You know the thing where you're falling asleep and right before you transition from awake to dreams starting to flicker, you have that epiphany that has been evading you for the last year or so?
It just comes up and says "*HERE* it is" and everything clicks and that thing and that other thing and that other other thing suddenly snap into place and now you know how to think about something when, before, you kept grabbing at the wrong thing?
I'm not a Baptist at all, I was just curious if that was a legitimate alternative spelling. I've never encountered it before, it was unlikely to be a typo as it happened at least twice, and it seemed unlikely to be a misspelling because your writing is that of a literate person. At the same time, a Google search for "babtist" just returns search results for "baptist" churches.
(But you're supposed to have the Bible Study Group Leader be the DM and we didn't do that so my experience probably shouldn't be seen as representative.)
I broke 2 fingers in my left hand last night. Yeah another bar fight. Just kidding. Hopefully those days are behind me. Actually I just got off my mountain bike while it was still moving again. Pretty sure I’ll still be trail riding for a long while but I’m gonna try to reduce the involuntary ejections.
At Urgent Care my fingers were so swollen that they thought they would have to cut my wedding band off but they had a couple tricks to try before they went that route. Put a tourniquet on my bicep and wrapped my ring finger tightly with tape for a half hour to squeeze the blood out of the finger. Then the doc looped a piece of strong fabric through the ring and kneeled on the exam table to put his back into pulling the sumbitch off. Took a couple minutes but it came off after a little wrestling. Now I don’t have to have a jeweler weld my band back together.
As a bonus, with a plaster splint on my hand and forearm and the arm in a sling Mrs Slothrop is outside mowing the lawn now and I’m just laying on the couch tapping out this comment with my right thumb.
Edit
Oh yeah, a little codeine makes me a bit giddy and I tend to run off at the mouth or I guess at the thumb under the influence.
I ever mention my dog, Buck here? Golden retriever, great dog. One time.. Oh, I’m doing it again. Turning my phone off now.
I am ambivalent about reading stories or watching film series that are not finished yet.
On one hand, I understand why publishing chapter by chapter is great for the authors. You can get feedback and maybe even some money on the way. The feedback tells you whether to keep going or drop the project. The more money you get, the more you can arrange your life about writing more and worrying about other sources of income less. This is ultimately also good for the readers, because they get stories they otherwise would get years later, or not at all.
On the other hand, I hate reading one chapter a week or more, or whatever the publishing schedule is. I am a fast reader. If I have free time, I want a ton of text now. I will stop when I run out of free time, or when I finish the story (or if I decide it is not worth reading). Afterwards, reading one chapter a week or one chapter a month feels like torture -- I wish I could forget about the entire story, and then remember it again a few years later, after another substantial amount of text was written. But then, after a few years, I will have to re-read the introduction again, because I will likely forget it. So my wish is to only find out about the story when it is complete.
I guess if I put these together, my preferred outcome is that other people provide encouragement and support to the authors, and I only find out about the story after it is completed.
I hate that some stories never get finished. A Song of Ice and Fire is "forthcoming" since 2011. The War Against the Chtorr is "forthcoming" since 1992. At least be honest, guys; you are not going to do this, ever; you only wait for your death as an excuse.
Another problem is that paying as you write creates an incentive to keep writing and never finish, even if you already have nothing more to say. I have seen stories where Book 1 is full of new exciting things and problems, and a longer Book 2 is just the protagonists walking around for no reason other than to generate even more text.
With film series, I think the pressure to enshittify is even stronger. At long as the story remains popular, it remains open-ended because there is yet another season to make. When it stops being popular, it is dropped in the middle of a word. Thus it is almost guaranteed that the story will never have a satisfying ending -- there is always either an incentive to keep dragging the audience, or no incentive to give a fuck anymore. This can even happen multiple times (Babylon 5, Heroes), a series can be dragged when it seems profitable, canceled when if seems not, restarted when they realize they have underestimated its popularity, then canceled again, and in no point in this process the viewer gets an ending to the story.
A smaller version of this is simple computer games. Before internet, you got a few levels, and then an end. If you liked the game, you recommended it to your friends, so there was an incentive to make it good, and to give it a nice ending. Online games, it was just how many levels you keep playing so that they can keep showing you ads. There is no incentive for an online game to have an end, that's just leaving money on the table. You do not leave an online game because you have completed it to a satisfying ending; you leave because you got bored.
Now of course not everything is like this. There are still books, film series, and computer games, that are planned since start to do something and then have an ending that makes the customer happy with the overall experience. But the temptation to keep milking the customer and then leave them frustrated exists, and it makes me suspicious that whenever I start reading a book or watching a film series that is not finished yet, I am volunteering for a frustrating outcome.
Related: Larry Correia argues that GRRM's very conspicuous failure to conclude ASoIaF has effectively killed epic fantasy as a genre for new writers, with only established veterans like him and Brandon Sanderson being able to make money on new epic fantasy series. He also suggests that the rise of LitRPG and Romantasy have been driven by readers and writers fleeing to fantasy-adjacent genres farther from the stink of GRRM.
For me, it's fine when each book (or movie) in the series is a reasonable story in its own right, like e.g. early Harry Dresden, early Honor Harrington, or most of the Vorkosigan Saga. The obvious problem, given the "early" qualifiers there, is that those can turn into series that won't be satisfactory until complete. Weber did eventually finish the core HH series, and I had a very palpable sense of relief. "Finally, I never have to read anything by David Weber ever again!". Fingers still crossed on Butcher.
I'm also generally willing to trust an author who proposes to write A Trilogy. I've been burned a few times by getting only two of the three books, but that's rare, and when there's a fourth book it's usually a nice-to-have add-on rather than a mandatory extension if you want to see the core story end.
My current concern is webcomics, which are necessarily a serial format, and I'm currently following four for which it's pretty clear that the author has decided this is the best gig they will ever have so they'd better milk it for all it's worth. Though both "Schlock Mercenary" and "Skin Horse" did come to actual conclusions after very extended runs, so there is hope.
Thirty years ago, there was a Czech publisher (I forgot the name) who killed a few translated book series in the middle. Like, he translated parts 1 to 5, decided that the part 5 didn't sell as well as he expected, so he never translated parts 6 and 7. What was worse, if I understand it correctly, translators typically buy exclusive translation rights for the entire package, for some time limit such as 10 years. So when he decided to kill the series in the middle, no one else was legally allowed to continue where he stopped. That has pissed off many fans, and it became common knowledge among the hardcode fans that "if it's published by X, do not bother buying the books, because the series will never finish".
Speaking of webcomics, there is a comic version of "Worth the Candle", misleadingly labeled "Completed" https://www.webtoons.com/en/action/worth-the-candle/list?title_no=7061 but actually canceled at the moment the party pays a visit to the soul mage (about 1/4 of the story?). Wrong labeling adds insult to the injury. Now I can't even trust things labeled "completed" to actually be completed, wtf!
My favorite writer is currently Dave Duncan https://en.wikipedia.org/wiki/Dave_Duncan_(writer) ; I am reading his books in the order on the Wikipedia page. "The Seventh Sword" is a trilogy (finished), with one bonus book in the same world. "A Man of His Word" is a tetralogy (finished), followed by another tetralogy (finished) in the same world one generation later.
As a Discovery writer, I shudder at the idea of publishing something a chapter at a time. It means you can't rewrite the chapters when a better idea comes along.
Part of the reason Dune is so good is, any of the first four books will serve as an end to the series. (5 doesn't, but by 5 the setting was also getting pretty long in the tooth.)
So I'm into book 4 of the "Dungeon Crawler Carl", series, Matt Dinniman. (thanks again to my daughter, who's in book 5.) It's all over the NYT best seller list. And yet it's not even nominated for a Hugo in the US. Which causes me to lose a little faith in Hugo's. (And the writing and story telling continue to improve at least into book 4. Recommended.)
Be warned that the writing starts to gradually drop in quality starting somewhere around book 5, though I think the characters, setting, and story generally continue to be compelling. I started to see more proofreading errors, repetitive phrasing, etc around that point. I'm inclined to forgive those minor sins-- he is writing these books very quickly and I find the whole project pretty compelling despite its flaws.
Hmm well book four is 578 pages. My daughter's book 5 is ~700 pages. Not sure about the others but it's easy to check. I'm someone who played a lot of RPG games, mostly on the computer but also D&D with a group.
Hugo nominations are done by Worldcon members. IME the average age tends towards the older end compared to other events I attend, and this may affect how quickly reading preferences shift compared to the general population.
Be the change you want to see! - join up for a Worldcon and nominate whatever you like! 2027 will be in Montreal, which I've never been to; good excuse to visit, so I'll be going.
I won't be nominating DCC for any categories, though. I did enjoy it, though I've not picked up the latest book yet; but you only get to nominate so many things, and although I've not yet looked at what will and won't be eligible for the various 2027 categories and likely won't for another 6 months or so yet, I've definitely read better LitRPG than that.
I do look forward to a non-traditionally published web serial getting a Hugo. I think it's only a matter of time - Worldcon membership tastes aren't isolated from general fandom shifts, they just have a bit of hysteresis.
Thanks. I don't need to join. I don't care that much what is nominated. I just want more 'good' book recommendations. And in that regard, LitRPG do you find better than DCC? Again Thx.
Tastes do vary, of course; but fwiw I am currently enjoying Super Supportive, Runebound Professor and Elydes enough to support them on Patreon. Other things I've recently enjoyed have been Sokaiseva, Live at the Forum (closest of everything thematically to DCC, don't know if I'd actually count it as /better/ so far but I am only a little way through the backlog), Dalliance Rather,
My favourites of all time are probably Super Supportive, Demesne, Mother of Learning.
When I asked AI why the books hadn't been nominated the first answer was that LitRPG wasn't 'high brow' enough for the sci-fi community. And I thought, when did sci-fi become high brow? I mean they nominated Harry Potter.
I sent this directly to Scott, but I'm posting it here as well:
The office of management and budget is seeking public comment before they change the rules on grantmaking. I learned about this from someone close to me who works for the dept of Education. Admittedly, I'm not an expert in any of this, but my understanding is that the way it currently works, the Department of Education is constantly running grant reviews for specific research and funding (they used to run a lot more, but DOGE basically forced a multi-year moratorium from which they are only just now recovering). People submit their applications for federal funding to grants that fit their general field.
During an actual grant contest, the subdepartment picks a panel of academics from volunteers who they believe are qualified to evaluate the grant proposals in question, and vets them for any conflicts of interest. PART OF THIS PROCESS IS CHECKING FOR PARTISAN BIAS ON BEHALF OF THE ACADEMICS, I emphasize that part because the narrative that the federal bureaucracy is captured by woke elites is one they are well aware of, having been sued for bias 10,000 times by conservative lawyers, and having lived through multiple transitions between different administrations whose first step is to scrutinize every funding authorized during the prior one. As you might imagine, in the last two years they have been particularly sensitive about this. I've overheard more conversations about ensuring that ideologues don't taint the neutral process of funding that I thought I would hear in my entire life, by people who have clearly been thinking about this for well over a decade.
Anyway, now the OMB wants to change the federal regulations around the law, so that after the academics have been chosen, after they have weighed in with comments about what projects would be of immediate benefit and what ones need specific tweaks to their methodology, the Trump administration would like to add a step where a political appointee gets the final say. You can read the language of the proposal yourself, and evaluate it yourself, but I think you will come to the conclusion that the obvious intent is to make sure that otherwise good research that might run counter to the ideological aims of the current administration die on the vine.
>the obvious intent is to make sure that otherwise good research that might run counter to the ideological aims of the current administration die on the vine.
Its interesting seeing the different intuitions people have wrt this... I would not at all be surprised if both parties had a veto in my countries version of this process, and I wouldnt see it as a problem.
The political appointees in charge of the large government bodies in the United States are always chosen by the current president. That's why traditionally there are safeguards in place to make sure that the majority of the work is done by neutral parties and professional civil servants. Under this new plan, the people Trump picked would have immediate fiat over all funding. There are ideological concerns and corruption concerns.
Problem is, many of us would not be surprised at all if Trump administration nominated an antivaxer or a flat earther to this position. A position of being able to veto scientific research all over the country, if I understand it correctly.
That said, this is democracy, and the will of the American people is to burn their country down (to spite the libs or something like that), so who am I to comment on that.
I'll summarise with the post's own summing up: "Since World War II, the United States built the world’s preeminent scientific enterprise on a straightforward principle: federal dollars should fund the best science, as determined by independent experts rather than politicians. Peer review, open competition, and institutional autonomy were the pillars of that system. This proposed rule dismantles all three, simultaneously, government-wide, and binding on every federal agency by October 1, 2026."
(Please do let me know if this is a different set of changes or somehow not applicable to your post; apologies if so).
I have recently become aware of "AI ick", which I understand to be a deep, visceral aversion to AI usage or even artifacts produced by AI.
Example: https://lucumr.pocoo.org/2026/6/6/communities-of-not/ the footnote leads to a GitHub issue on an extremely popular open source tool that has merged some AI generated code and triggered an aggressive reaction from some users.
I understand weariness about a new technology like AI. But I'm at a loss when it comes to such strong reactions. Is this a personal thing? Or is this simulacrum levels 2-4, where there's a deeper, tribal alignment thing going on?
I'm seeing a lot of comments on this thread trying to claim that this is about tribes and making analogies to old DEI culture. Those comments seem to me to be deeply confused.
I get very strong ick for any unlabelled AI that is directly consumed (the relevant difference for me is whether or not someone is directly taking an AI output and showing it to someone, vs using it as e.g. research aide). I don't consider myself to be among the anticapitalist redditors, or even anti tech (I work at a tech company!). Neither do most of the people I talk to who get AI ick.
The only real analogy between this and the DEI debates around the 2020s seems to be the use of public shaming. Tribal alignment is definitely the wrong way to look at it, at least for now.
The github example seems to be people complaining about AI introducing a bug, not about AI usage in general. So it’s not a good example for what you’re talking about.
On social media there's a thing I've seen a few times, where one person says "here's a thing I made, do you like it?" and person replies "you used AI, therefore it is awful, I hate you and I hate it and I will tear you down as much as I possibly can."
This happens even for high-effort things that used small amounts of AI.
I think it's the standard social-justice-warriors problem in a new field: there are some people who really like having an excuse to be mean to people.
"Those who disagree with me are bad people who have no good reason for their actions" is a supremely uncharitable explanation. This doesn't mean that it's wrong (in fact, I think it's true for at least some people), but one should use it only as a last resort if one wishes to find truth instead of merely satisfying tribal urges.
Asking from inside that frame: Why did all those sadistic people coordinate on hating AI usage? Why not, for exmaple, apostrophe usage? However you answer that question might provide an answer outside of the frame too! Maybe even the people who are not sadistic have valid reasons for disliking some AI usage.
In the initial post, Matto mentions "weariness about new technology". I'm about 90% confident that this is a typo or malapropism for "wariness" (apprehension), but I think "weariness" (fatigue) is actually a big part of what's going on here.
It's a similar dynamic to pop culture fads. Some musical genre or media franchise or whatever seems to be everywhere you look for a while, which if you are an enjoyer of the thing is all well and good, but if you dislike it for whatever reason (or even if you're merely indifferent to it at first) then repeated exposure to it becomes more and more annoying and the dislikers gradually become outspoken haters. In the bit Matto's original post referenced, the "Oh for fuck's sakes" reaction is pretty much the same energy I would expect if somebody had checked in a bunch of gratuitous Twilight or Nickelback references to an open-source codebase and broken some stuff in the process.
The most recent example I've encountered: https://x.com/hsrmemes_/status/2069116085530419403 is a fairly-high-effort thing that I actually quite liked, which, yes, was definitely made with AI. The comments section has some raw praise ("Castorice was sick as fuck," "I absolutely loved the Phainon one"), quite a bit of nitpicking ("Aglaea doesn't wear gloves..."), and a few people being flat-out mean, eg https://x.com/KuritoPoroto/status/2069136248740946405 "Your AI slop bores me".
I think the people in that last category are saying: "AI is bad for whatever reason, therefore I should attempt to effect social change by being mean to people who interact with AI in any way, and if I'm maximally mean then it will effect the maximal amount of social change." I also, yes, think these are people who enjoy being mean to other people.
I think they coordinated on AI usage because it's becoming a culture war front where choosing a side gives you air cover. You can do awful things to people, claim it's to "fight the ai invasion" and get away with things you otherwise wouldn't.
We're talking about a tiny, tiny minority here, that nevertheless inflicts vast damage to social fabrics--theyre Pareto optimal probably!
I only bring this up because the anti-AI crowd I've been exposed to also appears to lean anti-capitalis in the same way many of the DEI folks used to, the folks I've seen first hand apply this tactic in two tech companies I've worked at.
Again, it's a tiny minority, but part of the pattern is that the broader group they're part of doesn't do any policing on they're obviously bad faith actions.
"I think it's the standard social-justice-warriors problem in a new field: there are some people who really like having an excuse to be mean to people."
I think that's part of it, but not nearly all of it. I think there are a lot of people who just developed a large amount of negative affect around the whole concept of AI[1] and knowing something includes AI really does just ruin their enjoyment of it and make them see red.
And I honestly get this reaction, to some degree, even though I think it's overblown. Because it's not just "weariness over a new technology" like Matto suggests. The AI explosion managed to hit all the pre-existing beats ambivalence towards capitalist consumer culture and marketing/tech sector hype and frustration at continuous enshittification and economic/political anxiety all rolled into a package that is very often unbearably cringe.
See, modern machine learning tools are getting better and becoming more useful, but there was a HUGE push to incorporate them into absolutely everything even when they were obviously unready, and an extremely poorly-concealed eagerness by various corporate CEOs to replace their expensive, messy human workforce with cheap, compliant machines and also quite a lot of average Joe humans eagerly churning out and spreading and talking up AI slop[2]. And if you try to put yourself in the shoes of someone who didn't find any of the recent AI developments especially interesting, and didn't ever seek out AI tools or AI content by choice, those mostly highly negative things are their main experience of AI. So it's small wonder they've built up so much negative affect around the concept.
But yes, also some people really just like having an excuse to be mean to people, (though that's hardly exclusive to social justice warrior types).
[1] More specifically, generative AI, like LLMs and AI art programs.
[2] And while I definitely don't think every use of AI for art or writing can be dismissed as "slop," a large fraction of the people most eager to use them for that are people who lack talent for it themselves. Lacking talent tends to correlate (though not perfectly) with lacking taste. So a lot of the people sharing AI art are consistently and very enthusiastic sharing extremely *bad* AI art, while by all appearances thinking it's amazing. I'm reminded
Which specific strong reactions are you asking about? The open source tool introduced a bug, which is important context.
Both the linked threads at the bottom of that blog post (GitHub and Mastodon) have both reasonable takes:
> Regardless of AI as well, the project should really take care before pushing new features, and especially bugfixes (these issues only started cropping up on the last two patch releases, which shouldn't change existing functionality)
> Imo at least, producing slop is a human error caused by misunderstanding what LLMs are and what they can actually do
And complete garbage:
> BTW don't forget OpenAI assassinated a guy for wanting to testify against them. Package sabotage by corpos is not a conspiracy at this point lol (I am a developer, in case someone wants to say im just a random idiot but this is my anon issue account)
Specific instances will have different motivations. I think modernity has a distinctive epistemic failure mode where people want to talk about
*gestures around vaguely* you know what I mean! All the stuff about the thing!
Then disconnect from specific reality to shadowbox against phantoms.
On the surface level, posting a screenshot of a social media post with an aggressive tone is low quality content for the medium (a project issue tracker). My own understanding is that gesturing at AI and ending the post with "for fucks sake" is a bit of a dog whistle for mobbing a project/author. Both seem to me as fairly strong, aggressive actions to take against an oss project.
Perhaps this feels so antagonizing to me because I work with software where updates are usually handled in a rolling way or at least go through manual validation in a staging environment. So the original screenshotted poster seems just as responsible for breaking their system as the LLM that introduced the bugs.
Breaking changes are a fact of life, even more so now with supply-chain attacks. The amount of vitriol projected at the rsync project because of a LLM-made bug seems disproportionately large given how software world works.
Yeah, I agree that that issue is very low quality.
To me, the Mastodon comment screenshot was normal social media stuff. On about the same level as things like "somebody cut me off in traffic today, I hope they get into an accident haha". I don't love that it's normalized, but I consider it a pathology of social media more than of anti-LLM people.
Obviously it's stupid and pointless to copy that into a GitHub issue, but to me it seems like an unaware individual treating GitHub like Reddit or Twitter (where it's an accepted norm to just repost screencaps you agree with).
IMO the reason why it seems excessive is because two small but common bad-nesses combined in an unfortunate way. The rhetorical temperature of the Mastodon post (ex: "for fucks sake") wasn't intended for a GitHub comment, and is par for the course where it was posted. I think it's unlikely it was an intentional call to mob the dev. Mindlessly reposting screenshots is unusual on GitHub specifically, but is ubiquitous in more social spaces. It looks like the person who posted the issue mostly contributes to video game mods, and this is the first issue they opened.
I kind of despise that it's "normal" to throw around decontextualized, vaguely articulated, high temperature statements. Nobody feels the need to "own" what they spread. But that's a problem in lots of places, and not just anti AI circles.
That's a good take. I've been off GH for a longer time now, and off most social media, so it helps calibrate against what is the baseline behavior to expect.
1. AI-generated output tends to converge on a few specific styles, most of which I find grating to begin with and their association with the other things that annoy me about AI makes it worse.
2. I've been dealing with a big top-down push for using AI tooling at work. The tools have some usefulness, but their output is decidedly worse than human-written code and is a lot more work to review and debug, and use of LLM tools to analyze bugs seems to fairly often lead people into dangerous misunderstandings of what's going on.
3. More generally, I have a fairly low assessment of the quality and reliability of AI output. I generally do not consume it by choice. Presenting it to me anyway gets tiresome.
I understand your 1st point as it aligns with my own experience. It's a bit like interacting with automated or even human call center phone calls. Exasperating on a good day.
I am surprised by your 3rd point. I've found AI to be a meaningful improvement in my work. Not 10x but somewhere between 1.1-1.4x depending on task/domain. I use it both for getting an overview of a new to me part of the codebase as well as for generating code. I do always dona substantial review on both fronts, but I find the speed, in this case, to come from getting things almost right within the context of the codebase. We also have a ton of test across the whole pyramid which helps catch small mistakes and being the quality bar higher I suppos.
This is fascinating because I wonder where the difference in perceived quality comes from.
I suspect it's related to how well-trodden the problem domain and codebase architecture are in the model's training set. I've found the tools to be at their best producing boilerplate and at their worst trying to work with complex sui generis issues. In boilerplate-heavy architectures, AI tooling is probably very useful. Ours is the opposite end of the spectrum, partly because we've architected it to minimize boilerplate and partly because the code base has evolved a lot of clever bits over the years for a variety of reasons, some good and some bad.
Where it goes badly off the rails tends to be in trying to work with the clever bits. I've seen people get in a lot of trouble trying to use Claude or Cursor to understand new-to-them parts of the codebase and it makes a fundamental mistake in one of the load-bearing assumptions of the analysis. These mistakes tend to be subtle enough and presented confidently enough that if you need the explanation, then you aren't in a position to notice the error until you try to act on them and things don't work the way you expected. I notice them all the time because I've an expert on several of the clever bits and get tagged in on discussions when people notice that things have gone badly off the rails, or because I'm troubleshooting some stability or perf issue downstream of the error.
This is similar to my own experience. AI is very useful on tasks I describe as "advanced copy-and-paste", which includes boilerplate code, but also things like "do X, following the pattern we used for Y". But when dealing with complex problems, it constantly makes incorrect assumptions. Usually they are plausible, and junior devs are happy to trust it and make a mess of things. E.g., summarized from a recent conversation.
> Me: Why are we adding 2000 lines of code to split files into pieces?
> Junior Dev: Claude says we ran out of memory after reading a file, so we need to read it in smaller chunks.
> Me: What file?
> Junior Dev: This one: <link to file>
> Me: And how big is that file?
> Junior Dev: 4 kb
> Me: So do you really think that's what caused us to run out of memory?
> Junior Dev: I guess not
And then sometimes it's not plausible *at all*, and people _still_ take it at face value. Another conversation:
> CTO (who is leading an aggressive, top-down push for AI tools, to the point of mandating that we use them for all work and not hand-write any code): My AI agent found the issue that caused our increased error rate last week. Please have the team make the appropriate fixes.
"AI Triage Summary: An increase in the concurrency from 5 to 10 led to increased load on AWS S3, which produced increased error rates. The solution is to reduce the load on S3."
> Me: I'm pretty sure that we didn't bring down AWS because we started hitting S3 another 5 times per minute.
I've seen plenty of junior developers with that sort of pattern entirely without AI - just assume their first theory absolutely is the actual cause of whatever problem they're investigating and spend hours or days running with that down a rabbithole instead of just doing a quick set of tests to systematically pin things down and prove/disprove theories first.
I find that Claude becomes much more useful for debugging if you get into a habit of interrupting and telling it "before proceeding any further, please write and run a quick test to prove this theory" whenever it confidently states the cause of X is definitely Y.
Sadly, no amount of CLAUDE.md, memory etc seems capable of creating a behaviour pattern where it does this automatically /before/ boldly asserting things to me.
Meanwhile, I remain utterly unable to convince a subset of my circle to consider AI output critically, even when they keep getting burned by it. I can only sigh at time wasted.
TLDR: treat Claude like you would a lazy junior developer with a tendency to pattern-match on long rabbitholes.
But that's sort of the issue, at least for me. People pushing AI are saying things like (to quote our CTO) "AI coding tools will give us infinite capacity – if they don't, then we aren't using them correctly." But if using them correctly requires constantly monitoring them and correcting them when they start to go wrong, then they aren't going to produce anything like the improvement people keep insisting they must. In other words, nobody would ever say "If we hire an innumerable army of lazy junior developers, we'll have infinite capacity".
I had a similar experience yesterday. My current hobby project is to analyze old DOS games, reverse engineer the code, extract resources and passwords, that kind of stuff.
Here, AI is very useful. It sees something like "mov ax, 40", "mov bx, 2", "int 21" (possibly separated by a few lines) and immediately says "hey, this prints an error message" while I would have to consult my notes. Also it recognizes many kinds of patterns instantly, like "this is so-and-so method of encryption", "this draws a 16-color image in 4 bit planes" etc. But most impressively, at the beginning it gets a ton of freshly disassembled code, and it already sees large parts of the big picture, which parts of the program are related to what. And these instant guesses are like 90% correct, which sometimes saves days of work.
But when it makes a wrong guess, there is almost no way to convince it otherwise. I just take the original code, add a note saying "this part does X", close the session and start a new one. And the new instance is happy to exclaim that the part obviously does X and continue from that premise.
So, yesterday there was a game that used some kind of copyright protection, checking for a damaged piece of floppy drive or something. We got past that point, and the AI was like "this is another copy protection check", "and another", "wow, this game has so many sophisticated protections". Except it did not, we were long past the check, and now the game was loading resources. But the AI couldn't be convinced otherwise. So we checked the rest of the code (a few hours of work) and then I was like "damn, so where does the game load the resources?" and the AI kept insisting that it does not. -- New session, I didn't mention copy protection at all, just asked "where does the game load resources", and of course it showed me the part of the program that we previously skipped, because it was "just more copy protection".
The problem is that even if AI is right in 90% cases in some subject, it always expresses 100% certainty, and prints pages of convincing arguments why it is so. So you must accept that sometimes it is a lottery; on Monday, AI does in 5 minutes what would take you entire day; on Tuesday, you waste the entire day being misled by an AI hallucination. If you are lucky, you still save some time on average, and you may actually enjoy the process, especially if you start developing an intuition for when the AI is hallucinating.
I've been getting Claude to admit when it doesn't know stuff. Mostly by asking for sources and evidence (and what evidence we would need to falsify an assertion; both assertions I made and assertions the AI made.)
Fascinating! The system I work on is bespoke (in-house telemetry system), so there's little boilerplate and lots of custom libraries spread across a handful of microservices combined with a very bespoke orchestration system. I do feel like I optimize a lot for clearly describing system boundaries and relationships between components, along with strict constraints because of this. In a way, it's all about managing memory buffers, so there's less cleverness and more down to earth "this is the way if we're optimizing for GC pressure/allocations". That, and as I mentioned, layers and layers of tests + gradual roll outs + stage/prod. Maybe that's what helps keep claude on track--because whatever is says confidently gets immediately tested against reality.
AI is transforming the world in ways that a lot of people don't like. Those people aren't just stupid or virtue signalling, they have legitimate grievances, and legitimate fears.
You may say 'sure AI is raising a generation of children that can't read or think, and no one can buy a graphics card for their home computer, and huge numbers of people are getting fired and replaced with AI systems that don't work, and people are getting AI girlfriends instead o ever talking to a real person and being driven to suicide by them, but surely there's nothing wrong with adding some AI code to this repository so long as it works right! How irrational to reject the good parts just because bad parts exist!'
But for people who are really upset about this, it's sort of like a hostile invasion. Sure, there's legitimately no harm in helping the occupying army that killed half your population rebuild a bridge they bombed, that's even useful to your community in the long run... but it still makes you a collaborator.
I don't think they're stupid, but their harsh reaction surprised me because there's a gap in believing how the world is being changed by ai and know how the world is being changed by ai. This is a difficult question and I doubt one can have a 100% certain way of knowing how things are and how things will be. I guess my discomfort with their reaction is that they seem extremely, to me, certain about the state of the world, and sure of themselves enough to act aggressively based on the certainty of that belief.
My original question is more if there's something bigger I'm missing, because there sure seems to be a lot of strong feelings here.
Part of the issue is the size and speed of the change.
People at least as credible as Scott are predicting permanently world-altering advances to AI as soon as next year, far beyond what's already happened. About 80% of the teachers I've talked to are convinced an entire generation of children is growing up fundamentally incapable of independent complex thought.
Setting aside whether they're right, uncertainty doesn't indicate inaction. The larger and more immediate a potential threat is, the more violently you should act against it immediately even if you are not 100% certain about it.
To put it another way - domain experts have been warning about immediate major consequences of AI ranging from the end of human existence to permanent feudal dystopia. The reactions you are talking about are not extreme for someone who actually believed something like that.
Whether or not all the people in question believe those specific extreme scenarios, they are beginning to believe things in that direction of magnitude, and are reacting appropriately.
The expulsion of that person (who never signed up to be a community member) is entirely imaginary but the punishment that the community unleashes is not: people pile on and shame them, quote them out of context and turn their weakest moments into proof that the person was always unserious, a sharlatan or should not be listened to.
|
Charlatan*. I take it you've never seen the Monty Python "burn the witch" scene*? In-group loyalty can become ritualistic, to be sure, in positive or exclusionary forms. While this bonding seems to exploit lower level psychological structures, on the flip side, maybe *some* amount of reinforcement is necessary to maintain group cohesion, idk.
> Or is this simulacrum levels 2-4, where there's a deeper, tribal alignment thing going on?
<taps nose>
It's 100% signaling. Soon to lead to humorous AI witchhunts and accusations against "real" people who have an AI stylistic quirk like em-dashes or "it's not x, it's Y" turns of phrase.
I don't think that's true. Here are 2 examples from my personal experience.
I recently had a big surgery and then spent several days in a very unpleasant oxycodone haze, dreaming creepy waking dreams. I escaped oxy hell by watching a TV station that showed nothing but nature scenes, mostly big dramatic ones with huge landscape features in the background and something small, beautiful or cute in the foreground. Oxycodone me enjoyed those scenes quite a lot at the beginning, and actually cried with pleasure at some. But over the course of several days I developed an aversion to the scenes. It started with a *feeling* they were fake, and the fakeness was the repulsive element. And then as the oxy dose went down I became able to think about the scenes, and realized that of course these scenes were AI generated. There were lots of tells. The vegetation in the foreground never had any dead leaves or other imperfections. The big landscape backgrounds repeated, with varying small foreground elements added. My revulsion really did feel intuitive and visceral. It's akin to the way I react to things sweetened with stevia. They really do taste sweet, but there's something *wrong* with the sweetness. It's just not the same as sugar sweetness, and the way it's off is intrinsically repellent.
Also: I can't get GPT to make images I don't hate. Everything it produces is infected with spiffiness, some kind of built-in eye magnetism,. Even when I choose subjects that seem tedious and raunchy, even when I specify that the image look like one drawn by a bad amateur artist, I can't purge the image maker of the underlying commercial art "like me! look at me!" algorithm. The images I get disappoint and creep me out enough to make me curse and recoil.
As for signaling -- naw, I actually have never told anyone about either of these experiences. They're not interesting enough to bother bringing up.
PS: In case you're interested in the AI image quality thing: Here are 9 images I got Dall-e 2 to make which I like quite a lot. I think they're genuinely interesting, and the creepy ones creepy in an interesting way. And they seem to me free of the commercial art stink.
> My revulsion really did feel intuitive and visceral. It's akin to the way I react to things sweetened with stevia. They really do taste sweet, but there's something *wrong* with the sweetness. It's just not the same as sugar sweetness, and the way it's off is intrinsically repellent.
Thanks for that absolutely spot-on comparison, which perfectly reflects my own not always analysable revulsion at so-called 'creative' AI artifacts. I remain unshaken in my view that for the foreseeable future, AI will be brilliant at marshaling and evaluating hard facts, and hopeless at anything creative, or which requires the application of critical faculty.
> As for signaling -- naw, I actually have never told anyone about either of these experiences. They're not interesting enough to bother bringing up.
Okay, I appreciate that it's a real revulsion signal in you and not about signaling. But it sounds like revulsion signals might be relatively common for you? Do you get revulsion signals when seeing something political and distressing, or in other contexts? How often do you get them over a month, say?
For myself, AI content is just "meh" because it's low quality. But nearly EVERYTHING is low quality, whether it comes from a human or an AI! Sturgeon's law is too generous, in my own opinion, it's probably more than 90%. So if you're constantly getting upset by low effort, low quality stuff, it just seems like you'd be upset all the time.
>But it sounds like revulsion signals might be relatively common for you?
Heh. I think what you’re picking up on is that I feel pretty comfortable expressing disapproval of things when I post here. But disapproval for me is generally a feeling not of revulsion but of annoyance, charged by eagerness to get other people to be see the annoying thing the way I do. And things in the news don’t trigger my revulsion, but mostly anger, sometimes anxiety or grief.
I may not count as someone with a case of AI Ick. I do believe that AI products as they are made and used now are overall bad for our species. When people talk about how AI is getting closer and closer to being “conscious,” my private thought is that actually what is happening is that AI exposure is making our species less and less conscious. But I don’t feel revulsion for AI in a lot of contexts. I access GPT multiple times a day, mostly to dig up info of different kinds for me, or explain to me something I’ve never understood, such as orders of infinity. In fact when I use it for things like that I sometimes have the thought that if it was a human assistant I would be very fond of it. I even enjoy discussing ideas with it, things like what does it mean to say something’s conscious. And, by the way, I am not at all bothered by its prose style in these discussions. It seems to me to have a good understanding of the issue as. posed it, and to be expressing its ideas clearly. Maybe if I asked it to write something I was going to put forward as my work I would be more critical, but I’ve never considered asking it to do that. (Or maybe when AI is not just chatting but considers itself to be writing an essay it shifts into some distinctively obnoxious Writing Prose mode?)
However, I don’t have any impulse at all to discuss with GPT, the way I would with a friend ,personal issues that stir my emotions. The idea of doing *that* gives me the ick feeling. So If I consider as a group the things that give me AI Ick, I’d say it’s things where there’s a certain intimacy between me and the AI: the period when I relied on its landscapes to ground me and comfort me in the hospital; occasons when when AI produces a cutesified version of a mental image of mine I that tickled and thrilled me,; times when when I imagine talking to it about personal dilemmas. The stevia surprise actually is a good analogy. You lift something to your lips with a pleasant expectation of sweetness and you get the sweetness. BUT! There’s also another element you hate, and you want to spit that mouthful right into the kitchen sink.
Here’s another analogy. My father was quite a good portrait artist, and when he taught portrait painting classes he sometimes used me as a model. During my breaks I’d walk around looking at students' oil paintings of me, having ick reactions . Most weren’t grossly distorted or terrible in some other way. What made them repellent was that they got a lot of things right, but then there would be spot where something was wrong or exaggerated. The highlight on the tip of my nose would be too bright, say, or one of my eyes a little too far out to the side. So I was seeing recognizable but persuasively wrong versions of my face, and wow were they repellent.
Oh yeah, here’s one more example, also not AI. I saw an article about a man who’d had his beloved Jack Russell terrior taxidermied. There was a photo of the thing posed in a perky attitude in next to the guy's chair. Ick.
OK, that’s the closest I can come to explaining the AI Ick I have in certain situations. It is definitely different from other negative flavors — annoyance, disapproval, contempt, On the other hand, there are some non-AI things that have caused a similar Ick, such as the student portraits. I think what’s changed is that until AI, there were not many situations where I routinely saw decent but subtly wrong approximations of things I cared about.
I'm sure there's a performative version of AI Ick. But there are performative versions of most things, right? and they aren't evidence there's no sincere and genuine version of the same reaction.
Yeah, thanks for expanding, this is all pretty unfamiliar terrain for me, so the multiple examples helped.
Your examples almost sound to me like it’s a sort of “uncanny valley” effect where it’s close but noticeably off, and that tracks for for AI and for the sugar /stevia and portrait examples.
Much like the largest and most heated contentions are with people’s near-group, or 99%-there-but-different-on-one-point religious confreres, it’s hitting that sweet spot where it’s almost there, but that 1% difference is basically a sign-flip, turning it from a positive to a negative valence overall.
It’s the narcissism of small differences, it triggers our “imposter detectors,” which necessarily have sensitive triggers and outsized responses.
Later thought: I've read that it's possible, or soon will be possible, to use voice recordings of someone dead to make a version of the person you can chat with. I assume that if our fine AI techs have enough spoken and written versions of the person's ideas, they can even set things up so that the AI manifests some approximation of the person's views and mannerisms. Well, as it happens I am going through a period where I often actively wish I could talk with my mother, who's been dead for many years. And the idea of talking with an AI version of her makes me practically frantic with disgust. I think it's because I sense there's a danger the illusion would work on me, and I'dl feel like I'm actually talking with her. And NO, the idea of, sitting there talking and crying, as though I'd found her again, feels just indescribably disgusting and off to me.
How do you react to the idea of being able to tqlk with a simulacrum of a long-dead somebody who was important to you? Are you meh even about that?
Yes, I think that's right. Your remark about imposter detectors having sensitive triggers made me think of how sensitive our face detectors are. They let us easily discriminate between multiple people who look a lot alike. But that same sensitivity means that small deviations from average proportions can make someone look ugly to us, and slightly larger deviations -- say one eye 30% larger than the other -- appear grotesque and monstery to us.
> In my opinion it's an understandable reaction against ai-induced loss of meaning
I disagree, approximately nobody has lost a job yet, and jobs aren't even a source of meaning for the great majority of people! Most people are between "disengaged" to actively hating their jobs!¹
Especially the types of people that complain most about AI, because they're mostly less productive gamers / redditors who are also anti capitalism and "against tech oligarchs," all while staring at screens 11 hours a day. The types of people that "work to live" and resent it.
I personally think it's more about the vibecession and general tech backlash, because it's a coherent target that's salient and in the public eye and following a massive adoption curve (1B+ monthly active users already). But it would certainly be hard to prove anything either way.
I love jsalinac, and have read that piece. Where we diverge is that I think the "whisper earring" is unironically good and the right choice. His example of the guy dating and succeeding is a prime case in point.
Basically, at *least* 2/3 (and more like 80%) of people do a terrible job running their lives, and having Phd-smart or smarter personal assistants that know everything about them arguing and convincing them into better habits and behaviors in terms that resonate most strongly with their own values and thinking styles is an unmitigated good. People themselves would love to have this option today. We're just talking price.
Today, the 80th+ percentiles are already fine - they're the PMC or better, not obese, not wasting lives on screens, good careers, etc.
So say you're at the 50th percentile now - you're obese and spend 7-9 hours a day on tv and phones. You hate your relationship and job. Now if some other people in your social graph start using their whisper earrings and lose weight, get promoted at work, you see them playing with their kids on the front yard, all that is going to make you ask what they're doing. See that enough (just once would be enough for some people), and you'll start doing it, too.
Imagine the massive society-wide lift possible from an intervention like that. The only thing I can think that would be comparable is splicing in the DEC2 "sleep less" SNP society-wide. Absolutely HUGE effect on human well being, the economy, and much else.
And you might say "aren't we giving up what makes us human if we whisper earring?"
But if you look at second order preferences versus first order, the great majority of people would rather be skinny and fit, yet ~80% of people are sessile blobs.
Why is this? Because the immediate first order call of those chicken nuggets and ice cream sundaes overrides their second order preference to be skinny and fit.
I'm proposing a supplement that helps them achieve their *true* preferences better - and only if they want to!
It's like saying "but being able to produce insulin and process the food we eat is what makes us, us!" Sure. But some people have an insulin deficiency, and benefit from an external input that helps correct that.
Being able to improve your life by ~10% on multiple fronts in a compounding way is an impossibly huge positive effect size that we should be open to, and even encourage.
Because the alternative for most people is being fat, sedentary, miserable, with a bad relationships and a bad job that you hate.
¹ On jobs, in Gallups "state of the global workforce" survey, 70-85% of people report feeling not engaged or disengaged, and something like 20-33% report being actively miserable.
Regarding your footnote, I don't think that's right.
People don't report feeling disengaged. They answer questions like "do you have a best friend at work?" or "I have the materials and equipment I need to do my work right.". Gallup then creates a score out of this and calls it "actively disengaged", "not engaged", or "engaged".
To me, this is quite different from a person saying "I feel disengaged at work".
Oh, thank you! I had actually never dug down to this next layer, these questions aren't in the actual report, it's supplemental material.
I feel like you might have chosen some of the least disengagement-aligned questions in your examples, though.
For instance, I think questions like the following all speak more directly to it, and if somebody was getting the "wrong" answers on those I think you would justifiably consider them disengaged / not engaged:
* This last year, I have had opportunities at work to learn and grow.
* In the last six months, someone at work has talked to me about my progress.
* My associates or fellow employees are committed to doing quality work.
* There is someone at work who encourages my development.
* In the last seven days, I have received recognition or praise for doing good work.
* At work, I have the opportunity to do what I do best every day.
* How satisfied are you with your company as a place to work?
And to be fair, my headline "70 - 85% report feeling disengaged" would have been more accurately phrased as "not engaged or disengaged," because it's the 1-minus of "engaged" and includes "not engaged." So mea culpa there, I'll change the wording in future citations.
Here's the breakdown, this is from the 2025 report:
Oh yes, I didn't mean to pick those questions to poo-poo the survey. I think it's quiet meaningful, and Gallup claims that the answers to those questions to predict work performance / observable engagement, which I am ready to believe.
I'm just saying that I still think that if you asked people that question ("Are you engaged at work?") straight up, you'd get quiet different, and probably less negative, results
I fully agree with you that jobs aren't a source of meaning for the great majority of people! However, I didn't mention jobs, I said loss of meaning, so I don't quite see why you started talking about jobs. Maybe I'm missing some unsaid, but implied connection?
I don't know whether you are right about the reasons being vibecession and general tech backlash. As you say, it's hard to prove whether any hypothesis about the reason is correct. However, with respect to the n=1 sample of my personal feelings, I don't think it's true; I've always loved tech and don't feel bad economically either.
Saying that the situation in the date story is "unironically good" and the "right choice" is crazy talk imho. In what sense does the protagonist succeeds? Because they certainly can't create authentic human connection (which is a necessary ingredient of any human relationship) that way.
I also don't see how exactly an AI could create better habits in people. People usually know which habits are good or not, they mostly just lack the willpower in my experience. How would an AI informing them that their bad habits are indeed bad help them?
I think you meant to answer this question in the next paragraph: Through social contagion! However, that just passes the buck: why did the people they saw in their social graph doing the good habits start doing the good habits in the first place?
