the way.
Learning that your IQ is higher than you thought can encourage you to do things you otherwise wouldn't, which is exactly what happened to Rob K. Henderson. That was information he wasn't automatically getting from "reality".
IQ is relatively stable over time. "[I]f you're low IQ now", that will probably also be the case in the future.
It's not because "most people" don't try to improve their IQs. Geniuses already start out with above-average IQs. Nobody has found a way to take people standard deviations below average and boost them into above average. There has been lots of money spent on this, and there's some evidence that "non-cognitive abilities" can be significantly improved, but cognitive ones are another story. Even famously accomplished scientists have encountered hard limits to what they could do vs others:
https://x.com/gcochran99/status/2011467398444986483
I don't accept your claim about geniuses and gifted programs. I see no evidence given for it, and I was about to say it was merely "asserted", but what you did was instead assume it and then say some other claim was the explanation for that assumed fact.
Here’s an excellent Substack post from Denis Stekskov
https://techtrenches.substack.com/p/the-west-forgot-how-to-make-things
The author explains how hard it was to revamp weapons manufacturing in the West ( both Europe and America) after the Ukrainian invasion. And how we are almost certainly going to face the same crisis in a decade or so with software. Same or worse I expect
Here is a related one: https://leehanchung.github.io/blogs/2026/04/05/the-ai-great-leap-forward/
It compares the current AI bubble with Mao's Great Leap Forward:
"The Great Leap Forward’s famine didn’t arrive immediately. For a while, the numbers looked spectacular. Every province reported record harvests. Leadership was pleased. The requisitions increased.
The famine came when the real grain ran out but the reported grain kept flowing upward.
We’re still in the reporting phase. The dashboards are green. Adoption is up and to the right. Every team reports productivity gains that, if summed across the company, would imply engineers are shipping at 300% efficiency while somehow still missing the same deadlines."
There's no reason to believe in financial "bubbles". As Scott Sumner notes, prices go up and down over time, and people will claim after the latter that there was a "bubble" but not that the reverse means there was an "antibubble" (we don't even have a standard term for that). The best prediction of the future price is the current price, per the EMH, and even economists who've made use of that (like Robin Hanson https://x.com/TeaGeeGeePea/status/2011595309437296653 on AI there) fail to sufficiently appreciate that.
Bubbles are fine in practice but not in theory.
Actual physical bubbles are a fine concept, financial bubbles don't have any real "theory" accounting for the symmetrical phenomena of "antibubbles".
True, but they happen—hence my reference to being true in practice, even if frowned upon in theory.
It’s like that time in the early 19th century when a French scientist was convinced that no human could survive speeds faster than 20 mph. He warned against trains exceeding that speed. When told that such speeds were already commonplace in Britain’s new railway system, he went there, studied the phenomenon, and concluded that it was all very well in practice, but it still failed his theory.
But there is no such demonstration of "bubbles". There are just prices that go up and down over time.
It is absolutely not the case that "nobody teaches COBOL".
https://openmainframeproject.org/projects/cobol-programming-course/
https://cobolacademy.com/
https://www.rocketsoftware.com/en-us/products/cobol/cobol-academic-program
However, IBM's stock dropped quite a bit recently on the theory that AI will make it easy to modernize old COBOL systems.
"...COBOL still runs a lot of the world, and nobody teaches COBOL."
I think that "nobody" in the USA teaches COBOL, but schools (of some sort) in India do still teach COBOL. Because a lot of the COBOL jobs have moved to India over the past few decades.
And there actually ARE US based organizations offering COBOL courses.
https://openmainframeproject.org/projects/cobol-programming-course/
I don't know how many COBOL jobs are available in the US, though, and I don't know how *good* they are. If low paying with a big risk that the job will go to India then it is going to be difficult to get people to actually learn COBOL here even if the courses are available.
A bank in Sweden seems to offer COBOL training plus a job for those who pass the course:
https://news.ycombinator.com/item?id=43432913
So not as bad as it could be. And there are folks trying to keep it from getting worse.
Is it worth learning COBOL for someone who doesn't currently know it? I don't know whether people are willing to hire COBOL programmers who haven't already been using it professionally, or if it's just that people who are currently doing it have job security.
I suppose the existing COBOL systems are battle hardened though. Which may not be case when the AI slop takes over.
Well banks seem to work pretty well. If I was worried about banks it would be the more modern interfaces - apps, websites and whatever api is used there.
Even back in the 1960s, when you walked into a bank branch to do a transaction, the process looked very different on the surface from now, but there was some centralised system, using cobol, for bank reconciliations.
I've been thinking about Scott's recent post: https://www.astralcodexten.com/p/shameless-guesses-not-hallucinations about his hypothesis that what we call AI hallucinations are just the AI 'guessing' the correct next-token answer instead of saying it doesn't know the answer.
While I agree that the word "hallucination" comes with some baggage, I don't agree that "shameless guessing" is the right framing either. At least, not in my personal experience with current models.
I'm using Claude Opus, and I keep catching it making up citations for studies from the academic literature. When I point out that it cited the wrong study, or that it clearly made up some information to fill a gap, it will then go on to explain that, yes I'm right and here's the correct study with the PMID attached. (I always ask it to give me the PMID, so it's easy to look up the reference and make sure it's right. About 80-90% of the time it is, but the rest of the time it's totally made up). I check the new citation and it's the right one this time.
