This is basically a drug trip report, except it was all medically prescribed…
I have Graves’s disease (thyroid condition), and the endocrinologist decided to prescribe the heroically large dose of 80 mg/day carbimazole (plus 5 mg/day bisoprolol). It takes about two weeks for the drug dosage change to percolate through to a change in hormone levels, and I now have the blood test results:
T4 13.5 pmol/L
T3 5.5 pmol/L
Heart rate 80 bpm
… which is in the normal range.
So, is there a psychological effect? It is very strange. I feel like before the increased drug dosage (I was on 60 mg/day carbimazole + 7.5 mg bisoprolol) I felt a kind of fear-adjacent emotion, a pseudo-fear that felt physiologically a bit like fear without actually being fear. (What’s happening is the elevated T4 hormone levels are driving heart rate really high). It’s like the physiological response of fear without any associated cognitive content.
So anyway there it is, in the same spirit that the folks at erowid report on more exciting chemical experiences.
Today I had the pleasure of reading the following sentence on Psychotropical.com, the site of Ken Gillman, MD, MAOI expert: “Despite valiant efforts to make them more evidence-based, guidelines, recommendations and exercise of policy power unfortunately remain among the least evidence-based activities, impregnable strongholds of expert-based insolence and eminence-based innumeracy."
Even YouTube sharing half of advertising revenue with creators, in context, is less of a heartwarming story of non-greed. And more the story of a corporation maintaining monopolies that would have traditionally been regulated under 20th century antitrust norms.
Traditional antitrust enforcement targeted companies that quantitatively seized dominance in their industry and pushed out competing companies. 21st century antitrust theory is kind of faux-populist with its argument, "We don't care what kind of market share you have, as long as the customer is not being overly negatively impacted." How TOUCHING. When that conveniently abandons quantitative monitoring for a vague qualitative monitoring, based on gut judgements from Ayn Rand loving regulators. "This industry's consumers are still doing acceptably well. I holistically see that the benefits of this monopoly outweigh the downsides. In my subjective wisdom, anyone saying differently is being a negativistic whiner."
Antitrust enforcement as an art, rather than a science. How urbane and sophisticated. 🦝🐻🤔💀💀💀
To my great surprise, today I found out that there is a 2016 film adaptation of "The Man Who Was Thursday". I hadn't heard or read anything about this and seeing as how it's one of my favourite novels, I wondered was this any good.
Unfortunately, looking at the trailer, it seems to be the same old schlock about Catholicism. It is better described as "inspired by" the novel, as it takes the names and nothing else. I don't regret missing this, since it's just one more in the "supernatural horror sexy nuns" genre.
I had the same experience a few years ago, and had successfully blocked it out until you reminded me. It seems like they took the title from Thursday and attempted to write a Dan Brown movie around it except with hot lesbians?
An actual film adaptation of TMWWT sounds unlikely though, the ending is going to confuse and annoy any new audience member, and of course the whole thing is also very much inconsistent with Hollywood values.
The Department of Transport says they want to eliminate all of DC's speed traps. If they do this, this will of course greatly improve the lives of ordinary people in the DMV region, but what I like more is that you can pretty much see what precipitated this proposal: someone there got a speeding ticket, and with righteous indignation, resolved "Never again!" and decided to strike at that evil without worrying about "protocol" or whatever bullshit reason most other people make up for not doing confronting blatant injustice they have the power to immediately fix.
I suppose im not easily seeing what precipitated this. I recall speed traps being complained about all the time. I wouldnt attribute their persistence to ppl in denial of their ability to immediately fix things. Some agitators get lucky and are heard by just the right person at the right time. Some agitators are wise and plot a complex plan that works somehow. Lots of time the change has nothing to do with the specific agitators.
BUT rather than pretty much seeing it, you looked it up and shared the case study as a template for change, that would be cool?
Just have all DC speeding ticket revenue go to the federal budget. Do the same with every local government's traffic fine revenue. It won't have a noticeable effect on the budget but it will make the world a better place.
To answer your question directly. The basis for claiming otherwise is that speed variance kills more than the average level of speed. And that currently the way speed laws are enforced tends to increase variance therefore are on net bad. In particular, an optimal speed trap is a stretch of road that to the driver feels like it should be drove on faster than the marked signage ie a badly engineered road. Get rid of the speed traps. Get rid of the roads that invite them and adopt a plan for enforcement that reduces speed variance and youd be in better shape
Sounds like a call for spamming average speed enforcement cameras everywhere. Lots of UK highways are doing this now; and to be fair it does accomplish the goal of having drivers actually stick to speed limits without sudden changes of speed around the cameras.
Driving cars at *any* speed, will result in some people being killed. Either we ban driving outright, or we acknowledge that allowing these cars to be operated at this speed will result in this many deaths, and we're OK with that because the value of reasonably fast personal transportation outweighs the deaths, We have in fact decided on the latter course of action, and established at least a rough social consensus as to what the acceptable speeds are. The person saying "no, we should force people to drive slower than that because it will save lives", is doing net harm to society and should at minimum not indulge him.
It would be good if the consensus regarding acceptable driving speeds was codified in a clear and unambiguous law, i.e. a speed limit that we expect people to obey. In the United States, at least, we have for obscure but probably irreversible historical reasons decided not to do that. We have instead passed speed limits that are significantly lower than the speeds we expect people to drive at, and have empowered the police to stop and arrest literally any driver. Even drivers who are not exceeding the speed limit can be stopped and arrested by the police, on the grounds that everybody knows that everybody is supposed to drive faster than the speed limit so that one's decision to drive at the speed limit is probable cause to believe that they are a criminal trying to deny the police the ability to stop and arrest them. I wish I were making this up, but I'm not.
We at least used to trust that the police would use their discretion wisely and benevolently in this manner, stopping only the truly dangerous and/or probably criminal, and making the subsequent arrest as quick and unintrusive as possible. But we don't trust the police as much as we used to, and this sort of thing is part of the reason why. Wholly arbitrary law enforcement, and laws that can only be enforced in a wholly arbitrary way, are corrosive to public trust, and we are desperately short on public trust right now.
We would be better off with our existing consensus regarding acceptable speeds but literally no enforcement of speed limits short of reckless driving proven at trial. We aren't going to get that, but at the margin, any increase in enforcement of speed limits should be resisted and any relaxation of such enforcement welcomed. Maybe after a few decades with no limits, we would be able to reintroduce reasonable and fairly enforced speed limits.
Market approaches can only work for parties with a means to influence the market. Democratic approaches can only work for parties who get a vote. Consensus can only work for parties who get a voice.
Drivers get to decide vehicle speeds on the road. However, people wanting to drive at some speed are not the only users of the road network. Other road users' risk is an externality for the drivers, and absent some external incentive they have no reason to take it into account. Calls for speed limits to be imposed, lowered and/or actually enforced are also a signal; they are, in fact, the only input to the system that road users not in a vehicle have available to them. They are a mechanism by which drivers' personal risk/reward calculations can be made to take into account, however indirectly and imperfectly, their decisions' effect on squishier people around them.
The status quo vehicle speeds you observe are, yes, a de facto consensus - the result of a risk/reward calculation performed by the drivers, taking into account their risk preferences with respect to personal safety, and also their risk preferences with respect to police actions.
If you silence the latter, you are effectively silencing the road users who are not in charge of a vehicle, and the result is no longer "a rough social consensus" that takes into account everyone using the road. Only drivers' risk/reward preferences are now taken into account when determining acceptable speeds.
You then no longer have grounds to claim the outcome is the result of all road users' social consensus.
TLDR: do what you like on the highway, but please don't scrap speed limits on roads shared with pedestrians and/or cyclists.
When the federal 55 MPH speed limit was passed in 1973, fatal accidents per vehicle mile travelled went down a bit but returned to trend within a few years. When the federal limit was repealed in 1995 and many/most states raised their highway speed limits, there was no disernable effect on traffic fatalities.
My understanding is that both the setting and enforcement of speed limits, on the margins, has only a minor effect on how fast people actually drive. From what I gather, conventional wisdom is firmly in favor of "traffic calming" road design that induces people to slow down because driving (too) fast feels unsafe. This can be hard to retrofit onto existing roads, of course but represents a long-term opportunity when roads are being built or renovated anyway.
That said, there probably is are ways to improve how we enforce speed limits. I'd propose three things:
1. I suspect that many/most speed limits are set a bit too low, with the expectation that enforcers will exercise discretion in enforcing the laws only against people who are actually being dangerous. I expect that this underlies a lot of the opposition to speed cameras, as a speed limit set with a built-in cushion for human enforcement discretion is not well-suited towards automated enforcement. The obvious solution is to build the de-facto cushion into the law by raising the speed limit. I think some places with speed cameras do this in a de facto manner (e.g. if the posted speed limit is 35 MPH, only send tickets for people measured going more than 41 MPH), but this probably isn't well publicized.
2. The penalties for speeding are generally set assuming a very low chance of getting caught. Speed cameras make you much more likely to get caught, which could be good because more predictable penalties are a stronger deterrent, but if the same penalties are applied they're excessive at the higher likelihood of getting ticketed. I'd support widespread use of speed cameras if they were paired with lower fines (maybe 5-10% of current fines) and no license points except for particularly frequent or egregious offenders.
3. I suspect a large fraction of fatalities and serious injuries caused by speeding are caused by people who are speeding by a lot. No statistics to back this up, just intuition and anecdotes about people driving at absurdly high speeds. Go ahead and throw the book at people who are going at freeway speeds on residential side streets or going at racetrack speeds on highways.
>From what I gather, conventional wisdom is firmly in favor of "traffic calming" road design that induces people to slow down because driving (too) fast feels unsafe.
How does that work? I can imagine makeing a road that people feel like they have to drive slower on, but what you want is a road that will be *safer* at the level of speed people will choose on it. Insofar as peoples feelings are related to real risk levels, that seems difficult.
“Static hazards representative of the dynamic hazards” is a phrase sometimes used.
So you have dynamic hazards — things like children suddenly running out into the street.
And you have static hazards that don’t suddenly jump out at you, like street furniture, speed bumps etc. You see this as you approach, and slow down for them.
The idea is to put in place enough static stuff — the can always add a speed bump — to cause you to slow down appropriately for the likely sudden surprises.
(Of course, you also pay attention to things that are likely to suddenly become a hazard, like flocks of sheep, kids on the sidewalk, etc etc)
Traffic calming is most useful for roads around houses, schools, and shops with a lot of foot traffic.
If you've got a road in the middle of nowhere, the main risks of driving too fast are that you, your passengers, or the driver or passengers of another car are going to get hurt or killed in a collision. How fast you can reasonably drive is mostly constrained by sight lines, curve radius, and how separated you are from cars going the other direction. Most drivers tend to be pretty good about calibrating how fast they go to how fast the road is designed to accommodate around these sorts of constraints.
Take a road in the middle of nowhere where it's reasonably safe to drive 60 MPH and drop it into a residential area with houses and parks and stuff alongside it, and people will still want to drive 60 MPH on it. But you, as an urban planner, want people to drive more like 30 MPH because of risks of vehicle vs pedestrian accidents, so you put a 25 MPH speed limit (instead of the 55 MPH limit you would have put on the road in the middle of nowhere) there in hopes that people will only speed a little bit. The numbers I've seen most often are that lowering speed limits by 5 MPH lowers traffic speeds by 1-2 MPH, so your 25 MPH speed limit will lead to people driving more like 40 MPH. A big improvement, but not as much as you'd hoped. Also, you have people annoyed by a speed limit that feels unreasonably low even though it was set for good reason.
You get much better results by changing the road so people have to slow down to keep themselves safe, which also keeps the pedestrians safe. There are a bunch of things you can do to accomplish this. You can lay out the street so it curves more than necessary. You can put small roundabouts or four-way stops at intersections instead of traffic lights or two-way stops that prioritize traffic on the big road. You can make the road narrower. You can add speed bumps. And so on.
There's also some techniques that are getting more recognition and traction in recent years where you can make the road feel more dangerous to drive fast on without actually being more dangerous. I mentioned roundabouts before, which turn out to be an example: roundabouts are statistically safer than traditional intersections with stop signs or traffic lights, and have more throughput, but they feel more dangerous so people slow down around them. Another is to lay out roads with wide shoulders (good for safety, since this is a buffer between cars on one hand and pedestrians and building on the other hand) but add "chokers' (areas where a curb-height island blocks part of the shoulder) periodically so the street doesn't feel continuously wide. There's also stuff you can do with visual cues like how the road is painted to make it feel narrower than it actually is or it feels like you're going faster than you actually are.
The California Department of Transportation has a big guide about traffic calming techniques if you're interested in reading more:
The strategy is basically to make the road sinuous, bumpy, narrow, and obstacle-ridden. It's not just a feeling, they really DO have to drive slower to be safe. If you've heard of "risk compensation," where increasing the safety of some particular thing makes people do it more recklessly, and then you end up with more deaths/injuries, this is the same thing, but in reverse.
I do know of risk compensation, but you dont generally end up with *more* injuries/deaths from it. Most of the cases Ive heard of involve a close-to-constant level of risk, and a declining (but not in proportion to the "technical" improvement) risk is similarly possible.
That was the point: You can easily make people drive slower on a road, by making the road more dangerous - you need the slowdown to be overproportional to the danger increase, otherwise you havent reduced accidents, youve just made people drive slower at the same overall risk level.
It's true that lowering all speed limits to 15 mph including on highways would have fewer people die in traffic accidents. I do not believe that's a good tradeoff, and if there are bad laws, the enforcement of those laws is also bad.
But even specifically about the government of DC, I know from direct personal experience that it can and does set absurdly low limits, including on a particular broad four-lane straight and level stretch of road with no traffic lights or crossings, where the speed limit drops sharply just before the camera for no discernable reason, and then is raised back to a more natural speed a little farther ahead. A system that allows for such an abomination ought to be razed to the ground.
Yep fair enough. But here again we run into the classic American problem of overcorrecting for something by going too far in the other direction instead of finding a reasonable compromise.
The reasonable compromise is "have sensible speed limits and enforce them" instead of "have silly speed limits which exist only for revenue raising" versus "never enforce anything".
> Speeding is bad, actually. Enforcement of speeding laws are good, actually. Speed kills.
Okay, but do you know what *actually* kills? Homicide. And assault, and rape. Which have 40% / 50% / 33% clearance rates, respectively. As in, most people get away with them.
Do you know what cops spend ~80% of police hours on? Traffic stops. Pulling over soccer moms and working stiffs to revenue farm them.
I mean, I'm sure it makes sense to the cops - wouldn't you rather spend most of your time in air conditioned cars interacting with nice people with jobs and educations, versus trying to track down murderers and rapists, which undoubtedly has less congenial working conditions?
*Everyone* drives over the limit, including cops. If you're one of the rare people who don't, I'm sure you've been able to observe that you're by far the minority here. The entire system is dumb and literally set up for revenue farming.
Sure, there are some actually dangerous drivers out there, and they should be stopped and ticketed. That’s MAYBE 10%, probably less, of current traffic tickets, and can be accomplished with 10%-15% of the hours currently dedicated to it.
For all the rest, you *really* think we should be prioritizing giving rush hour commuters and busy moms traffic tickets over investigating and closing rapes, assaults, murders, thefts, and burglaries?? Because that’s what’s happening today!
Your post is not even making an argument for that 80% of police time is spent on traffic stops? Your assumptions show that ~65-70% is spent on admin, paperwork and commuting and then ~10-15% on traffic stops.
Your post centers on "reasonable assumptions" which appear to be based solely on your own impressions.
In that spirit -- I am related to one big-city cop and longtime friends with another (different cities); sent them your "reasonable assumptions" of how police officers spend their working time. Quickly got back LOLs from each.
The officer working in the very-large city that my wife and I live in added that the percentage of time officers spend doing traffic stops has been near zero for "years". This fits with my wife's and my experience. We do get speed-trap-camera tickets now and then (one of which I decided to fight, and successfully beat), but neither of us has been actually pulled over while living in our present home i.e. 15 years.
I'm struggling actually to recall the last time I even _saw_ a traffic stop in progress except on our local expressways which are part of the interstate highway system (meaning state troopers not city cops). When I was younger it was routine to see city-streets traffic stops in progress.
And then thinking on it further, the last couple times I did get pulled over were in suburbs/exurbs not the city, and were well before any jurisdiction around here had any speed-trap cameras.
> Your post centers on "reasonable assumptions" which appear to be based solely on your own impressions.
I triangulated from literal numbers of citations both nationwide and in the New York area, those numbers are actual data.
My additional assumptions go down from those top level numbers, they aren't sui generis. And I'd be interested in what specific numbers they don't agree with - the proportion spent doing paperwork? Because 30% seems light, if anything.
I agree that the Kelly Criterion does help with the absurd Pascal's Wager type issues, but I feel it doesn't go far enough - I advocate basically ruling out any scenarios which claim to result in a sufficiently large change as a default, and requiring an exponentially strong evidence base to even consider them. I *theoretically* admit there's a chance of, say, AI turning us all into eternally suffering immortal morons, but I refuse to put any resources whatsoever into it because the chances are pathetically tiny. The odds of more mundane bad outcomes, depending on the specific, might edge into worth thinking about, but I am quite tired or rehashes of the Rapture or damnation or the Revolution, unless they come with actually plausible ways that could actually happen based in actual reality, rather than "and then a miracle/disaster happened because superintelligence"
Something causing a very large change does not necessarily imply that the chance of it happening is very low.
If you were president in the 1950s-60s, would you act as if nuclear war was impossible (and take no action to avoid it), purely because it would have a large effect if it did happen?
Are you the US President during the 1950s? No. In fact vanishingly few people are in a position to have that kind of potential influence, and it's even more rare to be in a position of that consequence and not know it.
(Your comment was orphaned but I assume you're responding to me from here [0].)
Idk, i feel like "wild outcomes require wild evidence" just folds into the probability being infinitesimal. If P(x) is epsilon, then Kelly is even lower than epsilon. If P(doom) is epsilon, Kelly is less than epsilon. This all adds up to normality, so idk why you think Kelly's Criterion needs additional pleading.
I wrote a post about Charlie Kirk’s assassination 4 months later, I’m not the best writer so any tips on writing would be appreciated as well as general opinions.
I guess I'll do a writing critique in a separate post, since I've already got the political opinion finished.
The biggest problem is here.
>The position is correct, but there’s been a hijacking of what it means to just “have a different opinion”. Charlie Kirk was the CEO of the arguably one of if not the largest conservative movements in the United States. He’s published several books, and has been a spearhead of the MAGA movement as a whole. His viral debates have fueled the opinions of many Americans social media feeds.<
You're presenting this like it was some kind of supernatural phenomenon; a meteorite hit the Earth's atmosphere and granted Charlie Kirk superhuman powers of persuasion, that he was using to do evil. No one else stood a chance against this alchemy; the Democratic machinery were powerless before him. That wasn't what was happening.
The dude made money talking because he said things people wanted someone to say. So does every pundit; so do many Youtubers. Looking at a debate recording, he comes off as an asshole. If he wins the debates, it's because the students are bad at debating. https://www.youtube.com/watch?v=IP5LDxhBod0 ("How do you justify claiming murder is immoral," I mean Jesus, Student #2.) The dude was just one of many sparks in a country full of oily rags.
Never kill people who are willing to debate. Even if they weaponize the process with skewed facts and loud slogans, through the very act of showing up they're still promoting the process of seeking mutual, peaceful resolutions. It's a message to the common folks to get together and settle things with words at their own levels. If you kill them, then the people who were willing to talk will instead shut their mouths and load their guns and write their opinions in blood.
(The piece ends by saying you agree with this. The first writing complaint I'll make is that I have not been convinced you do.)
By this logic, any dictatorship would have fallen on day 1, because they killed someone and then the masses turned against them.
(If there is any lesson, it's more like "if you are the weaker side, do not give the stronger side a pretext by killing one of them", but even then it is not obvious that a pretext was needed or couldn't be manufactured otherwise.)
"who are willing to debate". Regimes fall because they're not willing to debate, they just do what they want and ignore objections, until the objections become violent and you get a dictatorship, who reigns by violence. Once you've killed the norm, it's just force of arms. And those arms keep getting weaker, because there's no cooperation.
I appreciate the reply, reading back over I actually don't make any case for why I even disagree with it in the first place, so this criticism is entirely valid. I'll be sure to revision it. I originally intended the article to be a devils advocate kind of deal.
It's easy to overcorrect when trying to present an opinion you don't have. I remember a teacher thinking I was a militant because I kept referring to the government as "law enforcement"; I was actually doing it because I thought the word "government" had negative connotations and was avoiding it.
I've messaged you my take on the writing itself, as a commented Google Doc.
“With God as my witness, we are going to use every resource we have at the Department of Justice, Homeland Security and throughout this government to identify, disrupt, dismantle and destroy these networks and make America safe again for the American people. It will happen, and we will do it Charlie’s name.”
As opposed to *before* Charlie Kirk's death, when the government was... *checks notes* ...deploying the National Guard to cities with liberal governments.
The whole "the liberals are making us do it" schtick on the right is very tiresome. They were already doing it before Kirk's assassin gave them an excuse, and Kirk's death doesn't seem to have changed that trajectory much.
Obviously they wanted to do it from the start, but it does allow them to justify it to their base. They still need support of the military, at the very least.
The most ironic thing about Kirk's death is that he's probably been more useful to his cause dead than alive. The violence he inspires in death will accomplish far more than his debate ever did.
I did a deep dive into some stats Kirk published about the effects of liberalizing gun laws. They were published in a widely circulated pamphlet. His stats are a textbook case of how to lie with statistics: really obvious, egregious cherry picking of stats, plus a couple of math errors that exaggerate the results in the direction of supporting his argument. His argument is that liberalizing gun laws actually makes people safer, because they are better able to protect themselves from violent crime. The actual data absolutely does not support that.
I was going to write that all up and post it here, but after my own upset about Kirk and people's reactions to it faded some I decided there wasn't a lot of point to doing that.. I have done all the math analysis, though, and am happy to send it to you if it would useful.
I did not notice, but thanks for letting me know. I've revised my article to provide more reasoning to oppose mocking his assassination but I'm interested in your analysis. Feel free to send it over.
Criticizing the actions or ideas or writings of a murder victim is not at all the same as mocking his assassination or cheering for his murder. Kirk in particular was a guy whose whole career was about political ideas and arguments, and it's 100% valid to criticize those ideas and arguments. Doing it the day he was murdered would seem kinda tasteless, but doing it months later seems entirely reasonable.
Hopefully, it's not too late in Open Thread 415 to post this. I just read a fascinating paper, "The Axioms of Cognitive Geometry: A Formal Model of Psychophysical Correlation" by Alexander Yiannopoulos. Creating a falsifiable mathematical model of consciousness is the first step toward understanding it. Yiannopoulos posits a dual-aspect or dual-substance quantum ontology, dressed in QM formalism, with separate Hamiltonians for the physical (outside world) and phenomenological (inner world). I may be missing something, because I don't see how he gets a causal coupling between the two, but I'm rereading the paper again to see if it was a comprehension issue on my part. But even so, it's pretty frigging cool that he attempts to model our internal phenomenological space as a Hilbert space. And then he provides a falsifiable framework. I love it! Be still, my Popperian heart! I'm not sure what sort of experimental setup he'd use. though? MEG? MRI?
Why do we need to modify the laws of physics to explain consciousness when a neural computational model seems perfectly sufficient? This does not seem like a parsimonious approach.
"The lived body is the pre-reflective 'zero-point of orientation' from which all experience is structured. We experience the world from here, where 'here' is constituted by our embodied situation."
I mean ... really? My prior for this being anything other than total nonsense is very, very, very low.
We shouldn’t have to modify the laws of physics to explain consciousness – and I hope I didn’t give you that impression. Unfortunately, it’s hard to step up a level of complexity without having some sort of rule-based understanding of the emergence of complexity. For instance, the standard model wouldn’t be predictive of the complexity of chemistry. And and knowing chemistry, we wouldn’t be able to predict the emergence of life (especially since the organic molecules required for life don’t spontaneously appear). The same goes for consciousness. It’s all very well to claim it’s just an illusion created by our neurons firing in certain sequences, but that doesn’t explain why we have the perception of red and can make the self-referential remark, "I see red, and it's pretty."
Prediction is much harder than explanation. No one would have been able to a priori predict consciousness-adjacent behaviors starting from the laws of physics, but from an ex-post position the laws of physics (or more precisely, the principles of computer science) seem more than sufficient to explain them.
>that doesn’t explain why we have the perception of red and can make the self-referential remark, "I see red, and it's pretty."
That's a very simple behavior that doesn't pose any conflict with physics. The neural mechanics of intersubjectivity haven't been worked out but I have no doubt that they will be eventually. In any case, how is a dualist explanation any better? Why is "this Hilbert space dimensionality reduction gives rise to the sensation of red" any more of an explanation than "this pattern of neural firings lets you perceive color"? It just camouflages the question with a new level of ambiguity. My view is that people have an irrational emotional resistance to understanding their subjective experience in purely mechanical terms. No one wants to see their wife turn into the green Matrix code. Sorry, but I'm pretty sure that's all we are. It feels magic to you because our brains didn't evolve to accurately perceive their own functioning.
Unless you can provide an objective, testable property of consciousness that resists a physical explanation then, as far as I'm concerned, it doesn't exist. Short of a behavior which can be constructively shown to be in conflict with the laws of physics, you're just making a god-of-the-gaps argument. Feelings of weirdness are irrelevant. "I like red" is something an LLM can say. There is no reason to presume that the subjective experience belongs to a fundamentally different category of explanation than the behaviors which are attached to it. Why would it?
Prediction is easier than explanation. because you can tell.exactly how successful you have been.
"That's a very simple behavior that doesn't pose any conflict with physics"
The puzzle -- the Hard Problem -- is not about explaining the report, it's about explaining the sensory quality reported.
". . .even when we have explained the performance of all the cognitive and behavioral functions in the vicinity of experience—perceptual discrimination, categorization, internal access, verbal report—there may still remain a further unanswered question: Why is the performance of these functions accompanied by experience?"
"Why is "this Hilbert space dimensionality reduction gives rise to the sensation of red" any more of an explanation than "this pattern of neural firings lets you perceive color"?
That isn't an explanation. Saying A causes B, somehow, is not explanatory. (Which is not to.say "it is done by a Hamiltonian, somehow" is.an explanation).
"Unless you can provide an objective, testable property of consciousness that resists a physical explanation then, as far as I'm concerned, it doesn't exist"
How do you everything is objective?
"Feelings of weirdness are irrelevant."
Qualia are directly accessible, the weirdness of Qualia is not.
Nothing is intrinsically weird. We judge weirdness by unusualness, and by theoretical expectation. These aren't the same. To encounter something every day is not to understand it .. although familiarlty creates an illusion of understanding. The layperson thinks magnetism is strange because they don't encounter it often, and don't notice that they experience gravity every moment -- yet gravitity is less well understood than electromagnetism to the physicist. To a young child, nothing is familiar or understood.,so everything is weird.
On the Weirdness of Qualia
Qualia are supposed to be immediately apparent, and the are supposed to be weird ... but the weirdness isn't supposed to be immediately apparent. Much confusion surrounds these points.
The weirdness isn't type 1 weirdness -- we literally experience qualita every waking moment. Instead it takes defiance of a theoretical expectation.
In the abaence of a science education , people tend to default to naive realism. For the naive realist, "red" is an entirely objective property of ripe tomatoes, etc. A ripe tomato "is" red , but looks blackish under green light. Dress illusion, etc
It takes some education (eg the Dress Illusion) to understand that qualia aren't objective.
(To the naive realist, it's thought that ls weird and insubstantial. The early mind-body problem, eg Descartes) was about thought. The invention of "thinking machines" dispelled that).
Moreover , a quale isn't a common or garden physical property that happens to be in the head. The expectation that everything in reality is fully described by physics , ie. that everything is quantifiable, objective and structural makes the existence. of qualia, which are qualitative, subjective and intrinsic surprising.
But without those realisations and expectations, there is nothing weird about qualia. Our ancestors were not puzzled by them.
Then why can everyone now deconstruct the causes of the 2008 GFC while only a small handful of people made a profit from it?
> The puzzle -- the Hard Problem -- is not about explaining the report, it's about explaining the sensory quality reported.
I’m well aware. One first has to demonstrate that there’s any reason to treat those two things differently. IMO no one has ever done a satisfactory job of that.
The rest of your comment reads like word salad to me. I genuinely have no idea what point you were trying to make.
It all boils down to an argument between people who think Qualia are something unique that needs to be explained, and people who think Qualia are just an emergent phenomenon of the nervous system that will eventually be explained in detail the same way we can explain, say, vertigo. Nobody ever convinces anybody in this argument!
If I'm reading that right, the claim is that if there is a certain mathematical relation between different neural processes this is going to show dualism is true and even how mind and matter interact? It seems a little like he's eulering himself. Did I get it right though?
That was my take. But I can’t see how he gets the physical and phenomenological Hamiltonians to interact using that QM formalism. I asked him on X but he hasn’t responded.
You post links to the greatest stuff! What’s your system for finding the diamonds in the dust heap?
The most interesting thing I’ve read recenly are some posts from Jordan Rubin’s blog. He’s building AI plug-in skills for mental moves one can make when ruminating: what’s an analogous thing, what’s the opposite? He’s get several posts about it, but this is the best one to start with because it includes an example of using one of the skills: https://jordanmrubin.substack.com/p/antithesize-this-raitah-edition
He put up a post about it here a couple weeks ago but it didn’t generate much interest. I think I was the only person to respond.
> You post links to the greatest stuff! What’s your system for finding the diamonds in the dust heap?
I have no life? I'm still over on X and follow a bunch of STEM and history posters there. There are still some virologists, epidemiologists, and biochemists on X, but most have moved to bsky due to the unrelenting harassment from COVID cranks.
Bsky for the virologists, epidemiologists, and biochemists no longer on X.
Substack is turning into an interesting intellectual resource.
I follow a bunch of YouTubers who produce videos on science and history, especially physics, archaeology, paleoanthropology, and linguistics. Sabine Hossenfelder is particularly good at bringing attention to new physics and physics-adjacent studies (and she rates them with her bullshit meter, which no doubt has contributed to her hat quotient).
I found this one over on X. Or, rather, it found me. I was arguing with some physicists physicalists who claim consciousness is weakly emergent and they'd have it all figured out in no time at all. Of course, I, being the Popperian Mystic provocateur, pushed back on that arrogance. In the course of the argument, someone posted that Sean Carroll says its weakly emergent, so it must be so. And I challenged them to post something other than epistemic handwaving. And Yiannopoulos replied: "How about this?" I don't think he expected me to read it.
On his show today Ben Shapiro described Cutis Yarvin as "a representative of a philosophy that is effectively a sort of right wing fascism" and part of the "right wing fascist opposition" to "left wing centralization". He criticized one of Yarvin's recent posts which called for Trump to dismantle the Constitution and build a "one party state". He pointed out that Yarvin has his fans within the Republican Party, including possibly J.D. Vance, and finished his segment by saying "In the same way that the left ought to divide off from its communist leaning splinter faction, the right needs to splinter off from this ridiculous neo-fascist trash. It's silly, and humoring it is incredibly silly. Not only silly, but dangerous, because humoring authoritarian nonsense from people who are purportedly on your side is a great way to be co-opted by those splinter groups."
Shapiro is certainly doing his best to defend American conservatism from its enemies to the right, be they Groypers or Neo-Reactionaries. I wish him the best of luck in doing so, and wanted to highlight an example of a pundit trying to police his own side of the aisle.
Devil's Advocate: To the contrary, communists and fascists and similar ones are the kind of thing that gets attention and votes in the Age of Clickbait.
(And yes, as a side effect they also give some attention and votes to the opposite side. That's how clickbait works. The losers are the moderate and sane ones.)
In what specific sense is Vance (possibly) a fan of Yarvin? I'm arguably a fan of lots of thinkers and talking heads with whom I nonetheless have significant disagreements, depending on the claim. Much of the time, I notice it's because they have phrased a particular idea in a way I find clear, compelling, or even just amusing. Yarvin is an example; I think his demotism is weird crankery, but I also admit it's sometimes fun to read his writing.
I think it's high time we all worked out a richer way to express distinctions between being a fan of some person and agreeing with all of their ideas.
I did a bit of searching, and it looks like Vance directly referred to Yarvin on a podcast in 2021, in the context of talking about how conservatives need to seize control of the bureaucracy by firing lots of people and replacing them with loyalists. Which the Trump admin has indeed done, so I don't think the fear that Yarvin influenced him is unfounded.
(Yarvin, for his part, said he doesn't think much of Vance and hasn't spoken to him directly. I did find one article that said he was in contact with Steve Bannon, though.)
I read the transcript of the podcast, and it was kind of an offhand mention, so I don't know how much the idea came from Yarvin vs. the idea just sort of being "in the water supply" on the far right. But I don't think there's a huge gulf in their viewpoints either - like, they agree on the need to purge the government of their enemies and neither one is very worried about the legalities of doing so, the question is just if Vance also agrees on the whole "American Caesar" thing.
I take all this as a question on whether the median American ought to be concerned about Vance becoming POTUS. The argument being suggested is that if Vance is a fan of Yarvin, then Vance might do anything just because Yarvin said it was a good idea; Yarvin said monarchy is a good idea; therefore, Vance will try to bring about a monarchy.
If this is the argument, then the reason I don't buy it is that "Vance will bring about monarchy" doesn't necessarily follow from "Vance is a fan of Yarvin". When resolving that argument, the claim that Vance is a fan of hiring loyalists ought to be irrelevant, because (1) there are multiple reasons both for and against hiring loyalists that have nothing to do with either Yarvin or bringing about monarchy, and (2) the people making the argument have been enjoying the benefits of loyalists in their administrations, which makes me think they don't really believe this argument.
Meanwhile, I haven't seen any evidence that Vance is any more a fan of monarchy than, say, any other future POTUS candidate. If the evidence that he does turns out to be solely "he's a fan of Yarvin", the evidence he's a fan is because he's a fan of hiring loyalists, and the people arguing this turn out to be okay with loyalists themselves, then it doesn't seem compelling.
Trump's purge of the bureaucracy has been much more far-reaching and politicized than any previous president's, so I don't really buy the argument that all presidents hire loyalists. For example, when the DOJ decided to start firing prosecutors until they found someone who was willing to drop the charges against Eric Adams, that was not normal operating procedure!
>I haven't seen any evidence that Vance is any more a fan of monarchy than, say, any other future POTUS candidate.
I disagree. I think Trump is much more a fan of monarchy than past or likely future presidential candidates, as evidence by both his own statements and his efforts to concentrate power in the Presidency, and that it's reasonable to believe that Trump's VP is on board with that.
Vance referring to something Yarvin wrote or said in a podcast is extremely weak evidence that Vance is strongly influenced by his ideas, or agrees with much he says.
It's... kinda sad that he doesn't realize he's already become the splinter faction now. Everyone else figured out that the old methods weren't working and have moved on to greener pastures, while he's still stuck on a lost cause. Oh well. Hopefully he realizes what's going on before it's too late.
The Groypers and the Neo-Reactionaries can't win in a general election: they can sabotage the Republican's chance of winning though, so I say Ben should fight to the hilt to keep them out of the seats of power on the Right.
You're far from the first person to claim they've got a clearer view of reality than the next guy over, very much including people whose view is nowhere as clear as they think. Meanwhile, you're on record claiming infinities don't exist and that any number of civilians can be killed to achieve a war aim without acknowledging any of the multiple usual objections.
How do we know you're not one of those people with an bad view who thinks it's clear?
That is a good question, how do you know? Not that the answer really matters, mind you, seeing as we'll see who's right soon enough. The people want change, and the current leadership wants change even more. It was only a matter of time until the legal and moral fictions that composed this society were unraveled.
Anyone have any suggestions on how to improve one's ability to ask good questions? My issue is mostly in academic settings but I should think most advice here would be fairly general.
You always hear that asking questions is essential to learn, willingness to ask stupid questions is a huge boon (https://danluu.com/look-stupid/), etc. and I can understand and agree with these points in the abstract, but when (for instance) I'm in class and a professor explains something, I often feel as though I've followed their explanation and, when they ask for questions, I struggle to think of anything to ask. Obviously this wouldn't be a problem if I did, in fact fully grok the subject on the first explanation, but often I find later when I try to recall information that it hasn't quite stuck with me despite feeling like I understood it to begin with. I assume that in these cases I've failed to really grasp the subject in the first place, and I suspect asking more questions would've helped me to do so, but I struggle to formulate these questions at the appropriate time.
Does anyone have any particular strategies they've found helpful for being able to quickly formulate questions which get to the meat of an issue and help to promote long-term internalization?
This is the point of problem sets. You don't really understand anything until you actually use it. I recommend working one lecture ahead in your textbook. Do some problems for the section he's going to be lecturing on. If you're unable to understand the material on your own then you'll come to lecture with ready questions.
* have already thought about something related to the topic
* tried to apply the topic to solving actual problems
...and maybe if your thinking is so much faster than the lesson, that you could do that during the lesson. But most likely it had to happen before the lesson.
Basically, the questions are about some tension between the lecture and something else you are thinking about. If there is only the lecture, there is nothing to contrast it against. (Unless the lecture is self-contradictory.)
I think classes are a very dumb way to transmit information, and I never not once in my entire life ever paid attention in any class, in either school or college. It's true I wasn't a high scorer in college, but I think that was more down to me having no study habits either.
Anyway, I think you should use office hours. Makes way more sense to have a go at digesting the material yourself, then you will be clear about what you don't understand and can go talk to the professor about it. Having to think up good questions off the cuff is a silly constraint.
When explaining things to other people irl, I've found that it's important to keep the topic grounded in what the other person already understands. For someone with more advanced knowledge, you can be relatively abstract. But for a beginner, you need to be very very concrete.
I don't know you, so obviously I could be wrong. But what I imagine is happening to you, is that the professor is talking quickly about a bunch of abstract topics, and each factoid sorta/kinda makes sense to you in isolation. But because the lecture isn't being related to your extant base of knowledge, the lecture has a "castles in the sky" quality, such that it immediately evaporates when you leave the room. You need the prof to anchor the lecture in concrete examples. So feel free to ask for concrete examples. In contrast, the professor naturally wants to be speedy and abstract, because it all feels boring and trivial from his perspective.
Shankar says "take better notes", and then doesn't elaborate. What he really means is that you should "annotate". "take notes" and "annotate" sounds like the same thing at first glance. But they have different connotations to my mind. When I hear "take notes", it means "write down what the professor says, word for word". But "annotate" means something closer to "just jot down whatever flits through your mind". E.g. if your bio professor is discussing the krebs cycle, feel free to jot down random garbage like "Oranges = citric acid. I ate orange last week lmao." This is useful because it grounds the topic in something concrete.
Meanwhile, Eremolalos says "don't take notes". She says this because most people just take notes to reread them. But rereading notes won't do jack squat, because it only tests your recognition. Recognition won't help you on the exam. *Recall* will help you on the exam. So yes, use flash cards. But you should still annotate. Not to reread them, but for the act of having written them.
Well if it's a small class (<20) then I would suggest asking questions right when they come to you. If you are not following some step in the explanation ask right away. In a big class constant disruptions are... disruptive. and you need to remember where to were uncertain... maybe write it down?
Take notes. Or better ones if you already are. It's easy to let the words wash over you leaving behind an illusion of understanding that fleets like memories of a dream when you next think about it. If you write down each point made as you hear and digest it, in addition to improving comprehension, you'll notice more quickly when it stops making sense.
It's been a few years since I was last a TA, so I don't if the trend I observed has held, but the professor making lecture slides available after, while certainly good and useful, doesn't obviate the need for taking notes yourself.
I am a big note-taker, although as you say I probably could have a better strategy for taking notes.
Generally I take notes in a somewhat stream-of-consciousness fashion, writing out sentences based on the content of the slides and what the professor says in the lecture, drawing lines between connected subjects in the margins. This seems to serve me well in history courses (my main area of study) but perhaps it is less suited to my more technical courses like biology, which are indeed where I most notice issues with my retention.
Do you have any particular note-taking strategies you recommend for more technical subjects?
Sorry, if you already consider yourself a reasonably good note taker (in your own judgment), I don't have any better advice to offer. What I said is only good for people who don't take notes at all, or so poorly — on the level of "Everyone else in class is taking notes, so I'll have do something too. Let me write down the title of every slide." — that they would themselves agree it doesn't even count.
But I'd say there are maybe three broad styles of taking notes.
- essentially transcribing every word the professor says, or replicating the content of the slides. Frankly, I think this is usually actively detrimental to your learning: when you're doing this, you can't pay attention to the informational content being delivered.
- As a way to keep your attention from drifting away from the topic, where you could chuck everything you've written in the wastepaper basket at the door as you leave class and lose little of value. My earlier advice was of this kind: hear the professor make a new point, understand it, and then quickly scribble a short enough précis of it that you can put your full attention to the next point as it's being made.
- A magical distillation of your state of mind, where a mere glance much later lets you load back perfect recollection of the material as if you had heard it just moments ago. If this is what you're looking for, I bet something close to this ability exists, but I don't have it.
For what it’s worth, I actually am against taking notes in class if it can be avoided. Taking notes is quite a different task from listening to the lecture, and it interferes with paying intelligent attention to the lecture just like sitting there painting with watercolors would.
The thing that helps turns immediate memory of a bunch of info into a solid grasp of it is working with the info. That’s what problem sets and thought questions at the end of textbook chapters are for. You can do a follow-up of the lecture that works the way those things do: Sit down after the lecture, ideally right after, and make additional notes, preferably right on the slides and lecture notes the prof provided. Write relevant stuff you remember right on them.
Doing that will also make you aware of things you don’t understand or remember well. Keep a record of those and ask the prob or TA later, or just look them up somewhere and find the answer.
Also, it’s important to be realistic about how much of a lecture can be remembered. There is no system that will leave you remembering every single detail. What you want to get from the lecture is big picture understanding of the system. Many of the details you don’t really need to remember.
If there are lots of details you actually must commit to memory, for instance the structure of all the amino acids, you won’t be able to learn those just by paying close attention while reading or listening to lectures. That kind of stuff you learn by putting it on flashcards and using something like Anki.
By the way, I’m a psychologist, so am not just speaking from personal experience. Still, I may not be right. Think of my take as one to try on.
"Taking notes is quite a different task from listening to the lecture, and it interferes with paying intelligent attention to the lecture just like sitting there painting with watercolors would."
...this is what I was getting at with "figuring out how to quickly [summarize the lecture] took all of my focus from actually understanding it". Glad to hear this is common knowledge elsewhere.
Agreed on "working with the info". (As I mentioned, I had the hardest time with English and history, and I think a lot of my trouble was because I couldn't do this like I could with a numerical method or a stochiometric equation.)
Big-picture understanding: another problem I had was recognizing which parts of lecture were load-bearing, as opposed to detail I could discard. I think this was mostly my lack of experience, coupled with different lecturers having widely different organization - some use BLUF, others start with a story and the Important Moral at the end, etc.
As a naturally combative person, my answer is "seek to destroy". Every answer is based in assumptions; find the assumptions and attack them. Finding them takes "why" questions, attacking them takes "why not" questions. I guess if you want to get better at questions in general, make a point of forming at least one of each in every lecture, and increase as fitting.
This seems helpful, framing it as "why" and "why not" questions seems to make sense.
Perhaps my approach has become entirely backwards, as I've gotten into some trouble in the past for applying this sort of mindset in social settings, but I've (not consciously) adopted a more deferential mode in academic settings.
Sometimes when listening to a lecture you’re left with a vagueness in your understanding rather than an actual question. Maybe you are having trouble formulating questions because you’re mostly having vaguenesses. In that case, ask the speaker to restate the relevant part briefly — or maybe to add a little
> Sometimes when listening to a lecture you’re left with a vagueness in your understanding rather than an actual question. Maybe you are having trouble formulating questions because you’re mostly having vaguenesses.
This seems to match with my experience, yes, although I suppose the crux of the issue is that I'm having trouble noticing these vaguenesses? Or rather, I guess in the past I have noticed these vaguenesses but considered them merely a symptom of my early stage of understanding, rather than a problem to be addressed... I will keep this in mind this semester.
The catch here is that ESR's guide assumes you know what your question is, and that you need help with how to ask it, and your middle paragraph indicates you don't even know what the question is yet.
The advice that springs to mind is to figure out a way to quickly explain the lecture content to yourself (without speaking, of course, if you're still in the room). The process of explaining often causes questions to naturally pop out. For example, suppose the lecture is about the Java programming language, and the topic that day is the 'interface' keyword. Explain what 'interface' does to yourself, and you might notice that C++ has a different way of handling it that might strike you as easier, and you might wonder why Java doesn't do that. That turns out to lead to a deep detail about the tradeoffs around multiple inheritance that you might find useful when thinking about OO languages in general. (It relates to how method pointers are organized within the compiled Java class file, but it's been two decades since I looked into it, so I can no longer tell the story.)
Despite not quite responding to my specific problem, I found ESR's text enriching, thank you for the link :)
As regards finding a way to quickly explain lecture content... hm, I guess it comes back to better note-taking, as Shankar suggests above. I'd welcome any suggestions for note-taking on more technical subjects.
Aye, as I said, yours is not quite the itch it scratches, but I felt you and other readers might find it germane enough to benefit elsewhere, so I'm glad you liked it.
I was notoriously bad at note taking in high school (my English teacher specifically called me out in public for appearing to do nothing during his lectures when I was just sitting there trying to listen and make sense of what he was saying), and never felt like I'd gotten the knack in college, so I can't be much help there (and also welcome anyone who cares to post about it).
The best I can give is an account of what happened when I tried: it became an exercise in writing whatever the lecturer was saying, but summarized. Talking is faster than writing, so my summary had to be fast enough to keep up. So it mostly became practice for my own version of shorthand. However, figuring out how to quickly build a new convention for shorthand took all of my focus from actually understanding it. And when I already had a convention for certain things, writing the notes typically resulted in my remembering what I heard for a long time, so I also got in the habit of never re-reading my notes, because it would have been a waste of time. I came to regard note taking as a sort of mantra that forced me to keep my attention on the lecture and not daydreaming about something else, and afterward, I could just throw the notes away.
Generally, I could tell that revisiting notes about a subject I wasn't interested in (such as English or history) wasn't going to help - it wouldn't improve whatever essay I was going to be assigned to write - and for subjects I was interested in, I found I learned a great deal more by putting my pencil down and just listening to the teacher.
I elaborate in my reply in another comment, but the mantra thing for focus and recall was what I was suggesting. We might mostly be in agreement, except I think that it'd be helpful even in subjects you ARE interested in.
"[China has] estimated 50% youth unemployment, general unemployment now illegal to publish (but you can still approximate by falling commuter numbers), some government workers haven’t been paid in more than a year (living entirely off extortion/bribery; includes police, teachers, and all healthcare), common experience in private sector to have your paycheck delayed by “2 weeks” that turns out to be 16 weeks. Starting salary offer for software engineer in Shanghai is <1/3rd legal minimum wage. You cannot partake of public services without home “ownership” (hukou system, and technically its a land lease). Private health insurance is overwhelmingly likely to be a scam and rarely used. Public coverage primarily covers TCM; most western medicine is out of pocket. If you cannot afford to pay - even in a trauma situation - you will be escorted off the property so your death doesn’t lower hospital mortality statistics."
I'm glad that I started a discussion that's spanned multiple Scott Alexander posts, but I'd really like to see a source for these claims. 50% youth unemployment is far higher than the 17% I typically see thrown around. Yes, I know we shouldn't believe Chinese government statistics, but that's not a reason to believe arbitrary figures.
The point about hukou is correct, but the hukou system has been around since 1958. It can hardly explain the Chinese people's nostalgia about the 2000s. Similarly, China's healthcare system has always been less than ideal, but the 21st century has seen vastly improved access due to a combination of urbanization, rising incomes, and expanding insurance coverage.
For the other claims, I'd like to see a reputable source. For example, I don't doubt that some wages are delayed, but where's the evidence that wage delays are widespread? Where did the OP hear that the starting salary offer for a software engineer in Shanghai is <1/3 the legal minimum wage?
An explanation for the discrepancy I saw elsewhere in this comment section is that the 50% figure includes people NOT actively looking for work, while the 17% uses the more conventional methodology of omitting them.
To some degree it makes sense. There is a difference between someone who wants to be e.g. a stay-at-home mom and never applies for a job... and a similar person who applies for a job but can't get it. For starters, the former is happy about the outcome, but the latter is unhappy.
But things get complicated, because both sides (the unemployed, and the government) have an incentive to distort the truth. For example, if you say "anyone who can't find a job, come here and get some money", the stay-at-home mom would also be eligible to take the free money. But if you say "no, you must be actively seeking for the job", it is easy (and tempting for the politician, to improve the numbers) to set the bar for "actively seeking" too high.
And there is a lot of gray area. Should someone who arrives drunk to every job interview really be counted as seriously trying to get a job? Should someone who needs to have two jobs, but could only find one, be counted as "unemployed"? What about a student who can't find a side job? A university-educated person who could only find manual jobs. Etc. Ultimately, every country counts this differently.
We have in the United States, thanks to the unintended consequences of welfare "reform", a significant number of people who were thrown off the rolls of those eligible for financial assistance to the unemployed because they could not find a job in six months or a year or whatever, and so to pay the rent were driven to join the rolls of those eligible for financial assistance because they are permanently disabled. Not by e.g. missing limbs, but by chronic fatigue syndrome or Long COVID or whatever, trust them, they've got it and they've got a letter from their doctor to prove it.
But in order to keep that going, they have to never get a job or even be seen applying for a job, because then it would be obvious that they aren't really permanently disabled. Do we count these people as "unemployed"?
If we want our unemployment statistics to look good, then of course we take the excuse we have been eagerly offered to say, nope, these people aren't unemployed. But if we want to know how many people could be working but aren't, or even how many people would prefer to be working but aren't, then we're hiding a big chunk of signal there.
Similarly, trying to get as many people to universities as possible is a clever way to reduce youth unemployment. (Well, it keeps the youth off the streets for most of the day, so I guess mission accomplished.)
Yes, that's a requirement in the definition used by the UN's International Labor Organization, the OECD, the US's Bureau of Labor Statistics, the EU's Eurostat. Even if there are countries that internally favor a different measure (Germany seems to be an example of this), any number you're likely to see would be harmonized to match this.