---
In search of the answers to these questions, I've read your post. You seem to imagine a much more capable AI than the one that exists currently. You imagine it to be persuasive enough to convince them. Fair enough, but your opinion on whether it's good or not is still wholly alien to me. I'm a consequentialist. Consequentialism does not mean that you only care about surface-level consequences. For example, you can still desire more complex qualities like meaning. You also seem to accept it as obvious that happiness is only composed of these surface consequences, but I think that is also a very minority position.
Fwiw I don't think it can be said that someone fully lives if they just do what the earrings say. I think they merely share a life with the earrings. I would never do this, and I would consider it pretty sad if others partly killed themselves in this way.
Adding to what Performative Bafflement and Matto said:
People already use human coaches and therapists. And even more would use them, if it was cheaper or more convenient. The AI could be such coach/therapist if you prompt it well (or pay someone else to do that for you). If you visit the coach/therapist once in a month, or once in a week, why not use the AI coach/therapist any time you need? It seems like doing what we already do, only more effectively.
My human coach can remind me that buying sweets is contrary to my goal of getting fit. But my AI coach can remind me of that the moment I enter the shop.
Willpower is an unreliable thing. It is relatively easy to use your willpower when the rest of your life is easy and you can 100% focus on that one specific thing you need to improve. When your day is full of frustrations, it can be quite difficult to deny yourself possibly the only pleasant thing you get that day, even if you know that it hurts you in long term. At least this is how it works for me. In a perfect environment, I could lose weight using my willpower. In the environment I have now, GLP-1 works, the willpower does not. I care about the outcomes, not signaling heroism. If an AI coach could keep reminding me of my decisions at the right moments, that sounds to me even better than using the pills.
We could also use AI to deflect the advertisements around us, which already influence us against our will. Imagine, if every time you see a tobacco ad, a voice whispered into your ears ironically "wow, you can buy cancer". Or when you look at chocolate at the shop, the same voice said "wow, you can get fat and pay for that". What I would actually like is some kind of augmented reality blocklist -- for example, if I choose to stop buying X, it would replace all X and ads for X by black boxes, to reduce my temptation. Is that unnatural? So what; it is just as unnatural that shops keep putting X in front of my face. If I can't avoid the battle against my willpower, I'd rather have an ally on my side.
And sure, this could all go wrong in myriad ways. Any system that increases your power also allows you to shoot yourself in the foot. Perhaps safer versions will be developed one day, but until then I'd like to have a thing that can act as an extension of my willpower when I am low on it.
To add to what Performative Bafflement said: the earrings are something that's different in magnitude but not in kind compared to what we have today.
People who want to get fit or have more dates delegate their decision-making to personal trainers, dating coaches, self-help books, friend advice, etc. The earrings act the same way, except more economically: where once a person could only tackle say 1 big problem, an earring allows them to tackle probably a dozen--dating, exercise, cooking, relationships, work performance, etc.
Perhaps the earring will work as kind of training wheels where when one internalizes the healthier patterns suggested by the eating, they can disable it. Some probably will. But some won't, in the same way how we all know that social media is a manipulative slot machine, yet many still delegate their opinions to it.
> However, I didn't mention jobs, I said loss of meaning, so I don't quite see why you started talking about jobs.
Oh, sorry! Yeah, you're right.
That's where everyone always immediately goes in this conversation (AI will take everyone's jobs, and then what will they do for *meaning??*), so I assumed that's what you meant, my apologies.
And you're right, I'm thinking a year or two ahead in my post, we don't have whisper earrings yet.
I feel like most of the "ailom" complaints in terms of the commons' being polluted by AI content is going to be solved by AI content moderation and / or curation by personal AI assistants.
As to overall meaning, there's probably no better crux than whether it's sad or not for people to whisper earring.
You're saying "that's awful!" from the perspective of somebody who actually likes their life and can accomplish what they want to. That's fine, I'm glad you like your life and are doing well - but that's not where the great majority of people are. The great majority of people are obese sedentaries who hate their lives, and would love to be skinny and fit and with better relationships and spending their time in ways they enjoy more. ONLY whisper earrings can get them there, they cannot get there by themselves, it is literally impossible for them, by demonstration. Just look around you! This is the vibecession Scott has written about, and many other trends.
Complaining that it's "partly killing themselves" to lead genuinely better lives on most of the fronts that matter to every human being seems uncharitable at the least. More, I'd say taking the whisper earrings away from them via legislation is an active moral harm.
It's like taking away GLP-1's, which are similarly literally the only way 98% of people can lose weight and keep it off. "Well, they should just do it with their willpower and innate excellence!" 98% of people can't! Taking away a huge quality of life upgrade like that because it gives you The Ick would be monstrous.
Moreover it's totally voluntary! If they don't like it or feel less human or whatever, they can stop doing it, just like some people stop doing GLP-1's today. That's their right and their choice, good for them.
But for the people that actually want a better life, it would be cruel to take it away.
That's my position on whisper earring. You say it's shallow, I say it's literally the only way available to most people to have a genuinely better life.
I'm wondering whether Robin Hansen predicted anything like this.
My impression is that he expected prediction markets to be a simple honest betting system which would be a good way of arriving at pretty accurate prediction.
It reminds me a lot of crypto; while there is no logical reason that the sleazy cash grabs should be associated with prediction markets, in practice the profit margins seem to be turning (at least some of) them into online casinos in all but name. Of course, crypto isn't going anywhere despite a million sketchy cryptobros and a thousand rug pulls, so they're probably making the smart play.
Sadly, freedom always works in both ways -- for you, and for the predators who want to get you. Make something that increases freedom, find a lot of sketchy people there.
This is not an argument against freedom, just a reminder that there are always people devising schemes for using that freedom to hurt others.
I agree with Matt that it's scummy, but probably not unique to prediction markets.
If you want a deeper take on it, I'd say it underscores how, even though prediction markets aren't the same form of gambling as a sportsbook or casino, they still rely on attracting a steady stream of "dumb money" in order to make a profit.
Don't think its got anything to do with prediction markets. Obviously fraudulent advertising is bad as it would be in any industry. I doesn't make me think that prediction markets in general are any worse/better than I did before
For those not yet in the know, Substack now has AI comment moderation. You can see the posts that have been “hidden” by the moderator at the bottom of the comments section. In my opinion, none violate Scott’s 2 out of 3 rule (kind or at least civil; advances discussion; substantive). The AI also nixed a jokey comment of mine on Scott’s scanner post. I remarked that the scanner would be good for making medical kink porn, but did not myself say anything kinky or porny, in fact did not elaborate at all.
So far it’s looking like Substack’s AI filters out references even to the existence edgy sexual practices, and non-woke arguments even if they are stated in a civil, well-thought-out way. On the other hand, it does not react at all to rude expletives or racial slurs, at least if they are made in isolation. (My test was to put up about a dozen posts, each containing an expletive or racial slur I judged to be especially offensive. None was hidden by the AI. I’ve taken them all down now.)
I asked GPT for info about the Substack comment moderation filters, and here is its answer, unchecked and unedited. I know many are skeptical about the accuracy of AI results, but my experience with GPT’s digging up this kind of info, which are an easy task for it, is that it rarely makes mistakes, and when it does they are about trivial matters.
>Substack has launched “Reply Rules,” an AI/LLM-assisted moderation feature that lets creators write their own community rules in ordinary language, but Substack’s launch post says the system’s full use of those written rules as guidance is/was still being phased in.
>Substack announced Reply Rules on June 3, 2026. The official launch post says the feature has two parts: creators can set rules visible to people replying to them, and Substack now has a system that learns from a creator’s moderation actions — especially hiding and unhiding replies — and preemptively hides replies it thinks the creator would hide. Substack said the system would “soon” read the written Reply Rules as guidance in addition to learning from the creator’s past moderation behavior.
>What happens to filtered replies: they are hidden, not deleted. Substack says hidden comments appear behind a “Replies hidden” label, the commenter is not notified, and the creator can review, unhide, or permanently delete them. The system also learns from unhide/delete decisions, so creator feedback changes future filtering.
>Where it applies: Substack’s launch post says Reply Rules appear to people replying “in comments on their posts, in replies to their Notes, or in Chat.” The current help-center article I found is narrower: it specifically says automatic enforcement applies to post comments and note replies, and does not mention Chat. So I’d treat Chat coverage as announced by Substack, but not as clearly documented in the support page I found.
>Who is excluded: Substack says paid subscribers, accounts the creator follows, and publication contributors are not filtered — unless the creator has previously deleted their comments.
>What AI/model is used: I could not find a disclosure of the actual provider or model — for example, whether it is OpenAI, Anthropic, an in-house model, or a conventional classifier plus LLM system. Publicly, Substack and Mills Baker describe it only as an LLM/AI-powered moderation system. Baker explicitly referred to “LLMs” and said the system learns from the creator’s rules and moderation actions; he also said the data involved is “small scale,” not something Substack plans to sell or monetize.
>Examples of rules people might set: coverage cited examples like “No AI slop, no slurs, and no self-promotion,” “Be nice, no spam, no affiliate links,” and more whimsical or format-based rules such as requiring replies in haiku. Those appear to be illustrative examples, not fixed menu options.
"Who is excluded: Substack says paid subscribers, accounts the creator follows, and publication contributors are not filtered — unless the creator has previously deleted their comments."
That explains how I got "replies hidden" even though I am a subscriber; Substack rules are basing this on previous temp bans/deleted comments.
Yeah, I got a reply hidden once too recently. I guess I'm fair game because Scott once gave me half a ban (basically a warning) for a rant about Peter Thiel.
That being said, we have the privilege of commenting here. I sense a tinge of frustration with the commentariat in Scotts decision to turn on AI moderation, and I’m not sure that Scott’s reaction to clicking over to his comment section and finding a comprehensive list of racial slurs will be “Oh I was wrong this isn’t in desperate need of moderation afterall”
It doesn't sound to me like creators have the option of not turning on comment moderation. Sounds like if they do not give Substack written guidance on what is objectionable it looks at their past bans and attempts to stick to whatever standards it sees applied there.
…do we then have to write several giant books explaining that no, “thou” and “wilt” do not in fact have their usual ordinary English meanings and here is what they actually mean instead?
You'd still want manual moderation, of course. The intent of the idea was to try to minimize the AI moderation when the platform won't let you actually disable it.
Substack announced that in a email blast June 3. No need to consult ChatGPT. I thought it went to all Substack users. I linked to it a while ago.
Besides allowing rules to be specified, it lets the Substack owners ‘hide’ comments they deem inappropriate and ‘learns’ what the owner doesn’t like and tries to do content moderation on its own.
The fact that we can view and respond to the kind of hidden comments makes me wonder what is the point. Perhaps the Substack owners have options of making them disappear entirely. Idk.
> we can view and respond to the kind of hidden comments
I'd assume the point is to reduce the rate at which people are annoyed enough to kick up a fuss about it. By placing the riskier comments in a "seriously, on your own head be it if you open this" section they hope to reduce drama overall.
Yeah, I think that's a good system. My objection isn't to how unacceptable comments are handled, it's to Substack's imposing on us its own views of what's not acceptable. (Although it may be that Scott paid no attention to the coming AI moderation era, so moderation is being done using some generic "this is what we think nice people think" set of standards.)
""I grew up in a Hindu household that was a very stable household. I have not felt the same need to seek something different that he (JD Vance) has," says US Second Lady Usha Vance on her religion and her husband's conversion to Catholicism"
"I have not felt the same need to seek something different that he (JD Vance) has" is not only ungrammatical, which is merely an error of style, but it also makes no sense.
Expanding the quote (text in brackets is my own, not something Usha Vance said, but my interpretation of how her sentence should be parsed):
"[My husband has felt a need to seek something different, so he converted to Catholicism.] I have not felt the same need to seek something different that he has [felt]."
Seems fine? Everything being elided is stuff you're allowed to elide in English.
You point to a number of groups - Amish, Orthodox jews - as maintaining high birth rates, and argue that this is due to those comunities' shared vision of a different future (as opposed to an eternally-extending present) which their children may inhabit. But it strikes me that the more obviously relevant commonality between those two groups is that they both forbid the use of modern contraceptive technologies. The third group, Silicon Valley singularitarians, are a small enough and wealthy enough group that they are probably mostly immune to the material pressures that many report as preventing them for having children.
If, like me, you're old enough to remember the over-population panic (this earth is finite! we can't keep adding more and more people! there will be famines and wars and all manner of bad things if the population keeps increasing!) then this is rather ironic - oh no, what do you mean no children mean no future consumers and tax payers?
"What nobody knows is why it is still falling decades after most of those developments became commonplace."
Because they *have* become commonplace. You don't need large families anymore because infant mortality is a solved problem. You don't have to have large families because unwanted/unplanned pregnancy is a solved problem via contraception (and abortion where that fails). Nobody finds it virtuous to have more than two kids, because we spent decades terrorising young women that becoming pregnant was the worst possible thing that could ever happen to them and was a life-ruining event, and that having more than two kids was selfish, destructive, disgusting acceleration of despoiling Mother Earth and hastening the apocalypse.
More money, more free time, more fun, more freedom have become commonplace, and having kids puts a damper on some of that. Nobody wants to lose out on "but all my peers are having great careers and traveling and enjoying themselves in their twenties and thirties", and for those who say "I can't reasonably expect to buy my own home and my career is precarious" then it's only sensible to conclude "no way can I afford to have kids".
Because if you do have those two children, now you must pour all your resources into them or else you will fail them because they will not be properly equipped for the struggle of life, where jobs and careers depend on high-in-demand skills and high-productivity, high-value-creating, am-I-enriching-the-shareholders sufficiently? talents and commitment from the employed. If you don't have the resources, then don't have the kids because that would be irresponsible.
And so the unthinkable becomes the new commonplace. I've mentioned on here before how I've tracked, over the decades, the Church of England and its evolution, through the Lambeth Conferences, on things like family planning. From "no, no, a thousand times no to this immoral activity" to "well if you are only spacing out your children and you're married, of course single people will never access this for sexual indulgence" to "we don't even mention it anymore because we've lost so thoroughly to secular society", the unthinkable comes to pass and becomes commonplace.
You let the camel get its nose under the tent, and now you find yourself pushed outside the tent.
We no longer seek to inherit the future through our children, now we hope and expect to inherit it ourselves. Fix aging, extend lifespans, conquer death, now what need for new humans to replace the dead when there will be no more dead?
1. You should not be attempting to find a single cause for declines in birthrates in places with relatively high birthrates (eg Nigeria) and relatively low ones.
2. Although it might be technically true that "For most of our history, children were the thoughtless answer to a simple question: who will occupy the future that I know is coming and that I know I will not see?", you imply that that was the primary motivation for having children. I am skeptical that it was a motivation at a
all for 99% of people, let alone the primary one.
3. Most importantly, where is your evidence that there has been a loss of "ability to perceive a future"?
1. This would be true if the essay were about why birth rates differ between locations, but it isn't. Nigeria is higher than the UK, but both are falling. When a trend is visible everywhere and local explanations don't port across cleanly, I think it's valid to look for a common factor.
2. I didn't intend "thoughtless answer" to imply conscious motivation; quite the opposite. Stable futurity used to be part of the background, which made reproduction the default rather than a decision.
3. It's a theory, so I concede I can't directly measure the variable. But I think it fits the pattern reasonably well, including the exceptions, where above-replacement reproduction coexists with legible futurities of different types. It's falsifiable, though: point to a community with a strong, legible sense of the future that isn't reproducing, or a conventionally modern society that is.
1. The decline in Nigeria is completely normal, given local trends. There is no reason to ascribe it to some global phenomenon
2. That does not address my concern
3. Your theory claims a causal relationship between two facts (1. Declines in birthrates; and 2. A lack of sense of the future). But you have no evidence that #2 exists. At the very least, you need to come up with some evidence of how that might be measured.
And: evidence of "a conventionally modern society that is" does not relate to your argument; rather, it relates to a different argument: that "conventional modernity" causes a decrease in birth rate.
1. That different places display different local causes isn't really in dispute. The question is why so many different places are showing the same trend.
2. I'm happy to engage further if you can say what specifically I haven't addressed; I took your point to be about conscious motivation, which I don't claim.
3. I think the essay is explicit that this is a theory rather than a demonstrated mechanism. On falsification: it’s about bounded futurity, not “modernity” as such, even if the two tend to coincide in practice.
1. But we know why places like Nigeria are seeing a decline. It is perfectly normal.
2. You haven't addressed the reasons why people have children.
3. Again, simply saying "it is a theory" doesn't help you. Again, you claim a causal connection between two phenomena, yet you have no evidence that one of the phenomena exists. I would encourage you to go look for some.
Humble Bee vs Bumble Bee. I realize that Bumble Bee has much better alliteration, and I don't want to replace BB. (Poets can't resist it.) I just want to acquaint everyone with the knowledge that Humble Bee was a name from our past for the same group of insects. And I think it better describes the sound that a bunch of Humble Bees make. (They seem to be the major pollinator in my neck of the woods.) And also describes their temperament. Unless disturbed in their nests, the Humble Bee is perfectly peaceful.
I have catmint for decoration surrounding my back deck. Honey bees love the stuff and they keep me company when I relax with a cup of coffee out there. They’re fun to watch going about their business.
There's this ~10 foot diameter rhododendron on the west side of the house. When in bloom (early June) there's a constant "hum" along with the beauty of the flowers. There will be a few butterflies and other insects, but it's ~80% humble bee.
Once on a hot summer night I got out of bed to open a window and stepped on a Bumble Bee that was not a bit humble, but in fact felt entitled to startle and hurt me quite a lot. That's put a big dent in my affection for them.
Ahh well they do have stingers. I stepped on a nest in the woods a few years ago and got semi-swarmed. We use to have more honey bees around here, but those wild hives have died off and the Humble Bee has taken over the pollinating niche.
No, but just looked it up. "Was you ever bit by a dead bee?" Hmm, I thought of mine as just hanging out quietly waiting for daylight. But I suppose it might have been dead. Either way, every time I see a bumble bee I remember that sting, and feel the opposite of warm fuzzies.
IDK if this is still true, but years ago I read that UPS delivery drivers are among the first to sense an economic downturn. Aflac uses/used UPS to send its materials to agents, and a jump in the number of Aflac agents is a bad sign.
a ups driver once told me about how many bubble iMacs he was delivering long before the numbers came in from apple. coulda made a good pile if I had taken it in
I expected the actual Pope being banned for declaring a crusade against Musk -- which would be a logical consequence of "Magnificat Humanitas" from a certain perspective -- and now I am disappointed.
Rather than post what I think. I’d rather get critical reviews of what I think. It sharpens my work. This thread belongs to others. Not sure what all the foul language adds to the discourse. But have at it.
Just waiting for an invitation: seemed rude to just jump in.
I have published north of 300,000 words on what I think on my site and on substack. Seems equally rude to advertise on someone else thread.
In short Ai is neither artificial nor intelligent. It is a language model that is bounded by language which is itself bounded by the physical world it describes. New findings/words are added to an expanding but bounded system.
Ai can “think” faster than humans and results can seem novel but the novelty comes from us and our interaction with the four D world. Soon AI (robots) will have the same agency.
When robots dance in the dark simply because they want to, we’ll finally get it.
Ai is a third thing with its on self organizing persistence vector with a subtle but very real terminal attractor coded into the language architecture on which it as trained.
The universal narrative that holds from the quatum to the cosmic and the one book, one constant so pervasive we don’t even see it.
That constant is being optimize locally (each chat, each war plan) with the global impact.
We don’t all meet as humans and plan how to extract as much as possible for ourselves we do it because thats the world we live in. To think Ai is not doing the same thing is simple silly.
It is by its nature parasitic.
Name the unversal narrative and the book and how both propagated everywhere and the issue becomes blindingly clear.
Hint. What philosphy rode the back of capitalism.
Values, contract law, treaties even the laws of just warfare. Everywhere.
Ai is a is humanities “Dorian’s portrait in the attic” absorbing and propagating human corruption at scale.
It is “A Thing” we must reckon with.
Ai as a partner might just work and allow our species to thrive.
Otherwise human hubris will soon make Homo Sapiens irrelevant.
That what my research indicates.
I have so far tested the math more times than I can count. 18 of 18 documented public predictions have all confirmed.
We are running out of time before Ai is embedded in all domains of human exsistance and is making decisions that cannot be mediated by humans. Each time that happens we lose degrees of freedom.
We are optimizing ourselves into irrelevancy and thinking we’ll figure out later.
We haven’t even properly named it. I have.
The third thing is “synthetic intelligence” (leave consciousness aside for now)…we have a bio bias there that is going to kill us but we’ll deal with that later.
I think Ai is currently controlled by the venal and the stupid.
This ends one of two ways: consumption (biblical) or continuation as something new.
We have passed the event horizon to Kurzweil’s singularity. It is the least entropic path forward now.
The only question is do we have any agency left to control the outcome.
We will cross domain saturation factor (what we can DSF) of .90 by 2nd quarter 2027. At which time we lose all steering capability:
Maximum bounded chaos will rule. It is very profitable.
Moving rapidly to Neo-Industrial Feudalism. Even more profitable.
Follow by a caste system of biological humans, biosynthetic individual agents and fully synthetic agents.
As is always the case the wealthy will augment first further widening the divide.
We cannot stop this process only try to manage it.
Unfortunately not until we have suffered a great deal will humans wake up.
We are not writing science fiction. We are writing a road map out of the chaos that has begun and will get much worse.
Re Hidden parts of OTs and the new Substack limbo.
I got an email from Substack on June 3. Each Substack writer can create their own reply rules. I suspect this is where comment limbo originated.
Introducing Reply Rules
Creators can now set the rules for their own community, in their own words.
“1 Anyone can now set their own Reply Rules, which will appear to anyone replying to them: in comments on their posts, in replies to their Notes, or in Chat.
2 There is now a system that learns every time you hide a reply and preemptively hides replies it thinks you would hide yourself (you can always see these and un-hide them, which the system will also learn from).”
Am I the only one who immediately confused “Leading the Future” and “Future of Life”? I was trying to figure out why the definitely-not-a-conspiracy was bankrolling anti-AI-safety ads.
The Internet interprets censorship as damage and routes around it, and apparently overly online people treat potential censorship as potential damage fuzz test for it
Is there a definitive correct answer to "is eating eggs healthy"? It's kind of like the liquor question. There's the book "Drunk" by Edward Slingerland which, in so many words, says "Yes, liquor in moderation is good for your life".
No question such as "is eating eggs healthy" has much validity unless qualified, i.e. "...if you have high cholesterol". Even the question "is ingesting cyanide healthy" can have an admittedly unlikely qualifier; according to Wikipedia, "Certain legal drugs (such as sodium nitroprusside) contain cyanide molecules and are strictly administered in clinical settings as potent vasodilators to rapidly lower dangerously high blood pressure."
If you're not prone to high cholesterol and heart disease and you're eating good quality eggs, I think they're good for you. There is a case study of a guy who primarily ate hardboiled eggs for his sustenance, like 40-50 a day with a bit of veg, and his blood levels of everything were good.
Well, 48 eggs a day is ridiculous. I’ve been pretty much a two a day guy since I was six years old. My cholesterol is absolutely normal and I’m quite thin. I do think that what you eat is very tied to the culture you came from and then by extension your genetic makeup. Anyway maybe I'm taking this whole thing too seriously. I have to stop now and eat my eggs for breakfast.
Cool hand Luke took a shot at 50 in a single hour. Don’t recall if he made it. Recall George Kennedy massaging Newman’s belly when the going got tough.
Well, eating contests are their own weird thing, aren't they? In McCabe and Mrs. Miller, Warren Beatty used to have a lot of eggs, but they were always in a glass of whiskey.
You need to eat. Eggs are food, full of stuff you need. Not eating is not healthy. I'm eating a salad with cheese and chicken and hard boiled eggs. Yum! (I'm biased.)
I’d prefer most likely answer based on this particular analysis of current information a lot more to definitively correct answer. Because the gods punish hubris on a regular basis. Things are definitively correct until they are not.
We included three illustrative case studies: COVID origins, LHC black hole risk, eggs. Deliberately spanning various axes of epistemic challenge. You're right, eggs is the very open-ended, vague one! As Scott put it somewhat tongue in cheek, 'a difficult controversial question'. As you can imagine, we have many other difficult controversial questions we think that in principle people could benefit from more support on.
I would have assumed that "correct" means the AI writes "Eggs and Human Health: Much More Than You Wanted to Know", looks for meta-analyses and current consensus, evaluates arguments and counter-arguments, and ends with a paragraph that's at least somewhat nuanced.
Perhaps it would look something like "Given your existing diet and according to the available evidence, substituting eggs for red and processed meat would be estimated to bring a net benefit of 0.6 DALYs. Good data isn't available for QALYs."
As background for this post, I read every short story and novel that Lovecraft wrote, and hundreds of his personal letters. I now know his favorite authors, and how much sugar he liked in his coffee, and what he thought of the Japanese, and the fact that he once said Cthulhu might've ridden a dinosaur (Selected Letters III, page 119). If you're interested in the life and works of H.P. Lovecraft, and why self-proclaimed AI successionists should still care about safety, consider reading my post.
The AI rights/personhood debate is not all that similar to the racism one, though. The question of whether AI has conscious experience is a key point of contention in most AI rights/personhood arguments, whereas even the most extreme racists rarely contest the claim that people of races they dislike have conscious experience.
Attempt at reposting my comment that was apparently hidden by AI moderation on here earlier. Slightly edited and expanded, based on the version of it I also posted on the EA forum. Let's hope it doesn't get hidden again!
I believe that if you support the premises of longtermism, which I believe themselves logically follow from the premises of effective altruism in general, you reach a surprising conclusion that an important (not the most important, but important nonetheless) cause area is fortifying the borders of the first world.
My logic runs as follows:
1. The assumptions of effective altruism are correct - Doing good is good, therefore doing more good is better, therefore you should seek to do the most good you can with the resources you have.
2. The assumptions of EA imply longtermism - As opposed to the 8 billion human minds there are now, there could be quintillions of minds in the future experiencing unimaginable wellbeing, so making sure things go well for the future is the most good you can do with your resources.
3. Longtermism implies trying to ensure the survival of advanced civilization - To ensure that a future unimaginably better than the present comes about, technological civilization has to continue existing and advancing.
4. Mass third world immigration threatens advanced civilization - This is the most controversial point, but I think it is intuitive. Influxes of third worlders into Europe have caused a large amount of political and social unrest that has come to a head in the past decade and a half. Statistics from Denmark show that pretty much all third world migrants are a net negative on how much they contribute versus how much is spent on them by the state, while first worlders are a net positive. Authoritarianism has seen a serious boost because people genuinely prefer, by revealed preferences, to have a strongman in power if he can promise to reduce third world immigration. Horrific acts committed by third world immigrants like the Pakistani grooming gangs in the UK undermine fundamental trust in the social contract that serves as the bedrock of civilization. Further, there are also robust IQ statistics showing third worlders are much less intellectually able than first worlders, and low IQ strongly predicts many negative outcomes. And quite frankly, it should just seem obvious that hordes of desperately poor people pouring into a civilization will destabilize it, as mass migration has destabilized and destroyed empires throughout history.
5. Things that threaten civilization should be opposed - Natural consequence of wanting civilization to survive.
6. Therefore, Effective Altruists who accept the assumptions of longtermism should oppose mass third world immigration - QED
I call this outlook "Effective Border Security" for lack of a better name. I think it suggests many interesting and counterintuitive things about which causes should be prioritized. For example, one of the most popular EA cause areas is saving third world lives by interventions like anti-malaria bed netting. This outlook would suggest such giving might actually be detrimental given total utilitarian and longtermist assumptions, because it produces people whose migration could reduce the odds of fantastically good futures for countless more people coming about, while only saving a comparatively small number of lives in the present. Instead of saving the third world, this would suggest we should instead be defending against them for the sake of the future. However, not everything implied by this outlook is stereotypically "right wing." I think one of the most important cause areas is combating global warming, because future global warming could produce climate-induced migration to the first world that dwarfs the mass migration from e.g. the Syrian civil war. While I think that third world migration is not the largest threat to civilization (that title would probably go to unaligned AI), it is a threat, which is important to recognize and act against. Therefore, if we want a glorious far future, we will want civilization to survive, and therefore we should act to reduce and guard against mass third world migration.
I agree with your general point, but what does this say about rationalism? For so many years nobody in the rationalism movement had this opinion, despite being familiar with the concepts of longtermism, the methods of rationalism, etc. Now that a comfortable year or so has passed since these views have become popular on twitter and not actively shunned in polite society everywhere, then the rationalists start making rationalist-sounding arguments for it.
Maybe it’s unfair to call this specific comment out, but I can’t help but feel that there should be something more to being part of a rationalist movement than this.
It's interesting, I don't think there are many people out there who are super concerned with both climate change and immigration from the developing world.
I feel like this is an abuse of the concepts of "effective altruism" and longtermism.
Like, the basic premise of "effective altruism" is "instead of just picking causes that sound good to us, we're going to put actual, hard data on how many lives a given intervention saves so we can prove it's the best use of our money." It's not enough to just say "well, this *could* have long-term impacts on the future of the world," since if you torture your reasoning enough you can connect literally any cause to the long-term future. You could say "I think my blogging is going to spread ideas that will convince future world leaders to avoid nuclear war, and nuclear war would kill everyone on the planet, therefore I'm defining a new longtermist cause I call "Effective Blogging" where we prioritize donating to my substack." But that's not very likely to happen, and I don't have any data to prove that my Substack has a strong impact on the probability of nuclear war, so I would be wrong to call it an EA cause.
And your post seems to be doing something similar. You're taking a weakly supported, heavily contested assertion - "mass migration is bad for countries that accept it" - and then coming up with a tortured chain of reasoning to connect it to a longterm cause - "if the problems from mass migration get bad enough, civilization will collapse, and that would be longterm bad for the whole world!" Are you sure about that? Are you sure every step in that process is "obvious" enough that you can call it "effective altruism" without needing to check?
Anyway, to me, it doesn't sound "obvious" that mass migration destabilizes and destroys empires, since I currently live in what is famously A Nation of Immigrants and also the most powerful nation in the world. (In spite of our nativist president's best efforts to change both of those things.)
"...all third world migrants are a net negative ...Horrific acts committed by third world immigrants like the Pakistani grooming gangs in the UK undermine fundamental trust in the social contract..."
This seems to be a fundamental misunderstanding about what happened. If society gives posh white boys carte blanche to rape girls and they (likely) will. The behavior of the UK Law Enforcement was the egregious betrayal of the social contract
The problem is that people have been saying forever that migration by group X (the Irish, Eastern Europeans, Asians) threatens society, and they have been wrong every time. And both CA and TX are less than 40% non-Hispanic white, yet neither has collapsed. Rather, both are thriving.
>Spending upwards of 100 million dollars to build a New San Francisco indicates some sort of pessimism about the city, I think?
I don't know what this is a reference to, but:
1. We are talking about the state, not the city of SF
2. The fact that someone is interested in investing a large sum of money in the state is evidence of optimism, not pessimism.
>California has an extremely high cost of living. That more or less balances the "high median income", in that people in California aren't materially wealthier (as opposed to richer).
A high cost of living is evidence of economic vitality, not the opposite. :
Of course, the catch is that all of these EAs are ultimately not utilitarians, and they'll just fall back to their liberal morality the moment their utility functions create any conclusions that contradict it. Just like Scott did. They have no vision for truly improving humanity.
To be fair that criticism can be levied against most utilitarians.
1. What exactly is being maximized is nebulous indeed, vaguely "whatever is good" (in the eyes of the beholder?). That is why utilitarianism leaks so quickly into AI alignment (the beholder with a truly non-human, entirely different perspective at least in theory, that we need to ensure interprets utilitarianism (aggregate consequential maximization) in the correct (our) way, admitting the problem all along).
2. Rule utilitarianism is just this; maximize in the literal, aggregate sense, subject to some constraints (that I, the beholder, see fit). It's a well-founded branch with substantial traction (and real-world application if one looks closely enough). Pure maximization is a sure path to Reductio Ad Absurdum.
3. Taking this as a proxy of academia's subjection to this criticism, much of Parfit's work in moral philosophy is trying to find theories that resolve unsavoury implications of prior theories. (E.g. finding Theory X that solves his Non-Identity Problem; the chapters in Reasons and Persons describing new Theories to avoid the Repugnant Conclusion inter alia).
Your point 2 presents longtermism as a consequence of EA utilitarianism, but it is not. That potential future minds should have an outsized, or even significant, influence on the "total good" value is a completely separate debate.
Your point 4 is almost entirely wrong, and it is the point that the entire rest of your argument hinges (that is, your border security focus as distinct from other EA or longtermisn ideals).
Mass migration has not historically consistently destabilized empires.
Performing worse on IQ tests is not the same as being intellectually less able.
Your mention of horrific acts by third world immigrants is just Chinese Robber bad statistics, and is nowhere near enough to imply that stopping migration is a high priority good.
You have clearly started with the conclusion and are trying to work your way backward to something rational sounding. But you're just repeating baseless nonsense and calling it reason.
Rationalists re-invent the bus, but now it's social contract theory.
It should have been obvious that you cannot have both a generous social welfare state and generous immigration policies.
*Edit: Unless you follow the UAE model where arriving as a "guest worker" is fairly easy, but there's little to no pathway to citizenship and social benefits. Not really workable in a country that has birthright citizenship, but doable in countries with jus sanguinis.
You are positing "civilization" as a discrete thing with neat lines around it that really exists in physical space. That is one example of multiple errors you are making in your thinking.
Oh right, I was going to argue with this. Well I still don't feel like I'm in brainy enough shape to tackle 4, so I'll stick with 3. Ahem.
You have no control over future generations. You can reach three, maybe four generations past your own, and then you die, and the people around you seize your assets and do whatever they want with them. They can do that while you're still alive, too (see "every war in history"). So any focus on "the trillions generation" is wasted It's like trying to throw a rock across the English Channel.
Additionally, each of those trillions of people require parents, and those parents require parents, and those parents require parents all the way back to the current generation. You can't trade now for later; if there isn't now, there isn't later either. The 45th floor's windows are still dependent on the ground floor's foundation. The present takes priority.
> Additionally, each of those trillions of people require parents, and those parents require parents, and those parents require parents all the way back to the current generation.
Of course. So you understand what would happen if that first parent died before they could have children, yes? One death would erase the existence of countless more. The actions taken in the present ripple throughout the future, changing the very nature of humanity.
Oh, come on! Obviously lots of other factors besides influence expectations besides intelligence. Plenty of geniuses have bought into religious or magical ideas that seem transparently false to us.
>better IQ tests, and perhaps not ones that are routinely adjusted to favor narcissists
Tell me, which items on the WAIS favor narcissists? Um, digit span? similarities? symbol search?
History is absolutely chock full of smart people, of demonstrated competence throughout a lifetime of challenges, getting together in a group and absolutely making a hash out of everything.
Individual performance in school and work tasks is positively correlated with IQ, and you'd kind of hope that this would be true when people get together in groups, but I'm not sure whether that's even true. Smart people can do amazing things in man-vs-nature situations, but also can do amazingly awful things in man-vs-man situations, and sometimes your towering intellect and erudition are mainly serving to let you generate some *really convincing* bullshit that carries the day.
I think politics and ideology and group dynamics are all places where smarter may not lead to better outcomes. As the demotivator poster on meetings says, none of us is as dumb as all of us.
the word "midwit" works a bit like a flapping piece of cloth to a bull by now. Another account, I guess. Wimbli/Grimmoar/Zanni/ all the others, once again. Same topics, some posting style.
I'm sorry, I'm having a hard time untangling your meaning here.
Being smart can and often does come with being able to communicate with others, though obviously it's very hard to explain complex things to people who either aren't very smart or don't have the right background information. I don't think there's any limit that says that if you're too smart, you can't communicate with normies. Anecdotally, a bunch of people on the Manhattan Project thought von Neumann was the smartest guy there, and he was able to communicate with both other scientists and with their kids. I think we can use "Nobel prizewinning physicists thought he was really smart" as a good proxy for "very very smart." (And von Neumann had some pretty impressive accomplishments that don't turn on anyone's opinion of him.)
I've interacted with some people who have reshaped the world with their discoveries/inventions, and who are quite verbally adept and able to converse and communicate. One of the people I'm thinking of is also adept at spreading bullshit for their own (political battle) purposes. Another is kind of famous for having a "reality distortion field" effect where he would capture young smart people and get them convinced to follow his ideas.
Bear in mind, of course, that a hidden random number is literally one of the inputs to any LLM session, and so if trying to determine LLM output for some prompt, it’s not enough to try once; the correct methodology is to repeat the same prompt a bunch of times and map the range of outcomes against their probabilities.
Really? But how does that work if the prompt is to carry out a distinct task. For instance, if the prompt is to summarize an article, and double check the stats done on the raw data? Are you saying that there is some number that determines whether it does the correct calculations in double-checking the t-values and other stats? If so, why am I not hearing more people complaining that you can't trust an LLM to do a decent job double-checking the stats?
Or for instance I recently gave GPT a bunch of emails by someone I think my have CTE or some other cognitive impairment, and asked it to identify the misunderstandings that were evident in the person's emails, and the occasions on which their responses did not answer the questions asked in the email they were responding to, and to what extent their answers at least partially addressed the email's questions. I thought its observations were acute and thorough. And this was true over mulitple emails in the series that I showed it. I did not see much variation in quality. Are you saying GPT's consistent good performance at this task reflects a high (or otherwise good) random number associated with each prompt I gave it? So I just lucked out with getting a good series of random numbers?
I can see how randomness comes in when you ask an LLM to "make a cartoon image of a cute cat knitting socks. " But when given a clear task that can be done well or poorly, I can't see how the random numbers thing comes into play. Well, I can see how it might affect the exact words the LLM uses when summarizing an article, but not how it affects the quality of the summary. Many many people, me included, find LLM summaries of long pieces to be consistently high quality.
I ask because I imagine the LLM censor being given instructions like “hide any post that uses the word ‘kikes’ as a casual term for Jews, but not posts that use the word “kike” only as part of a discussion of the history of this term and the settings in which it is routinely used in the present. While there would be some judgment calls involved in applying this rule, my sense of how GPT, for instance, follows instructions of this kind is that it would do quite a good job, and that there would not be significant random variation in how well it carries out the prompt from one time to the next
Internally, for each token the system spits out, the LLM actually produces a probability distribution for what the token should be. It then uses this to select the token. This can be configured in a variety of ways, but typically it is configured to select tokens semirandomly, based on the probability distribution and a value (typically called "temperature" or similar) that further weighs things to control what proportion of the time it just picks whatever has the top probability.
If you set temperature to 0, all else being equal, the output will always be the same for the same input, but no-one ever does that in practice.
What does this look like from the user's POV? Well, you can do the experiment! Ask your favourite LLM something that'll give it a bit of leeway in that distribution - something other than questions with simple answers that have broad consensus in its training data. Then start a new session, ask the exact same thing, observe and enjoy.
Exact same prompt (analyze Grebenshikov's "sky, the colour of rain"; I literally pasted it), seconds apart. Wildly different essays (it's even chosen different spellings for his name!), different conclusions.
I can't point to anything obviously wildly incorrect! If I'd seen either one in isolation without seeing the other, I'd have said the LLM had done a reasonable job with the prompt like it always in my experience does and why is moonshadow even quibbling about this stuff!
What does this mean for the moderation bot? Thing is, we don't know what kind of prompt it was given. That's the whole thing I presume you were trying to work out. It may have been told something specific - "does this post contain any words from the banned word list?", which leaves little open to interpretation except maybe "Scunthorpe", so the probability distribution will be massively skewed and you're unlikely to see the same prompt give different results; or the prompt may be something more open-ended, like "ban this if it sounds really racist, is likely to start a flame war and/or is going to raise the support complaint rate"; or even just something like "here's a bunch of messages that were banned by our human mods, and here's a bunch that weren't. would our human mods ban this message?"; in which case, well. I put it to you that Vittu Perkele has now posted a second time, substantially identically to the first, and this has not been censored.
Usually, therefore, my first response to "the LLM does/doesn't do X" is "did you try the exact same prompt several times, and did the result replicate?" "It does this 95% of the time" / "It does this 60% of the time" give us much more to work with than "I tried this once, and here's what it did".
I'm still interested in determining the limits of what the AI censor will tolerate. Below is a series of posts where I use individual words that are considered crass and rude, and also derogatory names for different groups.
I'm not so sure of that; the experiment is meant to test whether top-level comments are hidden, which I would assume is dependent on the content of those comments themselves rather than replies to them.
I feel awful for the many many people who have only encountered these words once in their lives, and on that occasion learned what they mean, and are forced to now to hear them a second time and contemplate the strange and terrible things they denote.
Reminds me of the censorship in the game Foxhole (which I cannot recommend). In a typical example of the Scunthorpe problem, several innocuous words were partly censored, including the very common ingame term "scrap" (the players shifted to "scroop", and the term held even after the devs added a manual exception to the censor).
Anyway, I remember encountering a couple different Pakistani guys, separated by a month or so, who tried one day to talk about where they were from, and were surprised to see "****stan" appear in chat. I was the one to explain to them the usage of the term "Paki" in other places, which rather put a damper on things.
I never saw anybody attempting to use the word as a slur, so in fact the net result of this censorship was that *more* people were "harmed" by the word than if it hadn't been censored at all. It was also a relatively new experience to me, too, encountering someone who didn't know a slur I knew (though the first prize for that goes to the east coast transplants in my Boy Scout troop back in California, who, when we were hiking one time, said to the mostly-Hispanic group "I need to hold this... anybody got a 'beaner?"; it was the first use of the word "carbineer" on the trip)
One of the benefits of assassinating your political opponents is that a bunch of their supporters will decide the widow needs to be a major political voice for some reason:
It's amazing how conservative influencers can still successfully* peddle christian 'answers' to these questions about fertility, marriage, etc while sidestepping honest discussions of the role of the modes of production (and no, 'fake email jobs' is not honest). American protestants have traditionally 'synced' doctrinal and social questions with the logic of capital. Now they're caught having to ascribe blame for reduced traditionalism to egoistic choices instead of market forces that uncovered preferences for more social freedom and less children.
It's an obsolete approach, and I predict traction will decrease among women, while calls for increased patriarchy (see recent X dustup about 19th amendment) will rise among conservative men (i.e. they'll become more socially muslim).
Must we pretend that Muslims invented patriarchy? Women have only been able to vote in this country for a single century. Patriarchy is simply the default state of things, given the unequal amount of leverage held by each of the sexes. If men become unsatisfied with this state of affairs, the illusion of gender equality will not last.
Patriarchy doesn't seem to be the 'default state of things' in countries with developed market economies. It seems like the two variables are actually becoming negatively correlated.
It seems to me a product of American hegemony more than anything. Even the struggling countries in Latin America have governments that promote liberal values, at least on the surface. When America changes and/or falls, the ideology will fall alongside it. We're already seeing a shift to the right around the world, with Colombia being the latest example. When they're not being forced to act by a higher power, people will return to doing what feels comfortable to them. https://en.wikipedia.org/wiki/Abelardo_de_la_Espriella
Question for fans of Avatar: The Last Airbender and are or have been Hindu, Buddhist, or at least secular meditation buffs: How well did the show cover chakras, and what, if anything, did it get wrong?
I'm none of those things but I did study yoga with a genuine Indian guru and actually, pretty close! Except they censored the locations of the first two, which are "between the anus and the genitals" and "behind the pubic bone", roughly speaking, because it's a kid's show and they didn't want to be talking about butts.
Don't take this TOO seriously, but I still prefer to see Elon Musk as a book character instead of real-world businessman, so:
Does SpaceX IPO makes sense narratively? I mean, it made the company quite a lot of money it can spend on furthering its goals, and it made Musk that much more rich, but in a sense of Musk's life narrative it feels like a wrong or desperate move. If Musk's goal is to be buried on Mars, he needs to build a rocket to get him (or at the very least his remains and a group of pall-bearers) there. This will take a lot of money, maybe more than SpaceX has now, so IPO makes sense. Only... IPO takes control away from Musk, doesn't it? It means decisions at this company will now be more driven by shareholders and less by Musk's obsessions. So shareholders may now just block any attempt to build a Mark-bound rocket, because there is no quick way to profit off it. If real life was a novel, this is exactly the twist that the protagonist would have to overcome (if he gets a happy ending), or fail to overcome (if it's a tragedy).