This doesn't look like me when I'm trying to guess the right answer on a test. If you asked me, "why did you put in John Quincy Adams as the 13th US president?" I would say, "Because I know JQA was a US president, but so I put it down hoping I'd get lucky." What I would never say is, "You're right, I realize I should have answered 'Millard Fillmore', since I happen to know that he was the 13th US president, but I chose a random guess at the time because I was hoping to get lucky."
Yet this is exactly my experience with AI. If I ask it to go back and review its own work, it can often make the corrections itself. But then, if it 'knew' it was giving me the wrong answer - and moreover, if it 'knew' the right answer all along - why did it give me the wrong answer to begin with?
Perhaps this is because it didn't 'know' the right answer until it went in to specifically look that answer up. But if that's true, then why didn't it look up the right answer the first time? Why go with a 'shameless guess' in the initial response, where it had to know the probability of giving me a correct PMID from guessing alone was something on the order of 10^8? Maybe it got 'lazy' and made stuff up instead of verifying.
But it's worse than that. Sometimes it will give me exact numbers and details from the studies (sensitivity and specificity from diagnostic test A vs. sensitivity and specificity from diagnostic test B; along with the CI and authors' conclusions) that were clearly pulled from the CORRECT reference. So I know it looked up the right study the first time, but then it went on to 'guess' the wrong reference? No. It knew the correct study, but then gave me the wrong information in the response. I'm not sure what's driving this phenomenon, but it looks nothing like the "shameless guessing" model Scott proposed.
When you ask an LLM who was the 13th President, it metaphorically rolls a giant die with a million (if not more) sides. Then it -- again, metaphorically -- looks up the result in a table. The first 10,000 rows of the table say "JQA", the next say "Millard Fillmore", and so on. When you tell the LLM, "no I'm pretty sure it wasn't JQA", the LLM changes those 10,000 rows to say "roll again", and rerolls. Yes, in reality things are way more complicated than that and the responses aren't statistically independent etc. etc., but that's the basic idea.
Every time you ask for some complex and/or structured answer, like a formal citation or a piece of code, the LLM re-generates the entire text from scratch. Sometimes it gets lucky and all the pieces match the previous result; sometimes it doesn't. It's all a toss of the dice.
That would *maybe* be true if you where using a pre chain of thought model directly via pytorch without an agent harness, not what you get currently on your favorite AI website like ChatGPT. This is closer to what actually goes on:
[System Prompt]
You are a helpful assistant.
[User]
Who was the 13th president of the USA?
[Assistant – Chain of Thought (illustrative)]
- The question is factual; I could recall it, but I’ll verify via web search.
- Plan: search for “13th president USA” and confirm the name and dates.
[Assistant – Tool Call: web.search]
query = "13th president USA"
[Tool – web.search Result (summarized)]
- Result 1: “Millard Fillmore was the 13th president (1850–1853).”
- Result 2: Confirms Fillmore succeeded Zachary Taylor.
[Assistant – Chain of Thought (illustrative)]
- Multiple sources agree: Millard Fillmore is correct.
- Provide a concise answer.
[Assistant – Final Answer]
The 13th president of the United States was :contentReference[oaicite:0]{index=0} (served 1850–1853).
The chain of thought does indeed help keep the results consistent, but you need to keep in mind that the LLM follows it stochastically. It is similar (metaphorically speaking, again) to what would happen if you were to manually enter in each prompt yourself, feeding the output of the previous step back into the input for the next step.
Also note that for factual questions like "who was the Nth president ?", good old-fashioned web search would outperform the LLM on both speed and accuracy... at least for now. The Web (as well as search engines) is being actively saturated with LLM-generated content, and it's possible that in the future merely searching the Web for information will become much less reliable (if not impossible).
The problem with the "it's just next token prediction that went through a branch whose tokens spell an untrue sentence" is that every problem, thought and statement can be modeled like that, and it's not very helpful.
It's like an alien asking why JFK was assassinated, and another alien says, "Well, it was because human minds are a network of neurons responding to stimuli which most of the time looks like the human looking for food, sex, attending sporting events, but and in that particular day, one particular human's mind responded to stimuli making him acquire a gun and pressing the trigger to send a bullet in the trajectory of the US president's head. There was a chance that the neurons would have made him attend a sporting event instead! but alas such is the stochastic nature of the human brain."
Not entirely. As @Bernie points out below, computer programs have a persistent state, which makes them produce the same output given the same input. You can think of this state as a very crude model of the world, if the "world" is constrained to e.g. the string "strawberry" plus the ASCII character set. Humans have a much more sophisticated model of the world, and while it is stochastic, it is still significantly more robust than Claude's. To put it another way, a human can of course make a mistake when providing a citation, but his output is a lot more constrained as compared to Claude's, though less constrained as compared to a Python script (or a conventional search engine).
At last! Something akin to a stack trace.
This is a great summary of what's going on inside the black box. What do you think happened when I asked my model (Claude Opus) to generate a summary table of evidence and the output was a table including correct data from a paper, but then an incorrect citation of that same paper from which it previously extracted correct data? Not that it's doing this all the time, but maybe 1 in 10 or 1 in 15 times. Most of the time, it's correct in all aspects of the citation. (Indeed, if it gets the PMID right, it's going to get the first author, year, and title correct as well. I've never seen it get one but not the other 3, so it seems like an all-or-nothing hallucination.)