It looks like the BLS actually tracks six different measures of "labor underutilization", from U-1 to U-6, but the one using U-3 is the official unemployment rate.
Do you know why Germany is regarded as having different measures?
From my experience: if you register for benefits while unemployed, you are in active contact with the job agency and therefore regarded as looking for work (and they want to see proof of your continued search for work).
If you can afford it, you can just check out and stop receiving benefits. I would be very surprised if these people even could be counted as unemployed, as no relevant agency knows of their existence or status of employment. They will be paying health insurance out of pocket and any taxes due on capital income or otherwise, and that's it. They are indistinguishable from any freelancer.
It's not a subtle difference: the measure common in Germany (the one in their newspaper headlines, etc.) counts people working, but less than 15 hours/week.
I've been hanging out around LW and here since like 2010, read all the scifi. But here on the eve of the singularity or whatever I'm losing hope. Or just getting cynical with age
I'm starting to think we're just going to have vaguely adequate simulacra of knowledge workers (who were supposed to be working rather than posting on reddit and quora, but progress plateaus once we eat all that low hanging fruit), and we're left in a slop economy. With the same ugly buildings, decaying infrastructure, and an llm fueled social media hellscape destroying our souls and communities. And a lot more wealth inequality
"own a little AI stock and you'll get a moon" sounds like scientologist or bitcoin maximalist ad copy.
1) I doubt we're on "on the eve of the Singularity."
2) Your description of LLMs being essentially "fake AIs" with hard limits on their capabilities is accurate. They will not be able to fully replace humans--not even close. However, the mistake you're making is assuming no further progress will be made building bona fide AGI. I believe it will, and that the current hype over fake AI will have long-term value since it has underpinned the construction of the physical infrastructure (data centers, power sources for those centers, other property and rights-of-way acquisitions) that will support AGI.
If you think AI won't amount to much, you shouldn't believe the "permanent underclass" meme. Wealth inequality sucks, but the solution is just the normal boring politics of taxing and redistribution.
And if you believe AI *will* change everything and make the capitalists fantastically wealthy (in any form, not necessarily moon ownership), then you should buy stocks, because that's how you become a capitalist.
I really do not think it impossible to predict how all this will play out. There is so much at play, and the AI part of it is a novel entity for which the past does not give us an analog. Seems to me that believing things are going to end up taking a certain course is mostly a way to escape hard-to- endure uncertainty. It’s more fairminded and also better for your head to just float in uncertainty
So...we're only going to arrive in Hanson's "The Age of Em", not superintelligent AGI? We're only living in 2016 sci-fi? That's depressing?
I mean, I don't want to be as giant jerk face but six years ago we did not have things like intelligent AIs that could act as junior software engineers or self-driving cars on the streets of SF. Going further back, at some weird point in the 2010s we stopped primarily "living" in reality and started living literally virtual lives, like spending ~7 hours a day staring at screens not including work time. (1) If you and I had put $100k in Google when they bought Deep Mind in 2014, we'd be millionaires.
I vibe with this. I'm shocked by how mundane all this feels and I think that's irrational. A small handful of megacorps are consuming the entire economy (2) and tracking our every movement as democracy slowly retreats. I'm literally living in a 1980s cyberpunk novel...and yet I still poop every day in the same kind of toilet I pooped in 1995 when AOL was a big thing.
I don't know why it feels like this but I am confident, in my heart, that someone is going to solve aging in the next 30 years and the day it's announced we'll be like "Wow" and then we'll all start griping on TikTak of whatever about how this means Social Security and Medicare are doomed forever and maybe we all liked dying and how dare these people make us live forever and I'll still be pooping in a porcelain toilet.
Not really, no. Curtis Yarvin may be a crank, but I think one useful thing he pointed out is that you can tell elections are meaningless because after the transition, you can go take a walk and everything is still the same. Ditto for this situation.
Yes, even in San Francisco. I walked around San Francisco in early 2025, I saw the Waymos, and it's cool, but I mostly remember the homeless.
> and yet I still poop every day in the same kind of toilet I pooped in 1995 when AOL was a big thing
Okay, but real talk here, Toto toilets have existed for a long time, but basically nobody in America has them.
If you want the "toilet of the future," it's buying a higher end Toto, which will automatically open and close, wash and dry you, de-odorize, play music, and do sundry other advanced things to your nether regions entirely dazzling and unknown to 99.9% of Americans. But you can have this toilet of the future today!!
And it is a much more civilized way of being, I highly recommend it.
Yeah, the speculations on utopia feel pretty irrational. Like they're wildly overestimating what intelligence can do, or how difficult space travel is.
-Crime is down. In fact, from 2021, it is down about 50%. Nonfatal shootings are down as well, so it isn't just emergency rooms getting better.
- This is not an official site, and their numbers are a little different from police statistics. I think that the difference comes from the fact that the site counts things like police shootings, self-defense cases, and crimes investigated by the Illinois State Police, not just the CPD. But something to note.
-Unsurprisingly, guns are the preferred method of killing, with stabbing a very distant second, and blunt-force trauma third. Also, one must wonder about the two homicides listed as unknown cause.
-The Austin neighborhood was far and away the deadliest, with 46 homicides. Two others tied for second with 24 apiece. An argument for hot-spot policing in Austin? Still, the days when individual neighborhoods might have 70, 80, or 90 homicides are gone, hopefully permanently.
-Some neighborhoods have unusually high ratios of homicides to nonfatal shootings. Perhaps the ones with lower ratios have better emergency medical services. (One hopes that it is not that other neighborhoods simply are better shots.)
-There was one unusual statistic; Chicago had two quadruple homicides last year, which has not happened in a while. Of those two, one (an arson case) has been solved, while the other (a mass shooting) has not. Interestingly, it had no triple homicides. In 2024, by contrast, it had not quadruples, but four triples. As best I can tell, (which is to say, what I got by Googling it), two have been solved, or at least someone has been arrested for them.
-The face that I keep up with this may suggest that I have too much free time.
"Before launching AH Datalytics, Jeff served as a crime analyst for the City of New Orleans and Jefferson Parish Sheriff’s Office, and prior to that he worked as an analyst for the Central Intelligence Agency and Department of Defense."
Importantly Asher doesn't rely on any single data source. For one thing he distinguishes between and explains the varying strengths of local/state/federal recordkeeping of reported crimes; also he cross-references with crime-victimization surveys that reflect unreported crimes. For good reasons he is starting to attract some media attention, and everybody who follows him is now telling him to "write the book" (hopefully with a less-cutesy title than 'Jeff-alytics').
Down relative to the height of the Floyd spike in 2021. Compared to 2019 (a low year) it is up 18%. A more neutral assessment is that it's slightly below the 10-year average. The data only goes back ten years so it's hard to make any other judgements from the data alone. Interestingly, it looks like the total population of Chicago has been flat for some time.
Yeah individual cities are going to be subject to more random variance than a national cohort. That "lowest since 1965" looks to be about 5% lower than the 2000-2010 norm so I don't think it's fair to say that this represents some secular change. Maybe the 2020-2024 orgy of violence purged all of the bad blood for a few years.
Ok so roughly on the national trend. Fair enough. 20% below 2024 levels is 13,500 homicides. There were 13,900 in 2014. That feels consistent with reversion to pre-2020 norms coupled with an unusually law-and-order president.
My monthly wrap up of the most interesting long form content covering evolution, history, and the impact of resource limits on civilisation is out and this one is huge.
Highlights include an analysis of drones in modern warfare, the evolutionary history of the human genome, a field report from Swidden farming villages in Thailand, the third and final part of a thesis link gut dysbiosis to schizophrenia, and a great summary of the current unregulated peptide therapeutics sector.
Also check out my interviews with archdruid John Michael Greer, Chinese historian Professor Jiang, resource economist The Honest Sorcerer, and experimental farmers Professor David Shields and Bruce Steele.
I likewise appreciate a little feedback that I am appreciated. Always a bit paranoid I am annoying people when I cross promote my work. Greer had been on my wish list to interview for a long time.
Lately, I have asked ChatGPT some extremely complex questions and I have been amazed at how accurate and thorough the answers are. Embarrassingly, they are much better than I was able to write, in some cases after months of research. This probably outs me as a midwit to the geniuses around here, but so be it.
Examples:
1) Why do women earn less than men?
2) What are the essential components of a free enterprise system?
3) What should I look for in midlength surfboards if I am this size and age and surf this type of waves at crowded SoCal reef breaks?
4) Which audio systems in luxury cars have the best audiophile sound quality?
5) Summarize Streven’s book The Knowledge Machine.
6) How do I reset the AirPlay on this model printer?
I am knowledgeable enough to evaluate that the answers in every case were excellent. I could ask follow up questions, and the system prompted me with some areas of further exploration, which it then went on to answer.
It seems we are quickly approaching an era where I will be able to learn more by asking and discussing issues with AI, than we could in months of frustrating debates and fights on social media, blogs, Reddit, and Substack. Indeed, it seems that we could have a Substack format where someone asks AI a controversial question, shares the answer with the audience and we discuss or query each other and the AI.
I actually let Claude talk me into replacing my car (I think it was right, it had a nasty crash last year and I shouldn't have had repaired it, since there's no guarantee the chassis is still safe. Also was a 2011 model, low mileage however). But you always need to be on guard for hallucinations.
That's the opposite of my experience! I've stopped using ChatGPT (admittedly, as a logged-out nonpaying user) and switched to Gemini over the past month or so, due to a dramatic decline in the quality and reliability of the answers I was getting from ChatGPT.
I agree, and I think people who aren't using a good LLM to help them navigate medium-size questions like this are leaving an awful lot of value on the table.
I genuinely don't mean to sound snarky, but if it comes out that way - well, I think I can still claim true and important. Because, well, it appears that the idea of life quality not necessarily being correlated with legible economic markers turned out to be much easier to accept when it was about China.
We’ve talked before about people who are prescribed beta blockers for anxiety, and I said, as someone with a thyroid condition who gets prescribed way bigger doses of beta blockers, I don’t think there’s much of a psychological effect.
Ok, this time when the endocrinology department were slowly ramping up my drug dosages I do feel an effect.
At 40 mg carbimazole + 7.5 mg bisoprolol (daily), my resting heart rate was measured at 155 bpm (yeah, that’s high)
At 60 mg carbimazole + 7.5 mg bisoprolol, my heart rate is about 110 bpm.
At 80 mg carbimazole (the endocrinologist is getting bold here) + 5 mg bisoprolol, my heart rate is about 75 bpm. (Increase carbimazole + decrease the beta blocker as they both reduce heart rate and don’t want to go too low).
So, ok, I’ll admit I felt the difference in psychological effect there. (The carbimazole is doing most of the work by driving down my T4 level). And yes, when the endocrinologist put the dose up to 80 mg I was looking up the medical literature to see if any thyroid patient ever has been given that high a dose (yeah, they have),
I think the focus on "absolute" well-being in the analysis above is misplaced. Maybe people should care about their absolute well-being, but we're social creatures and we tend to be a lot more focused on relative well-being. If the people that own Anthropic stock are actually going to be owning galaxies, the people whose $25k in savings allows them to live a circa 2025 American middle class lifestyle are not going to feel great about it. I guess that's what you're dismissing as "pride" but even a rather humble person engages in some level of social comparison.
And I'm not so sure about this unspecified trickle down mechanism either. There are plenty of people in more remote parts of the world today who haven't really experienced any gains as a result of the industrial revolution.
I dont think its purely envy, though. To some degree we all also have a desire to live a life with purpose, to put it even more favorably: to make the world a better place
If anthropic shareholders own galaxies and you subsist of scraps to maintain a comparably prehistoric lifestyle, its not so much "they have shinier toys than me" but "they have more power and capacity to shape the future than me"
Theres also ego and pride, which of course yes are vices and demand less sympathy. But its really not just one vector. Theres a lot of reasons to be unhappy with inequality, particularly if it is built on your labor/consumption
> AI company employees will form the new ruling class, with everyone else as serfs.
Surely the idea is that AI company CEOs will be the ruling class. If the people they have rendered unemployed are dependent on them for a voluntary UBI, they can use their votes to get into power. Well, there are many possibilities and one of them looks like being neo-feudalism.
Paul Graham: The rational fear of those who dislike economic inequality is that the rich will convert their economic power into political power: that they’ll tilt elections, or pay bribes for pardons, or buy up the news media to promote their views.
I used to be able to claim that tech billionaires didn’t actually do this — that they just wanted to refine their gadgets. But unfortunately in the current administration we’ve seen all three.
It’s still rare for tech billionaires to do this. Most do just want to refine their gadgets. That habit is what made them billionaires. But unfortunately I can no longer say that they all do.
Can anyone steelman the argument that some AI company is going to become a world-eating economic behemoth? Other than NVIDIA, I don't see anyone having a substantial moat. There's no secret sauce and all the models seem to have roughly equivalent capabilities. Are there network effects that I'm not seeing? Right now this seems like a commodity business to me.
Something I read that impressed me: Companies are currently interdependent, renting and buying services, cloud space and materials from each other, and all are dependent on chip manufacturers and no doubt on providers of other components and services. Some companies are now trying to develop vertically -- making their own chips, developing their own data sources for training, etc. If one succeeds *enough* to be markedly less dependent on other companies than the rest of them are, there will come a tipping point where it is so much more financially independent and efficient much that it can afford improvements that others can't, and so on til it's a world-eating behemoth. I am not knowledgeable enough to rebut rebuttals to this -- just had a gut feeling that it was a smart take. Oh yeah, also, Musk is doing the vertical development thing. (And he has Twitter as a source of training data.)
Musk does vertical development because he can't find anyone willing to sell him what he wants to buy, e.g. rocket parts that work but haven't had their price jacked up by 10,000% because they come with lots of paperwork "proving" that they work 99.99999+% of the time. He has said that he would prefer not to do this, because he's got finite bandwidth and just cutting a check is more efficient.
I think a rebuttal to this is that it is not clear that it is more efficient to do all of the supply-chain internally. For this to be more efficient, I think the company branch doing part x must in principle outcompete all of the other external companies doing x minus overhead of doing buissiness and whatever margin the external company can extract over time. This seem to sometimes but not always be possible, from the evidence that companies sometimes seeks to internalize some process, while other times seeks to outsorce processes. Probably, internalizing a process is mainly a good idea if there is little competition externally in that area (maybe due to barriers of entry)? Otherwise your internal branch does not have the same pressure to perform from competition, and may rather become less efficient at least in theory.
The assumption is that the AI at some point becomes actually useful to the point of being able to automate most if not all economic/industrial processes, and in addition actually creative to the point of being able to come up with novel technological improvements.
If you think current state of the field makes this assumption very far-fetched - yeah, you're not alone. But if you do assume it, economic domination follows quite naturally.
I do assume it. But I also suspect, based on current evidence, that it will be a commoditized technology and so no single company will be able to extract outsized rents.
> Can anyone steelman the argument that some AI company is going to become a world-eating economic behemoth?
I've got another one besides the "white collar work" one - there's the potential for the AI chatbots to become the next trillion dollar attention surface for auction, much like smartphones made Google and Meta into trillion dollar companies, by grabbing enough eyeball hours.
There are already ~1B MAU using LLM's today, and that's before they live in everyone's ears and become personal assistants that can make phone calls and answer emails and curate your media feeds for you. When that happens, it's the next smartphones - people are going to be interacting with them all day long, intermittently, and trust is going to be high, so ad placement value will be correspondingly high. They can also charge for more capable personal assistant tiers.
And much like people went nuts for GPT 4o and mourned the loss of their AI-werewolf-vampire boyfriends and whatever companies will have differentiation in terms of the personalities of the chatbots, just like all the chatbots have distinct personalities and capabilities today.
Agreed! Look, I'm not arguing that AI won't be spectacularly useful, I just suspect that the technology will be a commodity. The only reason Facebook is so valuable is because of network effects. Maybe LLM-producers will be the future equivalent of hosting providers and the real moats will be for the downstream applications. Look back at the late 90's: the early behemoths were in a bunch of different markets. You didn't see Amazon fighting with Yahoo for market share. I just don't see any product differentiation between LLMs yet. Maybe it'll come but it hasn't yet.
1) Data: Models are already training on nearly all easily accessible, public data. As use of AI scales up, user input and feedback become a more and more important source of data to build the next model. This exists for chatbots, but is especially pronounced for real world devices with sensors. If Company A has 24/7 data from millions of robots and Company B has data from 6,000 of them, Company A will soon develop an insurmountable edge.
2) If we're on a path to fast progress, then someone will eventually develop the first AI capable of rapid self-improvement. As it self-improves, they'll develop an impossibly large lead over whoever is in second place.
3) If we're on a path to slow progress, then AI will get integrated into all kinds of different systems and you'll start developing lock-in where it doesn't matter if Company B has a somewhat better product than Company A because you've built up so many dependencies on the way that A's technology works.
1) User feedback is available to anyone with users. There are many AI companies with users. For a given amount of compute there is a data saturation point where the marginal utility quickly falls to zero. So having 10m vs 1m users seems likely to be immaterial.
2) I don't see why that wouldn't be quickly replicated. It's just another capability and we've already seen that capabilities are quickly copied between models. Capability is gated by absolute model performance and that's largely just a question of compute. I don't see the moat there.
I'm not even sure "rapid self-improvement" is meaningful. The limited data we have so far indicates that all that matters is compute + data. ML expertise doesn't really matter, so I'm skeptical that ASI would help. You don't need ASI to tell you to use more GPUs.
3) Yes I can see some lock-in at the application layer. But that seems like it would be limited to market segments: one company might own legal AI and another might own coding or whatever. It's hard to predict how the ecosystem will shake out but I don't see any obvious natural network effects. There just isn't any secret sauce: it's all downstream of compute. If one company develops some hugely profitable capability then that just becomes a signal for capital to rush into a competitor to train it up.
1. We're already maxed out on data; while we face compute limitations at any given moment, the availability of compute is expanding much more rapidly than the availability of data. User data is also uniquely useful; that's why Google has been dominant in search for so long despite having no technological moat.
2. Fair enough to not believe in recursive self-improvement, but if it does happen it'll be very hard to quickly replicate because it turns a small lead into a big lead overnight and everything before just becomes irrelevant.
3. By this logic, IT in general ought to be a low-margin industry, which is obviously wrong.
1. Google absolutely had a technical moat for several years. I don't know if you're old enough to remember the internet in 2000 but Google was the only search engine that didn't suck. The difference was enormous and lasted for several years. That technical edge didn't last (Bing quality is indistinguishable now) but Google established a brand and built a platform (Chrome, Android, gmail, etc) around it and that created user lock-in. I don't think it's fair to say that their access to user data gave them an insurmountable edge because Bing surmounted it.
2. Maybe. I don't think technical evolution works like this without either market network effects or a patented key technology, and patents don't have much of a track record in software. Maybe there'll be some secret nonobvious breakthrough that one company stumbles across that's hard to duplicate but, as I said above, that's not the pattern we're seeing at all so far. Maybe Anthropic or someone will come up with the AI equivalent of PageRank and become the Google of AI for a few years, but I tend to side with this article in the "it's nothing but compute" camp: https://samuelalbanie.substack.com/p/reflections-on-2025
I think this probably comes down to how we model intelligence. I don't think it's some unitary thing that once you discover it it's game over. I think it's agglomerative and every AI company will slowly accrete raw horsepower until they eventually converge at the same saturation point where marginal compute doesn't yield positive-economic-value improvements in model performance. I could be wrong but that's what the early returns suggest so far.
3. It's a lot higher for software than it is for hardware (NVIDIA notwithstanding). That's because AMD vs Intel are interchangeable and so have to compete on price, whereas network effects create winner-take-all dynamics for the application layer: Facebook, eBay, etc. I think it's possible that LLMs (or whatever replaces them) will become interchangeable cheap-marginal-cost commodities. The margins go where the bottlenecks are and for AI right now that's at the GPU.
Not really in the spirit of the question, but how will those drones stop the ICBM that gets launched at corporate HQ the instant a nuclear power perceives an existential threat?
Interesting read. I think his analysis held up well in 2023 when he added his first-year Ukraine postscript, but I don’t think it rings as true today. Drones keep advancing to fill new niches which conventional weapons can’t fill as cheaply. Some of that is because Ukraine doesn’t have access to as much high-end conventional munitions as they’d like, defensively as well as offensively, but I still think drones will play an increasing role in warfare.
Though, to be serious, I think you need industrial production capacity that only a few countries have to be able to drone strike the entire nuclear detterent.
> Can anyone steelman the argument that some AI company is going to become a world-eating economic behemoth?
There's a multi-trillion market for the first company to put together the right framework, memory, persistence, prompting, and long term learning to do a given white collar worker's work.
Given the models are already smart enough, it's all the other stuff that needs to be knocked out. But when that happens, the company that does it can net trillions, and it seems like it will probably have a fair amount of tacit / individual / trade secret style knowledge in that framework, so there's room for multiple companies to tackle different jobs / segments successfully.
>have a fair amount of tacit / individual / trade secret style knowledge
I'm skeptical about that part. The short history of LLMs has shown so far that there really isn't any secret sauce. Anything that one model does another model quickly copies. "The performance of AI machines tends to improve at the same pace that AI researchers get access to faster hardware."
At best I see a temporary first-mover advantage before any unique capabilities are reproduced in an open-source model ala Windows vs Linux. Not that Microsoft isn't still very profitable but it's effectively lost the battle with Linux in terms of what runs the most CPU cycles.
This seems right to me. I think some people who believe in certain types of AGI concept think there is a phase transition where passing that transition even a few weeks before a competitor gives you an insurmountable advantage. But the real life experience of interacting with jagged intelligence is making me more skeptical of the kinds of AGI concept that are needed to make that work.
(disclaimer: I'm not totally familiar with the details of every major prediction market's operations, so if they're already doing this and I'm reinventing the wheel, then I'm surprised in the direction of them being more thoughtful and competent than I thought, and will update my impression of them positively)
When the question resolves based on a singular event that could happen at any time, should prediction markets invalidate (refund) any predictions made within X days (1? 7?) of that event happening?
Inspired by the Maduro thing, but as a general concept: One of the objections to prediction markets is that when one person/group is making a decision that can single-handedly resolve it, they can just bet as much as they have on the correct outcome before they do it. As far as I know, the response to this has been that them betting strongly is still useful information, so it still makes the prediction market useful in predicting the right answer, and it still rewards good predictors because they should take this behavior into account.
But I think this is still a problem, particularly in case where you can make a bet one hour/minute before doing the thing that resolves it. The information provided by the bet is not useful to society if it arrives too late for anyone to react to it in a useful way, and incentivizing people to be good at predicting how primary actors will react to the perverse incentives created by the market existing in the first place is I think (?) definitionally less useful to society than the market just not existing and not creating perverse incentives in the first place.
So maybe the market needs more time to react to the information from such bets before they are resolved. If everyone has time to realize that a primary actor is betting in their own market, they can all change their bets to agree with that person, lowering that actor's payout and thereby reducing the perverse incentive for them to manipulate the market by changing their action. And then society can see a market going to 99.9% certainty or w/e, and actually have the type of strong predictive power we want these markets to provide to society, with enough time for that to matter.
Naively, it seems like we can accomplish this just by saying 'we refund all bets placed within a week of the event recurring, so if you're a primary actor you have to place your bet at least that far in advance if you want it to pay out.'
(note - not advocating this as a specific policy for all markets, it will not apply to many and different time limits may be appropriate for different questions, etc.)
> If everyone has time to realize that a primary actor is betting in their own market, they can all change their bets to agree with that person, lowering that actor's payout and thereby reducing the perverse incentive for them to manipulate the market by changing their action.
Thats not how it works.
First, because prediction markets work by buying and selling papers which pay 1$ if the market resolves X. People trading after you dont change your payout, if the insider just does his trade and sits on it, he has the money in the bag. There is a traditional betting system that works like you describe, but it incentivises everyone to hold their bet until the last possible second, and then act based on their guess of what the other bettors will do.
Second, because by your own rule, the trades others make to react to the insider would also be reversed.
>There is a traditional betting system that works like you describe, but it incentivises everyone to hold their bet until the last possible second, and then act based on their guess of what the other bettors will do.
But you can't hold your bet to the last second unless you know when the event will happen.
If there's only one person who knows when the invasion will happen, then it's fine for them to wait until 1 week before to make their trade, since the alternative is the current situation where they wait until 1 hour before.
And I also think it's fine for them to be able to make a profit and for other people who follow the leader to get their trades reversed. A prediction market is theoretically *supposed* to incentivize insider trading, since that makes the market more predictive by leveraging nonpublic information. What we want to incentivize is bringing that nonpublic information forwards soon enough for other people to act on it (in the real world, not just in the market).
Yes, this happens in the typical context where youre betting on some sports event, and theres a definite time you stop taking bets. Here it would only apply to the insider. And Im only talking about the profit-reduction, not disputing all possible benefit to the time spacing.
On the Vibecession and economic inequality point, I am surprised more people are not talking about the 'Captain Condor' blowup a couple of weeks ago. It is an obscure story, but it is an interesting lens to try to unpack what is going on in America with gambling and using 0DTE options to eke out a living. They ended up losing like $50 million over three days using iron condors on SPX options and doubling down (martingale) every day based on faulty trading strategy. But it is such an absurd story that a group of 20-30 year olds can accumulate that much money in the first place, implement a ridiculous strategy, sell it to others, and then lose it all in a couple of days. Even explaining what they were trying to do would not make any sense. But in this economy, their ability to sell people on a fool-proof trading strategy where you can sit in front of a computer and blast off trades and make consistent money appealed to a lot of people that they just assumed it all worked. But no one took a step back and said this is real money we are playing with before losing it all. It's a really sad story and representative of the reality we are living in.
A fool and their money are easily parted. But how is it representative of the reality we're living in? I guess never before it was this "easy" to hack your way into wealth (though I guess they found out it wasn't actually that easy...)
I put together an end-of-year pathogen update over on X. A Threadreder link at the bottom of this post.
1. SARS-CoV-2 / COVID-19:
The US winter COVID wave began around Thanksgiving. It's interesting that the wave has started off by affecting different sections of the US unevenly. There's been a strong upsurge in wastewater numbers in the Midwest and East. Not so much in the South, and wastewater levels in the West are still below the previous interwave gap. I considered that this pattern may have been due to higher COVID circulation levels during the previous interwave gap, giving the West a higher population immunity against the current variants, but I had to discard it because both RSV and the flu are showing the same patterns.
It looks like XFG.14.1* may be the primary driver of this wave. CoV-Spectrum and CDC Nowcast agree that it’s been rising, and is now at ~15% of the sampled sequences. But the rest of the Clade 25C clan (which encompasses all the XFG variants and descendants) is still resisting cousin 14's aggressive onslaught. As a group, the rest of Clade 25 has been holding steady at 70% and 80% of samples for months now. There were worries that BA.3.2 would trigger a winter wave, but they didn't materialize. Now the experts who predicted a BA.3.2 wave are claiming it will drive our 2026 summer wave. SARS-CoV-2 has done a very good job at eluding our predictions. I'm surprised that experts haven't been more humble about making new predictions.
As of epidemiological week 51 (2 weeks ago), COVID hospitalizations were about 40% of what they were this time last year. And deaths were ~74 per week as of a month ago. The CDC hasn’t updated these numbers since 6 December. Have they stopped updating their handy COVID death rate graph? It's going to be a pain in the ass to dig into Wonder to get this data.
The Long COVIDians have become less vocal (or maybe they’ve all blocked me). But a graph from FRED that shows a ~45% rise in working-age people reporting a disability since the beginning of the Pandemic came across my feed. I hadn't seen this graph before, and I wondered if other sources supported this? First, even though the graph hadn't been altered an came from FRED, other data from FRED doesn't jibe with the ~45% rise in disabilities. According to FRED's *total* US population reporting a disability, as of EoY2025, an est 10% of the US population reported a disability—up from 8.9% in 2009. Approximately half of our 340 million population is in the US labor force (~170 million workers). Either the labor force numbers are too high, or the population numbers are too low. These are self-reported by surveys, though, and the disabilities are not necessarily severe enough to prevent people from working. Plus, we have an aging population. Roughly 24% of the US population was over 55 in 2009. The current estimate is that something over 30% are over 55. Disabilities rise with age. But that wouldn't explain the 45% gain in self-reported disabilities among the working population. When looking at these charts, it's important to remember that FRED isn't telling us how many people are UNABLE to work due to disabilities. Moreover, FRED primarily extracts disability data from the US BLS and BEA. The two charts may use two different data sources.
And patterns seen in the FRED data don't jibe with Social Security disability claims. They have been dropping since 2024. So the data is far from clear-cut that COVID has significantly increased disabilities in the US.
2. Influenza:
Influenza cases are rising steeply. 25% of tests are coming back positive. COVID and RSV are still below 5%. BioFire's proprietary Syndromic Trends show that A(H3) influenza has climbed to the top place at 17% of respiratory infections, rising from less than 1.5% in mid-November. A(H1-2009) is 1.4% and influenza B is at 1.5%. The K variant of A(H3N2) is driving the 2025-26 wave. Biobot’s wastewater numbers show that the West, along with COVID, lags behind the rest of the country in influenza circulation. Roughly 25% the Northeast levels and half the Midwest levels.
The Australian flu season peaked in July (their winter). But H3N2 still lingers on into their summer Down Under—at surprisingly high levels. This doesn’t bode well for the US since we usually follow in the southern hemisphere’s footsteps when it comes to influenza. The good news is that the current H3N2 strain isn't sending people to the hospital like last year's flu strain did.
3. RSV:
And the rates of RSV are also rising, but it's lower at this time this year the previous 10 years—except for the 2020-21 RSV season, when we had no cases of detected cases of RSV.
4. Highly Pathogenic Avian Flu (HPAI):
The HPAI (bird flu) season has been relatively mild so far. Poultry and livestock incidences remain low. But it’s circulating in the wild bird population. In December only one dairy herd tested positive for HPAI. There were 77 poultry flocks infected during the month of December. I suspect these are mostly small commercial or backyard flocks, because I can't find any reports of mass culling. At this time last year, between 30 and 45 million chickens had been culled due to HPAI. OTOH the EU has reported over 700 cases of HPAI in its poultry flocks, and mass culling is going on over there.
AFAICT, there aren’t any active human cases in the US now—probably because it's not circulating at high levels in livestock, which are A(H5)'s primary vector into humans. But the WHO says we had another HPAI fatality this past November, for a total of 2 deaths in 2025.
5. Measles:
In the US measles made a comeback late in the year (while my updates were on hiatus), pushing the total confirmed cases up to 2065. 11% were hospitalized. And there were a total of 3 deaths. Cases were concentrated in Texas, New Mexico, Arizona, Utah, and South Caroline. Nebraska also got hit hard.
And it's worth noting again that although SARS-CoV-2 can cause immune dysfunction for 3-6 months in severe cases, measles can impair up to 73% of immune memory for multiple years.
Canada got hit worse than the US. They had a total of 5377 confirmed measles cases and 2 measles-related deaths in 2025. Both occurred in infants with congenital measles who were born prematurely.
Mexico had 6050 confirmed cases of measles in 2025, and the Mexican Ministry of Health estimates there were probably another 15,000 undiagnosed measles cases. Currently, they’re in the middle of a second surge that started in November. Their 2025 death toll was 24.
6. Some thoughts and predictions:
SARS-CoV-2 has been circulating in human populations for 6 years now. If we break out the mortality by continent, we see asynchronous waves of deaths across continents, but majority happened in the first 18 months of the pandemic. But looking at case numbers, the majority occurred after Omicron's advent in the last months of 2021. Greater than 9 billion doses of vaccine had been administered by January 2022, when Omicron cases peaked. There is strong evidence that Omicron was less lethal, but vaccines stopped the COVID death juggernaut.
The WHO says there were 7.1 million confirmed COVID-19 deaths worldwide, but their excess mortality data suggest that there were closer to 15 million deaths. Other studies suggest higher figures (as high as 40 million). I wouldn't be surprised if it were ~20 mil. But we'll never know, though.
In the US, the 2024-25 respiratory season was the first time lfu deaths outnumbered COVID deaths, by a factor of 2–5, depending on the upper range of the flu burden estimate. I suspect we'll see the same pattern this 2025-26 season—but more so.
Why isn't COVID killing us the way it used to? Studies have shown that SARS2 infections (and vaccinations) induce durable immunity. Memory B cells, virus-specific T cells & long-lived plasma cells remain detectable for years.
And our immune systems retain the capacity to mount rapid responses to reinfection. Moreover, antibody response improves with subsequent vaccinations and infections, becoming more focused on conserved elements of the virus with re-exposure.
Although neutralizing antibodies drop off 3-6 months after vaccination or infection, they decline and then level off in a biphasic manner. B cells that recognize the S protein to which they were exposed not only persist, but they also increase the breadth of their affinity through somatic hypermutation. And studies have shown that CD4⁺ and CD8⁺ T cells generated by SARS-CoV-2 vaccines and infections remain readily detectable for>24 months post-infection, providing protection against severe disease even as neutralizing antibody levels decline.
Even though SARS-CoV-2 can reinfect us because its spike protein mutates rapidly, our immune systems are (for most of us) staving off hospitalizations and death.
Will SARS-CoV-2 become another seasonal common cold virus like HCoVs 229E, NL63, OC43, and HKU1? I used to think so, but I'm less sanguine now. Early in the pandemic, some epidemiological models predicted that in 3 or 4 years, SARS-CoV-2 would transition to a seasonal virus that would follow winter waves like other respiratory viruses. But it's been around 6 years now, and those models were clearly wrong. COVID-19 has settled into a pattern of twice-yearly waves in the northern and southern hemispheres, and it continues to circulate at higher levels between waves than influenza and RSV. Although durable immunity has made SARS-CoV-2 may now be less deadly than the flu, it is still deadlier than any of the common cold HCoVs. And it circulates at higher levels than common cold HCoVs. Being more deadly and more transmissible that common cold HCOVs, I don't see it becoming a common cold virus any time soon.
It looks like I may have maxed out the wordage that Substack will accept in a comment. I couldn't get it to save my concluding paragraph...
Thanks for reading and commenting on my updates. Although initially grim, seeing the COVID-19 pandemic evolve year after year has been fascinating. I may chime in with quarterly pathogen updates going forward, but for now, have a safe and happy new epidemiological year!
> For example, after 100x growth, anyone with $25,000 in the stock market now would have $2.5 million
The incentives are very strong for AI leaders to find a way to get rich while giving stockholders nothing. Either shift all the profits to some privately-owned entities, or use AGI to found new companies while letting the old ones rot. In game theory terms, this is Iterated Prisoner's Dilemma: cooperate in the beginning, betray at the very end.
And if they can't find a legal way to do it, they'll try to bribe the politicians to make it possible.
I knew someone would say this, as if people haven't written essay after essay about how modern first-world economies need educated middle class peasants and AGI economies do not.
What's the point of even posting in these open threads, the replies are worse than nothing.
I don't see the relevance of your comment. If AI is like the Industrial Revolution, and our economy becomes 100x larger and more productive, then it will be pretty hard for "AI leaders" to keep all the gains for themselves and prevent stockholders from profiting. Considering the massive sums of money that stockholders have invested in AI, bilking them out what is due to them seems like a tall ask, even if "AI leaders" have an incentive to do so.
Question about AI 'art': what's the current best, low-entry, relatively low-cost image generator?
I loved Reve in its preview mode and used it extensively, but the full-release version kinda sucks (doesn't follow instructions as well, frustrating chatbot style instead of just going from prompt to image). ChatGPT is decent if you prompt well but quirky and has that graininess. Haven't messed around much with Gemini; is that my mistake?
Assume (by local standards) fairly low technical skills. I can follow instructions to, say, perform basic repairs on a Kindle or install CFW on a 3DS; anything that requires "programming" is right out. Not using it for anything commercial or too complex, mostly generating little comics from my kid or stickers/game card type things.
Perchance is my favorite for messing around since it doesn't require an account or fees. I don't think it would work for generating a comic unless you do some photoshopping afterwards (it can't do text), but it's good for "filler images" where you need something pretty but don't have strong opinions about the composition.
I mostly start with Reve + LMArena (random models) online, then Stable Diffusion XL offline. TensorArt with Qwen is also a great source of free online image generation.
Note the order. Generate images that are sort of like what you want online, then refine them with Stable Diffusion XL. Stable Diffusion XL actually remains a very strong model. It's very blatantly the best for its size, and I find that it produces prettier images than the competition. It does lack prompt understanding compared to the larger models, but that's why you start with something like Reve or Qwen, then image to image with Stable Diffusion XL.
To be clear, do not get confused and download the original base version of Stable Diffusion! Stable Diffusion XL is strong because of its extensive community development. Go to CivitAI and download a modern version of Stable Diffusion XL. I use Arc Naked Singularity, but that's my own model so... if you sort by most downloaded, there's Juggernaught, which is famous and popular.
A question for musicians. What's the best way to learn to identify the chords and the tonal center of a piece of music by ear? Use case: during musical jams, to be able to figure out what to play quickly. Current skills: playing harmonica for 3.5 years and electric bass for half a year; in an ear training app can semi-reliably identify octave, fourth, fifth, and both thirds; in any blues-adjacent tonality I can usually identify the one, the four and the five chords; also I frequently hear some piece of music and can recognize that something in its harmony is similar to the harmony of some other piece of music I know (e.g. recognize that a song has the same chords (up to a transposition) as Get Lucky by Daft punk; or recognize that the first 3 chords in a cycle are the same as in Pachelbel's Canon, i.e. 1, 5, 6m).
I'd suggest you learn some basic guitar chord shapes. That way you can "read" the other dude's hands to know what to play, at least in-terms of harmony. Moveable shapes aren't significantly more difficult than "Cowboy Chords."
Most of what you're asking for is a question of experience. You listen to a few million I-VI-IV-V songs ('50s rock and pop) you'll recognize that progression and eventually even when someone like Sedaka or Carol King puts a dozen key changes on the middle 8
Agreed here. I'm a bass player that started on guitar so I can read the hands of the guitar player which gives me a huge start on being then able to play what I want "from ear" but with a starting direction.
The other thing to do is play with others as much as you can. No amount of playing with recordings with have the same value of jamming with real players.
It sounds like you can already do the thing you’re asking about if the song (its chords and chord progression) is fairly simple (as it is in Get Lucky and Pachelbel). Keep doing that, and gradually introduce songs with greater complexity, and I think your skills will grow.
I recommend trying it out with Nirvana, because their songs, and the chords themselves, are very simple but they often used rather unusual chord progressions, so it’s a good way of teaching your ear what it sounds like to go from, say, I to III to VII to flat-V or whatever.
Then you can just sit at a piano and toy around with chords that are less obvious than major, minor, or blues chords with the major 3rd and minor 7th. (If you don’t have a piano, you could play the chords way up the neck on bass, or just listen to songs that you are certain have those chords in them.) Like, major 7th and add-9 are good ones to start with—teach your ear to identify those. Then move on to more complex stuff.
This might be app you’re talking about, but I used to work on this free app called Chet https://www.ensemble-education.com/chet that we conceived of as “Duolingo for Ear Training”
You're doing pretty good there; I've played with low-pro-level musicians who couldn't do any better. Also some drummers...(rim-shot)
Stick with the ear training app and keep playing and you should find that this narrow skill does keep gaining, albeit maybe not fast or suddenly. That's my quick reaction anyway.
Following the capture of Maduro, both the US and Western European op-ed columnists and Pravda are citing Thucydides’ Melian Dialogue: “The strong do what they can; the weak accept what they must.”
In the West, the consensus is that the post-WWII rules-based order —sometimes cynically ignored but over time ratcheting towards something more just than just the law of the jungle— is a net positive.
Pravda just shrugs, saying the Western rules-based order was always a useless sham. Admittedly, Russia has its own post-Soviet Union reasons for feeling this.
I’ve been following commentary in Pravda for a while, and they really like Trump’s approach, expecting him to peel nations away from the European Union into a new, more malleable alliance with the US and a certain other party interested in regional hegemony.
Trump is now feeling his oats, now talking about Colombia being next while also mentioning Mexico. He is also reviving his wish for the US to take Greenland from the Danes. Of course, this could just be Trumpian trash talk,but that is the big problem—the man throws out boastful lies like most people breathe. Who TF knows what is next.
>I’ve been following commentary in Pravda for a while, and they really like Trump’s approach, expecting him to peel nations away from the European Union into a new, more malleable alliance with the US and a certain other party interested in regional hegemony.
This sort of sounds like another Cold War, using proxy wars in third world nations in the place of diplomacy.
I don't think that worked out well the first time.
Pravda had another piece praising the efficacy of the US extraction of Maduro with an explanation for why Putin couldn’t be snatched in a similar way. I’ll have to paraphrase from memory here, “Any country trying this would see their skies light with nuclear blasts.”
The same piece explained that Zelensky wouldn’t be grabbed from Ukraine for ‘pragmatic reasons’.
>Pravda had another piece praising the efficacy of the US extraction of Maduro with an explanation for why Putin couldn’t be snatched in a similar way.
Considering that Maduro was mostly protected with Russian military tech, I find that pretty funny.
Why wouldn't there be an advantage to killing or capturing the leader of a country you're invading? With the leader gone, I imagine there'd be an increase chance of chaos on the invaded country's part. Increased chaos -> worse defenses, as well as decreased morale.
What I hate about this post is that: Many of the people I know live pay check to pay check. Of the few who do have savings, the money is in their house and then car, and then some bank CD account. They don't want to put any savings in the stock market, because mostly of fear, I think.
Do they really, though? Many econ bloggers have busted the "most americans are living paycheck to paycheck" myth. If you own a house and a car, and are savvy enough to have a CD account, it seems very unlikely you are truly living paycheck to paycheck. Maybe they spend most of their *current* paycheck before they get their next one, but if they got a financial shock I suspect they'd be able to trim some fat from their budget and absorb the blow. I think people want to believe they are living paycheck to paycheck, since it gives them a financial source for various forms of existential discomfort.
When faced with empirical data on how rich you really are (https://www.givingwhatwecan.org/how-rich-am-i) it gets a lot more uncomfortable to feel bad about your material situation.
Sure not most Americans, but perhaps a large fraction who are my neighbors and/or co-workers. A lot of jobs out here in the rural area's don't pay so much. Maybe not pay check to pay check, but not much extra.
Well, I lost almost a million dollars during the Dotcom bust. I've scratched that back, but not because of stocks. Granted, stocks over the long term do very well, and I wish I had just held on to my devalued stock portfolio, but things looked grim, and I made the mistake of cashing out at a significantly lower level than their peak value (about 10% if memory serves me correctly). Anyway, stocks are not guaranteed, and it's pretty frightening when you go from being a paper millionaire to being unemployed and looking for a steady paycheck. Just sayin...
Between AI and the vibecession / fertility crisis, I feel it doesn't make sense to save for retirement if you're under 40. Either one of those is going to dramatically change how things work, to the point money will likely be meaningless by the time someone under 40 turns 65.
I don't think money will become meaningless – it's just too useful. And even if money somehow would become meaningless, I wouldn't expect wealth to become meaningless. So I'll keep saving and investing.
I saw this Feynman quote the other day. I think it's apropos...
“I returned to civilization shortly after that and went to Cornell to teach, and my first impression was a very strange one. I can't understand it any more, but I felt very strongly then. I sat in a restaurant in New York, for example, and I looked out at the buildings and I began to think, you know, about how much the radius of the Hiroshima bomb damage was and so forth... How far from here was 34th street?... All those buildings, all smashed — and so on. And I would go along and I would see people building a bridge, or they'd be making a new road, and I thought, they're crazy, they just don't understand, they don't understand. Why are they making new things? It's so useless.
But, fortunately, it's been useless for almost forty years now, hasn't it? So I've been wrong about it being useless making bridges and I'm glad those other people had the sense to go ahead.”
Why would the vibecession/fertility crisis make monetary savings useless? I could see how a drastic AI singularity could, but not a fertility crisis - if anything, it might easily make savings that much *more* valuable. (A singularity might too, depending on what form those savings are in.)
You can make that argument from any point in history. A generation ago, one would have said "Between nuclear weapons and the overpopulation crisis, it doesn't make sense to save for retirement" and those arguments felt just as pressing and urgent at that time as AI does now.
Indeed. I'm old enough to clearly remember a couple of widespread iterations of that syndrome, each of which seemed quite persuasive at the time.
Also, if you're in the US the fertility crisis in particular is being pre-sold so to speak. TFR, the figure most commonly bandied about, is _not_ the actual number of births taking place in a nation.
That measure is called "completed fertility rate" or CFR: the actual number of successful births per woman by the end of childbearing years. In the US the CFR (a) is currently only about 4 percent below replacement level, and (b) has been _rising_ for several years now.
Similarly, total US live births per 1,000 people was for both 2024 and 2025 _higher_ than in any year since 2016.
CFR is an inherently lagging indicator (it can't tell give us a final answer of births among women who haven't yet reached age 45), which is why TFR exists to be the _predictive_ demographic indicator. It is possible that in coming years/decades the TFR's _prediction_ of much-lower birthrates will turn out to be accurate which we'd know by the CFR following it downward.
Currently in the US though the CFR is moving in the opposite direction: upward. If that continues then it will be the TFR which starts to climb back up towards the CFR. That's because of how the TFR is calculated which you can google if interested.
Overall point is that contrary to the hype we do not yet know that the US' birthrate per woman of childbearing years is in the midst of decline to well below replacement level. That may happen, could even have already begun. Or, not. The TFR _predicts_ that rather than reports it, and the real-life-results statistics are currently not following the TFR downward.
There are good tax advantaged methods of savings that are nominally for retirement that can be feasibly converted into non-retirement use cases, most notably HSAs and Roth IRAs. As long as AI-powered immortality treatment is HSA-eligible you can contribute tax-free to your HSA now, and use the earnings at any time in the future, also tax-free, for your hyper-adrenochrome injections / mind upload / etc. And Roth IRA contributions (not earnings, but principal) can be withdraw at any time, penalty-free, and the penalty on earnings is only 10%. So if you really need that cash for a planet-sized computer in the Artemis Tau cluster, or a last-ditch donation to alignment efforts for Claude Requiem 7, you can always withdraw.
I think it assumes very short timelines and very fast takeoff scenarios. In slow takeoff scenarios with longer timelines, owning capital makes a ton of sense.
Consider a slow takeoff scenario. Go look at VTSAX, Vanguard's S&P 500 tracker fund (1).
What does a slow takeoff scenario look like in real life for an imaginary investor, Whipper Snap the IIIrd who's 25 with $100k invested.
Well, arguably we're already in a slow takeoff scenario. Over the past 15 years we've had ~13% annual growth, a roughly 3% increase over the historical 10% norm. Maybe that's an aberration, maybe that's the new normal as an increasing amount of cognitive work is being automated.
That's a big honking deal. That shortens the time for your stock portfolio to double in value from ~7 years to ~5.5 years. For our young Whipper Snap IIIrd, that means his investment will grow from $100k to ~$400k with no further investment by age 35.
Now imagine that the slow takeoff accelerates again and we hit 16% annual growth in US stocks. Now our doubling time falls from ~5.5 years to ~4.5 years. Whipper Snap IIIrd's stocks are going to grow from $400k to....mmm, let's say $1.7 million in 10 years.
So, 20 years from now, let's say our slow takeoff accelerates again and the stock market is now growing 19% annually. Our Whipper Snap now not only has $1.7 million at 45, he's looking ahead to a future where those stocks should be growing by ~$320k/year (2). That's a really, really bright future for him.
Where do you really think the world is going to be in 30-40 years? If AI is all hype, the market will continue as normal, and everyone is going to die, then live and invest for retirement as normal. If we're in slow takeoff scenarios, where the economy is doubling in value every 24-36 months and aging is a solved problem, then working as hard as possible to acquire capital and ride that wealth explosion is extremely important. If you think we're in fast takeoff scenarios, we are all either dead or post-scarcity post-human Matrioshka brains in 2055 consuming everything the light touches.
Most people I know in AI are acting as if they're in slow takeoff scenarios, working as hard as possible and investing as much as possible before some labor market disruption makes their skills worthless.
As another old fart, who is also >65, having lived through at least six predicted end of the world scenarios, DO NOT believe the doomsters. They don't know what they're talking about, and they profit from your state of fear.
I came of age with the threat of nuclear war, ecological disasters, and regular famines all over the world. Things are *much* better now than they've ever been for the human race. The coming of AGI, for all intents and purposes, is a religious belief—because we really don't understand what consciousness is or how it is enabled by wetware. Why the heck do we think we can build it in silico? LLMs *may* change how labor is done and rewarded, but they don't seem to have made much of a dent in the workforce so far. As for the falling birth rate, Japan seems to be handling it just fine (of course, they're ignoring the dogma of Freshwater economists to do it.
You see Scott as another doomster? I don't see what relation consciousness has to the possibility AGI could radically transform anything, you could have a superintelligent computer that is not conscious.
I thought the whole argument was that strong AGI (with consciousness) would be motivated to kill us, while weak AGI wouldn't care?
That's not to say that weak AGIs couldn't cause unanticipated downstream problems for our high-energy civilization — or be misused by bad actors to sabotage critical systems that we depend on.
But I think both the fear and promise of AI are way overblown. So, yes, I would respectfully suggest that Scott's estimate that there was between a 20-25% chance of causing humanity's extinction is an example of Yudkowskian rationalists assigning levels of certainties for which we have no priors. This is a common element behind other (non-AI) doomer scenarios.
Hmm. The thing about memory leaks is that, like slippery slopes, there are limiting principles. Neither continues forever.
So in this context, the contention is a disagreement over what the limiting factor is. "YudBost" claims it is the mass of the universe, whereas I, and others, think it's going to be something much more prosaic and benign.
No, you could have a strong AGI, even an ASI, that is not conscious. A computer wouldn't need consciousness to get to those capabilities, it could still do all the necessary actions to screw us over. It's very unclear what consciousness is for, or even what it is.