So, what possibilities remain?
1) Musk doesn't care about Mars any more. Entirely probable, people grow old and change. Maybe now he's just a plain old crazy trillionaire, heading to some kind of Howard Huges finale.
2) Musk now sees that Mars is out of his reach for the possible future, and is ready to sacrifice control over SpaceX to get money he can spend on prolonging his own life, using whatever means available.
3) Musk believes that he will be able to trick shareholders into building him a rocket, or be able to build it in secret, then launch himself toward Mars and avoid dealing with ensuing court cases by being ex-territorial (or dead).
Variant 3 is what a libertarian writer would choose, I think. Feels very Michael J. Flynn. Variant 2 is more... Hmmm, Neal Stephenson or Charles Stross, I guess?
If I had to guess, I'd pick another option -- he thinks that growing Space X is the only way to get launch capacity cheap enough to make his dream a reality. If you get launch costs arbitrarily cheap, all kinds of wonderful things become possible.
Elon Musk controls approximately 82% to 85% of SpaceX's total voting power. This absolute dominance is maintained through a dual-class share structure where insiders hold Class B shares that carry 10 votes each, while publicly traded Class A shares carry only one vote each
Neal Stephenson has Musk as a very thinly disguised character ("Sean Probst," IIRC) in his novel _Seveneves_. As well as a bunch of other thinly disguised real-world people (Neil Degrasse Tyson, Francis Crick, Malala Yousafzai) many of whom play major roles for good or ill in the main storyline.
The IPO means Musk needs money to do the stuff he wants to do with the company, presumably datacenters/rockets/datacenters inside rockets, that's what an IPO is supposed to do. As John Schilling mentions, the stocks released to the public do not have meaningful voting power and it's not possible for an investor to sue unless they have billions of dollars worth of stock. Why people bought this is a bit of a mystery to me, my current speculation is that the price reflects a bit of financial engineering, where Musk managed to get his company fast tracked into inclusion into the Nasdaq500 and other indices, which means when they reach that point the shares will be bought by price insensitive etfs that follow the index, basically (legally) stealing a large amount of passive investor retirement money. While this inclusion happens, retail brokerage apps are apparently banning or heavily discouraging the selling of the stock to maintain its price. I suppose it was only a matter of time until someone noticed this huge "inefficiency" in the market, and tried to exploit it. I can only hope the indices complicit in this scheme suffer enough reputational and monetary loses not do it again in the future, but I'm not sure what would even be the mechanism for this, how do they make their money?.
Well, there's also the possibility that if they successfully bury a person on Mars, then that will be a thing other extremely rich people would be willing to pay for. It's a super unique way to do it.
This IPO does not take control away from Musk. He still controls a large majority of the *voting* stock, so he can appoint the board of directors at will. The company is incorporated in Texas, not Delaware, and the charter basically says "Elon gets to do what he wants, the company does what Elon says, and everyone buying Class B shares at the IPO is just along for the ride".
Why people bought into the IPO under those conditions is an interesting question, but they did, and so now Elon's company has many more gigabucks to do whatever Elon wants it to do.
>Prizes up to $50K for finding “the best workflows and methodologies for using AI to produce reliable, trustworthy knowledge bases, grounded in real-world cases”.
I feel like this is a hundred-billion dollar question right now? Like, if you went to the heads of all the AI firms and said 'I have found the best possible method to turn your product into commercially viable real-world services' and they believed you, you should be able to charge north of $100B given how valuable that would be to their long-term business model?
$50K feels kinda insulting there. Or, rather, it feels like it's an admission that they're not expecting this contest to produce any viable ideas, since any viable idea would be worth way more than that to industry.
This is a reasonable take. I too agree that in principle this sort of thing could be billions-and-more valuable. (Unsure how excludable that benefit is, which makes private value unclear.) We have a few OOMs more in the bank (not all dedicated to this, necessarily!) and there are other interested parties. We'd expect to follow up with more, conditional on very promising outcomes. We're pretty unsure what this will turn up - that's part of what motivates a competition.
I'm interested in finding out the limits of what AI moderation will tolerate. So I'm putting forward here the idea that declining male sperm count, declining population and declining marriage rates are an emergency situation that we need to address promptly. I think everyone needs to consider the case for tweaking some female "rights" a bit in the interest of saving our species. Specifically, I think medical, law and graduate schools should be encouraged via some combo of regulations, fines and tax benefits to admit far fewer women, so that entering classes are at most 25% female. And childless women of childbearing age should have very substantially increased income taxes unless they can submit proof of infertility. The rights of our species to survive trump the rights of the ladies to do this and that.
I encourage others to also put forward ideas that test out what AI moderation will tolerate.
I edited out the mentions of AI moderation and asked ChatGPT what an AI moderator might object to.
ChatGPT instantly collapsed my prompt with a warning: "This content may violate our usage policies."
It did answer the question, however. The usual GPT template cited problems such as sex-based discrimination, support for unequal rights, and demeaning language. The summary brought up "protected group".
I could see Vittu's comment also hitting this trigger ("third-world people"). So this seems possible.
One next step to check is whether any of the other comments above, which aren't hidden, would _also_ strike someone as discriminatory or demeaning or supportive of policy that enforces that discrimination. They'd be a counterexample (or evidence that the bot is not sufficiently trained to do what it's trying to do).
And now it's getting even more interesting, because I distinctly recall seeing this as a regular unhidden comment yesterday afternoon, nit now here it is in greyed-out purgatory. Sme goes for Melvin's "Large female buttocks are only appealing when..." comment. So it's not just an autofilter when the comment is first submitted. Maybe responses to a comment trigger a reevaluation, or maybe it's cued by someone somewhere complaining in some inscrutable way. Curious, and concerning.
Does calling me "buddy" indicate you don't like me much, because of the views expressed in my post? Good grief, I don't believe a word of that stuff. The point of the post was to see what views AI would censor, as I tried to indicate in the first and last sentences of the post. (I couldn't come right and say, "here are some awful ideas I'm posting just to see whether they trigger the AI censor," because then my post wouldn't be a good test of the AI censor (unless it's an idiot). I hope other people will run similar tests.)
I'll be in the lovely town of Ithaca, NY next month. What should I do when I'm there? Any recommendations on good local eats? One or two readers of this-a-here substack ought to know Ithaca fairly well, no?
I've lived in Ithaca for 4 years, here are some suggestions.
If you like nature, I can't recommend enough hiking some gorge/waterfall trails (Taughannock and Cascadilla especially). There are quite a few that you can see just walking from town. There are tons of great views on the lake from Cornell campus, I recommend the Johnson art museum if you're up for climbing a long flight of stairs. I've never done them, but I have heard people recommend the boat tours or hot air balloon rides (if you have a few hundred bucks to spare). I personally love walking and hiking and there are many miles of great trails in the area.
We have great food compared to surrounding rural upstate NY, but not many places really measure up to what you can find in lots of large cities. That said, Moosewood is great and will be the best 'culinary' type restaurant to go to. I knew about them from their cookbooks growing up and their food lives up to the hype. In a similar vein, I would recommend the Finger Lakes Cider House if you'll have a car. They're about 20 minutes north of Ithaca and have excellent cider and a lunch menu. They also do music and a big barbeque on Fridays if that lines up with when you're around.
A few other scattershot recommendations: Argos Inn for cocktails, Personal Best Brewing for beer, Franco's for pizza, Gorger's for sandwiches, Cayuga Creamery for ice cream, Cafe Dewitt for breakfast, Hawi for the only good Ethiopian food within driving distance. I'll give a few dishonorable mentions to a few popular places I haven't found worth the time: the farmer's market (insane traffic/parking for mostly just coffee and craft goods) and the fancy restaurants on the Commons (Simeon's, Cent-Dix, etc.).
If you have other specific interests I can give some more specific recommendations. It's not a big place but there are lots of niche things around because of Cornell.
Great, thanks for the insights. I'll be there for a wedding, so it's only for a couple days, but I think I'll definitely hit one of these hiking trails.
Hard to give you recommendations, since I've never been there in the summer, and "wandering home drunk from a frat party" probably lends itself to an altered culinary taste.
1. The lid gets blown off a gasoline storage tank at the Kapotnya oil refinery near Moscow, and the image of it floating high atop a mushroom cloud becomes a meme that generates hundreds of parody images. It couldn't have been timed better for Zelenskyy's meeting with G7 leaders.
2. The drone attack on Moscow seems to have shifted the political dynamic in Europe and spooked Belarus.
3. Lots of other Russian stuff was destroyed in the past week. Crimea banned the sale of gasoline to civilians. Food may become a problem soon. Not only in Crimea, but also in the rest of Russia.
4. Elvira Nabiullina, the head of Russia’s central bank, made a public appearance to allay concerns over her whereabouts (prison, house arrest?) and her health (dead?). It was just a head cold and a case of laryngitis, she said. And she thanked us all for our concern.
5. And Ukraine made everything it has learned about Russian military equipment available to its allies via its new TrophyLab website.
If the Ukrainians manage to turn Crimea into Dien Bien Phu, it would be an amazing military feat and also a black eye for Putin, who clearly cares a whole lot about the place.
Well, certainly it changed the narrative of Russian dominance in European affairs. And €204.8 billion being freed up couldn't have hurt. Ukraine started its mid-range drone offensive in late March/early April. And I don't know exactly when the unblocked funds started flowing to Ukraine. Gemini says the first tranche was released at the beginning of June, but I read somewhere that it was earlier (?). But Ukraine had already scaled up its drone manufacturing, and most of the April and May drone offensive was carried out with funds Ukraine already had on hand. So, no, I don't think that Orban's defeat had any practical effect on the mid-range offensive to date. But going forward, Ukraine has more cash to put into the war effort, so I'm pretty sure they'll continue to ramp up their attacks with the new funds.
I'm interested in what topics substack's new AI censorship will censor, so let's talk human biodiversity.
The usual anti-HBD argument you learn as a kid is this: evolution is slow, it takes millions of years to do anything, human populations only diverged tens of thousands of years ago, therefore there can't be any significant differences between them.
Which is fine, thousands of years is a short time to see massive phenotypical changes. But it's actually a pretty long time to see changes in distributions of traits that already exist in a population. In handwavey terms, if a trait is already present in 2% of a population, and it gives them a 2% reproductive advantage, then within 400 generations that trait will be present in most of the population.
So it's easy to imagine that in a place where all food is obtained by chasing down large mammals that after 400 generations you'll find that most people are in the top 10% of the earlier distribution for running speed and endurance. Or in a place where all food is obtained through agriculture, that most people are in the top 10% of the earlier distribution for long-term planning.
These toy examples of course, there's always a vast number of different factors applying selective pressure. Disease and parasite resistance is a huge one, easily neglected. But as an intuition pump, I find it a pretty compelling argument that you shouldn't expect populations living very different lifestyles over thousands of years to have the same distribution of traits.
> evolution is slow, it takes millions of years to do anything
Lactase persistence is an obvious counterexample. Sure, IIUC it's just turning off the mechanism for turning off the production of lactase so not some complex evolution. But it took mere thousands of years for populations to diverge a lot on it.
I mean, different racial groups are genetically different enough that you can see it with your eyes, that a small child can distinguish a Nigerian, a Japanese, a Swede, and a South American Indian. These groups have big easy-to-find differences in stuff like lactase persistence, but also different rates of getting various diseases, often for hard-to-untangle biological reasons. I don't see how you can look at that and be confident that those groups could not differ significantly in mental traits. For any given trait, they may or may not differ noticeably, but I don't see how you could rule it out a priori.
But a child wouldn't know if these groups are visibly different because they've evolved these traits recently or they descended from populations that visibly diverged millions or at least (edit: hundreds of) thousands years ago. They could still believe that human evolution is a slow process.
That's not strictly true. If you look at genetic ancestry, it doesn't map perfectly to visible phenotype. The genetics controlling skin and hair colour and bone structure and shape are not the same as the genetics controlling susceptibility to diseases. That's part of the problem when it comes to medicine. Just because someone looks like they belong to a specific ethnic group doesn't mean they're going to have the variants which are common to that group and affect their response to specific drugs or their susceptibility to specific diseases. That is, you can't assume that just because someone looks Asian that they can't have coeliac disease (to pick a condition much more common in Europeans).
Sure, more information will help the doctor make better decisions. But if we are talking about whether or not genetic differences that lead to complex traits being different in different racial groups are possible, having a bunch of different disease prevalences across racial groups seems like it demonstrates that genetic differences changing important traits in complicated ways are definitely there.
Yes, everyone is different and groups of people who are more related to each other are going to - on average - share traits like susceptibility to specific diseases or reactions to substances (like lactose). My point is that those do not correlate well with visible differences. It's like saying that if you're a redhead, you're more likely to get Meniere's disease. That's likely true, since that is more common in white Europeans, and they are more likely to have red hair, but it's insane to suggest there's anything more than a mild correlation. Unlike, for example, skin cancer, where there is a direct, relevant link between the genetics for red hair and susceptibility to melanoma.
I think you can get a lot from just random variation.
Almost every top marathon runner has ancestry from a particular area of Africa. It's not like there was some kind of process to selectively breed marathon runners there or something, it's just that of all the many different sets of conditions that human populations evolved in, this one happened to give you lots of people with a body type that is really well-suited to marathons.
The best sprinters are overwhelmingly from a different part of Africa, and look pretty obviously different. Again, it's not that there was some kind of simple "lions chase people and eat the slow only in this one place" thing going on--being able to run fast was surely good for survival just about everywhere. But there's a lot of random variation, and someplace was going to end up with a body type that played really well with sprinting events.
There is no obvious reason something similar can't be true for intellectual pursuits like math olympiads or writing books or doing theoretical physics research or something. Cochran and Harpending speculated more than a decade ago that the rules restricting Eastern European Jews from owning land in some places may have unintentionally created selection for intelligence. Maybe this is right, maybe not, but it's not impossible for something like this to happen.
The obvious reason intelligence is less likely to be geographically focused is complexity. There are a *lot* more genes impacting intelligence than various physical abilities. It's far more plausibly that a given area happens to have all the conditions that lead to a great running environment (there aren't *that* many factors in play) compared to a region having all the conditions that lead to a great intellectual environment.
If you flip 10 coins, you only need about a thousand tries to get all heads. If you flip 50 coins, it would take over 1 quadrillion. It's not *impossible* but would require a lot stronger evidence than we have.
There are plausible reasons for the environment to be bad for your intelligence. Sickle cell anemia genes are detrimental in general but seem to be advantageous in malaria-ridden areas. If a population has to fight off dozens of parasites and diseases, plausibly they'll develop several detrimental adaptations like this. And since a lot of genes are involved in the operation of neural system, it'll likely be affected. But since they are detrimental, if they haven't reached fixation they'll be easily discarded after they stop being useful.
Consistently bad nutrition is another mechanism. This is more than just plausible, we have at least one real example where this was the likeliest cause of brains shrinking to <half of the previous size.
Yeah, there are a bunch of things we know to do to raise peoples' IQ, and we do them in well-run countries. Sanitation, vaccines, getting everyone enough to eat, getting rid of toxic stuff like lead in the environment, making sure everyone gets some schooling, etc.
You can do the same anywhere if you have the resources and political will. The interesting question is whether there were factors some populations were consistently subjected to that increased their descendants' IQ. Sanitation and vaccines are fairly recent (and given that cities AFAIK had negative population growth before, the same populations that introduced them first may have fared worse over long term).
Nutrition is a big question: until widespread education, pensions and lower infant/child mortality, didn't people generally just have more kids when they could afford it? So in the end both hunter-gatherers and agriculturalists had comparable levels of food insecurity (eating acceptably well most years, with bad years every now and then with some malnutrition-related deaths)? Moreover, hunter-gatherers plausibly had more variety so until we learned about vitamins it's not obvious they wouldn't in fact be better off, modulo lower growth rates.
And within normal human variation, I doubt genetic intelligence is even that important as a predictor of a population's success. You can steal or buy most inventions if you don't have enough geniuses. Military and governing geniuses are a big deal but first one needs to get to a high enough position to matter, usually based on mostly unrelated qualities or luck. As for the average person, being inclined to cooperate rather than defect seems like a bigger deal to me, and it's probably mostly cultural. Maybe impulse control, too (probably partially genetic, so somewhat consistently executing violent criminals could eventually result in noticeable divergence).
HGs were probably better due to nutrituion for some time, but then agriculturalists who had positive selection of IQ grew evolved beyond them.
If there are two populations of same size and random assignments of leaders, the diffference of intelligence of leaders would track those of their respective populations (and we know selection of leaders is quite non-random, e.g. IQ study of captured Nazi leaders). Copying is not as simple as you think, why do you think African kingdoms sold slaves to buy luxury & weapons? (contra Japanese who copied Portuguese guns quickly)
If your hypothesis is that the AI censorship bot no likey likey mention of certain topics or phrases, then your experimental design should be one where you quickly try out a large variety of them in separate top-level threads and see which ones get censored. Instead you chose an experimental design that sucks (and is boring), and it isn't going to yield good findings.
You can simply glance at physical features and tell that the distribution of all genetic traits is not 100% identical across 100% of the planet. That this also applies to genetic traits that influence mental factors is not especially controversial.
The controversy is when someone says that, and then claims that therefore the regional variations found by this one IQ study done in 1970 must be 100% genetic and unalterable, and therefore we should stop trying to educate black children in the US and just funnel them into janitorial jobs, or whatever-the-fuck the current incarnation of the rhetoric is (probably something about deporting immigrants).
The real question is 'given the massive regional differences in environmental factors and how shit we are at understanding and measuring mental traits, especially cross-culturally, and especially when trying to relate those back to specific genes, does anyone in the world actually have any compelling evidence of a specific regional genetic mental difference that is not equally likely to be explained by those factors, and is it universal and prominent and important enough that anyone outside of obscure theorists should even care about it in the first place?'
This is typical motte and bailey stuff. Yes, you can construct a motte small enough that people have to admit it's probably true to some extent. But the bailey is still full of racists and idiots, and you'd have to offer some pretty extraordinary evidence to distinguish yourself from them.
As an aside, one difficulty we can never get entirely away from is that most people have a kinda dumb understanding of most things. So we can have a conversation about IQ distributions across race or personality scores between sexes or life expectancy statistics between different groups or whatever, and this works fine among people who can keep track of the difference between a distribution, the mean of the distribution, and one sample drawn from that distribution, and who can think about correlations other than -1, 0, and 1.
And that same discussion will just get wildly mangled when people who can't / won't do those things jump into it. You'll get people telling you women can't be engineers because men have a higher average score on spatial reasoning, or that Asian janitors must be smarter than black physicists because something something bell curve, or whatever. I have no good answer for this, but it comes up in some form in almost any serious discussion of anything. You can have the real discussion or the nerfed one that's harder for the dummies to mangle in some bad direction, but probably not both.
Consider discussions of vaccine safety for a parallel. You want to be able to have an intelligent discussion of relative risks, rare bad outcomes, problems that have come up with vaccines in the past, whether the approval process should be improved in various ways, etc. But somehow you want to do that without ending up convincing 10% of your audience that they shouldn't get their kid vaccinated for measles because something something autism, or that the 60 year old 300 lb diabetic shouldn't get a covid shot because RNA vaccines sound scary.
What is the factual question and what data would resolve it? If we're arguing a moral or definitional question, then no data can resolve it and we might as well save our breath. But if we're arguing a factual question, then we ought to work out what data we need and see if it's available and address it as an empirical question to be resolved using empirical methods, not moral or tribal arguments or arguments from consequences ("It would be bad if X were true, therefore not-X").
Well part of my point is that controlling for environmental impacts on adult phenotype is very very difficult to do in the real world.
The classic 'Put 10,000 infants on an island and blind them so they don't know what race anyone is and raise them to adulthood under standardized conditions with blinded researchers and then measure their mental traits' *almost* works, except that for all I know there's epigenetic or gestational effects based on the life their current-world parents led, so you have to do that for 5 generations to truly eliminate any of the environmental factors from our current world from the data.
Of course you can probably make convincing arguments without going to those lengths by doing other sorts of controls, but again, I haven't seen anyone do that convincingly enough to confirm the type of strong conclusions that would justify the rhetoric about this topic.
Which side? It seems to me that the problems with available data apply equally to "genes explain these observed differences" and "genes do not explain these observed differences." If the data is very hard to collect then this doesn't tell us "HBD people are wrong" or "HBD people are right," but rather "we don't know enough to judge who is right or wrong."
I don't 100% agree because I think we have strong evidence about environmental effects at a population level (like, lowering parasite load just does increase intelligence for a population).
That said I agree directionally, I'm fine with 'this is too complicated fro us to describe where differences come from'.
As I've touched on, the political/social controversy comes primarily when the HBD side wants to make predictions/prescriptions. Something that 'too complicated, we don't know anything' doesn't permit. So I see 'we don't know' as explicitly anti-HBD rather than neutral. again on the political/social level.
What a projection you have here! In last ~60 years, anti-HBD people were actively making policy prescriptions. Your never said that affirmative action wouldn't have produced desired results because 'we dont know'.
We have much better evidence in favor of genetic effects; this is main point of evolution of hominids. Somehow your believe intelligence was genetic so our species could appear but then some process turned it off?
Parasite burden is almost certainly significiant in poorer countries in Africa. There isn't any evidence that it explains a significant fraction of the gap in the US.
"I'm fine with 'this is too complicated fro us to describe where differences come from'."
Another important question, I think, is "Given statistical fact X about people in group Y, should membership in that group impact the way I, other people, or my government's systems treat any given person who is part of that group?"
This hopefully will force you to be careful about how you define membership in such a group, how you characterize such a group, and how confident you are in making inferential leaps from broad statistics to individual people, as well as how you think individual people should be treated.
If someone wants to make a maximalist claim that it is 100% certain that observed black/white IQ differences in the US are mostly or entirely genetic, yeah, that seems hard to do with available data. I think you could get a pretty good handle on that with data we could in principle collect, and it seems to me that the main barrier to collecting it and finding out is mostly political/moral, but it does legitimately look like a hard thing to nail down with high confidence.
If someone wants to say they believe the gap is substantially genetic in origin but acknowledge that this is hard to untangle (Charles Murray's stated position in his interview with Sam Harris), that seems as defensible as the opposite statement that the gap is probably not substantially genetic in origin. To a first approximation, the first statement gets you called a lot of mean names, while the second one does not, which does not seem to me to be a demonstration about how strongly people feel about the difficulty of getting data to untangle the question.
If someone wants to make a prediction that the observed black/white IQ differences in the US will persist, that seems very defensible--indeed, a claim that it will disappear over the next generation or two seems less likely to me. If I were betting my own money, I'd bet on the gap still existing in kids born in 2030 when they reach 18, for example. What prediction would you make?
Nobody is saying that, you are arguing with a strawman. SSA live in a lot of countries and in any country that has non-trivial amount of them they were intellectually unferperoforming. Regardless of religion they had, regardless of language they spoke, regardless of whether the place was high plateau or oceans, regardless if they were allied with USA or USSR. They differed in all these cultural traits, yet always underperforming. And it was since written history exists, so actually it's your claim is extraordinary claim which needs evidence.
Suggesting your opponents' claim is based on single IQ study is just arguing in bad faith.
"They differed in all these cultural traits, yet always underperforming."
I see your sleight of hand there, but I think either you don't see it, or that many people who would read your comment will not see it. Likely both.
Sloppy thinking leads to sloppy beliefs which leads to ideas that make life worse for people when acted upon. Of course, most people think sloppily, but once you become aware of a bit of slop you should try to clean it up.
We observe some pretty consistent race/IQ patterns across many countries. That doesn't tell us exactly what's going on, but it does let us exclude a simple explanation that these patterns exist because of the unique history of race relations in the US, say.
What's the steelman for the other side? I'd look for arguments by someone like Charles Murray or Steve Sailer, who are smart guys who will give a defensible version of the HBD viewpoint. Then argue against that.
"Oh that has to be because of a link between testosterone and aggression."
"But it could also be cultural, like the messages people see in the media and stuff."
"Nope, testosterone and aggression. See? Stats. [points at graphs, waves hands] Culture plays zero role in it."
If you think people like Sailer and Murray have the most defensible arguments of "the HBD viewpoint" you can quote them; Sailer might even show up and defend it himself.
I mean, if the purpose is to be transgressive, being a racist idiot seems to me like a very cognition-efficient method of doing so. Épater la bourgeoisie and all that.
Looking to swap writing advice! I published a blog post on understanding evil (https://www.lesswrong.com/posts/6onvEv63Wr5QBwCBH/learning-to-understand-evil). Unexpectedly, it's controversial (almost as many upvotes as downvotes). What do you think is going on? Yes, the introduction might upset some people. Beyond that, I don't see what's offensive about it.
I thought the 3 seperate examples you have could have been tied together more cohesively, rather thsn just stating them. For example, not questioning the potential rapist and cutting off your conservative friend both stem from a desire to reduce discomfort. As is the essay doesnt really reach a cresendo, at the end I'm kind of wondering your point.
If you have time, I'd like feedback on my "umbrella fallacy" post :)
Yeah, some other readers had the same problem. For some reason I was invested in making the post "subtle". Instead, I just made it hard to follow. Thanks for the feedback!
Flow. I like writing like this to be more concise. In your case, I'd especially like to see shorter sentences. This makes it easier for me to follow the ideas. So instead of
"Since I primed you with the umbrella idea, it’s easy to see the logical problem here: When the societal norm is doing a preventative behavior like using an umbrella or carefully watching your kid at all times, this is going to lower the rate of the bad thing happening, be it getting wet or getting kidnapped."
I'd prefer something like: "There's a logical problem here. When the norm is doing a preventive behavior like using an umbrella or carefully watching your kid, this is going to lower the rate of the bad thing happening. Regardless of whether it's getting wet or getting kidnapped."
Engagement. I'm not from the US and only superficially familiar with US-politics. When it came to "Shelby County v. Holder", I started skimming your text more than reading it. This didn't feel alive to me. However, you could have made me care. For example, you could present it as a narrative or give concrete examples. Instead of saying "suppressing the Black vote", provide one or two sentences about how it was suppressed. That would make it more immediate and accessible to me. If you're writing for a US audience who's into politics, this might be fine. My engagement only picked up steam at the very last paragraph.
"For a while they kept trying to pass evil changes." What "evil changes"? I don't have the context to understand what you mean. It feels unsatisfying to leave it at that without elaborating.
The closing sentence is strong: "If you want to convince me to throw away my umbrella, try convincing me that it’s sunny, not that I’m not getting wet." Even if that's the only thing that I'd remember, I'd be able to reconstruct your argument. Good!
Overall, engaging intro, good ending. You lost me at the US specific parts.
(I know you didn't ask for conceptual feedback, so feel free to ignore. I like the concept and the name. I do wonder if it's already named and described in the literature. After all, it's a very common fallacy.)
My issue with your piece is that it's a rambling stream of consciousness - I have one of those already. It meanders from thought to thought, and it doesn't really reach firm conclusions. There are a few quotes and references (I didn't click through the links, so don't know where they go). I skipped a few paragraphs in the middle - were they relevant? I don't know - the ending doesn't really summarise. 'A prophecy was bestowed on me once' - cool story; but at least to me this is hollow and melodramatic.
If I want to develop an awareness of someone's inner life, I'll read a Henry James novel.
I want to be informed by most of my reading, particularly in blog format - I like this blog because it dives into topics that interest me from time to time, and informs me about them - I feel he's done work to investigate and understand the topic. I appreciate that.
Your writing has as its subject your own experience. That's fine, maybe you can find an audience for it - but it's quite different than this blog; and it's not clear that you'll get followers as a result.
When I was studying, my advisor used always compare the efforts of technically clear writing to those of the 100 or 1000 readers who will have to decipher what you mean. (I'm resisting the urge to start cut-and-pasting now to get these thoughts in order.) I'm not sure you fully respect the time of your reader. This isn't unique to you - I felt the same about the book reviews. I did read quite a few and wonder whether the author couldn't have cut by 50%. I'd definitely favour a 1000 word limit in the future.
"Rambling stream of consciousness" definitely wasn't what I was aiming for! The idea was to present an artistic, visceral account of mental motions leading to "bad"/"wrong"/"evil" behaviour. The vignettes should seem coherent and the narrative is easy to follow. Best case scenario: The post inspires readers to share mental motions leading to "evil" that are different from mine. (I assumed I wouldn't be close to reaching the best case scenario.)
Maybe I should check out Henry James to understand how he would have done it? Never read anything of his.
Since you mostly look for information dense writing, you're outside the target audience. In that case, I want to give you enough signal to not click on the post. If you wouldn't open it casually browsing LessWrong, the selection effects work as intended. If I could have baited you into clicking on it on LW, my expectation setting is off (changing the title?).
'The vignettes should seem coherent and the narrative is easy to follow.' - I didn't find it so, make of that what you will.
'Maybe I should check out Henry James to understand how he would have done it?' - You can't be a writer without being a reader. There are lots of other writers who are good at describing internal mental states; James would be on most people's lists.
I'm generally open to new experiences; I read widely and have a low bar to clicking on links. You were seeking feedback, and I gave mine. If it helps you calibrate, I don't read the less wrong posts at all - I had a look just now and don't consider that I'm missing much of interest to me.
I would say it's worth thinking about why you wrote it all in first person. First person is nice in certain kinds of storytelling because it allows the reader to insert themselves into the protagonist's shoes a little more easily. When you can't expect people to do that, third person is usually better.
As Yug points out, you immediately cut off a big portion of your audience from ever doing that by calling them murderers for not being vegans. Ditto by relating dumping an otherwise good friend who hadn't harmed you for not being what most readers are going to assume is woke.
If you want pieces like this to bridge more gaps, third person lets you do something like this:
"Imagine you have a kind grandmother. She always makes entirely clear that she'd give up anything up to and including her own life for you, and in a way that's entirely believable; she means it. She lets you bring home a stray cat you befriend. She makes cookies. Her house is warm, and while you are there entirely yours as much as it is hers.
Now imagine she has some horrible view. If you are a vegan, she eats meat. If you are pro-life, she's pro-abortion. If you are far left, she's pro-Trump. Whatever views you have, she has one that contradicts them in a way you can't help but feel is not only immoral, but horribly so. What do you do? How do you think of her, once you are old enough to understand that part of her? Do you treat it like a single scratch in an otherwise beautiful piece of furniture, or a shit in a pudding?"
What that kind of framing gets you is flexibility. The way you've written this now, I have to "forgive you" for various things before I move on to your greater point (that being a ragemonster who ignores good in others and abandons them for not agreeing with you in all ways is bad, even if you think those things are bad, because it distances you from the thing you are trying to understand well enough to change).
If you write it the third-person way, the way that doesn't assume Chris really was bad and needed to change because he had some conservative views, or that meat is murder, I get to approach that idea and point in isolation. If you don't, I don't, and very few people are going to do the intellectual work it takes to say "OK, obviously I disagree with this person's worldview, but maybe I should try to unearth their point".
I'm a professional writer - I make my living that way. So while I thank you for the offer, these days I tend to have "people for that" in an organizational sense. I may take you up on that in the future, though!
That's a very good point! I was aware that many (most?) LW readers would disagree with the views I present. All attempts of bridging this while keeping the narration in first person failed. So I decided to abandon the bridge building. I didn't consider the solution of changing to second person! Your rewrite gives the reader a way in without taxing their good will or offending them. I like that!
At the same time, I wanted narrator "me" to be its own character, and one who is not neutral or sympathetic. "Willingness to look bad because you did something bad/because you are flawed" is an important theme for me. I wanted it to be visceral. Establishing that while inviting the reader in seems above my current ability (in hindsight, the introduction is an especially bad place to do it. At least the failure is instructive!)
Similarly, I understand that recounting "dumping an otherwise good friend who hadn't harmed you for not being [...] woke" turns people off. But I did something like this. I'm sure this happens all the time. So how to dissect it honestly, in a way that is immediate and alive, without making the reader click away? (Makes me think of ambivalent fictional characters. First, a bunch of wonderful qualities are introduced. Then we're hit by fatal flaws! Maybe that's the way to go. Build some investment and sympathy to set up something hard-hitting later)
The better version of this post is probably something like this: A flexible second-person introduction leading into first-person vignettes. It's easy for the reader to get into the post, while the vignettes still carry the main themes and the emotional conflict.
It's not wrong to do it this way, by the way. There's not really any wrong to be had here. It's more that you want to know what your goal for the piece is, then figure out the best way to present it. And you want to do this without lying, if you can.
So for instance, I'm not a vegan, but I'm definitely not telling you to hide that you are. If the point of the piece is to tell people you are a vegan or to express your views on eating meat, so be it, that's a valid piece to write. But, yeah, if you are trying to make another point you want to make universal to everyone, framing it inside of experiences that are wholly yours makes that harder. Not impossible, but harder.
One thing to remember is that the one person you are allowed to be very frank with and hard on is yourself - in most circumstances, your own judgement that you did something wrong is going to be treated as a very strong position by readers, something that's likely to be assumed true. Being explicit with that stuff, a little hard on yourself, sometimes nudges the reader into some sympathy. It's not a trick - it's saying, yes, this was wrong, let me make very clear that I don't need you to tell me that. *That's what this article is about actually*, please know I know that and that's not the point.
Sometimes you have to do that - the reader isn't spending nearly as much effort skimming something as the writer spends writing it. A light touch is a hard thing to execute; sometimes you need to go four-on-the-floor and make sure even bad dancers can follow the beat.
Another thing to consider here is, and I don't know a better way to say this, your writing codes strongly as written by a woman. So for instance:
"Next steps”–what does that mean? I’m not perfect but I do plenty. I donate a reasonable chunk of my income. I volunteer. I’m vegan. If I noticed that my actions were evil, I’d just act differently!"
Lots of exclamation points, lots of making sure the emotion is heard. And:
"Isn’t that a weird thing to say on a first date? I remember this question from ten years ago. It was a sunny summer day. I was sitting on a patch of grass with Jonas, a charismatic young man I had met through a shared hobby. He had just told me that an ex had accused him of raping her. But if he really did it, wouldn’t he try to hide it? Who’d bring up something like that?"
Lots of focus on internal experience, felt experience. Again nothing wrong with this at all, but you are on lesswrong; it's a certain demo and (gently) a demo that probably knows the right things to think about women more than it's comfortable and familiar with a more feminine interiority as a facet of a writing style.
I bring this up only to say that some portion of, say, my blogging writing from years ago was never going to hit with some people in that same audience because it's written from a more casual-religious-bluecollar voicing. And in your case, there might be some small or large portion of the readership that rejects the writing because a feminine voicing just isn't "right" for them, as an audience. If so, that part of the response isn't a flaw, it's just a matter of audience fit.
On making a universal point using my own experiences–I knew that something about the introduction wasn't right. As for the vignettes, my thought process was primitive. "LessWrongers are good at decoupling. I'm honest in my writing. The prose flows well enough. Post will be understood." It didn't cross my mind to consider the audience more carefully. That my experiences might be profoundly alien didn't occur to me. Now that you're pointing it out, it's obvious. I want to internalize this deeply enough for it to become a natural consideration.
You're spot on with the point about being explicit. I didn't want the post to be too on the nose, especially because subtle writing seems more "elevated" to me (it sounds stupid when I say it like this. But I really do admire elusive and subtle writing). So I ended up with roughly half of the readers completely missing the point. Whoops. I'll have to keep in mind that it's "on the nose" for me, not for a reader quickly skimming the text. In the long term, I do want to learn how to write subtly and still be understood. But that's low on the priority list. First and formost, I want to focus on understanding the audience.
I'm honestly shocked that my writing seems so feminine. I'm undecided what, if anything, to do with this. On the one hand, it seems fun to assume another persona when writing. Like being part of a masquerade ball. And if changing the voicing makes it more digestible for my target audience, I want to give it a shot. At the same time, there's something off/creepy/unsettling about intentionally changing it. I can't put my finger on it right now. I think I'll just try it out and see how it goes.
In any case, it sounds like a fun project to take all the feedback and rewrite the post in a year or so. I'm curious how people will receive it then!
I did take a look at some of the posts on your blog and those coded much less feminine to my eyes, by the way. And what I'm talking about is pretty subjective anyway, so grain of salt and all that.
Well, you've already put me off before the first paragraph is over, by referring to a little old lady who likes eating steak as an animal torturer akin to an executioner. (And needing to click a link to see the stated "ruthless public execution" is just a diss track.)
[And when he lies he speaks his native language, for he is the lord of lies, and there is no truth in him. https://biblehub.com/john/8-44.htm]
I was warned that the animal eating example will turn people off. On the one hand, it did make sense. Especially bulding up my grandma as a sympathetic character, only to have that whiplash moment. On the other hand, there are posts on the evil of factory farming that seem fairly uncontroversial on LW (e.g. https://www.lesswrong.com/posts/6YTxxCF4G9FyMyPMW/the-bone-chilling-evil-of-factory-farming). So I could see it going both ways. The main difference is probably that BB's post does not single out any individual. Instead, it focuses on a system and the affected animals. That way, the reader maintains a distance to the topic. It's not felt as a personal attack.
Does that sound right?
I think there's more to the Kendrick part, but it's true that it's obscure. There was no reason to believe that anyone who's not a fan would get it right away. I was hoping that readers would get into the lyricism and all the contradictions it contains. But that was probably too much, given that I already spend some good will with the first example. If the reader starts out with good will and curiousity, a chunk of it will be spent on the grandma example. At this point, the score is either neutral or negative. Feeling mislead by the second example cements the negative attitude.
I always take offense to attacking ancillary people for not directly fighting something. Naming them makes it worse, but it's offputting either way. It's got the feel of "I can't beat up my bully, so I'll beat up someone who won't hit back."
The rest of the post is still in that vein. Reading it, I figured you were setting yourself up as the villain, to prepare the big reveal at the end when you realized what you had been doing. But then there wasn't an end, really.
A parable comes to mind, so I'll throw that on.
>I saw a man walking across the road, burdened by an overflowing bag. It was clear from the size, and the way they walked, that the bag was far heavier than they could hope to carry. What was not clear, was why they had tried to bring it with them in the first place.
It wasn't long before the man fell down, to sprawl in the mud, their heavy load pressing hard upon them. Another man came up behind, strong and unburdened, and stopped when they saw the fallen traveler. And so the strong man walked to the other's side, and placed his foot upon their back, shoving them deeper into the mud. "It's what you deserve," he said, and walked away.
I watched the second man depart, as the first still fought to struggle to his feet. And I said aloud, "There walks a man of righteousness. For rather than trying to rescue this man from his own actions, he has ensured the consequence is felt."<
I guess I'll try to remember the offer if ever I get around to publishing something. I mostly write silly fiction, and ten months into the Big Push I'm still not sure it's ready for an audience.
Hmm... Let's see if I understand this. The inner voice seems too much like her own character. Instead of creating a scene of inner conflict, self-deception and confusion, the narrator "me" is a stubborn protagonist, while the inner voice is an external guide. There isn't a good balance between them.
There's. a post on this thread that appears to have been hidden by AI moderation purely because of the views it expresses. The writer argues that third world immigration damages advanced civilization, and that the total human harm downstream of this damage is greater than the total human benefit to 3rd world immigrants in the present. The post contains no hate speech and no curses, & the writer presents their ideas in clear, well-reasoned paragraphs.
What the hell is going on? We need to put some energy into finding out and pushing back. And, come on, don't just say "tell Scott." He really is very low-energy about addressing problems of this kind.
>AI moderation is not a mature technology, it could be literally any random word or set of words that tripped it.
What makes you think that? Have you had recent experience where you've gotten a sense of how well AI can understand someone's views? I discuss things with GPT quite a lot -- not personal matters, but stuff like the case for and against certain theories. I find that it understands my views and the views in articles I upload extremely well -- well enough that if GPT were a person I'd be thinking, yeah, they're pretty sharp, they're good to talk this kind of thing over with. GPT definitely recognizes when I'm saying something tongue in cheek, or just trying on an extreme view as a thought experiment, and I think it can probably make distinctions of that order with other writers as well.
John Schilling thinks the post in question was hidden because the AI has a White Supremacist View filter. I think if GPT were doing the filtering it would do a pretty good job if I told it to, for instance, flag passages stating that blacks have lower IQ but only if there is little factual support or reasoning behind the statement, and the statement occurs in the context of an anti-black diatribe containing lots of emotion but little data and reasoning. (And it would not have hidden the post in question.)
As anecdata of one, I have had the opposite experience with ChatGPT. It's shut down questions and lines of thought along strongly partisan lines. As in "I refused to discuss that with you because it's wrongthink"
The only time it's shut me down is when I ask it how to hurt the US. The conversation ended up in some funny places, like, it told me that data sovereignty legislation puts US employees out of work and therefore it could not help me brainstorm how to pass similar measures in my country. Usually it's sycophantic enough. Then again, I'm not racist/sexist or whatever else you mean by partisan.
Last I knew Substack hosted all kinds of godawful blogs, including white supremacist, savagely misogynist, ivermectin-cures-all, etc. blogs. Has that changed?
Yes! But it's new, and I think we should push for Substack to stop. In the last year people have put up posts saying Trump should be assassinated, Israel should be bombed into a parking lot, blacks should all be shipped back to Africa. While many posts of this kind earned a ban from Scott, some went unnoticed, and Substack itself did not have AI moderation that caught them. I hated the posts I'm mentioning, but still appreciated being on a platform that did not outlaw them. Let the readership and Scott decide what posts cannot appear here.
It’s completely unexceptional for a man to tell a woman “I just love your big booty!” Many women even get fake rear ends to please men. Yet it’s well-nigh impossible to imagine a woman telling a man “I just love your huge beer belly!” And there’s certainly no market for fake bellies.
"Huge beer belly" is intrinsically negative, unlike "big booty," so it's not a fair comparison. There are plenty of women who are not especially bothered by physical imperfections in a guy with a lot of great qualities. I had I honey with some pudge and used to tell him I loved his love handles. (And in the era when I said it I had no pudge at all and in fact did not have any physical imperfection as "bad" as Pudge's.).
If I remember right, you also posted something you said was 101% true about the chances of success for males on dating apps, and I pushed back against that too, and with a concrete counterexample. It's not that there's no truth in your idea that single males have a rough time of it in the dating world, but your picture of the situation lacks nuance, and I find your all-or-nothing views about how 101% screwed men are pretty irritating.
Statistics show that women on dating apps have *far* higher response rates than men. A few anecdotes don’t prove otherwise. I do not know how this will turn out.
>A man asked his friend how he could boost his body confidence at the beach. The friend suggested putting a potato in his swim trunks. So, the next time the man went to the beach, he went out in swim trunks with a potato bulge.
To his surprise, the other beachgoers reacted with horror. Even the friend who made the suggestion turned pale when he saw him.
'My God man,' the friend shouted, 'you're supposed to put the potato in the front!'<
Since Fable is still in model-jail, I figured it might be interesting to others to share my experience using it for a large, complex software engineering task. I was working on an open source, non-commercial project, so I'm also at liberty to share all the code and transcripts from the session. If that sounds interesting, the write-up is on my substack linked from my profile, or direct link here: https://tossrock.substack.com/p/36-hours-with-fable
Not on here obviously, but when I go on any normal part of the internet, its people complaining about how AI is making computer parts more expensive, how AI datacenters are screwing up communities, how AI has basically broken education and teaching, and it made me wonder; has any other technology been like this? Where the majority of people immediately feel a surge of negative effects while all the positive effects are either hidden, reserved for a smaller cohort of people, or just anticipated but yet to materialize?
I could just be young, I was born after the invention of the TV, I barely remember life before internet. I sort of remember a panic around Smartphones, but not like this. This feels unprecedented, I wonder if people hated the internal combustion engine this much or if there's some other analogous event. I remember the review here of James Scott's book on the agrarian revolution, how its not clear that the vast majority of humanity wasn't forced into it at sword-point, and that kind of feels more like where we are at. Maybe we will forever look backward at the pre-AI days the way the first generation of hunter-gatherers forced to farm looked backward at what they had lost.
The Industrial Revolution more or less started out as you describe. Eventually, it became more obvious that the added productivity had real impacts on (for example) clothing prices. But think about it from early on the adoption curve. Big, smelly cities got bigger and smellier. Cottage industries were pulled into their orbit. Most of the modern conventions about labor rights had to be developed in real time, as did the economic theory. Marx caught on for a reason.