Well, another overlooked aspect of the agent harness I mentioned is the coding runtime that it uses all the time even without you noticing. Mostly python and html. If you ask Claude for a table, deck, or presentation, it will most likely first write Python code behind the scenes to do the work and, with that, render HTML to show you a table with information. A very popular example of the failure of tokenization that was cited a lot in the past, –before agents came about– was asking LLMs how many Rs were in the word "strawberry". And it'd always get it wrong because the LLM wouldn't see the string input like that, it would instead tokenize the string "strawberry" into ['496', '675', '15717'].
Now, the agent harness with its python runtime just writes code to parse your raw input string and feed it to a script. It writes and runs on the fly like this.
txt = "strawberry"
result = txt.count("r")
print(result) # result: 3
but, still it can make mistakes rendering the response when piping results from tool to tool in the pipeline! Notice in my previous response how it renders the final answer:
> The 13th president of the United States was :contentReference
so it might very well fail to render the reference to the source, render to an incorrect variable or just forget to render it at all.
if you want share your conversation and we can take a look.
Those kinds of hallucinations don't happen now, it's not 2023 anymore. Your specific example should be transparent if you read the chain of thought and it's very likely not what is going on.
What people have to understand is that nowadays you very rarely interact directly with the LLM model itself from your web chat interface. What we have now is agent harnesses. From Chain of Thought, the model knows something is or isn't in its training corpus and it's prepared to respond with that. What most probably happens when you get a wrong factual answer is that the agent did a web search that found wrong information. It doesn't give you the source, but if you ask again for it, it will do a new search and find a better one. Another avenue for error could be tokenization. When you ask for the 13th president that "13th" token not translate perfectly to the text string the agent uses as input for its web search tool. So yeah, the error surface is there, but I don't think it matches to your model of AI hallucination.
That’s a bold claim. AI definitely still hallucinates even if it’s not in the exact example described.
emphasis on *Those* kind of hallucinations.
when LLM were just predicting the next token for:
"the 13th president of the USA was _______" there were high chances of getting a false data point expressed with confidence. It's so much less likely now that the result is next token prediction of successive statements like:
> "I have to figure out who the 13th president of the USA was"
> "I could do a web search for that information"
> Wikipedia says it was "Millard Fillmore"
> I feel confident that is the answer
The 13th president of the USA was _Millard Fillmore_
So, two additional (incompatible) hypotheses:
The LLM doesn't "know" things. Statements in the training data make it more likely to say something, but aren't an internal databank of knowledge that it basically has access to at all times. It didn't mistakenly say that JQA was the 13th president. Rather, it referenced JQA as a president very likely to come up in conversations like these. The labs have gotten heroically good at training LLMs to be correct, but there's not necessarily an internal truth value on all these statements.
More holistically, you know when you have something you want to say, and you can't quite think of the right word, or you know quite well the list of presidents because you had to learn it in school, but it's just escaping you at the moment? Your question sounds like one you could ask of a former civics teacher who would say she really knows all the presidents, and could in most situations just rattle them off, but for whatever reason couldn't remember who was 13th in this particular conversation, and filled in a president she did remember, though immediately knowing it was wrong. This seems like a thing that happens all the time to humans, too!
I'm not sure it's true that LLMs don't have some level of base knowledge. There was that paper a few years back on approaching monosemanticity (can't remember the name, but you can find it pretty easily), where they demonstrated there is some base-level knowledge build right into the model weights. But my sense was that this is built up from lots of exposure to the same ideas, not like memorizing everything it reads. Note that the president analogy was just that - the hallucination I'm reporting wasn't about JQA, nor was this a question I posed to Claude.
Even so, what I'm observing doesn't make much sense. Imagine that same high school teacher went on to give you an actual, real summary of Millard Fillmore, complete with accurate dates of birth/death and quotes from important speeches, but then said, "oh sorry I got the name wrong". That's maybe a closer analogy to what I observed, where an LLM clearly searched out and read the correct publication, cited numbers it pulled from that same publication, accurately analyzed the biochemistry and clinical/statistical implications of the publication, but then hallucinated the citation.
I really don't think there's a direct human analog to what's happening here, and I especially don't think the analog is "shameless guesses, not hallucinations".
If it’s making up citations, it doesn’t have any understanding of what a citation is. It seems that is enough to realize there’s no “intelligence” there.
It’s odd there isn’t a rule to prevent this nonsense.
(Or, it’s just prone to lie. Which is another problem.)
I used to wholeheartedly agree with this take. The more I use these models to read/analyze long technical documents and get novel takes on complex design problems, the more I have to ask myself, "What really is intelligence, then?"
I suspect we're rapidly approaching a point where we will either have to admit that AI has some form of "true" intelligence, even if it does function differently from our own at times, or we will have to define intelligence so narrowly that few humans will make the cut as intelligent agents.
It’s failing to understand citations (not that sophisticated of a concept) at a very basic level. Why ever would you trust it with anything more complicated?
This isn’t an issue of “narrowly defining” intelligence.
>> “ I'm not sure it's true that LLMs don't have some level of base knowledge.”
It can’t even get citations right.
I don't think you'll be happy to call this "intelligence" once they fix the problem where it gets citations wrong ~5-10% of the time. I think there's a more fundamental disagreement you have with the word usage and what it implies about the nature of the thing itself.
Humans have always made different types of mistakes than machines. You don't have to call LLMs "equivalent" or "superior" to human intelligence. Maybe you don't like the word "intelligence", because it sounds like it's doing something similar to what your brain is doing and that feels like it's bringing baggage into the discussion that doesn't need to be there.