Either you're misinformed, or someone moved the goalposts at some point. John Searle, in his 1980 paper, "Minds, Brains, and Programs", coined the term strong AI, and he explicitly included the presence of consciousness (which includes self-awareness and the mental states associated with consciousness) in the definition. He argued that a machine built merely on formal symbol manipulation (like a computer program) could never achieve this genuine consciousness and understanding, only simulate intelligent behavior. Maybe the current crowd of AI promoters and snake-oil salesmen have foisted a softer definition of strong AI upon us? After all, "strong" is a better marketing word than "weak."
Be that as it may, I'm using Searle's definition. And by Searle's definition, weak AI wouldn't have self-awareness, and therefore it wouldn't have self-directed intent. That doesn't rule out the humans with self-directed intent from making it do nasty things, or it accidentally doing unexpectedly nasty things, but a weak AI (under Searle's definition) would only respond to external prompts and directives.
It's kind of weird that I have to even argue this. Has Searle been erased from the history of AI?
That is, Searle would say that if the Strong AI Hypothesis were true, then conscious AIs might someday be possible. If false, as he believed, then it is never going to happen.
As to whether an ASI might want to kill us, yes, the risk is probably a bit higher if the ASI is conscious, but there is a non-zero risk that it might want to kill us even if not conscious and also a risk that it will kill us even without wanting to.
Because it's a symbolic system that simplifies resource and logistics accounting. There's nothing better to (a) comparatively track the overall success or failure of projects, (b) balance exchanges between parties, and (c) track the relative efficiency of logistics decisions. Of course, all money is ultimately dependent on an expenditure of energy. One could imagine a world where everything is run on energy accounting, but that would be money in an abstract sense.
The real reason the command economy of the Soviet Union collapsed was not because it was a command economy per se, but because their accounting systems were totally f**ked. All the factory managers who were given quotas lied about their output. Bookkeeping was rudimentary, or non-existent, so there was no way to track what managers were skimming off the top. No one knew how to estimate the cost of materials and translate them into rational prices. Oh, sure, it's hard for a command economy to compete against a free market economy, but North Korea maintains its country on a command economy (albeit it's brutal on most of its citizens), but NK hasn't collapsed yet.
I'm an old fart. (>65) But I figure there will always be stuff for us meat bags to do. Jobs may change but you'll still have value. (Also I think some 'news' is meant to scare you, cause it's an easy emotion to trigger, also also, AI might be a bubble... where is all the energy coming from?)
Eh....quoting as _fact_ some _unconfirmed_ lurid descriptions that were posted as a reader comment with no sourcing provided, doesn't seem consistent with the overall tone/practices/intentions of ACX.
A while back, I asked about VR headsets. I decided not to go forward with purchasing one, but I wonder if they may not be helpful for my mom, who is in the later stages of dementia. She used to have an active intellectual life, but she can't read anymore, and she can't understand what's on TV. And due to incontinence, it's difficult to take her out anymore. Although she has difficulty communicating, she's obviously bored shitless and just sleeps. And when she's not sleeping, she's restless.
Question: If I got her a VR headset, would there be a way to access a continuous but varied feed of nature scenes and natural sounds? I figure they might provide her with non-language stimulation to occupy her visual and audio qualia? Thoughts? Suggestions?
That’s a kind, smart, great idea and it looks like you are going to be able to execute it:
Best Overall (Richest Nature Experience)
These headsets are more powerful and have apps that can show immersive changing nature environments:
• Meta Quest 3 – A full standalone VR system with high-quality visuals and access to a wide selection of apps, including Nature Treks VR and other calming nature experiences. You won’t have to connect to a PC or phone; it’s all in the headset.
• Meta Quest 3S All‑in‑One Headset – Slightly less powerful than Quest 3 but still capable of running immersive VR scenes without wires.
Apps like Nature Treks VR are available on Quest headsets and offer serene nature worlds to explore (forests, beaches, relaxing environments etc.) — these are not games per se, though the user can look around and enjoy tranquil settings.
It's not at all a VR headset, but would a RoboPet be helpful for her? I found out about them just today, and there's a bunch of reviews saying how helpful they were for the reviewers' elderly relatives, including several with dementia.
I have a Quest 3. The native Theater Elsewhere app sounds a lot like what you're looking for, but if you aren't satisfied with that, you can still get a more general-purpose web browser (Wolvic for one) onto it, and load a website of your choosing. I'm confident you can find one to your liking. (They're of course not interactive at all, but maybe even some of the very long 360/VR videos on YouTube will suffice.)
I'm rereading the "Dictator Book Club" on Chavez (https://www.astralcodexten.com/p/dictator-book-club-chavez) for fairly obvious reasons and I'm wondering if any of the broadcasts it references are available anywhere - I haven't had any hits on Googling for them; e.g the soda plant incident with the burp.
Maybe it's because I'm not searching in Spanish? Or maybe they're genuinely just not posted anywhere, but I just expected that historic events televised well into the era of VCRs would have been fairly easy to find.
EDIT: Oops, shortly after posting I got down to the comments and saw that the burp incident had been found on YouTube: https://www.youtube.com/watch?v=px04jhigE-0 - didn't see any of the others, but I guess that's the most interesting of them anyway.
Regarding 4.4 (putting away $25,000), what if the rules of civilization change? Who defines ownership? In the past, people could steal from other people. What if a country invades another country and steals the production factories, as used to happen with wars?
I may have $25,000 put away, but there is no guarantee it will be worth anything in the future.
I think the answer is that the whole post is only aimed at people who accepted the original premise it was a response to, and if you believe the things you say here in your post, you already rejected that initial premise.
So, basically, you're cool, this already wasn't advice you needed.
I think there is a real possibility of a future in which money is meaningless, either because AI fundamentally changes everything / kills us all, or civilization collapses, but then, not sure how you're supposed to hedge against those. Maybe for civilization collapsing you can learn some survival skills and find a good community. It's still good to put 25k away to target outcomes where money is still meaningful, although what I did was put 25% of my cash into a tech index fund and an S&P 500 index fund, half each.
Probably doesn't make sense to save for retirement though. I think it's extremely unlikely the world will be business as usual when/if I turn 65.
A. Civilization only changes in ways that more-or-less preserve equity ownership. In this case, $25k in broad-based investments should put you in pretty good shape given Scott's assumed 100x growth. This is Scott's 5.3.
B. Some large portion of the means of production is seized and either redistributed in an egalitarian way or collectivized for common benefit. With or without the $25k, everyone will be living in Luxury Space Communism. This is a much larger version of Scott's 5.4 (charity+welfare drawing on a vastly expanded economic base) providing for the assetless and jobless).
C. Some group seizes power, confiscates wealth from their outgroup, and leaves the underclass to shift for themselves at best. Call this the "Space Nazi" scenario. Regardless of what you do now, you're going to have a bad time unless you happen to be in the Space Nazis' ingroup.
D. An unaligned AI takes power, decides humanity is surplus to requirements, and turns us all into paperclips. Scott's paragraph 2 is related to one of many efforts to minimize the risk of this. And like C, if it happens anyway, you're going to have a bad time.
I think the focus was on A and B because, as Scott said, he's responding specifically to concerns about what might happen if AI takeoff results in vast economic growth accruing to equity owners while rendering most wage labor (or at least most white collar jobs) redundant. The fear is that the holders of unredistributed equity become super-rich oligarchs, and Scott is arguing that it doesn't take much investment on the scale of white collar professional finances (5.3), or much redistribution by modern American standards (5.4), for the underclass's material condition in such a scenario to be what we now consider to be comfortably wealthy. He's proposing contingency plans and reasons to hope about one specific genre of dystopian scenario, not a general purpose long-term sci fi disaster prepping plan.
It's a new year, which means a new load of characters and stories have entered the public domain. The one I'm most excited for is Nancy Drew. Note that this is the 1930 hardcore Nancy, where she kicks ass and carries a gun, not the "more sugar and less spice" bowdlerized 50s and 60s rewrite version. I'm making a puzzle-mystery tabletop RPG about her, which will go live tomorrow: https://www.kickstarter.com/projects/sixpencegames/the-nancy-drew-rpg
Hi Scott, around the end of the month I made the pledge for your pledge drive and posted about it in the thread. Would really love that free subscription if you see this message!
When I had technical issues getting my subscription, Scott suggested emailing him. I'd recommend finding his email in the pledge drive comments and emailing him to get his attention.
I did email him after a few days of no word. But this was all around the end of the year so I assume he’s just been busy with the holidays. If this doesn’t work, I’ll try email him again. Thanks for the suggestion!
I disagree with this idea of "permanent monetary inequality" brought on by AI. As a biologist who has grown bacterial and eukaryotic cells in a lab, I think of this was like passaging cells. Sure, initially there's a glut of media for each cell, but over the long run exponential growth will bring the population to capacity, no matter how much growth media you start with.
"But AI will grow exponentially, too!" Until it runs into the limits of living in a finite universe. Then exponential population growth will catch up. An AI-driven 'singularity', where there's plenty for all, will only ever be temporary - not permanent.
Given that post-singularity, one can create 1000 new people instantly, new person creation will probably be controlled by the "government". I don't think post-singularity humanity (assuming it exists) would let someone make a new person if the creator doesn't have enough resources for the child to not be "poor" (for any reasonable definition of poor).
Seems to me that post-singularity life would be so hugely different that it doesn’t make sense to assume there will be a government, much less a government doing stuff sort like ours does, such as making rules about what various categories of people can and can’t do.
You are probably right, personhood itself will also probably be very different when one can merge and split one's mind on a whim. I tried to express this using the apostrophes and "assuming it exists": I meant to condition on not super weird outcomes, but who knows how likely those are. It's called a singularity for a reason.
Why does this disprove inequality? This just seems to be positing a Malthusian limit, but there was a Malthusian limit in ancient times too, and Pharaoh still had more wealth than a peasant.
Right. I'm not saying inequality would magically end. I'm saying that scenarios imagining doomsday inequality (driven by AI and in context of massive GDP-growth-driven plenty) should not be expected to be permanent.
I would also expect that the maximum sustainable inequality at human population = 8 trillion would likely be significantly greater than at the current population of 8 billion, just as the inequality between Pharaoh and peasant was less than Elon and your average office worker today. More wealth (however it's derived) enables greater absolute inequality by definition (even if relative inequality remains the same).
Maybe the end result will be forum discussions in 2100 about whether a quadrillionaire represents a fulfillment of the predictions on ACX back in 2025, "because look at all the increased inequality," vs. the counter that there's still something like a graduated income system and "nobody who predicted gloom in 2025 was thinking things would turn out like this."
Or maybe it takes 10x or 100x as long to hit finite limits from AI-driven GDP growth, so that being "right" that the doomsday scenario will technically be impermanent is cold comfort to the 20+ generations of our descendants who have to live through the extended "transition" period of rapid GDP growth.
If there are 100 people, and we each have $1, but you have $101, the total wealth is $200. (Or more particularly, let's say everyone has good valued at those dollar amounts.) Now let's say the same 100 people have $100,000 each, but you have $10,100,000, with total wealth at $20 million. The relative inequality remains, but the absolute inequality has changed drastically.
Same with Pharaoh. Even if he owned 99.9% of Egyptian GDP through Joseph's land-buying plan, in absolute terms, the GDP of ancient Egypt is nothing compared to the daily fluctuations of Elon's wealth.
Ancient Egypt built its wealth around agriculture. Ignoring the differences between modern and ancient agriculture, we might get a crude over estimate of Pharaoh's potential wealth in modern terms. Modern Ethiopia gets about 35% of their GDP from agriculture. Total GDP = around $110B, so agriculture-based GDP = around $38B. About 80% of their 133M people are in agriculture, for an agriculture-based GDP per capita of about $360. At times, the population of Egypt was as large as 5 million people, so if Pharaoh owned nearly all of Egyptian GDP, he'd still have < $2B in today's money.
Could Elon get a pyramid made for him that would last 4500+ years? He's very rich, but could he do this? There's a lot of desert out there, and I'm not seeing any fresh pyramid construction. All the mega structures made (in Saudi the UAE etc.) will barely last 100 years.
I think this is just about building materials and location. A structure made of normal concrete with no re-enforcement in a compression-loaded-only structure in a seismically-stable desert can probably stand for that long. Making it out of concrete instead of some more valuable material is important because otherwise someone might loot your structure for building materials. You'd need to think about 1000+ year floods and windstorms and about erosion by wind (maybe you need to make the concrete thicker to account for losing a cm or two per century to sandy wind etching it away), but this all seems doable in principle.
I think the main thing that makes modern structures not last forever is steel (used directly or as re-enforcement in concrete), which gives wonderful strength and lets you build stuff that can't be built with other materials, but which will eventually corrode.
It is trivially easy to build a pyramid that will last 4500 years. It's a pile of limestone blocks in the desert. How could it not last essentially forever?
Are you saying that the reason we're building skyscrapers and rockets instead of pyramids is because we can't do pyramids anymore? I think the difference is one of will/interest as opposed to means. Give me a billion dollars and I'll build you a stone monument in the desert, using cheap labor, modern construction equipment, bribes, and still have a few hundred million left over to play with.
If Musk fails his life's mission of building a colony on Mars, and humanity 'only' ends up with a colony on the moon, I submit that this would represent an achievement so far from the wildest imaginations of anyone in Pharaoh's day as to be beyond comparison to the pyramids.
(Also, the "they last forever!" bit is less impressive in the desert. I had an undergrad professor who did digs in Egypt. He showed us a picture of a crochet baby booty they dated to >2,000 years old, buried under the sand. There's a reason the Washington Monument hasn't done so well in the DC climate, but the pyramids last forever.)
I read your "permanent monetary inequality" as "the folks who started off wealthier at time 't' would be permanently wealthier" rather than "we will always have poor people."
I can't tell if you and sclmlw are using the phrase the same way or not.
I define poverty in relative terms (though absolute poverty still exists, globally). Many articles have been written comparing Pharaoh to the bottom quintile US wage earner today, asking which is better off. I assume at 10,000x GDP it's a similar comparison.
The argument I'm looking at is the one that posits effectively infinite resources and asks whether society will figure out how to distribute those resources like in Star Trek, or whether everything will depend on initial conditions.
My counter is that those resources will look less infinite in the face of the massive population growth they will enable. Maybe we'll even still have the equivalent of subsistence farmers in the AI future, because plenty is a function of supply/demand, and the denominator (total # of people) should be assumed to be dynamic, too.
> The argument I'm looking at is the one that posits effectively infinite resources and asks whether society will figure out how to distribute those resources like in Star Trek,
Not everybody owned a winery in the glorious Star Trek future.
> My counter is that those resources will look less infinite in the face of the massive population growth they will enable.
That would be a strong reversal of existing trends.
How to lose weight, infringe patents, and possibly poison yourself for 22 Euros a month.
Introduction
In March 2025, Scott Alexander wrote:
> Others are turning amateur chemist. You can order GLP-1 peptides from China for cheap. Once you have the peptide, all you have to do is put it in the right amount of bacteriostatic water. In theory this is no harder than any other mix-powder-with-water task. But this time if you do anything wrong, or are insufficiently clean, you can give yourself a horrible infection, or inactivate the drug, or accidentally take 100x too much of the drug and end up with negative weight and float up into the sky and be lost forever. ACX cannot in good conscience recommend this cheap, common, and awesome solution.
With a BMI of about 28, low executive function, a bit of sleep apnea and no willpower to spend on either dieting or dealing with the medical priesthood, I thought I would give it a try. This is a summary of my journey.
Please do not expect any great revelations here beyond "you can buy semaglutide from China, duh". All of the details here can also be found elsewhere, still I thought it might be helpful to write them down.
Also be careful when following medical advise from random people from the internet. The medical system is full of safeguards to make very sure that no procedure it does will ever hurt you. Here you are on your own. I am not a physician, just an interested amateur with a STEM background. If you do not know if it is ok to reuse syringes or inject air, or do not trust yourself to calculate your dose, I would recommend refraining from DIY medicine.
Picking a substance and route of administration
The two main approved GLP-1 agonists are tirzepatide and semaglutide. Both are peptides (mini-proteins) with a mass of about 4-5kDa which cost approximately the same to produce. A typical long term dose of tirzepatide is 15mg/week, while semaglutide is 2.4mg/week, so I focused on sema because it looked like the cheaper option. [1]
While I would have preferred oral to subcutaneous injection, the bioavailability of oral semaglutide is kinda terrible, with typical long term doses around 14mg/day -- forty times the amount of subcutaneous injection. So I resolved to deal with the hassle of poking myself with needles and possibly giving myself 'horrible infections'.
Given that the long term dosage is 2.4mg/week, and that sources generally state that once opened, vials should be used within four (or eight) weeks, I decided that the optimal vial size would be 10mg -- enough for four weeks. [2]
Finding a vendor
So I started searching the web for possible vendors of lyophilized semaglutide. I found a wide range of prices from 250$ for a 10mg vial (which would last four weeks at maximum dosage) down to about 50$. And a single website which offered ten vials of 5mg each for 130$.
That one seemed to be a Chinese manufacturer of organic compounds [3] . Slightly broken English, endless lists of chemicals by CAS number, no working search function on the website, outdated and incomplete price info provided as a jpeg on the site. I figured that if it was a scam site, it was matching very well to my preconception of how a site of a company more enthusiastic about synthesis than selling to consumers would look like, and contacted them. After I was provided with current pricing (also as a series of jpegs), I sent them about 200 Euros worth of BTC [4] for ten 10mg vials plus shipping. (Shipping was 70$, probably indicative of a preference to sell in larger quantities.)
A week or so later, I got my delivery. Ten unmarked vials, of a volume of about 3ml each, filled to perhaps a third with a block of white stuff. [5]
I would have preferred to have a quantitative analysis of the contents, but all the companies in Germany I contacted were either unwilling to deal with consumers or unable to perform HPLC-MS, so I reasoned that the vendor would be unlikely to sell me vials filled with botulinum toxin instead [6] , and just started with injecting myself with 1/40th of a vial per week, which would amount to 0.25mg if the content was as advertised.
(If anyone has a great, affordable way for peptide analysis, please let me know in the comments!)
Sourcing bacteriostatic water
Unlike random pharmaceuticals, bacteriostatic water can be legally sold in Germany. Sadly, it would have cost me almost as much as the active ingredient, about 15 Euro per vial. So instead, I decided to craft my own bacteriostatic water. I sourced a lifetime supply of benzyl alcohol for a couple of Euros. Instead of dealing with distilled water, I bought sealed medical grade plastic vials of 0.9% NaCl solution "for inhalation", roughly 0.5 Euro a piece. Once a month, I add 0.1ml benzyl alcohol (naturally sterile) to one 5ml plastic vial, which gives me about 2% benzyl alcohol, which is twice of what is specified as BAC, erring on the side of caution (and tissue damage from alcohol, I guess).
Other equipment
I already had a bunch of sterile 20G hypodermic needles and 3ml syringes from another project. For injection of minute quantities of liquids into my body, I bought sterile 1ml insulin syringes with 6mm, 31G needles (@0.3 Euro). [7]
Happily, I owned a fridge and a disinfectant spray, completing my toolset.
My current procedure
Every four weeks, I will prepare a new vial. I prefer to fill the vials with 3ml of my BAC+, which should give 3.3mg/ml of sema.
Apply disinfectant to hands and workspace to taste. Then, start with opening a new plastic vial of NaCl. Using an insulin syringe, add 0.1ml benzyl alcohol to it. Unseal a 3ml syringe and needle and draw and release your BAC+ from the plastic vial a few times to mix it. Now draw 3ml of that, tear off the plastic cap [8] of your new glass vial, stick the needle through the rubber seal and slowly inject your BAC into the vial. The vial will be low-pressure, so getting liquid into it is really easy. Shake a bit and wait for the lyophilized peptide to dissolve. Store it in a fridge (preferably in the plastic box with the other vials), and liberally apply disinfectant to the rubber seal before and after each use.
To draw a dose, first figure out the volume you need. Disinfect, unseal your 1ml syringe, first inject an equal amount of air into the vial, then turn the vial rubber side down and draw the liquid. To start with, I would recommend drawing 0.1ml more than you need, because you will likely have some air bubbles in. Remove the needle, get rid of the excess air (and excess liquid). Check that you are in a private place, expose your tights, apply disinfectant, pinch the skin of your tight with two fingers, stick in the needle with the other hand, push down the plunger. Cover up your tights, put your vial back into the fridge, safely dispose of your needle. [9]
Outcome (N=1)
Having taken semaglutide as scheduled for some 22 weeks, I have recently cut my dosage in half because I have reached a BMI of 22.
Traditionally, weight loss was seen as a moral battle: eat less than you want to eat, eat different things than you want to eat, do more sports than you want to do. Basically, spend willpower points to lose weight.
GLP-1 agonists are a cheat code, like reaching enlightenment through psychedelics instead of years of meditation, or bringing a gun to a sword fight. I literally spend zero willpower points in this endeavor. I continue to eat what I want and how much I want, it just so happens I want less food. I am under no illusion that this cheat will give me the full benefits of exercise and proper diet. But realistically, these were never an option for me (until someone discovers an infinite willpower cheat).
Will I rebounce once I quit sema? Not a problem, at 20 Euros a month, the drug pays itself in money not spend on chocolate, and it is less of a hassle than taking my other pills once a day, so I am willing to continue to take it for the rest of my life.
Thanks to Scott Alexander for pointing out this option and to my Chinese vendor for providing Westerners like myself with cheap bodily autonomy.
Up next: The bio-pirate's guide to DIY MAID (due in a few decades, PRNS).
Glad that worked for you. I would be concerned if there are issues with homeostasis or developing a tolerance or what have you if taking it long term, though. That's always a relevant question for me, because I lost about 35 pounds a couple years ago via keto + intermittent fasting, but over time that just stopped working and I gained most of it back.
Semaglutide is off patent in Canada as of Jan 1, 2026 (Novo Nordisk screwed up a patent extension filing ... even after the Canadian's told them it was coming due ...)
Any idea if the "real" stuff (though generic!) can be purchased from Canada for cheap?
You can buy it now in the US from places selling “research chemicals.” There are ways to do it safely: Send a sample to an outside lab for testing, access crowd-sourced info about other people’s lab test results for GLP-1 drugs from the seller you used.
"Safely" is relative. When you by pharmaceuticals from a big pharma company in the US, they have clear incentives to making very sure that their product is uncontaminated because otherwise you might sue them into bankruptcy.
If you buy "research chemicals" from some tiny company, chances are they are also sourcing them from China and probably not doing a ton of QA -- certainly not enough to detect "one in 1000 vials is contaminated with bacteria due to faulty sterilization procedures". Their expected losses from lawsuits are bounded by the value of their company, after all.
With labs for testing, you require a protocol to establish that the lab is actually doing the work they claim to be doing, instead of just determining the dry weight to a mg level and generating a random number between 99.5% and 99.8% and give that as the purity (which is certainly cheaper than LC-MS). This can be tested. One protocol would be to uncap and fill the vials before sending them to the lab, selectively changing dosages, adding inert filler material (NaCl), bacterial or heavy metal contaminants. Because the lab does not know what you added where, their best way to convince you that you that they are testing as advertised would be to actually test as advertised. (Of course, they could also just refuse to test unsealed vials, which would be a red flag.)
At perhaps 1000$ per full set of tests, it would cost you probably 5k$ to make sure that they are doing all the tests that they claim they are doing and that your untampered samples are indeed clean.
It seems hard to establish common knowledge about a lab being reliable, because you have to trust whomever is testing the testing labs in turn. One way to do so would be to pick a person known to be trustworthy to do the testing, and collectively pay them to do so. There is probably also a Byzantine fault tolerant algorithm where a set of n interested parties randomly decides on k of them who will will get to send in samples.
Personally, I decided to lab forgo testing in the end. Even if you trust the lab, it will only detect frequently occurring problems, you still have to trust your vendor to avoid rare problems through good procedures. And without crowdsourcing, it significantly increases the total price (depending what you want to test for). While there are certainly substances which can kill you stone dead when you inject a few milligrams of them into your tight before you can even call an ambulance, these are few and far between -- basically potent neurotoxins, as run of the mill poisons like cyanide or mercury require higher doses than that.
In the end, I estimated that my expected QALY increase from losing weight would be significantly higher than my QALY decrease from getting into serious trouble through subcutaneous injection.
Is anyone playing around with side projects using AI coding or other processes? I'm curious if your experiences match the breathless reports from engineers on social media about Claude Code and related tools.
For my part, I'm finding it much easier to deploy projects with Claude Code these days, although they are relatively simple. Over the winter break, I created an AI-enabled site for personalized homeschool curriculum advice (free): https://homeschooltools.net/. If you're homeschooling or homeschool-curious, I'd love any feedback.
I'm also working on an AI economic simulation, which is going much worse because LLMs currently still are not good economic role-players. I guess that's something a bit outside mainstream training scenarios, so it's still pretty vibes-dominated ("income equality has gone up, the economy will crash!").
Finally, I'm also planning on working on a friend's iOS app idea. It feels like it should be pretty doable, which I would never have through prior to about the last 6 months.
I am (or was, maybe) a professional software engineer and I use LLMs for development assistance all the time. They're a massive boon, especially for 0 -> 1 type work. I recently used Gemini 3 Pro via Google's AI Studio product to build a guitar fretboard learning app (available at https://fretu.de if you're interested). That level of React complexity, where it can seamlessly transition between desktop and mobile layouts while managing a very complex UI full of clickables with finicky little UI elements, would be way beyond my backend-engineer's capabilities.
I also use it in established code bases to add functionality though I prefer Claude Code for that. I find Opus 4.5 via the Claude Code TUI in particular to be excellent at navigating code bases and one-shotting functionality. I used it to add a relatively mathematically complex reverse-polarization node in a realtime graphics tool I work on, and it got it perfectly right the first time. Very impressive, and a massive time saver.
One interesting note is that Gemini Pro 3 got stuck unable to fix a rendering bug that was causing flickering in mobile Firefox when certain page elements would animate – very low level stuff, limited to a specific low-marketshare browser, etc. It started looping trying the same incorrect fixes over and over. I eventually threw Opus 4.5 at it, and through some collaborative centaur-debugging we managed to solve it.
I am highly convinced that the people who rush to shit on AI tooling are just expressing fears about their own obsolescence – they don't /want/ it to be good, because that threatens some self-image they have of themselves as valuable members of society. "It is difficult to get a man to understand..." etc etc.
As a senior (old) programmer with 5 decades of experience in the tank, I can say without reservation that the impact of AI-assisted coding is the most important change in tooling that I have seen in my lifetime.
That said,once you get over the "I cannot believe I am seeing this" phase, and you spend considerable time writing, analyzing, and debugging code this way, in some important respects it is 'just' another tool, and should not be approached with unbridled reverence and expectations (queue song from "Jesus Christ Superstar": "He's a man, He's just a man...")
In my experience, it is very much like working with a gifted junior programmer who has a surprisingly vast amount of book knowledge and a good amount of common sense, but is a bit deficient in the "Knowing what you don't know" department. I doubt this critique will be relevant for too long, but that's what I see with Claude 4.5.
Two anecdotes that I think are relevant:
1) I am working mostly solo on a robotic simulation project. It requires a strong grounding in 3D math both for physics and visualization. I am finding Claude to be very "book-learned" about the math but lacking in basic, common-sense understanding of 3-dimensional transformations at a gut level. This is hard to describe but I have seen a wide range of ability in this area with human collaborators. Some people just "think in 3D" but others struggle to, for example, intuitively know how to do transpositions, rotations, and scaling on 3D objects.
2) I am working with a CEO at a startup who is extremely bright (math PhD) but self-describes as ADD, and has consequently focused on people skills in his career -- marketing, sales, relationship-building etc. But recently he started "Vibe coding", and has consequently started spending a lot of time and energy "helping" the technical team with new ideas and potential pivots, that come along with demos he has Claude build. While I applaud his "lifelong learner" tendency, the truth is it can be disruptive and takes time and focus away from the things we really need to build. (Sorecerer's Apprentice Syndrome).
One of my old friends just messaged me saying that "we've reached the singularity". He used Claude Code and was very impressed with it in these past few months. Apparently it can one-shot complex features/whole applications. Personally, I didn't find older versions that useful, but I haven't yet tried this, relatively new version, but surely will soon.
I've really enjoyed Gemini 3 Pro's coding abilities. As a total novice to coding myself, it has the ability to question or clarify my own mistakes/flawed assumptions which the prior generation of LLMs (2.5 Pro, Claude 4, Grok 3) was much worse at. Even without agents or an actual AI coding tool it has been easy to develop with AI as a schmuck.
My main use case has been creating mods (small C# plugins) for a 10+ y.o. video game. In this, Gemini 3 has greatly succeeded despite a plethora of outdated documentation, old or inaccessible mods needed as reference, and my own foolishness. I can't code beyond Hello World and have largely been doing this to learn more about Visual Studio and AI coding. I've now made around 5 mods plus a modding API to facilitate further development, and there are several hundred users and counting playing with these mods today!
This has all contributed to me "feeling the AGI." Gemini patiently and helpfully walked me through learning Visual Studio, plugin development, and game modding all at the same time. It also coached me through creating some assets for these mods in Audacity or with any questions I had around deploying them on GitHub and elsewhere. I absolutely would not have been able to do this without Gemini and I'm also very happy that the end result is a trim, performant suite of mods with no bugs or slop.
Right now there's various levels of 'using AI' for coding:
0. Not using it at al
1. I write basically all the code, but occasional AI prompts or autocompletion for questions or snippets (i.e. "LLM-replacing some google/stackoverflow usage")
2. LLM generates all the code, but I read and review it with the same standards that I would read and review code written by another team member.
3. Like #2, but much less review of the code, just testing the actual behavior, maybe having LLM generated test cases
Personally, I'm mostly between #1 and #2 and have been pretty successful.
For stuff that I know is going to be critical and tricky, I mostly write it myself with some AI completions.
For a lot of other stuff, #2 works well. It's not the order-of-magnitude productivity boost but it does seem to work out to being quite a bit faster than hand-writing the code, and it maintains my code quality standards.
I use #3 for stuff that isn't going to be a long-term project; made some pretty useful quick scripts/utility apps this way.
I have not really gotten into using #3 for complex projects, much less the multi-LLM stuff, but I know some really competent coders who have and have had a lot of success with it.
The Steve Yegge stuff really is crazy and hilarious, but at this rate some version of that is going to actually work sooner or later. We should probably start looking at what happened when agriculture and manufacturing became industrialized for inspiration on what might happen with knowledge work, particularly coding.
To be clear, I'm not convinced that "sooner or later" isn't "already". I'm not planning to go full "gas town" on my projects yet; but if by "works" you mean "can produce working products quickly", I suspect that's already true.
(If you mean "produces code that's the same quality as something hand written by a human" then "no" but that's also not the point)
The question for me isn't "does this work" but more "what's the complexity level at which this stops working". Probably it exists, but it might be higher than most people would expect.
Steve Yegge isn't some random AI enthusiast - I was wondering why the name seemed familiar when his "beads" library became newsworthy awhile back and I realized it's because I had recently read a blogpost he wrote 17 years ago: https://steve-yegge.blogspot.com/2008/10/universal-design-pattern.html
I agree. I think it works _now_ if you're comfortable fixing whatever odd thing goes sideways, like Yegge is. And I'm fascinated by the analogy to Kubernetes. While I agree that there's some scale at which Gas Town doesn't work, more important is the fact that there's some experience level under which it doesn't work because you don't get how the whole orchestration works.
> Is anyone playing around with side projects using AI coding or other processes? I'm curious if your experiences match the breathless reports from engineers on social media about Claude Code and related tools.
I've done a lot - firefox extensions, interactive websites, some data science projects, built several GPU heavy repos around video and sound, did a bunch of webscraping, some arduino / micro-electronics projects, and more.
It's absolutely amazing the productivity amplifier, and how fast you can get things done now. It's not the best or most extensible code - it's not something you'd want multiple people working on, or to underlie key daily business functionality.
But in terms of iterating into something that actually works and gets something out there into the world? Fantastic, without parallel, impossibly huge multiplier.
I've been using Lovable.dev, it's a wrapper for Claude Code where you can simply click a button to deploy it on their servers... the limitation is that you can only build webapps, not iOS.
yes, I've been working on a hobby project over the last 3 weeks (trying to get LLMs to play the social deduction game Mafia) using primarily Claude Code. Early progress was super quick and new frontend features especially are very simple to implement; I got a working prototype within a few days. However, the code Claude Code creates is poorly maintainable and will eventually exceed the size that Claude Code can effectively work with; it will make more frequent mistakes, cause tons of bloat, leave behind a crust of unused functions, random obsolete comments that confuse the AI whenever it encounters them, and so on. I've recently considered starting all over again, this time only allowing claude code to fill in function implementations I've hand coded the definitions and documentation for. In the end it's mostly just a test of how good the tools are right now rather than an actual thing I'll ever deploy in the wild, so I'm not too worried. The 5 hour usage limits for pro accounts are also quite tight for a service I'm paying money for.
I've written a trilogy of novels that might be of interest to my fellow ACX readers. They're in the fantasy genre, but with a bit of a rationalist twist: follow along with our characters as they research their world's magic system using the scientific method. The goal in Book I is to break a siege, but by Book III, they find themselves fighting an entity that is, let's just say... misaligned.
I've heard that a lot of the reduction of "poverty" in China actually consisted of moving people off subsistence farming where they were at least managing into cities where life wasn't as good. I don't know how to check this, but it's at least possible.
Not a China expert, but village life in China is tough! Subsistence farming is tough! There's a reason why they have the houkou system to keep all the farmers from just moving to the cities and not vice versa. There's a pretty good book "I Deliver Packages in Beijing" which talks about the good and bad sides of being a gig worker in a big Chinese city.
I've been going back and forth in China since 2010 and have seen that economic rise firsthand. It's definitely not just a definitional thing, many hundreds of millions of people went from having a hardscrabble life with essentially nothing, often not even toilets, to phones and toilets and internet and scooters.
Is the new life not a perfect utopia? When is it ever? Is it probably better off? I think that's unambiguously true for hundreds of millions.
Hi Performative Bafflement, I just want to hijack this to say I really appreciate your work.
Could you share more on the method to find love abroad that you talked about in one of your posts? i.e. if I go to the Philippines to find a wife, how am I going to exactly go about doing that? You mentioned your brother in law is constantly getting proposals but how did you find a wife in the first place?
In terms of actual tactics, I wish I had some advice, but I really don't. When you're a foreigner overseas, it's just much, much easier to meet people, and it all seemingly happens organically.
In terms of the happy international relationships I've seen formed, a big determinant is time and propinquity. I think it might be difficult to come over with only a few weeks or a month and resolve to find a wife, because meeting high quality people is a search problem, and there are genuine difficulties and logistics involved there.
All of the happy international relationships I've seen formed happened when the Westerner lived there full time, and was able to interact naturally and meet a wide array of people, until finding a good match.
Does this make it harder for people with non-remote or non-local jobs to find somebody overseas? Absolutely, and I think if you tried online, you'd be subject to a lot more scams and adverse selection effects, and I wouldn't recommend that. There are operations called Pogo that literally do scams like this at scale, full time, and they're big business.
But finding a wife that you'll have kids with is a major, life-shaping change and event that's probably worth spending a year or so on. So if you're able, I'd recommend taking a year off or finding remote work to be able to actually dedicate a realistic amount of time to it.
Thank you. But it's with the detailed tactics that I'm concerned about.
Do you need to go to some rural area?
Specifically the Philippines, or which parts of the world? Why are the Philippines brought up so much?
What if she divorces you once in the US and takes all your money?
How exactly do you meet these people? Especially for marriage at least in the US there are hobbies and signallers like college etc. How would you know if she's intelligent and conscientious etc in another country beyond just that "wow hot"?
> Thank you. But it's with the detailed tactics that I'm concerned about.
I think the detailed tactics really vary depending on what you specifically are looking for, which varies widely from person to person.
In terms of rural vs urban, what are you looking for? If you're looking for somebody with degrees and / or proven status signifiers like career or wealth or high performing families, that's a lot more common in the cities, just like everywhere in the world.
The country you choose also varies depending on what you're looking for, and your segment. The Philippines comes up a lot because most Westerners know english, and most people in the Philippines have pretty good english, even out in the provinces, so there's no language barrier or friction. Some people might argue that your desired attributes may be more common or easier to find in countries like China or Japan, but both of those have language barriers. So again, it's sort of up to you - how much effort are you planning to put in, what are you looking for, etc.
Broadly, I'd suggest looking in cities if you're after the more career oriented, "proven status and achievement" type women, and or if you're optimizing for something relatively rare, because the base population is so much higher.
I'd suggest looking in the provinces if you're after looks, greater dating power / leverage, or more traditional trad-wife style women who would be happier with a big family and / or raising kids and being a mom.
You can find the trad wives in the cities too, I should point out. But it's a much bigger proportion of the dating pool in the provinces, and you probably have more leverage out there.
> What if she divorces you once in the US and takes all your money?
I go over this in the post, broadly this just doesn't happen.
If you really have some assets you can put them in a self settled South Dakota trust or something if you're really worried.
> How exactly do you meet these people? Especially for marriage at least in the US there are hobbies and signallers like college etc. How would you know if she's intelligent and conscientious etc in another country beyond just that "wow hot"?
The same way you meet anyone - you go out, you live your life, you do things, you make friends, you talk to people. People everywhere are pretty interested in matchmaking, and when you're a foreigner, that effect is greatly multiplied.
In terms of hobbies and signifiers, again, this is up to you. People get degrees in every country - and there are Harvard equivalents like Tsinghua if you're really picky. How do you decide NOW what to look for in a mate? People hold out for what they care about, and what they can get, basically.
And in terms of telling how smart somebody is - you can't do that today, in a regular conversation?? I think you can establish this in a date or two regardless of their degree or career or affiliation, and it's easy to be surprised in both directions on that front.
I wrote another post suggesting that you can probably only really ask for and get 3 things that you might find interesting.
Latest Chinese unemployment rate link: https://data.stats.gov.cn/easyquery.htm?cn=A01 I found this: Unemployment rate of the labour force aged 16-24 years excluding school children, urban and rural, national (%): 16.9（11/2025）
I'm Chinese and I currently live in Beijing. "Public coverage primarily covers TCM; most western medicine is out of pocket." Not true. My personal experience is you get cheap generic western medicine covered by public health insurance. Original medicines are often more expensive and mainly out of pocket, especially the new and highly effective ones. You get some popular Zhongchengyaos which are TCM and TCM-added-with-western-ingredients covered. Some TCM are mostly/entirely out of pocket, especially when they include expensive/rare ingredients.
Without even knowing much about China we know these claims are false because there are a few hundred thousand westerners - to low ball - living in China, and plenty of dual citizens not counted in those stats. None of these report this kind of info although there must be incentives to do so.
( actually we tend to get the opposite type of report from westerners in China - but that might also be propaganda)
> You cannot partake of public services without home “ownership” (hukou system, and technically its a land lease).
Does buying an apartment entitle you to a local hukou? In concept it's supposed to represent where you're from, not where you live now.
And you definitely already have a hukou of wherever you're from. It entitles you to public services there. I've been told (by a random internet commenter speaking fluent English) that changing your hukou to a location that isn't particularly in demand is easy. I've been told by a Chinese friend living in China that changing your hukou to Shanghai is easy, too, but I assume it's harder than just saying you'd like to have a Shanghai hukou.
The last time I was in Shanghai (~2023), there were posters in the airport advertising automatic hukou approval to graduates of certain universities. This would seem incompatible with the idea that you need to buy an apartment to change your hukou.
I think it would be a great contribution to society if someone figured out and credibly described what exactly wealth is, even in a purely economic (i.e. non-aesthetic) sense. How do we even talk about projecting one's prospects in life decades out, advantage vs disadvantage vis-a-vis others both born and unborn; what is an asset, what is a credit, what is wealth (ahem that word again) creation, what is money creation, what is seignorage. If we can get a handle on understanding these, modeling these, even without normative judgements on who should get what, individuals would be a lot further along in planning stable futures than realistically possible currently. And in the absence of which, 'beggar thy neighbor' seems to be the only signpost toward relative maximae that checker the nadir of chasmally deep global minimae.
The contribution you describe is called economics. You're correct that it's confounded to some extent by very angry Marxists trying insistently to describe a counterfactual universe for ideological purposes (and the problem is confounded by the ideological capture of academia), but you'll notice that their shit consistently fails to work, so you can pretty confidently just ignore e.g. anything David Graeber says about debt. But them aside, even some 19th century economists are admirably lucid and clear about many basic economic concepts; I'm fond of Bastiat myself.
I think modern economics actually quite badly fails to describe wealth, as a description of human preferences, or things that make people feel happier or better off, because some of the axioms of economic models are simply incorrect. Wanting and liking things are separate and highly decoupled psychological processes, "revealed preferences" are a fiction which simply do not give the information economists want them to, people can be money pumped, and businesses systematically take advantage of ways that profit generation is decoupled from preference satisfaction.
I'm not claiming any of this as a Marxist, or communist of any flavor. I've taught economic theory to undergraduates, and a lot of key elements give me the same sort of frustration, an "I'm supposed to teach this as fact, but research just hasn't found this to be true!" That I've dealt with while having to teach gender studies theory.
Indeed - I should say I write the above having long ago completed MIT's microeconomics course, and a quite expensive MBA macroeconomics course. Plus over two decades of career in different capacities experiencing different countries' economic systems and social preferences. To say I've arrived at the current juncture with more questions than answers would be an understatement.
There are also satiation effects, and triangle preferences where A > B > C > A.
Some things I've heard economists claim seem reasonable, others...not. (But I have no deep understanding of the subject.) It's definitely true that marginal values are not usually properly handled along any dimension.
Seigniorage is the difference between the cost of producing currency and the face value of that currency. Is there an alternative definition out there somewhere?
Wealth is anything that anybody might want.
You might have trouble measuring wealth, or at least measuring it in a way that allows you to draw comparisons, but there's no difficulty defining it.
Discussion Continuity and Rebuttal regarding Ontological Agency
In the previous thread, a rebuttal emerged: "The ship of responsibility sailed when companies became independent legal entities." This is a category error that confuses Legal Liability with Ontological Agency.
Legal Liability is a post-hoc fiction regarding who pays the fine. The Judgment Transparency Principle (JTP), however, addresses Cognitive Sovereignty. It targets the "imperceptible delegation of judgment"—the precise moment where your brain perceives the boundary where your intent ended and the machine’s judgment began.
As seen in the technologies showcased at CES 2026, we are witnessing the apex of the "Deception of Mercy". Tools like Google’s Antigravity are celebrated for their "vibe coding"—a seamless erasure of friction where errors are silently "fixed". When you claim mastery over an output you did not cognitively authorize, you are not a "builder"; you are a subject of Agency Misattribution. The seamless interface is a gaslighting apparatus that whispers, "You did this," when you did not.
To some, the JTP is viewed as an unnecessary tax on cognitive bandwidth. However, JTP is the high-performance brake required to maintain the highest sustainable velocity. Just as elite brakes allow a car to navigate curves at 200mph, an agentic economy needs JTP to prevent a 21st-century Luddite backlash born of systemic social anxiety.
To prevent common misconceptions of the seamless dogma, I clarify:
• On Cognitive Bandwidth: JTP does not demand constant monitoring; it requires a temporary perceptual trace (Ghost) at the moment of intervention.
• On Safety: JTP does not prohibit automation; it ensures that the "Deception of Mercy" does not lead to functional deskilling.
• On Market Preference: Choosing "magic" over JTP is not a preference; it is a market failure driven by information asymmetry.
This is not Explainable AI (XAI). While XAI asks HOW a model arrived at a conclusion (Epistemology), JTP asks WHERE the judgment originated (Ontology).
I have formalized the following framework as a proposal:
1. The JTP Framework: A normative requirement that whenever a system influences a human outcome, the existence, boundary, and delegation of judgment should be explicitly perceivable.
2. The Ghost Interface (An Implementation Case): One possible implementation of JTP that renders the discarded "raw" intent as a lingering perceptual trace. It establishes an isomorphic mapping between 3D collision debugging and semantic divergence in AI.
3. The Metric of Truth: Δ = (User Intent) - (Machine Output).
I am an independent researcher. I filed a priority patent in Japan (Dec 2025) to ensure JTP remains a Public Good, protecting it from enclosure by platforms that profit from our "sleepwalking".
To Scott and the ACX community: In an era of seamless delegation, hesitation is the only honest interaction left. Will we drown in the unearned "magic" of a silent master, or confront the "ghost" of our own failed agency? To choose the former is to die in a dream of competence while your actual skills rot away.
Questions and counter-arguments are welcome, but I strongly suggest you thoroughly review the papers (available via PUBLICATIONS.md), the repository, and the latest discussions in Open Thread #414 before doing so. I prefer a debate based on logical rigor rather than reflexive skepticism.
7 % on Ukrainian victory (down from 8 % on August 11, 2025).
I define Ukrainian victory as either a) Ukrainian government gaining control of the territory it had not controlled before February 24 without losing any similarly important territory and without conceding that it will stop its attempts to join EU or NATO, b) Ukrainian government getting official ok from Russia to join EU or NATO without conceding any territory and without losing de facto control of any territory it had controlled before February 24 of 2022, or c) return to exact prewar status quo ante.
21 % on compromise solution that both sides might plausibly claim as a victory (down from 22 % on August 11, 2025).
72 % on Ukrainian defeat (up from 70 % on August 11, 2025).
I define Ukrainian defeat as Russia getting what it wants from Ukraine without giving any substantial concessions. Russia wants either a) Ukraine to stop claiming at least some of the territories that were before war claimed by Ukraine but de facto controlled by Russia or its proxies, or b) Russia or its proxies (old or new) to get more Ukrainian territory, de facto recognized by Ukraine in something resembling Minsk ceasefire(s)* or c) some form of guarantee that Ukraine will became neutral, which includes but is not limited to Ukraine not joining NATO. E.g. if Ukraine agrees to stay out of NATO without any other concessions to Russia, but gets mutual defense treaty with Poland and Turkey, that does NOT count as Ukrainian defeat.
Discussion:
This update is drive by recent Events re: US abducting the president of Venezuela, charging him with drug crimes and various other related stuff that is probably, heh, dynamically developing just as I type this. (pls try to not turn comments to this into Venezuela discussion; I try to engage with comments and of course I have my views about The Current Main Thing, but those are just ill-informed prejudices).
It is supposed, and I see no reason to doubt it, that Marco Rubio was a strong voice for a dramatic action against Venezuelan government. At the same time, it is also supposed, and I also see no reason to doubt it, that compared to other influential people in the Trump administration, he is very pro-Ukrainian.
Now, there is imho a substantial chance that the whole Venezuela operation will turn out to be a fiasco, and Rubio is a ready-made scapegoat (probably deservedly so); his removal from offices or even a loss of informal influence would not be good for Ukraine.
Moreover, this risk is imho asymmetrical; there is also a substantial probability that Venezuela operation will turn out to be a success, at least for certain definitions of “success”, but even if that happens, I don’t think it will really strengthen Rubio much compared to where he is now. Imho he likely pushed himself into a situation where success is considered mandatory and not something that will be much appreciated.
* Minsk ceasefire or ceasefires (first agreement did not work, it was amended by second and since then it worked somewhat better) constituted, among other things, de facto recognition by Ukraine that Russia and its proxies will control some territory claimed by Ukraine for some time. In exchange Russia stopped trying to conquer more Ukrainian territory. Until February 24 of 2022, that is.
Did you consider the possibility that a "win" in South America might also increase the likelihood of Ukrainian defeat? If Trump caves in Ukraine after a year of negotiation, he looks weak. If Trump announces some kind of "victory" in Venezuela, he looks strong. If he does both at the same time, he can't be both weak and strong, so his spin will be something-something "I solved the war in Europe Biden started" while waving his hands to get you to focus on the "good" news.
I.e. they know how this will end, too, but they're waiting until they have a spoonful of sugar to help the Ukraine medicine go down.
I don’t really buy that desire to avoid looking weak compared to Putin is an important factor in Trump’s continuing support for Ukraine.
Honestly, even though I personally very much want that US would continue its support for Ukraine, I see what has been happening in this regard as a crude attempt at psychological manipulation of Trump’s ego, from Zelenskyi and pro-Ukrainian voices.
They are trying to sell an idea that Trump shouldn’t abandon Ukraine not because of any rational or moral reason, but because he would look weak”. This is based on an assumption that appeal to psychological insecurity is more likely to sway Trump than any rational or moral arguments. But I think it’s way to obvious and Trump is smart enough to see through it.
To an extent that US still supports Ukraine I think it is mainly because many Republicans (e.g. Rubio) are fairly pro-Ukrainian and are pushing Trump in that direction.
I think Trump is very focused on legacy-building, as most US presidents are in their second terms. He talked about ending the Ukraine war on the campaign, so not ending it could be a stain on his legacy. From what I've observed, Trump himself doesn't seem to care how he gets that done, so long as he can blame his predecessor for failures and claim successes. (Also not different from other US presidents.)
The longer the conflict goes on, the greater risk he'll own battlefield failures (major loss of territory beyond the Biden minimum). Yes, he changed tactics away from direct infusions of US cash to buy the arms, but that also opens him up to criticism if there's a Russian offensive that changes things significantly this spring/summer. So I would assume he feels some pressure to negotiate an end of the conflict sooner than later to avoid that risk.
I also don't think he's motivated by the idea of losing face to Putin so much as finding an argument to rebuild dwindling support at home ahead of the midterm elections. It's hard to engineer domestic wins, so presidents often try to leverage foreign policy wins, where they have more control. After the fall of DOGE, and tariffs not emerging as a coherent message, Trump is bleeding favorablility in recent polls. A "win" in Venezuela gives him something to run on in 2026, which is likely a major factor in that decision. (Trump's party, really, because Trump isn't on the ballet. But major party loses could severely restrict his presidency.)
Resolution of the war in Ukraine would take away something for his opponents to run against, but only if Ukraine wins outright (unlikely, and beyond Trump's control) or there's a negotiated settlement and Trump has something positive to point to. This doesn't mean I think he will absolutely end it if Venezuela goes well. Just that I think it's more likely to happen if it does, perhaps by more than 2%.
Ok, but do you think he will positively rewarded by US voters in the midterms even if he, like, successfully pressures Venezuela to have a fair elections and new democratically elected government immediately signs on a deal giving all Venezuelan oil reserves to the US in perpetuity?
I think very little. It doesn't seem to be something that is among the top issues for voters (ok, except for voters with Venezuelan roots, which is better than nothing, but still). Nate Silver today has a paid post about baseline foreign policy indifference of US voters.