As I understand it, America dodged a lot of that unrest by merit of 1) a skilled labor shortage and 2) a certain unskilled labor institution. But you can see a similar pattern there. The American South felt threatened by rapid economic growth even when they could see a benefit, because wealth comes with political and social power.
To the extent that this is unprecedented, I think a large aspect is the role social and digital media plays, and the way it tends to be an amplifier for negativity. In the past people may have had issues with certain technological developments, but they weren't as empowered to spend practically 24/7 being angry about AI with other people who are angry about AI and getting selectively tailored headlines about how bad AI is.
This feels like the new "woke" to me. A way of socially signaling spurred on by groups of rage bait influencers. It feels like there's a feedback loop there and it's all performance.
I'm not saying there is zero point to the claims. But, I believe, no amount of whining is going to stop it.
“ Where the majority of people immediately feel a surge of negative effects while all the positive effects are either hidden, reserved for a smaller cohort of people, or just anticipated but yet to materialize?”
And yet where a majority of people do nothing to actually push back on the technology’s intrusion into their lives!
I do the sorts of trivial opt-out which you count as “pushing back.” When it’s even allowed. I don’t volunteer my data for training, I complain about and avoid the generative search stuff, I don’t talk to whatever bot people are using. I haven’t adopted it for my job, either, which is the one place I might manage to find some benefit.
This has approximately as much effect on the zeitgeist as installing an adblocker.
I do the sorts of trivial opt-out which you count as “pushing back.” When it’s even allowed. I don’t volunteer my data for training, I complain about and avoid the generative search stuff, I don’t talk to whatever bot people are using. I haven’t adopted it for my job, either, which is the one place I might manage to find some benefit.
This has approximately as much effect on the zeitgeist as installing an adblocker.
What are we supposed to do? Reproduce Google search from scratch? Reproduce computer parts manufacturing from scratch?
Global revolution?
The fact that things aren't bad enough to justify drastic measures (the digital world thankfully remains digital, and its enshittification predates AI anyway) doesn't disprove that they're bad.
AI has yet to intrude in my life in any significant way; just marginally more slop in some of my social-media feeds. But for people who are seeing an intrusion, what sort of pushback would you expect?
Simplistically speaking, to a high school or college system, AI offers them a faster and easier way to generate acceptable term papers for their classes, and threatens to consign them to the perpetual unemployed underclass when they graduate. They're probably going to see that as a net negative. But not using ChatGPT to generate term papers, won't move the needle on the "perpetual unemployed underclass" front, so what are they really going to do about it?
Short of going Full Luigi on Sam Altman, of course.
I don't think most people are thinking about it as deeply as you describe. AI showed up in their lives, they got a vague understanding of what it was, and (probably mostly based on what group they identify with) they internalized some kind of sentiment toward it, which they learned to parrot back in conversation and in response to pollsters.
By contrast, here's an anecdote from my life: a year or two ago my search engine of choice started showing me its AI-generated answer to my queries at the top of the results page, without me asking. I found a setting to make it so I don't get an AI answer unless I specifically click a dedicated button to get one. I avoid clicking that button in the vast majority of instances. (There are a select few types of questions where an AI-generated answer is both advantageous and scrutinizable for trustworthiness.)
This is the kind of pushback I'm talking about. Not marching with signs, not writing op eds, just being intentional about and sparing with the use of this new technology.
It's very simple and easy to do, but very few people do it. I find this fascinating.
A certain type of people spends a whole day on the settings menu of a new phone... I might even be one of them.
Yet the vast majority of people never go there and just use the defaults. Big Tech knows this and are not bothered by the minority of weirdos who get intentional instead of going with the flow.
Also, for the specific case of Google front-loading an AI response, most of us have already gotten used to just skipping over the front-loaded paid responses Google has been providing for many years, without an easy opt-out for obvious reasons. The simplest way to deal with front-loaded AI slop is to just add it to the list of things your eyes glide over without even registering.
Yes, the uncommonness of rejecting or at least scrutinizing default settings was covered in "Originals" by Adam Grant.
But what fascinates me in particular is how strong opinions about technology abound--predominate even!--simultaneously with a lack of intentionality or really doing *anything* that might be aligned with those opinions.
I see it all over the place, this cognitive dissonance. We all know social media is bad for you, everyone is on FB and Instagram and X. We all know staring at your phone all the time is a sign that your brain is turning to mush, people stare at their phones all the time even while driving. We all know AI is a slop-generator that's gunning for our jobs, even AI safety proponents like Scott keep using it and quoting it as an authoritative source of information. Etc.
Thinking about it more, I was reminded of a scene in Middlemarch where a railroad is being built through the middle of their county/region/whatever. Several people do get up in arms about it and even go so far as to march to where the railroad is under construction, but eventually its explained to them that their efforts are pointless and they go away. Thus it ever was, perhaps?
Yes, but my understanding is that the last time this was happening, during the industrial revolution, the trade-offs were obvious and almost immediate to the majority of people. We are being asked to deal with inflation and mass unemployment because.... fifteen years from now we might have vertical productivity?
Not sure the tradeoffs were obvious even then, though. Yes, now they are, and thus we (and "they system", whatever it may mean) prefer the narrative which says it was obvious, but the thing is, we have no clue how the whole thing looked like back then.
(Also see "Blood in the Machine", about the Luddites, who were actually not those mindless anti-machine mobs we all know them to be.)
Maybe it's just because you don't have access to the 1780s version of reddit, where all the gloomiest people with a lot of free time speculate about things.
Perhaps not identical, but I think the most disruptive technology that has occurred in human history that can be compared to this one, is the invention of writing.
Well I'm not sure that's what I had in mind. The invention of writing predates this fellow and I'm thinking more of the effect it had on creating larger civilizations and endowing a ruling class with power over everyone else
Grasping at straws, what would be a less popular invention than AI. Mustard gas perhaps? We can check quantitativelyish asterisk asterisk etc.
Mustard gas: 30% approve, 65% disapprove (during WWII) for -35
AI: 26% approve, 46% disapprove for -20
So fewer people are positive for AI than for mustard gas, but AI hasn't yet gotten that solid majority of people actually opposing it. Flattering for the chatbots!
A general recommendation to attend a FIFA Fan Festival if you live in a host city. I went to one yesterday, very cool, free fan event with overpriced food, some cool attractions, but the main thing is it's just so cool to sit on a lawn with a few thousand people and all watch the same giant screen and go "Oh" and "Ah" and "Oh no!" all together. Very wholesome sportsball stuff.
If you live in the best country, you can find a Fan Festival in:
-LA
-Houston
-Dallas
-Kansas City
-Atlanta
-Boston
-Philly
-Miami
If you live in the best food country, you can find a Fan Festival in:
-Guadalajara
-Mexico City
-Monterrey
If you're cursed to live in frozen commie land, you can temporarily salve your lifetime of disappointment in:
I found your point #1 in the original post, very intriguing. How do they define truth? The answer is often that "It depends".
Are eggs good for you? It depends. On your health history, your circumstances, what you enjoy...
Ancient Hindus believed there was no such thing as absolute truth. That is why they were never interested in proofs, even in math.
The Greeks were interested in iron-clad proofs, which has its own value. Knowing that a claim is always true. Like the Pythagoras theorem. True on the moon, true in a million years.
Indians had various approximations for Pi. They wanted an approximation that worked for an application they were currently pursuing. If their approximation didn't work, they'd refine pi.
Indian mathematical treatises are full of approximations. Like how to solve cubic equations with certain restrictions.
It's a very different way of thinking about knowledge, than the Greek way.
I would say the Indian approach happens to correspond to modern statistical ways. It's what GPT is doing anyway...
I like it because needing a proof is very limiting.
I realize this is what I need to change to write code. I can't seem to understand everything but need to progress in making the program work anyway. Very contingent knowledge.
A simple analogy is you don't need to understand how Excel spreadsheets work before using them.
The Indian thinking (i.e. no interest in proofs) is very well suited to modern ways of using knowledge, I think?
What are you talking about? Ancient Indians were the ones that came up with the concept of Maya. They came up with the concept of Om and Brahman and Atman. The rope looking like a snake analogy in the dark but still being a rope. Are these not eternal truth to you?
I think Indians thought of the rope-snake analogy in the following way : You operate under the assumption that it is a snake until you go closer and notice it is a rope.
If there is smoke, you assume there's a fire as a first pass.
And you're right about their saying Brahman is the ultimate reality of everything (a rough analogy, the way 0s and 1s are the ultimate reality of any software). They realized they had to stop somewhere.
For people who know more about generative AI, how much of the improvement in modern models is due to improvements in the "base" model vs preprocessing of the data, research and planning, and other more "agentic" stuff.
I downloaded and worked on a relatively small local LLM (1) and I was shocked by how much dumber it was than Claude Opus 4.8 in the browser. It wasn't just that it was dumb though, it was dumb in a way that models were dumb two years ago and aren't now. I don't know how to phrase it. Like...prompting doesn't matter with modern models. In a Claude chat, I just write and brain dump whatever I want and Claude "gets" it. Qwen doesn't. I'm not sure how much of that is just the raw Claude model being dumb and how much of it is, like, a Claude model reads in my input, rewrites my stupidity into a sensible question, recognizes it needs to go read website X, Y, and Z, and then feeds all that into the context window of the final Claude Opus 4.8 that actually writes the response.
So, like, when I'm "talking" with Claude in a chat window, is Claude so much better now because the "raw" LLM is better or because I'm actually talking with an agent that's doing a lot of post processing with my question and the underlying model is just marginally better?
some of claudes intelligence comes from the system prompts. you can try out using claude via API, then you'll get less of a systemprompt and a more "raw" experience. (you still get some system prompt, from antropic, albeit much leaner).
Afaik, one thing that makes claude much stronger than the competition, is that it is better to follow the systemprompt and reflect, which comes from antropics effort regarding alignment. (note: pure speculation).
Having said that, it's a bit of all that drives the improvements. If you want to see what a stand alone model can do, try one of the top Chinese ones (currently GLM 5.2 is t the top). If you want local, Gemma 4 or Qwen 3.6.
Yeah a loooot of people even technical people who should know better have the thinnest mental model of the capability distribution and rate of change in LLMs, even excluding frontier models.
Just looked it up. Qwen 2.5 is about a year and a half old (Jan 2025), and this was a distilled local model besides. By comparison, ChatGPT 4.5 (frontier model from Feb 2025) was the first one that reliably gave me not-terrible answers to the Kamikaze Pilot line of questioning.
I made a mobile app to answer “which of my friends is free to talk right now?” in less than 30 seconds.
It’s called Beacon! You send a beacon to a group and everyone gets pinged at the same time. The first person to answer the beacon gets connected for a 1-1 audio-only call.
(The signal drops silently for everyone else, no pressure to answer or missed call badges. There are quiet hours settings and of course you can manage inbound pings with OS-level focus modes)
It works best if join with at least four people you don't speak with as much as you'd like. I have a couple dozen connections on the app now and it feels like magic to me.
Would love feedback from ACX-ish people who still like phone calls, or wish they had more of them. You can get the iOS version on TestFlight right now here:
This is such a good idea and thank you for making this. I thought I would be the only person in the world who has this issue! Sending it to all my friends.
The only problem right now is several friends don't trust TestFlight or the app for some reason so aren't willing to get it. I will let you know how my experience goes.
Thanks! There are definitely a lot of us with this issue.
Curious if your friends would immediately trust the app if it was on the App Store instead of TestFlight, or if there is something about the app itself that would concern them even once it was on the App Store.
Happily accepting any feedback, bug reports, and feature requests :)
In honor of Father's day, here is a question for dads and others out there: what is a 'joke' you like that is amusing to you and the opposite of that to all the friends, family, co-workers and random acquaintances you inflict it on? I have a few:
Whenever my wife is working in the kitchen and happens to pick up a knife, I will scream "AHH KNIFE" and honestly it's a miracle she hasn't stabbed me yet
Whenever *anyone* comments 'Oh sorry about that' or similar, I'll immediately reply "I forgive you!", because I certainly don't want them to worry about me not forgiving their transgression (e.g., 30 seconds late joining a meeting, not bringing one of the drink orders, stabbing me with a knife)
Whenever I have a plausible reason to think somebody's face hurts, I always very calmly point it out (sometimes with a little bit of suggestive emphasis), and then don't say anything else.
Like, "Hey, does your face hurt? Because you're looking a bit sunburned."
A long time ago, I worked as an assistant editor for an Austrian fellow in Toronto, Canada. This was back in the film days before video, so at the end of an edit, we would have to send it to the lab so we could get a print, which took overnight. Every time that happened, this man would start singing, "One day my prince will come." It took me a long time to get into the spirit of it, but it still cracks me up when I think about it.
I wanted to thank you for starting this thread. I don't have anything to contribute (I'm German :P), but the responses here have been delightful to read so far.
Whenever my wife is doing tricks-for-treats with our family dog, after a couple of successful goes I say -- to the dog -- "Hey the training is going well, congrats!"
But the canonical recurring dad-joke in our household is: starting from when my youngest was maybe 5, when he tripped or whatever and landed hard I could be counted on to call from the next room, "Is the floor okay?" (That joke being a family heirloom which my younger brother and I first heard from our much-elder sibling and I don't think it was original to him.)
I over the years adapted its deployment to, I like to think, new applications. When my youngest as a grade-schooler collided with vegetation while trying to catch a thrown football he'd get, "Is the tree okay?" When wrassling with the pooch resulted in tumbling across his bed: "Is the pillow okay?" After taking an errant shot off his backside during a youth-hockey practice, me from the stands: "IS THE PUCK OKAY?" At the beach playing long-toss and the ball glances of his hand and then noggin, I waited until he'd thrown it back in, then visibly inspected it for a second and then called out, "IT'S ALRIGHT, THE BALL IS OKAY!" Etc, many other examples. To the point of for a while as a middle-schooler he'd pre-empt: "Dad don't you dare...."
Yesterday though in the car, and I don't think he was intentionally doing a Father's Day thing, came the apotheosis. My wife in adjusting the front seat fumbled the control briefly and said "Ow" from having banged her knee as the seat sprung forward. Our 14-year-old in the back seat leaned forward and deadpanned, "Is the car okay?"
You may now hum your favorite old passing-of-the-torch song. I'll need just a minute
Many, many years ago, I was playing in a D&D 3.5e game. One of the other players was playing an artificer and had just hit the level where he gets a robot companion. He was telling the GM that he'd chosen a robotic dog named "Spot".
GM: "Now you need to get a second one and name it Listen."
Me: "That only works if they're both from Prague."
"So, I want to build a Warforged character, but I'm debating whether or not I should take the mithril body upgrade. It costs a feat but it's a nice upgrade to my armor."
"Go with the all-metal design. And put some skill ranks into Craft (Alchemy)."
I once wanted to make a tanky Monk named Henry David, from the mythical Walden monastery. He'd spec to have high saving Thoreaus.
This went over about as well as me trying to explain to my GM for Pathfinder 2e that there's an arbitrage opportunity by buying all the individual components of an Adventurer's Kit separately, and therefore my character had cornered the world market and deserved a higher starting allotment of currency. I don't know if they kept that error(?) after the license revisions, but it'd make me sad if so.
To be fair, were I the GM, I'd totally let you do it... provided you could actually *do* it. You can't just tell me "I've cornered the world market" and treat it as fait accompli; that'd be like saying "oh yeah I totally killed the evil lv 17 Lich, trust me bro". But it might be fun to see how far you can get with obtaining startup capital, securing warehousing space, buying the goods in bulk, fighting off the local Thieves' Guild, committing tax evasion, hiring mercenaries to take out the competition, etc. etc. It sounds like a fun campaign premise, actually.
A character in a campaign I recently ran, did try an arbitrage on gems. I allowed him to do it, but ruled that there is a finite amount of gems in the areas he had access to.
A character concept I've come up with but haven't had a chance to play yet is a dog-man who's dual classed between Cleric and Rogue. His name is "Bandit Healer".
Oh that's good, and reminds me of something a couple players contrived of in one of my campaigns.
Using a combination of Magic Mouth and Animate Object, they would take a wood carving of a number four, which would then be named 'Number Five', and enchanted to say "I'm the number three!"
Resulting in the animated four, running through the town square with the wizard hot in pursuit yelling, 'Come back here Number Five!', while it shouted "I'm the number three!" and passerby's exclaimed 'that's the number four'. Repeat ad infinitum.
How does a tiny writer with a niche, hard-to-categorize publication become findable by the right readers on Substack without becoming annoying?
(ᕗ ͠° ਊ ͠° )ᕗ
I sometimes manage the first part at the expense of the second — as you may be discovering right now, dear reader.
⊂(◉‿◉)つ
Looking especially for advice beyond “post consistently” or “just comment more.” How do small, weird publications become legible to their natural audience without turning into a chatty billboard with thumbs saying "who's got a Substack and two thumbs? This guy!"
Say something like "I wrote a post about (topic)" and link to the post.
Maybe make a teaser. "I wrote a post about Masters of the Universe bombing at the box office despite the fact that it's arguably what people have wanted from a Masters of the Universe movie since the 80s. My theory is that Hollywood has screwed up so many times across so many reboots... Ghostbusters, Star Wars, Doctor Who, Rings of Power, Wheel of Time, the Terminator franchise... that by the time a team actually made a movie for established fans rather than the audience they wish they had instead, nobody was willing to show up. Twelve-times bitten, twice shy."
And people will click on the link if they want to read an essay about that and they will snort if they don't.
And then, next time an open thread rolls around, talk about your new post and give a nice and short teaser.
Don't say "SUBSCRIBE TO MY BLOG!"
That's annoying.
Say "I wrote an essay and it's about something that you will have an opinion about and here's the precis."
My suggestion would be to consider if you are a niche topic or a niche writer; if the former, emphasize quality and in depth to produce the definitive evergreen pieces which 'own' a point or topic, and if the latter, emphasize diversity and quantity to buy a lot of lottery tickets to discover what will surprisingly hit the spot. https://gwern.net/polish
I think "post consistently" and "just comment more" are both good advice.
Another good but often overlooked piece of advice is "produce genuinely excellent content." No, not good content, excellent. A hundred times more effort than the median post. I saw a link recently to a really detailed explanation of how mechanical watches worked, with tons of interactive animations showing the exact functions of each part. It's not a subject I'm interested in whatsoever, but just the sheer effort that went into it caught my attention, and I ended up learning a bit about a subject I never would have sought out.
Another, somewhat easier trick is "pick a fight with someone more famous than you." Bonus points if you're actually right about something, but if not, kayfabe and pointless drama always sell too. Depends how unscrupulous you're willing to be.
I recall a CGPGrey video with the same advice, but for YouTubers. He'd always see people talking about regular upload schedules, collaborations, &c, but to him all this came secondary to making content people wanted to watch.
I appreciate that advice. I've built content channels for other projects on other platforms in the past. Even more important than making things that people want to see, is working the algorithm the way you must, in order for it to even show you to enough people in the first place.
I've new'ish to Substack, and am happy to see the discovery algorithm here isn't skewed entirely away from benefiting creators and just slurping everything up for the mega-corporation.
IG for example is entirely dead for new creators unless you "work" the algorithm hard. You can spend a month putting out good daily content, and at the end of the months you'll have 30 posts, maybe 1-3 likes total, and a dozen or so views. And I'd only bother with Facebook if the plan was to advertise and pay for views.
For what it's worth, the channels I've watched and the blogs I've read for a long time have all been word-of-mouth. I found out about SlateStarCodex from a Reddit post in the Mass Effect subreddit, ACOUP from a comment on SSC or the SSC subreddit), RedLetterMedia from a CGPGrey video...
I don't remember how I found CGPGrey--that one may have been the algorithm (couldn't have been Reddit, because I found Reddit through one of his videos). If so, it's worth noting that I haven't been subscribed to CGPGrey, nor have I had a Reddit account, in years. Same goes for CinemaSins and Honest Trailers: those were for sure the algorithm, and I haven't been subscribed to either for the better part of a decade.
Anyway, all this to say word of mouth is also important, and that twisting your content to suit the algorithm might not lead to long-term gains. I've given money to the RedLetterMedia Patreon; I've never given money to anything shown to me by an algorithm.
I'd say publish less, actually. If you have some amount of effort to spend on things, use it on a small number of things that I'll be excited to see in my feed instead of a large number of things I'll be annoyed by.
That’s helpful, thanks. Do you think Substack Sections help solve the annoyance problem, or do they mostly create admin theater for the reader?
I’m curious whether readers really use the per-section/email toggles, or whether “too many emails” just gets processed as “this person posts too much” and leads to unsubscribes.
Does "hard-to-categorize" also mean "hard to describe"?
I don't know about other ACX readers, but a (short!) "sig file" attached to the bottom of each of someone's posts with some text and a link doesn't bother me. If you need a long paragraph to describe the publication then there may be issues ...
How did you guys found the competitors for this year's contest? At first I thought it was a poor batch, but I think it was just RNG messing with me, as I hit a few very good ones by the end of the deadline.
> How did you guys found the competitors for this year's contest?
I was recovering from a surgery, and so was in a place to read and rate 120 of them this batch. I'd say average quality is down - prior batches I had roughly 1.5 - 2x more 9's and 10's.
There were some truly incoherent and unreadable ones too, so the floor seemed lower too.
Going back and looking at my notes, I had eighteen 4-and-belows, and thirty six 8-and-highers (with only 2 tens and 10 nines), and the rest were between 5 and 7.
I feel the loss of very peak the most - that feeling when you're reading a great review, they've already surprised you twice and pointed out 3 interesting tangents you're going to follow later, and then they make a pivot or bring in an entirely new connection or point of view that catalyzes the whole thing, making their point and argument transformative and deep.
I pretty much live for those, and feel like we didn't have as good a showing this time as we did in some last rounds.
Still a great contest with many worthy entrants, and I'm certainly happy to have read the great majority of them. As always, kudos to Scott and his readership both, the incredibly high talent bar and selection effects are why we're all here.
I was dealing with some medical issues and didn't get to the reviews in time to vote--I went to cast my first vote on what I thought was the last day, but forgot about timezones--which is a shame, because I've run into several excellent ones.
Pleasantly surprised at the relative lack of (obvious) LLM usage, although it did ruin one particular review which would otherwise have been one of my top picks (especially aggravating because the good parts were human-written and the author clearly had the skill; the whole thing would've been improved by altogether deleting the LLM-generated paragraphs)
May have been mine :/ ran out of time and finished half the review day of the deadline. Very frustrating because I worked hard on it and if I’d had two weeks to edit I’m pretty sure it could have been a finalist, as is I’m almost certain it won’t be.
Without getting into details, the human-written parts were the author relating their personal experiences with the subject matter--interesting stories, though I found myself wanting more links to sources--while the LLM-generated portions were usually about synthesizing lessons from the stories.
I was a little disappointed until I looked at the two AI-produced reviews of the reviews. (See the comments in the post announcing the reviews and requesting reader rankings.) They explained the (as they see it) characteristics of a good ACX book review. And, looking at the reviews they recommended and looking through their eyes, I came to a greater appreciation.
Hm, I find it interesting how little it correlates with my own ratings. The first two are also exceptional in my eyes (10/10), but the other ones in the AI top 10 are only in the range of 5-8 stars for me.
That's way below my correlation with finalists in the last years. I think there were few finalists which I wouldn't rate at least as 8, and most of them as 9 or 10.
Well that’s an LLM for you. If I had a friend who somehow read all the entries and ranked them with notes, I’d probably disagree with a lot of their top ten picks as well.
I hope the AI predictions prove inaccurate, as they put my entry outside of the top 40. But there is no accounting for taste!
I thought both AIs ranked mine pretty fairly - I agree with the criticisms, and a skim through the ones it ranked definitively higher than mine showed a genuine difference in quality.
I read and voted on about ten, and it was roughly a bell curve. One bad, 4 below average, one very good, one excellent. My criteria are the things which bring me to SSC in the first place: interesting takes, solid reasoning, fun facts, good structure.
I found it still good. I had fewer mind-blowing experiences than in the first years of the competition, perhaps because those books are to some extent exhausted. But there are still many excellent reviews.
And I like the average book review a lot better than the general "everything-except-a-book" review that we had last year.
Have you seen the AI Pledge for Humanity in support of UBI as an AI dividend? I think that without this, the hatred directed at AI and data centers will only grow larger. aipledgeforhumanity.org
As of now, it seems unlikely that a UBI that would truly be distributed towards "humanity" and not just US citizens would find political support in the US. I also find it hard to imagine what form it could take.
We are mixing up the cost/benefits of a data center, which is entirely local, with the costs/benefits of AI, which is society-wide.
Do I want a building in my city that's a formless box that doesn't generate any economic activity after its built? Well, maybe, if it's generating a lot of direct revenue.
Argument by tweet isn’t all the convincing. The counter argument here is that .. it depends. On the actual tax revenue, the water availability and if the data centres are actually paying in total for grid upgrades, because the question is “does it pay enough to cover all the additional infrastructure and system costs it imposes on the grid”. And, I don’t know. And it might. It might not. A tweet isn’t going to be convincing on its own.
Not sure how the latter (i.e. that golf courses use an awful lot of water) makes anything less than it (e.g. less, but still enormous amounts of water) "negligible".
Not just golf courses. California almonds alone use up 7 times more water than all US data centers combined. Data centers make up 0.14% of our water use https://www.deseret.com/u-s-world/2026/06/21/what-environmental-costs-do-data-centers-have/ If anybody thinks the way to address our water consumption is to prevent something that takes up such small usage, and generates tax dollars (rather than receiving so many subsidies like agriculture) that could help pay for more water, you know not to take them seriously.
You’ve nevertheless engaged in a logical fallacy. As OhNoAnyway points out you compared a water user with golf courses which are very high users of water and declared that lower value negligible. You haven’t proven that grid infrastructure is helped by data centres or they if it is data centres are paying for it in full, and that’s the more pressing point.
I note pushback on social media government ID requirements is trying to get more organised / gain momentum with things like https://nevergivethemyourface.com/
To be frank, I don't know what they did differently from everyone else, but I've heard they also tuned toward their errors being false negatives (saying AI writing is human-written) instead of false positives (calling human writing AI-written). This helped them gain users, but them being generally very correct is what makes it worth using more often.
In my experience Pangram is much better than other sites. I tested it with a text of mine that was accused of being AI-written and Pangram said that it was "100% human written with high confidence" even though textdefend says it's 72% AI generated. Scott's post is also "100% human written with high confidence" on Pangram. You can also use it for free! You get 4 credits each day on the free plan.
I tried Pangram on my blog, it said 100% human, which is true.
(I expected it to be otherwise, after reading so many stories about false positives. So I am reporting a data point that in this case it worked OK.)
If someone has an account and/or patience, please try what happens when a genuine human text is *translated* by an AI to another language. I suspect that some false positives might happen there. Maybe even if you let an AI edit your text (such as "make it sound like written by a native speaker").
Ok. So maybe textdefend.com is not very accurate. I decided to try Scott's text on another free (as in doesn't ask me to register) AI scanner, quillbot.com. It said "0% AI generated". Great!
Going back to the original subject of this thread, I'm probably the target audience for nevergivethemyourface. I'm pretty paranoid when it comes to privacy and I hate the direction that internet regulation seems to be taking in many countries.
And I hate having to create a new account on a site just so I can test their service.
Explicit note for those who need it: obviously I only put this into Pangram because it read like AI to me. It will read like AI to others. Borrow phrases, don't copy sentences.
If the author doesn't find writing the post worthwhile, why should I find reading it worthwhile? A research summary I'll give you: I read AI research summaries and use them to find sources. But a public awareness campaign should be written by a person, because *the whole point* was to demonstrate that a person cared about the cause!
I can read a lot about a topic, make notes, write an outline of a lecture, and give the lecture... all of that is easy. But when I start writing a blogpost, it is slow, and I am second guessing myself after every sentence. Perhaps this gets better with a *lot* of practice, but I am not there yet.
So -- although I have never actually done this, yet -- I sometimes wonder whether it would make more sense to e.g. record my speech, and then tell an AI to make a blogpost out of it. Then I would check it for mistakes, of course. I think this should still count as a human produced text, but I can imagine that the AI editing would trigger some alarms.
You will never be as bad at writing as an LLM. If your ideas have any merit at all, you will be able to do a better job of expressing them than any computer program, and developing your writing skills further will pay off in the long run.
I get the same feeling reading comments like this as I do when my coworker casually mentions thinking about getting a facelift (she looks fine and, thankfully, my other coworkers are just as horrified). No surprise she's the same one who runs all her emails through ChatGPT now--she was capable of writing for the past few decades, but now she's helpless.
I think what matters is how closely the AI follows the speech. On one extreme, if it simply writes down what was said, that's completely okay. However, if it incorporates its own style and comes up with its own points, then it quickly gets much less acceptable imo. Basically, it depends on how much the points in this great comment apply: https://www.lesswrong.com/posts/pvGavgxY5mQcyvPQa/g-wood-s-shortform#3DKQ7W8iGCETxoJer
whatever the explanation is, I go to this blog for high quality interactions with humans. this comment section is made worse the more I have to second guess if what I'm reading is AI.
I must be getting old, because I'm more or less resigned to "age verification" taking over the mainstream internet. I only hope that when this inevitably pushes teens to shady tactics and unmoderated sites, the process teaches them some technical skills and social sense. Social media has become too easy to use, what we need is a new generation of hackers.
Years ago Robin Hanson proposed a more decentralized approach to verifying physical identity https://www.overcomingbias.com/p/reliable-private-human-identityhtml My understanding is that some places with laws are relying on a third-party site, but I don't think there's some standard protocol permitting competing sites.
There's probably a way of checking ID in anonymous way using things like zero knowledge proofs. But this is not the way it's going to be implemented, so that's moot anyway.
Jeez, that proposal sounds pretty awful. Mandatory RFID tags on all people sending to central databases (yes, technically he gives an ancap solution where there are many companies with different databases, but almost certainly there wouldn't be enough decentralization to matter) is definitely not privacy preserving, no matter what extra idea goes into it. People will change the rules and abuse the system to remove the privacy preserving parts, and all that will be left is "there is an RFID tag on every person".
I like physical IDs where there actually is a need for a method of identification. That way whoever needs the ID needs to go out of their way to ask you. If whoever is asking for your ID puts it in a database, it's probably their own special database, which law enforcement would have to individually subpoena. If you make tracking people a pain in the ass, then you could probably still catch the criminals who actually matter, while making it harder to stoop to enforcement of pettier things.
What's going to happen with $current_generation AI hardware in the next year or two?
I'm wondering if we should expect a drop of cheaper used server hardware like ram, graphics cards, and maybe even whole rack units in the next 12 months once data centers start upgrading from hopper (this year?) and Blackwell (next year?)
> My understanding is that the $current_generation − 1 and maybe even $current_generation − 2 hardware is still in use.
Oh, it's much farther back than that. A100's at 40-80gb of vram are still being used in data centers for inference, and there are ~zero on the used market compared to the absolute volume bought and deployed in data centers (~2M or so).
That's 2020, the era of Nvidia 3090 consumer GPU's, and basically 4 generations back from Rubin GPU's coming out in 2026.
Speaking as somebody who's gone the local model route to test them out, here's my own research results, if it helps:
If you want an easy plug-and-go solution, the Mac M1-M4 workstations have unified memory and ultra high GPU/CPU bandwidth that can run 70B models. An Apple Pro Ultra runs about $7k new and can run a 70B model. The earliest M1 generation goes for about $3500 used and can also run 70B models, just with lower t/s.
People will argue that this money is better spent on getting a regular desktop with two 4090's or 3090's, because you can upgrade the GPU's later, but that's more work to put together a system, and some people like Macs. My theory is you can always sell the Mac workstation and get a real GPU box if you get into it enough that you want to upgrade.
You can also run smaller than 70B models on lesser Mac hardware than Ultras - plenty of people run 7B and 13B models on Macbook Pros and Airs with decent ram (ie 32gb and up, I think max is 128gb in M4 macbook pros, and 96gb in earlier M's).
Here's some different model size benchmarks on an M2 Max for comparison you might find interesting:
If you want to go the GPU route, used server stuff a couple generations back is getting pretty cheap now - you can get EPYC 7702 boards with 128 PCIE lanes and CPU's and ram included for a couple thousand bucks now on Ebay, and can fit 6-8 GPU's in them with risers and splitters.
If you were a crypto bro and have a bunch of GPU's, you might want to look at Juicelabs' repo, which allows you to do GPU-over-IP:
The Juice Labs solution appeals to me because I have a ton of GPU's and mining board setups from an old crypto mining operation I put together - all I need to do is put a linux live usb in with the Juice server / client package and I should be able to do a lot with the hardware I have already.
In terms of multi-gpu setups here's the order of things that matter:
> First GPU VRAM (more the better)
> 2nd GPU VRAM (more the better, ideally NVLINK to 1st GPU)
> Ram GB / speed (there are arguments that 2 48gb sticks are actually better at 96GB total vs 4 at 192GB total, because the 2 can run at full DDR5 or DDR4 speed, versus your speed gets discounted in almost all mobos with more than 2 sticks of ram)
> CPU / Ram / GPU bus' bandwidth in gb / s - where M1/M2 macs really shine, with huge bus speeds up to 800 gb / s for the Ultra M2
> CPU speed
> PCIE speed
When you have multi-gpu's doing inference the main talking they do back and forth is sending hidden weight updates to each other, which is actually a very small amount of bandwidth, so PCIE bandwidth / speed doesn't end up mattering much compared to VRAM and ram.
Dang, thanks for sharing! I need to digest all this info--I just started digging into this stuff after realizing it would be nice to run a model at home and try to get some simple automation set up that is a lot more private than sending over all my data to one of the cloud operators. Not sure how much utility I'll get out of using it but evening setting it up would be a nice exercise in home labbing
usually datacenters swap their hardware, because old hardware requires more electric power per processing, i.e. it's not worth to pay for power for old hardware, when newer hardware exists. iirc they used to update hardware about once per 4 yearss. That was in the past, where 1 FLOP from some CPU was equal to 1 FLOP from some other CPU. With AI loads, this changes, because some GPU have architecture that is much better suited to some models, than other.
So in addition to the electric power argument from the past, datacenters now have an other incentive to update. So expect more hardware updates and not less
> I was hoping to nab some server grade hardware on the cheap
When I nabbed old hardware in the past, it always was via inofficial channels (i.e. the person in charge of recycling gave a hint)
edit: i never got free hardware from datacenters, only from offices. I don't know what datacenters do with their old hardware
Where compute per se is the bottleneck (i.e., not power or rack space), I'd guess the older GPUs et al. would be used for smaller models (e.g., Haiku) in order to reserve the newer hardware for the largest ones.
Alternatively, prioritize higher-margin API traffic for the fastest hardware and let the subsidized subscription users make do.
Where are people hosting and reading medium-length web fiction these days?
Some writers I know of post fiction on Substack, but it really doesn't seem suited for it, especially stories longer than 10k words or so. I'm also not too keen on posting fiction outside of narrow genres on my main blog, or creating a separate Substack. RoyalRoad seems to be geared for very long-form fiction.
You shouldn't copy my example, but I write my web serial on schlaugh.com. That gets you absolutely no audience (schlaugh is absurdly tiny for a social media site), but one thing I really like about it is that my posts are all scheduled to go up at a particular time of day and I don't even need to configure that - it's just inherent to the schlaugh platform, which only shows new content (including private messages, profile updates, etc) once a day tops.
I am posting on Substack. I tried Royal Road, Wattpad, and Vocal, and found no readers and no engagement. On Substack, I was able to build a readership. My work includes short stories, novellas, and full-length novels that I serialize. Substack is not super-well-suited for serializing - you have to do a lot of the work yourself, like the navigation and landing pages. Still, it seems like the place to be if you want your work to be seen.
yes, and substack bounces me off it for any extended reading due to not playing nice with dark mode settings and not having its own dark mode version. I tried to read some of your stuff but couldn't stare at the bright white background for long enough to concentrate!
i can handle it for short articles, or here at ACX its like a pleasant light blue background which is more tolerable, but the burn your eyes out white background with black text is a no go for me. I know you can change the colors in the code on your browser but that's way too much effort lol
Related: I really want to read a street wizard/hacker/conman-esque cyberpunk story, but it's about implausibly skilled prompt engineers expertly manipulating the low-tier-AGI overlords in a future surveillance state. You people know the genre I'm getting at. Does this exist, or do I need to write it myself?
If we have the technology to allow adults to change their body type then we probably have the technology to treat back and joint problems a whole lot better.
If I can grow my bones longer (or shorter!) then I can probably repair my spine.
I guess broadly at the centuries-scale you'd expect those medical techniques to improve in parallel. But on the ground year by year, progress seems to be pretty uneven and isolated between disciplines. Today we're still prescribing physical therapy, cortisone injections and finally things like spinal fusions when people have chronic problems that less-invasive treatments don't resolve, which have tradeoffs and aren't 100% reliable. This is a discussion about sci-fi world anyway though, so I suppose we're free to speculate however we like.
I'll bet that will stop being fun really quick for most people -- the males choosing basketball player, the females choosing the rhythmic gymnast look. Right now, people who have one of those favored body conformations are rare and envied. When 75% of us look that way they won't be. On the other hand, there will be a few people choosing to look like Truman Capote, or a red-headed Eleanor Roosebelt, or just weird -- coal black skin, yellow hair, lavender eyes, extra-large hands . . .-- and that's kind of fun.
Just in case the autofilter is triggered by someone replying to the original post, here I am replying to this original post. If that "works", I'll reply to all the others so we won't have a stream of context-free slurs and expletives cluttering up the thread.
What, did the AI bot infer that you were trying to get it in trouble and so prudently refuse to moderate you? A win for automated moderation, a lose for AI alignment....
I will be in the Bay for the AI Engineer conference tomorrow - friday. If you're also attending, reach out!
Okay.
You know the thing where you're falling asleep and right before you transition from awake to dreams starting to flicker, you have that epiphany that has been evading you for the last year or so?
It just comes up and says "*HERE* it is" and everything clicks and that thing and that other thing and that other other thing suddenly snap into place and now you know how to think about something when, before, you kept grabbing at the wrong thing?
That happened to me last night.
I wrote an essay today.
https://ordinary-times.com/2026/06/28/reflections-on-converting-to-atheism/
Sorry to be a nitpicker but is Babtist some kind of niche branch if Christianity or did you misspell Baptist a few times?
Ah, you must be a Northern Baptist.
I'm not a Baptist at all, I was just curious if that was a legitimate alternative spelling. I've never encountered it before, it was unlikely to be a typo as it happened at least twice, and it seemed unlikely to be a misspelling because your writing is that of a literate person. At the same time, a Google search for "babtist" just returns search results for "baptist" churches.
Around Southern Babtists, you'll find that that's generally how they pronounce their denomination.
It's a deliberate transliteration.
Now I want to try https://en.wikipedia.org/wiki/DragonRaid :D
It's less fun than it sounds.
(But you're supposed to have the Bible Study Group Leader be the DM and we didn't do that so my experience probably shouldn't be seen as representative.)
Either the Group Leader, or the drunkest guy you can find.
I broke 2 fingers in my left hand last night. Yeah another bar fight. Just kidding. Hopefully those days are behind me. Actually I just got off my mountain bike while it was still moving again. Pretty sure I’ll still be trail riding for a long while but I’m gonna try to reduce the involuntary ejections.
At Urgent Care my fingers were so swollen that they thought they would have to cut my wedding band off but they had a couple tricks to try before they went that route. Put a tourniquet on my bicep and wrapped my ring finger tightly with tape for a half hour to squeeze the blood out of the finger. Then the doc looped a piece of strong fabric through the ring and kneeled on the exam table to put his back into pulling the sumbitch off. Took a couple minutes but it came off after a little wrestling. Now I don’t have to have a jeweler weld my band back together.
As a bonus, with a plaster splint on my hand and forearm and the arm in a sling Mrs Slothrop is outside mowing the lawn now and I’m just laying on the couch tapping out this comment with my right thumb.
Edit
Oh yeah, a little codeine makes me a bit giddy and I tend to run off at the mouth or I guess at the thumb under the influence.
I ever mention my dog, Buck here? Golden retriever, great dog. One time.. Oh, I’m doing it again. Turning my phone off now.
This is a good reminder to use silicone wedding rings for everyday use. They are like 5 bucks for a 10 pack on amazon.
I am ambivalent about reading stories or watching film series that are not finished yet.
On one hand, I understand why publishing chapter by chapter is great for the authors. You can get feedback and maybe even some money on the way. The feedback tells you whether to keep going or drop the project. The more money you get, the more you can arrange your life about writing more and worrying about other sources of income less. This is ultimately also good for the readers, because they get stories they otherwise would get years later, or not at all.
On the other hand, I hate reading one chapter a week or more, or whatever the publishing schedule is. I am a fast reader. If I have free time, I want a ton of text now. I will stop when I run out of free time, or when I finish the story (or if I decide it is not worth reading). Afterwards, reading one chapter a week or one chapter a month feels like torture -- I wish I could forget about the entire story, and then remember it again a few years later, after another substantial amount of text was written. But then, after a few years, I will have to re-read the introduction again, because I will likely forget it. So my wish is to only find out about the story when it is complete.
I guess if I put these together, my preferred outcome is that other people provide encouragement and support to the authors, and I only find out about the story after it is completed.
I hate that some stories never get finished. A Song of Ice and Fire is "forthcoming" since 2011. The War Against the Chtorr is "forthcoming" since 1992. At least be honest, guys; you are not going to do this, ever; you only wait for your death as an excuse.
Another problem is that paying as you write creates an incentive to keep writing and never finish, even if you already have nothing more to say. I have seen stories where Book 1 is full of new exciting things and problems, and a longer Book 2 is just the protagonists walking around for no reason other than to generate even more text.
With film series, I think the pressure to enshittify is even stronger. At long as the story remains popular, it remains open-ended because there is yet another season to make. When it stops being popular, it is dropped in the middle of a word. Thus it is almost guaranteed that the story will never have a satisfying ending -- there is always either an incentive to keep dragging the audience, or no incentive to give a fuck anymore. This can even happen multiple times (Babylon 5, Heroes), a series can be dragged when it seems profitable, canceled when if seems not, restarted when they realize they have underestimated its popularity, then canceled again, and in no point in this process the viewer gets an ending to the story.
A smaller version of this is simple computer games. Before internet, you got a few levels, and then an end. If you liked the game, you recommended it to your friends, so there was an incentive to make it good, and to give it a nice ending. Online games, it was just how many levels you keep playing so that they can keep showing you ads. There is no incentive for an online game to have an end, that's just leaving money on the table. You do not leave an online game because you have completed it to a satisfying ending; you leave because you got bored.
Now of course not everything is like this. There are still books, film series, and computer games, that are planned since start to do something and then have an ending that makes the customer happy with the overall experience. But the temptation to keep milking the customer and then leave them frustrated exists, and it makes me suspicious that whenever I start reading a book or watching a film series that is not finished yet, I am volunteering for a frustrating outcome.
Related: Larry Correia argues that GRRM's very conspicuous failure to conclude ASoIaF has effectively killed epic fantasy as a genre for new writers, with only established veterans like him and Brandon Sanderson being able to make money on new epic fantasy series. He also suggests that the rise of LitRPG and Romantasy have been driven by readers and writers fleeing to fantasy-adjacent genres farther from the stink of GRRM.
https:/x.com/monsterhunter45/status/2070508567531319413#m
For me, it's fine when each book (or movie) in the series is a reasonable story in its own right, like e.g. early Harry Dresden, early Honor Harrington, or most of the Vorkosigan Saga. The obvious problem, given the "early" qualifiers there, is that those can turn into series that won't be satisfactory until complete. Weber did eventually finish the core HH series, and I had a very palpable sense of relief. "Finally, I never have to read anything by David Weber ever again!". Fingers still crossed on Butcher.
I'm also generally willing to trust an author who proposes to write A Trilogy. I've been burned a few times by getting only two of the three books, but that's rare, and when there's a fourth book it's usually a nice-to-have add-on rather than a mandatory extension if you want to see the core story end.