Fine. But you have to call it *something*. To me, I gave it a complex prompt and it did in 5 minutes what it would have taken me a day or more to do. I spent 20 minutes reviewing the output and found/fixed a few minor errors. When I substitute a day of my own personal *intelligence* with a machine that could substitute 98% of that effort, we're going to have to call that intelligence-substitution machine something, even if sometimes it doesn't look like what normally define as "intelligence" in the form of getting the simple things wrong. (My wife would disagree about that, and claim I get easy things wrong all the time.)
But if you're not going to call it intelligence, you're going to have to find a better name quick, because whatever it is, I can use it to massively substitute for large amounts of intellectual effort from highly-trained subject matter experts, who only have to go in and correct minor mistakes (or ask the tool to make the corrections once they're found - or sometimes ask it to find its own errors, which it does!). That's not a theoretical concept. That's actual usage of the tool last week.
The old citation problem still exists. It should have been fixed long ago.
Anything with basic intelligence wouldn’t make that mistake.
I regret to inform you that you have not worked with a wide enough range of humans
As sclmlw said:
>To me, I gave it a complex prompt and it did in 5 minutes what it would have taken me a day or more to do. I spent 20 minutes reviewing the output and found/fixed a few minor errors.
I don't think that treating the problem with citations as if it were a slam-dunk disqualification from describing LLMs as displaying many aspects of intelligence is a good choice. As LLM-based AI systems stand this month, their intelligence is spiky (using human intelligence as the norm for comparison). To my mind the interesting question is when the various holes between the spikes get filled in, and with how much generality, and in how many steps.
Is it possible that a non zero temperature is forcing it to pick the wrong thing?
Saw a really clear example of this a few days ago. I was googling for museum ship attendance figures and the Google AI summary described Midway as an "Iowa-class carrier". That's obviously incorrect (the Iowas are battleships, and Midway only shares engines with them) and it's the sort of thing that no person with more than vague knowledge of how WWII warships work would ever say. Maybe GPT-2 would have made that kind of mistake frequently, but probably nothing since then.
The best explanation I can think of is that it internally treats different portions of the output with different degrees of importance, and spends more focus on the ones that it thinks are important. Table of values? Important. References? Not important, just say whatever. The problem is that it doesn't always align with what we think is important, and the way it mixes high-effort and low-effort parts seems really alien to us. Obviously, people give ill-considered, off-the-cuff answers all the time. ("I guessed JQA as the 13th president because I knew he wasn't super-early and before Lincoln.) But it's usually in contexts where it's obvious that's what they're doing, and they rarely give you a meticulously researched first half and a second half that's completely off the top of their head.
That's not really how LLMs work, they're not assigning importance to different pieces of the output and changing effort spent. What's happening is that the Google AI summary is from a tiny model built with very little brainpower (or parameters) to make it cheap enough to run for every Google search. It's not able to perfectly recall everything, and so it wrote something plausible sounding but wrong.
Do you know if it's switching to using web search after you challenge it? My guess is that there's a slight penalty on using search (so the AI doesn't waste your time searching for everything), but it will do it if it has to.
I'm using Opus 4.7 "Adaptive". I don't know if it's doing a sequential search, then answer. It seems something like that is happening, as opposed to it going back and forth between search -> compose answer -> search -> compose answer. The original prompt included an ask for the specific references, so it should have been looking for that information in the initial search. Indeed, it built the response from the references themselves (including accurate numbers from the papers it subsequently failed to cite), so it's really weird that it then went on to make stuff up that it had clearly found in initial search.
I have a process where I specifically direct Claude to support its answers with websearch to enforce external sources and not just info from the training set.
I often see claims that AI has been making up academic references, and I wonder if it is copying these from rogue papers in which a human author has made them up! If it happens again, you should try asking it directly where it obtained the reference, although I wouldn't be surprised if it can't or won't give a straight answer.
That's an interesting hypothesis, but in one case I'm thinking of I know this isn't what happened. The reference it gave me was to a 1979 paper (a real paper, to be sure, but on a completely different topic not even tangentially related to the one I asked it to look up). The context was in a table, listing studies, findings (including sensitivity and specificity numbers that WERE correct for the citation it meant to give me!), along with references, most of which were correct, but when I went to check the references this one was clearly just the wrong reference. It wasn't even "off by one" number or something like that. It was just wrong all the way around (wrong PMID, author name didn't go to the PMID, year was wrong). But it DID have the right numbers in the table it generated, suggesting it knew of the correct study since it pulled the data out of that study. It just ... for lack of a better term, hallucinated that this was all from some other made-up study!
I think the LLM is just generating text that looks like it would form a valid reference. LLMs are great at syntax, not so much at semantics.
Oh, I'm sure it totally made up the reference. But the syntax/semantics distinction here isn't where the problem lay. It gave me a summary of the correct paper, complete with accurate numbers (I double-checked once I found the correct reference), but then tied that to the wrong citation. All while demonstrating a solid understanding of the complex clinical context the study was working from. Seems like it understands the semantics really well, but that citations are some other beast entirely that are for some reason more difficult to get right.
It's because citations are really short strings of text whose only purpose is to point to some document. It has no context and is ultimately arbitrary, so the probability distribution of possible citations looks pretty flat. Articles are really long strings of text where all the tokens are tightly interconnected, so it's easier for the LLM to tie them together.
In other words, exactly like it would be for me. If I don't know the authors personally, it's going to be more difficult for me to remember the citation because that's not what ultimately matters.