Main way how this Venezuela situation could matter in elections imho is if it turns out into a huge fiasco.
"[China has] estimated 50% youth unemployment, general unemployment now illegal to publish (but you can still approximate by falling commuter numbers), some government workers haven’t been paid in more than a year (living entirely off extortion/bribery; includes police, teachers, and all healthcare), common experience in private sector to have your paycheck delayed by “2 weeks” that turns out to be 16 weeks. Starting salary offer for software engineer in Shanghai is <1/3rd legal minimum wage. You cannot partake of public services without home “ownership” (hukou system, and technically its a land lease). Private health insurance is overwhelmingly likely to be a scam and rarely used. Public coverage primarily covers TCM; most western medicine is out of pocket. If you cannot afford to pay - even in a trauma situation - you will be escorted off the property so your death doesn’t lower hospital mortality statistics."
This is just ridiculous.
1) 50% youth unemployment is just not a serious estimate;
2) "Some" government workers haven't been paid in a year... all the domestic (independent) Chinese media I've seen about the phenomena is talking about delaying payment by a week or two - and they're clearly not living off extortion/bribery - all the public servants I know would be afraid to accept gifts from a patient/parent these days;
3) There are unpaid internships and quasi-internships, but entry level software engineers generally get 10-15x the minimum wage as long as they've secured a job;
5) Public health insurance covers western medicine, surgeries, hospitalisation, with co-pays and caps (https://www.nhsa.gov.cn/art/2025/8/23/art_14_17675.html), and there are many reputable private health insurance providers
Its the opinion of one professor, in particular they arrived at the 50% figure by including people who are not actively seeking work.
There's a couple of things to unpack there:
Indeed, while it may line up with peoples intuitive interpretation of the term "unemployment", most official unemployment figures around the world don't include people not seeking work.
That being said, it does imply a massive amount of young chinese people who are simply doing nothing.
Yeah. I’m surprised it was included. The extremely high youth unemployment rate is incredulous without a lot of supplementary information.
There’s definitely misreporting on China’s economic situation, but exaggerating the situation and just making up information about how bad it is doesn’t help that. You can’t fight misinformation with more misinformation in the other direction.
90% of the economic reporting on China is clearly false. And it seems co-ordinated. There was a ridiculous paper suggesting that the lack of street lights in China meant their GDP was over estimated, rather than a country that hasn’t got around to building out as many such lights as countries that have been developed for decade. Obscure journals and university theses are always there, paper never refuses ink, it’s how this spreads is interesting.
The once great Economist magazine has gone to the dogs on this.
I have a bit of epistemic learned helplessness on China. On one hand, you have what you mentioned, and like Peter Zeihan constantly predicting China's collapse. It's patently false. China has advanced a lot in GDP, living standards, infrastructure, science (genuine publications in top journals) etc.
On the other hand, I read things from Steve Hsu, Karl Zha, and certain subreddits and it's like China can do no wrong. They can cheaply mass manufacture anything in a short period of time. Any challenges they face have all been planned for and will be well dealt with, especially compared to the sclerotic west. The governments are more responsive to the people than western ones, etc.
Some examples, that DeepSeek was basically a weekend project for some hedge fund workers who used a box of scraps to best multi-billion dollar American companies. Outdated Chinese jets are better than French SOTA ones, Or that China will eclipse the west in all sciences and manufacturing. It seem to stem from some HBD thoughts, "first principal" thinking and biases for STEM-types. I.e. that China is lead by engineers while the west is lead by "retards" and "midwits" (their words) lawyers, and the west faces existential crises that they will not be able to avoid.
Latest Chinese unemployment rate link: https://data.stats.gov.cn/easyquery.htm?cn=A01 I found this: Unemployment rate of the labour force aged 16-24 years excluding school children, urban and rural, national (%): 16.9（11/2025）
> If we don’t get a crazy AI future, then human labor won’t be obsolete, and you won’t be in a permanent underclass (at least for that reason)
Wait, where does that assumption come from? It's perfectly plausible that AI capabilities stagnate at above-replacement levels, but fails to perform any miracles. Even with "free" labor, energy and resources are still greatly limited without any breakthroughs. We're already seeing issues today with the AI industry taking away resources from producing consumer electronics...
I think if AI works like other technological advances, then it replaces some existing labor but new jobs arise and we're overall just richer--nobody makes a living as a truck driver or call center employee anymore, but plenty of other jobs are still human-only. OTOH, if AI ends up improving until it is better than humans at almost every job, then the pattern won't follow what we've seen with other new technologies, and perhaps most or all humans will simply not be employable anymore, as there just aren't very many things that any human can do better (at the same cost) than a machine. Perhaps there are still rare super talented people who can outperform any machine in some domain, but if that's like 1% of humanity and 99% of humanity has nothing we can do better than a machine, that's enough to break the old pattern of "new technology takes many jobs but then new jobs arise and the world comes back into balance."
Then we have to think about whether or not there are jobs for which the demand is based on it being done by a human, which could be good (chess grandmasters and some pro athletes get paid for doing things a machine could do better now) or bad (maybe the thing you really want is the ability to lord it over your human sex slaves, knowing that their misery is real instead of simulated is the thing that makes them more fun than just lording it over a sexbot whose misery is probably just an LLM telling you what you want to hear).
Don't forget bio-research! With so many people who are suddenly now completely expendable, there will finally be an opportunity to gain comprehensive knowledge on the functions of the human body! For example, they could finally verify data from that nearly century-old hypothermia study!
Completely agree. Being at a level of capabilities which makes humans basically worthless economically is still "in distribution", AI would not need any fundamental new capabilities, just be much more efficient than humans. The jump to "then we also will have space travel, abundant resources, much better political coordination, etc" simply does not follow, as it requires fundamentally new "out of distribution"-capabilities.
Venezuela: "Power is the chance(probability) that orders are carried out"- I was kinda surprised Trump seems to mix up catching Maduro with having gained the power to run Venezuela. Now, some say, the Vice-president and some elites are willing to do Trumps's will. With just some public pretense of "we are independent". One wonders.
Ehhh, I'm not sure Trump had a plan, this seems like more of a "you can just do stuff" thing.
Like, Trump just took out the head of state at virtually no cost, no invasion, just snatch the president, swag, and leave.
Will Venezuela improve? I dunno. It's hard to imagine Venezuela getting worse. Maduro seems pretty uniquely bad in terms of economic performance, democratic legitimacy, and just good governance.
In basic risk-reward, if the cost for removing Maduro is effectively zero and the benefit is potentially really high for the US and Venezuelan people...why not? If we get Neo-Chavez or somebody, has anybody really lost anything? Versus potentially getting someone competent enough to get the oil flowing again and the economy back to some semi-reasonable functioning.
Just a reminder how much Venezuela's GDP per capita has cratered under Maduro:
I think the cost to the US is pretty clear, and the benefit pretty minuscule, actually, and I'm puzzled why the warmongers think otherwise.
(Think how eminently successful most woke outrage mobs were, in both purging people and controlling the narrative to make it clear the victims deserved it. Now think about the end result after a few years and multiple iterations.)
It's demonstrated that the US is still a country willing to force regime change if they decide they don't like your government and they think they can get away with it.
A lot of the anti-US people I meet cite past US regime changes in SA as a major reason for their attitude and I used to be able to say "Well that was forever ago, nowadays the US doesn't do that kind of shit anymore".
Now I'm thinking they probably had a point and I was more pro-US than was warranted. The US+Europe alliance certainly has a lot of perks but this kind of shit still makes it hard to stomach.
I hope it leads to a surge of pro-EU / against reliance on the US sentiment, even if that will cost us. But frankly I'm not hopeful.
> Well that was forever ago, nowadays the US doesn't do that kind of shit anymore
You're not counting the failed one in Bolivia in 2019?
And don't worry, if you thought that argument persuasive last week, you can use it again as early as this time next year, when the 120th Congress is sworn in and the US fundamentally changes.
I'm murky on the specifics since my view was always "even granting that happened, it was forever ago and the US is no longer doing it" so I never needed the details to argue my point.
That said I ctrl+f-ed the wiki page¹ "US involvement in regime change" for CIA and there seems to be enough that qualifies as "close enough". I'm happy to concede that most entries on the list are not literally regime change in SA, depending on how you count you might find several.
But in any case, the sentiment I want to communicate is that the US was unilaterally deciding they don't like your country or government and then decided to do something about it. For example I would also count the following quote as expression of that sentiment, despite not being regime change: "The U.S. government ran a psy ops action in Chile from 1963 until the coup d'état in 1973, and the CIA was involved in every Chilean election during that time."
> But in any case, the sentiment I want to communicate is that the US was unilaterally deciding they don't like your country or government and then decided to do something about it.
I assume that's the major reason for the attitude of anti-US people from SA. But then why don't Americans have anti-Russian political parties (with anti-Russianism a major plank), or anti-Iranian or anti-Israel? All of those countries seem to have intervened in American elections several times: https://en.wikipedia.org/wiki/List_of_foreign_electoral_interventions#United_States
Why isn't anti-X country sentiment more widespread in the U.S.? I've met a fair share of Americans and I don't recall ever meeting an anti-Russian one, while based on you account - and a few others I've heard over the years - it's pretty easy for an American who has some contact with South Americans to find anti-US people.
I find the anti-US sentiment derives from American intervention explanation lacking.
You might consider actually reading what said "intervention" actually involved instead of simply the fact that they're mentioned on the list. If you do, you might find that adequately explains the observed difference in anti-X country sentiment. I'd actually say that shows disproportionately high levels of such sentiment in the US.
To clarify, I am from Germany and have no special insight into what people in SA believe.
That said, it's perfectly consistent to condemn what the US has been doing and also condemn Russia (, Israel, Iran) when they're doing the same.
And I see no issue with the US taking measures to limit foreign influence on their elections, in fact that seems like a good idea in general.
But if the question is why the people tend to mind it more when the US does it then when, say, Russia does it... I can only speculate but somewhat high up on my list on candidates is "because the US says they're our friends, and we say we're their friends, and yet still they act like that".
> In basic risk-reward, if the cost for removing Maduro is effectively zero and the benefit is potentially really high for the US and Venezuelan people...why not? If we get Neo-Chavez or somebody, has anybody really lost anything?
Plenty of room for worse outcomes, like "Venezuela's military fragments into competing groups and has a decade-long civil war that brings even more instability to the region".
If they don't follow orders, the army's going to keep killing people until someone does. It's not any more complicated than that. After all, the prize is Venezuela's resources, not its people.
Basically, if you think you can enforce specific policies by drone strikes against leaders, you are sorely mistaken. How many Taliban leaders did the US kill by bombing weddings, again? Did it result in the Taliban becoming docile and submitting to the US?
That the US was able to kidnap Maduro is surprising in itself. Sure, they might have taken out all the AA, but preventing infantry from firing rockets at your helicopters once they are within range seems hard. It seems entirely possible that they had cut some deal where the VZ military was not resisting to the best of their ability.
Nor does Trump have the freedom of action to depopulate Venezuela. His adventure was branded as a DoJ operation. Drone striking their VP will be harder to sell.
In the end, the only way to impose his will on a VZ bent on resisting him would be a full invasion. And that would be a long and messy affair which will almost certainly be net negative for the US.
To some extent it had to be an inside job. I know the US had a source that would tell them exactly where they could find Maduro and exactly what the room looked like so they built one to practice the attack. The fact that the US is willing to embrace the vice president tells me that she's not too far from the center of that. I also think that her protests in public about American hegemony is a little bit of kabuki.
This could easily go awry, and given our history, that's the way to bet. But ISTM as someone with no relevant information or experience like the snatch-and-grab operation had to use some insiders who were telling us where to find him and perhaps quietly turning off some cameras or removing some guards or something. And the person lined up to take power after Maduro is on his way to the US is a pretty plausible guess about who might have been in on this.
> Nor does Trump have the freedom of action to depopulate Venezuela. His adventure was branded as a DoJ operation. Drone striking their VP will be harder to sell.
I think it's far too early to say that. That entirely depends on the hold he and his administration has over the military. The choice ultimately lies with them, not Congress.
Any Venezuela that gives us cheap oil, cooperates in anti-narco interdictions, and aligns against China can probably do whatever else it wants. This isn't a situation like Afghanistan where we were attempting to forcibly westernize a backwards medieval population of tribal warlords and Islamic theocrats. Venezuela was already a civilized western nation, at least in the way South American ones are, filled with decent modern people who have motivations and concerns and values pretty similar to ours. There's no reason to think it wouldn't be normal again with Maduro removed, and reversion to the South American norm is all we need there. It has a bunch of resources and doesn't really need to placate the narcos. Whoever's playing the Vicki Nuland role in this coup can probably find somebody who'll be accepted and operate a basically normal country, they'll know they're in our sphere and better not play footsie with China but other than that do whatever they want. In Afghanistan we were asking for a 500 year leap in what "normal" was, no conceivable leader could have done that in a generation.
> No civilian populace (note: not military junta or dictator) has ever been persuaded to give up
So this seems to be one of those "by definition" style arguments, where if you DO successfully get them to adopt some favored policy of yours through bombs/drone-strikes, that just means they were a dictatorship.
That does not logically follow from your quote, no. The article did not say what kind of government can be persuaded to give up through only bombardment. It does say that if a government gives up the fight after bombardment, the pressure to do so did not come from the population.
At some point that will end up with American soldiers being killed. Or a complete collapse of authority in Venezuela, at which point securing the oil fields will become a gun fight and it will be a horrendous task to get anything built or any oil out.
It feels more like Trump wanted his typical attention grabbing act, and then will simply declare victory and move on regardless of the actual facts on the ground. Look at how they cracked down on any suggestion that the bombing of Iran was not fully successful and how they haven't followed up on whether they are still trying to build a bomb.
Right. As some former military-intel people have been saying today: Bush43 et al had a poor operational plan for Iraq, Trump et al have just no plan at all for Venezuela.
What worth are guns against bombs and drones? Worst case scenario, it shouldn't be too difficult to clear the necessary areas out. But the Venezuelans are a reasonable people, right? Surely they're not so stubborn as to value their autonomy over their lives...
Either way, this is proof that the military is capable of operating efficiently as long it isn't obsessed over collateral damage. The US always had it in them to win a war, they just needed to rid themselves of their needless sentimentality.
Drones are not very good at e.g. seeing what's inside a building. And they suck at talking to people.
You'll eventually infer, from the fact that the main cracking tower at the local oil refinery just blew up, that somewhere there's a building in which an insurgent bombmaker fabricated an IED tailored to fit a disgruntled refinery worker's lunchpail or whatever. But that won't tell you which building, or which disgruntled worker.
For that, you'll need to search buildings, and talk to oil-refinery workers. The people you send to do that, can be shot with guns. By other people who looked just like any other random Venezuelan until they shot your people. Eventually, you'll run out of people willing to be shot for the sake of American oil-company profits, and then what?
> Drones are not very good at e.g. seeing what's inside a building. And they suck at talking to people.
Isn't that what the strategy used in Gaza is for? People can't hide in buildings if there are no buildings. If there's a big enough buffer zone between the oil fields and the rest of civilization, it won't even be possible to approach it without getting shot. Besides, there's no reason the people who eventually drill the oil need to be Venezuelans. That might even be an opportunity to create American jobs.
If the plan is to ethnically cleanse Venezuela (or even just the good parts thereof) of Venezuelans so that proper Americans can go have good jobs there without being killed by people peeved about all the ethnic cleansing, then no, that's not going to work.
It didn't even work for Israel in Gaza, in spite of their having (as you note) flattened a significant fraction of the buildings. And they can only survive the blowback from that because they've got a patron some thirty times larger and more powerful than themselves. Unless Trump has secured the support of the Galactic Empire recently, that's not going to work for us.
>the military is capable of operating efficiently as long it isn't obsessed over collateral damage
And if my grandmother had wheels, she would be a bicycle. Being concerned with excessive collateral damage is one of the things that distinguishes a professional military from an unprofessional one, a concept that in the US dates to the Civil War. https://avalon.law.yale.edu/19th_century/lieber.asp
And how did that "professionalism" work out for them? Nearly a hundred years of zero meaningful victories under their belt due to constant half-measures. It's no coincidence that the last war they won involved a nuclear bomb. That is the lengths the military went to when it was fighting a war that it wanted to win.
Seems more like we usually win the war and lose the peace, to be honest. We're good at the killing people and breaking things part of war, but often not so good at the nation-building and winning-hearts-and-minds part, probably because being really wealthy and good at technology and organization helps more with the first one than the second.
Leaving aside the obvious incorrectness of your empirical claim about zero meaningful victories -- Saddam Hussein (twice), Mullah Omar, whomever was in charge of North Korean troops at Inchon, etc would beg to differ -- what exactly is your claim? You seem to say that any war aim, no matter how trivial, is worth the death of an infinite number of civilians. Why?
Bombs and drones can't secure an oil refinery, at least not if you want it to be able to produce oil afterwards. You can try to order the Venezuelan military to secure it for you, but the more you bomb them the less capable they are of doing that.
Also, even if you think we can just keep drone striking Presidents forever until we find someone pliable, you need the President's orders to be obeyed. If the President is widely seen as illegitimate - e.g., because everyone knows he's an American puppet, or because the government is unable to function because their leaders keep exploding every week, or because the cartels or a revolutionary movement has decided to take advantage of the chaos - then it doesn't matter how many Presidents you bomb, you won't be able to use them to direct the people manning the oil wells.
Also also, the political will for this currently rests on the idea that this is a quick in-n-out operation, and if we're blowing up a President every week (or God forbid, trying to bomb entire cities into submission) that quickly stops being true.
To put it another way, they Venezuelans don't have to be so stubborn they'd rather die than obey your orders, they just have to be stubborn enough that you have to send men with guns to see if they're obeying or not.
It did work in Iraq eventually, although it took a lot of time to defeat the insurgency.
People seem to think that Iraq is still a bullet-ridden hellhole full of suicide bombers. In reality, it disappeared from the news precisely because the country stabilised itself into a mostly boring status quo, with lots of small parties squabbling in the parliament. It even attracts quite a lot of foreign investment, which is the best indicator of the general trust that the current stability will continue.
But, it kind of did work in both places. The current Iraqi government structure is the one established by the 2005 constitution. And the US-supported Afghan government would have stayed in power for as long as the US military remained.
That being said, OP is being extraordinarily glib.
"Surely they're not so stubborn as to value their autonomy over their lives..."
It's reasonable to suppose that they aren't, since they didn't hurl themselves in human waves at the military in a display of ahimsa after Maduro rigged last year's elections. Few people are, when it comes down to it.
> Surely they're not so stubborn as to value their autonomy over their lives...
Is this stupid only for Venezuelans or do you think the founding fathers were misguided. Or any anti imperialist movement for that matter?
Anyway I don’t know if the people of Venezuela will be happy with this transition or not - the lifting of the sanctions will help economically. Not being autonomous will rankle after a while, particularly as the profits from the oil probably won’t be kept in the country.
I think it's silly when you're comically outgunned to the extent that Venezuela is. Their military budget isn't even a thousandth of their invaders'. The US is going to get what it wants one way or another, but it's better for everyone involved if no one has to die.
The Taliban were "comically outgunned" compared to the US military. But they're running Afghanistan, not the US, and everybody who was on our side when we were trying to run the place is now desperately trying to find asylum in the United States.
That isn’t the way human nature works though. Or else empires would never fall. The question is whether the Venezuelans are happy with limited autonomy or not. If the US isn’t going to run the army there then at some point they will most likely just declare full independence.
> If we do get a crazy AI future, and the economy grows 100x (Industrial Revolution scale) or 1000000x (solar system colonization scale) in your lifetime, then you only need a little capital to remain as absolutely well-off as you are today.
This is technically true but somewhat misleading. During the Industrial Revolution, all one needed to have in order to secure menial employment (barely enough to make ends meet) were good health, the clothes on one's back, access to a shack, and maybe a shovel. Today, one needs all these things as well, but also basic education, a cellphone (and thus a cellphone plan), and in some places a functioning car. On the plus side, you can sleep in the car, maybe. Similarly, being middle class back then required a lot less capital than it does now. It stands to reason that in a non-Singularity AI future (i.e. a future where humans are still alive and functioning in the world, vs. being dead and/or uploaded into infinite bliss simulations), even more capital would be required in order to remain at the same level of prosperity relative to the human average.
> B2B SAAS
I get it, B2B SAAS is not in any way sexy and super boring and why would you do that when you could be working on blockchain AI-driven biohacks or whatever... but... as it turns out a lot of that stuff is actually pretty useful. It's something that provides real value to lots of real businesses (and therefore people) today, right now; and when it works well it's completely invisible, as all good machinery is. The difference between working on sexy blockhain biohacks and B2B SAAS is akin to the difference between working on a space mission to Mars, and working on building a really nice and affordable car. One could make arguments for both being needed, but last time I checked everyone is driving cars and no one lives on Mars.
I think #4 is a good point. If we're talking about an actual beneficial AI + Robotics take-off, then it's trivially cheap to provide every existing human with a life-style that would be the envy of any modern plutocrat. They might not all be able to have mansion estates in Beverly Hills (land on Earth is still scarce), but they could build an extremely plausible replica of it in a giant space colony.
Although I tend to think betting on an investment portfolio in an all-out AI take-off is like assuming that getting yourself an extra ten acres of farm land is going to save you from the disruptions of the Industrial Revolution. Anyone particularly rich in particular is going to be a prime target for malicious AI scamming/kidnapping/crime efforts, plus any AI given orders by violent political dissidents.
No, but I've got a "to do" to set up a Claude nutritionist prompt and I'll see how that works.
The plan is basically, once a month, upload all my Fitbit and smart scale data (steps, exercise, sleep, weight, body fat%) and my caloric intake and see how well it does. I have not been...super impressed by nutritionists in the past, especially given their cost.
So the Claude agent should be able to see, for every day, my weight, body fat comp, steps taken, sleep score, stress level, activity time/score/whatevs and potentially make some recommendations from there. Manually logging calories is still a challenge (1) but I *think* the agent should be able to provide some actionable insights based on the above.
(1) Calorie counting is challenging because of social obligations. Most days I can calorie count to within ~50 calories (Me like ham sandwich, me eat ham sandwich every day) but my Saturdays are like "Alright, I went to poker night and had one martini, one gimlet, 2 chocolate cigars....2 glasses of wine....maybe half-ish a bottle of sake...I think I had some peanuts, did I eat anything else? Oh yeah, that weird soup, it tasted like beef and butter, so I didn't eat much. How many calories is that?"
Yeah. Don't get me wrong, it's extremely valuable but it stinks and it's disruptive in that "just one minute" is the death of productivity way. No one pulls up those stupid little apps and goes "alright, say one shot of gin, 80 calories, half a lime of squeezed lime juice, alright, it's not in the app, say 20 calories" and enjoys it. Other stuff, like lifting weights, is objectively harder but it's also, like, "Oh yeah! Rob Zombie and lifting heavy rocks time! Life is gud." On the other hand, CICO isn't...strictly true but it's like 90% true and miscounting your caloric intake is the easiest and most common way to screw up weight loss.
So yeah, counting calories: more valuable than running 3 miles/day, also sucks way worse than running 3 miles/day.
This reader experience with China is directionally right but wrong by an order of magnitude from my personal experience (I didn't search for data but not a single thing he mentions corroborates with the daily life of the people I know in China living in the 3 biggest cities)
One important point about China is that it is pretty common in major cities to work 12h a day (like probably half of the white collar workers in the private sector or more work that much from what I am hearing), this was a deal easier to accept for people when it could buy them literally anything, but harder now when their spending power and their wealth has diminished.
It would be historically normal for the VP to be the favorite at this point, the only times in my life it wasn't involved Biden not intending to run because Hillary had arranged her coronation and Dick Cheney signaling early he had no intention. George HW Bush was a favorite, Al Gore was a favorite, both faced no serious opposition.
In this case however, Trump is mercurial and has enough control over the party apparatus that he might be able to throw his support to somebody else successfully and toss JD under the bus. 2027 will be interesting to watch, because after the midterms Trump would typically enter lame duck status and the party would be looking to the future, and he'll be pretty old, if there's a schism and Vance isn't the consensus there could be a lot of drama with who is willing to publicly defect from Trump and whether the floodgates open.
Of the last several Presidents who were still in office at the end of a term and either did not run for reelection or dropped out early in the nomination process:
Obama (two full terms): Biden declined to run in 2016.
Bush the Younger (two full terms): Cheney declined to run in 2008.
Clinton (two full terms): Gore ran in 200, was nominated, and lost the general election.
Reagan (two full terms): Bush the Elder ran in 1988, was nominated without serious opposition, and won the general election.
LBJ (1.5 terms): Humphrey ran in 1968, was nominated, and lost the general election.
Eisenhower (two full terms): Nixon ran in 1960, was nominated, and narrowly lost the general election.
Truman (1.95 terms): Barkley ran in 1952, lost the nomination.
Coolidge (1.5 terms): Dawes declined to run in 1928.
Wilson (2 terms): Marshall ran in 1920, lost the nomination.
TR (1.8 terms): Fairbanks ran in 1908, lost the nomination.
----
Overall, that's 70% running, 40% winning the nomination (57% of the ones who ran), and 10% (just Bush the Elder) winning the general election. Although Biden and Nixon did go on to run and win the Presidency after sitting out a term or two.
Caveats: small sample size, long time window (n.b. that the three failed runs for the nomination were all at the beginning of the sample and the four successful ones all towards the end), and the results are not very robust to contingency (Gore in 2000 and Nixon in 1960 both came very close to winning the general election, especially Gore who lost by much less than the margin of error in the counting procedures).
I misread the original post as favorite to be the party's nominee, not to win the general election. Good summary.
However, I still think you would expect that individual to have the highest chance of any named person, even if he should be under 50% vs "the field". Gore certainly would've been the prediction market favorite in 1998, even putting aside the possibility he might've been an incumbent president by the end of 1998, because there were no serious challengers staking out a lane for themselves against him whereas the GOP expected a crowded field. And in 1986 it would've been obvious that Bush had a clear path and the Dems had no obvious torch-carrier after Mondale went down. In '66 LBJ himself would've been the favorite because he was expected to run again. In '58 Nixon was sitting VP for a popular administration, Rockefeller didn't run his exploratory tour until the following year, whereas the Dems had multiple plausible candidates.
I think sitting VP as the odds-leader would be the default case 2 years out in a primary system, even if VPs have underperformed relative to their stated chances. (And as you noted, not by much in small sample, Gore won the popular vote and Nixon has long been said to have lost only because of the political machine in a couple midwestern cities, so this is a hair from being 30%.)
If we're doing popular vote, it's a slightly larger hair from being 40%. Nixon trounced Humphrey in the Electoral College in 1968 because George Wallace carried most of the Deep South, and had a small but healthy margin in the tipping-point state (2.28% in Ohio, which put Nixon over the top for an outright majority in the electoral college), but only won a plurality of the popular vote by a 0.7% margin (43.4% to 42.7%).
By comparison, the 1960 election had Kennedy with a 0.2% margin in the national popular vote and a 0.52% margin in the tipping point state for an outright electoral college majority (0.8% in the tipping point state for Nixon to win an EC majority, since Alabama and Mississippi elected Dixiecrat unpledged electors who wound up casting protest votes for Senator Harry Byrd. And in 2000, Gore won a 0.5% plurality in the national popular vote but lost the tipping point state by 0.009% in the official count.
---
Yes, I agree that it's reasonable for Vance to be around 30% to win the general election and around 55% for the nomination. VPs in recent decades do usually win the nomination when they seek it, and Vance doesn't currently seem to have signs of not being apt to seek it: he definitely isn't too old, seems to be in good health, and doesn't seem to be massively unpopular with his party's primary voters. OTOH, there's plenty of room for uncertainty and for someone other than Vance to have a shot at the nomination.
Frankly I don't see anybody else coming to rise in the Republican Party, now that it's so intertwined with the MAGA movement and they want to have the closest "link" to Trump as possible.
I think this is normally the case 3 years out from an election. A year is a long time in politics, at this point it is very hard to say who will be the leading candidate come 2027. Vance is the closest to power right now so he's the only obvious candidate.
I would classify it as interesting-but-not-surprising. He seems to be around a 30% chance on most markets, followed by Newsom, Rubio, Cortez, and a long tail.
This seems like a not-unreasonable number for Vance (though I think Rubio and Cortez are way too high). At this stage it's a tossup whether the winner is an R or a D, and saying that Vance as incumbent VP (and a young ambitious man) has a 60% chance of being the nominee sounds about right too.
Winning the presidency usually requires a candidate with broad, charismatic appeal, and Vance doesn’t seem especially strong on that dimension. He’s also taken positions that could narrow his coalition: leaning toward including far-right elements in the party, aligning with a more restrictive approach to foreign policy, and emphasizing the U.S. as a Christian country in a way that may cost him swing voters.
It’s also notable that Newsom isn’t priced higher in Democratic nomination markets, since he has been positioning himself for a national run for some time and seems to have meaningful support with Democratic voters.
I expect it to be like the 2024 primary, in which there was one frontrunner who was clearly winning and a bunch of people trying to get a small polling percentage, but who were never really in the race.
That's 30% to be the winner of the 2028 general election. For the nomination, the odds are:
Vance: 55%
Rubio: 14%
Trump: 4%
DeSantis: 4%
Other (total): ~23%
In normal circumstances, Trump being on the list at all would be surprised because he's termed out, but Trump has been intentmittentlt making noises about trying to ignore term limits on some pretense or other.
Besides that, it looks like the market is pricing in a real possibility of a competitive primary but consider it far from certain.
> In pre-trial proceedings, the government stated that Noriega had received $322,000 from the U.S. Army and the CIA. Noriega insisted that he had in fact been paid close to $10,000,000, and that he should be allowed to testify about the work he had done for the U.S. government. The district court held that information about the operations in which Noriega had played a part supposedly in return for payment from the U.S. was not relevant to his defense.
Getting a "trial" with an actual jury doesn't mean much. Some dialogue from Bridge of Spies:
[scene]
[Jim's boss] It was important to us - it's important to our country, Jim, that this man is seen as getting a fair shake. American justice will be on trial.
[...]
Jim, look at the situation: the man is publicly reviled.
[Jim] And I will be too!
Yes, in more ignorant quarters. But that's exactly why this has to be done. And capably done. It can't look like our justice system tosses people on the ash heap.
[scene]
[Jim] I just don't think that three weeks is going to do it here, We've got a massive amount of evidence--
[the judge] You want to postpone.
Six weeks.
Jim, is this serious?
Sir?
Is this serious?
Yes, yes, indeed it is. You can see in the filing that--
Jim, this man is a 𝘚𝘰𝘷𝘪𝘦𝘵 𝘴𝘱𝘺.
Allegedly--
Come on, counselor!
Your honor--
Of course, I salute you - we all salute you - for taking on a thankless task. This man has to have due process. But let's not kid each other. He'll receive a capable defense. And god willing, he'll be convicted. Come on, counselor. Let's not play games with this, not in my courtroom. We have a date, and we're going to trial.
[scene]
[Jim's boss] Jim, you did a great job. You fulfilled your mandate and then some. But the man is a spy. The verdict is correct and there's no reason to appeal it.
[Jim] There's ample procedural reason. We know the search is tainted, and fourth amendment issues will always weigh more heavily in an appellate forum. We've got a good shot.
What the goddamn hell are you talking about? We were supposed to show he had a 𝘤𝘢𝘱𝘢𝘣𝘭𝘦 𝘥𝘦𝘧𝘦𝘯𝘴𝘦, which we did. Why are you citing the goddamn constitution at me?
---
I remember seeing an opinion essay by Eric Posner arguing how important it was to try 9/11 conspirators in whatever was the most court-like forum that guaranteed a 100% chance of conviction. You've got to keep up appearances, after all.
I think there is a world of a difference between the level of courts operating at the height of the red scare to make sure that a commie spy is convicted and Maduro getting accused of illegal possession of machine guns and narcotics trafficking.
If Putin had pulled this stunt, then I am sure that he could just call the judge and tell him what verdict he would prefer, and will get exactly that verdict. By contrast, the federal courts are thankfully much more independent of Trump.
Maduro is not reviled by the American public to the same level as Bin Laden was. Sure, if the DoJ presents solid evidence entangling him with smuggling drugs into the US, they would very likely feel obliged to convict him, but if it turns out that there is no great evidence, then I don't think a court would have a problem with acquitting him.
Why is there so little recent data on early miscarriages (<20 or <22 wks depending on country) considering early miscarriages compose 98-99% of all miscarriages? Good data is collected on abortion, std's, etc. Why is there not a collective interest in this topic, particularly considering the potential effects of COVID and consequent health measures on reproductive health? The Scandinavian countries are the only countries I could find that collect data, but they haven't published any data on early miscarriages since early 2021. I have heard bold pronouncements both directions, particularly around the vaccines, but there is, as far as I can tell, an absolute absence of data with which I can address the question.
How would we collect data on early miscarriages? The modal early miscarriage occurs before the mother realizes she's pregnant. You'd need to institute frequent, regular pregnancy tests on everyone who could menstruate.
Miscarriage rate per all pregnancies. Even better would disaggregated rates among relevant categories such as gestational age, maternal vaccine status, and maternal characteristics such as race, BMI, residence, etc.
I asked GPT5 your question, and it located numerous articles. Looked to me like most were published 2023-2025. Below is its summary. which includes links to articles. In addition to summary it also listed 2 big-picture summary articles.
I made a game for guessing where people are from based on their accents! Feel free to try it at accentguessr.io. Let me know if you have any requests for features or other areas for improvement
If effective altruism seeks to maximize wellbeing relative to resource input, it would seem that things that have a non-zero chance of producing infinite wellbeing swamp everything else in expected value and are worth arbitrarily large resource expenditure. From this you get “effective evangelization,” trying to save souls because an infinite life in heaven has infinite value, there’s a non-zero chance a given religion is true, and therefore saving souls should be maximized. However, the problem with this is that salvation usually relies on converting to and maintaining a lifelong belief, which is a non-trivial burden to expect out of people. This is where Pure Land Buddhism comes in. In Pure Land, faith is effective, but it does not require unwavering, perpetual devotion. It’s said in one of the Buddha Amitabha’s vows that people who sincerely recite “I take refuge in Amitabha Buddha” even as few as ten times will be reborn in his Pure Land. This is a land of perfect bliss where achieving Nirvana is easy. You will never be reborn in a lower realm again, your future is guaranteed to be bliss followed by enlightenment. So, if Pure Land Buddhism has a non-zero chance of being true, and if it has one of the lowest barriers to salvation out of all faith-based religions (saying “I take refuge in Amitabha Buddha” ten times), then, as effective evangelists, ought we not try to spread this practice as far as possible? It would seem to be an effective cause area to tell as many people the good news that they can forever escape the suffering of Samsara just by repeating a saying ten times. If taken seriously, those that want to maximize value should put at least some resources into Pure Land evangelization, because it offers the possibility of infinite reward for trivial commitment. So, what do you all think? Should we take out online ads and billboards telling people to repeat Amitabha's name, or some other way of turning money into mantra recitations?
Same objections as Pascal's Wager, the space of possible consequences is larger than the hypothetical.
The True God might get jealous and damn you for worshiping false idols, but not for living a good life without worshiping anything.
By calling out to random supernatural entities you might invite demons to snatch dying souls from our world when we'd otherwise not come to their attention.
Money spent on making people's mortal lives better might give them the slack to actually live better and more moral lives, and that might get more of them into a positive afterlife than spending the same money on random preaching.
Etc.
If you feel like there is actual evidence for any one specific religious tradition having a chance of being correct, then that may swamp those hypothetical concerns and justify you spending money on that specific tradition. But I think that's true whether or not you use a utilitarian framework.
If you don't see an real evidence for anything, though, I don't think this is a meaningful argument for speculating.
Answer is no; Bayesian reasoning doesn't handle infinities properly. We need a better way to express "there is zero evidence for this incredibly-unlikely theory" than assigning it a very low (but nonzero) P. Until we find that, we have to reject incredibly-unlikely theories with zero supporting evidence manually.
I mostly manage this by ruling out by default (barring very strong non-probabilistic evidence) any payoffs of more than 100x, and increasingly down-weighting payoffs as they increase (so a 90x payoff with say a 10% chance I downweight further than having a 10% chance, closer to 1%).
Also I have a very strong time discounting, to the point where I basically ignore all non-trivial predictions beyond about a 10 year horizon, regardless of their payoff/EV.
While this is unprincipled and crude, it does effectively stop me needing to worry about stupid speculative rapture/damnation scenarios like "what if AI makes us all immoral geniuses in constant hyper-bliss" in favour of, say, "can I pay my rent next month"
Contra grumboid, the issue isn't Bayes. The issue is EV. In contrast, Kelly's Criterion handles Pascalian scenarios gracefully by automatically putting an asymptotic upper bound on the size of the wager, based on P(x). E.g. if P(x) = 2% and the payoff is infinite, you should only bet 2% of your bankroll. For finite payoffs, the wager will be a fraction of 2%.
Even with only finite numbers, you run into problems. The expected positive and negative value grows superexponentially with K-complexity. It grows faster than any computable function. So if you're basing the probability of something purely on the complexity and even if you're ignoring infinities, expected utility diverges.
You could assign probability based on how good or bad it is, but why should we expect the world to care about our utility function? Different people would have different utility functions. Does that mean they have to have different prior probabilities?
Enough people have reported deathbed experiences of Amitabha that I don't think it's fair to say there's "zero supporting evidence" at all. It might be weak evidence by your epistemic standards, but it's evidence nonetheless. Given this, a non-zero probability seems proper.
If something has been vetted, we trust it more. But if something has been doctored, we trust it less. This suggests that MDs seem more nefarious than DVMs.
Your question is not clear. On the surface, there is definitely no connection between "MD" (𝗠edicinae 𝗗octor, "'doctor' of medicine") and "meddle".
There also isn't a connection between "𝗺𝗲𝗱icine" and "𝗺𝗲𝗱dle". Wiktionary has meddle << Latin misceo ("mix") << PIE *meyḱ- ("mix"); medicine << Latin medicus ("doctor") < Latin medeor ("heal [someone]") << PIE *med- ("heal").
Note that while "doctor" is a Latin word, it isn't the Latin word for doctor; in Latin it means "teacher". This makes sense if you view a degree as a credential for professorship. A doctor of medicine is someone who teaches medicine to others.
If you were asking about some other potential linguistic relationship, you'll need to be more explicit.
I looked it up, and kinda maybe. There are several proposed etymologies for the verb floating around, but one of the more popular ones suggests that it comes from use of "to doctor" with a straightforward meaning of "to perform medical treatment upon". Possibly by way of analogy between "doctoring" a drink by adulterating it or watering it down and the process of preparing a medication for a patient. Or maybe just a general analogy of medical treatment as an invasive alteration.
Two others caught my eye. One was from 15th century use of "doctor" as a noun to refer to loaded dice (not sure where that term comes from). The other is independent derivation from the Latin root doceo, which originally meant to teach (doctor acquired the modern meaning due to the overlap between teachers and scholars) or to direct a play or other performance. Doceo is closely related to decet (to adorn, to give the proper appearance), which is in turn related to the root of the word "decorate". "Deceit" sounds like it should also be related, but that appears to be a misleading coincidence.
According to Scott’s most recent post every comment here stands the chance to become a cherished item in the reliquary of an EA space monastery (they will have turned it into an NFT).
I think humans should eat out of a spout on top of our head and eject waste from our toes: feces out of the big eight, pee from the pinkies. Placing the entry and exit points for food at the extreme distal sections of the body maximizes the surface area available for nutrient absorption and dispersal, while simultaneously allowing for more intuitive hygiene, since feces is ejected directly adjacent to the ground, which people generally treat as unclean anyway. Awkward crouching is no longer necessary when dropping a deuce, you can just leave a little trail of nuggets behind while walking. There's also a higher potential for personalized musical expression when farting, since each toe would naturally produce a different pitch when utilized.
A separate digestive tract also allows for much more specialization in the respiratory and reproductive systems, removing the perpetual need to switch between systems situationally to avoid contamination or strangulation.
Plus, I really want to learn how to juggle dumplings, then land them in a stack on top of my head, then open up my spout and make them all disappear in one big gulp.
This is basically a drug trip report, except it was all medically prescribed…
I have Graves’s disease (thyroid condition), and the endocrinologist decided to prescribe the heroically large dose of 80 mg/day carbimazole (plus 5 mg/day bisoprolol). It takes about two weeks for the drug dosage change to percolate through to a change in hormone levels, and I now have the blood test results:
T4 13.5 pmol/L
T3 5.5 pmol/L
Heart rate 80 bpm
… which is in the normal range.
So, is there a psychological effect? It is very strange. I feel like before the increased drug dosage (I was on 60 mg/day carbimazole + 7.5 mg bisoprolol) I felt a kind of fear-adjacent emotion, a pseudo-fear that felt physiologically a bit like fear without actually being fear. (What’s happening is the elevated T4 hormone levels are driving heart rate really high). It’s like the physiological response of fear without any associated cognitive content.
So anyway there it is, in the same spirit that the folks at erowid report on more exciting chemical experiences.
Today I had the pleasure of reading the following sentence on Psychotropical.com, the site of Ken Gillman, MD, MAOI expert: “Despite valiant efforts to make them more evidence-based, guidelines, recommendations and exercise of policy power unfortunately remain among the least evidence-based activities, impregnable strongholds of expert-based insolence and eminence-based innumeracy."
Even YouTube sharing half of advertising revenue with creators, in context, is less of a heartwarming story of non-greed. And more the story of a corporation maintaining monopolies that would have traditionally been regulated under 20th century antitrust norms.
Traditional antitrust enforcement targeted companies that quantitatively seized dominance in their industry and pushed out competing companies. 21st century antitrust theory is kind of faux-populist with its argument, "We don't care what kind of market share you have, as long as the customer is not being overly negatively impacted." How TOUCHING. When that conveniently abandons quantitative monitoring for a vague qualitative monitoring, based on gut judgements from Ayn Rand loving regulators. "This industry's consumers are still doing acceptably well. I holistically see that the benefits of this monopoly outweigh the downsides. In my subjective wisdom, anyone saying differently is being a negativistic whiner."
Antitrust enforcement as an art, rather than a science. How urbane and sophisticated. 🦝🐻🤔💀💀💀
To my great surprise, today I found out that there is a 2016 film adaptation of "The Man Who Was Thursday". I hadn't heard or read anything about this and seeing as how it's one of my favourite novels, I wondered was this any good.
Unfortunately, looking at the trailer, it seems to be the same old schlock about Catholicism. It is better described as "inspired by" the novel, as it takes the names and nothing else. I don't regret missing this, since it's just one more in the "supernatural horror sexy nuns" genre.
https://www.youtube.com/watch?v=wKyVyrYWP6k
https://web.archive.org/web/20161005142327/http://www.edfilmfest.org.uk/films/2016/the-man-who-was-thursday
A proper adaptation would be great, but until then I'll have to stick with the printed book and the BBC radio plays adaptation.
I had the same experience a few years ago, and had successfully blocked it out until you reminded me. It seems like they took the title from Thursday and attempted to write a Dan Brown movie around it except with hot lesbians?
An actual film adaptation of TMWWT sounds unlikely though, the ending is going to confuse and annoy any new audience member, and of course the whole thing is also very much inconsistent with Hollywood values.
I agree this would be an excellent adaptation and it's on my nonchalant to-do-list. The big challenge is the narrator has all the best lines.
The Department of Transport says they want to eliminate all of DC's speed traps. If they do this, this will of course greatly improve the lives of ordinary people in the DMV region, but what I like more is that you can pretty much see what precipitated this proposal: someone there got a speeding ticket, and with righteous indignation, resolved "Never again!" and decided to strike at that evil without worrying about "protocol" or whatever bullshit reason most other people make up for not doing confronting blatant injustice they have the power to immediately fix.
I suppose im not easily seeing what precipitated this. I recall speed traps being complained about all the time. I wouldnt attribute their persistence to ppl in denial of their ability to immediately fix things. Some agitators get lucky and are heard by just the right person at the right time. Some agitators are wise and plot a complex plan that works somehow. Lots of time the change has nothing to do with the specific agitators.
BUT rather than pretty much seeing it, you looked it up and shared the case study as a template for change, that would be cool?
Just have all DC speeding ticket revenue go to the federal budget. Do the same with every local government's traffic fine revenue. It won't have a noticeable effect on the budget but it will make the world a better place.
Speeding is bad, actually. Enforcement of speeding laws are good, actually. Speed kills. What is the basis for claiming otherwise?
To answer your question directly. The basis for claiming otherwise is that speed variance kills more than the average level of speed. And that currently the way speed laws are enforced tends to increase variance therefore are on net bad. In particular, an optimal speed trap is a stretch of road that to the driver feels like it should be drove on faster than the marked signage ie a badly engineered road. Get rid of the speed traps. Get rid of the roads that invite them and adopt a plan for enforcement that reduces speed variance and youd be in better shape
Sounds like a call for spamming average speed enforcement cameras everywhere. Lots of UK highways are doing this now; and to be fair it does accomplish the goal of having drivers actually stick to speed limits without sudden changes of speed around the cameras.
Driving cars at *any* speed, will result in some people being killed. Either we ban driving outright, or we acknowledge that allowing these cars to be operated at this speed will result in this many deaths, and we're OK with that because the value of reasonably fast personal transportation outweighs the deaths, We have in fact decided on the latter course of action, and established at least a rough social consensus as to what the acceptable speeds are. The person saying "no, we should force people to drive slower than that because it will save lives", is doing net harm to society and should at minimum not indulge him.
It would be good if the consensus regarding acceptable driving speeds was codified in a clear and unambiguous law, i.e. a speed limit that we expect people to obey. In the United States, at least, we have for obscure but probably irreversible historical reasons decided not to do that. We have instead passed speed limits that are significantly lower than the speeds we expect people to drive at, and have empowered the police to stop and arrest literally any driver. Even drivers who are not exceeding the speed limit can be stopped and arrested by the police, on the grounds that everybody knows that everybody is supposed to drive faster than the speed limit so that one's decision to drive at the speed limit is probable cause to believe that they are a criminal trying to deny the police the ability to stop and arrest them. I wish I were making this up, but I'm not.
We at least used to trust that the police would use their discretion wisely and benevolently in this manner, stopping only the truly dangerous and/or probably criminal, and making the subsequent arrest as quick and unintrusive as possible. But we don't trust the police as much as we used to, and this sort of thing is part of the reason why. Wholly arbitrary law enforcement, and laws that can only be enforced in a wholly arbitrary way, are corrosive to public trust, and we are desperately short on public trust right now.
We would be better off with our existing consensus regarding acceptable speeds but literally no enforcement of speed limits short of reckless driving proven at trial. We aren't going to get that, but at the margin, any increase in enforcement of speed limits should be resisted and any relaxation of such enforcement welcomed. Maybe after a few decades with no limits, we would be able to reintroduce reasonable and fairly enforced speed limits.
> established at least a rough social consensus
Market approaches can only work for parties with a means to influence the market. Democratic approaches can only work for parties who get a vote. Consensus can only work for parties who get a voice.
Drivers get to decide vehicle speeds on the road. However, people wanting to drive at some speed are not the only users of the road network. Other road users' risk is an externality for the drivers, and absent some external incentive they have no reason to take it into account. Calls for speed limits to be imposed, lowered and/or actually enforced are also a signal; they are, in fact, the only input to the system that road users not in a vehicle have available to them. They are a mechanism by which drivers' personal risk/reward calculations can be made to take into account, however indirectly and imperfectly, their decisions' effect on squishier people around them.
The status quo vehicle speeds you observe are, yes, a de facto consensus - the result of a risk/reward calculation performed by the drivers, taking into account their risk preferences with respect to personal safety, and also their risk preferences with respect to police actions.
If you silence the latter, you are effectively silencing the road users who are not in charge of a vehicle, and the result is no longer "a rough social consensus" that takes into account everyone using the road. Only drivers' risk/reward preferences are now taken into account when determining acceptable speeds.
You then no longer have grounds to claim the outcome is the result of all road users' social consensus.
TLDR: do what you like on the highway, but please don't scrap speed limits on roads shared with pedestrians and/or cyclists.
When the federal 55 MPH speed limit was passed in 1973, fatal accidents per vehicle mile travelled went down a bit but returned to trend within a few years. When the federal limit was repealed in 1995 and many/most states raised their highway speed limits, there was no disernable effect on traffic fatalities.
https://www.researchgate.net/figure/US-Fatality-Rate-per-100-Million-Vehicle-Miles-Traveled-Source-US-Department-of_fig1_228775388
My understanding is that both the setting and enforcement of speed limits, on the margins, has only a minor effect on how fast people actually drive. From what I gather, conventional wisdom is firmly in favor of "traffic calming" road design that induces people to slow down because driving (too) fast feels unsafe. This can be hard to retrofit onto existing roads, of course but represents a long-term opportunity when roads are being built or renovated anyway.
That said, there probably is are ways to improve how we enforce speed limits. I'd propose three things:
1. I suspect that many/most speed limits are set a bit too low, with the expectation that enforcers will exercise discretion in enforcing the laws only against people who are actually being dangerous. I expect that this underlies a lot of the opposition to speed cameras, as a speed limit set with a built-in cushion for human enforcement discretion is not well-suited towards automated enforcement. The obvious solution is to build the de-facto cushion into the law by raising the speed limit. I think some places with speed cameras do this in a de facto manner (e.g. if the posted speed limit is 35 MPH, only send tickets for people measured going more than 41 MPH), but this probably isn't well publicized.
2. The penalties for speeding are generally set assuming a very low chance of getting caught. Speed cameras make you much more likely to get caught, which could be good because more predictable penalties are a stronger deterrent, but if the same penalties are applied they're excessive at the higher likelihood of getting ticketed. I'd support widespread use of speed cameras if they were paired with lower fines (maybe 5-10% of current fines) and no license points except for particularly frequent or egregious offenders.
3. I suspect a large fraction of fatalities and serious injuries caused by speeding are caused by people who are speeding by a lot. No statistics to back this up, just intuition and anecdotes about people driving at absurdly high speeds. Go ahead and throw the book at people who are going at freeway speeds on residential side streets or going at racetrack speeds on highways.