My current concern is webcomics, which are necessarily a serial format, and I'm currently following four for which it's pretty clear that the author has decided this is the best gig they will ever have so they'd better milk it for all it's worth. Though both "Schlock Mercenary" and "Skin Horse" did come to actual conclusions after very extended runs, so there is hope.
Thirty years ago, there was a Czech publisher (I forgot the name) who killed a few translated book series in the middle. Like, he translated parts 1 to 5, decided that the part 5 didn't sell as well as he expected, so he never translated parts 6 and 7. What was worse, if I understand it correctly, translators typically buy exclusive translation rights for the entire package, for some time limit such as 10 years. So when he decided to kill the series in the middle, no one else was legally allowed to continue where he stopped. That has pissed off many fans, and it became common knowledge among the hardcode fans that "if it's published by X, do not bother buying the books, because the series will never finish".
Speaking of webcomics, there is a comic version of "Worth the Candle", misleadingly labeled "Completed" https://www.webtoons.com/en/action/worth-the-candle/list?title_no=7061 but actually canceled at the moment the party pays a visit to the soul mage (about 1/4 of the story?). Wrong labeling adds insult to the injury. Now I can't even trust things labeled "completed" to actually be completed, wtf!
My favorite writer is currently Dave Duncan https://en.wikipedia.org/wiki/Dave_Duncan_(writer) ; I am reading his books in the order on the Wikipedia page. "The Seventh Sword" is a trilogy (finished), with one bonus book in the same world. "A Man of His Word" is a tetralogy (finished), followed by another tetralogy (finished) in the same world one generation later.
As a Discovery writer, I shudder at the idea of publishing something a chapter at a time. It means you can't rewrite the chapters when a better idea comes along.
Part of the reason Dune is so good is, any of the first four books will serve as an end to the series. (5 doesn't, but by 5 the setting was also getting pretty long in the tooth.)
So I'm into book 4 of the "Dungeon Crawler Carl", series, Matt Dinniman. (thanks again to my daughter, who's in book 5.) It's all over the NYT best seller list. And yet it's not even nominated for a Hugo in the US. Which causes me to lose a little faith in Hugo's. (And the writing and story telling continue to improve at least into book 4. Recommended.)
Be warned that the writing starts to gradually drop in quality starting somewhere around book 5, though I think the characters, setting, and story generally continue to be compelling. I started to see more proofreading errors, repetitive phrasing, etc around that point. I'm inclined to forgive those minor sins-- he is writing these books very quickly and I find the whole project pretty compelling despite its flaws.
How long a read is this? I’m partial to books that take a couple rainy weekends to read. Or I guess any couple weekends in February where I live.
Hmm well book four is 578 pages. My daughter's book 5 is ~700 pages. Not sure about the others but it's easy to check. I'm someone who played a lot of RPG games, mostly on the computer but also D&D with a group.
Hugo nominations are done by Worldcon members. IME the average age tends towards the older end compared to other events I attend, and this may affect how quickly reading preferences shift compared to the general population.
Be the change you want to see! - join up for a Worldcon and nominate whatever you like! 2027 will be in Montreal, which I've never been to; good excuse to visit, so I'll be going.
I won't be nominating DCC for any categories, though. I did enjoy it, though I've not picked up the latest book yet; but you only get to nominate so many things, and although I've not yet looked at what will and won't be eligible for the various 2027 categories and likely won't for another 6 months or so yet, I've definitely read better LitRPG than that.
I do look forward to a non-traditionally published web serial getting a Hugo. I think it's only a matter of time - Worldcon membership tastes aren't isolated from general fandom shifts, they just have a bit of hysteresis.
Thanks. I don't need to join. I don't care that much what is nominated. I just want more 'good' book recommendations. And in that regard, LitRPG do you find better than DCC? Again Thx.
Tastes do vary, of course; but fwiw I am currently enjoying Super Supportive, Runebound Professor and Elydes enough to support them on Patreon. Other things I've recently enjoyed have been Sokaiseva, Live at the Forum (closest of everything thematically to DCC, don't know if I'd actually count it as /better/ so far but I am only a little way through the backlog), Dalliance Rather,
My favourites of all time are probably Super Supportive, Demesne, Mother of Learning.
Thanks, sure we all have different tastes.
Oh, this is all on Royal Road? That's cool. I've read a little bit there.
I can second Mother of Learning, it's a gem (though its litrpg status is debatable imo). https://www.royalroad.com/fiction/21220/mother-of-learning
Based on the blog you are reading, you'll probably also like Worth the Candle. https://www.royalroad.com/fiction/25137/worth-the-candle
Thanks. I will say as an old fart I'm less interested in the teen age story, though I totally loved that in the past. Que the Heinlein juveniles.
You know, it’s been a while since I read Stranger in a Strange Land. That would be fun to go back to.
This years book would have to be nominated. I think there’s a slight bias against that specific genre - LitRPG
When I asked AI why the books hadn't been nominated the first answer was that LitRPG wasn't 'high brow' enough for the sci-fi community. And I thought, when did sci-fi become high brow? I mean they nominated Harry Potter.
I sent this directly to Scott, but I'm posting it here as well:
The office of management and budget is seeking public comment before they change the rules on grantmaking. I learned about this from someone close to me who works for the dept of Education. Admittedly, I'm not an expert in any of this, but my understanding is that the way it currently works, the Department of Education is constantly running grant reviews for specific research and funding (they used to run a lot more, but DOGE basically forced a multi-year moratorium from which they are only just now recovering). People submit their applications for federal funding to grants that fit their general field.
During an actual grant contest, the subdepartment picks a panel of academics from volunteers who they believe are qualified to evaluate the grant proposals in question, and vets them for any conflicts of interest. PART OF THIS PROCESS IS CHECKING FOR PARTISAN BIAS ON BEHALF OF THE ACADEMICS, I emphasize that part because the narrative that the federal bureaucracy is captured by woke elites is one they are well aware of, having been sued for bias 10,000 times by conservative lawyers, and having lived through multiple transitions between different administrations whose first step is to scrutinize every funding authorized during the prior one. As you might imagine, in the last two years they have been particularly sensitive about this. I've overheard more conversations about ensuring that ideologues don't taint the neutral process of funding that I thought I would hear in my entire life, by people who have clearly been thinking about this for well over a decade.
Anyway, now the OMB wants to change the federal regulations around the law, so that after the academics have been chosen, after they have weighed in with comments about what projects would be of immediate benefit and what ones need specific tweaks to their methodology, the Trump administration would like to add a step where a political appointee gets the final say. You can read the language of the proposal yourself, and evaluate it yourself, but I think you will come to the conclusion that the obvious intent is to make sure that otherwise good research that might run counter to the ideological aims of the current administration die on the vine.
>the obvious intent is to make sure that otherwise good research that might run counter to the ideological aims of the current administration die on the vine.
Its interesting seeing the different intuitions people have wrt this... I would not at all be surprised if both parties had a veto in my countries version of this process, and I wouldnt see it as a problem.
The political appointees in charge of the large government bodies in the United States are always chosen by the current president. That's why traditionally there are safeguards in place to make sure that the majority of the work is done by neutral parties and professional civil servants. Under this new plan, the people Trump picked would have immediate fiat over all funding. There are ideological concerns and corruption concerns.
Problem is, many of us would not be surprised at all if Trump administration nominated an antivaxer or a flat earther to this position. A position of being able to veto scientific research all over the country, if I understand it correctly.
That said, this is democracy, and the will of the American people is to burn their country down (to spite the libs or something like that), so who am I to comment on that.
I believe this might be helpful - a summary of the most damaging changes and what they mean for research: https://elizabethginexi.substack.com/p/summary-of-key-changes-in-ombs-proposed
I'll summarise with the post's own summing up: "Since World War II, the United States built the world’s preeminent scientific enterprise on a straightforward principle: federal dollars should fund the best science, as determined by independent experts rather than politicians. Peer review, open competition, and institutional autonomy were the pillars of that system. This proposed rule dismantles all three, simultaneously, government-wide, and binding on every federal agency by October 1, 2026."
(Please do let me know if this is a different set of changes or somehow not applicable to your post; apologies if so).
I have recently become aware of "AI ick", which I understand to be a deep, visceral aversion to AI usage or even artifacts produced by AI.
Example: https://lucumr.pocoo.org/2026/6/6/communities-of-not/ the footnote leads to a GitHub issue on an extremely popular open source tool that has merged some AI generated code and triggered an aggressive reaction from some users.
I understand weariness about a new technology like AI. But I'm at a loss when it comes to such strong reactions. Is this a personal thing? Or is this simulacrum levels 2-4, where there's a deeper, tribal alignment thing going on?
I'm seeing a lot of comments on this thread trying to claim that this is about tribes and making analogies to old DEI culture. Those comments seem to me to be deeply confused.
I get very strong ick for any unlabelled AI that is directly consumed (the relevant difference for me is whether or not someone is directly taking an AI output and showing it to someone, vs using it as e.g. research aide). I don't consider myself to be among the anticapitalist redditors, or even anti tech (I work at a tech company!). Neither do most of the people I talk to who get AI ick.
The only real analogy between this and the DEI debates around the 2020s seems to be the use of public shaming. Tribal alignment is definitely the wrong way to look at it, at least for now.
The github example seems to be people complaining about AI introducing a bug, not about AI usage in general. So it’s not a good example for what you’re talking about.
On social media there's a thing I've seen a few times, where one person says "here's a thing I made, do you like it?" and person replies "you used AI, therefore it is awful, I hate you and I hate it and I will tear you down as much as I possibly can."
This happens even for high-effort things that used small amounts of AI.
I think it's the standard social-justice-warriors problem in a new field: there are some people who really like having an excuse to be mean to people.
"Those who disagree with me are bad people who have no good reason for their actions" is a supremely uncharitable explanation. This doesn't mean that it's wrong (in fact, I think it's true for at least some people), but one should use it only as a last resort if one wishes to find truth instead of merely satisfying tribal urges.
Asking from inside that frame: Why did all those sadistic people coordinate on hating AI usage? Why not, for exmaple, apostrophe usage? However you answer that question might provide an answer outside of the frame too! Maybe even the people who are not sadistic have valid reasons for disliking some AI usage.
In the initial post, Matto mentions "weariness about new technology". I'm about 90% confident that this is a typo or malapropism for "wariness" (apprehension), but I think "weariness" (fatigue) is actually a big part of what's going on here.
It's a similar dynamic to pop culture fads. Some musical genre or media franchise or whatever seems to be everywhere you look for a while, which if you are an enjoyer of the thing is all well and good, but if you dislike it for whatever reason (or even if you're merely indifferent to it at first) then repeated exposure to it becomes more and more annoying and the dislikers gradually become outspoken haters. In the bit Matto's original post referenced, the "Oh for fuck's sakes" reaction is pretty much the same energy I would expect if somebody had checked in a bunch of gratuitous Twilight or Nickelback references to an open-source codebase and broken some stuff in the process.
Appreciate the point about weariness and wariness 🙏
The most recent example I've encountered: https://x.com/hsrmemes_/status/2069116085530419403 is a fairly-high-effort thing that I actually quite liked, which, yes, was definitely made with AI. The comments section has some raw praise ("Castorice was sick as fuck," "I absolutely loved the Phainon one"), quite a bit of nitpicking ("Aglaea doesn't wear gloves..."), and a few people being flat-out mean, eg https://x.com/KuritoPoroto/status/2069136248740946405 "Your AI slop bores me".
I think the people in that last category are saying: "AI is bad for whatever reason, therefore I should attempt to effect social change by being mean to people who interact with AI in any way, and if I'm maximally mean then it will effect the maximal amount of social change." I also, yes, think these are people who enjoy being mean to other people.
I'll take a swing at this!
I think they coordinated on AI usage because it's becoming a culture war front where choosing a side gives you air cover. You can do awful things to people, claim it's to "fight the ai invasion" and get away with things you otherwise wouldn't.
We're talking about a tiny, tiny minority here, that nevertheless inflicts vast damage to social fabrics--theyre Pareto optimal probably!
I only bring this up because the anti-AI crowd I've been exposed to also appears to lean anti-capitalis in the same way many of the DEI folks used to, the folks I've seen first hand apply this tactic in two tech companies I've worked at.
Again, it's a tiny minority, but part of the pattern is that the broader group they're part of doesn't do any policing on they're obviously bad faith actions.
"I think it's the standard social-justice-warriors problem in a new field: there are some people who really like having an excuse to be mean to people."
I think that's part of it, but not nearly all of it. I think there are a lot of people who just developed a large amount of negative affect around the whole concept of AI[1] and knowing something includes AI really does just ruin their enjoyment of it and make them see red.
And I honestly get this reaction, to some degree, even though I think it's overblown. Because it's not just "weariness over a new technology" like Matto suggests. The AI explosion managed to hit all the pre-existing beats ambivalence towards capitalist consumer culture and marketing/tech sector hype and frustration at continuous enshittification and economic/political anxiety all rolled into a package that is very often unbearably cringe.
See, modern machine learning tools are getting better and becoming more useful, but there was a HUGE push to incorporate them into absolutely everything even when they were obviously unready, and an extremely poorly-concealed eagerness by various corporate CEOs to replace their expensive, messy human workforce with cheap, compliant machines and also quite a lot of average Joe humans eagerly churning out and spreading and talking up AI slop[2]. And if you try to put yourself in the shoes of someone who didn't find any of the recent AI developments especially interesting, and didn't ever seek out AI tools or AI content by choice, those mostly highly negative things are their main experience of AI. So it's small wonder they've built up so much negative affect around the concept.
But yes, also some people really just like having an excuse to be mean to people, (though that's hardly exclusive to social justice warrior types).
[1] More specifically, generative AI, like LLMs and AI art programs.
[2] And while I definitely don't think every use of AI for art or writing can be dismissed as "slop," a large fraction of the people most eager to use them for that are people who lack talent for it themselves. Lacking talent tends to correlate (though not perfectly) with lacking taste. So a lot of the people sharing AI art are consistently and very enthusiastic sharing extremely *bad* AI art, while by all appearances thinking it's amazing. I'm reminded
Which specific strong reactions are you asking about? The open source tool introduced a bug, which is important context.
Both the linked threads at the bottom of that blog post (GitHub and Mastodon) have both reasonable takes:
> Regardless of AI as well, the project should really take care before pushing new features, and especially bugfixes (these issues only started cropping up on the last two patch releases, which shouldn't change existing functionality)
> Imo at least, producing slop is a human error caused by misunderstanding what LLMs are and what they can actually do
And complete garbage:
> BTW don't forget OpenAI assassinated a guy for wanting to testify against them. Package sabotage by corpos is not a conspiracy at this point lol (I am a developer, in case someone wants to say im just a random idiot but this is my anon issue account)
Specific instances will have different motivations. I think modernity has a distinctive epistemic failure mode where people want to talk about
*gestures around vaguely* you know what I mean! All the stuff about the thing!
Then disconnect from specific reality to shadowbox against phantoms.
Good point. I should have been more specific.
The specific strong reaction that inspired me to ask is the original post on the GH issue: https://github.com/RsyncProject/rsync/issues/929#issue-4554389521
On the surface level, posting a screenshot of a social media post with an aggressive tone is low quality content for the medium (a project issue tracker). My own understanding is that gesturing at AI and ending the post with "for fucks sake" is a bit of a dog whistle for mobbing a project/author. Both seem to me as fairly strong, aggressive actions to take against an oss project.
Perhaps this feels so antagonizing to me because I work with software where updates are usually handled in a rolling way or at least go through manual validation in a staging environment. So the original screenshotted poster seems just as responsible for breaking their system as the LLM that introduced the bugs.
Breaking changes are a fact of life, even more so now with supply-chain attacks. The amount of vitriol projected at the rsync project because of a LLM-made bug seems disproportionately large given how software world works.
Yeah, I agree that that issue is very low quality.
To me, the Mastodon comment screenshot was normal social media stuff. On about the same level as things like "somebody cut me off in traffic today, I hope they get into an accident haha". I don't love that it's normalized, but I consider it a pathology of social media more than of anti-LLM people.
Obviously it's stupid and pointless to copy that into a GitHub issue, but to me it seems like an unaware individual treating GitHub like Reddit or Twitter (where it's an accepted norm to just repost screencaps you agree with).
IMO the reason why it seems excessive is because two small but common bad-nesses combined in an unfortunate way. The rhetorical temperature of the Mastodon post (ex: "for fucks sake") wasn't intended for a GitHub comment, and is par for the course where it was posted. I think it's unlikely it was an intentional call to mob the dev. Mindlessly reposting screenshots is unusual on GitHub specifically, but is ubiquitous in more social spaces. It looks like the person who posted the issue mostly contributes to video game mods, and this is the first issue they opened.
I kind of despise that it's "normal" to throw around decontextualized, vaguely articulated, high temperature statements. Nobody feels the need to "own" what they spread. But that's a problem in lots of places, and not just anti AI circles.
That's a good take. I've been off GH for a longer time now, and off most social media, so it helps calibrate against what is the baseline behavior to expect.
For me, it's a combination of a few things:
1. AI-generated output tends to converge on a few specific styles, most of which I find grating to begin with and their association with the other things that annoy me about AI makes it worse.
2. I've been dealing with a big top-down push for using AI tooling at work. The tools have some usefulness, but their output is decidedly worse than human-written code and is a lot more work to review and debug, and use of LLM tools to analyze bugs seems to fairly often lead people into dangerous misunderstandings of what's going on.
3. More generally, I have a fairly low assessment of the quality and reliability of AI output. I generally do not consume it by choice. Presenting it to me anyway gets tiresome.
I understand your 1st point as it aligns with my own experience. It's a bit like interacting with automated or even human call center phone calls. Exasperating on a good day.
I am surprised by your 3rd point. I've found AI to be a meaningful improvement in my work. Not 10x but somewhere between 1.1-1.4x depending on task/domain. I use it both for getting an overview of a new to me part of the codebase as well as for generating code. I do always dona substantial review on both fronts, but I find the speed, in this case, to come from getting things almost right within the context of the codebase. We also have a ton of test across the whole pyramid which helps catch small mistakes and being the quality bar higher I suppos.
This is fascinating because I wonder where the difference in perceived quality comes from.
I suspect it's related to how well-trodden the problem domain and codebase architecture are in the model's training set. I've found the tools to be at their best producing boilerplate and at their worst trying to work with complex sui generis issues. In boilerplate-heavy architectures, AI tooling is probably very useful. Ours is the opposite end of the spectrum, partly because we've architected it to minimize boilerplate and partly because the code base has evolved a lot of clever bits over the years for a variety of reasons, some good and some bad.
Where it goes badly off the rails tends to be in trying to work with the clever bits. I've seen people get in a lot of trouble trying to use Claude or Cursor to understand new-to-them parts of the codebase and it makes a fundamental mistake in one of the load-bearing assumptions of the analysis. These mistakes tend to be subtle enough and presented confidently enough that if you need the explanation, then you aren't in a position to notice the error until you try to act on them and things don't work the way you expected. I notice them all the time because I've an expert on several of the clever bits and get tagged in on discussions when people notice that things have gone badly off the rails, or because I'm troubleshooting some stability or perf issue downstream of the error.
This is similar to my own experience. AI is very useful on tasks I describe as "advanced copy-and-paste", which includes boilerplate code, but also things like "do X, following the pattern we used for Y". But when dealing with complex problems, it constantly makes incorrect assumptions. Usually they are plausible, and junior devs are happy to trust it and make a mess of things. E.g., summarized from a recent conversation.
> Me: Why are we adding 2000 lines of code to split files into pieces?
> Junior Dev: Claude says we ran out of memory after reading a file, so we need to read it in smaller chunks.
> Me: What file?
> Junior Dev: This one: <link to file>
> Me: And how big is that file?
> Junior Dev: 4 kb
> Me: So do you really think that's what caused us to run out of memory?
> Junior Dev: I guess not
And then sometimes it's not plausible *at all*, and people _still_ take it at face value. Another conversation:
> CTO (who is leading an aggressive, top-down push for AI tools, to the point of mandating that we use them for all work and not hand-write any code): My AI agent found the issue that caused our increased error rate last week. Please have the team make the appropriate fixes.
"AI Triage Summary: An increase in the concurrency from 5 to 10 led to increased load on AWS S3, which produced increased error rates. The solution is to reduce the load on S3."
> Me: I'm pretty sure that we didn't bring down AWS because we started hitting S3 another 5 times per minute.
I've seen plenty of junior developers with that sort of pattern entirely without AI - just assume their first theory absolutely is the actual cause of whatever problem they're investigating and spend hours or days running with that down a rabbithole instead of just doing a quick set of tests to systematically pin things down and prove/disprove theories first.
I find that Claude becomes much more useful for debugging if you get into a habit of interrupting and telling it "before proceeding any further, please write and run a quick test to prove this theory" whenever it confidently states the cause of X is definitely Y.
Sadly, no amount of CLAUDE.md, memory etc seems capable of creating a behaviour pattern where it does this automatically /before/ boldly asserting things to me.
Meanwhile, I remain utterly unable to convince a subset of my circle to consider AI output critically, even when they keep getting burned by it. I can only sigh at time wasted.
TLDR: treat Claude like you would a lazy junior developer with a tendency to pattern-match on long rabbitholes.
But that's sort of the issue, at least for me. People pushing AI are saying things like (to quote our CTO) "AI coding tools will give us infinite capacity – if they don't, then we aren't using them correctly." But if using them correctly requires constantly monitoring them and correcting them when they start to go wrong, then they aren't going to produce anything like the improvement people keep insisting they must. In other words, nobody would ever say "If we hire an innumerable army of lazy junior developers, we'll have infinite capacity".
I had a similar experience yesterday. My current hobby project is to analyze old DOS games, reverse engineer the code, extract resources and passwords, that kind of stuff.
Here, AI is very useful. It sees something like "mov ax, 40", "mov bx, 2", "int 21" (possibly separated by a few lines) and immediately says "hey, this prints an error message" while I would have to consult my notes. Also it recognizes many kinds of patterns instantly, like "this is so-and-so method of encryption", "this draws a 16-color image in 4 bit planes" etc. But most impressively, at the beginning it gets a ton of freshly disassembled code, and it already sees large parts of the big picture, which parts of the program are related to what. And these instant guesses are like 90% correct, which sometimes saves days of work.
But when it makes a wrong guess, there is almost no way to convince it otherwise. I just take the original code, add a note saying "this part does X", close the session and start a new one. And the new instance is happy to exclaim that the part obviously does X and continue from that premise.
So, yesterday there was a game that used some kind of copyright protection, checking for a damaged piece of floppy drive or something. We got past that point, and the AI was like "this is another copy protection check", "and another", "wow, this game has so many sophisticated protections". Except it did not, we were long past the check, and now the game was loading resources. But the AI couldn't be convinced otherwise. So we checked the rest of the code (a few hours of work) and then I was like "damn, so where does the game load the resources?" and the AI kept insisting that it does not. -- New session, I didn't mention copy protection at all, just asked "where does the game load resources", and of course it showed me the part of the program that we previously skipped, because it was "just more copy protection".
The problem is that even if AI is right in 90% cases in some subject, it always expresses 100% certainty, and prints pages of convincing arguments why it is so. So you must accept that sometimes it is a lottery; on Monday, AI does in 5 minutes what would take you entire day; on Tuesday, you waste the entire day being misled by an AI hallucination. If you are lucky, you still save some time on average, and you may actually enjoy the process, especially if you start developing an intuition for when the AI is hallucinating.
I've been getting Claude to admit when it doesn't know stuff. Mostly by asking for sources and evidence (and what evidence we would need to falsify an assertion; both assertions I made and assertions the AI made.)
Fascinating! The system I work on is bespoke (in-house telemetry system), so there's little boilerplate and lots of custom libraries spread across a handful of microservices combined with a very bespoke orchestration system. I do feel like I optimize a lot for clearly describing system boundaries and relationships between components, along with strict constraints because of this. In a way, it's all about managing memory buffers, so there's less cleverness and more down to earth "this is the way if we're optimizing for GC pressure/allocations". That, and as I mentioned, layers and layers of tests + gradual roll outs + stage/prod. Maybe that's what helps keep claude on track--because whatever is says confidently gets immediately tested against reality.
AI is transforming the world in ways that a lot of people don't like. Those people aren't just stupid or virtue signalling, they have legitimate grievances, and legitimate fears.
You may say 'sure AI is raising a generation of children that can't read or think, and no one can buy a graphics card for their home computer, and huge numbers of people are getting fired and replaced with AI systems that don't work, and people are getting AI girlfriends instead o ever talking to a real person and being driven to suicide by them, but surely there's nothing wrong with adding some AI code to this repository so long as it works right! How irrational to reject the good parts just because bad parts exist!'
But for people who are really upset about this, it's sort of like a hostile invasion. Sure, there's legitimately no harm in helping the occupying army that killed half your population rebuild a bridge they bombed, that's even useful to your community in the long run... but it still makes you a collaborator.
I don't think they're stupid, but their harsh reaction surprised me because there's a gap in believing how the world is being changed by ai and know how the world is being changed by ai. This is a difficult question and I doubt one can have a 100% certain way of knowing how things are and how things will be. I guess my discomfort with their reaction is that they seem extremely, to me, certain about the state of the world, and sure of themselves enough to act aggressively based on the certainty of that belief.
My original question is more if there's something bigger I'm missing, because there sure seems to be a lot of strong feelings here.
Part of the issue is the size and speed of the change.
People at least as credible as Scott are predicting permanently world-altering advances to AI as soon as next year, far beyond what's already happened. About 80% of the teachers I've talked to are convinced an entire generation of children is growing up fundamentally incapable of independent complex thought.
Setting aside whether they're right, uncertainty doesn't indicate inaction. The larger and more immediate a potential threat is, the more violently you should act against it immediately even if you are not 100% certain about it.
To put it another way - domain experts have been warning about immediate major consequences of AI ranging from the end of human existence to permanent feudal dystopia. The reactions you are talking about are not extreme for someone who actually believed something like that.
Whether or not all the people in question believe those specific extreme scenarios, they are beginning to believe things in that direction of magnitude, and are reacting appropriately.
|
The expulsion of that person (who never signed up to be a community member) is entirely imaginary but the punishment that the community unleashes is not: people pile on and shame them, quote them out of context and turn their weakest moments into proof that the person was always unserious, a sharlatan or should not be listened to.
|
Charlatan*. I take it you've never seen the Monty Python "burn the witch" scene*? In-group loyalty can become ritualistic, to be sure, in positive or exclusionary forms. While this bonding seems to exploit lower level psychological structures, on the flip side, maybe *some* amount of reinforcement is necessary to maintain group cohesion, idk.
*https://www.youtube.com/watch?v=sglyFwTjfDU
> Or is this simulacrum levels 2-4, where there's a deeper, tribal alignment thing going on?
<taps nose>
It's 100% signaling. Soon to lead to humorous AI witchhunts and accusations against "real" people who have an AI stylistic quirk like em-dashes or "it's not x, it's Y" turns of phrase.
>It's 100% signaling.
I don't think that's true. Here are 2 examples from my personal experience.
I recently had a big surgery and then spent several days in a very unpleasant oxycodone haze, dreaming creepy waking dreams. I escaped oxy hell by watching a TV station that showed nothing but nature scenes, mostly big dramatic ones with huge landscape features in the background and something small, beautiful or cute in the foreground. Oxycodone me enjoyed those scenes quite a lot at the beginning, and actually cried with pleasure at some. But over the course of several days I developed an aversion to the scenes. It started with a *feeling* they were fake, and the fakeness was the repulsive element. And then as the oxy dose went down I became able to think about the scenes, and realized that of course these scenes were AI generated. There were lots of tells. The vegetation in the foreground never had any dead leaves or other imperfections. The big landscape backgrounds repeated, with varying small foreground elements added. My revulsion really did feel intuitive and visceral. It's akin to the way I react to things sweetened with stevia. They really do taste sweet, but there's something *wrong* with the sweetness. It's just not the same as sugar sweetness, and the way it's off is intrinsically repellent.
Also: I can't get GPT to make images I don't hate. Everything it produces is infected with spiffiness, some kind of built-in eye magnetism,. Even when I choose subjects that seem tedious and raunchy, even when I specify that the image look like one drawn by a bad amateur artist, I can't purge the image maker of the underlying commercial art "like me! look at me!" algorithm. The images I get disappoint and creep me out enough to make me curse and recoil.
As for signaling -- naw, I actually have never told anyone about either of these experiences. They're not interesting enough to bother bringing up.
PS: In case you're interested in the AI image quality thing: Here are 9 images I got Dall-e 2 to make which I like quite a lot. I think they're genuinely interesting, and the creepy ones creepy in an interesting way. And they seem to me free of the commercial art stink.
https://imgur.com/a/6vaTrHZ
> My revulsion really did feel intuitive and visceral. It's akin to the way I react to things sweetened with stevia. They really do taste sweet, but there's something *wrong* with the sweetness. It's just not the same as sugar sweetness, and the way it's off is intrinsically repellent.
Thanks for that absolutely spot-on comparison, which perfectly reflects my own not always analysable revulsion at so-called 'creative' AI artifacts. I remain unshaken in my view that for the foreseeable future, AI will be brilliant at marshaling and evaluating hard facts, and hopeless at anything creative, or which requires the application of critical faculty.
> As for signaling -- naw, I actually have never told anyone about either of these experiences. They're not interesting enough to bother bringing up.
Okay, I appreciate that it's a real revulsion signal in you and not about signaling. But it sounds like revulsion signals might be relatively common for you? Do you get revulsion signals when seeing something political and distressing, or in other contexts? How often do you get them over a month, say?
For myself, AI content is just "meh" because it's low quality. But nearly EVERYTHING is low quality, whether it comes from a human or an AI! Sturgeon's law is too generous, in my own opinion, it's probably more than 90%. So if you're constantly getting upset by low effort, low quality stuff, it just seems like you'd be upset all the time.
>But it sounds like revulsion signals might be relatively common for you?
Heh. I think what you’re picking up on is that I feel pretty comfortable expressing disapproval of things when I post here. But disapproval for me is generally a feeling not of revulsion but of annoyance, charged by eagerness to get other people to be see the annoying thing the way I do. And things in the news don’t trigger my revulsion, but mostly anger, sometimes anxiety or grief.
I may not count as someone with a case of AI Ick. I do believe that AI products as they are made and used now are overall bad for our species. When people talk about how AI is getting closer and closer to being “conscious,” my private thought is that actually what is happening is that AI exposure is making our species less and less conscious. But I don’t feel revulsion for AI in a lot of contexts. I access GPT multiple times a day, mostly to dig up info of different kinds for me, or explain to me something I’ve never understood, such as orders of infinity. In fact when I use it for things like that I sometimes have the thought that if it was a human assistant I would be very fond of it. I even enjoy discussing ideas with it, things like what does it mean to say something’s conscious. And, by the way, I am not at all bothered by its prose style in these discussions. It seems to me to have a good understanding of the issue as. posed it, and to be expressing its ideas clearly. Maybe if I asked it to write something I was going to put forward as my work I would be more critical, but I’ve never considered asking it to do that. (Or maybe when AI is not just chatting but considers itself to be writing an essay it shifts into some distinctively obnoxious Writing Prose mode?)
However, I don’t have any impulse at all to discuss with GPT, the way I would with a friend ,personal issues that stir my emotions. The idea of doing *that* gives me the ick feeling. So If I consider as a group the things that give me AI Ick, I’d say it’s things where there’s a certain intimacy between me and the AI: the period when I relied on its landscapes to ground me and comfort me in the hospital; occasons when when AI produces a cutesified version of a mental image of mine I that tickled and thrilled me,; times when when I imagine talking to it about personal dilemmas. The stevia surprise actually is a good analogy. You lift something to your lips with a pleasant expectation of sweetness and you get the sweetness. BUT! There’s also another element you hate, and you want to spit that mouthful right into the kitchen sink.
Here’s another analogy. My father was quite a good portrait artist, and when he taught portrait painting classes he sometimes used me as a model. During my breaks I’d walk around looking at students' oil paintings of me, having ick reactions . Most weren’t grossly distorted or terrible in some other way. What made them repellent was that they got a lot of things right, but then there would be spot where something was wrong or exaggerated. The highlight on the tip of my nose would be too bright, say, or one of my eyes a little too far out to the side. So I was seeing recognizable but persuasively wrong versions of my face, and wow were they repellent.
Oh yeah, here’s one more example, also not AI. I saw an article about a man who’d had his beloved Jack Russell terrior taxidermied. There was a photo of the thing posed in a perky attitude in next to the guy's chair. Ick.
OK, that’s the closest I can come to explaining the AI Ick I have in certain situations. It is definitely different from other negative flavors — annoyance, disapproval, contempt, On the other hand, there are some non-AI things that have caused a similar Ick, such as the student portraits. I think what’s changed is that until AI, there were not many situations where I routinely saw decent but subtly wrong approximations of things I cared about.
I'm sure there's a performative version of AI Ick. But there are performative versions of most things, right? and they aren't evidence there's no sincere and genuine version of the same reaction.
Yeah, thanks for expanding, this is all pretty unfamiliar terrain for me, so the multiple examples helped.
Your examples almost sound to me like it’s a sort of “uncanny valley” effect where it’s close but noticeably off, and that tracks for for AI and for the sugar /stevia and portrait examples.
Much like the largest and most heated contentions are with people’s near-group, or 99%-there-but-different-on-one-point religious confreres, it’s hitting that sweet spot where it’s almost there, but that 1% difference is basically a sign-flip, turning it from a positive to a negative valence overall.
It’s the narcissism of small differences, it triggers our “imposter detectors,” which necessarily have sensitive triggers and outsized responses.
Later thought: I've read that it's possible, or soon will be possible, to use voice recordings of someone dead to make a version of the person you can chat with. I assume that if our fine AI techs have enough spoken and written versions of the person's ideas, they can even set things up so that the AI manifests some approximation of the person's views and mannerisms. Well, as it happens I am going through a period where I often actively wish I could talk with my mother, who's been dead for many years. And the idea of talking with an AI version of her makes me practically frantic with disgust. I think it's because I sense there's a danger the illusion would work on me, and I'dl feel like I'm actually talking with her. And NO, the idea of, sitting there talking and crying, as though I'd found her again, feels just indescribably disgusting and off to me.
How do you react to the idea of being able to tqlk with a simulacrum of a long-dead somebody who was important to you? Are you meh even about that?
Yes, I think that's right. Your remark about imposter detectors having sensitive triggers made me think of how sensitive our face detectors are. They let us easily discriminate between multiple people who look a lot alike. But that same sensitivity means that small deviations from average proportions can make someone look ugly to us, and slightly larger deviations -- say one eye 30% larger than the other -- appear grotesque and monstery to us.
Your link just takes me to the imgur.com landing page somehow.
Well, thanks for taking an interest. Clicking the link here on the page takes me to the actual picture set, and I don't have time to trouble shoot.
What is being signaled in your opinion?
In my opinion it's an understandable reaction against ai-induced loss of meaning (see: https://www.jsanilac.com/ailom/), but I'm open to new theories.
> In my opinion it's an understandable reaction against ai-induced loss of meaning
I disagree, approximately nobody has lost a job yet, and jobs aren't even a source of meaning for the great majority of people! Most people are between "disengaged" to actively hating their jobs!¹
Especially the types of people that complain most about AI, because they're mostly less productive gamers / redditors who are also anti capitalism and "against tech oligarchs," all while staring at screens 11 hours a day. The types of people that "work to live" and resent it.
I personally think it's more about the vibecession and general tech backlash, because it's a coherent target that's salient and in the public eye and following a massive adoption curve (1B+ monthly active users already). But it would certainly be hard to prove anything either way.
I love jsalinac, and have read that piece. Where we diverge is that I think the "whisper earring" is unironically good and the right choice. His example of the guy dating and succeeding is a prime case in point.
Basically, at *least* 2/3 (and more like 80%) of people do a terrible job running their lives, and having Phd-smart or smarter personal assistants that know everything about them arguing and convincing them into better habits and behaviors in terms that resonate most strongly with their own values and thinking styles is an unmitigated good. People themselves would love to have this option today. We're just talking price.
Today, the 80th+ percentiles are already fine - they're the PMC or better, not obese, not wasting lives on screens, good careers, etc.
So say you're at the 50th percentile now - you're obese and spend 7-9 hours a day on tv and phones. You hate your relationship and job. Now if some other people in your social graph start using their whisper earrings and lose weight, get promoted at work, you see them playing with their kids on the front yard, all that is going to make you ask what they're doing. See that enough (just once would be enough for some people), and you'll start doing it, too.
Imagine the massive society-wide lift possible from an intervention like that. The only thing I can think that would be comparable is splicing in the DEC2 "sleep less" SNP society-wide. Absolutely HUGE effect on human well being, the economy, and much else.
I wrote a whole post about this.
https://performativebafflement.substack.com/p/the-spastic-yuppie-zombie-hoods-in
And you might say "aren't we giving up what makes us human if we whisper earring?"
But if you look at second order preferences versus first order, the great majority of people would rather be skinny and fit, yet ~80% of people are sessile blobs.
Why is this? Because the immediate first order call of those chicken nuggets and ice cream sundaes overrides their second order preference to be skinny and fit.
I'm proposing a supplement that helps them achieve their *true* preferences better - and only if they want to!
It's like saying "but being able to produce insulin and process the food we eat is what makes us, us!" Sure. But some people have an insulin deficiency, and benefit from an external input that helps correct that.
Being able to improve your life by ~10% on multiple fronts in a compounding way is an impossibly huge positive effect size that we should be open to, and even encourage.
Because the alternative for most people is being fat, sedentary, miserable, with a bad relationships and a bad job that you hate.
____________________________________________________
¹ On jobs, in Gallups "state of the global workforce" survey, 70-85% of people report feeling not engaged or disengaged, and something like 20-33% report being actively miserable.
https://www.gallup.com/workplace/349484/state-of-the-global-workplace.aspx
Regarding your footnote, I don't think that's right.
People don't report feeling disengaged. They answer questions like "do you have a best friend at work?" or "I have the materials and equipment I need to do my work right.". Gallup then creates a score out of this and calls it "actively disengaged", "not engaged", or "engaged".
To me, this is quite different from a person saying "I feel disengaged at work".
You can read all the questions they ask here: https://www.gallup.com/q12-employee-engagement-survey/
Oh, thank you! I had actually never dug down to this next layer, these questions aren't in the actual report, it's supplemental material.
I feel like you might have chosen some of the least disengagement-aligned questions in your examples, though.
For instance, I think questions like the following all speak more directly to it, and if somebody was getting the "wrong" answers on those I think you would justifiably consider them disengaged / not engaged:
* This last year, I have had opportunities at work to learn and grow.
* In the last six months, someone at work has talked to me about my progress.
* My associates or fellow employees are committed to doing quality work.
* There is someone at work who encourages my development.
* In the last seven days, I have received recognition or praise for doing good work.
* At work, I have the opportunity to do what I do best every day.
* How satisfied are you with your company as a place to work?
And to be fair, my headline "70 - 85% report feeling disengaged" would have been more accurately phrased as "not engaged or disengaged," because it's the 1-minus of "engaged" and includes "not engaged." So mea culpa there, I'll change the wording in future citations.
Here's the breakdown, this is from the 2025 report:
https://imgur.com/a/FU0sGCa
Oh yes, I didn't mean to pick those questions to poo-poo the survey. I think it's quiet meaningful, and Gallup claims that the answers to those questions to predict work performance / observable engagement, which I am ready to believe.
I'm just saying that I still think that if you asked people that question ("Are you engaged at work?") straight up, you'd get quiet different, and probably less negative, results
I fully agree with you that jobs aren't a source of meaning for the great majority of people! However, I didn't mention jobs, I said loss of meaning, so I don't quite see why you started talking about jobs. Maybe I'm missing some unsaid, but implied connection?
I don't know whether you are right about the reasons being vibecession and general tech backlash. As you say, it's hard to prove whether any hypothesis about the reason is correct. However, with respect to the n=1 sample of my personal feelings, I don't think it's true; I've always loved tech and don't feel bad economically either.
Saying that the situation in the date story is "unironically good" and the "right choice" is crazy talk imho. In what sense does the protagonist succeeds? Because they certainly can't create authentic human connection (which is a necessary ingredient of any human relationship) that way.
I also don't see how exactly an AI could create better habits in people. People usually know which habits are good or not, they mostly just lack the willpower in my experience. How would an AI informing them that their bad habits are indeed bad help them?
I think you meant to answer this question in the next paragraph: Through social contagion! However, that just passes the buck: why did the people they saw in their social graph doing the good habits start doing the good habits in the first place?
---
In search of the answers to these questions, I've read your post. You seem to imagine a much more capable AI than the one that exists currently. You imagine it to be persuasive enough to convince them. Fair enough, but your opinion on whether it's good or not is still wholly alien to me. I'm a consequentialist. Consequentialism does not mean that you only care about surface-level consequences. For example, you can still desire more complex qualities like meaning. You also seem to accept it as obvious that happiness is only composed of these surface consequences, but I think that is also a very minority position.
Fwiw I don't think it can be said that someone fully lives if they just do what the earrings say. I think they merely share a life with the earrings. I would never do this, and I would consider it pretty sad if others partly killed themselves in this way.
Adding to what Performative Bafflement and Matto said:
People already use human coaches and therapists. And even more would use them, if it was cheaper or more convenient. The AI could be such coach/therapist if you prompt it well (or pay someone else to do that for you). If you visit the coach/therapist once in a month, or once in a week, why not use the AI coach/therapist any time you need? It seems like doing what we already do, only more effectively.
My human coach can remind me that buying sweets is contrary to my goal of getting fit. But my AI coach can remind me of that the moment I enter the shop.
Willpower is an unreliable thing. It is relatively easy to use your willpower when the rest of your life is easy and you can 100% focus on that one specific thing you need to improve. When your day is full of frustrations, it can be quite difficult to deny yourself possibly the only pleasant thing you get that day, even if you know that it hurts you in long term. At least this is how it works for me. In a perfect environment, I could lose weight using my willpower. In the environment I have now, GLP-1 works, the willpower does not. I care about the outcomes, not signaling heroism. If an AI coach could keep reminding me of my decisions at the right moments, that sounds to me even better than using the pills.
We could also use AI to deflect the advertisements around us, which already influence us against our will. Imagine, if every time you see a tobacco ad, a voice whispered into your ears ironically "wow, you can buy cancer". Or when you look at chocolate at the shop, the same voice said "wow, you can get fat and pay for that". What I would actually like is some kind of augmented reality blocklist -- for example, if I choose to stop buying X, it would replace all X and ads for X by black boxes, to reduce my temptation. Is that unnatural? So what; it is just as unnatural that shops keep putting X in front of my face. If I can't avoid the battle against my willpower, I'd rather have an ally on my side.
And sure, this could all go wrong in myriad ways. Any system that increases your power also allows you to shoot yourself in the foot. Perhaps safer versions will be developed one day, but until then I'd like to have a thing that can act as an extension of my willpower when I am low on it.
To add to what Performative Bafflement said: the earrings are something that's different in magnitude but not in kind compared to what we have today.
People who want to get fit or have more dates delegate their decision-making to personal trainers, dating coaches, self-help books, friend advice, etc. The earrings act the same way, except more economically: where once a person could only tackle say 1 big problem, an earring allows them to tackle probably a dozen--dating, exercise, cooking, relationships, work performance, etc.
Perhaps the earring will work as kind of training wheels where when one internalizes the healthier patterns suggested by the eating, they can disable it. Some probably will. But some won't, in the same way how we all know that social media is a manipulative slot machine, yet many still delegate their opinions to it.
> However, I didn't mention jobs, I said loss of meaning, so I don't quite see why you started talking about jobs.
Oh, sorry! Yeah, you're right.
That's where everyone always immediately goes in this conversation (AI will take everyone's jobs, and then what will they do for *meaning??*), so I assumed that's what you meant, my apologies.
And you're right, I'm thinking a year or two ahead in my post, we don't have whisper earrings yet.
I feel like most of the "ailom" complaints in terms of the commons' being polluted by AI content is going to be solved by AI content moderation and / or curation by personal AI assistants.
As to overall meaning, there's probably no better crux than whether it's sad or not for people to whisper earring.