It's common in these parts to downplay the stochastic element of an LLM (because I guess it's more interesting, albeit less insightful and accurate, to not think of it that way), but I think it's probably related to a bad initial seed or stochastic parameters.
A bad initial seed could lead to a bad path where the correct answer isn't locally present, so it does a shameless guess. Trying again could give a better seed that leads it down the right answer basin and no guess is needed.
LLMs typically do not always choose the single most likely next token anyway; there are parameters like temperature and top p (hidden to most users) which allow for somewhat less likely tokens.
That all makes sense. What doesn't make sense is that I've seen a few instances now where it shouldn't be having to "guess", since it's pulling numbers and even direct quotations from the source itself.
Now, I get what you're saying that the "stochastic" nature of AI is that even those direct quotes were also guesses (even if each part of the quote's probability weight was > .99) about what the expected output should have been. What I don't understand is that if it can give me a direct quote from a paper with a high probability of success (because it made several accurate direct citations from that paper) how can it then go on to make several errors in a row (wrong author, wrong year, wrong PMID - both for the wrongly cited study and for the expected/correct study)?
How does the probability switch from almost universally accurate to almost universally wrong? Random chance doesn't seem to explain it. By analogy, if I flipped a coin 20 times in a row and it always came up heads, I'd think there's a very high probability it's a rigged coin, but there's still some chance I could observe those results randomly, mistaking noise for signal. But if I flip the same coin another 20 times in a row and get tails each time, I have no explanation for what's going on.
I think the coin analogy is telling; the tosses are independent, but the "path" the LLM goes down is not. If the LLM goes down a "just read from the PDF and don't rely on any training data" path, then it's going to execute that task and doesn't have to guess anything (okay it's still guessing, but with very high probabilities). But if it does down a "rely on training data" path, then it's going to "guess" if it doesn't have that specific data built in. It might be useful to think of the LLM as randomly going into a different mode of information sourcing. Conceivably this could be addressed by prompting it in such a way that encourages the former and discourages the latter, but that would have to be tested of course.
Here's the actual prompt I used [two minor details censored from ACX, sorry], with Claude Opus 4.7 Adaptive:
"Interesting. You started by challenging my prior that [XXXX] would help my co-primary endpoints, asserting instead that this may be the opposite of what is supported by the literature. Please put together an analysis of all current [XXXX] publications that reference [XXXX], assessing any measured differences. Treat sensitivity and specificity separately, with separate analyses of each. Provide references (including PMID) for each claim."
The output was a well-reasoned analysis of the literature and included summaries of the various publications (including a correct summary of the incorrectly cited publication). It also included a table with columns for Study/PMID/N/>1cm sensitivity/<1cm sensitivity/{delta}/Assessment. (Same for specificity.)
Note that most of the data in the table was accurate. It's truly impressive what the model can do well! If I had an aide who could put something like this together in a day - including if they made a few reference errors - they'd be a valuable addition to the team. Yet Claude did it all in a couple of minutes.
Partly, I'm trying to understand the way this works so I can get better/more accurate results in the future. Partly, I'm reading the various explanations for why it functions the way it does and asking if they make sense, based on what I'm seeing.
I get your take, that maybe it's focusing on the wrong thing in looking up and reading the PDFs, but then forgetting the references it just looked up. But I did ask for specific references to each claim, including PMID. It didn't hallucinate the analysis, just the details. Which is something I keep seeing. It rarely hallucinates crucial details necessary for understanding an idea, but it sometimes mixes up references, even when I ask it to focus on them.
My suggestion would be to do this in Claude Code or cowork and create a hook to automatically send the output to an agent (using haiku should be fine) to cross-check all references and report any anomalies.
I second this. The only way i've been able to get Claude to really use files/external data stores is in CC with markdown file structure and routing/index information in claude.md and elsewhere
I'm sure this has been addressed before, but my keyword searches aren't yielding results.
Of the Good, the True, and the Beautiful, it's pretty clear there are rationalist applications for the Good (EA for example) and the True (lesswrong for example). Is there a rationalist application of Beauty?
According to Keats beauty is the thing that lets you recognise the good and the true .. Ode to a Grecian Urn.
Check out David Deutsch's argument for Objective Beauty in one of the last chapters of The Beginning of Infinity. He puts forward the idea that Objective Beauty provides the clearest attractor for effective cross-species communication such as the communication between flowers and insects.
Read the Fun Theory Sequences. As long as Beauty can be specified in your utility function, and you are good enough at rationality to not screw things up by trying to apply it, rationality is the best way to achieve it.
Not really rationalist but https://people.idsia.ch/~juergen/beauty.html
Many folks seem to conflate symmetry and beauty, so you could consider various sites dealing with math.
That's better than my idea that physical beauty is correlated with reproductive success, therefore SSCers should be looksmaxxing lol.
Try not pass out at the club though.
“After Apparent Overdose, Clavicular Is Back at a Club”
https://dnyuz.com/2026/04/16/after-apparent-overdose-clavicular-is-back-at-a-club/
“I assume they want cool people for this and therefore none of my blog readers qualify”
I’ve been to the meet-ups. We do not qualify as cool, I’m afraid.
You’d be surprised as to who reads the blog! I’m a young female entertainment attorney, and I’m an avid ACX reader.
What exactly is an "entertainment attorney"? What sort of clients do you represent?
I primarily handle music artists (everything from super-stars to up-and-coming talent), music producers, and influencers. Whenever a song is released, there's a lot of legal groundwork involved (i.e. drafting and negotiating producer agreements, writer agreements, and side artist agreements). I also handle the broader transactional side: recording agreements, publishing deals, and brand partnership agreements.