>From what I gather, conventional wisdom is firmly in favor of "traffic calming" road design that induces people to slow down because driving (too) fast feels unsafe.
How does that work? I can imagine makeing a road that people feel like they have to drive slower on, but what you want is a road that will be *safer* at the level of speed people will choose on it. Insofar as peoples feelings are related to real risk levels, that seems difficult.
“Static hazards representative of the dynamic hazards” is a phrase sometimes used.
So you have dynamic hazards — things like children suddenly running out into the street.
And you have static hazards that don’t suddenly jump out at you, like street furniture, speed bumps etc. You see this as you approach, and slow down for them.
The idea is to put in place enough static stuff — the can always add a speed bump — to cause you to slow down appropriately for the likely sudden surprises.
(Of course, you also pay attention to things that are likely to suddenly become a hazard, like flocks of sheep, kids on the sidewalk, etc etc)
Traffic calming is most useful for roads around houses, schools, and shops with a lot of foot traffic.
If you've got a road in the middle of nowhere, the main risks of driving too fast are that you, your passengers, or the driver or passengers of another car are going to get hurt or killed in a collision. How fast you can reasonably drive is mostly constrained by sight lines, curve radius, and how separated you are from cars going the other direction. Most drivers tend to be pretty good about calibrating how fast they go to how fast the road is designed to accommodate around these sorts of constraints.
Take a road in the middle of nowhere where it's reasonably safe to drive 60 MPH and drop it into a residential area with houses and parks and stuff alongside it, and people will still want to drive 60 MPH on it. But you, as an urban planner, want people to drive more like 30 MPH because of risks of vehicle vs pedestrian accidents, so you put a 25 MPH speed limit (instead of the 55 MPH limit you would have put on the road in the middle of nowhere) there in hopes that people will only speed a little bit. The numbers I've seen most often are that lowering speed limits by 5 MPH lowers traffic speeds by 1-2 MPH, so your 25 MPH speed limit will lead to people driving more like 40 MPH. A big improvement, but not as much as you'd hoped. Also, you have people annoyed by a speed limit that feels unreasonably low even though it was set for good reason.
You get much better results by changing the road so people have to slow down to keep themselves safe, which also keeps the pedestrians safe. There are a bunch of things you can do to accomplish this. You can lay out the street so it curves more than necessary. You can put small roundabouts or four-way stops at intersections instead of traffic lights or two-way stops that prioritize traffic on the big road. You can make the road narrower. You can add speed bumps. And so on.
There's also some techniques that are getting more recognition and traction in recent years where you can make the road feel more dangerous to drive fast on without actually being more dangerous. I mentioned roundabouts before, which turn out to be an example: roundabouts are statistically safer than traditional intersections with stop signs or traffic lights, and have more throughput, but they feel more dangerous so people slow down around them. Another is to lay out roads with wide shoulders (good for safety, since this is a buffer between cars on one hand and pedestrians and building on the other hand) but add "chokers' (areas where a curb-height island blocks part of the shoulder) periodically so the street doesn't feel continuously wide. There's also stuff you can do with visual cues like how the road is painted to make it feel narrower than it actually is or it feels like you're going faster than you actually are.
The California Department of Transportation has a big guide about traffic calming techniques if you're interested in reading more:
https://dot.ca.gov/-/media/dot-media/programs/safety-programs/documents/traffic-calming/final-traffic-calming-guide_v2-a11y.pdf
The strategy is basically to make the road sinuous, bumpy, narrow, and obstacle-ridden. It's not just a feeling, they really DO have to drive slower to be safe. If you've heard of "risk compensation," where increasing the safety of some particular thing makes people do it more recklessly, and then you end up with more deaths/injuries, this is the same thing, but in reverse.
I do know of risk compensation, but you dont generally end up with *more* injuries/deaths from it. Most of the cases Ive heard of involve a close-to-constant level of risk, and a declining (but not in proportion to the "technical" improvement) risk is similarly possible.
That was the point: You can easily make people drive slower on a road, by making the road more dangerous - you need the slowdown to be overproportional to the danger increase, otherwise you havent reduced accidents, youve just made people drive slower at the same overall risk level.
For reference, neighboring Maryland has a 12mph cushion, and no points on the license. And lower fines, but not as low as you suggest.
It's true that lowering all speed limits to 15 mph including on highways would have fewer people die in traffic accidents. I do not believe that's a good tradeoff, and if there are bad laws, the enforcement of those laws is also bad.
But even specifically about the government of DC, I know from direct personal experience that it can and does set absurdly low limits, including on a particular broad four-lane straight and level stretch of road with no traffic lights or crossings, where the speed limit drops sharply just before the camera for no discernable reason, and then is raised back to a more natural speed a little farther ahead. A system that allows for such an abomination ought to be razed to the ground.
See also: https://www.youtube.com/watch?v=eXWhbUUE4ko
Yep fair enough. But here again we run into the classic American problem of overcorrecting for something by going too far in the other direction instead of finding a reasonable compromise.
The reasonable compromise is "have sensible speed limits and enforce them" instead of "have silly speed limits which exist only for revenue raising" versus "never enforce anything".
> Speeding is bad, actually. Enforcement of speeding laws are good, actually. Speed kills.
Okay, but do you know what *actually* kills? Homicide. And assault, and rape. Which have 40% / 50% / 33% clearance rates, respectively. As in, most people get away with them.
Do you know what cops spend ~80% of police hours on? Traffic stops. Pulling over soccer moms and working stiffs to revenue farm them.
I mean, I'm sure it makes sense to the cops - wouldn't you rather spend most of your time in air conditioned cars interacting with nice people with jobs and educations, versus trying to track down murderers and rapists, which undoubtedly has less congenial working conditions?
*Everyone* drives over the limit, including cops. If you're one of the rare people who don't, I'm sure you've been able to observe that you're by far the minority here. The entire system is dumb and literally set up for revenue farming.
Sure, there are some actually dangerous drivers out there, and they should be stopped and ticketed. That’s MAYBE 10%, probably less, of current traffic tickets, and can be accomplished with 10%-15% of the hours currently dedicated to it.
For all the rest, you *really* think we should be prioritizing giving rush hour commuters and busy moms traffic tickets over investigating and closing rapes, assaults, murders, thefts, and burglaries?? Because that’s what’s happening today!
From a post I did on this topic:
https://performativebafflement.substack.com/p/more-than-80-of-police-hours-are
Your post is not even making an argument for that 80% of police time is spent on traffic stops? Your assumptions show that ~65-70% is spent on admin, paperwork and commuting and then ~10-15% on traffic stops.
How you get from there to 80% is traffic stops?!
"Traffic stops or overhead."
The imbalance between "time spent solving actual crimes" and "time spent on traffic stops" is huge and largely unappreciated by most people.
None of this seems to apply to automated traffic cameras, which is what I was condemning.
Your post centers on "reasonable assumptions" which appear to be based solely on your own impressions.
In that spirit -- I am related to one big-city cop and longtime friends with another (different cities); sent them your "reasonable assumptions" of how police officers spend their working time. Quickly got back LOLs from each.
The officer working in the very-large city that my wife and I live in added that the percentage of time officers spend doing traffic stops has been near zero for "years". This fits with my wife's and my experience. We do get speed-trap-camera tickets now and then (one of which I decided to fight, and successfully beat), but neither of us has been actually pulled over while living in our present home i.e. 15 years.
I'm struggling actually to recall the last time I even _saw_ a traffic stop in progress except on our local expressways which are part of the interstate highway system (meaning state troopers not city cops). When I was younger it was routine to see city-streets traffic stops in progress.
And then thinking on it further, the last couple times I did get pulled over were in suburbs/exurbs not the city, and were well before any jurisdiction around here had any speed-trap cameras.
All anecdata of course, YMMV etc.
> Your post centers on "reasonable assumptions" which appear to be based solely on your own impressions.
I triangulated from literal numbers of citations both nationwide and in the New York area, those numbers are actual data.
My additional assumptions go down from those top level numbers, they aren't sui generis. And I'd be interested in what specific numbers they don't agree with - the proportion spent doing paperwork? Because 30% seems light, if anything.
Indeed, we need more speed cameras
I agree that the Kelly Criterion does help with the absurd Pascal's Wager type issues, but I feel it doesn't go far enough - I advocate basically ruling out any scenarios which claim to result in a sufficiently large change as a default, and requiring an exponentially strong evidence base to even consider them. I *theoretically* admit there's a chance of, say, AI turning us all into eternally suffering immortal morons, but I refuse to put any resources whatsoever into it because the chances are pathetically tiny. The odds of more mundane bad outcomes, depending on the specific, might edge into worth thinking about, but I am quite tired or rehashes of the Rapture or damnation or the Revolution, unless they come with actually plausible ways that could actually happen based in actual reality, rather than "and then a miracle/disaster happened because superintelligence"
Something causing a very large change does not necessarily imply that the chance of it happening is very low.
If you were president in the 1950s-60s, would you act as if nuclear war was impossible (and take no action to avoid it), purely because it would have a large effect if it did happen?
Are you the US President during the 1950s? No. In fact vanishingly few people are in a position to have that kind of potential influence, and it's even more rare to be in a position of that consequence and not know it.
(Your comment was orphaned but I assume you're responding to me from here [0].)
Idk, i feel like "wild outcomes require wild evidence" just folds into the probability being infinitesimal. If P(x) is epsilon, then Kelly is even lower than epsilon. If P(doom) is epsilon, Kelly is less than epsilon. This all adds up to normality, so idk why you think Kelly's Criterion needs additional pleading.
[0] https://www.astralcodexten.com/p/open-thread-415/comment/195451507
Yeah, I screwed up the comment reply XD
I feel like I'm struggling to frame my thoughts that is helpful and not-unkind, so I'm going to gently bow out.
I wrote a post about Charlie Kirk’s assassination 4 months later, I’m not the best writer so any tips on writing would be appreciated as well as general opinions.
https://open.substack.com/pub/mumbiss/p/4-months-later-charlie-kirks-assassination?utm_source=app-post-stats-page&r=1r3bff&utm_medium=ios
I guess I'll do a writing critique in a separate post, since I've already got the political opinion finished.
The biggest problem is here.
>The position is correct, but there’s been a hijacking of what it means to just “have a different opinion”. Charlie Kirk was the CEO of the arguably one of if not the largest conservative movements in the United States. He’s published several books, and has been a spearhead of the MAGA movement as a whole. His viral debates have fueled the opinions of many Americans social media feeds.<
You're presenting this like it was some kind of supernatural phenomenon; a meteorite hit the Earth's atmosphere and granted Charlie Kirk superhuman powers of persuasion, that he was using to do evil. No one else stood a chance against this alchemy; the Democratic machinery were powerless before him. That wasn't what was happening.
(Re-reading it, you even give him credit for Donald Trump's election. That's just nonsense. Trump was a #1 contender in 2011, a year before Turning Point existed. https://news.gallup.com/poll/147233/huckabee-trump-romney-pace-gop-field-2012.aspx)
The dude made money talking because he said things people wanted someone to say. So does every pundit; so do many Youtubers. Looking at a debate recording, he comes off as an asshole. If he wins the debates, it's because the students are bad at debating. https://www.youtube.com/watch?v=IP5LDxhBod0 ("How do you justify claiming murder is immoral," I mean Jesus, Student #2.) The dude was just one of many sparks in a country full of oily rags.
((Also this is the process Eliezer Yudkowsky called for in Let Them Debate College Students, three years before Charlie started debating college students. https://www.lesswrong.com/posts/yN38rRLzyuvNnhqr3/let-them-debate-college-students If it backfired, it's Lesswrong's fault.))
Never kill people who are willing to debate. Even if they weaponize the process with skewed facts and loud slogans, through the very act of showing up they're still promoting the process of seeking mutual, peaceful resolutions. It's a message to the common folks to get together and settle things with words at their own levels. If you kill them, then the people who were willing to talk will instead shut their mouths and load their guns and write their opinions in blood.
(The piece ends by saying you agree with this. The first writing complaint I'll make is that I have not been convinced you do.)
By this logic, any dictatorship would have fallen on day 1, because they killed someone and then the masses turned against them.
(If there is any lesson, it's more like "if you are the weaker side, do not give the stronger side a pretext by killing one of them", but even then it is not obvious that a pretext was needed or couldn't be manufactured otherwise.)
"who are willing to debate". Regimes fall because they're not willing to debate, they just do what they want and ignore objections, until the objections become violent and you get a dictatorship, who reigns by violence. Once you've killed the norm, it's just force of arms. And those arms keep getting weaker, because there's no cooperation.
I appreciate the reply, reading back over I actually don't make any case for why I even disagree with it in the first place, so this criticism is entirely valid. I'll be sure to revision it. I originally intended the article to be a devils advocate kind of deal.
It's easy to overcorrect when trying to present an opinion you don't have. I remember a teacher thinking I was a militant because I kept referring to the government as "law enforcement"; I was actually doing it because I thought the word "government" had negative connotations and was avoiding it.
I've messaged you my take on the writing itself, as a commented Google Doc.
> the people who were willing to talk will instead shut their mouths and load their guns and write their opinions in blood.
In Stephen Miller's words, https://x.com/RapidResponse47/status/1967627684886061110
“With God as my witness, we are going to use every resource we have at the Department of Justice, Homeland Security and throughout this government to identify, disrupt, dismantle and destroy these networks and make America safe again for the American people. It will happen, and we will do it Charlie’s name.”
As opposed to *before* Charlie Kirk's death, when the government was... *checks notes* ...deploying the National Guard to cities with liberal governments.
The whole "the liberals are making us do it" schtick on the right is very tiresome. They were already doing it before Kirk's assassin gave them an excuse, and Kirk's death doesn't seem to have changed that trajectory much.
Obviously they wanted to do it from the start, but it does allow them to justify it to their base. They still need support of the military, at the very least.
The most ironic thing about Kirk's death is that he's probably been more useful to his cause dead than alive. The violence he inspires in death will accomplish far more than his debate ever did.
I did a deep dive into some stats Kirk published about the effects of liberalizing gun laws. They were published in a widely circulated pamphlet. His stats are a textbook case of how to lie with statistics: really obvious, egregious cherry picking of stats, plus a couple of math errors that exaggerate the results in the direction of supporting his argument. His argument is that liberalizing gun laws actually makes people safer, because they are better able to protect themselves from violent crime. The actual data absolutely does not support that.
I was going to write that all up and post it here, but after my own upset about Kirk and people's reactions to it faded some I decided there wasn't a lot of point to doing that.. I have done all the math analysis, though, and am happy to send it to you if it would useful.
I did not notice, but thanks for letting me know. I've revised my article to provide more reasoning to oppose mocking his assassination but I'm interested in your analysis. Feel free to send it over.
Criticizing the actions or ideas or writings of a murder victim is not at all the same as mocking his assassination or cheering for his murder. Kirk in particular was a guy whose whole career was about political ideas and arguments, and it's 100% valid to criticize those ideas and arguments. Doing it the day he was murdered would seem kinda tasteless, but doing it months later seems entirely reasonable.
Hopefully, it's not too late in Open Thread 415 to post this. I just read a fascinating paper, "The Axioms of Cognitive Geometry: A Formal Model of Psychophysical Correlation" by Alexander Yiannopoulos. Creating a falsifiable mathematical model of consciousness is the first step toward understanding it. Yiannopoulos posits a dual-aspect or dual-substance quantum ontology, dressed in QM formalism, with separate Hamiltonians for the physical (outside world) and phenomenological (inner world). I may be missing something, because I don't see how he gets a causal coupling between the two, but I'm rereading the paper again to see if it was a comprehension issue on my part. But even so, it's pretty frigging cool that he attempts to model our internal phenomenological space as a Hilbert space. And then he provides a falsifiable framework. I love it! Be still, my Popperian heart! I'm not sure what sort of experimental setup he'd use. though? MEG? MRI?
https://philpapers.org/archive/YIATAO.pdf
Oh, and I don't quite see how he can get from the perception of red to the self-referential "I see red, and it's pretty." But I nitpick. Cheers!
Why do we need to modify the laws of physics to explain consciousness when a neural computational model seems perfectly sufficient? This does not seem like a parsimonious approach.
"The lived body is the pre-reflective 'zero-point of orientation' from which all experience is structured. We experience the world from here, where 'here' is constituted by our embodied situation."
I mean ... really? My prior for this being anything other than total nonsense is very, very, very low.
We shouldn’t have to modify the laws of physics to explain consciousness – and I hope I didn’t give you that impression. Unfortunately, it’s hard to step up a level of complexity without having some sort of rule-based understanding of the emergence of complexity. For instance, the standard model wouldn’t be predictive of the complexity of chemistry. And and knowing chemistry, we wouldn’t be able to predict the emergence of life (especially since the organic molecules required for life don’t spontaneously appear). The same goes for consciousness. It’s all very well to claim it’s just an illusion created by our neurons firing in certain sequences, but that doesn’t explain why we have the perception of red and can make the self-referential remark, "I see red, and it's pretty."
Prediction is much harder than explanation. No one would have been able to a priori predict consciousness-adjacent behaviors starting from the laws of physics, but from an ex-post position the laws of physics (or more precisely, the principles of computer science) seem more than sufficient to explain them.
>that doesn’t explain why we have the perception of red and can make the self-referential remark, "I see red, and it's pretty."
That's a very simple behavior that doesn't pose any conflict with physics. The neural mechanics of intersubjectivity haven't been worked out but I have no doubt that they will be eventually. In any case, how is a dualist explanation any better? Why is "this Hilbert space dimensionality reduction gives rise to the sensation of red" any more of an explanation than "this pattern of neural firings lets you perceive color"? It just camouflages the question with a new level of ambiguity. My view is that people have an irrational emotional resistance to understanding their subjective experience in purely mechanical terms. No one wants to see their wife turn into the green Matrix code. Sorry, but I'm pretty sure that's all we are. It feels magic to you because our brains didn't evolve to accurately perceive their own functioning.
Unless you can provide an objective, testable property of consciousness that resists a physical explanation then, as far as I'm concerned, it doesn't exist. Short of a behavior which can be constructively shown to be in conflict with the laws of physics, you're just making a god-of-the-gaps argument. Feelings of weirdness are irrelevant. "I like red" is something an LLM can say. There is no reason to presume that the subjective experience belongs to a fundamentally different category of explanation than the behaviors which are attached to it. Why would it?
Prediction is easier than explanation. because you can tell.exactly how successful you have been.
"That's a very simple behavior that doesn't pose any conflict with physics"
The puzzle -- the Hard Problem -- is not about explaining the report, it's about explaining the sensory quality reported.
". . .even when we have explained the performance of all the cognitive and behavioral functions in the vicinity of experience—perceptual discrimination, categorization, internal access, verbal report—there may still remain a further unanswered question: Why is the performance of these functions accompanied by experience?"
"Why is "this Hilbert space dimensionality reduction gives rise to the sensation of red" any more of an explanation than "this pattern of neural firings lets you perceive color"?
That isn't an explanation. Saying A causes B, somehow, is not explanatory. (Which is not to.say "it is done by a Hamiltonian, somehow" is.an explanation).
"Unless you can provide an objective, testable property of consciousness that resists a physical explanation then, as far as I'm concerned, it doesn't exist"
How do you everything is objective?
"Feelings of weirdness are irrelevant."
Qualia are directly accessible, the weirdness of Qualia is not.
Nothing is intrinsically weird. We judge weirdness by unusualness, and by theoretical expectation. These aren't the same. To encounter something every day is not to understand it .. although familiarlty creates an illusion of understanding. The layperson thinks magnetism is strange because they don't encounter it often, and don't notice that they experience gravity every moment -- yet gravitity is less well understood than electromagnetism to the physicist. To a young child, nothing is familiar or understood.,so everything is weird.
On the Weirdness of Qualia
Qualia are supposed to be immediately apparent, and the are supposed to be weird ... but the weirdness isn't supposed to be immediately apparent. Much confusion surrounds these points.
The weirdness isn't type 1 weirdness -- we literally experience qualita every waking moment. Instead it takes defiance of a theoretical expectation.
In the abaence of a science education , people tend to default to naive realism. For the naive realist, "red" is an entirely objective property of ripe tomatoes, etc. A ripe tomato "is" red , but looks blackish under green light. Dress illusion, etc
It takes some education (eg the Dress Illusion) to understand that qualia aren't objective.
(To the naive realist, it's thought that ls weird and insubstantial. The early mind-body problem, eg Descartes) was about thought. The invention of "thinking machines" dispelled that).
Moreover , a quale isn't a common or garden physical property that happens to be in the head. The expectation that everything in reality is fully described by physics , ie. that everything is quantifiable, objective and structural makes the existence. of qualia, which are qualitative, subjective and intrinsic surprising.
But without those realisations and expectations, there is nothing weird about qualia. Our ancestors were not puzzled by them.
> Prediction is easier than explanation.
Then why can everyone now deconstruct the causes of the 2008 GFC while only a small handful of people made a profit from it?
> The puzzle -- the Hard Problem -- is not about explaining the report, it's about explaining the sensory quality reported.
I’m well aware. One first has to demonstrate that there’s any reason to treat those two things differently. IMO no one has ever done a satisfactory job of that.
The rest of your comment reads like word salad to me. I genuinely have no idea what point you were trying to make.
>Then why can everyone now deconstruct the causes of the 2008 GFC while only a small handful of people made a profit from it?
Confabulating fictitious s,explanations is easy, but it's not explaining, any more than false prophecies are predictions.
>One first has to demonstrate that there’s any reason to treat those two things differently
They are different. You can have the one without the other.
It all boils down to an argument between people who think Qualia are something unique that needs to be explained, and people who think Qualia are just an emergent phenomenon of the nervous system that will eventually be explained in detail the same way we can explain, say, vertigo. Nobody ever convinces anybody in this argument!
If I'm reading that right, the claim is that if there is a certain mathematical relation between different neural processes this is going to show dualism is true and even how mind and matter interact? It seems a little like he's eulering himself. Did I get it right though?
That was my take. But I can’t see how he gets the physical and phenomenological Hamiltonians to interact using that QM formalism. I asked him on X but he hasn’t responded.
You post links to the greatest stuff! What’s your system for finding the diamonds in the dust heap?
The most interesting thing I’ve read recenly are some posts from Jordan Rubin’s blog. He’s building AI plug-in skills for mental moves one can make when ruminating: what’s an analogous thing, what’s the opposite? He’s get several posts about it, but this is the best one to start with because it includes an example of using one of the skills: https://jordanmrubin.substack.com/p/antithesize-this-raitah-edition
He put up a post about it here a couple weeks ago but it didn’t generate much interest. I think I was the only person to respond.
> You post links to the greatest stuff! What’s your system for finding the diamonds in the dust heap?
I have no life? I'm still over on X and follow a bunch of STEM and history posters there. There are still some virologists, epidemiologists, and biochemists on X, but most have moved to bsky due to the unrelenting harassment from COVID cranks.
Bsky for the virologists, epidemiologists, and biochemists no longer on X.
Substack is turning into an interesting intellectual resource.
I follow a bunch of YouTubers who produce videos on science and history, especially physics, archaeology, paleoanthropology, and linguistics. Sabine Hossenfelder is particularly good at bringing attention to new physics and physics-adjacent studies (and she rates them with her bullshit meter, which no doubt has contributed to her hat quotient).
I found this one over on X. Or, rather, it found me. I was arguing with some physicists physicalists who claim consciousness is weakly emergent and they'd have it all figured out in no time at all. Of course, I, being the Popperian Mystic provocateur, pushed back on that arrogance. In the course of the argument, someone posted that Sean Carroll says its weakly emergent, so it must be so. And I challenged them to post something other than epistemic handwaving. And Yiannopoulos replied: "How about this?" I don't think he expected me to read it.
On his show today Ben Shapiro described Cutis Yarvin as "a representative of a philosophy that is effectively a sort of right wing fascism" and part of the "right wing fascist opposition" to "left wing centralization". He criticized one of Yarvin's recent posts which called for Trump to dismantle the Constitution and build a "one party state". He pointed out that Yarvin has his fans within the Republican Party, including possibly J.D. Vance, and finished his segment by saying "In the same way that the left ought to divide off from its communist leaning splinter faction, the right needs to splinter off from this ridiculous neo-fascist trash. It's silly, and humoring it is incredibly silly. Not only silly, but dangerous, because humoring authoritarian nonsense from people who are purportedly on your side is a great way to be co-opted by those splinter groups."
Shapiro is certainly doing his best to defend American conservatism from its enemies to the right, be they Groypers or Neo-Reactionaries. I wish him the best of luck in doing so, and wanted to highlight an example of a pundit trying to police his own side of the aisle.
Devil's Advocate: To the contrary, communists and fascists and similar ones are the kind of thing that gets attention and votes in the Age of Clickbait.
(And yes, as a side effect they also give some attention and votes to the opposite side. That's how clickbait works. The losers are the moderate and sane ones.)
In what specific sense is Vance (possibly) a fan of Yarvin? I'm arguably a fan of lots of thinkers and talking heads with whom I nonetheless have significant disagreements, depending on the claim. Much of the time, I notice it's because they have phrased a particular idea in a way I find clear, compelling, or even just amusing. Yarvin is an example; I think his demotism is weird crankery, but I also admit it's sometimes fun to read his writing.
I think it's high time we all worked out a richer way to express distinctions between being a fan of some person and agreeing with all of their ideas.
I did a bit of searching, and it looks like Vance directly referred to Yarvin on a podcast in 2021, in the context of talking about how conservatives need to seize control of the bureaucracy by firing lots of people and replacing them with loyalists. Which the Trump admin has indeed done, so I don't think the fear that Yarvin influenced him is unfounded.
(Yarvin, for his part, said he doesn't think much of Vance and hasn't spoken to him directly. I did find one article that said he was in contact with Steve Bannon, though.)
I read the transcript of the podcast, and it was kind of an offhand mention, so I don't know how much the idea came from Yarvin vs. the idea just sort of being "in the water supply" on the far right. But I don't think there's a huge gulf in their viewpoints either - like, they agree on the need to purge the government of their enemies and neither one is very worried about the legalities of doing so, the question is just if Vance also agrees on the whole "American Caesar" thing.
I take all this as a question on whether the median American ought to be concerned about Vance becoming POTUS. The argument being suggested is that if Vance is a fan of Yarvin, then Vance might do anything just because Yarvin said it was a good idea; Yarvin said monarchy is a good idea; therefore, Vance will try to bring about a monarchy.
If this is the argument, then the reason I don't buy it is that "Vance will bring about monarchy" doesn't necessarily follow from "Vance is a fan of Yarvin". When resolving that argument, the claim that Vance is a fan of hiring loyalists ought to be irrelevant, because (1) there are multiple reasons both for and against hiring loyalists that have nothing to do with either Yarvin or bringing about monarchy, and (2) the people making the argument have been enjoying the benefits of loyalists in their administrations, which makes me think they don't really believe this argument.
Meanwhile, I haven't seen any evidence that Vance is any more a fan of monarchy than, say, any other future POTUS candidate. If the evidence that he does turns out to be solely "he's a fan of Yarvin", the evidence he's a fan is because he's a fan of hiring loyalists, and the people arguing this turn out to be okay with loyalists themselves, then it doesn't seem compelling.
Trump's purge of the bureaucracy has been much more far-reaching and politicized than any previous president's, so I don't really buy the argument that all presidents hire loyalists. For example, when the DOJ decided to start firing prosecutors until they found someone who was willing to drop the charges against Eric Adams, that was not normal operating procedure!
>I haven't seen any evidence that Vance is any more a fan of monarchy than, say, any other future POTUS candidate.
I disagree. I think Trump is much more a fan of monarchy than past or likely future presidential candidates, as evidence by both his own statements and his efforts to concentrate power in the Presidency, and that it's reasonable to believe that Trump's VP is on board with that.
Vance referring to something Yarvin wrote or said in a podcast is extremely weak evidence that Vance is strongly influenced by his ideas, or agrees with much he says.
It's... kinda sad that he doesn't realize he's already become the splinter faction now. Everyone else figured out that the old methods weren't working and have moved on to greener pastures, while he's still stuck on a lost cause. Oh well. Hopefully he realizes what's going on before it's too late.
The Groypers and the Neo-Reactionaries can't win in a general election: they can sabotage the Republican's chance of winning though, so I say Ben should fight to the hilt to keep them out of the seats of power on the Right.
I’m afraid we are living through a post-Judeo Christian, post-Enlightenment moment. Shapiro is being savaged by the battier part of the Right.
You're far from the first person to claim they've got a clearer view of reality than the next guy over, very much including people whose view is nowhere as clear as they think. Meanwhile, you're on record claiming infinities don't exist and that any number of civilians can be killed to achieve a war aim without acknowledging any of the multiple usual objections.
How do we know you're not one of those people with an bad view who thinks it's clear?
That is a good question, how do you know? Not that the answer really matters, mind you, seeing as we'll see who's right soon enough. The people want change, and the current leadership wants change even more. It was only a matter of time until the legal and moral fictions that composed this society were unraveled.
Thank you.
Though Shapiro's "Israel first, America second" approach doesn't seem promising either.
https://youtu.be/bKgWm5TNeBA
Anyone here with law-enforcement experience?
ACX-launched nonprofit Legal Impact for Chickens seeks a Humane Officer! 🚔🐥❤️
https://www.legalimpactforchickens.org/humane-officer
Anyone have any suggestions on how to improve one's ability to ask good questions? My issue is mostly in academic settings but I should think most advice here would be fairly general.
You always hear that asking questions is essential to learn, willingness to ask stupid questions is a huge boon (https://danluu.com/look-stupid/), etc. and I can understand and agree with these points in the abstract, but when (for instance) I'm in class and a professor explains something, I often feel as though I've followed their explanation and, when they ask for questions, I struggle to think of anything to ask. Obviously this wouldn't be a problem if I did, in fact fully grok the subject on the first explanation, but often I find later when I try to recall information that it hasn't quite stuck with me despite feeling like I understood it to begin with. I assume that in these cases I've failed to really grasp the subject in the first place, and I suspect asking more questions would've helped me to do so, but I struggle to formulate these questions at the appropriate time.
Does anyone have any particular strategies they've found helpful for being able to quickly formulate questions which get to the meat of an issue and help to promote long-term internalization?
Here's how I tend to do it (varies a bit depending on the subject/topic/lecture style):
I try to:
- build a model in my head.
- jot down just the most essential points and relationships between them.
- ask myself three questions:
- Can I "steal" this/ use this somewhere? (including: Can I use this for a bad joke/pun? (Basically my main mnemonic system :) ))
- Can I disprove this?
- Could I explain this to a 12 year old? (basically Feynman technique, shows you your gaps.)
The two latter ones are great ways to generate helpful questions.
This is the point of problem sets. You don't really understand anything until you actually use it. I recommend working one lecture ahead in your textbook. Do some problems for the section he's going to be lecturing on. If you're unable to understand the material on your own then you'll come to lecture with ready questions.
Yep. You get good questions when you:
* have already thought about the topic
* have already thought about something related to the topic
* tried to apply the topic to solving actual problems
...and maybe if your thinking is so much faster than the lesson, that you could do that during the lesson. But most likely it had to happen before the lesson.
Basically, the questions are about some tension between the lecture and something else you are thinking about. If there is only the lecture, there is nothing to contrast it against. (Unless the lecture is self-contradictory.)
I think classes are a very dumb way to transmit information, and I never not once in my entire life ever paid attention in any class, in either school or college. It's true I wasn't a high scorer in college, but I think that was more down to me having no study habits either.
Anyway, I think you should use office hours. Makes way more sense to have a go at digesting the material yourself, then you will be clear about what you don't understand and can go talk to the professor about it. Having to think up good questions off the cuff is a silly constraint.
When explaining things to other people irl, I've found that it's important to keep the topic grounded in what the other person already understands. For someone with more advanced knowledge, you can be relatively abstract. But for a beginner, you need to be very very concrete.
I don't know you, so obviously I could be wrong. But what I imagine is happening to you, is that the professor is talking quickly about a bunch of abstract topics, and each factoid sorta/kinda makes sense to you in isolation. But because the lecture isn't being related to your extant base of knowledge, the lecture has a "castles in the sky" quality, such that it immediately evaporates when you leave the room. You need the prof to anchor the lecture in concrete examples. So feel free to ask for concrete examples. In contrast, the professor naturally wants to be speedy and abstract, because it all feels boring and trivial from his perspective.
Shankar says "take better notes", and then doesn't elaborate. What he really means is that you should "annotate". "take notes" and "annotate" sounds like the same thing at first glance. But they have different connotations to my mind. When I hear "take notes", it means "write down what the professor says, word for word". But "annotate" means something closer to "just jot down whatever flits through your mind". E.g. if your bio professor is discussing the krebs cycle, feel free to jot down random garbage like "Oranges = citric acid. I ate orange last week lmao." This is useful because it grounds the topic in something concrete.
Meanwhile, Eremolalos says "don't take notes". She says this because most people just take notes to reread them. But rereading notes won't do jack squat, because it only tests your recognition. Recognition won't help you on the exam. *Recall* will help you on the exam. So yes, use flash cards. But you should still annotate. Not to reread them, but for the act of having written them.
Well if it's a small class (<20) then I would suggest asking questions right when they come to you. If you are not following some step in the explanation ask right away. In a big class constant disruptions are... disruptive. and you need to remember where to were uncertain... maybe write it down?
Take notes. Or better ones if you already are. It's easy to let the words wash over you leaving behind an illusion of understanding that fleets like memories of a dream when you next think about it. If you write down each point made as you hear and digest it, in addition to improving comprehension, you'll notice more quickly when it stops making sense.
It's been a few years since I was last a TA, so I don't if the trend I observed has held, but the professor making lecture slides available after, while certainly good and useful, doesn't obviate the need for taking notes yourself.
See my comment near here about note taking - I'd also like to hear about good techniques for this.
Replied to Lucas about this.
I am a big note-taker, although as you say I probably could have a better strategy for taking notes.
Generally I take notes in a somewhat stream-of-consciousness fashion, writing out sentences based on the content of the slides and what the professor says in the lecture, drawing lines between connected subjects in the margins. This seems to serve me well in history courses (my main area of study) but perhaps it is less suited to my more technical courses like biology, which are indeed where I most notice issues with my retention.
Do you have any particular note-taking strategies you recommend for more technical subjects?
Sorry, if you already consider yourself a reasonably good note taker (in your own judgment), I don't have any better advice to offer. What I said is only good for people who don't take notes at all, or so poorly — on the level of "Everyone else in class is taking notes, so I'll have do something too. Let me write down the title of every slide." — that they would themselves agree it doesn't even count.
But I'd say there are maybe three broad styles of taking notes.
- essentially transcribing every word the professor says, or replicating the content of the slides. Frankly, I think this is usually actively detrimental to your learning: when you're doing this, you can't pay attention to the informational content being delivered.
- As a way to keep your attention from drifting away from the topic, where you could chuck everything you've written in the wastepaper basket at the door as you leave class and lose little of value. My earlier advice was of this kind: hear the professor make a new point, understand it, and then quickly scribble a short enough précis of it that you can put your full attention to the next point as it's being made.
- A magical distillation of your state of mind, where a mere glance much later lets you load back perfect recollection of the material as if you had heard it just moments ago. If this is what you're looking for, I bet something close to this ability exists, but I don't have it.
For what it’s worth, I actually am against taking notes in class if it can be avoided. Taking notes is quite a different task from listening to the lecture, and it interferes with paying intelligent attention to the lecture just like sitting there painting with watercolors would.
The thing that helps turns immediate memory of a bunch of info into a solid grasp of it is working with the info. That’s what problem sets and thought questions at the end of textbook chapters are for. You can do a follow-up of the lecture that works the way those things do: Sit down after the lecture, ideally right after, and make additional notes, preferably right on the slides and lecture notes the prof provided. Write relevant stuff you remember right on them.
Doing that will also make you aware of things you don’t understand or remember well. Keep a record of those and ask the prob or TA later, or just look them up somewhere and find the answer.
Also, it’s important to be realistic about how much of a lecture can be remembered. There is no system that will leave you remembering every single detail. What you want to get from the lecture is big picture understanding of the system. Many of the details you don’t really need to remember.
If there are lots of details you actually must commit to memory, for instance the structure of all the amino acids, you won’t be able to learn those just by paying close attention while reading or listening to lectures. That kind of stuff you learn by putting it on flashcards and using something like Anki.
By the way, I’m a psychologist, so am not just speaking from personal experience. Still, I may not be right. Think of my take as one to try on.
Probably depends on the kind of lecture.
If it is difficult, you better pay 100% attention to it.
If it is long and boring, you can spend the extra attention trying to summarize it.
"Taking notes is quite a different task from listening to the lecture, and it interferes with paying intelligent attention to the lecture just like sitting there painting with watercolors would."
...this is what I was getting at with "figuring out how to quickly [summarize the lecture] took all of my focus from actually understanding it". Glad to hear this is common knowledge elsewhere.
Agreed on "working with the info". (As I mentioned, I had the hardest time with English and history, and I think a lot of my trouble was because I couldn't do this like I could with a numerical method or a stochiometric equation.)
Big-picture understanding: another problem I had was recognizing which parts of lecture were load-bearing, as opposed to detail I could discard. I think this was mostly my lack of experience, coupled with different lecturers having widely different organization - some use BLUF, others start with a story and the Important Moral at the end, etc.
As a naturally combative person, my answer is "seek to destroy". Every answer is based in assumptions; find the assumptions and attack them. Finding them takes "why" questions, attacking them takes "why not" questions. I guess if you want to get better at questions in general, make a point of forming at least one of each in every lecture, and increase as fitting.
Take caution in applying this to social events.
This seems helpful, framing it as "why" and "why not" questions seems to make sense.
Perhaps my approach has become entirely backwards, as I've gotten into some trouble in the past for applying this sort of mindset in social settings, but I've (not consciously) adopted a more deferential mode in academic settings.
Sometimes when listening to a lecture you’re left with a vagueness in your understanding rather than an actual question. Maybe you are having trouble formulating questions because you’re mostly having vaguenesses. In that case, ask the speaker to restate the relevant part briefly — or maybe to add a little
graph or something.
> Sometimes when listening to a lecture you’re left with a vagueness in your understanding rather than an actual question. Maybe you are having trouble formulating questions because you’re mostly having vaguenesses.
This seems to match with my experience, yes, although I suppose the crux of the issue is that I'm having trouble noticing these vaguenesses? Or rather, I guess in the past I have noticed these vaguenesses but considered them merely a symptom of my early stage of understanding, rather than a problem to be addressed... I will keep this in mind this semester.
To flatly answer your first question as asked: my go-to is ESR's How to Ask Questions the Smart Way. The Github version: https://github.com/selfteaching/How-To-Ask-Questions-The-Smart-Way?tab=readme-ov-file The original is linked from there.
The catch here is that ESR's guide assumes you know what your question is, and that you need help with how to ask it, and your middle paragraph indicates you don't even know what the question is yet.
The advice that springs to mind is to figure out a way to quickly explain the lecture content to yourself (without speaking, of course, if you're still in the room). The process of explaining often causes questions to naturally pop out. For example, suppose the lecture is about the Java programming language, and the topic that day is the 'interface' keyword. Explain what 'interface' does to yourself, and you might notice that C++ has a different way of handling it that might strike you as easier, and you might wonder why Java doesn't do that. That turns out to lead to a deep detail about the tradeoffs around multiple inheritance that you might find useful when thinking about OO languages in general. (It relates to how method pointers are organized within the compiled Java class file, but it's been two decades since I looked into it, so I can no longer tell the story.)
Despite not quite responding to my specific problem, I found ESR's text enriching, thank you for the link :)
As regards finding a way to quickly explain lecture content... hm, I guess it comes back to better note-taking, as Shankar suggests above. I'd welcome any suggestions for note-taking on more technical subjects.
Aye, as I said, yours is not quite the itch it scratches, but I felt you and other readers might find it germane enough to benefit elsewhere, so I'm glad you liked it.
I was notoriously bad at note taking in high school (my English teacher specifically called me out in public for appearing to do nothing during his lectures when I was just sitting there trying to listen and make sense of what he was saying), and never felt like I'd gotten the knack in college, so I can't be much help there (and also welcome anyone who cares to post about it).
The best I can give is an account of what happened when I tried: it became an exercise in writing whatever the lecturer was saying, but summarized. Talking is faster than writing, so my summary had to be fast enough to keep up. So it mostly became practice for my own version of shorthand. However, figuring out how to quickly build a new convention for shorthand took all of my focus from actually understanding it. And when I already had a convention for certain things, writing the notes typically resulted in my remembering what I heard for a long time, so I also got in the habit of never re-reading my notes, because it would have been a waste of time. I came to regard note taking as a sort of mantra that forced me to keep my attention on the lecture and not daydreaming about something else, and afterward, I could just throw the notes away.
Generally, I could tell that revisiting notes about a subject I wasn't interested in (such as English or history) wasn't going to help - it wouldn't improve whatever essay I was going to be assigned to write - and for subjects I was interested in, I found I learned a great deal more by putting my pencil down and just listening to the teacher.
I elaborate in my reply in another comment, but the mantra thing for focus and recall was what I was suggesting. We might mostly be in agreement, except I think that it'd be helpful even in subjects you ARE interested in.
"[China has] estimated 50% youth unemployment, general unemployment now illegal to publish (but you can still approximate by falling commuter numbers), some government workers haven’t been paid in more than a year (living entirely off extortion/bribery; includes police, teachers, and all healthcare), common experience in private sector to have your paycheck delayed by “2 weeks” that turns out to be 16 weeks. Starting salary offer for software engineer in Shanghai is <1/3rd legal minimum wage. You cannot partake of public services without home “ownership” (hukou system, and technically its a land lease). Private health insurance is overwhelmingly likely to be a scam and rarely used. Public coverage primarily covers TCM; most western medicine is out of pocket. If you cannot afford to pay - even in a trauma situation - you will be escorted off the property so your death doesn’t lower hospital mortality statistics."
I'm glad that I started a discussion that's spanned multiple Scott Alexander posts, but I'd really like to see a source for these claims. 50% youth unemployment is far higher than the 17% I typically see thrown around. Yes, I know we shouldn't believe Chinese government statistics, but that's not a reason to believe arbitrary figures.
The point about hukou is correct, but the hukou system has been around since 1958. It can hardly explain the Chinese people's nostalgia about the 2000s. Similarly, China's healthcare system has always been less than ideal, but the 21st century has seen vastly improved access due to a combination of urbanization, rising incomes, and expanding insurance coverage.
For the other claims, I'd like to see a reputable source. For example, I don't doubt that some wages are delayed, but where's the evidence that wage delays are widespread? Where did the OP hear that the starting salary offer for a software engineer in Shanghai is <1/3 the legal minimum wage?
I'm sorry do you have any references for these claims?
That's what I'm asking for, yes.
An explanation for the discrepancy I saw elsewhere in this comment section is that the 50% figure includes people NOT actively looking for work, while the 17% uses the more conventional methodology of omitting them.
People not actively looking for work aren't conventionally included in these statistics?
Yes.
To some degree it makes sense. There is a difference between someone who wants to be e.g. a stay-at-home mom and never applies for a job... and a similar person who applies for a job but can't get it. For starters, the former is happy about the outcome, but the latter is unhappy.
But things get complicated, because both sides (the unemployed, and the government) have an incentive to distort the truth. For example, if you say "anyone who can't find a job, come here and get some money", the stay-at-home mom would also be eligible to take the free money. But if you say "no, you must be actively seeking for the job", it is easy (and tempting for the politician, to improve the numbers) to set the bar for "actively seeking" too high.
And there is a lot of gray area. Should someone who arrives drunk to every job interview really be counted as seriously trying to get a job? Should someone who needs to have two jobs, but could only find one, be counted as "unemployed"? What about a student who can't find a side job? A university-educated person who could only find manual jobs. Etc. Ultimately, every country counts this differently.
We have in the United States, thanks to the unintended consequences of welfare "reform", a significant number of people who were thrown off the rolls of those eligible for financial assistance to the unemployed because they could not find a job in six months or a year or whatever, and so to pay the rent were driven to join the rolls of those eligible for financial assistance because they are permanently disabled. Not by e.g. missing limbs, but by chronic fatigue syndrome or Long COVID or whatever, trust them, they've got it and they've got a letter from their doctor to prove it.
But in order to keep that going, they have to never get a job or even be seen applying for a job, because then it would be obvious that they aren't really permanently disabled. Do we count these people as "unemployed"?
If we want our unemployment statistics to look good, then of course we take the excuse we have been eagerly offered to say, nope, these people aren't unemployed. But if we want to know how many people could be working but aren't, or even how many people would prefer to be working but aren't, then we're hiding a big chunk of signal there.
Similarly, trying to get as many people to universities as possible is a clever way to reduce youth unemployment. (Well, it keeps the youth off the streets for most of the day, so I guess mission accomplished.)
Yes, that's a requirement in the definition used by the UN's International Labor Organization, the OECD, the US's Bureau of Labor Statistics, the EU's Eurostat. Even if there are countries that internally favor a different measure (Germany seems to be an example of this), any number you're likely to see would be harmonized to match this.
It looks like the BLS actually tracks six different measures of "labor underutilization", from U-1 to U-6, but the one using U-3 is the official unemployment rate.
Do you know why Germany is regarded as having different measures?
From my experience: if you register for benefits while unemployed, you are in active contact with the job agency and therefore regarded as looking for work (and they want to see proof of your continued search for work).
If you can afford it, you can just check out and stop receiving benefits. I would be very surprised if these people even could be counted as unemployed, as no relevant agency knows of their existence or status of employment. They will be paying health insurance out of pocket and any taxes due on capital income or otherwise, and that's it. They are indistinguishable from any freelancer.
It's not a subtle difference: the measure common in Germany (the one in their newspaper headlines, etc.) counts people working, but less than 15 hours/week.
I don't know about China, but that is certianly the case for United States unemployment statistics.
I've been hanging out around LW and here since like 2010, read all the scifi. But here on the eve of the singularity or whatever I'm losing hope. Or just getting cynical with age
I'm starting to think we're just going to have vaguely adequate simulacra of knowledge workers (who were supposed to be working rather than posting on reddit and quora, but progress plateaus once we eat all that low hanging fruit), and we're left in a slop economy. With the same ugly buildings, decaying infrastructure, and an llm fueled social media hellscape destroying our souls and communities. And a lot more wealth inequality
"own a little AI stock and you'll get a moon" sounds like scientologist or bitcoin maximalist ad copy.
1) I doubt we're on "on the eve of the Singularity."
2) Your description of LLMs being essentially "fake AIs" with hard limits on their capabilities is accurate. They will not be able to fully replace humans--not even close. However, the mistake you're making is assuming no further progress will be made building bona fide AGI. I believe it will, and that the current hype over fake AI will have long-term value since it has underpinned the construction of the physical infrastructure (data centers, power sources for those centers, other property and rights-of-way acquisitions) that will support AGI.
If you think AI won't amount to much, you shouldn't believe the "permanent underclass" meme. Wealth inequality sucks, but the solution is just the normal boring politics of taxing and redistribution.
And if you believe AI *will* change everything and make the capitalists fantastically wealthy (in any form, not necessarily moon ownership), then you should buy stocks, because that's how you become a capitalist.
I really do not think it impossible to predict how all this will play out. There is so much at play, and the AI part of it is a novel entity for which the past does not give us an analog. Seems to me that believing things are going to end up taking a certain course is mostly a way to escape hard-to- endure uncertainty. It’s more fairminded and also better for your head to just float in uncertainty
So...we're only going to arrive in Hanson's "The Age of Em", not superintelligent AGI? We're only living in 2016 sci-fi? That's depressing?
I mean, I don't want to be as giant jerk face but six years ago we did not have things like intelligent AIs that could act as junior software engineers or self-driving cars on the streets of SF. Going further back, at some weird point in the 2010s we stopped primarily "living" in reality and started living literally virtual lives, like spending ~7 hours a day staring at screens not including work time. (1) If you and I had put $100k in Google when they bought Deep Mind in 2014, we'd be millionaires.
I vibe with this. I'm shocked by how mundane all this feels and I think that's irrational. A small handful of megacorps are consuming the entire economy (2) and tracking our every movement as democracy slowly retreats. I'm literally living in a 1980s cyberpunk novel...and yet I still poop every day in the same kind of toilet I pooped in 1995 when AOL was a big thing.
I don't know why it feels like this but I am confident, in my heart, that someone is going to solve aging in the next 30 years and the day it's announced we'll be like "Wow" and then we'll all start griping on TikTak of whatever about how this means Social Security and Medicare are doomed forever and maybe we all liked dying and how dare these people make us live forever and I'll still be pooping in a porcelain toilet.
(1) https://backlinko.com/screen-time-statistics
(2) Like, as a share of total S&P 500 value consuming the economy.
> I'm literally living in a 1980s cyberpunk novel
Not really, no. Curtis Yarvin may be a crank, but I think one useful thing he pointed out is that you can tell elections are meaningless because after the transition, you can go take a walk and everything is still the same. Ditto for this situation.
Yes, even in San Francisco. I walked around San Francisco in early 2025, I saw the Waymos, and it's cool, but I mostly remember the homeless.
> and yet I still poop every day in the same kind of toilet I pooped in 1995 when AOL was a big thing
Okay, but real talk here, Toto toilets have existed for a long time, but basically nobody in America has them.
If you want the "toilet of the future," it's buying a higher end Toto, which will automatically open and close, wash and dry you, de-odorize, play music, and do sundry other advanced things to your nether regions entirely dazzling and unknown to 99.9% of Americans. But you can have this toilet of the future today!!
And it is a much more civilized way of being, I highly recommend it.
I also highly recommend.
It can also bring great pleasure in the form of orgasms.
Yeah, the speculations on utopia feel pretty irrational. Like they're wildly overestimating what intelligence can do, or how difficult space travel is.
> and you'll get a moon" sounds like scientologist
More Mormon than Scientology, no?
https://heyjackass.com/ has summarized its 2025 statistics on Chicago violence. Some notes.
-Crime is down. In fact, from 2021, it is down about 50%. Nonfatal shootings are down as well, so it isn't just emergency rooms getting better.
- This is not an official site, and their numbers are a little different from police statistics. I think that the difference comes from the fact that the site counts things like police shootings, self-defense cases, and crimes investigated by the Illinois State Police, not just the CPD. But something to note.