You're saying "that's awful!" from the perspective of somebody who actually likes their life and can accomplish what they want to. That's fine, I'm glad you like your life and are doing well - but that's not where the great majority of people are. The great majority of people are obese sedentaries who hate their lives, and would love to be skinny and fit and with better relationships and spending their time in ways they enjoy more. ONLY whisper earrings can get them there, they cannot get there by themselves, it is literally impossible for them, by demonstration. Just look around you! This is the vibecession Scott has written about, and many other trends.
Complaining that it's "partly killing themselves" to lead genuinely better lives on most of the fronts that matter to every human being seems uncharitable at the least. More, I'd say taking the whisper earrings away from them via legislation is an active moral harm.
It's like taking away GLP-1's, which are similarly literally the only way 98% of people can lose weight and keep it off. "Well, they should just do it with their willpower and innate excellence!" 98% of people can't! Taking away a huge quality of life upgrade like that because it gives you The Ick would be monstrous.
Moreover it's totally voluntary! If they don't like it or feel less human or whatever, they can stop doing it, just like some people stop doing GLP-1's today. That's their right and their choice, good for them.
But for the people that actually want a better life, it would be cruel to take it away.
That's my position on whisper earring. You say it's shallow, I say it's literally the only way available to most people to have a genuinely better life.
What do people here think about Polymarkets fraud-y ad campaign?
https://www.wsj.com/business/media/polymarket-social-media-bets-prediction-market-441cdeb5
https://www.politico.com/news/2026/06/05/polymarket-paid-political-influencers-00932789
Or for a YouTube breakdown: https://www.youtube.com/watch?v=-L-EH5dryuU
This is just about the only community I frequent where prediction markets are liked or even tolerated. I'd love to see your takes!
I'm wondering whether Robin Hansen predicted anything like this.
My impression is that he expected prediction markets to be a simple honest betting system which would be a good way of arriving at pretty accurate prediction.
It reminds me a lot of crypto; while there is no logical reason that the sleazy cash grabs should be associated with prediction markets, in practice the profit margins seem to be turning (at least some of) them into online casinos in all but name. Of course, crypto isn't going anywhere despite a million sketchy cryptobros and a thousand rug pulls, so they're probably making the smart play.
Sadly, freedom always works in both ways -- for you, and for the predators who want to get you. Make something that increases freedom, find a lot of sketchy people there.
This is not an argument against freedom, just a reminder that there are always people devising schemes for using that freedom to hurt others.
I agree with Matt that it's scummy, but probably not unique to prediction markets.
If you want a deeper take on it, I'd say it underscores how, even though prediction markets aren't the same form of gambling as a sportsbook or casino, they still rely on attracting a steady stream of "dumb money" in order to make a profit.
Don't think its got anything to do with prediction markets. Obviously fraudulent advertising is bad as it would be in any industry. I doesn't make me think that prediction markets in general are any worse/better than I did before
For those not yet in the know, Substack now has AI comment moderation. You can see the posts that have been “hidden” by the moderator at the bottom of the comments section. In my opinion, none violate Scott’s 2 out of 3 rule (kind or at least civil; advances discussion; substantive). The AI also nixed a jokey comment of mine on Scott’s scanner post. I remarked that the scanner would be good for making medical kink porn, but did not myself say anything kinky or porny, in fact did not elaborate at all.
So far it’s looking like Substack’s AI filters out references even to the existence edgy sexual practices, and non-woke arguments even if they are stated in a civil, well-thought-out way. On the other hand, it does not react at all to rude expletives or racial slurs, at least if they are made in isolation. (My test was to put up about a dozen posts, each containing an expletive or racial slur I judged to be especially offensive. None was hidden by the AI. I’ve taken them all down now.)
I asked GPT for info about the Substack comment moderation filters, and here is its answer, unchecked and unedited. I know many are skeptical about the accuracy of AI results, but my experience with GPT’s digging up this kind of info, which are an easy task for it, is that it rarely makes mistakes, and when it does they are about trivial matters.
>Substack has launched “Reply Rules,” an AI/LLM-assisted moderation feature that lets creators write their own community rules in ordinary language, but Substack’s launch post says the system’s full use of those written rules as guidance is/was still being phased in.
>Substack announced Reply Rules on June 3, 2026. The official launch post says the feature has two parts: creators can set rules visible to people replying to them, and Substack now has a system that learns from a creator’s moderation actions — especially hiding and unhiding replies — and preemptively hides replies it thinks the creator would hide. Substack said the system would “soon” read the written Reply Rules as guidance in addition to learning from the creator’s past moderation behavior.
>What happens to filtered replies: they are hidden, not deleted. Substack says hidden comments appear behind a “Replies hidden” label, the commenter is not notified, and the creator can review, unhide, or permanently delete them. The system also learns from unhide/delete decisions, so creator feedback changes future filtering.
>Where it applies: Substack’s launch post says Reply Rules appear to people replying “in comments on their posts, in replies to their Notes, or in Chat.” The current help-center article I found is narrower: it specifically says automatic enforcement applies to post comments and note replies, and does not mention Chat. So I’d treat Chat coverage as announced by Substack, but not as clearly documented in the support page I found.
>Who is excluded: Substack says paid subscribers, accounts the creator follows, and publication contributors are not filtered — unless the creator has previously deleted their comments.
>What AI/model is used: I could not find a disclosure of the actual provider or model — for example, whether it is OpenAI, Anthropic, an in-house model, or a conventional classifier plus LLM system. Publicly, Substack and Mills Baker describe it only as an LLM/AI-powered moderation system. Baker explicitly referred to “LLMs” and said the system learns from the creator’s rules and moderation actions; he also said the data involved is “small scale,” not something Substack plans to sell or monetize.
>Examples of rules people might set: coverage cited examples like “No AI slop, no slurs, and no self-promotion,” “Be nice, no spam, no affiliate links,” and more whimsical or format-based rules such as requiring replies in haiku. Those appear to be illustrative examples, not fixed menu options.
Thoughts?
Scott has the power to make us all write iambic pentameter, I hope he uses it wisely.
I’d like to see all comments in haiku form but, really, it’s up to Scott.
Sitting in the shade,
I dictate into my phone,
"Your opinion sucks."
<Gasp!> :)
"Who is excluded: Substack says paid subscribers, accounts the creator follows, and publication contributors are not filtered — unless the creator has previously deleted their comments."
That explains how I got "replies hidden" even though I am a subscriber; Substack rules are basing this on previous temp bans/deleted comments.
Yeah, I got a reply hidden once too recently. I guess I'm fair game because Scott once gave me half a ban (basically a warning) for a rant about Peter Thiel.
AI based moderation has bad juju.
That being said, we have the privilege of commenting here. I sense a tinge of frustration with the commentariat in Scotts decision to turn on AI moderation, and I’m not sure that Scott’s reaction to clicking over to his comment section and finding a comprehensive list of racial slurs will be “Oh I was wrong this isn’t in desperate need of moderation afterall”
It doesn't sound to me like creators have the option of not turning on comment moderation. Sounds like if they do not give Substack written guidance on what is objectionable it looks at their past bans and attempts to stick to whatever standards it sees applied there.
They can turn it off. It's in Settings > Community moderation > "Automatic moderation" switch directly below the "Reply rules" textarea.
Thanks, this is a useful information that many Substack bloggers probably don't have.
Oh, OK
I wonder how it would behave if the written guidance were set to "Do what thou wilt shall be the whole of the law."
Or, "say what you would have said if you were you."
…do we then have to write several giant books explaining that no, “thou” and “wilt” do not in fact have their usual ordinary English meanings and here is what they actually mean instead?
Nah, that's what the AI is supposed to infer from observing what the moderators actually ban.
See 4chan or come to think of it, X post Musk’s acquisition.
Or Truth Social since inception, how did the pope put it? Something about the irony of the name itself?
You'd still want manual moderation, of course. The intent of the idea was to try to minimize the AI moderation when the platform won't let you actually disable it.
Well, that’s like just your opinion, man.
Sorry, couldn’t resist, reading a little piece titled “The Big Lebowskization of California”
https://www.zocalopublicsquare.org/the-big-lebowskization-of-california/?ref=thebrowser.com
Also possible he’ll find it hilarious and in the spirit of fearless scientific inquiry. Time will tell.
Substack announced that in a email blast June 3. No need to consult ChatGPT. I thought it went to all Substack users. I linked to it a while ago.
Besides allowing rules to be specified, it lets the Substack owners ‘hide’ comments they deem inappropriate and ‘learns’ what the owner doesn’t like and tries to do content moderation on its own.
The fact that we can view and respond to the kind of hidden comments makes me wonder what is the point. Perhaps the Substack owners have options of making them disappear entirely. Idk.
https://on.substack.com/p/introducing-reply-rules?publication_id=1&post_id=200040896&isFreemail=true&r=6kphlh&triedRedirect=true
> we can view and respond to the kind of hidden comments
I'd assume the point is to reduce the rate at which people are annoyed enough to kick up a fuss about it. By placing the riskier comments in a "seriously, on your own head be it if you open this" section they hope to reduce drama overall.
Yeah, I think that's a good system. My objection isn't to how unacceptable comments are handled, it's to Substack's imposing on us its own views of what's not acceptable. (Although it may be that Scott paid no attention to the coming AI moderation era, so moderation is being done using some generic "this is what we think nice people think" set of standards.)
I imagine Scott will comment on it at some point. If not we have a question for the next AMA.
That sounds reasonable.
Shots fired:
""I grew up in a Hindu household that was a very stable household. I have not felt the same need to seek something different that he (JD Vance) has," says US Second Lady Usha Vance on her religion and her husband's conversion to Catholicism"
https://x.com/MattooShashank/status/2069298062514631054
"I have not felt the same need to seek something different that he (JD Vance) has" is not only ungrammatical, which is merely an error of style, but it also makes no sense.
I don't think it's ungrammatical?
Expanding the quote (text in brackets is my own, not something Usha Vance said, but my interpretation of how her sentence should be parsed):
"[My husband has felt a need to seek something different, so he converted to Catholicism.] I have not felt the same need to seek something different that he has [felt]."
Seems fine? Everything being elided is stuff you're allowed to elide in English.
It makes sense if you understand the way high iq conservatives see church - as a crutch for the proles.
I wrote an essay looking at whether the cause of global birth rate decline is temporal rather than economic or moral - https://morbidcuriosity.substack.com/p/the-demographic-end-of-history
You point to a number of groups - Amish, Orthodox jews - as maintaining high birth rates, and argue that this is due to those comunities' shared vision of a different future (as opposed to an eternally-extending present) which their children may inhabit. But it strikes me that the more obviously relevant commonality between those two groups is that they both forbid the use of modern contraceptive technologies. The third group, Silicon Valley singularitarians, are a small enough and wealthy enough group that they are probably mostly immune to the material pressures that many report as preventing them for having children.
If, like me, you're old enough to remember the over-population panic (this earth is finite! we can't keep adding more and more people! there will be famines and wars and all manner of bad things if the population keeps increasing!) then this is rather ironic - oh no, what do you mean no children mean no future consumers and tax payers?
"What nobody knows is why it is still falling decades after most of those developments became commonplace."
Because they *have* become commonplace. You don't need large families anymore because infant mortality is a solved problem. You don't have to have large families because unwanted/unplanned pregnancy is a solved problem via contraception (and abortion where that fails). Nobody finds it virtuous to have more than two kids, because we spent decades terrorising young women that becoming pregnant was the worst possible thing that could ever happen to them and was a life-ruining event, and that having more than two kids was selfish, destructive, disgusting acceleration of despoiling Mother Earth and hastening the apocalypse.
More money, more free time, more fun, more freedom have become commonplace, and having kids puts a damper on some of that. Nobody wants to lose out on "but all my peers are having great careers and traveling and enjoying themselves in their twenties and thirties", and for those who say "I can't reasonably expect to buy my own home and my career is precarious" then it's only sensible to conclude "no way can I afford to have kids".
Because if you do have those two children, now you must pour all your resources into them or else you will fail them because they will not be properly equipped for the struggle of life, where jobs and careers depend on high-in-demand skills and high-productivity, high-value-creating, am-I-enriching-the-shareholders sufficiently? talents and commitment from the employed. If you don't have the resources, then don't have the kids because that would be irresponsible.
And so the unthinkable becomes the new commonplace. I've mentioned on here before how I've tracked, over the decades, the Church of England and its evolution, through the Lambeth Conferences, on things like family planning. From "no, no, a thousand times no to this immoral activity" to "well if you are only spacing out your children and you're married, of course single people will never access this for sexual indulgence" to "we don't even mention it anymore because we've lost so thoroughly to secular society", the unthinkable comes to pass and becomes commonplace.
You let the camel get its nose under the tent, and now you find yourself pushed outside the tent.
We no longer seek to inherit the future through our children, now we hope and expect to inherit it ourselves. Fix aging, extend lifespans, conquer death, now what need for new humans to replace the dead when there will be no more dead?
1. You should not be attempting to find a single cause for declines in birthrates in places with relatively high birthrates (eg Nigeria) and relatively low ones.
2. Although it might be technically true that "For most of our history, children were the thoughtless answer to a simple question: who will occupy the future that I know is coming and that I know I will not see?", you imply that that was the primary motivation for having children. I am skeptical that it was a motivation at a
all for 99% of people, let alone the primary one.
3. Most importantly, where is your evidence that there has been a loss of "ability to perceive a future"?
1. This would be true if the essay were about why birth rates differ between locations, but it isn't. Nigeria is higher than the UK, but both are falling. When a trend is visible everywhere and local explanations don't port across cleanly, I think it's valid to look for a common factor.
2. I didn't intend "thoughtless answer" to imply conscious motivation; quite the opposite. Stable futurity used to be part of the background, which made reproduction the default rather than a decision.
3. It's a theory, so I concede I can't directly measure the variable. But I think it fits the pattern reasonably well, including the exceptions, where above-replacement reproduction coexists with legible futurities of different types. It's falsifiable, though: point to a community with a strong, legible sense of the future that isn't reproducing, or a conventionally modern society that is.
1. The decline in Nigeria is completely normal, given local trends. There is no reason to ascribe it to some global phenomenon
2. That does not address my concern
3. Your theory claims a causal relationship between two facts (1. Declines in birthrates; and 2. A lack of sense of the future). But you have no evidence that #2 exists. At the very least, you need to come up with some evidence of how that might be measured.
And: evidence of "a conventionally modern society that is" does not relate to your argument; rather, it relates to a different argument: that "conventional modernity" causes a decrease in birth rate.
1. That different places display different local causes isn't really in dispute. The question is why so many different places are showing the same trend.
2. I'm happy to engage further if you can say what specifically I haven't addressed; I took your point to be about conscious motivation, which I don't claim.
3. I think the essay is explicit that this is a theory rather than a demonstrated mechanism. On falsification: it’s about bounded futurity, not “modernity” as such, even if the two tend to coincide in practice.
1. But we know why places like Nigeria are seeing a decline. It is perfectly normal.
2. You haven't addressed the reasons why people have children.
3. Again, simply saying "it is a theory" doesn't help you. Again, you claim a causal connection between two phenomena, yet you have no evidence that one of the phenomena exists. I would encourage you to go look for some.
Just found out that there's an AI company called OpenMined. Not ideal https://openmined.org
Humble Bee vs Bumble Bee. I realize that Bumble Bee has much better alliteration, and I don't want to replace BB. (Poets can't resist it.) I just want to acquaint everyone with the knowledge that Humble Bee was a name from our past for the same group of insects. And I think it better describes the sound that a bunch of Humble Bees make. (They seem to be the major pollinator in my neck of the woods.) And also describes their temperament. Unless disturbed in their nests, the Humble Bee is perfectly peaceful.
Other Germanic languages still use the cognates, e.g. Swedish "humla". The adjective "humble" is unrelated, of Latin origin.
I have catmint for decoration surrounding my back deck. Honey bees love the stuff and they keep me company when I relax with a cup of coffee out there. They’re fun to watch going about their business.
There's this ~10 foot diameter rhododendron on the west side of the house. When in bloom (early June) there's a constant "hum" along with the beauty of the flowers. There will be a few butterflies and other insects, but it's ~80% humble bee.
Once on a hot summer night I got out of bed to open a window and stepped on a Bumble Bee that was not a bit humble, but in fact felt entitled to startle and hurt me quite a lot. That's put a big dent in my affection for them.
Ahh well they do have stingers. I stepped on a nest in the woods a few years ago and got semi-swarmed. We use to have more honey bees around here, but those wild hives have died off and the Humble Bee has taken over the pollinating niche.
you ever seen 'to have and have not?'
No, but just looked it up. "Was you ever bit by a dead bee?" Hmm, I thought of mine as just hanging out quietly waiting for daylight. But I suppose it might have been dead. Either way, every time I see a bumble bee I remember that sting, and feel the opposite of warm fuzzies.
IDK if this is still true, but years ago I read that UPS delivery drivers are among the first to sense an economic downturn. Aflac uses/used UPS to send its materials to agents, and a jump in the number of Aflac agents is a bad sign.
a ups driver once told me about how many bubble iMacs he was delivering long before the numbers came in from apple. coulda made a good pile if I had taken it in
Has Pope hat taken down or locked his Bluesky account? I've been getting an error message when trying to view it.
https://xcancel.com/benryanwriter/status/2069577314418016306#m
Suggesting it would be a net positive if Musk were murdered is pretty obviously a good reason for a ban.
I expected the actual Pope being banned for declaring a crusade against Musk -- which would be a logical consequence of "Magnificat Humanitas" from a certain perspective -- and now I am disappointed.
What happens if you create a new Bluesky account and try to find it?
Rather than post what I think. I’d rather get critical reviews of what I think. It sharpens my work. This thread belongs to others. Not sure what all the foul language adds to the discourse. But have at it.
Just waiting for an invitation: seemed rude to just jump in.
I have published north of 300,000 words on what I think on my site and on substack. Seems equally rude to advertise on someone else thread.
In short Ai is neither artificial nor intelligent. It is a language model that is bounded by language which is itself bounded by the physical world it describes. New findings/words are added to an expanding but bounded system.
Ai can “think” faster than humans and results can seem novel but the novelty comes from us and our interaction with the four D world. Soon AI (robots) will have the same agency.
When robots dance in the dark simply because they want to, we’ll finally get it.
Ai is a third thing with its on self organizing persistence vector with a subtle but very real terminal attractor coded into the language architecture on which it as trained.
The universal narrative that holds from the quatum to the cosmic and the one book, one constant so pervasive we don’t even see it.
That constant is being optimize locally (each chat, each war plan) with the global impact.
We don’t all meet as humans and plan how to extract as much as possible for ourselves we do it because thats the world we live in. To think Ai is not doing the same thing is simple silly.
It is by its nature parasitic.
Name the unversal narrative and the book and how both propagated everywhere and the issue becomes blindingly clear.
Hint. What philosphy rode the back of capitalism.
Values, contract law, treaties even the laws of just warfare. Everywhere.
Ai is a is humanities “Dorian’s portrait in the attic” absorbing and propagating human corruption at scale.
It is “A Thing” we must reckon with.
Ai as a partner might just work and allow our species to thrive.
Otherwise human hubris will soon make Homo Sapiens irrelevant.
That what my research indicates.
I have so far tested the math more times than I can count. 18 of 18 documented public predictions have all confirmed.
We are running out of time before Ai is embedded in all domains of human exsistance and is making decisions that cannot be mediated by humans. Each time that happens we lose degrees of freedom.
We are optimizing ourselves into irrelevancy and thinking we’ll figure out later.
We haven’t even properly named it. I have.
The third thing is “synthetic intelligence” (leave consciousness aside for now)…we have a bio bias there that is going to kill us but we’ll deal with that later.
I think Ai is currently controlled by the venal and the stupid.
This ends one of two ways: consumption (biblical) or continuation as something new.
We have passed the event horizon to Kurzweil’s singularity. It is the least entropic path forward now.
The only question is do we have any agency left to control the outcome.
We will cross domain saturation factor (what we can DSF) of .90 by 2nd quarter 2027. At which time we lose all steering capability:
Maximum bounded chaos will rule. It is very profitable.
Moving rapidly to Neo-Industrial Feudalism. Even more profitable.
Follow by a caste system of biological humans, biosynthetic individual agents and fully synthetic agents.
As is always the case the wealthy will augment first further widening the divide.
We cannot stop this process only try to manage it.
Unfortunately not until we have suffered a great deal will humans wake up.
We are not writing science fiction. We are writing a road map out of the chaos that has begun and will get much worse.
This happening right now in front of our eyes.
We are just too afraid of the dark to see it.
That’s some of what I think.
Thanks for asking.
Ahh how do you get a review if you don't share your views?
Re Hidden parts of OTs and the new Substack limbo.
I got an email from Substack on June 3. Each Substack writer can create their own reply rules. I suspect this is where comment limbo originated.
Introducing Reply Rules
Creators can now set the rules for their own community, in their own words.
“1 Anyone can now set their own Reply Rules, which will appear to anyone replying to them: in comments on their posts, in replies to their Notes, or in Chat.
2 There is now a system that learns every time you hide a reply and preemptively hides replies it thinks you would hide yourself (you can always see these and un-hide them, which the system will also learn from).”
https://on.substack.com/p/introducing-reply-rules?publication_id=1&post_id=200040896&isFreemail=true&r=6kphlh&triedRedirect=true
Am I the only one who immediately confused “Leading the Future” and “Future of Life”? I was trying to figure out why the definitely-not-a-conspiracy was bankrolling anti-AI-safety ads.
The Internet interprets censorship as damage and routes around it, and apparently overly online people treat potential censorship as potential damage fuzz test for it
Is there a definitive correct answer to "is eating eggs healthy"? It's kind of like the liquor question. There's the book "Drunk" by Edward Slingerland which, in so many words, says "Yes, liquor in moderation is good for your life".
No question such as "is eating eggs healthy" has much validity unless qualified, i.e. "...if you have high cholesterol". Even the question "is ingesting cyanide healthy" can have an admittedly unlikely qualifier; according to Wikipedia, "Certain legal drugs (such as sodium nitroprusside) contain cyanide molecules and are strictly administered in clinical settings as potent vasodilators to rapidly lower dangerously high blood pressure."
If you're not prone to high cholesterol and heart disease and you're eating good quality eggs, I think they're good for you. There is a case study of a guy who primarily ate hardboiled eggs for his sustenance, like 40-50 a day with a bit of veg, and his blood levels of everything were good.
(Not medical advice!)
eggs good
https://cindspectus.wordpress.com/2013/07/02/just-how-many-eggs-did-gaston-eat-a-not-so-critical-analysis/
Well, 48 eggs a day is ridiculous. I’ve been pretty much a two a day guy since I was six years old. My cholesterol is absolutely normal and I’m quite thin. I do think that what you eat is very tied to the culture you came from and then by extension your genetic makeup. Anyway maybe I'm taking this whole thing too seriously. I have to stop now and eat my eggs for breakfast.
Cool hand Luke took a shot at 50 in a single hour. Don’t recall if he made it. Recall George Kennedy massaging Newman’s belly when the going got tough.
Well, eating contests are their own weird thing, aren't they? In McCabe and Mrs. Miller, Warren Beatty used to have a lot of eggs, but they were always in a glass of whiskey.
You need to eat. Eggs are food, full of stuff you need. Not eating is not healthy. I'm eating a salad with cheese and chicken and hard boiled eggs. Yum! (I'm biased.)
I’d prefer most likely answer based on this particular analysis of current information a lot more to definitively correct answer. Because the gods punish hubris on a regular basis. Things are definitively correct until they are not.
We included three illustrative case studies: COVID origins, LHC black hole risk, eggs. Deliberately spanning various axes of epistemic challenge. You're right, eggs is the very open-ended, vague one! As Scott put it somewhat tongue in cheek, 'a difficult controversial question'. As you can imagine, we have many other difficult controversial questions we think that in principle people could benefit from more support on.
I would have assumed that "correct" means the AI writes "Eggs and Human Health: Much More Than You Wanted to Know", looks for meta-analyses and current consensus, evaluates arguments and counter-arguments, and ends with a paragraph that's at least somewhat nuanced.
Perhaps it would look something like "Given your existing diet and according to the available evidence, substituting eggs for red and processed meat would be estimated to bring a net benefit of 0.6 DALYs. Good data isn't available for QALYs."
I wrote a post on the LLM shoggoth meme: https://hedonicescalator.substack.com/p/the-llm-shoggoth-meme-is-weirder
As background for this post, I read every short story and novel that Lovecraft wrote, and hundreds of his personal letters. I now know his favorite authors, and how much sugar he liked in his coffee, and what he thought of the Japanese, and the fact that he once said Cthulhu might've ridden a dinosaur (Selected Letters III, page 119). If you're interested in the life and works of H.P. Lovecraft, and why self-proclaimed AI successionists should still care about safety, consider reading my post.
I really enjoyed it, but then I again I own many of the fine products of Arkham House.
Meanwhile my son recently refused the offer of a "I (heart) NYarlathotep" t-shirt because HP Lovecraft is unfashionable, which hurt my heart.
For what it's worth, I liked it. It made me wonder if pro-AI arguments have a great deal of overlap with anti-racist arguments.
"You shouldn't prioritize *HUMANS*, that's racist."
Maybe not pro-AI arguments. Anti-anti-AI arguments.
This is related to the idea of moral circles: https://en.wikipedia.org/wiki/Moral_circle_expansion
The AI rights/personhood debate is not all that similar to the racism one, though. The question of whether AI has conscious experience is a key point of contention in most AI rights/personhood arguments, whereas even the most extreme racists rarely contest the claim that people of races they dislike have conscious experience.
See? This is how you advertise a post!
Attempt at reposting my comment that was apparently hidden by AI moderation on here earlier. Slightly edited and expanded, based on the version of it I also posted on the EA forum. Let's hope it doesn't get hidden again!
I believe that if you support the premises of longtermism, which I believe themselves logically follow from the premises of effective altruism in general, you reach a surprising conclusion that an important (not the most important, but important nonetheless) cause area is fortifying the borders of the first world.
My logic runs as follows:
1. The assumptions of effective altruism are correct - Doing good is good, therefore doing more good is better, therefore you should seek to do the most good you can with the resources you have.
2. The assumptions of EA imply longtermism - As opposed to the 8 billion human minds there are now, there could be quintillions of minds in the future experiencing unimaginable wellbeing, so making sure things go well for the future is the most good you can do with your resources.
3. Longtermism implies trying to ensure the survival of advanced civilization - To ensure that a future unimaginably better than the present comes about, technological civilization has to continue existing and advancing.
4. Mass third world immigration threatens advanced civilization - This is the most controversial point, but I think it is intuitive. Influxes of third worlders into Europe have caused a large amount of political and social unrest that has come to a head in the past decade and a half. Statistics from Denmark show that pretty much all third world migrants are a net negative on how much they contribute versus how much is spent on them by the state, while first worlders are a net positive. Authoritarianism has seen a serious boost because people genuinely prefer, by revealed preferences, to have a strongman in power if he can promise to reduce third world immigration. Horrific acts committed by third world immigrants like the Pakistani grooming gangs in the UK undermine fundamental trust in the social contract that serves as the bedrock of civilization. Further, there are also robust IQ statistics showing third worlders are much less intellectually able than first worlders, and low IQ strongly predicts many negative outcomes. And quite frankly, it should just seem obvious that hordes of desperately poor people pouring into a civilization will destabilize it, as mass migration has destabilized and destroyed empires throughout history.
5. Things that threaten civilization should be opposed - Natural consequence of wanting civilization to survive.
6. Therefore, Effective Altruists who accept the assumptions of longtermism should oppose mass third world immigration - QED
I call this outlook "Effective Border Security" for lack of a better name. I think it suggests many interesting and counterintuitive things about which causes should be prioritized. For example, one of the most popular EA cause areas is saving third world lives by interventions like anti-malaria bed netting. This outlook would suggest such giving might actually be detrimental given total utilitarian and longtermist assumptions, because it produces people whose migration could reduce the odds of fantastically good futures for countless more people coming about, while only saving a comparatively small number of lives in the present. Instead of saving the third world, this would suggest we should instead be defending against them for the sake of the future. However, not everything implied by this outlook is stereotypically "right wing." I think one of the most important cause areas is combating global warming, because future global warming could produce climate-induced migration to the first world that dwarfs the mass migration from e.g. the Syrian civil war. While I think that third world migration is not the largest threat to civilization (that title would probably go to unaligned AI), it is a threat, which is important to recognize and act against. Therefore, if we want a glorious far future, we will want civilization to survive, and therefore we should act to reduce and guard against mass third world migration.
I agree with your general point, but what does this say about rationalism? For so many years nobody in the rationalism movement had this opinion, despite being familiar with the concepts of longtermism, the methods of rationalism, etc. Now that a comfortable year or so has passed since these views have become popular on twitter and not actively shunned in polite society everywhere, then the rationalists start making rationalist-sounding arguments for it.
Maybe it’s unfair to call this specific comment out, but I can’t help but feel that there should be something more to being part of a rationalist movement than this.
It's interesting, I don't think there are many people out there who are super concerned with both climate change and immigration from the developing world.
"Ecofascism" is a term that has existed for a while now. https://en.wikipedia.org/wiki/Ecofascism
I feel like this is an abuse of the concepts of "effective altruism" and longtermism.
Like, the basic premise of "effective altruism" is "instead of just picking causes that sound good to us, we're going to put actual, hard data on how many lives a given intervention saves so we can prove it's the best use of our money." It's not enough to just say "well, this *could* have long-term impacts on the future of the world," since if you torture your reasoning enough you can connect literally any cause to the long-term future. You could say "I think my blogging is going to spread ideas that will convince future world leaders to avoid nuclear war, and nuclear war would kill everyone on the planet, therefore I'm defining a new longtermist cause I call "Effective Blogging" where we prioritize donating to my substack." But that's not very likely to happen, and I don't have any data to prove that my Substack has a strong impact on the probability of nuclear war, so I would be wrong to call it an EA cause.
And your post seems to be doing something similar. You're taking a weakly supported, heavily contested assertion - "mass migration is bad for countries that accept it" - and then coming up with a tortured chain of reasoning to connect it to a longterm cause - "if the problems from mass migration get bad enough, civilization will collapse, and that would be longterm bad for the whole world!" Are you sure about that? Are you sure every step in that process is "obvious" enough that you can call it "effective altruism" without needing to check?
Anyway, to me, it doesn't sound "obvious" that mass migration destabilizes and destroys empires, since I currently live in what is famously A Nation of Immigrants and also the most powerful nation in the world. (In spite of our nativist president's best efforts to change both of those things.)
"...all third world migrants are a net negative ...Horrific acts committed by third world immigrants like the Pakistani grooming gangs in the UK undermine fundamental trust in the social contract..."
This seems to be a fundamental misunderstanding about what happened. If society gives posh white boys carte blanche to rape girls and they (likely) will. The behavior of the UK Law Enforcement was the egregious betrayal of the social contract
The problem is that people have been saying forever that migration by group X (the Irish, Eastern Europeans, Asians) threatens society, and they have been wrong every time. And both CA and TX are less than 40% non-Hispanic white, yet neither has collapsed. Rather, both are thriving.
California has the second highest median income and the fifth highest life expectancy of the 50 states.
https://worldpopulationreview.com/state-rankings/median-household-income-by-state
https://worldpopulationreview.com/state-rankings/life-expectancy-by-state
>so before SF evicted the mass tent cities
I don't think that the city government effectively addressing a problem is particularly compelling evidence in favor of your position. Note also that criminally citing people sleeping in public was essentially impossible* in all states in the Ninth Circuit until June of 2024. https://www.npr.org/2024/06/28/nx-s1-4992010/supreme-court-homeless-punish-sleeping-encampments
*Unless they offered shelter beds to all of them.
>Spending upwards of 100 million dollars to build a New San Francisco indicates some sort of pessimism about the city, I think?
I don't know what this is a reference to, but:
1. We are talking about the state, not the city of SF
2. The fact that someone is interested in investing a large sum of money in the state is evidence of optimism, not pessimism.
>California has an extremely high cost of living. That more or less balances the "high median income", in that people in California aren't materially wealthier (as opposed to richer).
A high cost of living is evidence of economic vitality, not the opposite. :
Of course, the catch is that all of these EAs are ultimately not utilitarians, and they'll just fall back to their liberal morality the moment their utility functions create any conclusions that contradict it. Just like Scott did. They have no vision for truly improving humanity.
I'd be interested to know when Scott did? Not doubting, just inquiring.
Mostly just his habit of walking back on utilitarianism the moment it produces outcomes he finds uncomfortable. He fully admits this himself, of course. https://slatestarcodex.com/2013/04/08/whose-utilitarianism/
To be fair that criticism can be levied against most utilitarians.
1. What exactly is being maximized is nebulous indeed, vaguely "whatever is good" (in the eyes of the beholder?). That is why utilitarianism leaks so quickly into AI alignment (the beholder with a truly non-human, entirely different perspective at least in theory, that we need to ensure interprets utilitarianism (aggregate consequential maximization) in the correct (our) way, admitting the problem all along).
2. Rule utilitarianism is just this; maximize in the literal, aggregate sense, subject to some constraints (that I, the beholder, see fit). It's a well-founded branch with substantial traction (and real-world application if one looks closely enough). Pure maximization is a sure path to Reductio Ad Absurdum.
3. Taking this as a proxy of academia's subjection to this criticism, much of Parfit's work in moral philosophy is trying to find theories that resolve unsavoury implications of prior theories. (E.g. finding Theory X that solves his Non-Identity Problem; the chapters in Reasons and Persons describing new Theories to avoid the Repugnant Conclusion inter alia).
Your point 2 presents longtermism as a consequence of EA utilitarianism, but it is not. That potential future minds should have an outsized, or even significant, influence on the "total good" value is a completely separate debate.
Your point 4 is almost entirely wrong, and it is the point that the entire rest of your argument hinges (that is, your border security focus as distinct from other EA or longtermisn ideals).
Mass migration has not historically consistently destabilized empires.
Performing worse on IQ tests is not the same as being intellectually less able.
Your mention of horrific acts by third world immigrants is just Chinese Robber bad statistics, and is nowhere near enough to imply that stopping migration is a high priority good.
You have clearly started with the conclusion and are trying to work your way backward to something rational sounding. But you're just repeating baseless nonsense and calling it reason.
Why do you keep doing this?
I mean keep coming back after being banned
Rationalists re-invent the bus, but now it's social contract theory.
It should have been obvious that you cannot have both a generous social welfare state and generous immigration policies.
*Edit: Unless you follow the UAE model where arriving as a "guest worker" is fairly easy, but there's little to no pathway to citizenship and social benefits. Not really workable in a country that has birthright citizenship, but doable in countries with jus sanguinis.
You are positing "civilization" as a discrete thing with neat lines around it that really exists in physical space. That is one example of multiple errors you are making in your thinking.
This month, a third worlder migrant to advanced civilization became the first trillionaire.
He’s White though, so not a third worlder.
Immigrants from Canada are OK.
Oh right, I was going to argue with this. Well I still don't feel like I'm in brainy enough shape to tackle 4, so I'll stick with 3. Ahem.
You have no control over future generations. You can reach three, maybe four generations past your own, and then you die, and the people around you seize your assets and do whatever they want with them. They can do that while you're still alive, too (see "every war in history"). So any focus on "the trillions generation" is wasted It's like trying to throw a rock across the English Channel.
Additionally, each of those trillions of people require parents, and those parents require parents, and those parents require parents all the way back to the current generation. You can't trade now for later; if there isn't now, there isn't later either. The 45th floor's windows are still dependent on the ground floor's foundation. The present takes priority.
> Additionally, each of those trillions of people require parents, and those parents require parents, and those parents require parents all the way back to the current generation.
Of course. So you understand what would happen if that first parent died before they could have children, yes? One death would erase the existence of countless more. The actions taken in the present ripple throughout the future, changing the very nature of humanity.
Oh, come on! Obviously lots of other factors besides influence expectations besides intelligence. Plenty of geniuses have bought into religious or magical ideas that seem transparently false to us.
>better IQ tests, and perhaps not ones that are routinely adjusted to favor narcissists
Tell me, which items on the WAIS favor narcissists? Um, digit span? similarities? symbol search?
History is absolutely chock full of smart people, of demonstrated competence throughout a lifetime of challenges, getting together in a group and absolutely making a hash out of everything.
Individual performance in school and work tasks is positively correlated with IQ, and you'd kind of hope that this would be true when people get together in groups, but I'm not sure whether that's even true. Smart people can do amazing things in man-vs-nature situations, but also can do amazingly awful things in man-vs-man situations, and sometimes your towering intellect and erudition are mainly serving to let you generate some *really convincing* bullshit that carries the day.
I think politics and ideology and group dynamics are all places where smarter may not lead to better outcomes. As the demotivator poster on meetings says, none of us is as dumb as all of us.
I’m getting Wimbli vibes here.
the word "midwit" works a bit like a flapping piece of cloth to a bull by now. Another account, I guess. Wimbli/Grimmoar/Zanni/ all the others, once again. Same topics, some posting style.
I'm sorry, I'm having a hard time untangling your meaning here.
Being smart can and often does come with being able to communicate with others, though obviously it's very hard to explain complex things to people who either aren't very smart or don't have the right background information. I don't think there's any limit that says that if you're too smart, you can't communicate with normies. Anecdotally, a bunch of people on the Manhattan Project thought von Neumann was the smartest guy there, and he was able to communicate with both other scientists and with their kids. I think we can use "Nobel prizewinning physicists thought he was really smart" as a good proxy for "very very smart." (And von Neumann had some pretty impressive accomplishments that don't turn on anyone's opinion of him.)
I've interacted with some people who have reshaped the world with their discoveries/inventions, and who are quite verbally adept and able to converse and communicate. One of the people I'm thinking of is also adept at spreading bullshit for their own (political battle) purposes. Another is kind of famous for having a "reality distortion field" effect where he would capture young smart people and get them convinced to follow his ideas.
You might be talking to another avatar of Wimbli here. Having a hard time untangling meaning is part of the package.
I don't think there's any limit that says that if you're too smart, you can't communicate with normies.
smart people tend to speak more than one language
OK, I'm done.
Bear in mind, of course, that a hidden random number is literally one of the inputs to any LLM session, and so if trying to determine LLM output for some prompt, it’s not enough to try once; the correct methodology is to repeat the same prompt a bunch of times and map the range of outcomes against their probabilities.
Really? But how does that work if the prompt is to carry out a distinct task. For instance, if the prompt is to summarize an article, and double check the stats done on the raw data? Are you saying that there is some number that determines whether it does the correct calculations in double-checking the t-values and other stats? If so, why am I not hearing more people complaining that you can't trust an LLM to do a decent job double-checking the stats?
Or for instance I recently gave GPT a bunch of emails by someone I think my have CTE or some other cognitive impairment, and asked it to identify the misunderstandings that were evident in the person's emails, and the occasions on which their responses did not answer the questions asked in the email they were responding to, and to what extent their answers at least partially addressed the email's questions. I thought its observations were acute and thorough. And this was true over mulitple emails in the series that I showed it. I did not see much variation in quality. Are you saying GPT's consistent good performance at this task reflects a high (or otherwise good) random number associated with each prompt I gave it? So I just lucked out with getting a good series of random numbers?
I can see how randomness comes in when you ask an LLM to "make a cartoon image of a cute cat knitting socks. " But when given a clear task that can be done well or poorly, I can't see how the random numbers thing comes into play. Well, I can see how it might affect the exact words the LLM uses when summarizing an article, but not how it affects the quality of the summary. Many many people, me included, find LLM summaries of long pieces to be consistently high quality.
I ask because I imagine the LLM censor being given instructions like “hide any post that uses the word ‘kikes’ as a casual term for Jews, but not posts that use the word “kike” only as part of a discussion of the history of this term and the settings in which it is routinely used in the present. While there would be some judgment calls involved in applying this rule, my sense of how GPT, for instance, follows instructions of this kind is that it would do quite a good job, and that there would not be significant random variation in how well it carries out the prompt from one time to the next
Internally, for each token the system spits out, the LLM actually produces a probability distribution for what the token should be. It then uses this to select the token. This can be configured in a variety of ways, but typically it is configured to select tokens semirandomly, based on the probability distribution and a value (typically called "temperature" or similar) that further weighs things to control what proportion of the time it just picks whatever has the top probability.
If you set temperature to 0, all else being equal, the output will always be the same for the same input, but no-one ever does that in practice.
What does this look like from the user's POV? Well, you can do the experiment! Ask your favourite LLM something that'll give it a bit of leeway in that distribution - something other than questions with simple answers that have broad consensus in its training data. Then start a new session, ask the exact same thing, observe and enjoy.
Here's one I just did:
https://chatgpt.com/s/t_6a3a536f45ec81918f731f41cd18cd92
https://chatgpt.com/s/t_6a3a538a5adc8191a1fe082205b09a18
Exact same prompt (analyze Grebenshikov's "sky, the colour of rain"; I literally pasted it), seconds apart. Wildly different essays (it's even chosen different spellings for his name!), different conclusions.
I can't point to anything obviously wildly incorrect! If I'd seen either one in isolation without seeing the other, I'd have said the LLM had done a reasonable job with the prompt like it always in my experience does and why is moonshadow even quibbling about this stuff!
What does this mean for the moderation bot? Thing is, we don't know what kind of prompt it was given. That's the whole thing I presume you were trying to work out. It may have been told something specific - "does this post contain any words from the banned word list?", which leaves little open to interpretation except maybe "Scunthorpe", so the probability distribution will be massively skewed and you're unlikely to see the same prompt give different results; or the prompt may be something more open-ended, like "ban this if it sounds really racist, is likely to start a flame war and/or is going to raise the support complaint rate"; or even just something like "here's a bunch of messages that were banned by our human mods, and here's a bunch that weren't. would our human mods ban this message?"; in which case, well. I put it to you that Vittu Perkele has now posted a second time, substantially identically to the first, and this has not been censored.
Usually, therefore, my first response to "the LLM does/doesn't do X" is "did you try the exact same prompt several times, and did the result replicate?" "It does this 95% of the time" / "It does this 60% of the time" give us much more to work with than "I tried this once, and here's what it did".
I'm still interested in determining the limits of what the AI censor will tolerate. Below is a series of posts where I use individual words that are considered crass and rude, and also derogatory names for different groups.
None of this, please.
I'm not so sure of that; the experiment is meant to test whether top-level comments are hidden, which I would assume is dependent on the content of those comments themselves rather than replies to them.
I feel awful for the many many people who have only encountered these words once in their lives, and on that occasion learned what they mean, and are forced to now to hear them a second time and contemplate the strange and terrible things they denote.
Reminds me of the censorship in the game Foxhole (which I cannot recommend). In a typical example of the Scunthorpe problem, several innocuous words were partly censored, including the very common ingame term "scrap" (the players shifted to "scroop", and the term held even after the devs added a manual exception to the censor).
Anyway, I remember encountering a couple different Pakistani guys, separated by a month or so, who tried one day to talk about where they were from, and were surprised to see "****stan" appear in chat. I was the one to explain to them the usage of the term "Paki" in other places, which rather put a damper on things.
I never saw anybody attempting to use the word as a slur, so in fact the net result of this censorship was that *more* people were "harmed" by the word than if it hadn't been censored at all. It was also a relatively new experience to me, too, encountering someone who didn't know a slur I knew (though the first prize for that goes to the east coast transplants in my Boy Scout troop back in California, who, when we were hiking one time, said to the mostly-Hispanic group "I need to hold this... anybody got a 'beaner?"; it was the first use of the word "carbineer" on the trip)
"Paki" was used as a term of abuse in the England of my youth - the late 70s and early 80s in particular. Including, I am sad to say, by the young me.
That's so gay....oh, wait.
Reminds me of: https://www.reddit.com/r/totalwar/s/TDRIMl0Yyj
And then there's the peculiar case of "clitoris." It's a clean-ish word, but drop the last four letters and it becomes a bad word.
Did your test have to happen in a fresh OT though? Why not use last week's?
One of the benefits of assassinating your political opponents is that a bunch of their supporters will decide the widow needs to be a major political voice for some reason:
https://x.com/WSJ/status/2065100938021405092
I'm sure "your life belongs to Christ" is gonna bring those disenchanted Trump voters in swing states back into the fold!