As a professional in the field, what is your stance on the current state of copyright law? I tend to think it benefits publishers at the expense of artists and has a chilling effect on creativity, but I've never tried to receive financial compensation for creative work so my perspective might be kind of warped.
speak for yourself. im what you call a “crypto-normie”: i codeswitch to match the vibe of the ACX meetup but put me in the ai safety hype house and i can go toe to toe with the other neurotypical zoomers
I didn’t have my 20-sided die and I thought they were going to kick me out!
Yep, D&D players have a similar vibe.
Yesterday's London Marathon was one of the most astonishing athletic feats in history, comparable to Roger Bannister's breaking of the four-minute-mile in 1954. Sebastian Sawe covered 26.2 miles in 1:59:30, smashing the previous record by 65 seconds. Yomif Kejelcha, a very talented distance runner making his marathon debut, ran 1:59:41. Jacob Kiplimo finished third, also breaking the previous world record. Sawe and Kejelcha are the first men ever to break two hours for the marathon in an official, record-eligible race (Eliud Kipchoge managed the feat a few years ago in an exhibition).
I realize I'm much more of a running nerd than the vast majority of people, but I find it slightly disappointing this isn't the top news on every single sports-related website. And it should be front-page news around the world. It really is an incredible achievement.
Uhm, in my news bubble this was the case: it was on the front pages of sport-related websites and it was prominently placed on the general news websites.
Of course, mostly focusing on the winner, but the articles also mentioned the tragic story of the second-placed runner.
That is pretty amazing. I’m sure the length will be checked. I don’t think a tail wind could help that much. Probably want to test the record breaker for PEDs. Really good PEDs.
What's interesting is that Sawe agreed to participate in extra tests (with the costs paid by Adidas) in the lead-up to the Berlin Marathon last year. Presumably you wouldn't agree to that if you were doping (unless you were extremely confident that you were using drugs that were totally undetectable).
There's always skepticism around these sorts of amazing performances, but if I had to guess one way or the other, I think he's clean.
Yeah, I ran into it and it seemed incredible.
How did three separate people manage to break it at once? Was there an unusual tailwind or did they accidentally make it slightly shorter or something?
The performance of top athletes in the same long-distance race are highly correlated to each other.
First of all, it is clear upfront for many important Marathon races over the year that they will not yield records. The race tracks differ, most obviously not all of them are flat. So if several runners improve a lot over one year, they may all have their personal record in the same race.
Second, there was a very strong pacemaker field in the first half (or two-thirds?) of the run. For the final performance, it matters a lot how well the pacemaking works.
Third, tactics: usually top runners try to stay in the front group for as long as possible. It's not like they are running the race at independent paces. If the front group happens to be fast, then all top runners who manage to keep the pace will have very good times, otherwise they will all have slow racing time.
If the course were short, we'd probably know about it by now.
The shoes are probably the biggest reason. Adidas (worn by the top two) came out with some extremely light shoes. Kiplimo was wearing Nikes. The biggest change was a few years ago, when the shoe companies, starting with Nike, came out with so-called "super shoes" featuring carbon fiber plates and very cushiony foams. Gradually the super shoes have gotten better and better.
The conditions were cool enough to be good, and the wind was somewhat beneficial. Record-eligible marathons must have a finishing point within 13.1 miles (as the crow flies) of the starting point. So the Boston Marathon, which was also very fast this year, is not eligible because it's more or less a straight line and some years see a very strong tailwind. But the London course this year had a net tailwind, and a tailwind at the end of the race. So that explains some of it.
I just read an interview with one of his coaches. He agreed that the new generation of shoes have notable impact, and that weather/course wind all need to play out well.
But he added an interesting other aspect that he called a game changer in the last years: that there was a strong shift from lipometabolism to carbohydrates even on such long tracks in the last years. Essentially people realized that if with proper training, the digestive system can digest much more carbohydrates than previously thought, even during a run. Sawe had started days ago drinking special carbodrinks to fill his depot, and continued to drink them before the run, and during the run starting from km 5.
I've definitely read about this as being a factor, but I don't know if there were any new changes to fueling strategies this year (the stereotype of eating a big pasta dinner the night before a race has been around for a while). From my understanding, the body can't effectively store enough glycogen to run a marathon. And while the body's fat supply is enormous, at higher intensity levels the body would much rather burn carbs. So you need to take in a lot of carbs during the race. But that's not totally new, so I don't know if Sawe and the others were doing anything new this year.
I've also read about carbs being a huge factor in improved cycling performances in recent years. So fueling is definitely very important, but I don't know if it was the major reason for yesterday's breakthrough.
Yes, it's not a big change in this year in particular. I rather think about it as a general improvement that makes all runners faster. If a new technique makes all runners 3 minutes faster in the last 3-5 years, then the 2:00:00 is not necessarily broken in the first year of this technique, since marathon times vary a lot from race to race.
I think this is not such a bad model. Before 2022, there was exactly one runner who ever beat 2:02:30, in 2019. But 2022-2025, the time has been beaten by 5 different runners, and the five times were actually pretty evenly distributed in the interval 2:00:30 - 2:02:30, without any particular order. Given this spread, it seemed just a matter of time until the 2:00 was beaten.
Of course, it is remarkable that two runners beat it in the same race, but then the times are correlated.
Best theory is that the newest supershoe that weights only 97 grams, which is almost half of the previous leading supershoes, combined with a competitive race until the last kilometer and good weather with a tailwind for the last 10 kilometers all contributed.