-Unsurprisingly, guns are the preferred method of killing, with stabbing a very distant second, and blunt-force trauma third. Also, one must wonder about the two homicides listed as unknown cause.
-The Austin neighborhood was far and away the deadliest, with 46 homicides. Two others tied for second with 24 apiece. An argument for hot-spot policing in Austin? Still, the days when individual neighborhoods might have 70, 80, or 90 homicides are gone, hopefully permanently.
-Some neighborhoods have unusually high ratios of homicides to nonfatal shootings. Perhaps the ones with lower ratios have better emergency medical services. (One hopes that it is not that other neighborhoods simply are better shots.)
-There was one unusual statistic; Chicago had two quadruple homicides last year, which has not happened in a while. Of those two, one (an arson case) has been solved, while the other (a mass shooting) has not. Interestingly, it had no triple homicides. In 2024, by contrast, it had not quadruples, but four triples. As best I can tell, (which is to say, what I got by Googling it), two have been solved, or at least someone has been arrested for them.
-The face that I keep up with this may suggest that I have too much free time.
https://substack.com/@jasher/note/c-191250583
Jeff Asher is an experienced crime-data analyst who's been publishing very-high-quality analysis on Substack for a couple years now:
https://jasher.substack.com/
https://jasher.substack.com/notes
https://jasher.substack.com/p/the-jeff-alytics-year-in-review
"Before launching AH Datalytics, Jeff served as a crime analyst for the City of New Orleans and Jefferson Parish Sheriff’s Office, and prior to that he worked as an analyst for the Central Intelligence Agency and Department of Defense."
Importantly Asher doesn't rely on any single data source. For one thing he distinguishes between and explains the varying strengths of local/state/federal recordkeeping of reported crimes; also he cross-references with crime-victimization surveys that reflect unreported crimes. For good reasons he is starting to attract some media attention, and everybody who follows him is now telling him to "write the book" (hopefully with a less-cutesy title than 'Jeff-alytics').
Down relative to the height of the Floyd spike in 2021. Compared to 2019 (a low year) it is up 18%. A more neutral assessment is that it's slightly below the 10-year average. The data only goes back ten years so it's hard to make any other judgements from the data alone. Interestingly, it looks like the total population of Chicago has been flat for some time.
>In fact, from 2021, it is down about 50%.
That's mostly reversion to the mean. Homicides went up about 40% post George Floyd and remained elevated until late last year.
https://substack.com/@jasher/note/c-194021875
Yeah individual cities are going to be subject to more random variance than a national cohort. That "lowest since 1965" looks to be about 5% lower than the 2000-2010 norm so I don't think it's fair to say that this represents some secular change. Maybe the 2020-2024 orgy of violence purged all of the bad blood for a few years.
"individual cities are going to be subject to more random variance than a national cohort"....
https://substack.com/@jasher/note/c-191250583
https://jasher.substack.com/p/2025-year-in-review-a-remarkable
Ok so roughly on the national trend. Fair enough. 20% below 2024 levels is 13,500 homicides. There were 13,900 in 2014. That feels consistent with reversion to pre-2020 norms coupled with an unusually law-and-order president.
The current decline began in 2022, following 2020's record rise. I'm trying to remember now, who was president in 2020?
My monthly wrap up of the most interesting long form content covering evolution, history, and the impact of resource limits on civilisation is out and this one is huge.
Highlights include an analysis of drones in modern warfare, the evolutionary history of the human genome, a field report from Swidden farming villages in Thailand, the third and final part of a thesis link gut dysbiosis to schizophrenia, and a great summary of the current unregulated peptide therapeutics sector.
Also check out my interviews with archdruid John Michael Greer, Chinese historian Professor Jiang, resource economist The Honest Sorcerer, and experimental farmers Professor David Shields and Bruce Steele.
https://open.substack.com/pub/recombinationnation/p/the-long-forum-january-2026?r=f45kp&utm_campaign=post&utm_medium=web&showWelcomeOnShare=true
Great!
I appreciate these roundups of yours and John Michael Greer is always a treat so thank you.
I likewise appreciate a little feedback that I am appreciated. Always a bit paranoid I am annoying people when I cross promote my work. Greer had been on my wish list to interview for a long time.
Lately, I have asked ChatGPT some extremely complex questions and I have been amazed at how accurate and thorough the answers are. Embarrassingly, they are much better than I was able to write, in some cases after months of research. This probably outs me as a midwit to the geniuses around here, but so be it.
Examples:
1) Why do women earn less than men?
2) What are the essential components of a free enterprise system?
3) What should I look for in midlength surfboards if I am this size and age and surf this type of waves at crowded SoCal reef breaks?
4) Which audio systems in luxury cars have the best audiophile sound quality?
5) Summarize Streven’s book The Knowledge Machine.
6) How do I reset the AirPlay on this model printer?
I am knowledgeable enough to evaluate that the answers in every case were excellent. I could ask follow up questions, and the system prompted me with some areas of further exploration, which it then went on to answer.
It seems we are quickly approaching an era where I will be able to learn more by asking and discussing issues with AI, than we could in months of frustrating debates and fights on social media, blogs, Reddit, and Substack. Indeed, it seems that we could have a Substack format where someone asks AI a controversial question, shares the answer with the audience and we discuss or query each other and the AI.
Thoughts?
(And no I haven’t asked ChatGPt this …yet)
I actually let Claude talk me into replacing my car (I think it was right, it had a nasty crash last year and I shouldn't have had repaired it, since there's no guarantee the chassis is still safe. Also was a 2011 model, low mileage however). But you always need to be on guard for hallucinations.
That's the opposite of my experience! I've stopped using ChatGPT (admittedly, as a logged-out nonpaying user) and switched to Gemini over the past month or so, due to a dramatic decline in the quality and reliability of the answers I was getting from ChatGPT.
I agree, and I think people who aren't using a good LLM to help them navigate medium-size questions like this are leaving an awful lot of value on the table.
I genuinely don't mean to sound snarky, but if it comes out that way - well, I think I can still claim true and important. Because, well, it appears that the idea of life quality not necessarily being correlated with legible economic markers turned out to be much easier to accept when it was about China.
We’ve talked before about people who are prescribed beta blockers for anxiety, and I said, as someone with a thyroid condition who gets prescribed way bigger doses of beta blockers, I don’t think there’s much of a psychological effect.
Ok, this time when the endocrinology department were slowly ramping up my drug dosages I do feel an effect.
At 40 mg carbimazole + 7.5 mg bisoprolol (daily), my resting heart rate was measured at 155 bpm (yeah, that’s high)
At 60 mg carbimazole + 7.5 mg bisoprolol, my heart rate is about 110 bpm.
At 80 mg carbimazole (the endocrinologist is getting bold here) + 5 mg bisoprolol, my heart rate is about 75 bpm. (Increase carbimazole + decrease the beta blocker as they both reduce heart rate and don’t want to go too low).
So, ok, I’ll admit I felt the difference in psychological effect there. (The carbimazole is doing most of the work by driving down my T4 level). And yes, when the endocrinologist put the dose up to 80 mg I was looking up the medical literature to see if any thyroid patient ever has been given that high a dose (yeah, they have),
I think the focus on "absolute" well-being in the analysis above is misplaced. Maybe people should care about their absolute well-being, but we're social creatures and we tend to be a lot more focused on relative well-being. If the people that own Anthropic stock are actually going to be owning galaxies, the people whose $25k in savings allows them to live a circa 2025 American middle class lifestyle are not going to feel great about it. I guess that's what you're dismissing as "pride" but even a rather humble person engages in some level of social comparison.
And I'm not so sure about this unspecified trickle down mechanism either. There are plenty of people in more remote parts of the world today who haven't really experienced any gains as a result of the industrial revolution.
I dont think its purely envy, though. To some degree we all also have a desire to live a life with purpose, to put it even more favorably: to make the world a better place
If anthropic shareholders own galaxies and you subsist of scraps to maintain a comparably prehistoric lifestyle, its not so much "they have shinier toys than me" but "they have more power and capacity to shape the future than me"
Theres also ego and pride, which of course yes are vices and demand less sympathy. But its really not just one vector. Theres a lot of reasons to be unhappy with inequality, particularly if it is built on your labor/consumption
>we're social creatures and we tend to be a lot more focused on relative well-being.
Yes: which is why the envy remains a constant temptation towards sin, no matter how rich we become. Still a sin though!
> AI company employees will form the new ruling class, with everyone else as serfs.
Surely the idea is that AI company CEOs will be the ruling class. If the people they have rendered unemployed are dependent on them for a voluntary UBI, they can use their votes to get into power. Well, there are many possibilities and one of them looks like being neo-feudalism.
Paul Graham: The rational fear of those who dislike economic inequality is that the rich will convert their economic power into political power: that they’ll tilt elections, or pay bribes for pardons, or buy up the news media to promote their views.
I used to be able to claim that tech billionaires didn’t actually do this — that they just wanted to refine their gadgets. But unfortunately in the current administration we’ve seen all three.
It’s still rare for tech billionaires to do this. Most do just want to refine their gadgets. That habit is what made them billionaires. But unfortunately I can no longer say that they all do.
Can anyone steelman the argument that some AI company is going to become a world-eating economic behemoth? Other than NVIDIA, I don't see anyone having a substantial moat. There's no secret sauce and all the models seem to have roughly equivalent capabilities. Are there network effects that I'm not seeing? Right now this seems like a commodity business to me.
Something I read that impressed me: Companies are currently interdependent, renting and buying services, cloud space and materials from each other, and all are dependent on chip manufacturers and no doubt on providers of other components and services. Some companies are now trying to develop vertically -- making their own chips, developing their own data sources for training, etc. If one succeeds *enough* to be markedly less dependent on other companies than the rest of them are, there will come a tipping point where it is so much more financially independent and efficient much that it can afford improvements that others can't, and so on til it's a world-eating behemoth. I am not knowledgeable enough to rebut rebuttals to this -- just had a gut feeling that it was a smart take. Oh yeah, also, Musk is doing the vertical development thing. (And he has Twitter as a source of training data.)
Musk does vertical development because he can't find anyone willing to sell him what he wants to buy, e.g. rocket parts that work but haven't had their price jacked up by 10,000% because they come with lots of paperwork "proving" that they work 99.99999+% of the time. He has said that he would prefer not to do this, because he's got finite bandwidth and just cutting a check is more efficient.
I think a rebuttal to this is that it is not clear that it is more efficient to do all of the supply-chain internally. For this to be more efficient, I think the company branch doing part x must in principle outcompete all of the other external companies doing x minus overhead of doing buissiness and whatever margin the external company can extract over time. This seem to sometimes but not always be possible, from the evidence that companies sometimes seeks to internalize some process, while other times seeks to outsorce processes. Probably, internalizing a process is mainly a good idea if there is little competition externally in that area (maybe due to barriers of entry)? Otherwise your internal branch does not have the same pressure to perform from competition, and may rather become less efficient at least in theory.
The assumption is that the AI at some point becomes actually useful to the point of being able to automate most if not all economic/industrial processes, and in addition actually creative to the point of being able to come up with novel technological improvements.
If you think current state of the field makes this assumption very far-fetched - yeah, you're not alone. But if you do assume it, economic domination follows quite naturally.
The question of how much economic value it generates vs how much AI companies can capture are very different
I've had the same idea, it's looking very much like a commodity to me.
I do assume it. But I also suspect, based on current evidence, that it will be a commoditized technology and so no single company will be able to extract outsized rents.
> Can anyone steelman the argument that some AI company is going to become a world-eating economic behemoth?
I've got another one besides the "white collar work" one - there's the potential for the AI chatbots to become the next trillion dollar attention surface for auction, much like smartphones made Google and Meta into trillion dollar companies, by grabbing enough eyeball hours.
There are already ~1B MAU using LLM's today, and that's before they live in everyone's ears and become personal assistants that can make phone calls and answer emails and curate your media feeds for you. When that happens, it's the next smartphones - people are going to be interacting with them all day long, intermittently, and trust is going to be high, so ad placement value will be correspondingly high. They can also charge for more capable personal assistant tiers.
And much like people went nuts for GPT 4o and mourned the loss of their AI-werewolf-vampire boyfriends and whatever companies will have differentiation in terms of the personalities of the chatbots, just like all the chatbots have distinct personalities and capabilities today.
Agreed! Look, I'm not arguing that AI won't be spectacularly useful, I just suspect that the technology will be a commodity. The only reason Facebook is so valuable is because of network effects. Maybe LLM-producers will be the future equivalent of hosting providers and the real moats will be for the downstream applications. Look back at the late 90's: the early behemoths were in a bunch of different markets. You didn't see Amazon fighting with Yahoo for market share. I just don't see any product differentiation between LLMs yet. Maybe it'll come but it hasn't yet.
1) Data: Models are already training on nearly all easily accessible, public data. As use of AI scales up, user input and feedback become a more and more important source of data to build the next model. This exists for chatbots, but is especially pronounced for real world devices with sensors. If Company A has 24/7 data from millions of robots and Company B has data from 6,000 of them, Company A will soon develop an insurmountable edge.
2) If we're on a path to fast progress, then someone will eventually develop the first AI capable of rapid self-improvement. As it self-improves, they'll develop an impossibly large lead over whoever is in second place.
3) If we're on a path to slow progress, then AI will get integrated into all kinds of different systems and you'll start developing lock-in where it doesn't matter if Company B has a somewhat better product than Company A because you've built up so many dependencies on the way that A's technology works.
1) User feedback is available to anyone with users. There are many AI companies with users. For a given amount of compute there is a data saturation point where the marginal utility quickly falls to zero. So having 10m vs 1m users seems likely to be immaterial.
2) I don't see why that wouldn't be quickly replicated. It's just another capability and we've already seen that capabilities are quickly copied between models. Capability is gated by absolute model performance and that's largely just a question of compute. I don't see the moat there.
I'm not even sure "rapid self-improvement" is meaningful. The limited data we have so far indicates that all that matters is compute + data. ML expertise doesn't really matter, so I'm skeptical that ASI would help. You don't need ASI to tell you to use more GPUs.
3) Yes I can see some lock-in at the application layer. But that seems like it would be limited to market segments: one company might own legal AI and another might own coding or whatever. It's hard to predict how the ecosystem will shake out but I don't see any obvious natural network effects. There just isn't any secret sauce: it's all downstream of compute. If one company develops some hugely profitable capability then that just becomes a signal for capital to rush into a competitor to train it up.
1. We're already maxed out on data; while we face compute limitations at any given moment, the availability of compute is expanding much more rapidly than the availability of data. User data is also uniquely useful; that's why Google has been dominant in search for so long despite having no technological moat.
2. Fair enough to not believe in recursive self-improvement, but if it does happen it'll be very hard to quickly replicate because it turns a small lead into a big lead overnight and everything before just becomes irrelevant.
3. By this logic, IT in general ought to be a low-margin industry, which is obviously wrong.
1. Google absolutely had a technical moat for several years. I don't know if you're old enough to remember the internet in 2000 but Google was the only search engine that didn't suck. The difference was enormous and lasted for several years. That technical edge didn't last (Bing quality is indistinguishable now) but Google established a brand and built a platform (Chrome, Android, gmail, etc) around it and that created user lock-in. I don't think it's fair to say that their access to user data gave them an insurmountable edge because Bing surmounted it.
2. Maybe. I don't think technical evolution works like this without either market network effects or a patented key technology, and patents don't have much of a track record in software. Maybe there'll be some secret nonobvious breakthrough that one company stumbles across that's hard to duplicate but, as I said above, that's not the pattern we're seeing at all so far. Maybe Anthropic or someone will come up with the AI equivalent of PageRank and become the Google of AI for a few years, but I tend to side with this article in the "it's nothing but compute" camp: https://samuelalbanie.substack.com/p/reflections-on-2025
I think this probably comes down to how we model intelligence. I don't think it's some unitary thing that once you discover it it's game over. I think it's agglomerative and every AI company will slowly accrete raw horsepower until they eventually converge at the same saturation point where marginal compute doesn't yield positive-economic-value improvements in model performance. I could be wrong but that's what the early returns suggest so far.
3. It's a lot higher for software than it is for hardware (NVIDIA notwithstanding). That's because AMD vs Intel are interchangeable and so have to compete on price, whereas network effects create winner-take-all dynamics for the application layer: Facebook, eBay, etc. I think it's possible that LLMs (or whatever replaces them) will become interchangeable cheap-marginal-cost commodities. The margins go where the bottlenecks are and for AI right now that's at the GPU.
The first company that invents fully-autonomous and loyal murder drones will be able to take over the world in a fortnight.
Not really in the spirit of the question, but how will those drones stop the ICBM that gets launched at corporate HQ the instant a nuclear power perceives an existential threat?
You Zerg rush all the nuclear missiles with the drones before they can launch.
There's already an article about why this is unlikely.
https://www.navalgazing.net/The-Drone-Revolution
Interesting read. I think his analysis held up well in 2023 when he added his first-year Ukraine postscript, but I don’t think it rings as true today. Drones keep advancing to fill new niches which conventional weapons can’t fill as cheaply. Some of that is because Ukraine doesn’t have access to as much high-end conventional munitions as they’d like, defensively as well as offensively, but I still think drones will play an increasing role in warfare.
What if someone full-walls?
Your underwater drones are, of course, shadowing the nuclear submarines that are under the polar ice.
Though, to be serious, I think you need industrial production capacity that only a few countries have to be able to drone strike the entire nuclear detterent.
> Can anyone steelman the argument that some AI company is going to become a world-eating economic behemoth?
There's a multi-trillion market for the first company to put together the right framework, memory, persistence, prompting, and long term learning to do a given white collar worker's work.
Given the models are already smart enough, it's all the other stuff that needs to be knocked out. But when that happens, the company that does it can net trillions, and it seems like it will probably have a fair amount of tacit / individual / trade secret style knowledge in that framework, so there's room for multiple companies to tackle different jobs / segments successfully.
>have a fair amount of tacit / individual / trade secret style knowledge
I'm skeptical about that part. The short history of LLMs has shown so far that there really isn't any secret sauce. Anything that one model does another model quickly copies. "The performance of AI machines tends to improve at the same pace that AI researchers get access to faster hardware."
At best I see a temporary first-mover advantage before any unique capabilities are reproduced in an open-source model ala Windows vs Linux. Not that Microsoft isn't still very profitable but it's effectively lost the battle with Linux in terms of what runs the most CPU cycles.
This seems right to me. I think some people who believe in certain types of AGI concept think there is a phase transition where passing that transition even a few weeks before a competitor gives you an insurmountable advantage. But the real life experience of interacting with jagged intelligence is making me more skeptical of the kinds of AGI concept that are needed to make that work.
(disclaimer: I'm not totally familiar with the details of every major prediction market's operations, so if they're already doing this and I'm reinventing the wheel, then I'm surprised in the direction of them being more thoughtful and competent than I thought, and will update my impression of them positively)
When the question resolves based on a singular event that could happen at any time, should prediction markets invalidate (refund) any predictions made within X days (1? 7?) of that event happening?
Inspired by the Maduro thing, but as a general concept: One of the objections to prediction markets is that when one person/group is making a decision that can single-handedly resolve it, they can just bet as much as they have on the correct outcome before they do it. As far as I know, the response to this has been that them betting strongly is still useful information, so it still makes the prediction market useful in predicting the right answer, and it still rewards good predictors because they should take this behavior into account.
But I think this is still a problem, particularly in case where you can make a bet one hour/minute before doing the thing that resolves it. The information provided by the bet is not useful to society if it arrives too late for anyone to react to it in a useful way, and incentivizing people to be good at predicting how primary actors will react to the perverse incentives created by the market existing in the first place is I think (?) definitionally less useful to society than the market just not existing and not creating perverse incentives in the first place.
So maybe the market needs more time to react to the information from such bets before they are resolved. If everyone has time to realize that a primary actor is betting in their own market, they can all change their bets to agree with that person, lowering that actor's payout and thereby reducing the perverse incentive for them to manipulate the market by changing their action. And then society can see a market going to 99.9% certainty or w/e, and actually have the type of strong predictive power we want these markets to provide to society, with enough time for that to matter.
Naively, it seems like we can accomplish this just by saying 'we refund all bets placed within a week of the event recurring, so if you're a primary actor you have to place your bet at least that far in advance if you want it to pay out.'
(note - not advocating this as a specific policy for all markets, it will not apply to many and different time limits may be appropriate for different questions, etc.)
> If everyone has time to realize that a primary actor is betting in their own market, they can all change their bets to agree with that person, lowering that actor's payout and thereby reducing the perverse incentive for them to manipulate the market by changing their action.
Thats not how it works.
First, because prediction markets work by buying and selling papers which pay 1$ if the market resolves X. People trading after you dont change your payout, if the insider just does his trade and sits on it, he has the money in the bag. There is a traditional betting system that works like you describe, but it incentivises everyone to hold their bet until the last possible second, and then act based on their guess of what the other bettors will do.
Second, because by your own rule, the trades others make to react to the insider would also be reversed.
>There is a traditional betting system that works like you describe, but it incentivises everyone to hold their bet until the last possible second, and then act based on their guess of what the other bettors will do.
But you can't hold your bet to the last second unless you know when the event will happen.
If there's only one person who knows when the invasion will happen, then it's fine for them to wait until 1 week before to make their trade, since the alternative is the current situation where they wait until 1 hour before.
And I also think it's fine for them to be able to make a profit and for other people who follow the leader to get their trades reversed. A prediction market is theoretically *supposed* to incentivize insider trading, since that makes the market more predictive by leveraging nonpublic information. What we want to incentivize is bringing that nonpublic information forwards soon enough for other people to act on it (in the real world, not just in the market).
Yes, this happens in the typical context where youre betting on some sports event, and theres a definite time you stop taking bets. Here it would only apply to the insider. And Im only talking about the profit-reduction, not disputing all possible benefit to the time spacing.
On the Vibecession and economic inequality point, I am surprised more people are not talking about the 'Captain Condor' blowup a couple of weeks ago. It is an obscure story, but it is an interesting lens to try to unpack what is going on in America with gambling and using 0DTE options to eke out a living. They ended up losing like $50 million over three days using iron condors on SPX options and doubling down (martingale) every day based on faulty trading strategy. But it is such an absurd story that a group of 20-30 year olds can accumulate that much money in the first place, implement a ridiculous strategy, sell it to others, and then lose it all in a couple of days. Even explaining what they were trying to do would not make any sense. But in this economy, their ability to sell people on a fool-proof trading strategy where you can sit in front of a computer and blast off trades and make consistent money appealed to a lot of people that they just assumed it all worked. But no one took a step back and said this is real money we are playing with before losing it all. It's a really sad story and representative of the reality we are living in.
A fool and their money are easily parted. But how is it representative of the reality we're living in? I guess never before it was this "easy" to hack your way into wealth (though I guess they found out it wasn't actually that easy...)
I put together an end-of-year pathogen update over on X. A Threadreder link at the bottom of this post.
1. SARS-CoV-2 / COVID-19:
The US winter COVID wave began around Thanksgiving. It's interesting that the wave has started off by affecting different sections of the US unevenly. There's been a strong upsurge in wastewater numbers in the Midwest and East. Not so much in the South, and wastewater levels in the West are still below the previous interwave gap. I considered that this pattern may have been due to higher COVID circulation levels during the previous interwave gap, giving the West a higher population immunity against the current variants, but I had to discard it because both RSV and the flu are showing the same patterns.
It looks like XFG.14.1* may be the primary driver of this wave. CoV-Spectrum and CDC Nowcast agree that it’s been rising, and is now at ~15% of the sampled sequences. But the rest of the Clade 25C clan (which encompasses all the XFG variants and descendants) is still resisting cousin 14's aggressive onslaught. As a group, the rest of Clade 25 has been holding steady at 70% and 80% of samples for months now. There were worries that BA.3.2 would trigger a winter wave, but they didn't materialize. Now the experts who predicted a BA.3.2 wave are claiming it will drive our 2026 summer wave. SARS-CoV-2 has done a very good job at eluding our predictions. I'm surprised that experts haven't been more humble about making new predictions.
As of epidemiological week 51 (2 weeks ago), COVID hospitalizations were about 40% of what they were this time last year. And deaths were ~74 per week as of a month ago. The CDC hasn’t updated these numbers since 6 December. Have they stopped updating their handy COVID death rate graph? It's going to be a pain in the ass to dig into Wonder to get this data.
The Long COVIDians have become less vocal (or maybe they’ve all blocked me). But a graph from FRED that shows a ~45% rise in working-age people reporting a disability since the beginning of the Pandemic came across my feed. I hadn't seen this graph before, and I wondered if other sources supported this? First, even though the graph hadn't been altered an came from FRED, other data from FRED doesn't jibe with the ~45% rise in disabilities. According to FRED's *total* US population reporting a disability, as of EoY2025, an est 10% of the US population reported a disability—up from 8.9% in 2009. Approximately half of our 340 million population is in the US labor force (~170 million workers). Either the labor force numbers are too high, or the population numbers are too low. These are self-reported by surveys, though, and the disabilities are not necessarily severe enough to prevent people from working. Plus, we have an aging population. Roughly 24% of the US population was over 55 in 2009. The current estimate is that something over 30% are over 55. Disabilities rise with age. But that wouldn't explain the 45% gain in self-reported disabilities among the working population. When looking at these charts, it's important to remember that FRED isn't telling us how many people are UNABLE to work due to disabilities. Moreover, FRED primarily extracts disability data from the US BLS and BEA. The two charts may use two different data sources.
And patterns seen in the FRED data don't jibe with Social Security disability claims. They have been dropping since 2024. So the data is far from clear-cut that COVID has significantly increased disabilities in the US.
2. Influenza:
Influenza cases are rising steeply. 25% of tests are coming back positive. COVID and RSV are still below 5%. BioFire's proprietary Syndromic Trends show that A(H3) influenza has climbed to the top place at 17% of respiratory infections, rising from less than 1.5% in mid-November. A(H1-2009) is 1.4% and influenza B is at 1.5%. The K variant of A(H3N2) is driving the 2025-26 wave. Biobot’s wastewater numbers show that the West, along with COVID, lags behind the rest of the country in influenza circulation. Roughly 25% the Northeast levels and half the Midwest levels.
The Australian flu season peaked in July (their winter). But H3N2 still lingers on into their summer Down Under—at surprisingly high levels. This doesn’t bode well for the US since we usually follow in the southern hemisphere’s footsteps when it comes to influenza. The good news is that the current H3N2 strain isn't sending people to the hospital like last year's flu strain did.
3. RSV:
And the rates of RSV are also rising, but it's lower at this time this year the previous 10 years—except for the 2020-21 RSV season, when we had no cases of detected cases of RSV.
4. Highly Pathogenic Avian Flu (HPAI):
The HPAI (bird flu) season has been relatively mild so far. Poultry and livestock incidences remain low. But it’s circulating in the wild bird population. In December only one dairy herd tested positive for HPAI. There were 77 poultry flocks infected during the month of December. I suspect these are mostly small commercial or backyard flocks, because I can't find any reports of mass culling. At this time last year, between 30 and 45 million chickens had been culled due to HPAI. OTOH the EU has reported over 700 cases of HPAI in its poultry flocks, and mass culling is going on over there.
AFAICT, there aren’t any active human cases in the US now—probably because it's not circulating at high levels in livestock, which are A(H5)'s primary vector into humans. But the WHO says we had another HPAI fatality this past November, for a total of 2 deaths in 2025.
5. Measles:
In the US measles made a comeback late in the year (while my updates were on hiatus), pushing the total confirmed cases up to 2065. 11% were hospitalized. And there were a total of 3 deaths. Cases were concentrated in Texas, New Mexico, Arizona, Utah, and South Caroline. Nebraska also got hit hard.
And it's worth noting again that although SARS-CoV-2 can cause immune dysfunction for 3-6 months in severe cases, measles can impair up to 73% of immune memory for multiple years.
Canada got hit worse than the US. They had a total of 5377 confirmed measles cases and 2 measles-related deaths in 2025. Both occurred in infants with congenital measles who were born prematurely.
Mexico had 6050 confirmed cases of measles in 2025, and the Mexican Ministry of Health estimates there were probably another 15,000 undiagnosed measles cases. Currently, they’re in the middle of a second surge that started in November. Their 2025 death toll was 24.
6. Some thoughts and predictions:
SARS-CoV-2 has been circulating in human populations for 6 years now. If we break out the mortality by continent, we see asynchronous waves of deaths across continents, but majority happened in the first 18 months of the pandemic. But looking at case numbers, the majority occurred after Omicron's advent in the last months of 2021. Greater than 9 billion doses of vaccine had been administered by January 2022, when Omicron cases peaked. There is strong evidence that Omicron was less lethal, but vaccines stopped the COVID death juggernaut.
The WHO says there were 7.1 million confirmed COVID-19 deaths worldwide, but their excess mortality data suggest that there were closer to 15 million deaths. Other studies suggest higher figures (as high as 40 million). I wouldn't be surprised if it were ~20 mil. But we'll never know, though.
In the US, the 2024-25 respiratory season was the first time lfu deaths outnumbered COVID deaths, by a factor of 2–5, depending on the upper range of the flu burden estimate. I suspect we'll see the same pattern this 2025-26 season—but more so.
Why isn't COVID killing us the way it used to? Studies have shown that SARS2 infections (and vaccinations) induce durable immunity. Memory B cells, virus-specific T cells & long-lived plasma cells remain detectable for years.
https://tinyurl.com/2fpfzwsn
And our immune systems retain the capacity to mount rapid responses to reinfection. Moreover, antibody response improves with subsequent vaccinations and infections, becoming more focused on conserved elements of the virus with re-exposure.
https://tinyurl.com/yc38jn83
Although neutralizing antibodies drop off 3-6 months after vaccination or infection, they decline and then level off in a biphasic manner. B cells that recognize the S protein to which they were exposed not only persist, but they also increase the breadth of their affinity through somatic hypermutation. And studies have shown that CD4⁺ and CD8⁺ T cells generated by SARS-CoV-2 vaccines and infections remain readily detectable for>24 months post-infection, providing protection against severe disease even as neutralizing antibody levels decline.
https://tinyurl.com/yc445kwb
Even though SARS-CoV-2 can reinfect us because its spike protein mutates rapidly, our immune systems are (for most of us) staving off hospitalizations and death.
Will SARS-CoV-2 become another seasonal common cold virus like HCoVs 229E, NL63, OC43, and HKU1? I used to think so, but I'm less sanguine now. Early in the pandemic, some epidemiological models predicted that in 3 or 4 years, SARS-CoV-2 would transition to a seasonal virus that would follow winter waves like other respiratory viruses. But it's been around 6 years now, and those models were clearly wrong. COVID-19 has settled into a pattern of twice-yearly waves in the northern and southern hemispheres, and it continues to circulate at higher levels between waves than influenza and RSV. Although durable immunity has made SARS-CoV-2 may now be less deadly than the flu, it is still deadlier than any of the common cold HCoVs. And it circulates at higher levels than common cold HCoVs. Being more deadly and more transmissible that common cold HCOVs, I don't see it becoming a common cold virus any time soon.
Threadreader unroll here... https://t.co/ZM9d7aZU1g
It looks like I may have maxed out the wordage that Substack will accept in a comment. I couldn't get it to save my concluding paragraph...
Thanks for reading and commenting on my updates. Although initially grim, seeing the COVID-19 pandemic evolve year after year has been fascinating. I may chime in with quarterly pathogen updates going forward, but for now, have a safe and happy new epidemiological year!
Thanks for all this! I’d be interested in reading the updates when you do them - can you reply to this comment when you do so I can see them?
Is there a chance that lower RSV numbers this year are due to vaccination, given that the vaccine wasn’t available until the last year or two?
> Is there a chance that lower RSV numbers this year are due to vaccination, given that the vaccine wasn’t available until the last year or two?
Hard to say, because uptake has been so uneven. No doubt it helped, but I'm afraid I was too lazy to see if I could detect a pattern.
I'll try to reply to this comment when and if I put together a new update.
Thanks!
> For example, after 100x growth, anyone with $25,000 in the stock market now would have $2.5 million
The incentives are very strong for AI leaders to find a way to get rich while giving stockholders nothing. Either shift all the profits to some privately-owned entities, or use AGI to found new companies while letting the old ones rot. In game theory terms, this is Iterated Prisoner's Dilemma: cooperate in the beginning, betray at the very end.
And if they can't find a legal way to do it, they'll try to bribe the politicians to make it possible.
>The incentives are very strong for AI leaders to find a way to get rich while giving stockholders nothing.
The incentives were very strong for the robber barons of the Industrial Revolution to do the same, yet society became 100x richer regardless.
I knew someone would say this, as if people haven't written essay after essay about how modern first-world economies need educated middle class peasants and AGI economies do not.
What's the point of even posting in these open threads, the replies are worse than nothing.
I don't see the relevance of your comment. If AI is like the Industrial Revolution, and our economy becomes 100x larger and more productive, then it will be pretty hard for "AI leaders" to keep all the gains for themselves and prevent stockholders from profiting. Considering the massive sums of money that stockholders have invested in AI, bilking them out what is due to them seems like a tall ask, even if "AI leaders" have an incentive to do so.
Not to mention, if you looked up what the Iterated Prisoner's Dilemma is, you'd have your answer there too ffs.
Question about AI 'art': what's the current best, low-entry, relatively low-cost image generator?
I loved Reve in its preview mode and used it extensively, but the full-release version kinda sucks (doesn't follow instructions as well, frustrating chatbot style instead of just going from prompt to image). ChatGPT is decent if you prompt well but quirky and has that graininess. Haven't messed around much with Gemini; is that my mistake?
Assume (by local standards) fairly low technical skills. I can follow instructions to, say, perform basic repairs on a Kindle or install CFW on a 3DS; anything that requires "programming" is right out. Not using it for anything commercial or too complex, mostly generating little comics from my kid or stickers/game card type things.
Perchance is my favorite for messing around since it doesn't require an account or fees. I don't think it would work for generating a comic unless you do some photoshopping afterwards (it can't do text), but it's good for "filler images" where you need something pretty but don't have strong opinions about the composition.
https://perchance.org/ai-text-to-image-generator
I mostly start with Reve + LMArena (random models) online, then Stable Diffusion XL offline. TensorArt with Qwen is also a great source of free online image generation.
Note the order. Generate images that are sort of like what you want online, then refine them with Stable Diffusion XL. Stable Diffusion XL actually remains a very strong model. It's very blatantly the best for its size, and I find that it produces prettier images than the competition. It does lack prompt understanding compared to the larger models, but that's why you start with something like Reve or Qwen, then image to image with Stable Diffusion XL.
To be clear, do not get confused and download the original base version of Stable Diffusion! Stable Diffusion XL is strong because of its extensive community development. Go to CivitAI and download a modern version of Stable Diffusion XL. I use Arc Naked Singularity, but that's my own model so... if you sort by most downloaded, there's Juggernaught, which is famous and popular.
A question for musicians. What's the best way to learn to identify the chords and the tonal center of a piece of music by ear? Use case: during musical jams, to be able to figure out what to play quickly. Current skills: playing harmonica for 3.5 years and electric bass for half a year; in an ear training app can semi-reliably identify octave, fourth, fifth, and both thirds; in any blues-adjacent tonality I can usually identify the one, the four and the five chords; also I frequently hear some piece of music and can recognize that something in its harmony is similar to the harmony of some other piece of music I know (e.g. recognize that a song has the same chords (up to a transposition) as Get Lucky by Daft punk; or recognize that the first 3 chords in a cycle are the same as in Pachelbel's Canon, i.e. 1, 5, 6m).
I'd suggest you learn some basic guitar chord shapes. That way you can "read" the other dude's hands to know what to play, at least in-terms of harmony. Moveable shapes aren't significantly more difficult than "Cowboy Chords."
Most of what you're asking for is a question of experience. You listen to a few million I-VI-IV-V songs ('50s rock and pop) you'll recognize that progression and eventually even when someone like Sedaka or Carol King puts a dozen key changes on the middle 8
Agreed here. I'm a bass player that started on guitar so I can read the hands of the guitar player which gives me a huge start on being then able to play what I want "from ear" but with a starting direction.
The other thing to do is play with others as much as you can. No amount of playing with recordings with have the same value of jamming with real players.
Good Luck!
It sounds like you can already do the thing you’re asking about if the song (its chords and chord progression) is fairly simple (as it is in Get Lucky and Pachelbel). Keep doing that, and gradually introduce songs with greater complexity, and I think your skills will grow.
I recommend trying it out with Nirvana, because their songs, and the chords themselves, are very simple but they often used rather unusual chord progressions, so it’s a good way of teaching your ear what it sounds like to go from, say, I to III to VII to flat-V or whatever.
Then you can just sit at a piano and toy around with chords that are less obvious than major, minor, or blues chords with the major 3rd and minor 7th. (If you don’t have a piano, you could play the chords way up the neck on bass, or just listen to songs that you are certain have those chords in them.) Like, major 7th and add-9 are good ones to start with—teach your ear to identify those. Then move on to more complex stuff.
This might be app you’re talking about, but I used to work on this free app called Chet https://www.ensemble-education.com/chet that we conceived of as “Duolingo for Ear Training”
You're doing pretty good there; I've played with low-pro-level musicians who couldn't do any better. Also some drummers...(rim-shot)
Stick with the ear training app and keep playing and you should find that this narrow skill does keep gaining, albeit maybe not fast or suddenly. That's my quick reaction anyway.
Following the capture of Maduro, both the US and Western European op-ed columnists and Pravda are citing Thucydides’ Melian Dialogue: “The strong do what they can; the weak accept what they must.”
In the West, the consensus is that the post-WWII rules-based order —sometimes cynically ignored but over time ratcheting towards something more just than just the law of the jungle— is a net positive.
Pravda just shrugs, saying the Western rules-based order was always a useless sham. Admittedly, Russia has its own post-Soviet Union reasons for feeling this.
I’ve been following commentary in Pravda for a while, and they really like Trump’s approach, expecting him to peel nations away from the European Union into a new, more malleable alliance with the US and a certain other party interested in regional hegemony.
Trump is now feeling his oats, now talking about Colombia being next while also mentioning Mexico. He is also reviving his wish for the US to take Greenland from the Danes. Of course, this could just be Trumpian trash talk,but that is the big problem—the man throws out boastful lies like most people breathe. Who TF knows what is next.
>I’ve been following commentary in Pravda for a while, and they really like Trump’s approach, expecting him to peel nations away from the European Union into a new, more malleable alliance with the US and a certain other party interested in regional hegemony.
This sort of sounds like another Cold War, using proxy wars in third world nations in the place of diplomacy.
I don't think that worked out well the first time.
Pravda had another piece praising the efficacy of the US extraction of Maduro with an explanation for why Putin couldn’t be snatched in a similar way. I’ll have to paraphrase from memory here, “Any country trying this would see their skies light with nuclear blasts.”
The same piece explained that Zelensky wouldn’t be grabbed from Ukraine for ‘pragmatic reasons’.
>Pravda had another piece praising the efficacy of the US extraction of Maduro with an explanation for why Putin couldn’t be snatched in a similar way.
Considering that Maduro was mostly protected with Russian military tech, I find that pretty funny.
>The same piece explained that Zelensky wouldn’t be grabbed from Ukraine for ‘pragmatic reasons’.
The pragmatic reason being that doing so would have no effect on the outcome of the war?
It was obvious face saving BS.
If they could grab Zelensky the way the US grabbed Maduro they would do it.
The article also mentions that Putin had promised not to kill Zelensky.
To quote our current president regarding another Putin statement, “I don’t know why he would lie.” Putin must have promised very strongly.
Again, I don't believe they would grab Zelensky, nor kill him, because there is no military advantage in doing so.
Why wouldn't there be an advantage to killing or capturing the leader of a country you're invading? With the leader gone, I imagine there'd be an increase chance of chaos on the invaded country's part. Increased chaos -> worse defenses, as well as decreased morale.
You may have more detailed knowledge than me. My first thought was of the demoralizing propaganda the Russians could develop if nothing else.
What I hate about this post is that: Many of the people I know live pay check to pay check. Of the few who do have savings, the money is in their house and then car, and then some bank CD account. They don't want to put any savings in the stock market, because mostly of fear, I think.
Talk to your friends about compound interest.
Oh there's lots or things they could do to save more money.
Buy a used car/ no car payments
Live more frugally. (lots of things in this bag)
Do they really, though? Many econ bloggers have busted the "most americans are living paycheck to paycheck" myth. If you own a house and a car, and are savvy enough to have a CD account, it seems very unlikely you are truly living paycheck to paycheck. Maybe they spend most of their *current* paycheck before they get their next one, but if they got a financial shock I suspect they'd be able to trim some fat from their budget and absorb the blow. I think people want to believe they are living paycheck to paycheck, since it gives them a financial source for various forms of existential discomfort.
When faced with empirical data on how rich you really are (https://www.givingwhatwecan.org/how-rich-am-i) it gets a lot more uncomfortable to feel bad about your material situation.
Measures seem to vary a little bit, but according to Federal Reserve data only 58% of Americans have direct or indirect exposure to the stock market (https://www.pewresearch.org/short-reads/2024/03/06/a-booming-us-stock-market-doesnt-benefit-all-racial-and-ethnic-groups-equally/).
Sure not most Americans, but perhaps a large fraction who are my neighbors and/or co-workers. A lot of jobs out here in the rural area's don't pay so much. Maybe not pay check to pay check, but not much extra.
Nonetheless there's a lot of people with nothing in the stock market, and very little in savings.
Well, I lost almost a million dollars during the Dotcom bust. I've scratched that back, but not because of stocks. Granted, stocks over the long term do very well, and I wish I had just held on to my devalued stock portfolio, but things looked grim, and I made the mistake of cashing out at a significantly lower level than their peak value (about 10% if memory serves me correctly). Anyway, stocks are not guaranteed, and it's pretty frightening when you go from being a paper millionaire to being unemployed and looking for a steady paycheck. Just sayin...
Between AI and the vibecession / fertility crisis, I feel it doesn't make sense to save for retirement if you're under 40. Either one of those is going to dramatically change how things work, to the point money will likely be meaningless by the time someone under 40 turns 65.
What do you think?
I don't think money will become meaningless – it's just too useful. And even if money somehow would become meaningless, I wouldn't expect wealth to become meaningless. So I'll keep saving and investing.
I saw this Feynman quote the other day. I think it's apropos...
“I returned to civilization shortly after that and went to Cornell to teach, and my first impression was a very strange one. I can't understand it any more, but I felt very strongly then. I sat in a restaurant in New York, for example, and I looked out at the buildings and I began to think, you know, about how much the radius of the Hiroshima bomb damage was and so forth... How far from here was 34th street?... All those buildings, all smashed — and so on. And I would go along and I would see people building a bridge, or they'd be making a new road, and I thought, they're crazy, they just don't understand, they don't understand. Why are they making new things? It's so useless.
But, fortunately, it's been useless for almost forty years now, hasn't it? So I've been wrong about it being useless making bridges and I'm glad those other people had the sense to go ahead.”
― Richard P. Feynman
Why would the vibecession/fertility crisis make monetary savings useless? I could see how a drastic AI singularity could, but not a fertility crisis - if anything, it might easily make savings that much *more* valuable. (A singularity might too, depending on what form those savings are in.)
You can make that argument from any point in history. A generation ago, one would have said "Between nuclear weapons and the overpopulation crisis, it doesn't make sense to save for retirement" and those arguments felt just as pressing and urgent at that time as AI does now.
Indeed. I'm old enough to clearly remember a couple of widespread iterations of that syndrome, each of which seemed quite persuasive at the time.
Also, if you're in the US the fertility crisis in particular is being pre-sold so to speak. TFR, the figure most commonly bandied about, is _not_ the actual number of births taking place in a nation.
That measure is called "completed fertility rate" or CFR: the actual number of successful births per woman by the end of childbearing years. In the US the CFR (a) is currently only about 4 percent below replacement level, and (b) has been _rising_ for several years now.
Similarly, total US live births per 1,000 people was for both 2024 and 2025 _higher_ than in any year since 2016.
CFR is an inherently lagging indicator (it can't tell give us a final answer of births among women who haven't yet reached age 45), which is why TFR exists to be the _predictive_ demographic indicator. It is possible that in coming years/decades the TFR's _prediction_ of much-lower birthrates will turn out to be accurate which we'd know by the CFR following it downward.
Currently in the US though the CFR is moving in the opposite direction: upward. If that continues then it will be the TFR which starts to climb back up towards the CFR. That's because of how the TFR is calculated which you can google if interested.
Overall point is that contrary to the hype we do not yet know that the US' birthrate per woman of childbearing years is in the midst of decline to well below replacement level. That may happen, could even have already begun. Or, not. The TFR _predicts_ that rather than reports it, and the real-life-results statistics are currently not following the TFR downward.
There are good tax advantaged methods of savings that are nominally for retirement that can be feasibly converted into non-retirement use cases, most notably HSAs and Roth IRAs. As long as AI-powered immortality treatment is HSA-eligible you can contribute tax-free to your HSA now, and use the earnings at any time in the future, also tax-free, for your hyper-adrenochrome injections / mind upload / etc. And Roth IRA contributions (not earnings, but principal) can be withdraw at any time, penalty-free, and the penalty on earnings is only 10%. So if you really need that cash for a planet-sized computer in the Artemis Tau cluster, or a last-ditch donation to alignment efforts for Claude Requiem 7, you can always withdraw.
I think it assumes very short timelines and very fast takeoff scenarios. In slow takeoff scenarios with longer timelines, owning capital makes a ton of sense.
Consider a slow takeoff scenario. Go look at VTSAX, Vanguard's S&P 500 tracker fund (1).
What does a slow takeoff scenario look like in real life for an imaginary investor, Whipper Snap the IIIrd who's 25 with $100k invested.
Well, arguably we're already in a slow takeoff scenario. Over the past 15 years we've had ~13% annual growth, a roughly 3% increase over the historical 10% norm. Maybe that's an aberration, maybe that's the new normal as an increasing amount of cognitive work is being automated.
That's a big honking deal. That shortens the time for your stock portfolio to double in value from ~7 years to ~5.5 years. For our young Whipper Snap IIIrd, that means his investment will grow from $100k to ~$400k with no further investment by age 35.
Now imagine that the slow takeoff accelerates again and we hit 16% annual growth in US stocks. Now our doubling time falls from ~5.5 years to ~4.5 years. Whipper Snap IIIrd's stocks are going to grow from $400k to....mmm, let's say $1.7 million in 10 years.
So, 20 years from now, let's say our slow takeoff accelerates again and the stock market is now growing 19% annually. Our Whipper Snap now not only has $1.7 million at 45, he's looking ahead to a future where those stocks should be growing by ~$320k/year (2). That's a really, really bright future for him.
Where do you really think the world is going to be in 30-40 years? If AI is all hype, the market will continue as normal, and everyone is going to die, then live and invest for retirement as normal. If we're in slow takeoff scenarios, where the economy is doubling in value every 24-36 months and aging is a solved problem, then working as hard as possible to acquire capital and ride that wealth explosion is extremely important. If you think we're in fast takeoff scenarios, we are all either dead or post-scarcity post-human Matrioshka brains in 2055 consuming everything the light touches.
Most people I know in AI are acting as if they're in slow takeoff scenarios, working as hard as possible and investing as much as possible before some labor market disruption makes their skills worthless.
(1) https://investor.vanguard.com/investment-products/mutual-funds/profile/vtsax?msockid=2e100ccded096fe107bc1adaec406e4b
(2) 1,700,000 * 0.19=$323,000
As another old fart, who is also >65, having lived through at least six predicted end of the world scenarios, DO NOT believe the doomsters. They don't know what they're talking about, and they profit from your state of fear.
I came of age with the threat of nuclear war, ecological disasters, and regular famines all over the world. Things are *much* better now than they've ever been for the human race. The coming of AGI, for all intents and purposes, is a religious belief—because we really don't understand what consciousness is or how it is enabled by wetware. Why the heck do we think we can build it in silico? LLMs *may* change how labor is done and rewarded, but they don't seem to have made much of a dent in the workforce so far. As for the falling birth rate, Japan seems to be handling it just fine (of course, they're ignoring the dogma of Freshwater economists to do it.
You see Scott as another doomster? I don't see what relation consciousness has to the possibility AGI could radically transform anything, you could have a superintelligent computer that is not conscious.
I thought the whole argument was that strong AGI (with consciousness) would be motivated to kill us, while weak AGI wouldn't care?
That's not to say that weak AGIs couldn't cause unanticipated downstream problems for our high-energy civilization — or be misused by bad actors to sabotage critical systems that we depend on.
But I think both the fear and promise of AI are way overblown. So, yes, I would respectfully suggest that Scott's estimate that there was between a 20-25% chance of causing humanity's extinction is an example of Yudkowskian rationalists assigning levels of certainties for which we have no priors. This is a common element behind other (non-AI) doomer scenarios.
The traditional Yudkowskian/Bostromian position is the "paperclip maximizer" argument.
A) Matter is rivalrous.
B) AI has goals (e.g. tiling the universe with paperclips).
C) AI needs matter for goals.
D) Therefore AI is dangerous.
E) NOTE: The AI need not be conscious or malicious to be dangerous.
They don't fear the Terminator or Skynet or HAL, they fear the Von-Neumann Probe and the Morris Worm.
So it boils down to that AI will become a giant resource "memory leak". Well, it's already become that, so I guess I can't argue with that idea. :-)
Hmm. The thing about memory leaks is that, like slippery slopes, there are limiting principles. Neither continues forever.
So in this context, the contention is a disagreement over what the limiting factor is. "YudBost" claims it is the mass of the universe, whereas I, and others, think it's going to be something much more prosaic and benign.
No, you could have a strong AGI, even an ASI, that is not conscious. A computer wouldn't need consciousness to get to those capabilities, it could still do all the necessary actions to screw us over. It's very unclear what consciousness is for, or even what it is.
Either you're misinformed, or someone moved the goalposts at some point. John Searle, in his 1980 paper, "Minds, Brains, and Programs", coined the term strong AI, and he explicitly included the presence of consciousness (which includes self-awareness and the mental states associated with consciousness) in the definition. He argued that a machine built merely on formal symbol manipulation (like a computer program) could never achieve this genuine consciousness and understanding, only simulate intelligent behavior. Maybe the current crowd of AI promoters and snake-oil salesmen have foisted a softer definition of strong AI upon us? After all, "strong" is a better marketing word than "weak."