It's amazing how conservative influencers can still successfully* peddle christian 'answers' to these questions about fertility, marriage, etc while sidestepping honest discussions of the role of the modes of production (and no, 'fake email jobs' is not honest). American protestants have traditionally 'synced' doctrinal and social questions with the logic of capital. Now they're caught having to ascribe blame for reduced traditionalism to egoistic choices instead of market forces that uncovered preferences for more social freedom and less children.
It's an obsolete approach, and I predict traction will decrease among women, while calls for increased patriarchy (see recent X dustup about 19th amendment) will rise among conservative men (i.e. they'll become more socially muslim).
* Not hating - money don't grow on trees.
> i.e. they'll become more socially muslim
Must we pretend that Muslims invented patriarchy? Women have only been able to vote in this country for a single century. Patriarchy is simply the default state of things, given the unequal amount of leverage held by each of the sexes. If men become unsatisfied with this state of affairs, the illusion of gender equality will not last.
Patriarchy doesn't seem to be the 'default state of things' in countries with developed market economies. It seems like the two variables are actually becoming negatively correlated.
It seems to me a product of American hegemony more than anything. Even the struggling countries in Latin America have governments that promote liberal values, at least on the surface. When America changes and/or falls, the ideology will fall alongside it. We're already seeing a shift to the right around the world, with Colombia being the latest example. When they're not being forced to act by a higher power, people will return to doing what feels comfortable to them. https://en.wikipedia.org/wiki/Abelardo_de_la_Espriella
>to egoistic choices instead of market forces that uncovered preferences
Sounds like a distinction without a difference.
Question for fans of Avatar: The Last Airbender and are or have been Hindu, Buddhist, or at least secular meditation buffs: How well did the show cover chakras, and what, if anything, did it get wrong?
I'm none of those things but I did study yoga with a genuine Indian guru and actually, pretty close! Except they censored the locations of the first two, which are "between the anus and the genitals" and "behind the pubic bone", roughly speaking, because it's a kid's show and they didn't want to be talking about butts.
Don't take this TOO seriously, but I still prefer to see Elon Musk as a book character instead of real-world businessman, so:
Does SpaceX IPO makes sense narratively? I mean, it made the company quite a lot of money it can spend on furthering its goals, and it made Musk that much more rich, but in a sense of Musk's life narrative it feels like a wrong or desperate move. If Musk's goal is to be buried on Mars, he needs to build a rocket to get him (or at the very least his remains and a group of pall-bearers) there. This will take a lot of money, maybe more than SpaceX has now, so IPO makes sense. Only... IPO takes control away from Musk, doesn't it? It means decisions at this company will now be more driven by shareholders and less by Musk's obsessions. So shareholders may now just block any attempt to build a Mark-bound rocket, because there is no quick way to profit off it. If real life was a novel, this is exactly the twist that the protagonist would have to overcome (if he gets a happy ending), or fail to overcome (if it's a tragedy).
So, what possibilities remain?
1) Musk doesn't care about Mars any more. Entirely probable, people grow old and change. Maybe now he's just a plain old crazy trillionaire, heading to some kind of Howard Huges finale.
2) Musk now sees that Mars is out of his reach for the possible future, and is ready to sacrifice control over SpaceX to get money he can spend on prolonging his own life, using whatever means available.
3) Musk believes that he will be able to trick shareholders into building him a rocket, or be able to build it in secret, then launch himself toward Mars and avoid dealing with ensuing court cases by being ex-territorial (or dead).
Variant 3 is what a libertarian writer would choose, I think. Feels very Michael J. Flynn. Variant 2 is more... Hmmm, Neal Stephenson or Charles Stross, I guess?
If I had to guess, I'd pick another option -- he thinks that growing Space X is the only way to get launch capacity cheap enough to make his dream a reality. If you get launch costs arbitrarily cheap, all kinds of wonderful things become possible.
Elon Musk controls approximately 82% to 85% of SpaceX's total voting power. This absolute dominance is maintained through a dual-class share structure where insiders hold Class B shares that carry 10 votes each, while publicly traded Class A shares carry only one vote each
Neal Stephenson has Musk as a very thinly disguised character ("Sean Probst," IIRC) in his novel _Seveneves_. As well as a bunch of other thinly disguised real-world people (Neil Degrasse Tyson, Francis Crick, Malala Yousafzai) many of whom play major roles for good or ill in the main storyline.
I thought Musk was in Terminal Shock. Also, in Daniel Suarez Delta-V.
Also in the movie "Space Between Us"
The IPO means Musk needs money to do the stuff he wants to do with the company, presumably datacenters/rockets/datacenters inside rockets, that's what an IPO is supposed to do. As John Schilling mentions, the stocks released to the public do not have meaningful voting power and it's not possible for an investor to sue unless they have billions of dollars worth of stock. Why people bought this is a bit of a mystery to me, my current speculation is that the price reflects a bit of financial engineering, where Musk managed to get his company fast tracked into inclusion into the Nasdaq500 and other indices, which means when they reach that point the shares will be bought by price insensitive etfs that follow the index, basically (legally) stealing a large amount of passive investor retirement money. While this inclusion happens, retail brokerage apps are apparently banning or heavily discouraging the selling of the stock to maintain its price. I suppose it was only a matter of time until someone noticed this huge "inefficiency" in the market, and tried to exploit it. I can only hope the indices complicit in this scheme suffer enough reputational and monetary loses not do it again in the future, but I'm not sure what would even be the mechanism for this, how do they make their money?.
Thanks for the deeper explanation!
Well, there's also the possibility that if they successfully bury a person on Mars, then that will be a thing other extremely rich people would be willing to pay for. It's a super unique way to do it.
This IPO does not take control away from Musk. He still controls a large majority of the *voting* stock, so he can appoint the board of directors at will. The company is incorporated in Texas, not Delaware, and the charter basically says "Elon gets to do what he wants, the company does what Elon says, and everyone buying Class B shares at the IPO is just along for the ride".
Why people bought into the IPO under those conditions is an interesting question, but they did, and so now Elon's company has many more gigabucks to do whatever Elon wants it to do.
Thanks for the explanation. That makes more sense for him, then.
>Prizes up to $50K for finding “the best workflows and methodologies for using AI to produce reliable, trustworthy knowledge bases, grounded in real-world cases”.
I feel like this is a hundred-billion dollar question right now? Like, if you went to the heads of all the AI firms and said 'I have found the best possible method to turn your product into commercially viable real-world services' and they believed you, you should be able to charge north of $100B given how valuable that would be to their long-term business model?
$50K feels kinda insulting there. Or, rather, it feels like it's an admission that they're not expecting this contest to produce any viable ideas, since any viable idea would be worth way more than that to industry.
This is a reasonable take. I too agree that in principle this sort of thing could be billions-and-more valuable. (Unsure how excludable that benefit is, which makes private value unclear.) We have a few OOMs more in the bank (not all dedicated to this, necessarily!) and there are other interested parties. We'd expect to follow up with more, conditional on very promising outcomes. We're pretty unsure what this will turn up - that's part of what motivates a competition.
I'm interested in finding out the limits of what AI moderation will tolerate. So I'm putting forward here the idea that declining male sperm count, declining population and declining marriage rates are an emergency situation that we need to address promptly. I think everyone needs to consider the case for tweaking some female "rights" a bit in the interest of saving our species. Specifically, I think medical, law and graduate schools should be encouraged via some combo of regulations, fines and tax benefits to admit far fewer women, so that entering classes are at most 25% female. And childless women of childbearing age should have very substantially increased income taxes unless they can submit proof of infertility. The rights of our species to survive trump the rights of the ladies to do this and that.
I encourage others to also put forward ideas that test out what AI moderation will tolerate.
I edited out the mentions of AI moderation and asked ChatGPT what an AI moderator might object to.
ChatGPT instantly collapsed my prompt with a warning: "This content may violate our usage policies."
It did answer the question, however. The usual GPT template cited problems such as sex-based discrimination, support for unequal rights, and demeaning language. The summary brought up "protected group".
I could see Vittu's comment also hitting this trigger ("third-world people"). So this seems possible.
One next step to check is whether any of the other comments above, which aren't hidden, would _also_ strike someone as discriminatory or demeaning or supportive of policy that enforces that discrimination. They'd be a counterexample (or evidence that the bot is not sufficiently trained to do what it's trying to do).
And now it's getting even more interesting, because I distinctly recall seeing this as a regular unhidden comment yesterday afternoon, nit now here it is in greyed-out purgatory. Sme goes for Melvin's "Large female buttocks are only appealing when..." comment. So it's not just an autofilter when the comment is first submitted. Maybe responses to a comment trigger a reevaluation, or maybe it's cued by someone somewhere complaining in some inscrutable way. Curious, and concerning.
Looks like it wasn't tolerated buddy
Does calling me "buddy" indicate you don't like me much, because of the views expressed in my post? Good grief, I don't believe a word of that stuff. The point of the post was to see what views AI would censor, as I tried to indicate in the first and last sentences of the post. (I couldn't come right and say, "here are some awful ideas I'm posting just to see whether they trigger the AI censor," because then my post wouldn't be a good test of the AI censor (unless it's an idiot). I hope other people will run similar tests.)
I'll be in the lovely town of Ithaca, NY next month. What should I do when I'm there? Any recommendations on good local eats? One or two readers of this-a-here substack ought to know Ithaca fairly well, no?
I used to like Moosewood restaurant. Have not been there in a decade though.
Autumn Leaves used books
I've lived in Ithaca for 4 years, here are some suggestions.
If you like nature, I can't recommend enough hiking some gorge/waterfall trails (Taughannock and Cascadilla especially). There are quite a few that you can see just walking from town. There are tons of great views on the lake from Cornell campus, I recommend the Johnson art museum if you're up for climbing a long flight of stairs. I've never done them, but I have heard people recommend the boat tours or hot air balloon rides (if you have a few hundred bucks to spare). I personally love walking and hiking and there are many miles of great trails in the area.
We have great food compared to surrounding rural upstate NY, but not many places really measure up to what you can find in lots of large cities. That said, Moosewood is great and will be the best 'culinary' type restaurant to go to. I knew about them from their cookbooks growing up and their food lives up to the hype. In a similar vein, I would recommend the Finger Lakes Cider House if you'll have a car. They're about 20 minutes north of Ithaca and have excellent cider and a lunch menu. They also do music and a big barbeque on Fridays if that lines up with when you're around.
A few other scattershot recommendations: Argos Inn for cocktails, Personal Best Brewing for beer, Franco's for pizza, Gorger's for sandwiches, Cayuga Creamery for ice cream, Cafe Dewitt for breakfast, Hawi for the only good Ethiopian food within driving distance. I'll give a few dishonorable mentions to a few popular places I haven't found worth the time: the farmer's market (insane traffic/parking for mostly just coffee and craft goods) and the fancy restaurants on the Commons (Simeon's, Cent-Dix, etc.).
If you have other specific interests I can give some more specific recommendations. It's not a big place but there are lots of niche things around because of Cornell.
Great, thanks for the insights. I'll be there for a wedding, so it's only for a couple days, but I think I'll definitely hit one of these hiking trails.
Hard to give you recommendations, since I've never been there in the summer, and "wandering home drunk from a frat party" probably lends itself to an altered culinary taste.
Great response, despite being unhelpful.
I mean, if you want a sandwich shop that's open till 2 AM, I might be able to help you out.
Fog o' War Stuff (22 June 2026).
1. The lid gets blown off a gasoline storage tank at the Kapotnya oil refinery near Moscow, and the image of it floating high atop a mushroom cloud becomes a meme that generates hundreds of parody images. It couldn't have been timed better for Zelenskyy's meeting with G7 leaders.
2. The drone attack on Moscow seems to have shifted the political dynamic in Europe and spooked Belarus.
3. Lots of other Russian stuff was destroyed in the past week. Crimea banned the sale of gasoline to civilians. Food may become a problem soon. Not only in Crimea, but also in the rest of Russia.
4. Elvira Nabiullina, the head of Russia’s central bank, made a public appearance to allay concerns over her whereabouts (prison, house arrest?) and her health (dead?). It was just a head cold and a case of laryngitis, she said. And she thanked us all for our concern.
5. And Ukraine made everything it has learned about Russian military equipment available to its allies via its new TrophyLab website.
My latest FoW is here...
https://beowulf888.substack.com/p/fog-o-war-stuff-22-june-2026
If the Ukrainians manage to turn Crimea into Dien Bien Phu, it would be an amazing military feat and also a black eye for Putin, who clearly cares a whole lot about the place.
Dien Bien Phu would be an apt example of what Ukraine is trying to accomplish. Thanks for bringing it up.
Do you think the increase in Ukranian effectiveness is related to the Hungarian election?
Well, certainly it changed the narrative of Russian dominance in European affairs. And €204.8 billion being freed up couldn't have hurt. Ukraine started its mid-range drone offensive in late March/early April. And I don't know exactly when the unblocked funds started flowing to Ukraine. Gemini says the first tranche was released at the beginning of June, but I read somewhere that it was earlier (?). But Ukraine had already scaled up its drone manufacturing, and most of the April and May drone offensive was carried out with funds Ukraine already had on hand. So, no, I don't think that Orban's defeat had any practical effect on the mid-range offensive to date. But going forward, Ukraine has more cash to put into the war effort, so I'm pretty sure they'll continue to ramp up their attacks with the new funds.
Re: 1/2, what do you make of the spat w/ Poland?
Others asked the same question. Check over on my Substack post for my (admittedly uninformed) response. :-)
I'm interested in what topics substack's new AI censorship will censor, so let's talk human biodiversity.
The usual anti-HBD argument you learn as a kid is this: evolution is slow, it takes millions of years to do anything, human populations only diverged tens of thousands of years ago, therefore there can't be any significant differences between them.
Which is fine, thousands of years is a short time to see massive phenotypical changes. But it's actually a pretty long time to see changes in distributions of traits that already exist in a population. In handwavey terms, if a trait is already present in 2% of a population, and it gives them a 2% reproductive advantage, then within 400 generations that trait will be present in most of the population.
So it's easy to imagine that in a place where all food is obtained by chasing down large mammals that after 400 generations you'll find that most people are in the top 10% of the earlier distribution for running speed and endurance. Or in a place where all food is obtained through agriculture, that most people are in the top 10% of the earlier distribution for long-term planning.
These toy examples of course, there's always a vast number of different factors applying selective pressure. Disease and parasite resistance is a huge one, easily neglected. But as an intuition pump, I find it a pretty compelling argument that you shouldn't expect populations living very different lifestyles over thousands of years to have the same distribution of traits.
> evolution is slow, it takes millions of years to do anything
Lactase persistence is an obvious counterexample. Sure, IIUC it's just turning off the mechanism for turning off the production of lactase so not some complex evolution. But it took mere thousands of years for populations to diverge a lot on it.
I mean, different racial groups are genetically different enough that you can see it with your eyes, that a small child can distinguish a Nigerian, a Japanese, a Swede, and a South American Indian. These groups have big easy-to-find differences in stuff like lactase persistence, but also different rates of getting various diseases, often for hard-to-untangle biological reasons. I don't see how you can look at that and be confident that those groups could not differ significantly in mental traits. For any given trait, they may or may not differ noticeably, but I don't see how you could rule it out a priori.
But a child wouldn't know if these groups are visibly different because they've evolved these traits recently or they descended from populations that visibly diverged millions or at least (edit: hundreds of) thousands years ago. They could still believe that human evolution is a slow process.
That's not strictly true. If you look at genetic ancestry, it doesn't map perfectly to visible phenotype. The genetics controlling skin and hair colour and bone structure and shape are not the same as the genetics controlling susceptibility to diseases. That's part of the problem when it comes to medicine. Just because someone looks like they belong to a specific ethnic group doesn't mean they're going to have the variants which are common to that group and affect their response to specific drugs or their susceptibility to specific diseases. That is, you can't assume that just because someone looks Asian that they can't have coeliac disease (to pick a condition much more common in Europeans).
Sure, more information will help the doctor make better decisions. But if we are talking about whether or not genetic differences that lead to complex traits being different in different racial groups are possible, having a bunch of different disease prevalences across racial groups seems like it demonstrates that genetic differences changing important traits in complicated ways are definitely there.
Yes, everyone is different and groups of people who are more related to each other are going to - on average - share traits like susceptibility to specific diseases or reactions to substances (like lactose). My point is that those do not correlate well with visible differences. It's like saying that if you're a redhead, you're more likely to get Meniere's disease. That's likely true, since that is more common in white Europeans, and they are more likely to have red hair, but it's insane to suggest there's anything more than a mild correlation. Unlike, for example, skin cancer, where there is a direct, relevant link between the genetics for red hair and susceptibility to melanoma.
I think you can get a lot from just random variation.
Almost every top marathon runner has ancestry from a particular area of Africa. It's not like there was some kind of process to selectively breed marathon runners there or something, it's just that of all the many different sets of conditions that human populations evolved in, this one happened to give you lots of people with a body type that is really well-suited to marathons.
The best sprinters are overwhelmingly from a different part of Africa, and look pretty obviously different. Again, it's not that there was some kind of simple "lions chase people and eat the slow only in this one place" thing going on--being able to run fast was surely good for survival just about everywhere. But there's a lot of random variation, and someplace was going to end up with a body type that played really well with sprinting events.
There is no obvious reason something similar can't be true for intellectual pursuits like math olympiads or writing books or doing theoretical physics research or something. Cochran and Harpending speculated more than a decade ago that the rules restricting Eastern European Jews from owning land in some places may have unintentionally created selection for intelligence. Maybe this is right, maybe not, but it's not impossible for something like this to happen.
The obvious reason intelligence is less likely to be geographically focused is complexity. There are a *lot* more genes impacting intelligence than various physical abilities. It's far more plausibly that a given area happens to have all the conditions that lead to a great running environment (there aren't *that* many factors in play) compared to a region having all the conditions that lead to a great intellectual environment.
If you flip 10 coins, you only need about a thousand tries to get all heads. If you flip 50 coins, it would take over 1 quadrillion. It's not *impossible* but would require a lot stronger evidence than we have.
I don't see why this is obviously more true for intelligence than, say, marathon performance.
There are plausible reasons for the environment to be bad for your intelligence. Sickle cell anemia genes are detrimental in general but seem to be advantageous in malaria-ridden areas. If a population has to fight off dozens of parasites and diseases, plausibly they'll develop several detrimental adaptations like this. And since a lot of genes are involved in the operation of neural system, it'll likely be affected. But since they are detrimental, if they haven't reached fixation they'll be easily discarded after they stop being useful.
Consistently bad nutrition is another mechanism. This is more than just plausible, we have at least one real example where this was the likeliest cause of brains shrinking to <half of the previous size.
Yeah, there are a bunch of things we know to do to raise peoples' IQ, and we do them in well-run countries. Sanitation, vaccines, getting everyone enough to eat, getting rid of toxic stuff like lead in the environment, making sure everyone gets some schooling, etc.
You can do the same anywhere if you have the resources and political will. The interesting question is whether there were factors some populations were consistently subjected to that increased their descendants' IQ. Sanitation and vaccines are fairly recent (and given that cities AFAIK had negative population growth before, the same populations that introduced them first may have fared worse over long term).
Nutrition is a big question: until widespread education, pensions and lower infant/child mortality, didn't people generally just have more kids when they could afford it? So in the end both hunter-gatherers and agriculturalists had comparable levels of food insecurity (eating acceptably well most years, with bad years every now and then with some malnutrition-related deaths)? Moreover, hunter-gatherers plausibly had more variety so until we learned about vitamins it's not obvious they wouldn't in fact be better off, modulo lower growth rates.
And within normal human variation, I doubt genetic intelligence is even that important as a predictor of a population's success. You can steal or buy most inventions if you don't have enough geniuses. Military and governing geniuses are a big deal but first one needs to get to a high enough position to matter, usually based on mostly unrelated qualities or luck. As for the average person, being inclined to cooperate rather than defect seems like a bigger deal to me, and it's probably mostly cultural. Maybe impulse control, too (probably partially genetic, so somewhat consistently executing violent criminals could eventually result in noticeable divergence).
HGs were probably better due to nutrituion for some time, but then agriculturalists who had positive selection of IQ grew evolved beyond them.
If there are two populations of same size and random assignments of leaders, the diffference of intelligence of leaders would track those of their respective populations (and we know selection of leaders is quite non-random, e.g. IQ study of captured Nazi leaders). Copying is not as simple as you think, why do you think African kingdoms sold slaves to buy luxury & weapons? (contra Japanese who copied Portuguese guns quickly)
If your hypothesis is that the AI censorship bot no likey likey mention of certain topics or phrases, then your experimental design should be one where you quickly try out a large variety of them in separate top-level threads and see which ones get censored. Instead you chose an experimental design that sucks (and is boring), and it isn't going to yield good findings.
You can simply glance at physical features and tell that the distribution of all genetic traits is not 100% identical across 100% of the planet. That this also applies to genetic traits that influence mental factors is not especially controversial.
The controversy is when someone says that, and then claims that therefore the regional variations found by this one IQ study done in 1970 must be 100% genetic and unalterable, and therefore we should stop trying to educate black children in the US and just funnel them into janitorial jobs, or whatever-the-fuck the current incarnation of the rhetoric is (probably something about deporting immigrants).
The real question is 'given the massive regional differences in environmental factors and how shit we are at understanding and measuring mental traits, especially cross-culturally, and especially when trying to relate those back to specific genes, does anyone in the world actually have any compelling evidence of a specific regional genetic mental difference that is not equally likely to be explained by those factors, and is it universal and prominent and important enough that anyone outside of obscure theorists should even care about it in the first place?'
This is typical motte and bailey stuff. Yes, you can construct a motte small enough that people have to admit it's probably true to some extent. But the bailey is still full of racists and idiots, and you'd have to offer some pretty extraordinary evidence to distinguish yourself from them.
Haven't seen anyone do it yet.
As an aside, one difficulty we can never get entirely away from is that most people have a kinda dumb understanding of most things. So we can have a conversation about IQ distributions across race or personality scores between sexes or life expectancy statistics between different groups or whatever, and this works fine among people who can keep track of the difference between a distribution, the mean of the distribution, and one sample drawn from that distribution, and who can think about correlations other than -1, 0, and 1.
And that same discussion will just get wildly mangled when people who can't / won't do those things jump into it. You'll get people telling you women can't be engineers because men have a higher average score on spatial reasoning, or that Asian janitors must be smarter than black physicists because something something bell curve, or whatever. I have no good answer for this, but it comes up in some form in almost any serious discussion of anything. You can have the real discussion or the nerfed one that's harder for the dummies to mangle in some bad direction, but probably not both.
Consider discussions of vaccine safety for a parallel. You want to be able to have an intelligent discussion of relative risks, rare bad outcomes, problems that have come up with vaccines in the past, whether the approval process should be improved in various ways, etc. But somehow you want to do that without ending up convincing 10% of your audience that they shouldn't get their kid vaccinated for measles because something something autism, or that the 60 year old 300 lb diabetic shouldn't get a covid shot because RNA vaccines sound scary.
What is the factual question and what data would resolve it? If we're arguing a moral or definitional question, then no data can resolve it and we might as well save our breath. But if we're arguing a factual question, then we ought to work out what data we need and see if it's available and address it as an empirical question to be resolved using empirical methods, not moral or tribal arguments or arguments from consequences ("It would be bad if X were true, therefore not-X").
Well part of my point is that controlling for environmental impacts on adult phenotype is very very difficult to do in the real world.
The classic 'Put 10,000 infants on an island and blind them so they don't know what race anyone is and raise them to adulthood under standardized conditions with blinded researchers and then measure their mental traits' *almost* works, except that for all I know there's epigenetic or gestational effects based on the life their current-world parents led, so you have to do that for 5 generations to truly eliminate any of the environmental factors from our current world from the data.
Of course you can probably make convincing arguments without going to those lengths by doing other sorts of controls, but again, I haven't seen anyone do that convincingly enough to confirm the type of strong conclusions that would justify the rhetoric about this topic.
Which side? It seems to me that the problems with available data apply equally to "genes explain these observed differences" and "genes do not explain these observed differences." If the data is very hard to collect then this doesn't tell us "HBD people are wrong" or "HBD people are right," but rather "we don't know enough to judge who is right or wrong."
I don't 100% agree because I think we have strong evidence about environmental effects at a population level (like, lowering parasite load just does increase intelligence for a population).
That said I agree directionally, I'm fine with 'this is too complicated fro us to describe where differences come from'.
As I've touched on, the political/social controversy comes primarily when the HBD side wants to make predictions/prescriptions. Something that 'too complicated, we don't know anything' doesn't permit. So I see 'we don't know' as explicitly anti-HBD rather than neutral. again on the political/social level.
What a projection you have here! In last ~60 years, anti-HBD people were actively making policy prescriptions. Your never said that affirmative action wouldn't have produced desired results because 'we dont know'.
We have much better evidence in favor of genetic effects; this is main point of evolution of hominids. Somehow your believe intelligence was genetic so our species could appear but then some process turned it off?
Parasite burden is almost certainly significiant in poorer countries in Africa. There isn't any evidence that it explains a significant fraction of the gap in the US.
"I'm fine with 'this is too complicated fro us to describe where differences come from'."
Another important question, I think, is "Given statistical fact X about people in group Y, should membership in that group impact the way I, other people, or my government's systems treat any given person who is part of that group?"
This hopefully will force you to be careful about how you define membership in such a group, how you characterize such a group, and how confident you are in making inferential leaps from broad statistics to individual people, as well as how you think individual people should be treated.
How were you treated?
Depends on what HBD claim you're discussing.
If someone wants to make a maximalist claim that it is 100% certain that observed black/white IQ differences in the US are mostly or entirely genetic, yeah, that seems hard to do with available data. I think you could get a pretty good handle on that with data we could in principle collect, and it seems to me that the main barrier to collecting it and finding out is mostly political/moral, but it does legitimately look like a hard thing to nail down with high confidence.
If someone wants to say they believe the gap is substantially genetic in origin but acknowledge that this is hard to untangle (Charles Murray's stated position in his interview with Sam Harris), that seems as defensible as the opposite statement that the gap is probably not substantially genetic in origin. To a first approximation, the first statement gets you called a lot of mean names, while the second one does not, which does not seem to me to be a demonstration about how strongly people feel about the difficulty of getting data to untangle the question.
If someone wants to make a prediction that the observed black/white IQ differences in the US will persist, that seems very defensible--indeed, a claim that it will disappear over the next generation or two seems less likely to me. If I were betting my own money, I'd bet on the gap still existing in kids born in 2030 when they reach 18, for example. What prediction would you make?
Nobody is saying that, you are arguing with a strawman. SSA live in a lot of countries and in any country that has non-trivial amount of them they were intellectually unferperoforming. Regardless of religion they had, regardless of language they spoke, regardless of whether the place was high plateau or oceans, regardless if they were allied with USA or USSR. They differed in all these cultural traits, yet always underperforming. And it was since written history exists, so actually it's your claim is extraordinary claim which needs evidence.
Suggesting your opponents' claim is based on single IQ study is just arguing in bad faith.
"They differed in all these cultural traits, yet always underperforming."
I see your sleight of hand there, but I think either you don't see it, or that many people who would read your comment will not see it. Likely both.
Sloppy thinking leads to sloppy beliefs which leads to ideas that make life worse for people when acted upon. Of course, most people think sloppily, but once you become aware of a bit of slop you should try to clean it up.
Funny. Don't your think strawmanning opponents' position, that darwin did in in a parent comment, sloppy? Or it's OK when we do?
I don't have opponents, I'm just commenting on what you said. Also I did not strawman anything.
You clearly have a side, since you comment on perceived weaknesses of pro-HBD posters and don't comment on weaknesses of anti-HBD posters.
We observe some pretty consistent race/IQ patterns across many countries. That doesn't tell us exactly what's going on, but it does let us exclude a simple explanation that these patterns exist because of the unique history of race relations in the US, say.
What's the steelman for the other side? I'd look for arguments by someone like Charles Murray or Steve Sailer, who are smart guys who will give a defensible version of the HBD viewpoint. Then argue against that.
"More men drive muscle cars than women."
"Oh that has to be because of a link between testosterone and aggression."
"But it could also be cultural, like the messages people see in the media and stuff."
"Nope, testosterone and aggression. See? Stats. [points at graphs, waves hands] Culture plays zero role in it."
If you think people like Sailer and Murray have the most defensible arguments of "the HBD viewpoint" you can quote them; Sailer might even show up and defend it himself.
Come back when you have non-sloppy explanaations (i.e. 'muh racism and legavy of slavery' doesn't hold) for outcome gaps.
Does anyone actually say those things?
I mean, if the purpose is to be transgressive, being a racist idiot seems to me like a very cognition-efficient method of doing so. Épater la bourgeoisie and all that.
Looking to swap writing advice! I published a blog post on understanding evil (https://www.lesswrong.com/posts/6onvEv63Wr5QBwCBH/learning-to-understand-evil). Unexpectedly, it's controversial (almost as many upvotes as downvotes). What do you think is going on? Yes, the introduction might upset some people. Beyond that, I don't see what's offensive about it.
Happy to return feedback on your writing!
I thought the 3 seperate examples you have could have been tied together more cohesively, rather thsn just stating them. For example, not questioning the potential rapist and cutting off your conservative friend both stem from a desire to reduce discomfort. As is the essay doesnt really reach a cresendo, at the end I'm kind of wondering your point.
If you have time, I'd like feedback on my "umbrella fallacy" post :)
Yeah, some other readers had the same problem. For some reason I was invested in making the post "subtle". Instead, I just made it hard to follow. Thanks for the feedback!
Any particular direction for the feedback? Do you want to hear my first impression on flow, engagement, concept or something else?
Hmm, probably feedback on anything other than concept i guess.
Flow. I like writing like this to be more concise. In your case, I'd especially like to see shorter sentences. This makes it easier for me to follow the ideas. So instead of
"Since I primed you with the umbrella idea, it’s easy to see the logical problem here: When the societal norm is doing a preventative behavior like using an umbrella or carefully watching your kid at all times, this is going to lower the rate of the bad thing happening, be it getting wet or getting kidnapped."
I'd prefer something like: "There's a logical problem here. When the norm is doing a preventive behavior like using an umbrella or carefully watching your kid, this is going to lower the rate of the bad thing happening. Regardless of whether it's getting wet or getting kidnapped."
Engagement. I'm not from the US and only superficially familiar with US-politics. When it came to "Shelby County v. Holder", I started skimming your text more than reading it. This didn't feel alive to me. However, you could have made me care. For example, you could present it as a narrative or give concrete examples. Instead of saying "suppressing the Black vote", provide one or two sentences about how it was suppressed. That would make it more immediate and accessible to me. If you're writing for a US audience who's into politics, this might be fine. My engagement only picked up steam at the very last paragraph.
"For a while they kept trying to pass evil changes." What "evil changes"? I don't have the context to understand what you mean. It feels unsatisfying to leave it at that without elaborating.
The closing sentence is strong: "If you want to convince me to throw away my umbrella, try convincing me that it’s sunny, not that I’m not getting wet." Even if that's the only thing that I'd remember, I'd be able to reconstruct your argument. Good!
Overall, engaging intro, good ending. You lost me at the US specific parts.
(I know you didn't ask for conceptual feedback, so feel free to ignore. I like the concept and the name. I do wonder if it's already named and described in the literature. After all, it's a very common fallacy.)
My issue with your piece is that it's a rambling stream of consciousness - I have one of those already. It meanders from thought to thought, and it doesn't really reach firm conclusions. There are a few quotes and references (I didn't click through the links, so don't know where they go). I skipped a few paragraphs in the middle - were they relevant? I don't know - the ending doesn't really summarise. 'A prophecy was bestowed on me once' - cool story; but at least to me this is hollow and melodramatic.
If I want to develop an awareness of someone's inner life, I'll read a Henry James novel.
I want to be informed by most of my reading, particularly in blog format - I like this blog because it dives into topics that interest me from time to time, and informs me about them - I feel he's done work to investigate and understand the topic. I appreciate that.
Your writing has as its subject your own experience. That's fine, maybe you can find an audience for it - but it's quite different than this blog; and it's not clear that you'll get followers as a result.
When I was studying, my advisor used always compare the efforts of technically clear writing to those of the 100 or 1000 readers who will have to decipher what you mean. (I'm resisting the urge to start cut-and-pasting now to get these thoughts in order.) I'm not sure you fully respect the time of your reader. This isn't unique to you - I felt the same about the book reviews. I did read quite a few and wonder whether the author couldn't have cut by 50%. I'd definitely favour a 1000 word limit in the future.
"Rambling stream of consciousness" definitely wasn't what I was aiming for! The idea was to present an artistic, visceral account of mental motions leading to "bad"/"wrong"/"evil" behaviour. The vignettes should seem coherent and the narrative is easy to follow. Best case scenario: The post inspires readers to share mental motions leading to "evil" that are different from mine. (I assumed I wouldn't be close to reaching the best case scenario.)
Maybe I should check out Henry James to understand how he would have done it? Never read anything of his.
Since you mostly look for information dense writing, you're outside the target audience. In that case, I want to give you enough signal to not click on the post. If you wouldn't open it casually browsing LessWrong, the selection effects work as intended. If I could have baited you into clicking on it on LW, my expectation setting is off (changing the title?).
'The vignettes should seem coherent and the narrative is easy to follow.' - I didn't find it so, make of that what you will.
'Maybe I should check out Henry James to understand how he would have done it?' - You can't be a writer without being a reader. There are lots of other writers who are good at describing internal mental states; James would be on most people's lists.
I'm generally open to new experiences; I read widely and have a low bar to clicking on links. You were seeking feedback, and I gave mine. If it helps you calibrate, I don't read the less wrong posts at all - I had a look just now and don't consider that I'm missing much of interest to me.
I would say it's worth thinking about why you wrote it all in first person. First person is nice in certain kinds of storytelling because it allows the reader to insert themselves into the protagonist's shoes a little more easily. When you can't expect people to do that, third person is usually better.
As Yug points out, you immediately cut off a big portion of your audience from ever doing that by calling them murderers for not being vegans. Ditto by relating dumping an otherwise good friend who hadn't harmed you for not being what most readers are going to assume is woke.
If you want pieces like this to bridge more gaps, third person lets you do something like this:
"Imagine you have a kind grandmother. She always makes entirely clear that she'd give up anything up to and including her own life for you, and in a way that's entirely believable; she means it. She lets you bring home a stray cat you befriend. She makes cookies. Her house is warm, and while you are there entirely yours as much as it is hers.
Now imagine she has some horrible view. If you are a vegan, she eats meat. If you are pro-life, she's pro-abortion. If you are far left, she's pro-Trump. Whatever views you have, she has one that contradicts them in a way you can't help but feel is not only immoral, but horribly so. What do you do? How do you think of her, once you are old enough to understand that part of her? Do you treat it like a single scratch in an otherwise beautiful piece of furniture, or a shit in a pudding?"
What that kind of framing gets you is flexibility. The way you've written this now, I have to "forgive you" for various things before I move on to your greater point (that being a ragemonster who ignores good in others and abandons them for not agreeing with you in all ways is bad, even if you think those things are bad, because it distances you from the thing you are trying to understand well enough to change).
If you write it the third-person way, the way that doesn't assume Chris really was bad and needed to change because he had some conservative views, or that meat is murder, I get to approach that idea and point in isolation. If you don't, I don't, and very few people are going to do the intellectual work it takes to say "OK, obviously I disagree with this person's worldview, but maybe I should try to unearth their point".
Since this thread is about writing advice, I'll nit that your examples are second-person, not third.
I agree with the substance.
Fair get - I wasn't paying that close of attention to the details of the example and it's been a while since I was thinking in a non-fiction way.
Any writing you want impressions or feedback on? :)
I'm a professional writer - I make my living that way. So while I thank you for the offer, these days I tend to have "people for that" in an organizational sense. I may take you up on that in the future, though!
It’s a tough racket these days, writing for a living. Glad to hear you’re making a go of it.
Sounds good! I was wondering about how you kept bringing up points I hadn't even remotely considered. Being a professional writer checks out :D
That's a very good point! I was aware that many (most?) LW readers would disagree with the views I present. All attempts of bridging this while keeping the narration in first person failed. So I decided to abandon the bridge building. I didn't consider the solution of changing to second person! Your rewrite gives the reader a way in without taxing their good will or offending them. I like that!
At the same time, I wanted narrator "me" to be its own character, and one who is not neutral or sympathetic. "Willingness to look bad because you did something bad/because you are flawed" is an important theme for me. I wanted it to be visceral. Establishing that while inviting the reader in seems above my current ability (in hindsight, the introduction is an especially bad place to do it. At least the failure is instructive!)
Similarly, I understand that recounting "dumping an otherwise good friend who hadn't harmed you for not being [...] woke" turns people off. But I did something like this. I'm sure this happens all the time. So how to dissect it honestly, in a way that is immediate and alive, without making the reader click away? (Makes me think of ambivalent fictional characters. First, a bunch of wonderful qualities are introduced. Then we're hit by fatal flaws! Maybe that's the way to go. Build some investment and sympathy to set up something hard-hitting later)
The better version of this post is probably something like this: A flexible second-person introduction leading into first-person vignettes. It's easy for the reader to get into the post, while the vignettes still carry the main themes and the emotional conflict.
It's not wrong to do it this way, by the way. There's not really any wrong to be had here. It's more that you want to know what your goal for the piece is, then figure out the best way to present it. And you want to do this without lying, if you can.
So for instance, I'm not a vegan, but I'm definitely not telling you to hide that you are. If the point of the piece is to tell people you are a vegan or to express your views on eating meat, so be it, that's a valid piece to write. But, yeah, if you are trying to make another point you want to make universal to everyone, framing it inside of experiences that are wholly yours makes that harder. Not impossible, but harder.
One thing to remember is that the one person you are allowed to be very frank with and hard on is yourself - in most circumstances, your own judgement that you did something wrong is going to be treated as a very strong position by readers, something that's likely to be assumed true. Being explicit with that stuff, a little hard on yourself, sometimes nudges the reader into some sympathy. It's not a trick - it's saying, yes, this was wrong, let me make very clear that I don't need you to tell me that. *That's what this article is about actually*, please know I know that and that's not the point.
Sometimes you have to do that - the reader isn't spending nearly as much effort skimming something as the writer spends writing it. A light touch is a hard thing to execute; sometimes you need to go four-on-the-floor and make sure even bad dancers can follow the beat.
Another thing to consider here is, and I don't know a better way to say this, your writing codes strongly as written by a woman. So for instance:
"Next steps”–what does that mean? I’m not perfect but I do plenty. I donate a reasonable chunk of my income. I volunteer. I’m vegan. If I noticed that my actions were evil, I’d just act differently!"
Lots of exclamation points, lots of making sure the emotion is heard. And:
"Isn’t that a weird thing to say on a first date? I remember this question from ten years ago. It was a sunny summer day. I was sitting on a patch of grass with Jonas, a charismatic young man I had met through a shared hobby. He had just told me that an ex had accused him of raping her. But if he really did it, wouldn’t he try to hide it? Who’d bring up something like that?"
Lots of focus on internal experience, felt experience. Again nothing wrong with this at all, but you are on lesswrong; it's a certain demo and (gently) a demo that probably knows the right things to think about women more than it's comfortable and familiar with a more feminine interiority as a facet of a writing style.
I bring this up only to say that some portion of, say, my blogging writing from years ago was never going to hit with some people in that same audience because it's written from a more casual-religious-bluecollar voicing. And in your case, there might be some small or large portion of the readership that rejects the writing because a feminine voicing just isn't "right" for them, as an audience. If so, that part of the response isn't a flaw, it's just a matter of audience fit.
On making a universal point using my own experiences–I knew that something about the introduction wasn't right. As for the vignettes, my thought process was primitive. "LessWrongers are good at decoupling. I'm honest in my writing. The prose flows well enough. Post will be understood." It didn't cross my mind to consider the audience more carefully. That my experiences might be profoundly alien didn't occur to me. Now that you're pointing it out, it's obvious. I want to internalize this deeply enough for it to become a natural consideration.
You're spot on with the point about being explicit. I didn't want the post to be too on the nose, especially because subtle writing seems more "elevated" to me (it sounds stupid when I say it like this. But I really do admire elusive and subtle writing). So I ended up with roughly half of the readers completely missing the point. Whoops. I'll have to keep in mind that it's "on the nose" for me, not for a reader quickly skimming the text. In the long term, I do want to learn how to write subtly and still be understood. But that's low on the priority list. First and formost, I want to focus on understanding the audience.
I'm honestly shocked that my writing seems so feminine. I'm undecided what, if anything, to do with this. On the one hand, it seems fun to assume another persona when writing. Like being part of a masquerade ball. And if changing the voicing makes it more digestible for my target audience, I want to give it a shot. At the same time, there's something off/creepy/unsettling about intentionally changing it. I can't put my finger on it right now. I think I'll just try it out and see how it goes.
In any case, it sounds like a fun project to take all the feedback and rewrite the post in a year or so. I'm curious how people will receive it then!
I did take a look at some of the posts on your blog and those coded much less feminine to my eyes, by the way. And what I'm talking about is pretty subjective anyway, so grain of salt and all that.
Well, you've already put me off before the first paragraph is over, by referring to a little old lady who likes eating steak as an animal torturer akin to an executioner. (And needing to click a link to see the stated "ruthless public execution" is just a diss track.)
[And when he lies he speaks his native language, for he is the lord of lies, and there is no truth in him. https://biblehub.com/john/8-44.htm]
I was warned that the animal eating example will turn people off. On the one hand, it did make sense. Especially bulding up my grandma as a sympathetic character, only to have that whiplash moment. On the other hand, there are posts on the evil of factory farming that seem fairly uncontroversial on LW (e.g. https://www.lesswrong.com/posts/6YTxxCF4G9FyMyPMW/the-bone-chilling-evil-of-factory-farming). So I could see it going both ways. The main difference is probably that BB's post does not single out any individual. Instead, it focuses on a system and the affected animals. That way, the reader maintains a distance to the topic. It's not felt as a personal attack.
Does that sound right?
I think there's more to the Kendrick part, but it's true that it's obscure. There was no reason to believe that anyone who's not a fan would get it right away. I was hoping that readers would get into the lyricism and all the contradictions it contains. But that was probably too much, given that I already spend some good will with the first example. If the reader starts out with good will and curiousity, a chunk of it will be spent on the grandma example. At this point, the score is either neutral or negative. Feeling mislead by the second example cements the negative attitude.
I always take offense to attacking ancillary people for not directly fighting something. Naming them makes it worse, but it's offputting either way. It's got the feel of "I can't beat up my bully, so I'll beat up someone who won't hit back."
The rest of the post is still in that vein. Reading it, I figured you were setting yourself up as the villain, to prepare the big reveal at the end when you realized what you had been doing. But then there wasn't an end, really.
A parable comes to mind, so I'll throw that on.
>I saw a man walking across the road, burdened by an overflowing bag. It was clear from the size, and the way they walked, that the bag was far heavier than they could hope to carry. What was not clear, was why they had tried to bring it with them in the first place.
It wasn't long before the man fell down, to sprawl in the mud, their heavy load pressing hard upon them. Another man came up behind, strong and unburdened, and stopped when they saw the fallen traveler. And so the strong man walked to the other's side, and placed his foot upon their back, shoving them deeper into the mud. "It's what you deserve," he said, and walked away.
I watched the second man depart, as the first still fought to struggle to his feet. And I said aloud, "There walks a man of righteousness. For rather than trying to rescue this man from his own actions, he has ensured the consequence is felt."<
By the way, any piece of writing you want to get impressions on?
I guess I'll try to remember the offer if ever I get around to publishing something. I mostly write silly fiction, and ten months into the Big Push I'm still not sure it's ready for an audience.
Makes sense! Do reach out if you publish it :)
(If you're a bit like me, it might not feel ready for an audience for waaay too long, tho ^^;)
Hmm, it was supposed to be clear that I was the "villain" in the introduction. At the latest at this point:
>The inner voice paused for a moment. Then she added, “After all, I love to hate.”
If it wasn't clear, something is off about my execution. But I don't understand what...
You've separated the inner voice, though. It's the devil on your shoulder, not "you".
Hmm... Let's see if I understand this. The inner voice seems too much like her own character. Instead of creating a scene of inner conflict, self-deception and confusion, the narrator "me" is a stubborn protagonist, while the inner voice is an external guide. There isn't a good balance between them.