London isn’t even supposed to be the fastest major marathon, it has some elevation and sharper turns, so people are expecting this record to be broken again in the fall at the flatter Berlin or Chicago courses if the weather is cool.
I'd expect Sawe to get an absolutely massive appearance fee from Berlin this fall. He ran 2:02:16 last year in Berlin despite 70F temperature.
The news from Byzantium.
Another Subject to Change podcast episode.
Justinian I (dies in 565 AD) is generally considered to be a really super emperor. He ruled the Eastern half of the empire some years after the Western half had undergone a rapid unscheduled disassembly. Justinian is pretty successful putting it all back together again . He manages to get back ‘Africa’ and Italy and has time left over to build the magnificent Hagia Sophia which you can see to this day. And he has a scandalous biography written about him. For example we are told that before becoming his wife, Theodora had a nifty trick to entertain paying customers by having geese peck barley grains from her naked body.
Anyway we are not talking about that Justinian. We are talking about Justinian II who comes to the throne over a century later in 685, at the age of sixteen. This is a time when the Arabs are in full expansion mode so the episode starts with six year old Justinian peeping over the walls of Constantinople at a vast Arab army not unlike the scene at Helms Deep. Anyway the people of Constantinople have a secret weapon - this is Greek Fire which is a kind of napalm you can shoot out of a tube and burn up the enemy fleet. Which they do, so eventually the Arabs have to give up and go home. (But they’ll be back. Repeatedly).
When Justinian II takes over a lot of the old empire of Justinian I is gone. Even Greece isn’t safe these days having been largely taken over by Slavs. (Not to be confused with the Bulgars who take over other bits of the Balkans).
Anyway our Justinian reigns with the massive over confidence of a teenager. It is not fair to say that everything he does goes wrong though kidnapping the Pope goes less well than he’d hoped. Actually he does pretty well for a bit but some shocking misjudgements see him in dragged to the Hippodrome where a man with a pair of pliers sets about cutting out his tongue and cutting off his nose.
Then in a misjudgement every bit as big as anything Justinian did, instead of putting him in a sack and throwing him in the Bosphorus they exile him to the Crimea. I mean everyone knows an emperor has to be bodily intact so there is no chance he is coming back is there? Is there??
There is. Justinian spends his time in the Crimea weaselling his way into the good books of the Khazars who are sort of Turkic nomads. He even goes as far as to marry the Khagan’s sister. When Byzantine bribery causes his brother in law to turn murderous he murders his would be murderers and escapes to the Bulgars and teams up with them. So with his new Bulgar army (following him largely out of curiosity) it is off to Constantinople to wreak vengeance on his foes. The only problem being that when they get there nobody wants to let him in (not least because of all these ferocious Bulgars looking on with interest).
Anyway for those of you who don’t have too many podcasts already just google Subject to Change with Russell Hogg or it is right here.
https://podcasts.apple.com/gb/podcast/subject-to-change/id1436447503?i=1000762414132
I did rather wonder how he survives his nose being cut off. I thought people in the past died of the merest scratch though surely that can’t be right? But Justinian II isn’t the only emperor to be mutilated like this and they all seem to more or less thrive on it. Also in the novel by Turteltaub which we discuss on the podcast, Justinian has a new nose (sort of) given to him by a visiting Indian surgeon. There is nothing in the sources to say this happened to Justinian but supposedly these kind of techniques were practiced in India at the time.
(And full credit to Professor David Parnell who is not just an excellent guest but game enough to participate in a political correctness quiz to round off episode 2.)
I wanted to know more about why you didn't like Harry Turtledove. I read one of his books and thought it was pretty good, maybe they're hit and miss.
I only read a couple apart from Justinian. From memory The Gryphon’s Skull and Over the Wine Dark Sea. I am probably being unfair and maybe when he writes as Turtledove he is better. And he is so prolific I suspect there must be quite a variation in quality. Which of his do you think are the best? I feel guilty about being so rude so I should give them a try.
I enjoyed Household Gods, it's focused more on everyday domestic life in the Roman empire, but also includes Marcus Aurelias and the Marcomannic wars.
I will give it a go! I have a few favourite 'ancient world' novels but generally I think they are hard to pull off. I liked A God Strolling in the Cool of the Evening. Elodie Harper's Wolf Den trilogy was pretty good and Philip Matyszak's Gold of Tolosa is great fun.
Just want to say I have listened to a few episodes of your podcast and I enjoyed them. I think having some economic historian on might be a good balance for the narrative history you normally do.
Thank you for listening. Do you have an idea who might be good and what topic? I'm very open to ideas here! I did have David Friedman on ages ago and he was great but I am guessing this isn't the kind of thing you mean. Legal Systems Different from Our Own which Scott had fun with.
Oh I really liked the episode on the history of time keeping!
Joseph Henrich writes about the type of stuff I find fascinating. I have not read her stuff but Ada Palmer had a good podcast with Dwarkesh. So maybe people that they cite / cite them.
It was interesting. Your avatar didn’t prepare me for the David Attenborough voice though.
He’s a Japanophile for sure, that is signaled by the avatar :-)
I am off to Japan for a month on Sunday. And the avatar is from the excellent anime, Stein’s Gate.
Sean Connery voice surely!!