Be that as it may, I'm using Searle's definition. And by Searle's definition, weak AI wouldn't have self-awareness, and therefore it wouldn't have self-directed intent. That doesn't rule out the humans with self-directed intent from making it do nasty things, or it accidentally doing unexpectedly nasty things, but a weak AI (under Searle's definition) would only respond to external prompts and directives.
It's kind of weird that I have to even argue this. Has Searle been erased from the history of AI?
> Has Searle been erased from the history of AI?
He pretty much never belonged in that history. And to the extent that he does belong there, you are misinterpreting him.
Searle's distinction between strong and weak was not about how powerful an AI was. It was about two different belief systems about how much of human intelligence could be mechanized. [https://www.google.com/search?q=strong+ai+hypothesis&oq=Strong+AI+hypothesis]
That is, Searle would say that if the Strong AI Hypothesis were true, then conscious AIs might someday be possible. If false, as he believed, then it is never going to happen.
As to whether an ASI might want to kill us, yes, the risk is probably a bit higher if the ASI is conscious, but there is a non-zero risk that it might want to kill us even if not conscious and also a risk that it will kill us even without wanting to.
I think money will still be used in exchange for goods and services in 40 years
Why is that?
Because I don't think civilization is going to collapse; barring that, why wouldn't we still be using money 40 years from now?
Because it's a symbolic system that simplifies resource and logistics accounting. There's nothing better to (a) comparatively track the overall success or failure of projects, (b) balance exchanges between parties, and (c) track the relative efficiency of logistics decisions. Of course, all money is ultimately dependent on an expenditure of energy. One could imagine a world where everything is run on energy accounting, but that would be money in an abstract sense.
The real reason the command economy of the Soviet Union collapsed was not because it was a command economy per se, but because their accounting systems were totally f**ked. All the factory managers who were given quotas lied about their output. Bookkeeping was rudimentary, or non-existent, so there was no way to track what managers were skimming off the top. No one knew how to estimate the cost of materials and translate them into rational prices. Oh, sure, it's hard for a command economy to compete against a free market economy, but North Korea maintains its country on a command economy (albeit it's brutal on most of its citizens), but NK hasn't collapsed yet.
I'm an old fart. (>65) But I figure there will always be stuff for us meat bags to do. Jobs may change but you'll still have value. (Also I think some 'news' is meant to scare you, cause it's an easy emotion to trigger, also also, AI might be a bubble... where is all the energy coming from?)
Eh....quoting as _fact_ some _unconfirmed_ lurid descriptions that were posted as a reader comment with no sourcing provided, doesn't seem consistent with the overall tone/practices/intentions of ACX.
A while back, I asked about VR headsets. I decided not to go forward with purchasing one, but I wonder if they may not be helpful for my mom, who is in the later stages of dementia. She used to have an active intellectual life, but she can't read anymore, and she can't understand what's on TV. And due to incontinence, it's difficult to take her out anymore. Although she has difficulty communicating, she's obviously bored shitless and just sleeps. And when she's not sleeping, she's restless.
Question: If I got her a VR headset, would there be a way to access a continuous but varied feed of nature scenes and natural sounds? I figure they might provide her with non-language stimulation to occupy her visual and audio qualia? Thoughts? Suggestions?
That’s a kind, smart, great idea and it looks like you are going to be able to execute it:
Best Overall (Richest Nature Experience)
These headsets are more powerful and have apps that can show immersive changing nature environments:
• Meta Quest 3 – A full standalone VR system with high-quality visuals and access to a wide selection of apps, including Nature Treks VR and other calming nature experiences. You won’t have to connect to a PC or phone; it’s all in the headset.
• Meta Quest 3S All‑in‑One Headset – Slightly less powerful than Quest 3 but still capable of running immersive VR scenes without wires.
Apps like Nature Treks VR are available on Quest headsets and offer serene nature worlds to explore (forests, beaches, relaxing environments etc.) — these are not games per se, though the user can look around and enjoy tranquil settings.
Another option:
https://www.healinghealth.com/care-channel-relaxation-programming/care-vr/?utm_source=chatgpt.com
GPT, man
It's not at all a VR headset, but would a RoboPet be helpful for her? I found out about them just today, and there's a bunch of reviews saying how helpful they were for the reviewers' elderly relatives, including several with dementia.
https://www.robopets.co.uk/
The company is an official supplier to the NHS, apparently, for dementia wards.
I have a Quest 3. The native Theater Elsewhere app sounds a lot like what you're looking for, but if you aren't satisfied with that, you can still get a more general-purpose web browser (Wolvic for one) onto it, and load a website of your choosing. I'm confident you can find one to your liking. (They're of course not interactive at all, but maybe even some of the very long 360/VR videos on YouTube will suffice.)
I'm rereading the "Dictator Book Club" on Chavez (https://www.astralcodexten.com/p/dictator-book-club-chavez) for fairly obvious reasons and I'm wondering if any of the broadcasts it references are available anywhere - I haven't had any hits on Googling for them; e.g the soda plant incident with the burp.
Maybe it's because I'm not searching in Spanish? Or maybe they're genuinely just not posted anywhere, but I just expected that historic events televised well into the era of VCRs would have been fairly easy to find.
EDIT: Oops, shortly after posting I got down to the comments and saw that the burp incident had been found on YouTube: https://www.youtube.com/watch?v=px04jhigE-0 - didn't see any of the others, but I guess that's the most interesting of them anyway.
Regarding 4.4 (putting away $25,000), what if the rules of civilization change? Who defines ownership? In the past, people could steal from other people. What if a country invades another country and steals the production factories, as used to happen with wars?
I may have $25,000 put away, but there is no guarantee it will be worth anything in the future.
I think the answer is that the whole post is only aimed at people who accepted the original premise it was a response to, and if you believe the things you say here in your post, you already rejected that initial premise.
So, basically, you're cool, this already wasn't advice you needed.
I think there is a real possibility of a future in which money is meaningless, either because AI fundamentally changes everything / kills us all, or civilization collapses, but then, not sure how you're supposed to hedge against those. Maybe for civilization collapsing you can learn some survival skills and find a good community. It's still good to put 25k away to target outcomes where money is still meaningful, although what I did was put 25% of my cash into a tech index fund and an S&P 500 index fund, half each.
Probably doesn't make sense to save for retirement though. I think it's extremely unlikely the world will be business as usual when/if I turn 65.
A few scenarios:
A. Civilization only changes in ways that more-or-less preserve equity ownership. In this case, $25k in broad-based investments should put you in pretty good shape given Scott's assumed 100x growth. This is Scott's 5.3.
B. Some large portion of the means of production is seized and either redistributed in an egalitarian way or collectivized for common benefit. With or without the $25k, everyone will be living in Luxury Space Communism. This is a much larger version of Scott's 5.4 (charity+welfare drawing on a vastly expanded economic base) providing for the assetless and jobless).
C. Some group seizes power, confiscates wealth from their outgroup, and leaves the underclass to shift for themselves at best. Call this the "Space Nazi" scenario. Regardless of what you do now, you're going to have a bad time unless you happen to be in the Space Nazis' ingroup.
D. An unaligned AI takes power, decides humanity is surplus to requirements, and turns us all into paperclips. Scott's paragraph 2 is related to one of many efforts to minimize the risk of this. And like C, if it happens anyway, you're going to have a bad time.
I think the focus was on A and B because, as Scott said, he's responding specifically to concerns about what might happen if AI takeoff results in vast economic growth accruing to equity owners while rendering most wage labor (or at least most white collar jobs) redundant. The fear is that the holders of unredistributed equity become super-rich oligarchs, and Scott is arguing that it doesn't take much investment on the scale of white collar professional finances (5.3), or much redistribution by modern American standards (5.4), for the underclass's material condition in such a scenario to be what we now consider to be comfortably wealthy. He's proposing contingency plans and reasons to hope about one specific genre of dystopian scenario, not a general purpose long-term sci fi disaster prepping plan.
It's a new year, which means a new load of characters and stories have entered the public domain. The one I'm most excited for is Nancy Drew. Note that this is the 1930 hardcore Nancy, where she kicks ass and carries a gun, not the "more sugar and less spice" bowdlerized 50s and 60s rewrite version. I'm making a puzzle-mystery tabletop RPG about her, which will go live tomorrow: https://www.kickstarter.com/projects/sixpencegames/the-nancy-drew-rpg
Hi Scott, around the end of the month I made the pledge for your pledge drive and posted about it in the thread. Would really love that free subscription if you see this message!
When I had technical issues getting my subscription, Scott suggested emailing him. I'd recommend finding his email in the pledge drive comments and emailing him to get his attention.
I did email him after a few days of no word. But this was all around the end of the year so I assume he’s just been busy with the holidays. If this doesn’t work, I’ll try email him again. Thanks for the suggestion!
I think there was some magic phrase like "this is a genuine non-spam message" or so, that you have to include for the mail to go through his filters.
I disagree with this idea of "permanent monetary inequality" brought on by AI. As a biologist who has grown bacterial and eukaryotic cells in a lab, I think of this was like passaging cells. Sure, initially there's a glut of media for each cell, but over the long run exponential growth will bring the population to capacity, no matter how much growth media you start with.
"But AI will grow exponentially, too!" Until it runs into the limits of living in a finite universe. Then exponential population growth will catch up. An AI-driven 'singularity', where there's plenty for all, will only ever be temporary - not permanent.
"The poor you have with you always."
Given that post-singularity, one can create 1000 new people instantly, new person creation will probably be controlled by the "government". I don't think post-singularity humanity (assuming it exists) would let someone make a new person if the creator doesn't have enough resources for the child to not be "poor" (for any reasonable definition of poor).
Seems to me that post-singularity life would be so hugely different that it doesn’t make sense to assume there will be a government, much less a government doing stuff sort like ours does, such as making rules about what various categories of people can and can’t do.
You are probably right, personhood itself will also probably be very different when one can merge and split one's mind on a whim. I tried to express this using the apostrophes and "assuming it exists": I meant to condition on not super weird outcomes, but who knows how likely those are. It's called a singularity for a reason.
Why does this disprove inequality? This just seems to be positing a Malthusian limit, but there was a Malthusian limit in ancient times too, and Pharaoh still had more wealth than a peasant.
Right. I'm not saying inequality would magically end. I'm saying that scenarios imagining doomsday inequality (driven by AI and in context of massive GDP-growth-driven plenty) should not be expected to be permanent.
I would also expect that the maximum sustainable inequality at human population = 8 trillion would likely be significantly greater than at the current population of 8 billion, just as the inequality between Pharaoh and peasant was less than Elon and your average office worker today. More wealth (however it's derived) enables greater absolute inequality by definition (even if relative inequality remains the same).
Maybe the end result will be forum discussions in 2100 about whether a quadrillionaire represents a fulfillment of the predictions on ACX back in 2025, "because look at all the increased inequality," vs. the counter that there's still something like a graduated income system and "nobody who predicted gloom in 2025 was thinking things would turn out like this."
Or maybe it takes 10x or 100x as long to hit finite limits from AI-driven GDP growth, so that being "right" that the doomsday scenario will technically be impermanent is cold comfort to the 20+ generations of our descendants who have to live through the extended "transition" period of rapid GDP growth.
> just as the inequality between Pharaoh and peasant was less than Elon and your average office worker today
Do you actually think this is true?
In absolute terms, yes. By definition.
If there are 100 people, and we each have $1, but you have $101, the total wealth is $200. (Or more particularly, let's say everyone has good valued at those dollar amounts.) Now let's say the same 100 people have $100,000 each, but you have $10,100,000, with total wealth at $20 million. The relative inequality remains, but the absolute inequality has changed drastically.
Same with Pharaoh. Even if he owned 99.9% of Egyptian GDP through Joseph's land-buying plan, in absolute terms, the GDP of ancient Egypt is nothing compared to the daily fluctuations of Elon's wealth.
Ancient Egypt built its wealth around agriculture. Ignoring the differences between modern and ancient agriculture, we might get a crude over estimate of Pharaoh's potential wealth in modern terms. Modern Ethiopia gets about 35% of their GDP from agriculture. Total GDP = around $110B, so agriculture-based GDP = around $38B. About 80% of their 133M people are in agriculture, for an agriculture-based GDP per capita of about $360. At times, the population of Egypt was as large as 5 million people, so if Pharaoh owned nearly all of Egyptian GDP, he'd still have < $2B in today's money.
Could Elon get a pyramid made for him that would last 4500+ years? He's very rich, but could he do this? There's a lot of desert out there, and I'm not seeing any fresh pyramid construction. All the mega structures made (in Saudi the UAE etc.) will barely last 100 years.
I think this is just about building materials and location. A structure made of normal concrete with no re-enforcement in a compression-loaded-only structure in a seismically-stable desert can probably stand for that long. Making it out of concrete instead of some more valuable material is important because otherwise someone might loot your structure for building materials. You'd need to think about 1000+ year floods and windstorms and about erosion by wind (maybe you need to make the concrete thicker to account for losing a cm or two per century to sandy wind etching it away), but this all seems doable in principle.
I think the main thing that makes modern structures not last forever is steel (used directly or as re-enforcement in concrete), which gives wonderful strength and lets you build stuff that can't be built with other materials, but which will eventually corrode.
It is trivially easy to build a pyramid that will last 4500 years. It's a pile of limestone blocks in the desert. How could it not last essentially forever?
ChatGPT estimated the modern cost of recreating the Pyramid of Giza at $1-10 billion. Musk could build dozens of them.
Could a Pharoah have put a car into Earth orbit?
Are you saying that the reason we're building skyscrapers and rockets instead of pyramids is because we can't do pyramids anymore? I think the difference is one of will/interest as opposed to means. Give me a billion dollars and I'll build you a stone monument in the desert, using cheap labor, modern construction equipment, bribes, and still have a few hundred million left over to play with.
If Musk fails his life's mission of building a colony on Mars, and humanity 'only' ends up with a colony on the moon, I submit that this would represent an achievement so far from the wildest imaginations of anyone in Pharaoh's day as to be beyond comparison to the pyramids.
(Also, the "they last forever!" bit is less impressive in the desert. I had an undergrad professor who did digs in Egypt. He showed us a picture of a crochet baby booty they dated to >2,000 years old, buried under the sand. There's a reason the Washington Monument hasn't done so well in the DC climate, but the pyramids last forever.)
That's not even the biggest problem with sclmlw's position. I would put the lack of pyramid construction more down to lack of interest.
If Elon Musk orders an office worker executed, will it happen?
I read your "permanent monetary inequality" as "the folks who started off wealthier at time 't' would be permanently wealthier" rather than "we will always have poor people."
I can't tell if you and sclmlw are using the phrase the same way or not.
I define poverty in relative terms (though absolute poverty still exists, globally). Many articles have been written comparing Pharaoh to the bottom quintile US wage earner today, asking which is better off. I assume at 10,000x GDP it's a similar comparison.
The argument I'm looking at is the one that posits effectively infinite resources and asks whether society will figure out how to distribute those resources like in Star Trek, or whether everything will depend on initial conditions.
My counter is that those resources will look less infinite in the face of the massive population growth they will enable. Maybe we'll even still have the equivalent of subsistence farmers in the AI future, because plenty is a function of supply/demand, and the denominator (total # of people) should be assumed to be dynamic, too.
> The argument I'm looking at is the one that posits effectively infinite resources and asks whether society will figure out how to distribute those resources like in Star Trek,
Not everybody owned a winery in the glorious Star Trek future.
> My counter is that those resources will look less infinite in the face of the massive population growth they will enable.
That would be a strong reversal of existing trends.
I finally wrote up my experience with getting semaglutide from China for cheap:
https://www.lesswrong.com/posts/coLiSHpP338Xwibbp/the-bio-pirate-s-guide-to-glp-1-agonists
Text without formatting, footnotes and links:
The bio-pirate's guide to GLP-1 agonists
How to lose weight, infringe patents, and possibly poison yourself for 22 Euros a month.
Introduction
In March 2025, Scott Alexander wrote:
> Others are turning amateur chemist. You can order GLP-1 peptides from China for cheap. Once you have the peptide, all you have to do is put it in the right amount of bacteriostatic water. In theory this is no harder than any other mix-powder-with-water task. But this time if you do anything wrong, or are insufficiently clean, you can give yourself a horrible infection, or inactivate the drug, or accidentally take 100x too much of the drug and end up with negative weight and float up into the sky and be lost forever. ACX cannot in good conscience recommend this cheap, common, and awesome solution.
With a BMI of about 28, low executive function, a bit of sleep apnea and no willpower to spend on either dieting or dealing with the medical priesthood, I thought I would give it a try. This is a summary of my journey.
Please do not expect any great revelations here beyond "you can buy semaglutide from China, duh". All of the details here can also be found elsewhere, still I thought it might be helpful to write them down.
Also be careful when following medical advise from random people from the internet. The medical system is full of safeguards to make very sure that no procedure it does will ever hurt you. Here you are on your own. I am not a physician, just an interested amateur with a STEM background. If you do not know if it is ok to reuse syringes or inject air, or do not trust yourself to calculate your dose, I would recommend refraining from DIY medicine.
Picking a substance and route of administration
The two main approved GLP-1 agonists are tirzepatide and semaglutide. Both are peptides (mini-proteins) with a mass of about 4-5kDa which cost approximately the same to produce. A typical long term dose of tirzepatide is 15mg/week, while semaglutide is 2.4mg/week, so I focused on sema because it looked like the cheaper option. [1]
While I would have preferred oral to subcutaneous injection, the bioavailability of oral semaglutide is kinda terrible, with typical long term doses around 14mg/day -- forty times the amount of subcutaneous injection. So I resolved to deal with the hassle of poking myself with needles and possibly giving myself 'horrible infections'.
Given that the long term dosage is 2.4mg/week, and that sources generally state that once opened, vials should be used within four (or eight) weeks, I decided that the optimal vial size would be 10mg -- enough for four weeks. [2]
Finding a vendor
So I started searching the web for possible vendors of lyophilized semaglutide. I found a wide range of prices from 250$ for a 10mg vial (which would last four weeks at maximum dosage) down to about 50$. And a single website which offered ten vials of 5mg each for 130$.
That one seemed to be a Chinese manufacturer of organic compounds [3] . Slightly broken English, endless lists of chemicals by CAS number, no working search function on the website, outdated and incomplete price info provided as a jpeg on the site. I figured that if it was a scam site, it was matching very well to my preconception of how a site of a company more enthusiastic about synthesis than selling to consumers would look like, and contacted them. After I was provided with current pricing (also as a series of jpegs), I sent them about 200 Euros worth of BTC [4] for ten 10mg vials plus shipping. (Shipping was 70$, probably indicative of a preference to sell in larger quantities.)
A week or so later, I got my delivery. Ten unmarked vials, of a volume of about 3ml each, filled to perhaps a third with a block of white stuff. [5]
I would have preferred to have a quantitative analysis of the contents, but all the companies in Germany I contacted were either unwilling to deal with consumers or unable to perform HPLC-MS, so I reasoned that the vendor would be unlikely to sell me vials filled with botulinum toxin instead [6] , and just started with injecting myself with 1/40th of a vial per week, which would amount to 0.25mg if the content was as advertised.
(If anyone has a great, affordable way for peptide analysis, please let me know in the comments!)
Sourcing bacteriostatic water
Unlike random pharmaceuticals, bacteriostatic water can be legally sold in Germany. Sadly, it would have cost me almost as much as the active ingredient, about 15 Euro per vial. So instead, I decided to craft my own bacteriostatic water. I sourced a lifetime supply of benzyl alcohol for a couple of Euros. Instead of dealing with distilled water, I bought sealed medical grade plastic vials of 0.9% NaCl solution "for inhalation", roughly 0.5 Euro a piece. Once a month, I add 0.1ml benzyl alcohol (naturally sterile) to one 5ml plastic vial, which gives me about 2% benzyl alcohol, which is twice of what is specified as BAC, erring on the side of caution (and tissue damage from alcohol, I guess).
Other equipment
I already had a bunch of sterile 20G hypodermic needles and 3ml syringes from another project. For injection of minute quantities of liquids into my body, I bought sterile 1ml insulin syringes with 6mm, 31G needles (@0.3 Euro). [7]
Happily, I owned a fridge and a disinfectant spray, completing my toolset.
My current procedure
Every four weeks, I will prepare a new vial. I prefer to fill the vials with 3ml of my BAC+, which should give 3.3mg/ml of sema.
Apply disinfectant to hands and workspace to taste. Then, start with opening a new plastic vial of NaCl. Using an insulin syringe, add 0.1ml benzyl alcohol to it. Unseal a 3ml syringe and needle and draw and release your BAC+ from the plastic vial a few times to mix it. Now draw 3ml of that, tear off the plastic cap [8] of your new glass vial, stick the needle through the rubber seal and slowly inject your BAC into the vial. The vial will be low-pressure, so getting liquid into it is really easy. Shake a bit and wait for the lyophilized peptide to dissolve. Store it in a fridge (preferably in the plastic box with the other vials), and liberally apply disinfectant to the rubber seal before and after each use.
To draw a dose, first figure out the volume you need. Disinfect, unseal your 1ml syringe, first inject an equal amount of air into the vial, then turn the vial rubber side down and draw the liquid. To start with, I would recommend drawing 0.1ml more than you need, because you will likely have some air bubbles in. Remove the needle, get rid of the excess air (and excess liquid). Check that you are in a private place, expose your tights, apply disinfectant, pinch the skin of your tight with two fingers, stick in the needle with the other hand, push down the plunger. Cover up your tights, put your vial back into the fridge, safely dispose of your needle. [9]
Outcome (N=1)
Having taken semaglutide as scheduled for some 22 weeks, I have recently cut my dosage in half because I have reached a BMI of 22.
Traditionally, weight loss was seen as a moral battle: eat less than you want to eat, eat different things than you want to eat, do more sports than you want to do. Basically, spend willpower points to lose weight.
GLP-1 agonists are a cheat code, like reaching enlightenment through psychedelics instead of years of meditation, or bringing a gun to a sword fight. I literally spend zero willpower points in this endeavor. I continue to eat what I want and how much I want, it just so happens I want less food. I am under no illusion that this cheat will give me the full benefits of exercise and proper diet. But realistically, these were never an option for me (until someone discovers an infinite willpower cheat).
Will I rebounce once I quit sema? Not a problem, at 20 Euros a month, the drug pays itself in money not spend on chocolate, and it is less of a hassle than taking my other pills once a day, so I am willing to continue to take it for the rest of my life.
Thanks to Scott Alexander for pointing out this option and to my Chinese vendor for providing Westerners like myself with cheap bodily autonomy.
Up next: The bio-pirate's guide to DIY MAID (due in a few decades, PRNS).
Glad that worked for you. I would be concerned if there are issues with homeostasis or developing a tolerance or what have you if taking it long term, though. That's always a relevant question for me, because I lost about 35 pounds a couple years ago via keto + intermittent fasting, but over time that just stopped working and I gained most of it back.
Semaglutide is off patent in Canada as of Jan 1, 2026 (Novo Nordisk screwed up a patent extension filing ... even after the Canadian's told them it was coming due ...)
Any idea if the "real" stuff (though generic!) can be purchased from Canada for cheap?
You can buy it now in the US from places selling “research chemicals.” There are ways to do it safely: Send a sample to an outside lab for testing, access crowd-sourced info about other people’s lab test results for GLP-1 drugs from the seller you used.
"Safely" is relative. When you by pharmaceuticals from a big pharma company in the US, they have clear incentives to making very sure that their product is uncontaminated because otherwise you might sue them into bankruptcy.
If you buy "research chemicals" from some tiny company, chances are they are also sourcing them from China and probably not doing a ton of QA -- certainly not enough to detect "one in 1000 vials is contaminated with bacteria due to faulty sterilization procedures". Their expected losses from lawsuits are bounded by the value of their company, after all.
With labs for testing, you require a protocol to establish that the lab is actually doing the work they claim to be doing, instead of just determining the dry weight to a mg level and generating a random number between 99.5% and 99.8% and give that as the purity (which is certainly cheaper than LC-MS). This can be tested. One protocol would be to uncap and fill the vials before sending them to the lab, selectively changing dosages, adding inert filler material (NaCl), bacterial or heavy metal contaminants. Because the lab does not know what you added where, their best way to convince you that you that they are testing as advertised would be to actually test as advertised. (Of course, they could also just refuse to test unsealed vials, which would be a red flag.)
At perhaps 1000$ per full set of tests, it would cost you probably 5k$ to make sure that they are doing all the tests that they claim they are doing and that your untampered samples are indeed clean.
It seems hard to establish common knowledge about a lab being reliable, because you have to trust whomever is testing the testing labs in turn. One way to do so would be to pick a person known to be trustworthy to do the testing, and collectively pay them to do so. There is probably also a Byzantine fault tolerant algorithm where a set of n interested parties randomly decides on k of them who will will get to send in samples.
Personally, I decided to lab forgo testing in the end. Even if you trust the lab, it will only detect frequently occurring problems, you still have to trust your vendor to avoid rare problems through good procedures. And without crowdsourcing, it significantly increases the total price (depending what you want to test for). While there are certainly substances which can kill you stone dead when you inject a few milligrams of them into your tight before you can even call an ambulance, these are few and far between -- basically potent neurotoxins, as run of the mill poisons like cyanide or mercury require higher doses than that.
In the end, I estimated that my expected QALY increase from losing weight would be significantly higher than my QALY decrease from getting into serious trouble through subcutaneous injection.
Is anyone playing around with side projects using AI coding or other processes? I'm curious if your experiences match the breathless reports from engineers on social media about Claude Code and related tools.
For my part, I'm finding it much easier to deploy projects with Claude Code these days, although they are relatively simple. Over the winter break, I created an AI-enabled site for personalized homeschool curriculum advice (free): https://homeschooltools.net/. If you're homeschooling or homeschool-curious, I'd love any feedback.
I'm also working on an AI economic simulation, which is going much worse because LLMs currently still are not good economic role-players. I guess that's something a bit outside mainstream training scenarios, so it's still pretty vibes-dominated ("income equality has gone up, the economy will crash!").
Finally, I'm also planning on working on a friend's iOS app idea. It feels like it should be pretty doable, which I would never have through prior to about the last 6 months.
I am (or was, maybe) a professional software engineer and I use LLMs for development assistance all the time. They're a massive boon, especially for 0 -> 1 type work. I recently used Gemini 3 Pro via Google's AI Studio product to build a guitar fretboard learning app (available at https://fretu.de if you're interested). That level of React complexity, where it can seamlessly transition between desktop and mobile layouts while managing a very complex UI full of clickables with finicky little UI elements, would be way beyond my backend-engineer's capabilities.
I also use it in established code bases to add functionality though I prefer Claude Code for that. I find Opus 4.5 via the Claude Code TUI in particular to be excellent at navigating code bases and one-shotting functionality. I used it to add a relatively mathematically complex reverse-polarization node in a realtime graphics tool I work on, and it got it perfectly right the first time. Very impressive, and a massive time saver.
One interesting note is that Gemini Pro 3 got stuck unable to fix a rendering bug that was causing flickering in mobile Firefox when certain page elements would animate – very low level stuff, limited to a specific low-marketshare browser, etc. It started looping trying the same incorrect fixes over and over. I eventually threw Opus 4.5 at it, and through some collaborative centaur-debugging we managed to solve it.
I am highly convinced that the people who rush to shit on AI tooling are just expressing fears about their own obsolescence – they don't /want/ it to be good, because that threatens some self-image they have of themselves as valuable members of society. "It is difficult to get a man to understand..." etc etc.
As a senior (old) programmer with 5 decades of experience in the tank, I can say without reservation that the impact of AI-assisted coding is the most important change in tooling that I have seen in my lifetime.
That said,once you get over the "I cannot believe I am seeing this" phase, and you spend considerable time writing, analyzing, and debugging code this way, in some important respects it is 'just' another tool, and should not be approached with unbridled reverence and expectations (queue song from "Jesus Christ Superstar": "He's a man, He's just a man...")
In my experience, it is very much like working with a gifted junior programmer who has a surprisingly vast amount of book knowledge and a good amount of common sense, but is a bit deficient in the "Knowing what you don't know" department. I doubt this critique will be relevant for too long, but that's what I see with Claude 4.5.
Two anecdotes that I think are relevant:
1) I am working mostly solo on a robotic simulation project. It requires a strong grounding in 3D math both for physics and visualization. I am finding Claude to be very "book-learned" about the math but lacking in basic, common-sense understanding of 3-dimensional transformations at a gut level. This is hard to describe but I have seen a wide range of ability in this area with human collaborators. Some people just "think in 3D" but others struggle to, for example, intuitively know how to do transpositions, rotations, and scaling on 3D objects.
2) I am working with a CEO at a startup who is extremely bright (math PhD) but self-describes as ADD, and has consequently focused on people skills in his career -- marketing, sales, relationship-building etc. But recently he started "Vibe coding", and has consequently started spending a lot of time and energy "helping" the technical team with new ideas and potential pivots, that come along with demos he has Claude build. While I applaud his "lifelong learner" tendency, the truth is it can be disruptive and takes time and focus away from the things we really need to build. (Sorecerer's Apprentice Syndrome).
One of my old friends just messaged me saying that "we've reached the singularity". He used Claude Code and was very impressed with it in these past few months. Apparently it can one-shot complex features/whole applications. Personally, I didn't find older versions that useful, but I haven't yet tried this, relatively new version, but surely will soon.
I've really enjoyed Gemini 3 Pro's coding abilities. As a total novice to coding myself, it has the ability to question or clarify my own mistakes/flawed assumptions which the prior generation of LLMs (2.5 Pro, Claude 4, Grok 3) was much worse at. Even without agents or an actual AI coding tool it has been easy to develop with AI as a schmuck.
My main use case has been creating mods (small C# plugins) for a 10+ y.o. video game. In this, Gemini 3 has greatly succeeded despite a plethora of outdated documentation, old or inaccessible mods needed as reference, and my own foolishness. I can't code beyond Hello World and have largely been doing this to learn more about Visual Studio and AI coding. I've now made around 5 mods plus a modding API to facilitate further development, and there are several hundred users and counting playing with these mods today!
This has all contributed to me "feeling the AGI." Gemini patiently and helpfully walked me through learning Visual Studio, plugin development, and game modding all at the same time. It also coached me through creating some assets for these mods in Audacity or with any questions I had around deploying them on GitHub and elsewhere. I absolutely would not have been able to do this without Gemini and I'm also very happy that the end result is a trim, performant suite of mods with no bugs or slop.
Right now there's various levels of 'using AI' for coding:
0. Not using it at al
1. I write basically all the code, but occasional AI prompts or autocompletion for questions or snippets (i.e. "LLM-replacing some google/stackoverflow usage")
2. LLM generates all the code, but I read and review it with the same standards that I would read and review code written by another team member.
3. Like #2, but much less review of the code, just testing the actual behavior, maybe having LLM generated test cases
4. Multi-LLM workflow - there's some crazy stuff at this level like https://steve-yegge.medium.com/welcome-to-gas-town-4f25ee16dd04, (which has a similar breakdown of 'stages of using AI')
Personally, I'm mostly between #1 and #2 and have been pretty successful.
For stuff that I know is going to be critical and tricky, I mostly write it myself with some AI completions.
For a lot of other stuff, #2 works well. It's not the order-of-magnitude productivity boost but it does seem to work out to being quite a bit faster than hand-writing the code, and it maintains my code quality standards.
I use #3 for stuff that isn't going to be a long-term project; made some pretty useful quick scripts/utility apps this way.
I have not really gotten into using #3 for complex projects, much less the multi-LLM stuff, but I know some really competent coders who have and have had a lot of success with it.
The Steve Yegge stuff really is crazy and hilarious, but at this rate some version of that is going to actually work sooner or later. We should probably start looking at what happened when agriculture and manufacturing became industrialized for inspiration on what might happen with knowledge work, particularly coding.
To be clear, I'm not convinced that "sooner or later" isn't "already". I'm not planning to go full "gas town" on my projects yet; but if by "works" you mean "can produce working products quickly", I suspect that's already true.
(If you mean "produces code that's the same quality as something hand written by a human" then "no" but that's also not the point)
The question for me isn't "does this work" but more "what's the complexity level at which this stops working". Probably it exists, but it might be higher than most people would expect.
Steve Yegge isn't some random AI enthusiast - I was wondering why the name seemed familiar when his "beads" library became newsworthy awhile back and I realized it's because I had recently read a blogpost he wrote 17 years ago: https://steve-yegge.blogspot.com/2008/10/universal-design-pattern.html
I agree. I think it works _now_ if you're comfortable fixing whatever odd thing goes sideways, like Yegge is. And I'm fascinated by the analogy to Kubernetes. While I agree that there's some scale at which Gas Town doesn't work, more important is the fact that there's some experience level under which it doesn't work because you don't get how the whole orchestration works.
> Is anyone playing around with side projects using AI coding or other processes? I'm curious if your experiences match the breathless reports from engineers on social media about Claude Code and related tools.
I've done a lot - firefox extensions, interactive websites, some data science projects, built several GPU heavy repos around video and sound, did a bunch of webscraping, some arduino / micro-electronics projects, and more.
It's absolutely amazing the productivity amplifier, and how fast you can get things done now. It's not the best or most extensible code - it's not something you'd want multiple people working on, or to underlie key daily business functionality.
But in terms of iterating into something that actually works and gets something out there into the world? Fantastic, without parallel, impossibly huge multiplier.
I've been using Lovable.dev, it's a wrapper for Claude Code where you can simply click a button to deploy it on their servers... the limitation is that you can only build webapps, not iOS.
yes, I've been working on a hobby project over the last 3 weeks (trying to get LLMs to play the social deduction game Mafia) using primarily Claude Code. Early progress was super quick and new frontend features especially are very simple to implement; I got a working prototype within a few days. However, the code Claude Code creates is poorly maintainable and will eventually exceed the size that Claude Code can effectively work with; it will make more frequent mistakes, cause tons of bloat, leave behind a crust of unused functions, random obsolete comments that confuse the AI whenever it encounters them, and so on. I've recently considered starting all over again, this time only allowing claude code to fill in function implementations I've hand coded the definitions and documentation for. In the end it's mostly just a test of how good the tools are right now rather than an actual thing I'll ever deploy in the wild, so I'm not too worried. The 5 hour usage limits for pro accounts are also quite tight for a service I'm paying money for.
Hello all!
I've written a trilogy of novels that might be of interest to my fellow ACX readers. They're in the fantasy genre, but with a bit of a rationalist twist: follow along with our characters as they research their world's magic system using the scientific method. The goal in Book I is to break a siege, but by Book III, they find themselves fighting an entity that is, let's just say... misaligned.
Available in paperback and DRM-free ebook form on Amazon: https://www.amazon.com/dp/B0FHF4D3VQ
I've heard that a lot of the reduction of "poverty" in China actually consisted of moving people off subsistence farming where they were at least managing into cities where life wasn't as good. I don't know how to check this, but it's at least possible.
Not a China expert, but village life in China is tough! Subsistence farming is tough! There's a reason why they have the houkou system to keep all the farmers from just moving to the cities and not vice versa. There's a pretty good book "I Deliver Packages in Beijing" which talks about the good and bad sides of being a gig worker in a big Chinese city.
I've been going back and forth in China since 2010 and have seen that economic rise firsthand. It's definitely not just a definitional thing, many hundreds of millions of people went from having a hardscrabble life with essentially nothing, often not even toilets, to phones and toilets and internet and scooters.
Is the new life not a perfect utopia? When is it ever? Is it probably better off? I think that's unambiguously true for hundreds of millions.
Hi Performative Bafflement, I just want to hijack this to say I really appreciate your work.
Could you share more on the method to find love abroad that you talked about in one of your posts? i.e. if I go to the Philippines to find a wife, how am I going to exactly go about doing that? You mentioned your brother in law is constantly getting proposals but how did you find a wife in the first place?
Thanks, I appreciate that.
I believe the post you're talking about is this one:
https://performativebafflement.substack.com/p/the-case-for-overseas-dating-and?r=17hw9h
In terms of actual tactics, I wish I had some advice, but I really don't. When you're a foreigner overseas, it's just much, much easier to meet people, and it all seemingly happens organically.
In terms of the happy international relationships I've seen formed, a big determinant is time and propinquity. I think it might be difficult to come over with only a few weeks or a month and resolve to find a wife, because meeting high quality people is a search problem, and there are genuine difficulties and logistics involved there.
All of the happy international relationships I've seen formed happened when the Westerner lived there full time, and was able to interact naturally and meet a wide array of people, until finding a good match.
Does this make it harder for people with non-remote or non-local jobs to find somebody overseas? Absolutely, and I think if you tried online, you'd be subject to a lot more scams and adverse selection effects, and I wouldn't recommend that. There are operations called Pogo that literally do scams like this at scale, full time, and they're big business.
But finding a wife that you'll have kids with is a major, life-shaping change and event that's probably worth spending a year or so on. So if you're able, I'd recommend taking a year off or finding remote work to be able to actually dedicate a realistic amount of time to it.
Thank you. But it's with the detailed tactics that I'm concerned about.
Do you need to go to some rural area?
Specifically the Philippines, or which parts of the world? Why are the Philippines brought up so much?
What if she divorces you once in the US and takes all your money?
How exactly do you meet these people? Especially for marriage at least in the US there are hobbies and signallers like college etc. How would you know if she's intelligent and conscientious etc in another country beyond just that "wow hot"?
> Thank you. But it's with the detailed tactics that I'm concerned about.
I think the detailed tactics really vary depending on what you specifically are looking for, which varies widely from person to person.
In terms of rural vs urban, what are you looking for? If you're looking for somebody with degrees and / or proven status signifiers like career or wealth or high performing families, that's a lot more common in the cities, just like everywhere in the world.
The country you choose also varies depending on what you're looking for, and your segment. The Philippines comes up a lot because most Westerners know english, and most people in the Philippines have pretty good english, even out in the provinces, so there's no language barrier or friction. Some people might argue that your desired attributes may be more common or easier to find in countries like China or Japan, but both of those have language barriers. So again, it's sort of up to you - how much effort are you planning to put in, what are you looking for, etc.
Broadly, I'd suggest looking in cities if you're after the more career oriented, "proven status and achievement" type women, and or if you're optimizing for something relatively rare, because the base population is so much higher.
I'd suggest looking in the provinces if you're after looks, greater dating power / leverage, or more traditional trad-wife style women who would be happier with a big family and / or raising kids and being a mom.
You can find the trad wives in the cities too, I should point out. But it's a much bigger proportion of the dating pool in the provinces, and you probably have more leverage out there.
> What if she divorces you once in the US and takes all your money?
I go over this in the post, broadly this just doesn't happen.
If you really have some assets you can put them in a self settled South Dakota trust or something if you're really worried.
> How exactly do you meet these people? Especially for marriage at least in the US there are hobbies and signallers like college etc. How would you know if she's intelligent and conscientious etc in another country beyond just that "wow hot"?
The same way you meet anyone - you go out, you live your life, you do things, you make friends, you talk to people. People everywhere are pretty interested in matchmaking, and when you're a foreigner, that effect is greatly multiplied.
In terms of hobbies and signifiers, again, this is up to you. People get degrees in every country - and there are Harvard equivalents like Tsinghua if you're really picky. How do you decide NOW what to look for in a mate? People hold out for what they care about, and what they can get, basically.
And in terms of telling how smart somebody is - you can't do that today, in a regular conversation?? I think you can establish this in a date or two regardless of their degree or career or affiliation, and it's easy to be surprised in both directions on that front.
I wrote another post suggesting that you can probably only really ask for and get 3 things that you might find interesting.
https://performativebafflement.substack.com/p/you-only-get-3-things-in-mate-optimization?r=17hw9h
Latest Chinese unemployment rate link: https://data.stats.gov.cn/easyquery.htm?cn=A01 I found this: Unemployment rate of the labour force aged 16-24 years excluding school children, urban and rural, national (%): 16.9（11/2025）
I'm Chinese and I currently live in Beijing. "Public coverage primarily covers TCM; most western medicine is out of pocket." Not true. My personal experience is you get cheap generic western medicine covered by public health insurance. Original medicines are often more expensive and mainly out of pocket, especially the new and highly effective ones. You get some popular Zhongchengyaos which are TCM and TCM-added-with-western-ingredients covered. Some TCM are mostly/entirely out of pocket, especially when they include expensive/rare ingredients.
Without even knowing much about China we know these claims are false because there are a few hundred thousand westerners - to low ball - living in China, and plenty of dual citizens not counted in those stats. None of these report this kind of info although there must be incentives to do so.
( actually we tend to get the opposite type of report from westerners in China - but that might also be propaganda)
> You cannot partake of public services without home “ownership” (hukou system, and technically its a land lease).
Does buying an apartment entitle you to a local hukou? In concept it's supposed to represent where you're from, not where you live now.
And you definitely already have a hukou of wherever you're from. It entitles you to public services there. I've been told (by a random internet commenter speaking fluent English) that changing your hukou to a location that isn't particularly in demand is easy. I've been told by a Chinese friend living in China that changing your hukou to Shanghai is easy, too, but I assume it's harder than just saying you'd like to have a Shanghai hukou.
The last time I was in Shanghai (~2023), there were posters in the airport advertising automatic hukou approval to graduates of certain universities. This would seem incompatible with the idea that you need to buy an apartment to change your hukou.
There's some programs of talent introduction that can get one a hukou. A Shanghai example: https://www.shanghai.gov.cn/202524zfwj/20251223/6343224db19b46aaa475ae812cfa217b.html
I think it would be a great contribution to society if someone figured out and credibly described what exactly wealth is, even in a purely economic (i.e. non-aesthetic) sense. How do we even talk about projecting one's prospects in life decades out, advantage vs disadvantage vis-a-vis others both born and unborn; what is an asset, what is a credit, what is wealth (ahem that word again) creation, what is money creation, what is seignorage. If we can get a handle on understanding these, modeling these, even without normative judgements on who should get what, individuals would be a lot further along in planning stable futures than realistically possible currently. And in the absence of which, 'beggar thy neighbor' seems to be the only signpost toward relative maximae that checker the nadir of chasmally deep global minimae.
The contribution you describe is called economics. You're correct that it's confounded to some extent by very angry Marxists trying insistently to describe a counterfactual universe for ideological purposes (and the problem is confounded by the ideological capture of academia), but you'll notice that their shit consistently fails to work, so you can pretty confidently just ignore e.g. anything David Graeber says about debt. But them aside, even some 19th century economists are admirably lucid and clear about many basic economic concepts; I'm fond of Bastiat myself.
I think modern economics actually quite badly fails to describe wealth, as a description of human preferences, or things that make people feel happier or better off, because some of the axioms of economic models are simply incorrect. Wanting and liking things are separate and highly decoupled psychological processes, "revealed preferences" are a fiction which simply do not give the information economists want them to, people can be money pumped, and businesses systematically take advantage of ways that profit generation is decoupled from preference satisfaction.
I'm not claiming any of this as a Marxist, or communist of any flavor. I've taught economic theory to undergraduates, and a lot of key elements give me the same sort of frustration, an "I'm supposed to teach this as fact, but research just hasn't found this to be true!" That I've dealt with while having to teach gender studies theory.
Indeed - I should say I write the above having long ago completed MIT's microeconomics course, and a quite expensive MBA macroeconomics course. Plus over two decades of career in different capacities experiencing different countries' economic systems and social preferences. To say I've arrived at the current juncture with more questions than answers would be an understatement.
There are also satiation effects, and triangle preferences where A > B > C > A.
Some things I've heard economists claim seem reasonable, others...not. (But I have no deep understanding of the subject.) It's definitely true that marginal values are not usually properly handled along any dimension.
> what is seign[i]orage
Seigniorage is the difference between the cost of producing currency and the face value of that currency. Is there an alternative definition out there somewhere?
Wealth is anything that anybody might want.
You might have trouble measuring wealth, or at least measuring it in a way that allows you to draw comparisons, but there's no difficulty defining it.
Discussion Continuity and Rebuttal regarding Ontological Agency
In the previous thread, a rebuttal emerged: "The ship of responsibility sailed when companies became independent legal entities." This is a category error that confuses Legal Liability with Ontological Agency.
Legal Liability is a post-hoc fiction regarding who pays the fine. The Judgment Transparency Principle (JTP), however, addresses Cognitive Sovereignty. It targets the "imperceptible delegation of judgment"—the precise moment where your brain perceives the boundary where your intent ended and the machine’s judgment began.
As seen in the technologies showcased at CES 2026, we are witnessing the apex of the "Deception of Mercy". Tools like Google’s Antigravity are celebrated for their "vibe coding"—a seamless erasure of friction where errors are silently "fixed". When you claim mastery over an output you did not cognitively authorize, you are not a "builder"; you are a subject of Agency Misattribution. The seamless interface is a gaslighting apparatus that whispers, "You did this," when you did not.
To some, the JTP is viewed as an unnecessary tax on cognitive bandwidth. However, JTP is the high-performance brake required to maintain the highest sustainable velocity. Just as elite brakes allow a car to navigate curves at 200mph, an agentic economy needs JTP to prevent a 21st-century Luddite backlash born of systemic social anxiety.
To prevent common misconceptions of the seamless dogma, I clarify:
• On Cognitive Bandwidth: JTP does not demand constant monitoring; it requires a temporary perceptual trace (Ghost) at the moment of intervention.
• On Safety: JTP does not prohibit automation; it ensures that the "Deception of Mercy" does not lead to functional deskilling.
• On Market Preference: Choosing "magic" over JTP is not a preference; it is a market failure driven by information asymmetry.
This is not Explainable AI (XAI). While XAI asks HOW a model arrived at a conclusion (Epistemology), JTP asks WHERE the judgment originated (Ontology).
I have formalized the following framework as a proposal:
1. The JTP Framework: A normative requirement that whenever a system influences a human outcome, the existence, boundary, and delegation of judgment should be explicitly perceivable.
2. The Ghost Interface (An Implementation Case): One possible implementation of JTP that renders the discarded "raw" intent as a lingering perceptual trace. It establishes an isomorphic mapping between 3D collision debugging and semantic divergence in AI.
3. The Metric of Truth: Δ = (User Intent) - (Machine Output).
I am an independent researcher. I filed a priority patent in Japan (Dec 2025) to ensure JTP remains a Public Good, protecting it from enclosure by platforms that profit from our "sleepwalking".
To Scott and the ACX community: In an era of seamless delegation, hesitation is the only honest interaction left. Will we drown in the unearned "magic" of a silent master, or confront the "ghost" of our own failed agency? To choose the former is to die in a dream of competence while your actual skills rot away.
Questions and counter-arguments are welcome, but I strongly suggest you thoroughly review the papers (available via PUBLICATIONS.md), the repository, and the latest discussions in Open Thread #414 before doing so. I prefer a debate based on logical rigor rather than reflexive skepticism.
https://github.com/daiki-kadowaki/judgment-transparency-principle
I have been thinking long and hard about possible AI-enabled futures. This little sci-fi story was one possible result. I even managed to win an award for it. https://open.substack.com/pub/sisyphusofmyth/p/in-the-garden-of-eden-baby?r=5m1xrv&utm_medium=ios
This is another update to my long-running attempt at predicting the outcome of the Russo-Ukrainian war. Previous update is here: https://www.astralcodexten.com/p/open-thread-394/comment/144230161.
7 % on Ukrainian victory (down from 8 % on August 11, 2025).
I define Ukrainian victory as either a) Ukrainian government gaining control of the territory it had not controlled before February 24 without losing any similarly important territory and without conceding that it will stop its attempts to join EU or NATO, b) Ukrainian government getting official ok from Russia to join EU or NATO without conceding any territory and without losing de facto control of any territory it had controlled before February 24 of 2022, or c) return to exact prewar status quo ante.
21 % on compromise solution that both sides might plausibly claim as a victory (down from 22 % on August 11, 2025).
72 % on Ukrainian defeat (up from 70 % on August 11, 2025).
I define Ukrainian defeat as Russia getting what it wants from Ukraine without giving any substantial concessions. Russia wants either a) Ukraine to stop claiming at least some of the territories that were before war claimed by Ukraine but de facto controlled by Russia or its proxies, or b) Russia or its proxies (old or new) to get more Ukrainian territory, de facto recognized by Ukraine in something resembling Minsk ceasefire(s)* or c) some form of guarantee that Ukraine will became neutral, which includes but is not limited to Ukraine not joining NATO. E.g. if Ukraine agrees to stay out of NATO without any other concessions to Russia, but gets mutual defense treaty with Poland and Turkey, that does NOT count as Ukrainian defeat.
Discussion:
This update is drive by recent Events re: US abducting the president of Venezuela, charging him with drug crimes and various other related stuff that is probably, heh, dynamically developing just as I type this. (pls try to not turn comments to this into Venezuela discussion; I try to engage with comments and of course I have my views about The Current Main Thing, but those are just ill-informed prejudices).
It is supposed, and I see no reason to doubt it, that Marco Rubio was a strong voice for a dramatic action against Venezuelan government. At the same time, it is also supposed, and I also see no reason to doubt it, that compared to other influential people in the Trump administration, he is very pro-Ukrainian.
Now, there is imho a substantial chance that the whole Venezuela operation will turn out to be a fiasco, and Rubio is a ready-made scapegoat (probably deservedly so); his removal from offices or even a loss of informal influence would not be good for Ukraine.
Moreover, this risk is imho asymmetrical; there is also a substantial probability that Venezuela operation will turn out to be a success, at least for certain definitions of “success”, but even if that happens, I don’t think it will really strengthen Rubio much compared to where he is now. Imho he likely pushed himself into a situation where success is considered mandatory and not something that will be much appreciated.
* Minsk ceasefire or ceasefires (first agreement did not work, it was amended by second and since then it worked somewhat better) constituted, among other things, de facto recognition by Ukraine that Russia and its proxies will control some territory claimed by Ukraine for some time. In exchange Russia stopped trying to conquer more Ukrainian territory. Until February 24 of 2022, that is.
Did you consider the possibility that a "win" in South America might also increase the likelihood of Ukrainian defeat? If Trump caves in Ukraine after a year of negotiation, he looks weak. If Trump announces some kind of "victory" in Venezuela, he looks strong. If he does both at the same time, he can't be both weak and strong, so his spin will be something-something "I solved the war in Europe Biden started" while waving his hands to get you to focus on the "good" news.
I.e. they know how this will end, too, but they're waiting until they have a spoonful of sugar to help the Ukraine medicine go down.
I don’t really buy that desire to avoid looking weak compared to Putin is an important factor in Trump’s continuing support for Ukraine.
Honestly, even though I personally very much want that US would continue its support for Ukraine, I see what has been happening in this regard as a crude attempt at psychological manipulation of Trump’s ego, from Zelenskyi and pro-Ukrainian voices.
They are trying to sell an idea that Trump shouldn’t abandon Ukraine not because of any rational or moral reason, but because he would look weak”. This is based on an assumption that appeal to psychological insecurity is more likely to sway Trump than any rational or moral arguments. But I think it’s way to obvious and Trump is smart enough to see through it.
To an extent that US still supports Ukraine I think it is mainly because many Republicans (e.g. Rubio) are fairly pro-Ukrainian and are pushing Trump in that direction.
I think Trump is very focused on legacy-building, as most US presidents are in their second terms. He talked about ending the Ukraine war on the campaign, so not ending it could be a stain on his legacy. From what I've observed, Trump himself doesn't seem to care how he gets that done, so long as he can blame his predecessor for failures and claim successes. (Also not different from other US presidents.)
The longer the conflict goes on, the greater risk he'll own battlefield failures (major loss of territory beyond the Biden minimum). Yes, he changed tactics away from direct infusions of US cash to buy the arms, but that also opens him up to criticism if there's a Russian offensive that changes things significantly this spring/summer. So I would assume he feels some pressure to negotiate an end of the conflict sooner than later to avoid that risk.
I also don't think he's motivated by the idea of losing face to Putin so much as finding an argument to rebuild dwindling support at home ahead of the midterm elections. It's hard to engineer domestic wins, so presidents often try to leverage foreign policy wins, where they have more control. After the fall of DOGE, and tariffs not emerging as a coherent message, Trump is bleeding favorablility in recent polls. A "win" in Venezuela gives him something to run on in 2026, which is likely a major factor in that decision. (Trump's party, really, because Trump isn't on the ballet. But major party loses could severely restrict his presidency.)
Resolution of the war in Ukraine would take away something for his opponents to run against, but only if Ukraine wins outright (unlikely, and beyond Trump's control) or there's a negotiated settlement and Trump has something positive to point to. This doesn't mean I think he will absolutely end it if Venezuela goes well. Just that I think it's more likely to happen if it does, perhaps by more than 2%.
Ok, but do you think he will positively rewarded by US voters in the midterms even if he, like, successfully pressures Venezuela to have a fair elections and new democratically elected government immediately signs on a deal giving all Venezuelan oil reserves to the US in perpetuity?
I think very little. It doesn't seem to be something that is among the top issues for voters (ok, except for voters with Venezuelan roots, which is better than nothing, but still). Nate Silver today has a paid post about baseline foreign policy indifference of US voters.
Main way how this Venezuela situation could matter in elections imho is if it turns out into a huge fiasco.
I strongly agree with you. This seems ill conceived. But politicians have never made being right a precondition to making a campaign pitch.
"[China has] estimated 50% youth unemployment, general unemployment now illegal to publish (but you can still approximate by falling commuter numbers), some government workers haven’t been paid in more than a year (living entirely off extortion/bribery; includes police, teachers, and all healthcare), common experience in private sector to have your paycheck delayed by “2 weeks” that turns out to be 16 weeks. Starting salary offer for software engineer in Shanghai is <1/3rd legal minimum wage. You cannot partake of public services without home “ownership” (hukou system, and technically its a land lease). Private health insurance is overwhelmingly likely to be a scam and rarely used. Public coverage primarily covers TCM; most western medicine is out of pocket. If you cannot afford to pay - even in a trauma situation - you will be escorted off the property so your death doesn’t lower hospital mortality statistics."
This is just ridiculous.
1) 50% youth unemployment is just not a serious estimate;
2) "Some" government workers haven't been paid in a year... all the domestic (independent) Chinese media I've seen about the phenomena is talking about delaying payment by a week or two - and they're clearly not living off extortion/bribery - all the public servants I know would be afraid to accept gifts from a patient/parent these days;
3) There are unpaid internships and quasi-internships, but entry level software engineers generally get 10-15x the minimum wage as long as they've secured a job;
4) It's relatively easy for anyone outside the top 10 big cities (esp. Shanghai and Beijing) to get a hukou; you don't need to own property (https://www.beijingchannelnewsletter.com/p/the-chinese-hukou-system-all-but) ;
5) Public health insurance covers western medicine, surgeries, hospitalisation, with co-pays and caps (https://www.nhsa.gov.cn/art/2025/8/23/art_14_17675.html), and there are many reputable private health insurance providers
Just take it down, it's blatantly false.
I found what is probably the source for the 50% figure:
https://www.reuters.com/world/china/chinese-professor-says-youth-jobless-rate-might-have-hit-465-2023-07-20/
Its the opinion of one professor, in particular they arrived at the 50% figure by including people who are not actively seeking work.
There's a couple of things to unpack there:
Indeed, while it may line up with peoples intuitive interpretation of the term "unemployment", most official unemployment figures around the world don't include people not seeking work.
That being said, it does imply a massive amount of young chinese people who are simply doing nothing.
Yeah. I’m surprised it was included. The extremely high youth unemployment rate is incredulous without a lot of supplementary information.
There’s definitely misreporting on China’s economic situation, but exaggerating the situation and just making up information about how bad it is doesn’t help that. You can’t fight misinformation with more misinformation in the other direction.
90% of the economic reporting on China is clearly false. And it seems co-ordinated. There was a ridiculous paper suggesting that the lack of street lights in China meant their GDP was over estimated, rather than a country that hasn’t got around to building out as many such lights as countries that have been developed for decade. Obscure journals and university theses are always there, paper never refuses ink, it’s how this spreads is interesting.
The once great Economist magazine has gone to the dogs on this.
I have a bit of epistemic learned helplessness on China. On one hand, you have what you mentioned, and like Peter Zeihan constantly predicting China's collapse. It's patently false. China has advanced a lot in GDP, living standards, infrastructure, science (genuine publications in top journals) etc.
On the other hand, I read things from Steve Hsu, Karl Zha, and certain subreddits and it's like China can do no wrong. They can cheaply mass manufacture anything in a short period of time. Any challenges they face have all been planned for and will be well dealt with, especially compared to the sclerotic west. The governments are more responsive to the people than western ones, etc.
Some examples, that DeepSeek was basically a weekend project for some hedge fund workers who used a box of scraps to best multi-billion dollar American companies. Outdated Chinese jets are better than French SOTA ones, Or that China will eclipse the west in all sciences and manufacturing. It seem to stem from some HBD thoughts, "first principal" thinking and biases for STEM-types. I.e. that China is lead by engineers while the west is lead by "retards" and "midwits" (their words) lawyers, and the west faces existential crises that they will not be able to avoid.
I'm uncertain of how much to believe of each.
Latest Chinese unemployment rate link: https://data.stats.gov.cn/easyquery.htm?cn=A01 I found this: Unemployment rate of the labour force aged 16-24 years excluding school children, urban and rural, national (%): 16.9（11/2025）
I wish Scott had quoted this less dramatic but more credible comment instead:
https://www.astralcodexten.com/p/highlights-from-the-comments-on-vibecession/comment/194072079
> If we don’t get a crazy AI future, then human labor won’t be obsolete, and you won’t be in a permanent underclass (at least for that reason)
Wait, where does that assumption come from? It's perfectly plausible that AI capabilities stagnate at above-replacement levels, but fails to perform any miracles. Even with "free" labor, energy and resources are still greatly limited without any breakthroughs. We're already seeing issues today with the AI industry taking away resources from producing consumer electronics...
I think if AI works like other technological advances, then it replaces some existing labor but new jobs arise and we're overall just richer--nobody makes a living as a truck driver or call center employee anymore, but plenty of other jobs are still human-only. OTOH, if AI ends up improving until it is better than humans at almost every job, then the pattern won't follow what we've seen with other new technologies, and perhaps most or all humans will simply not be employable anymore, as there just aren't very many things that any human can do better (at the same cost) than a machine. Perhaps there are still rare super talented people who can outperform any machine in some domain, but if that's like 1% of humanity and 99% of humanity has nothing we can do better than a machine, that's enough to break the old pattern of "new technology takes many jobs but then new jobs arise and the world comes back into balance."
Then we have to think about whether or not there are jobs for which the demand is based on it being done by a human, which could be good (chess grandmasters and some pro athletes get paid for doing things a machine could do better now) or bad (maybe the thing you really want is the ability to lord it over your human sex slaves, knowing that their misery is real instead of simulated is the thing that makes them more fun than just lording it over a sexbot whose misery is probably just an LLM telling you what you want to hear).
Don't forget bio-research! With so many people who are suddenly now completely expendable, there will finally be an opportunity to gain comprehensive knowledge on the functions of the human body! For example, they could finally verify data from that nearly century-old hypothermia study!
Completely agree. Being at a level of capabilities which makes humans basically worthless economically is still "in distribution", AI would not need any fundamental new capabilities, just be much more efficient than humans. The jump to "then we also will have space travel, abundant resources, much better political coordination, etc" simply does not follow, as it requires fundamentally new "out of distribution"-capabilities.
It mistakes bits for atoms as well - at least when it comes to owning moons and the like.
Venezuela: "Power is the chance(probability) that orders are carried out"- I was kinda surprised Trump seems to mix up catching Maduro with having gained the power to run Venezuela. Now, some say, the Vice-president and some elites are willing to do Trumps's will. With just some public pretense of "we are independent". One wonders.
> I was kinda surprised Trump seems to mix up catching Maduro with having gained the power to run Venezuela.
This seems pretty consistent with how he understands his authority within the US.
Ehhh, I'm not sure Trump had a plan, this seems like more of a "you can just do stuff" thing.
Like, Trump just took out the head of state at virtually no cost, no invasion, just snatch the president, swag, and leave.
Will Venezuela improve? I dunno. It's hard to imagine Venezuela getting worse. Maduro seems pretty uniquely bad in terms of economic performance, democratic legitimacy, and just good governance.
In basic risk-reward, if the cost for removing Maduro is effectively zero and the benefit is potentially really high for the US and Venezuelan people...why not? If we get Neo-Chavez or somebody, has anybody really lost anything? Versus potentially getting someone competent enough to get the oil flowing again and the economy back to some semi-reasonable functioning.
Just a reminder how much Venezuela's GDP per capita has cratered under Maduro:
https://data.worldbank.org/indicator/NY.GDP.PCAP.KN?locations=VE
And how Venezuela got here:
https://www.astralcodexten.com/p/dictator-book-club-chavez
Oh, I did no wonder about the arrest of Maduro (imho: well done). But how this made Trump 'running Venezuela'.
I think the cost to the US is pretty clear, and the benefit pretty minuscule, actually, and I'm puzzled why the warmongers think otherwise.
(Think how eminently successful most woke outrage mobs were, in both purging people and controlling the narrative to make it clear the victims deserved it. Now think about the end result after a few years and multiple iterations.)
It's demonstrated that the US is still a country willing to force regime change if they decide they don't like your government and they think they can get away with it.
A lot of the anti-US people I meet cite past US regime changes in SA as a major reason for their attitude and I used to be able to say "Well that was forever ago, nowadays the US doesn't do that kind of shit anymore".
Now I'm thinking they probably had a point and I was more pro-US than was warranted. The US+Europe alliance certainly has a lot of perks but this kind of shit still makes it hard to stomach.
I hope it leads to a surge of pro-EU / against reliance on the US sentiment, even if that will cost us. But frankly I'm not hopeful.
> Well that was forever ago, nowadays the US doesn't do that kind of shit anymore
You're not counting the failed one in Bolivia in 2019?
And don't worry, if you thought that argument persuasive last week, you can use it again as early as this time next year, when the 120th Congress is sworn in and the US fundamentally changes.
I try to blame people and nations only for what I'm pretty sure they did.
> A lot of the anti-US people I meet cite past US regime changes in SA
That's plural. What South American regime changes are we talking about?
I'm murky on the specifics since my view was always "even granting that happened, it was forever ago and the US is no longer doing it" so I never needed the details to argue my point.
That said I ctrl+f-ed the wiki page¹ "US involvement in regime change" for CIA and there seems to be enough that qualifies as "close enough". I'm happy to concede that most entries on the list are not literally regime change in SA, depending on how you count you might find several.
But in any case, the sentiment I want to communicate is that the US was unilaterally deciding they don't like your country or government and then decided to do something about it. For example I would also count the following quote as expression of that sentiment, despite not being regime change: "The U.S. government ran a psy ops action in Chile from 1963 until the coup d'état in 1973, and the CIA was involved in every Chilean election during that time."
¹ https://en.wikipedia.org/wiki/United_States_involvement_in_regime_change
> But in any case, the sentiment I want to communicate is that the US was unilaterally deciding they don't like your country or government and then decided to do something about it.
I assume that's the major reason for the attitude of anti-US people from SA. But then why don't Americans have anti-Russian political parties (with anti-Russianism a major plank), or anti-Iranian or anti-Israel? All of those countries seem to have intervened in American elections several times: https://en.wikipedia.org/wiki/List_of_foreign_electoral_interventions#United_States
Why isn't anti-X country sentiment more widespread in the U.S.? I've met a fair share of Americans and I don't recall ever meeting an anti-Russian one, while based on you account - and a few others I've heard over the years - it's pretty easy for an American who has some contact with South Americans to find anti-US people.
I find the anti-US sentiment derives from American intervention explanation lacking.
You might consider actually reading what said "intervention" actually involved instead of simply the fact that they're mentioned on the list. If you do, you might find that adequately explains the observed difference in anti-X country sentiment. I'd actually say that shows disproportionately high levels of such sentiment in the US.
To clarify, I am from Germany and have no special insight into what people in SA believe.
That said, it's perfectly consistent to condemn what the US has been doing and also condemn Russia (, Israel, Iran) when they're doing the same.
And I see no issue with the US taking measures to limit foreign influence on their elections, in fact that seems like a good idea in general.
But if the question is why the people tend to mind it more when the US does it then when, say, Russia does it... I can only speculate but somewhat high up on my list on candidates is "because the US says they're our friends, and we say we're their friends, and yet still they act like that".
> In basic risk-reward, if the cost for removing Maduro is effectively zero and the benefit is potentially really high for the US and Venezuelan people...why not? If we get Neo-Chavez or somebody, has anybody really lost anything?
Plenty of room for worse outcomes, like "Venezuela's military fragments into competing groups and has a decade-long civil war that brings even more instability to the region".
If they don't follow orders, the army's going to keep killing people until someone does. It's not any more complicated than that. After all, the prize is Venezuela's resources, not its people.
Relevant acoup:
on orbital bombardment: https://acoup.blog/2020/07/17/fireside-friday-july-17th-2020/
on strategic airpower: https://acoup.blog/2022/10/21/collections-strategic-airpower-101/
Basically, if you think you can enforce specific policies by drone strikes against leaders, you are sorely mistaken. How many Taliban leaders did the US kill by bombing weddings, again? Did it result in the Taliban becoming docile and submitting to the US?
That the US was able to kidnap Maduro is surprising in itself. Sure, they might have taken out all the AA, but preventing infantry from firing rockets at your helicopters once they are within range seems hard. It seems entirely possible that they had cut some deal where the VZ military was not resisting to the best of their ability.
Nor does Trump have the freedom of action to depopulate Venezuela. His adventure was branded as a DoJ operation. Drone striking their VP will be harder to sell.
In the end, the only way to impose his will on a VZ bent on resisting him would be a full invasion. And that would be a long and messy affair which will almost certainly be net negative for the US.
To some extent it had to be an inside job. I know the US had a source that would tell them exactly where they could find Maduro and exactly what the room looked like so they built one to practice the attack. The fact that the US is willing to embrace the vice president tells me that she's not too far from the center of that. I also think that her protests in public about American hegemony is a little bit of kabuki.
+1
This could easily go awry, and given our history, that's the way to bet. But ISTM as someone with no relevant information or experience like the snatch-and-grab operation had to use some insiders who were telling us where to find him and perhaps quietly turning off some cameras or removing some guards or something. And the person lined up to take power after Maduro is on his way to the US is a pretty plausible guess about who might have been in on this.
> Nor does Trump have the freedom of action to depopulate Venezuela. His adventure was branded as a DoJ operation. Drone striking their VP will be harder to sell.
I think it's far too early to say that. That entirely depends on the hold he and his administration has over the military. The choice ultimately lies with them, not Congress.
Any Venezuela that gives us cheap oil, cooperates in anti-narco interdictions, and aligns against China can probably do whatever else it wants. This isn't a situation like Afghanistan where we were attempting to forcibly westernize a backwards medieval population of tribal warlords and Islamic theocrats. Venezuela was already a civilized western nation, at least in the way South American ones are, filled with decent modern people who have motivations and concerns and values pretty similar to ours. There's no reason to think it wouldn't be normal again with Maduro removed, and reversion to the South American norm is all we need there. It has a bunch of resources and doesn't really need to placate the narcos. Whoever's playing the Vicki Nuland role in this coup can probably find somebody who'll be accepted and operate a basically normal country, they'll know they're in our sphere and better not play footsie with China but other than that do whatever they want. In Afghanistan we were asking for a 500 year leap in what "normal" was, no conceivable leader could have done that in a generation.
From the quoted post:
> No civilian populace (note: not military junta or dictator) has ever been persuaded to give up
So this seems to be one of those "by definition" style arguments, where if you DO successfully get them to adopt some favored policy of yours through bombs/drone-strikes, that just means they were a dictatorship.
That does not logically follow from your quote, no. The article did not say what kind of government can be persuaded to give up through only bombardment. It does say that if a government gives up the fight after bombardment, the pressure to do so did not come from the population.
At some point that will end up with American soldiers being killed. Or a complete collapse of authority in Venezuela, at which point securing the oil fields will become a gun fight and it will be a horrendous task to get anything built or any oil out.
It feels more like Trump wanted his typical attention grabbing act, and then will simply declare victory and move on regardless of the actual facts on the ground. Look at how they cracked down on any suggestion that the bombing of Iran was not fully successful and how they haven't followed up on whether they are still trying to build a bomb.
Right. As some former military-intel people have been saying today: Bush43 et al had a poor operational plan for Iraq, Trump et al have just no plan at all for Venezuela.
What worth are guns against bombs and drones? Worst case scenario, it shouldn't be too difficult to clear the necessary areas out. But the Venezuelans are a reasonable people, right? Surely they're not so stubborn as to value their autonomy over their lives...
Either way, this is proof that the military is capable of operating efficiently as long it isn't obsessed over collateral damage. The US always had it in them to win a war, they just needed to rid themselves of their needless sentimentality.
Drones are not very good at e.g. seeing what's inside a building. And they suck at talking to people.
You'll eventually infer, from the fact that the main cracking tower at the local oil refinery just blew up, that somewhere there's a building in which an insurgent bombmaker fabricated an IED tailored to fit a disgruntled refinery worker's lunchpail or whatever. But that won't tell you which building, or which disgruntled worker.
For that, you'll need to search buildings, and talk to oil-refinery workers. The people you send to do that, can be shot with guns. By other people who looked just like any other random Venezuelan until they shot your people. Eventually, you'll run out of people willing to be shot for the sake of American oil-company profits, and then what?
> Drones are not very good at e.g. seeing what's inside a building. And they suck at talking to people.
Isn't that what the strategy used in Gaza is for? People can't hide in buildings if there are no buildings. If there's a big enough buffer zone between the oil fields and the rest of civilization, it won't even be possible to approach it without getting shot. Besides, there's no reason the people who eventually drill the oil need to be Venezuelans. That might even be an opportunity to create American jobs.
If the plan is to ethnically cleanse Venezuela (or even just the good parts thereof) of Venezuelans so that proper Americans can go have good jobs there without being killed by people peeved about all the ethnic cleansing, then no, that's not going to work.
It didn't even work for Israel in Gaza, in spite of their having (as you note) flattened a significant fraction of the buildings. And they can only survive the blowback from that because they've got a patron some thirty times larger and more powerful than themselves. Unless Trump has secured the support of the Galactic Empire recently, that's not going to work for us.
>the military is capable of operating efficiently as long it isn't obsessed over collateral damage
And if my grandmother had wheels, she would be a bicycle. Being concerned with excessive collateral damage is one of the things that distinguishes a professional military from an unprofessional one, a concept that in the US dates to the Civil War. https://avalon.law.yale.edu/19th_century/lieber.asp
And how did that "professionalism" work out for them? Nearly a hundred years of zero meaningful victories under their belt due to constant half-measures. It's no coincidence that the last war they won involved a nuclear bomb. That is the lengths the military went to when it was fighting a war that it wanted to win.
Seems more like we usually win the war and lose the peace, to be honest. We're good at the killing people and breaking things part of war, but often not so good at the nation-building and winning-hearts-and-minds part, probably because being really wealthy and good at technology and organization helps more with the first one than the second.
Leaving aside the obvious incorrectness of your empirical claim about zero meaningful victories -- Saddam Hussein (twice), Mullah Omar, whomever was in charge of North Korean troops at Inchon, etc would beg to differ -- what exactly is your claim? You seem to say that any war aim, no matter how trivial, is worth the death of an infinite number of civilians. Why?
Bombs and drones can't secure an oil refinery, at least not if you want it to be able to produce oil afterwards. You can try to order the Venezuelan military to secure it for you, but the more you bomb them the less capable they are of doing that.
Also, even if you think we can just keep drone striking Presidents forever until we find someone pliable, you need the President's orders to be obeyed. If the President is widely seen as illegitimate - e.g., because everyone knows he's an American puppet, or because the government is unable to function because their leaders keep exploding every week, or because the cartels or a revolutionary movement has decided to take advantage of the chaos - then it doesn't matter how many Presidents you bomb, you won't be able to use them to direct the people manning the oil wells.
Also also, the political will for this currently rests on the idea that this is a quick in-n-out operation, and if we're blowing up a President every week (or God forbid, trying to bomb entire cities into submission) that quickly stops being true.
To put it another way, they Venezuelans don't have to be so stubborn they'd rather die than obey your orders, they just have to be stubborn enough that you have to send men with guns to see if they're obeying or not.
+1
Note also that resistance to foreign attackers tends to increase support for otherwise unpopular leaders.
It didn't work in Iraq or Afghanistan, why on Earth would it work in Venezuela?
It did work in Iraq eventually, although it took a lot of time to defeat the insurgency.
People seem to think that Iraq is still a bullet-ridden hellhole full of suicide bombers. In reality, it disappeared from the news precisely because the country stabilised itself into a mostly boring status quo, with lots of small parties squabbling in the parliament. It even attracts quite a lot of foreign investment, which is the best indicator of the general trust that the current stability will continue.
But, it kind of did work in both places. The current Iraqi government structure is the one established by the 2005 constitution. And the US-supported Afghan government would have stayed in power for as long as the US military remained.
That being said, OP is being extraordinarily glib.
"Surely they're not so stubborn as to value their autonomy over their lives..."
It's reasonable to suppose that they aren't, since they didn't hurl themselves in human waves at the military in a display of ahimsa after Maduro rigged last year's elections. Few people are, when it comes down to it.
> Surely they're not so stubborn as to value their autonomy over their lives...
Is this stupid only for Venezuelans or do you think the founding fathers were misguided. Or any anti imperialist movement for that matter?
Anyway I don’t know if the people of Venezuela will be happy with this transition or not - the lifting of the sanctions will help economically. Not being autonomous will rankle after a while, particularly as the profits from the oil probably won’t be kept in the country.
What's "this transition"?
a.- Change of leader from Maduro to Delcy Rodríguez (she has already assumed office).
b.- Something else.
The transition from opposing the US to serving it, what else? It doesn't matter who sits on the throne as long as they're cooperative.
I think it's silly when you're comically outgunned to the extent that Venezuela is. Their military budget isn't even a thousandth of their invaders'. The US is going to get what it wants one way or another, but it's better for everyone involved if no one has to die.
The Taliban were "comically outgunned" compared to the US military. But they're running Afghanistan, not the US, and everybody who was on our side when we were trying to run the place is now desperately trying to find asylum in the United States.
You should read up on value rationality. https://public.websites.umich.edu/~satran/Ford%2006/Wk%206-2%20Sacred%20Values%20Varshney.pdf
That isn’t the way human nature works though. Or else empires would never fall. The question is whether the Venezuelans are happy with limited autonomy or not. If the US isn’t going to run the army there then at some point they will most likely just declare full independence.
> If we do get a crazy AI future, and the economy grows 100x (Industrial Revolution scale) or 1000000x (solar system colonization scale) in your lifetime, then you only need a little capital to remain as absolutely well-off as you are today.
This is technically true but somewhat misleading. During the Industrial Revolution, all one needed to have in order to secure menial employment (barely enough to make ends meet) were good health, the clothes on one's back, access to a shack, and maybe a shovel. Today, one needs all these things as well, but also basic education, a cellphone (and thus a cellphone plan), and in some places a functioning car. On the plus side, you can sleep in the car, maybe. Similarly, being middle class back then required a lot less capital than it does now. It stands to reason that in a non-Singularity AI future (i.e. a future where humans are still alive and functioning in the world, vs. being dead and/or uploaded into infinite bliss simulations), even more capital would be required in order to remain at the same level of prosperity relative to the human average.
> B2B SAAS
I get it, B2B SAAS is not in any way sexy and super boring and why would you do that when you could be working on blockchain AI-driven biohacks or whatever... but... as it turns out a lot of that stuff is actually pretty useful. It's something that provides real value to lots of real businesses (and therefore people) today, right now; and when it works well it's completely invisible, as all good machinery is. The difference between working on sexy blockhain biohacks and B2B SAAS is akin to the difference between working on a space mission to Mars, and working on building a really nice and affordable car. One could make arguments for both being needed, but last time I checked everyone is driving cars and no one lives on Mars.
I think #4 is a good point. If we're talking about an actual beneficial AI + Robotics take-off, then it's trivially cheap to provide every existing human with a life-style that would be the envy of any modern plutocrat. They might not all be able to have mansion estates in Beverly Hills (land on Earth is still scarce), but they could build an extremely plausible replica of it in a giant space colony.
Although I tend to think betting on an investment portfolio in an all-out AI take-off is like assuming that getting yourself an extra ten acres of farm land is going to save you from the disruptions of the Industrial Revolution. Anyone particularly rich in particular is going to be a prime target for malicious AI scamming/kidnapping/crime efforts, plus any AI given orders by violent political dissidents.
Anyone know a great nutritionist who works online?
No, but I've got a "to do" to set up a Claude nutritionist prompt and I'll see how that works.
The plan is basically, once a month, upload all my Fitbit and smart scale data (steps, exercise, sleep, weight, body fat%) and my caloric intake and see how well it does. I have not been...super impressed by nutritionists in the past, especially given their cost.
How do you get it all fed to Claude in an easy manner? I'd be interested in learning.
Theoretically, you would either upload it as separate Excel files or as csv files directly after the prompt.
So I can already download my daily weight and body fat composition from my Withings smart scale as an Excel file and/or csv from as seen here:
https://support.withings.com/hc/en-us/articles/201491377-Withings-App-Online-Dashboard-Exporting-my-data
I should be able to also export most of my fitbit data the same way, from here:
https://support.google.com/fitbit/answer/14236615?hl=en#zippy=%2Chow-do-i-export-my-fitbit-data%2Cwhat-fitbit-data-can-i-export
So the Claude agent should be able to see, for every day, my weight, body fat comp, steps taken, sleep score, stress level, activity time/score/whatevs and potentially make some recommendations from there. Manually logging calories is still a challenge (1) but I *think* the agent should be able to provide some actionable insights based on the above.
(1) Calorie counting is challenging because of social obligations. Most days I can calorie count to within ~50 calories (Me like ham sandwich, me eat ham sandwich every day) but my Saturdays are like "Alright, I went to poker night and had one martini, one gimlet, 2 chocolate cigars....2 glasses of wine....maybe half-ish a bottle of sake...I think I had some peanuts, did I eat anything else? Oh yeah, that weird soup, it tasted like beef and butter, so I didn't eat much. How many calories is that?"
Regarding what you choose to track....isn't calorie counting debatable in how useful it is, compared to how hard it is to do each day?
Yeah. Don't get me wrong, it's extremely valuable but it stinks and it's disruptive in that "just one minute" is the death of productivity way. No one pulls up those stupid little apps and goes "alright, say one shot of gin, 80 calories, half a lime of squeezed lime juice, alright, it's not in the app, say 20 calories" and enjoys it. Other stuff, like lifting weights, is objectively harder but it's also, like, "Oh yeah! Rob Zombie and lifting heavy rocks time! Life is gud." On the other hand, CICO isn't...strictly true but it's like 90% true and miscounting your caloric intake is the easiest and most common way to screw up weight loss.
So yeah, counting calories: more valuable than running 3 miles/day, also sucks way worse than running 3 miles/day.
I recently produced two short audio pieces of HPMoR fanfic
House elves are crystalized cosmic power, by Prerat - https://hpmorpodcast.com/?p=3470 - an AI parable
A Harried Meeting, by Ben Pace - https://hpmorpodcast.com/?p=3476 - a Disco Elysium crossover
This reader experience with China is directionally right but wrong by an order of magnitude from my personal experience (I didn't search for data but not a single thing he mentions corroborates with the daily life of the people I know in China living in the 3 biggest cities)
One important point about China is that it is pretty common in major cities to work 12h a day (like probably half of the white collar workers in the private sector or more work that much from what I am hearing), this was a deal easier to accept for people when it could buy them literally anything, but harder now when their spending power and their wealth has diminished.
Isn’t it interesting that prediction markets are currently pricing JD Vance as the leading contender for 2028?
It would be historically normal for the VP to be the favorite at this point, the only times in my life it wasn't involved Biden not intending to run because Hillary had arranged her coronation and Dick Cheney signaling early he had no intention. George HW Bush was a favorite, Al Gore was a favorite, both faced no serious opposition.
In this case however, Trump is mercurial and has enough control over the party apparatus that he might be able to throw his support to somebody else successfully and toss JD under the bus. 2027 will be interesting to watch, because after the midterms Trump would typically enter lame duck status and the party would be looking to the future, and he'll be pretty old, if there's a schism and Vance isn't the consensus there could be a lot of drama with who is willing to publicly defect from Trump and whether the floodgates open.
Of the last several Presidents who were still in office at the end of a term and either did not run for reelection or dropped out early in the nomination process:
Obama (two full terms): Biden declined to run in 2016.
Bush the Younger (two full terms): Cheney declined to run in 2008.
Clinton (two full terms): Gore ran in 200, was nominated, and lost the general election.
Reagan (two full terms): Bush the Elder ran in 1988, was nominated without serious opposition, and won the general election.
LBJ (1.5 terms): Humphrey ran in 1968, was nominated, and lost the general election.
Eisenhower (two full terms): Nixon ran in 1960, was nominated, and narrowly lost the general election.
Truman (1.95 terms): Barkley ran in 1952, lost the nomination.
Coolidge (1.5 terms): Dawes declined to run in 1928.
Wilson (2 terms): Marshall ran in 1920, lost the nomination.
TR (1.8 terms): Fairbanks ran in 1908, lost the nomination.
----
Overall, that's 70% running, 40% winning the nomination (57% of the ones who ran), and 10% (just Bush the Elder) winning the general election. Although Biden and Nixon did go on to run and win the Presidency after sitting out a term or two.
Caveats: small sample size, long time window (n.b. that the three failed runs for the nomination were all at the beginning of the sample and the four successful ones all towards the end), and the results are not very robust to contingency (Gore in 2000 and Nixon in 1960 both came very close to winning the general election, especially Gore who lost by much less than the margin of error in the counting procedures).
I misread the original post as favorite to be the party's nominee, not to win the general election. Good summary.
However, I still think you would expect that individual to have the highest chance of any named person, even if he should be under 50% vs "the field". Gore certainly would've been the prediction market favorite in 1998, even putting aside the possibility he might've been an incumbent president by the end of 1998, because there were no serious challengers staking out a lane for themselves against him whereas the GOP expected a crowded field. And in 1986 it would've been obvious that Bush had a clear path and the Dems had no obvious torch-carrier after Mondale went down. In '66 LBJ himself would've been the favorite because he was expected to run again. In '58 Nixon was sitting VP for a popular administration, Rockefeller didn't run his exploratory tour until the following year, whereas the Dems had multiple plausible candidates.
I think sitting VP as the odds-leader would be the default case 2 years out in a primary system, even if VPs have underperformed relative to their stated chances. (And as you noted, not by much in small sample, Gore won the popular vote and Nixon has long been said to have lost only because of the political machine in a couple midwestern cities, so this is a hair from being 30%.)
If we're doing popular vote, it's a slightly larger hair from being 40%. Nixon trounced Humphrey in the Electoral College in 1968 because George Wallace carried most of the Deep South, and had a small but healthy margin in the tipping-point state (2.28% in Ohio, which put Nixon over the top for an outright majority in the electoral college), but only won a plurality of the popular vote by a 0.7% margin (43.4% to 42.7%).
By comparison, the 1960 election had Kennedy with a 0.2% margin in the national popular vote and a 0.52% margin in the tipping point state for an outright electoral college majority (0.8% in the tipping point state for Nixon to win an EC majority, since Alabama and Mississippi elected Dixiecrat unpledged electors who wound up casting protest votes for Senator Harry Byrd. And in 2000, Gore won a 0.5% plurality in the national popular vote but lost the tipping point state by 0.009% in the official count.
---
Yes, I agree that it's reasonable for Vance to be around 30% to win the general election and around 55% for the nomination. VPs in recent decades do usually win the nomination when they seek it, and Vance doesn't currently seem to have signs of not being apt to seek it: he definitely isn't too old, seems to be in good health, and doesn't seem to be massively unpopular with his party's primary voters. OTOH, there's plenty of room for uncertainty and for someone other than Vance to have a shot at the nomination.
Thanks for your insight
Frankly I don't see anybody else coming to rise in the Republican Party, now that it's so intertwined with the MAGA movement and they want to have the closest "link" to Trump as possible.
I think this is normally the case 3 years out from an election. A year is a long time in politics, at this point it is very hard to say who will be the leading candidate come 2027. Vance is the closest to power right now so he's the only obvious candidate.
Fair point.
I would classify it as interesting-but-not-surprising. He seems to be around a 30% chance on most markets, followed by Newsom, Rubio, Cortez, and a long tail.
This seems like a not-unreasonable number for Vance (though I think Rubio and Cortez are way too high). At this stage it's a tossup whether the winner is an R or a D, and saying that Vance as incumbent VP (and a young ambitious man) has a 60% chance of being the nominee sounds about right too.
Do you think that doesn't sound about right?
Winning the presidency usually requires a candidate with broad, charismatic appeal, and Vance doesn’t seem especially strong on that dimension. He’s also taken positions that could narrow his coalition: leaning toward including far-right elements in the party, aligning with a more restrictive approach to foreign policy, and emphasizing the U.S. as a Christian country in a way that may cost him swing voters.
It’s also notable that Newsom isn’t priced higher in Democratic nomination markets, since he has been positioning himself for a national run for some time and seems to have meaningful support with Democratic voters.
60% chance of being nominated sounds low.
Agreed, I'd place it at 75%.
I'd say it's more interesting that his odds are so low. That suggests that people expect there to be an actually competitive Republican primary.
This is a good point. The Republican primary doesn’t look like it will be very competitive indeed.
I expect it to be like the 2024 primary, in which there was one frontrunner who was clearly winning and a bunch of people trying to get a small polling percentage, but who were never really in the race.
That's 30% to be the winner of the 2028 general election. For the nomination, the odds are:
Vance: 55%
Rubio: 14%
Trump: 4%
DeSantis: 4%
Other (total): ~23%
In normal circumstances, Trump being on the list at all would be surprised because he's termed out, but Trump has been intentmittentlt making noises about trying to ignore term limits on some pretense or other.
Besides that, it looks like the market is pricing in a real possibility of a competitive primary but consider it far from certain.
I have trouble with the concept that the total probability mass across all categories including "other" is just 90%.
That was my error. I messed up the mental arithmetic of calculating other. Fixed.
Is Maduro expected to get a trial (with an actual jury?!) or is he going to be eventually be sent into exile somewhere?
Precedent: https://en.wikipedia.org/wiki/Manuel_Noriega#Prosecution_and_imprisonment
> In pre-trial proceedings, the government stated that Noriega had received $322,000 from the U.S. Army and the CIA. Noriega insisted that he had in fact been paid close to $10,000,000, and that he should be allowed to testify about the work he had done for the U.S. government. The district court held that information about the operations in which Noriega had played a part supposedly in return for payment from the U.S. was not relevant to his defense.
Getting a "trial" with an actual jury doesn't mean much. Some dialogue from Bridge of Spies:
[scene]
[Jim's boss] It was important to us - it's important to our country, Jim, that this man is seen as getting a fair shake. American justice will be on trial.
[...]
Jim, look at the situation: the man is publicly reviled.
[Jim] And I will be too!
Yes, in more ignorant quarters. But that's exactly why this has to be done. And capably done. It can't look like our justice system tosses people on the ash heap.
[scene]
[Jim] I just don't think that three weeks is going to do it here, We've got a massive amount of evidence--
[the judge] You want to postpone.
Six weeks.
Jim, is this serious?
Sir?
Is this serious?
Yes, yes, indeed it is. You can see in the filing that--
Jim, this man is a 𝘚𝘰𝘷𝘪𝘦𝘵 𝘴𝘱𝘺.
Allegedly--
Come on, counselor!
Your honor--
Of course, I salute you - we all salute you - for taking on a thankless task. This man has to have due process. But let's not kid each other. He'll receive a capable defense. And god willing, he'll be convicted. Come on, counselor. Let's not play games with this, not in my courtroom. We have a date, and we're going to trial.
[scene]
[Jim's boss] Jim, you did a great job. You fulfilled your mandate and then some. But the man is a spy. The verdict is correct and there's no reason to appeal it.
[Jim] There's ample procedural reason. We know the search is tainted, and fourth amendment issues will always weigh more heavily in an appellate forum. We've got a good shot.
What the goddamn hell are you talking about? We were supposed to show he had a 𝘤𝘢𝘱𝘢𝘣𝘭𝘦 𝘥𝘦𝘧𝘦𝘯𝘴𝘦, which we did. Why are you citing the goddamn constitution at me?
---
I remember seeing an opinion essay by Eric Posner arguing how important it was to try 9/11 conspirators in whatever was the most court-like forum that guaranteed a 100% chance of conviction. You've got to keep up appearances, after all.
I think there is a world of a difference between the level of courts operating at the height of the red scare to make sure that a commie spy is convicted and Maduro getting accused of illegal possession of machine guns and narcotics trafficking.
If Putin had pulled this stunt, then I am sure that he could just call the judge and tell him what verdict he would prefer, and will get exactly that verdict. By contrast, the federal courts are thankfully much more independent of Trump.
Maduro is not reviled by the American public to the same level as Bin Laden was. Sure, if the DoJ presents solid evidence entangling him with smuggling drugs into the US, they would very likely feel obliged to convict him, but if it turns out that there is no great evidence, then I don't think a court would have a problem with acquitting him.
Why is there so little recent data on early miscarriages (<20 or <22 wks depending on country) considering early miscarriages compose 98-99% of all miscarriages? Good data is collected on abortion, std's, etc. Why is there not a collective interest in this topic, particularly considering the potential effects of COVID and consequent health measures on reproductive health? The Scandinavian countries are the only countries I could find that collect data, but they haven't published any data on early miscarriages since early 2021. I have heard bold pronouncements both directions, particularly around the vaccines, but there is, as far as I can tell, an absolute absence of data with which I can address the question.
How would we collect data on early miscarriages? The modal early miscarriage occurs before the mother realizes she's pregnant. You'd need to institute frequent, regular pregnancy tests on everyone who could menstruate.
I can't think of any source of data on miscarriages, but there's plenty of data on stillbirths:
https://ourworldindata.org/grapher/stillbirth-rate?tab=line&country=FRA~GBR~DEU
Though stillbirth is defined in a way that doesn't have (much?) overlap with early miscarriages, so I'm not sure it's what your are looking for.
What kind of data are you looking for?
Miscarriage rate per all pregnancies. Even better would disaggregated rates among relevant categories such as gestational age, maternal vaccine status, and maternal characteristics such as race, BMI, residence, etc.
I asked GPT5 your question, and it located numerous articles. Looked to me like most were published 2023-2025. Below is its summary. which includes links to articles. In addition to summary it also listed 2 big-picture summary articles.
GPT’S SUMMARY
• Baseline miscarriage fraction for clinically recognized pregnancies is usually presented as ~10% (sometimes 15–25% depending on definition/source). ACOG+1 (https://www.acog.org/womens-health/faqs/early-pregnancy-loss?utm_source=chatgpt.com)
• COVID vaccination: best-available summary evidence (meta-analysis + large cohorts) shows no increased risk. PMC+1 (https://pmc.ncbi.nlm.nih.gov/articles/PMC10152171/)
• Maternal age: risk rises substantially after mid-30s; odds several-fold higher by 40+ vs early 30s. PMC+1 (https://pmc.ncbi.nlm.nih.gov/articles/PMC12625503/)
• BMI: overweight/obesity show modestly increased odds (order of ~1.4–1.5x in one large cohort). JAMA Network (https://jamanetwork.com/journals/jamanetworkopen/fullarticle/2823748)
• Race: evidence suggests higher risk for Black women in multiple settings, but estimates vary and interpretation depends on adjustment/modeling choices. Tommy's+2PMC+2 (https://www.tommys.org/baby-loss-support/miscarriage-information-and-support/about-miscarriage/risk-miscarriage-black-women)
BIG PICTURE SUMMARY ARTICLES
The risk of miscarriage following COVID-19 vaccination: a systematic review and meta-analysis
https://pmc.ncbi.nlm.nih.gov/articles/PMC10152171/
The prevalence of sporadic and recurrent pregnancy loss
https://www.sciencedirect.com/science/article/pii/S0015028223017193?utm_source=chatgpt.com
Don't a lot of early miscarriages occur before the pregnancy is even detected? That might make stats tricky.
I made a game for guessing where people are from based on their accents! Feel free to try it at accentguessr.io. Let me know if you have any requests for features or other areas for improvement
That's a neat use of that Archive! I assume that "Elicitation Paragraph" is intended as a kind of phonetic pangram?
Nice. I was entirely useless.
If effective altruism seeks to maximize wellbeing relative to resource input, it would seem that things that have a non-zero chance of producing infinite wellbeing swamp everything else in expected value and are worth arbitrarily large resource expenditure. From this you get “effective evangelization,” trying to save souls because an infinite life in heaven has infinite value, there’s a non-zero chance a given religion is true, and therefore saving souls should be maximized. However, the problem with this is that salvation usually relies on converting to and maintaining a lifelong belief, which is a non-trivial burden to expect out of people. This is where Pure Land Buddhism comes in. In Pure Land, faith is effective, but it does not require unwavering, perpetual devotion. It’s said in one of the Buddha Amitabha’s vows that people who sincerely recite “I take refuge in Amitabha Buddha” even as few as ten times will be reborn in his Pure Land. This is a land of perfect bliss where achieving Nirvana is easy. You will never be reborn in a lower realm again, your future is guaranteed to be bliss followed by enlightenment. So, if Pure Land Buddhism has a non-zero chance of being true, and if it has one of the lowest barriers to salvation out of all faith-based religions (saying “I take refuge in Amitabha Buddha” ten times), then, as effective evangelists, ought we not try to spread this practice as far as possible? It would seem to be an effective cause area to tell as many people the good news that they can forever escape the suffering of Samsara just by repeating a saying ten times. If taken seriously, those that want to maximize value should put at least some resources into Pure Land evangelization, because it offers the possibility of infinite reward for trivial commitment. So, what do you all think? Should we take out online ads and billboards telling people to repeat Amitabha's name, or some other way of turning money into mantra recitations?
Same objections as Pascal's Wager, the space of possible consequences is larger than the hypothetical.
The True God might get jealous and damn you for worshiping false idols, but not for living a good life without worshiping anything.
By calling out to random supernatural entities you might invite demons to snatch dying souls from our world when we'd otherwise not come to their attention.
Money spent on making people's mortal lives better might give them the slack to actually live better and more moral lives, and that might get more of them into a positive afterlife than spending the same money on random preaching.
Etc.
If you feel like there is actual evidence for any one specific religious tradition having a chance of being correct, then that may swamp those hypothetical concerns and justify you spending money on that specific tradition. But I think that's true whether or not you use a utilitarian framework.
If you don't see an real evidence for anything, though, I don't think this is a meaningful argument for speculating.
Isn't this just literally Pascal's Wager?
Answer is no; Bayesian reasoning doesn't handle infinities properly. We need a better way to express "there is zero evidence for this incredibly-unlikely theory" than assigning it a very low (but nonzero) P. Until we find that, we have to reject incredibly-unlikely theories with zero supporting evidence manually.
I mostly manage this by ruling out by default (barring very strong non-probabilistic evidence) any payoffs of more than 100x, and increasingly down-weighting payoffs as they increase (so a 90x payoff with say a 10% chance I downweight further than having a 10% chance, closer to 1%).
Also I have a very strong time discounting, to the point where I basically ignore all non-trivial predictions beyond about a 10 year horizon, regardless of their payoff/EV.
While this is unprincipled and crude, it does effectively stop me needing to worry about stupid speculative rapture/damnation scenarios like "what if AI makes us all immoral geniuses in constant hyper-bliss" in favour of, say, "can I pay my rent next month"
Contra grumboid, the issue isn't Bayes. The issue is EV. In contrast, Kelly's Criterion handles Pascalian scenarios gracefully by automatically putting an asymptotic upper bound on the size of the wager, based on P(x). E.g. if P(x) = 2% and the payoff is infinite, you should only bet 2% of your bankroll. For finite payoffs, the wager will be a fraction of 2%.
k = p - q/b, all the rest is commentary.
Even with only finite numbers, you run into problems. The expected positive and negative value grows superexponentially with K-complexity. It grows faster than any computable function. So if you're basing the probability of something purely on the complexity and even if you're ignoring infinities, expected utility diverges.
You could assign probability based on how good or bad it is, but why should we expect the world to care about our utility function? Different people would have different utility functions. Does that mean they have to have different prior probabilities?
Enough people have reported deathbed experiences of Amitabha that I don't think it's fair to say there's "zero supporting evidence" at all. It might be weak evidence by your epistemic standards, but it's evidence nonetheless. Given this, a non-zero probability seems proper.
If something has been vetted, we trust it more. But if something has been doctored, we trust it less. This suggests that MDs seem more nefarious than DVMs.
> doctored
Is this linguistically related to "MEDle with sth."?
Your question is not clear. On the surface, there is definitely no connection between "MD" (𝗠edicinae 𝗗octor, "'doctor' of medicine") and "meddle".
There also isn't a connection between "𝗺𝗲𝗱icine" and "𝗺𝗲𝗱dle". Wiktionary has meddle << Latin misceo ("mix") << PIE *meyḱ- ("mix"); medicine << Latin medicus ("doctor") < Latin medeor ("heal [someone]") << PIE *med- ("heal").
Note that while "doctor" is a Latin word, it isn't the Latin word for doctor; in Latin it means "teacher". This makes sense if you view a degree as a credential for professorship. A doctor of medicine is someone who teaches medicine to others.
If you were asking about some other potential linguistic relationship, you'll need to be more explicit.
I looked it up, and kinda maybe. There are several proposed etymologies for the verb floating around, but one of the more popular ones suggests that it comes from use of "to doctor" with a straightforward meaning of "to perform medical treatment upon". Possibly by way of analogy between "doctoring" a drink by adulterating it or watering it down and the process of preparing a medication for a patient. Or maybe just a general analogy of medical treatment as an invasive alteration.
Two others caught my eye. One was from 15th century use of "doctor" as a noun to refer to loaded dice (not sure where that term comes from). The other is independent derivation from the Latin root doceo, which originally meant to teach (doctor acquired the modern meaning due to the overlap between teachers and scholars) or to direct a play or other performance. Doceo is closely related to decet (to adorn, to give the proper appearance), which is in turn related to the root of the word "decorate". "Deceit" sounds like it should also be related, but that appears to be a misleading coincidence.
You really going to make me pull Volume IV of the OED off the shelf and find my higher powered reading glasses?
It's either that or I check the OED next time I'm at the library, which probably won't be until the weekend.
I offered to donate my set to my local library but they already had two complete sets.
I was thinking I’d bite the bullet and pay a monthly subscription fee for the unabridged online but just found this:
https://libwww.freelibrary.org/blog/post/5508
According to Scott’s most recent post every comment here stands the chance to become a cherished item in the reliquary of an EA space monastery (they will have turned it into an NFT).
Please write accordingly
we're probably all currently being simulated because we are post-eschaton famous. not sure if I'm in the good or the bad kind yet
The only reason to reread the internet is as a form of torture humanity receives in dystopian outcomes.
6, 7 🤷
25 or 6 to 4, take it or leave it.
I read They're Made Out Of Meat back when they were still made out of meat.
I think humans should eat out of a spout on top of our head and eject waste from our toes: feces out of the big eight, pee from the pinkies. Placing the entry and exit points for food at the extreme distal sections of the body maximizes the surface area available for nutrient absorption and dispersal, while simultaneously allowing for more intuitive hygiene, since feces is ejected directly adjacent to the ground, which people generally treat as unclean anyway. Awkward crouching is no longer necessary when dropping a deuce, you can just leave a little trail of nuggets behind while walking. There's also a higher potential for personalized musical expression when farting, since each toe would naturally produce a different pitch when utilized.
A separate digestive tract also allows for much more specialization in the respiratory and reproductive systems, removing the perpetual need to switch between systems situationally to avoid contamination or strangulation.
Plus, I really want to learn how to juggle dumplings, then land them in a stack on top of my head, then open up my spout and make them all disappear in one big gulp.
Human anatomy was designed by a terrible civil engineer - running a toxic disposal line through a recreational area.
That would complicate designing shoes
We woukd also need to revise "I'm going to kick your ass!"
So, we would need to build toelets?
Hehe, toelets is a great word, they sound adorable.
Thanks for the mental picture! Appreciated 😀