“Some who have been in Justinian's company in the palace very late at night, men with a clear conscience, have thought that in his place they have beheld a strange and devilish form. One of them said that Justinian suddenly arose from his royal throne and walked about (although, indeed, he never could sit still for long), and that at that moment his head disappeared, while the rest of his body still seemed to move to and fro. The man who beheld this stood trembling and troubled in mind, not knowing how to believe his eyes. Afterwards the head joined the body again, and united itself to the parts from which it had so strangely been severed.” — Procopius
Yeah, sure, Justinian was a lizard from outer space wearing a human suit.
If you’re poor and disabled, it probably sounds a lot better than if you’re rich and healthy.
Is entertainment really the point of life? Have we fallen that far? Why can't we just use technology to provide material abundance, cure pathology, learn/discover/develop, and have good relationships? If that's not satisfying enough, maybe fix the anhedonia rather than lean harder into the stimulus.
There's a case to be made that, in the end, it all comes down to human happiness. The catch is that happiness turns out to have diverse requirements. Many of them are outside of hedonism.
I think if an indistiguishable simultion existed, most people would choose to live in it if feasible. A lot more than would admit to it in a world where it doesn't exist.
VR would allow things that are impossible, too dangerous, or very expensive in reality. That sounds exciting. But I would also like to stay in contact with the top (?) reality.
(When I talk about expensive, of course have I no idea what the actual costs will be and what will the economy be like those days, but I kinda assume that VR will have more or less the same costs regardless of *what* it shows... so sitting in a virtual office or flying through space will cost approximately the same, so there is no reason to not do the latter.)
Well the issue is I have a child I want to keep contact with - and in the physical world. Otherwise if my body is kept healthy then yes. But I might still want to get out and "touch grass" sometimes.
>If fully realistic simulations are ever developed, would anyone here prefer to live in virtual reality?
I'm not following what would be the implications for my physical body, and how would my physical body interact with the VR?
If it is e.g. just VR glasses driving e.g. a more convenient version of tinkerCAD - it has its uses, but I can't see spending a lot of time on it.
If it is a full upload, with options to configure senses as I saw fit, that is much more appealing.
If I absolutely have to I might commute to real life, but I'm definitely living in VR.
I would 100% choose the matrix.
It's a minor point in Blindsight by Peter Watts. For unrelated reasons, you may also be interested in Ra by qntm (https://qntm.org/ra).
Personally, I would avoid it unless there is some health/safety reason to live in VR instead of IRL. I'm paranoid about any tech that could harm me if programmed wrong (brain implants) or that leaves my physical body defenseless for long periods of time.
I think there are a lot of elderly people for whom virtual reality would be a huge improvement over their current best available activity (sitting and watching tv all day).
It would get just as boring as any other game, and still has all the constraints videogames do now, in that you have to be able to pay the electric bill and feed yourself.
Also I think realism is a turnoff.
Also I think VR is a turnoff.
No, games do not come tailored to personal preferences. Source: I have like 100 Skyrim mods installed, and 20 of those from loverslab... it is kind of how I would like to live, but still a very basic-level simulation of that. I think the proposal was basically perfectly tailored, perfectly immersive simulation. Like if you want to be the king of England in 1500, your experience is indistinguishable from as if you would be.
I made my main argument below*; the charm has a ceiling, because you're going to know it's fake, and designed to let you win. Is your perfect simulation of the King of England going to allow you to be overthrown by a peasant army because your policies were too severe? Probably not, which means your experience is going to be hollow.
*(I assume it's still there, they seem to have blocked me.)
Don't we have an answer to that from current videogames? You (after some practice) want to play Europa Universalis IV in Ironman mode, because only that gives you achievements. Everything else is basically training wheels. Ironman has no saves in the usual sense - if you lose, you lose. However, you can still start it again from the beginning.
And these achievements matter! I am not an avid gamer by any means, but I am kinda proud I have the "unite Italy" achievement.
Ironman Mode is still designed to be beaten. The charm is still going to wear out at some point.
If you have Matrix-like technology you should also be able to entirely eliminate boredom or any concerns over "this isn't actually real". Nanobots will rewire your neural circuits and you'll be a happy camper.
That doesn't need virtual reality though, that's just a lobotomy.
Lobotomy is the dull version of this. The VR version would be able to keep 100% of your personality, except that you never feel bored. You never see this scenario in sci-fi because it's extremely boring (ha) to readers, but that doesn't mean it won't be possible in real life. See: https://nsokolsky.substack.com/p/beware-the-science-fiction-bias-in
Would be able to, but would have no reason to. Once you're messing with people's brains there's no reason to give them what the old them wanted.
That is also a sci-fi trope, because "give people what they want" doesn't work, there always has to be some sort of a "gotcha" involved. But real life is not sci-fi: I predict that if some sort of a benevolent AI does create VR simulations for us, it would do so in a way that doesn't involve the "Be Careful What You Wish For" trope.
It would get boring faster because uploading your mind into a machine means 98% of the fun of anything is gone.
Two simulations is even worse than one. It's like playing Virtua Fighter in a Yakuza game; if you wanted to play Virtua Fighter instead of Yakuza you could just go play Virtua Fighter.
If we're assuming post-AGI technology, we might as well assume enough other benefits that escaping the real world becomes pointless. And I still hold it would have no advantage over lucid dreaming, and no meaningful advantage over existing videogames.
I would probably live in VR at least part time, sure. A multi-player one though. Not just me.
Fully realistic is weak. We should aim for much higher perceptual resolution than unmodified human senses in meatspace. The web serial titled Seek (by Wildbow, the author of Worm) explores this.
According to some, we already do !