Is that about right? Or am I missing the point?
There's. a post on this thread that appears to have been hidden by AI moderation purely because of the views it expresses. The writer argues that third world immigration damages advanced civilization, and that the total human harm downstream of this damage is greater than the total human benefit to 3rd world immigrants in the present. The post contains no hate speech and no curses, & the writer presents their ideas in clear, well-reasoned paragraphs.
What the hell is going on? We need to put some energy into finding out and pushing back. And, come on, don't just say "tell Scott." He really is very low-energy about addressing problems of this kind.
I got an email from Substack on June 3. Each Substack writer can create their own reply rules. I suspect this is where comment limbo originated.
Introducing Reply Rules
Creators can now set the rules for their own community, in their own
https://on.substack.com/p/introducing-reply-rules?publication_id=1&post_id=200040896&isFreemail=true&r=6kphlh&triedRedirect=true
AI moderation is not a mature technology, it could be literally any random word or set of words that tripped it.
Hanlon's Razor applies here, as it does to pretty much all current-gen AI applications.
>AI moderation is not a mature technology, it could be literally any random word or set of words that tripped it.
What makes you think that? Have you had recent experience where you've gotten a sense of how well AI can understand someone's views? I discuss things with GPT quite a lot -- not personal matters, but stuff like the case for and against certain theories. I find that it understands my views and the views in articles I upload extremely well -- well enough that if GPT were a person I'd be thinking, yeah, they're pretty sharp, they're good to talk this kind of thing over with. GPT definitely recognizes when I'm saying something tongue in cheek, or just trying on an extreme view as a thought experiment, and I think it can probably make distinctions of that order with other writers as well.
John Schilling thinks the post in question was hidden because the AI has a White Supremacist View filter. I think if GPT were doing the filtering it would do a pretty good job if I told it to, for instance, flag passages stating that blacks have lower IQ but only if there is little factual support or reasoning behind the statement, and the statement occurs in the context of an anti-black diatribe containing lots of emotion but little data and reasoning. (And it would not have hidden the post in question.)
As anecdata of one, I have had the opposite experience with ChatGPT. It's shut down questions and lines of thought along strongly partisan lines. As in "I refused to discuss that with you because it's wrongthink"
The only time it's shut me down is when I ask it how to hurt the US. The conversation ended up in some funny places, like, it told me that data sovereignty legislation puts US employees out of work and therefore it could not help me brainstorm how to pass similar measures in my country. Usually it's sycophantic enough. Then again, I'm not racist/sexist or whatever else you mean by partisan.
If I had to guess, I'd bet that the sentence on national or racial IQ statistics triggered a "White Supremacy Wrongthink" autofilter.
Last I knew Substack hosted all kinds of godawful blogs, including white supremacist, savagely misogynist, ivermectin-cures-all, etc. blogs. Has that changed?
Isn't it Substack that does that automatically?
Yes! But it's new, and I think we should push for Substack to stop. In the last year people have put up posts saying Trump should be assassinated, Israel should be bombed into a parking lot, blacks should all be shipped back to Africa. While many posts of this kind earned a ban from Scott, some went unnoticed, and Substack itself did not have AI moderation that caught them. I hated the posts I'm mentioning, but still appreciated being on a platform that did not outlaw them. Let the readership and Scott decide what posts cannot appear here.
It’s completely unexceptional for a man to tell a woman “I just love your big booty!” Many women even get fake rear ends to please men. Yet it’s well-nigh impossible to imagine a woman telling a man “I just love your huge beer belly!” And there’s certainly no market for fake bellies.
Why the double standard?
Reportedly, men who want a big belly consume large amounts of blood and milk: https://www.dailymail.com/news/article-4360060/Ethiopian-tribesmen-compete-fat-possible.html
You need a female equivalent of Bubba Sparxxx to sing a magnum opus to the beer belly.
"Huge beer belly" is intrinsically negative, unlike "big booty," so it's not a fair comparison. There are plenty of women who are not especially bothered by physical imperfections in a guy with a lot of great qualities. I had I honey with some pudge and used to tell him I loved his love handles. (And in the era when I said it I had no pudge at all and in fact did not have any physical imperfection as "bad" as Pudge's.).
If I remember right, you also posted something you said was 101% true about the chances of success for males on dating apps, and I pushed back against that too, and with a concrete counterexample. It's not that there's no truth in your idea that single males have a rough time of it in the dating world, but your picture of the situation lacks nuance, and I find your all-or-nothing views about how 101% screwed men are pretty irritating.
I checked, and your counter example involved a match between your daughter and a man who was tall and attractive. That proves my point.
Statistics show that women on dating apps have *far* higher response rates than men. A few anecdotes don’t prove otherwise. I do not know how this will turn out.
I'm combining this post with the Dad Joke post.
>A man asked his friend how he could boost his body confidence at the beach. The friend suggested putting a potato in his swim trunks. So, the next time the man went to the beach, he went out in swim trunks with a potato bulge.
To his surprise, the other beachgoers reacted with horror. Even the friend who made the suggestion turned pale when he saw him.
'My God man,' the friend shouted, 'you're supposed to put the potato in the front!'<
Possibly more well known example would be "Baby Got Back" by Sir Mix-a-lot
Since Fable is still in model-jail, I figured it might be interesting to others to share my experience using it for a large, complex software engineering task. I was working on an open source, non-commercial project, so I'm also at liberty to share all the code and transcripts from the session. If that sounds interesting, the write-up is on my substack linked from my profile, or direct link here: https://tossrock.substack.com/p/36-hours-with-fable
Not on here obviously, but when I go on any normal part of the internet, its people complaining about how AI is making computer parts more expensive, how AI datacenters are screwing up communities, how AI has basically broken education and teaching, and it made me wonder; has any other technology been like this? Where the majority of people immediately feel a surge of negative effects while all the positive effects are either hidden, reserved for a smaller cohort of people, or just anticipated but yet to materialize?
I could just be young, I was born after the invention of the TV, I barely remember life before internet. I sort of remember a panic around Smartphones, but not like this. This feels unprecedented, I wonder if people hated the internal combustion engine this much or if there's some other analogous event. I remember the review here of James Scott's book on the agrarian revolution, how its not clear that the vast majority of humanity wasn't forced into it at sword-point, and that kind of feels more like where we are at. Maybe we will forever look backward at the pre-AI days the way the first generation of hunter-gatherers forced to farm looked backward at what they had lost.
The Industrial Revolution more or less started out as you describe. Eventually, it became more obvious that the added productivity had real impacts on (for example) clothing prices. But think about it from early on the adoption curve. Big, smelly cities got bigger and smellier. Cottage industries were pulled into their orbit. Most of the modern conventions about labor rights had to be developed in real time, as did the economic theory. Marx caught on for a reason.
As I understand it, America dodged a lot of that unrest by merit of 1) a skilled labor shortage and 2) a certain unskilled labor institution. But you can see a similar pattern there. The American South felt threatened by rapid economic growth even when they could see a benefit, because wealth comes with political and social power.
To the extent that this is unprecedented, I think a large aspect is the role social and digital media plays, and the way it tends to be an amplifier for negativity. In the past people may have had issues with certain technological developments, but they weren't as empowered to spend practically 24/7 being angry about AI with other people who are angry about AI and getting selectively tailored headlines about how bad AI is.
This feels like the new "woke" to me. A way of socially signaling spurred on by groups of rage bait influencers. It feels like there's a feedback loop there and it's all performance.
I'm not saying there is zero point to the claims. But, I believe, no amount of whining is going to stop it.
“ Where the majority of people immediately feel a surge of negative effects while all the positive effects are either hidden, reserved for a smaller cohort of people, or just anticipated but yet to materialize?”
And yet where a majority of people do nothing to actually push back on the technology’s intrusion into their lives!
I do the sorts of trivial opt-out which you count as “pushing back.” When it’s even allowed. I don’t volunteer my data for training, I complain about and avoid the generative search stuff, I don’t talk to whatever bot people are using. I haven’t adopted it for my job, either, which is the one place I might manage to find some benefit.
This has approximately as much effect on the zeitgeist as installing an adblocker.
I do the sorts of trivial opt-out which you count as “pushing back.” When it’s even allowed. I don’t volunteer my data for training, I complain about and avoid the generative search stuff, I don’t talk to whatever bot people are using. I haven’t adopted it for my job, either, which is the one place I might manage to find some benefit.
This has approximately as much effect on the zeitgeist as installing an adblocker.
What are we supposed to do? Reproduce Google search from scratch? Reproduce computer parts manufacturing from scratch?
Global revolution?
The fact that things aren't bad enough to justify drastic measures (the digital world thankfully remains digital, and its enshittification predates AI anyway) doesn't disprove that they're bad.
AI has yet to intrude in my life in any significant way; just marginally more slop in some of my social-media feeds. But for people who are seeing an intrusion, what sort of pushback would you expect?
Simplistically speaking, to a high school or college system, AI offers them a faster and easier way to generate acceptable term papers for their classes, and threatens to consign them to the perpetual unemployed underclass when they graduate. They're probably going to see that as a net negative. But not using ChatGPT to generate term papers, won't move the needle on the "perpetual unemployed underclass" front, so what are they really going to do about it?
Short of going Full Luigi on Sam Altman, of course.
I don't think most people are thinking about it as deeply as you describe. AI showed up in their lives, they got a vague understanding of what it was, and (probably mostly based on what group they identify with) they internalized some kind of sentiment toward it, which they learned to parrot back in conversation and in response to pollsters.
By contrast, here's an anecdote from my life: a year or two ago my search engine of choice started showing me its AI-generated answer to my queries at the top of the results page, without me asking. I found a setting to make it so I don't get an AI answer unless I specifically click a dedicated button to get one. I avoid clicking that button in the vast majority of instances. (There are a select few types of questions where an AI-generated answer is both advantageous and scrutinizable for trustworthiness.)
This is the kind of pushback I'm talking about. Not marching with signs, not writing op eds, just being intentional about and sparing with the use of this new technology.
It's very simple and easy to do, but very few people do it. I find this fascinating.
A certain type of people spends a whole day on the settings menu of a new phone... I might even be one of them.
Yet the vast majority of people never go there and just use the defaults. Big Tech knows this and are not bothered by the minority of weirdos who get intentional instead of going with the flow.
Also, for the specific case of Google front-loading an AI response, most of us have already gotten used to just skipping over the front-loaded paid responses Google has been providing for many years, without an easy opt-out for obvious reasons. The simplest way to deal with front-loaded AI slop is to just add it to the list of things your eyes glide over without even registering.
Yes, the uncommonness of rejecting or at least scrutinizing default settings was covered in "Originals" by Adam Grant.
But what fascinates me in particular is how strong opinions about technology abound--predominate even!--simultaneously with a lack of intentionality or really doing *anything* that might be aligned with those opinions.
I see it all over the place, this cognitive dissonance. We all know social media is bad for you, everyone is on FB and Instagram and X. We all know staring at your phone all the time is a sign that your brain is turning to mush, people stare at their phones all the time even while driving. We all know AI is a slop-generator that's gunning for our jobs, even AI safety proponents like Scott keep using it and quoting it as an authoritative source of information. Etc.
Thinking about it more, I was reminded of a scene in Middlemarch where a railroad is being built through the middle of their county/region/whatever. Several people do get up in arms about it and even go so far as to march to where the railroad is under construction, but eventually its explained to them that their efforts are pointless and they go away. Thus it ever was, perhaps?
People are afraid AI will take their jobs.
Yes, but my understanding is that the last time this was happening, during the industrial revolution, the trade-offs were obvious and almost immediate to the majority of people. We are being asked to deal with inflation and mass unemployment because.... fifteen years from now we might have vertical productivity?
How about https://en.wikipedia.org/wiki/Swing_Riots or https://en.wikipedia.org/wiki/Canut_revolts or https://en.wikipedia.org/wiki/Silesian_weavers%27_uprising
Not sure the tradeoffs were obvious even then, though. Yes, now they are, and thus we (and "they system", whatever it may mean) prefer the narrative which says it was obvious, but the thing is, we have no clue how the whole thing looked like back then.
(Also see "Blood in the Machine", about the Luddites, who were actually not those mindless anti-machine mobs we all know them to be.)
Why is that your understanding though?
Maybe it's just because you don't have access to the 1780s version of reddit, where all the gloomiest people with a lot of free time speculate about things.
Perhaps not identical, but I think the most disruptive technology that has occurred in human history that can be compared to this one, is the invention of writing.
Ah yes, Phaedrus rears his head again.
Well I'm not sure that's what I had in mind. The invention of writing predates this fellow and I'm thinking more of the effect it had on creating larger civilizations and endowing a ruling class with power over everyone else
Grasping at straws, what would be a less popular invention than AI. Mustard gas perhaps? We can check quantitativelyish asterisk asterisk etc.
Mustard gas: 30% approve, 65% disapprove (during WWII) for -35
AI: 26% approve, 46% disapprove for -20
So fewer people are positive for AI than for mustard gas, but AI hasn't yet gotten that solid majority of people actually opposing it. Flattering for the chatbots!
https://www.nationalww2museum.org/war/wwii-polls/roper-polls-poison-gas
https://www.pewresearch.org/short-reads/2026/03/12/key-findings-about-how-americans-view-artificial-intelligence/
Surely nuclear power is a better comparison.
Nuclear plants are way more popular than AI, its not close
https://www.bisconti.com/blog/public-opinion-2025
A general recommendation to attend a FIFA Fan Festival if you live in a host city. I went to one yesterday, very cool, free fan event with overpriced food, some cool attractions, but the main thing is it's just so cool to sit on a lawn with a few thousand people and all watch the same giant screen and go "Oh" and "Ah" and "Oh no!" all together. Very wholesome sportsball stuff.
If you live in the best country, you can find a Fan Festival in:
-LA
-Houston
-Dallas
-Kansas City
-Atlanta
-Boston
-Philly
-Miami
If you live in the best food country, you can find a Fan Festival in:
-Guadalajara
-Mexico City
-Monterrey
If you're cursed to live in frozen commie land, you can temporarily salve your lifetime of disappointment in:
-Vancouver
-Toronto
https://www.fifa.com/en/tournaments/mens/worldcup/canadamexicousa2026/articles/biggest-selection-of-host-city-fan-events-in-world-cup-history-to-ignite-celebrations
Ah yeah, fun fact, we call that "public viewing", and I understand that in English this means viewing of a dead body https://en.wikipedia.org/wiki/Public_and_private_screening
Okay this is cool, and so far as I'm aware absolutely unique:
https://substack.com/inbox/post/188261171
That url isn't very informative, so I think it's better to link to what that redirects to:
https://jasher.substack.com/p/announcing-the-crime-index
Let me guess: another substack author reinventing racial profiling?
Edit: I shouldn’t have been so snarky. It’s a neat data-indexing project making use of historic pdfs. I admire the effort that went into it.
I found your point #1 in the original post, very intriguing. How do they define truth? The answer is often that "It depends".
Are eggs good for you? It depends. On your health history, your circumstances, what you enjoy...
Ancient Hindus believed there was no such thing as absolute truth. That is why they were never interested in proofs, even in math.
The Greeks were interested in iron-clad proofs, which has its own value. Knowing that a claim is always true. Like the Pythagoras theorem. True on the moon, true in a million years.
Indians had various approximations for Pi. They wanted an approximation that worked for an application they were currently pursuing. If their approximation didn't work, they'd refine pi.
Indian mathematical treatises are full of approximations. Like how to solve cubic equations with certain restrictions.
It's a very different way of thinking about knowledge, than the Greek way.
I would say the Indian approach happens to correspond to modern statistical ways. It's what GPT is doing anyway...
I like it because needing a proof is very limiting.
I realize this is what I need to change to write code. I can't seem to understand everything but need to progress in making the program work anyway. Very contingent knowledge.
A simple analogy is you don't need to understand how Excel spreadsheets work before using them.
The Indian thinking (i.e. no interest in proofs) is very well suited to modern ways of using knowledge, I think?
What are you talking about? Ancient Indians were the ones that came up with the concept of Maya. They came up with the concept of Om and Brahman and Atman. The rope looking like a snake analogy in the dark but still being a rope. Are these not eternal truth to you?
I think Indians thought of the rope-snake analogy in the following way : You operate under the assumption that it is a snake until you go closer and notice it is a rope.
If there is smoke, you assume there's a fire as a first pass.
And you're right about their saying Brahman is the ultimate reality of everything (a rough analogy, the way 0s and 1s are the ultimate reality of any software). They realized they had to stop somewhere.
For people who know more about generative AI, how much of the improvement in modern models is due to improvements in the "base" model vs preprocessing of the data, research and planning, and other more "agentic" stuff.
I downloaded and worked on a relatively small local LLM (1) and I was shocked by how much dumber it was than Claude Opus 4.8 in the browser. It wasn't just that it was dumb though, it was dumb in a way that models were dumb two years ago and aren't now. I don't know how to phrase it. Like...prompting doesn't matter with modern models. In a Claude chat, I just write and brain dump whatever I want and Claude "gets" it. Qwen doesn't. I'm not sure how much of that is just the raw Claude model being dumb and how much of it is, like, a Claude model reads in my input, rewrites my stupidity into a sensible question, recognizes it needs to go read website X, Y, and Z, and then feeds all that into the context window of the final Claude Opus 4.8 that actually writes the response.
So, like, when I'm "talking" with Claude in a chat window, is Claude so much better now because the "raw" LLM is better or because I'm actually talking with an agent that's doing a lot of post processing with my question and the underlying model is just marginally better?
(1) Qwen 2.5, 14B I think
some of claudes intelligence comes from the system prompts. you can try out using claude via API, then you'll get less of a systemprompt and a more "raw" experience. (you still get some system prompt, from antropic, albeit much leaner).
Afaik, one thing that makes claude much stronger than the competition, is that it is better to follow the systemprompt and reflect, which comes from antropics effort regarding alignment. (note: pure speculation).
Qwen 2.5 is *ancient*.
Having said that, it's a bit of all that drives the improvements. If you want to see what a stand alone model can do, try one of the top Chinese ones (currently GLM 5.2 is t the top). If you want local, Gemma 4 or Qwen 3.6.
Yeah a loooot of people even technical people who should know better have the thinnest mental model of the capability distribution and rate of change in LLMs, even excluding frontier models.
Just looked it up. Qwen 2.5 is about a year and a half old (Jan 2025), and this was a distilled local model besides. By comparison, ChatGPT 4.5 (frontier model from Feb 2025) was the first one that reliably gave me not-terrible answers to the Kamikaze Pilot line of questioning.
I made a mobile app to answer “which of my friends is free to talk right now?” in less than 30 seconds.
It’s called Beacon! You send a beacon to a group and everyone gets pinged at the same time. The first person to answer the beacon gets connected for a 1-1 audio-only call.
(The signal drops silently for everyone else, no pressure to answer or missed call badges. There are quiet hours settings and of course you can manage inbound pings with OS-level focus modes)
It works best if join with at least four people you don't speak with as much as you'd like. I have a couple dozen connections on the app now and it feels like magic to me.
Would love feedback from ACX-ish people who still like phone calls, or wish they had more of them. You can get the iOS version on TestFlight right now here:
https://trybeacon.chat/
I have an Android version too but it's an earlier build so still adding people manually for now. If you really want the Android version let me know!
This is such a good idea and thank you for making this. I thought I would be the only person in the world who has this issue! Sending it to all my friends.
The only problem right now is several friends don't trust TestFlight or the app for some reason so aren't willing to get it. I will let you know how my experience goes.
Thanks! There are definitely a lot of us with this issue.
Curious if your friends would immediately trust the app if it was on the App Store instead of TestFlight, or if there is something about the app itself that would concern them even once it was on the App Store.
Happily accepting any feedback, bug reports, and feature requests :)
Yeah I think they’d be fine with it on the App Store, especially if it had some reviews
Does your Android version allow cross-platform communication with iOS users?
Yes, definitely!
In honor of Father's day, here is a question for dads and others out there: what is a 'joke' you like that is amusing to you and the opposite of that to all the friends, family, co-workers and random acquaintances you inflict it on? I have a few:
Whenever my wife is working in the kitchen and happens to pick up a knife, I will scream "AHH KNIFE" and honestly it's a miracle she hasn't stabbed me yet
Whenever *anyone* comments 'Oh sorry about that' or similar, I'll immediately reply "I forgive you!", because I certainly don't want them to worry about me not forgiving their transgression (e.g., 30 seconds late joining a meeting, not bringing one of the drink orders, stabbing me with a knife)
What are yours?
What do get when you cross a mosquito and a mountain climber?
Punchline - you can't cross a vector and a scalar.
The readership here is more likely to get it than the normal distribution of the population.
all time favorite:
> A: oh my god
> B: yes?
"I'm gonna run to the store."
"You should drive, it's hot out."
Whenever I have a plausible reason to think somebody's face hurts, I always very calmly point it out (sometimes with a little bit of suggestive emphasis), and then don't say anything else.
Like, "Hey, does your face hurt? Because you're looking a bit sunburned."
Thanks for this. For a punish 'sci fi' book I recommend "Callahan's Crosstime Salon" by Spider Robinson.
I think my diarrhea is genetic,
it runs in my jeans.
All of my jokes are funny. If I have a joke that's not funny to anyone else, I stop telling it. I save it for the bathroom. Mmm.
A long time ago, I worked as an assistant editor for an Austrian fellow in Toronto, Canada. This was back in the film days before video, so at the end of an edit, we would have to send it to the lab so we could get a print, which took overnight. Every time that happened, this man would start singing, "One day my prince will come." It took me a long time to get into the spirit of it, but it still cracks me up when I think about it.
I wanted to thank you for starting this thread. I don't have anything to contribute (I'm German :P), but the responses here have been delightful to read so far.
When our son says “I’m hungry/tired/bored,” my husband will extend his hand to shake and reply, “Nice to meet you, Hungry! I’m Daddy!”
It drives our son nuts. I occasionally do it too.
Not particularly a Dad Joke, but if someone ever says "no pun intended" I say "none taken", and they usually give me a blank stare.
Reminds me of the time I entered a punning contest. I submitted 10 different puns hoping at least one of them would win but, unfortunately...
Whenever my wife is doing tricks-for-treats with our family dog, after a couple of successful goes I say -- to the dog -- "Hey the training is going well, congrats!"
But the canonical recurring dad-joke in our household is: starting from when my youngest was maybe 5, when he tripped or whatever and landed hard I could be counted on to call from the next room, "Is the floor okay?" (That joke being a family heirloom which my younger brother and I first heard from our much-elder sibling and I don't think it was original to him.)
I over the years adapted its deployment to, I like to think, new applications. When my youngest as a grade-schooler collided with vegetation while trying to catch a thrown football he'd get, "Is the tree okay?" When wrassling with the pooch resulted in tumbling across his bed: "Is the pillow okay?" After taking an errant shot off his backside during a youth-hockey practice, me from the stands: "IS THE PUCK OKAY?" At the beach playing long-toss and the ball glances of his hand and then noggin, I waited until he'd thrown it back in, then visibly inspected it for a second and then called out, "IT'S ALRIGHT, THE BALL IS OKAY!" Etc, many other examples. To the point of for a while as a middle-schooler he'd pre-empt: "Dad don't you dare...."
Yesterday though in the car, and I don't think he was intentionally doing a Father's Day thing, came the apotheosis. My wife in adjusting the front seat fumbled the control briefly and said "Ow" from having banged her knee as the seat sprung forward. Our 14-year-old in the back seat leaned forward and deadpanned, "Is the car okay?"
You may now hum your favorite old passing-of-the-torch song. I'll need just a minute
Brings a tear to the eye it does (because of the banged knee)
I'm still in love with a pun from one of Scott's posts a long while back:
When Immanuel Kant died, he left his machine gun to the leader of a coup; willing that his Maxim should make a general rule.
I've told it countless times, and it has been appreciated approximately 3. That has not and will not stop me from continuing to tell it.
I. Kant let you do that, Fox. (Sadly, philosophy majors don't seem to play videogames, and vice versa...)
Oh that's hideous, and will now be deployed upon my eldest who was a philosophy minor in college.
Many, many years ago, I was playing in a D&D 3.5e game. One of the other players was playing an artificer and had just hit the level where he gets a robot companion. He was telling the GM that he'd chosen a robotic dog named "Spot".
GM: "Now you need to get a second one and name it Listen."
Me: "That only works if they're both from Prague."
GM: "Why?"
Me: "So they'll be Czechs."
GM: *headdesk*
Me: "You started it."
One exchange I heard from a D&D forum years ago:
"So, I want to build a Warforged character, but I'm debating whether or not I should take the mithril body upgrade. It costs a feat but it's a nice upgrade to my armor."
"Go with the all-metal design. And put some skill ranks into Craft (Alchemy)."
"What? Why would I do that?"
"So that you can be a fullmetal alchemist."
I once wanted to make a tanky Monk named Henry David, from the mythical Walden monastery. He'd spec to have high saving Thoreaus.
This went over about as well as me trying to explain to my GM for Pathfinder 2e that there's an arbitrage opportunity by buying all the individual components of an Adventurer's Kit separately, and therefore my character had cornered the world market and deserved a higher starting allotment of currency. I don't know if they kept that error(?) after the license revisions, but it'd make me sad if so.
To be fair, were I the GM, I'd totally let you do it... provided you could actually *do* it. You can't just tell me "I've cornered the world market" and treat it as fait accompli; that'd be like saying "oh yeah I totally killed the evil lv 17 Lich, trust me bro". But it might be fun to see how far you can get with obtaining startup capital, securing warehousing space, buying the goods in bulk, fighting off the local Thieves' Guild, committing tax evasion, hiring mercenaries to take out the competition, etc. etc. It sounds like a fun campaign premise, actually.
A character in a campaign I recently ran, did try an arbitrage on gems. I allowed him to do it, but ruled that there is a finite amount of gems in the areas he had access to.
A character concept I've come up with but haven't had a chance to play yet is a dog-man who's dual classed between Cleric and Rogue. His name is "Bandit Healer".
When he is diagnosing a patient, does he open his eyes wide and tilt his head sideways ? :-)
Oh that's good, and reminds me of something a couple players contrived of in one of my campaigns.
Using a combination of Magic Mouth and Animate Object, they would take a wood carving of a number four, which would then be named 'Number Five', and enchanted to say "I'm the number three!"
Resulting in the animated four, running through the town square with the wizard hot in pursuit yelling, 'Come back here Number Five!', while it shouted "I'm the number three!" and passerby's exclaimed 'that's the number four'. Repeat ad infinitum.
Hehehe. That kinda reminds me of this bit from the White Knight chapter of Alice Through the Looking Glass:
------
“The name of the song is called ‘Haddocks’ Eyes.’”
“Oh, that’s the name of the song, is it?” Alice said, trying to feel interested.
“No, you don’t understand,” the Knight said, looking a little vexed. “That’s what the name is called. The name really is ‘The Aged Aged Man.’”
“Then I ought to have said ‘That’s what the song is called’?” Alice corrected herself.
“No, you oughtn’t: that’s quite another thing! The song is called ‘Ways and Means’: but that’s only what it’s called, you know!”
“Well, what is the song, then?” said Alice, who was by this time completely bewildered.
“I was coming to that,” the Knight said. “The song really is ‘A-sitting On A Gate’: and the tune’s my own invention.”
That actually sounds like a lot of fun.
Almost all They Might Be Giants songs have at least one hilarious line that my family refuses to acknowledge.
I think the following can be repurposed by most parents:
Kid: there's nothing to eat! (note: they are standing in front of full refrigerator)
Dad: Oh no! We're going to starve to death, ahhhhh!
(advanced move: flail arms and run around the room while continuing to express dismay about the unfolding tragedy)
I do like those guys.
ACX Bangkok chapter meets again on Sunday 28 June at 1pm. We'll be at Thong Kee restaurant: https://maps.app.goo.gl/ufPk6SrwCaCk6Tb59
Come join us for good food and conversation. You don't have to be hardcore rat or even based in Bangkok.
I *may* bring a cat plushie as a mascot but you can also just look for the loud English speaking table
How does a tiny writer with a niche, hard-to-categorize publication become findable by the right readers on Substack without becoming annoying?
(ᕗ ͠° ਊ ͠° )ᕗ
I sometimes manage the first part at the expense of the second — as you may be discovering right now, dear reader.
⊂(◉‿◉)つ
Looking especially for advice beyond “post consistently” or “just comment more.” How do small, weird publications become legible to their natural audience without turning into a chatty billboard with thumbs saying "who's got a Substack and two thumbs? This guy!"
Say something like "I wrote a post about (topic)" and link to the post.
Maybe make a teaser. "I wrote a post about Masters of the Universe bombing at the box office despite the fact that it's arguably what people have wanted from a Masters of the Universe movie since the 80s. My theory is that Hollywood has screwed up so many times across so many reboots... Ghostbusters, Star Wars, Doctor Who, Rings of Power, Wheel of Time, the Terminator franchise... that by the time a team actually made a movie for established fans rather than the audience they wish they had instead, nobody was willing to show up. Twelve-times bitten, twice shy."
And people will click on the link if they want to read an essay about that and they will snort if they don't.
And then, next time an open thread rolls around, talk about your new post and give a nice and short teaser.
Don't say "SUBSCRIBE TO MY BLOG!"
That's annoying.
Say "I wrote an essay and it's about something that you will have an opinion about and here's the precis."
And most people won't click on it but some will.
My suggestion would be to consider if you are a niche topic or a niche writer; if the former, emphasize quality and in depth to produce the definitive evergreen pieces which 'own' a point or topic, and if the latter, emphasize diversity and quantity to buy a lot of lottery tickets to discover what will surprisingly hit the spot. https://gwern.net/polish
I think "post consistently" and "just comment more" are both good advice.
Another good but often overlooked piece of advice is "produce genuinely excellent content." No, not good content, excellent. A hundred times more effort than the median post. I saw a link recently to a really detailed explanation of how mechanical watches worked, with tons of interactive animations showing the exact functions of each part. It's not a subject I'm interested in whatsoever, but just the sheer effort that went into it caught my attention, and I ended up learning a bit about a subject I never would have sought out.
Another, somewhat easier trick is "pick a fight with someone more famous than you." Bonus points if you're actually right about something, but if not, kayfabe and pointless drama always sell too. Depends how unscrupulous you're willing to be.
I recall a CGPGrey video with the same advice, but for YouTubers. He'd always see people talking about regular upload schedules, collaborations, &c, but to him all this came secondary to making content people wanted to watch.
I appreciate that advice. I've built content channels for other projects on other platforms in the past. Even more important than making things that people want to see, is working the algorithm the way you must, in order for it to even show you to enough people in the first place.
I've new'ish to Substack, and am happy to see the discovery algorithm here isn't skewed entirely away from benefiting creators and just slurping everything up for the mega-corporation.
IG for example is entirely dead for new creators unless you "work" the algorithm hard. You can spend a month putting out good daily content, and at the end of the months you'll have 30 posts, maybe 1-3 likes total, and a dozen or so views. And I'd only bother with Facebook if the plan was to advertise and pay for views.
For what it's worth, the channels I've watched and the blogs I've read for a long time have all been word-of-mouth. I found out about SlateStarCodex from a Reddit post in the Mass Effect subreddit, ACOUP from a comment on SSC or the SSC subreddit), RedLetterMedia from a CGPGrey video...
I don't remember how I found CGPGrey--that one may have been the algorithm (couldn't have been Reddit, because I found Reddit through one of his videos). If so, it's worth noting that I haven't been subscribed to CGPGrey, nor have I had a Reddit account, in years. Same goes for CinemaSins and Honest Trailers: those were for sure the algorithm, and I haven't been subscribed to either for the better part of a decade.
Anyway, all this to say word of mouth is also important, and that twisting your content to suit the algorithm might not lead to long-term gains. I've given money to the RedLetterMedia Patreon; I've never given money to anything shown to me by an algorithm.
So you're saying I need friends with bigger mouths. :) noted.
I'd say publish less, actually. If you have some amount of effort to spend on things, use it on a small number of things that I'll be excited to see in my feed instead of a large number of things I'll be annoyed by.
That’s helpful, thanks. Do you think Substack Sections help solve the annoyance problem, or do they mostly create admin theater for the reader?
I’m curious whether readers really use the per-section/email toggles, or whether “too many emails” just gets processed as “this person posts too much” and leads to unsubscribes.
They help! You probably have to point it out to your readers though. I wouldn't know it existed if Nate Silver hadn't explicitly mentioned it.
Why read the rest when you can read the best?
And I agree with you — scruples only get in the way 👺
Does "hard-to-categorize" also mean "hard to describe"?
I don't know about other ACX readers, but a (short!) "sig file" attached to the bottom of each of someone's posts with some text and a link doesn't bother me. If you need a long paragraph to describe the publication then there may be issues ...
Oh, I could describe for days.
Most succinctly, I'm a bird-watcher, but for arguments. And I'm writing a field guide.
…I’ll polish that before production — but I get what you're saying. That's good advice.
Is this a bald eagle argument or a tit argument sort of thing?
I study the yellow-bellied tit pedant, and the New World argumentative warbler.
I'm not a birdwatcher, but I'm tempted. Related, I started reading The Peregrine by JA Baker and gosh it's evocative.
I just read the synopsis - it looks like an interesting way to communicate avian conservation.
It's often interesting listening to an expert being passionate about their expertise even if it's well outside your own realm of typical interests.
How did you guys found the competitors for this year's contest? At first I thought it was a poor batch, but I think it was just RNG messing with me, as I hit a few very good ones by the end of the deadline.
> How did you guys found the competitors for this year's contest?
I was recovering from a surgery, and so was in a place to read and rate 120 of them this batch. I'd say average quality is down - prior batches I had roughly 1.5 - 2x more 9's and 10's.
There were some truly incoherent and unreadable ones too, so the floor seemed lower too.
Going back and looking at my notes, I had eighteen 4-and-belows, and thirty six 8-and-highers (with only 2 tens and 10 nines), and the rest were between 5 and 7.
I feel the loss of very peak the most - that feeling when you're reading a great review, they've already surprised you twice and pointed out 3 interesting tangents you're going to follow later, and then they make a pivot or bring in an entirely new connection or point of view that catalyzes the whole thing, making their point and argument transformative and deep.
I pretty much live for those, and feel like we didn't have as good a showing this time as we did in some last rounds.
Still a great contest with many worthy entrants, and I'm certainly happy to have read the great majority of them. As always, kudos to Scott and his readership both, the incredibly high talent bar and selection effects are why we're all here.
I was dealing with some medical issues and didn't get to the reviews in time to vote--I went to cast my first vote on what I thought was the last day, but forgot about timezones--which is a shame, because I've run into several excellent ones.
Pleasantly surprised at the relative lack of (obvious) LLM usage, although it did ruin one particular review which would otherwise have been one of my top picks (especially aggravating because the good parts were human-written and the author clearly had the skill; the whole thing would've been improved by altogether deleting the LLM-generated paragraphs)
May have been mine :/ ran out of time and finished half the review day of the deadline. Very frustrating because I worked hard on it and if I’d had two weeks to edit I’m pretty sure it could have been a finalist, as is I’m almost certain it won’t be.
Without getting into details, the human-written parts were the author relating their personal experiences with the subject matter--interesting stories, though I found myself wanting more links to sources--while the LLM-generated portions were usually about synthesizing lessons from the stories.
Not mine then, but I imagine a significant portion of AI usage is owed to similar reasons.
I was a little disappointed until I looked at the two AI-produced reviews of the reviews. (See the comments in the post announcing the reviews and requesting reader rankings.) They explained the (as they see it) characteristics of a good ACX book review. And, looking at the reviews they recommended and looking through their eyes, I came to a greater appreciation.
I checked the comments on the announcement post but couldn't find the AI reviews of the reviews that you mentioned. Do you have a link?
EDIT: Never mind, I found it here: https://www.astralcodexten.com/p/choose-book-review-finalists-2026/comment/270116545
I was searching for "AI" but the comment used "LLM" so I missed it the first time.
Hm, I find it interesting how little it correlates with my own ratings. The first two are also exceptional in my eyes (10/10), but the other ones in the AI top 10 are only in the range of 5-8 stars for me.
That's way below my correlation with finalists in the last years. I think there were few finalists which I wouldn't rate at least as 8, and most of them as 9 or 10.
I am not quite sure what to make of it.
Well that’s an LLM for you. If I had a friend who somehow read all the entries and ranked them with notes, I’d probably disagree with a lot of their top ten picks as well.
I hope the AI predictions prove inaccurate, as they put my entry outside of the top 40. But there is no accounting for taste!
I thought both AIs ranked mine pretty fairly - I agree with the criticisms, and a skim through the ones it ranked definitively higher than mine showed a genuine difference in quality.
Not me, I’m brilliant and the LLM clearly doesn’t know what it’s talking about. I hope, anyway.
I read and voted on about ten, and it was roughly a bell curve. One bad, 4 below average, one very good, one excellent. My criteria are the things which bring me to SSC in the first place: interesting takes, solid reasoning, fun facts, good structure.
This mirrors my experience. I didn't set out to mark along a bell curve, it was happenstance.
I found it still good. I had fewer mind-blowing experiences than in the first years of the competition, perhaps because those books are to some extent exhausted. But there are still many excellent reviews.
And I like the average book review a lot better than the general "everything-except-a-book" review that we had last year.
Side note, I thought your username was "Demotist" and I was really appreciating the callback to those days.
No, my nick has several overlapping meanings, including a character from Ender's game, but "Demotist" is not one of them.
Have you seen the AI Pledge for Humanity in support of UBI as an AI dividend? I think that without this, the hatred directed at AI and data centers will only grow larger. aipledgeforhumanity.org
As of now, it seems unlikely that a UBI that would truly be distributed towards "humanity" and not just US citizens would find political support in the US. I also find it hard to imagine what form it could take.
We are mixing up the cost/benefits of a data center, which is entirely local, with the costs/benefits of AI, which is society-wide.
Do I want a building in my city that's a formless box that doesn't generate any economic activity after its built? Well, maybe, if it's generating a lot of direct revenue.
Virginia is benefiting locally. The opposition to data centers is more from ignorance than bad experiences.
Maybe the experience of Virginia will make other places jealous over time https://x.com/LoganDobson/status/2068747822774530484
Argument by tweet isn’t all the convincing. The counter argument here is that .. it depends. On the actual tax revenue, the water availability and if the data centres are actually paying in total for grid upgrades, because the question is “does it pay enough to cover all the additional infrastructure and system costs it imposes on the grid”. And, I don’t know. And it might. It might not. A tweet isn’t going to be convincing on its own.
Water availability is a negligible issue for data centers. They use the tiniest fraction of what golf courses use.
Not sure how the latter (i.e. that golf courses use an awful lot of water) makes anything less than it (e.g. less, but still enormous amounts of water) "negligible".
Not just golf courses. California almonds alone use up 7 times more water than all US data centers combined. Data centers make up 0.14% of our water use https://www.deseret.com/u-s-world/2026/06/21/what-environmental-costs-do-data-centers-have/ If anybody thinks the way to address our water consumption is to prevent something that takes up such small usage, and generates tax dollars (rather than receiving so many subsidies like agriculture) that could help pay for more water, you know not to take them seriously.
You’ve nevertheless engaged in a logical fallacy. As OhNoAnyway points out you compared a water user with golf courses which are very high users of water and declared that lower value negligible. You haven’t proven that grid infrastructure is helped by data centres or they if it is data centres are paying for it in full, and that’s the more pressing point.
There's a utilitarian alien invasion short story that might be amusing to some here up at my author site, https://rauhala.org/2026/01/07/story-the-end-of-suffering/
Cheers.
Can confirm I found it amusing! Thanks for sharing :D
I note pushback on social media government ID requirements is trying to get more organised / gain momentum with things like https://nevergivethemyourface.com/
Pangram says 82% generated. Why can't people be assed to write a website anymore?
https://www.pangram.com/history/8f9cebb3-0257-493e-8858-c05f2fd94c91?ucc=hw0Nsipa4CD
textdefend.com says the text of Scott's post is 74% AI generated (I can't run it through Pangram because I don't have an account)
I know these are different services, and the comparison might not be valid, but is there a meaningful difference between 74% and 82%?
(There's a meaningful difference between Pangram and any other site)
I hadn't tried any AI scanning site before today, but I'm beginning to believe that is the case (see my comment below). Do you know why?
To be frank, I don't know what they did differently from everyone else, but I've heard they also tuned toward their errors being false negatives (saying AI writing is human-written) instead of false positives (calling human writing AI-written). This helped them gain users, but them being generally very correct is what makes it worth using more often.
In my experience Pangram is much better than other sites. I tested it with a text of mine that was accused of being AI-written and Pangram said that it was "100% human written with high confidence" even though textdefend says it's 72% AI generated. Scott's post is also "100% human written with high confidence" on Pangram. You can also use it for free! You get 4 credits each day on the free plan.
I tried Pangram on my blog, it said 100% human, which is true.
(I expected it to be otherwise, after reading so many stories about false positives. So I am reporting a data point that in this case it worked OK.)
If someone has an account and/or patience, please try what happens when a genuine human text is *translated* by an AI to another language. I suspect that some false positives might happen there. Maybe even if you let an AI edit your text (such as "make it sound like written by a native speaker").
Huh, this was interesting.
Using AI, I translated one of my comments to English and checked it with Pangram and it was 100% human-written, but only with medium confidence.
Then, I told the AI to "Please make it sound like a native speaker" and then it was 100% human-written with low-confidence.
Then, I told the AI to "Please make it even more native-liek" and I got the same: 100% human-written with low-confidence.
Then, I told the AI to "Please spice it up to be more interesting" and then it was 100% AI-written with low-confidence.
The combination of "low confidence" and "100%" is interesting.
Apparently AIs and Pangram disagree on what is spicy. Or both agree that humans are not.
Ok. So maybe textdefend.com is not very accurate. I decided to try Scott's text on another free (as in doesn't ask me to register) AI scanner, quillbot.com. It said "0% AI generated". Great!
I then tried the first ten paragraphs of nevergivethemyourface.com on quillbot.com and it says "0% AI generated". ?!?!
Are all free AI scanners crap?
Pangram is free! For 4,000 words a day
Going back to the original subject of this thread, I'm probably the target audience for nevergivethemyourface. I'm pretty paranoid when it comes to privacy and I hate the direction that internet regulation seems to be taking in many countries.
And I hate having to create a new account on a site just so I can test their service.
Explicit note for those who need it: obviously I only put this into Pangram because it read like AI to me. It will read like AI to others. Borrow phrases, don't copy sentences.
It won't read like AI to me, because I won't read it. Thank you for the heads-up!
If the author doesn't find writing the post worthwhile, why should I find reading it worthwhile? A research summary I'll give you: I read AI research summaries and use them to find sources. But a public awareness campaign should be written by a person, because *the whole point* was to demonstrate that a person cared about the cause!
Some people are bad at writing.
I can read a lot about a topic, make notes, write an outline of a lecture, and give the lecture... all of that is easy. But when I start writing a blogpost, it is slow, and I am second guessing myself after every sentence. Perhaps this gets better with a *lot* of practice, but I am not there yet.
So -- although I have never actually done this, yet -- I sometimes wonder whether it would make more sense to e.g. record my speech, and then tell an AI to make a blogpost out of it. Then I would check it for mistakes, of course. I think this should still count as a human produced text, but I can imagine that the AI editing would trigger some alarms.
You will never be as bad at writing as an LLM. If your ideas have any merit at all, you will be able to do a better job of expressing them than any computer program, and developing your writing skills further will pay off in the long run.
I get the same feeling reading comments like this as I do when my coworker casually mentions thinking about getting a facelift (she looks fine and, thankfully, my other coworkers are just as horrified). No surprise she's the same one who runs all her emails through ChatGPT now--she was capable of writing for the past few decades, but now she's helpless.
I think what matters is how closely the AI follows the speech. On one extreme, if it simply writes down what was said, that's completely okay. However, if it incorporates its own style and comes up with its own points, then it quickly gets much less acceptable imo. Basically, it depends on how much the points in this great comment apply: https://www.lesswrong.com/posts/pvGavgxY5mQcyvPQa/g-wood-s-shortform#3DKQ7W8iGCETxoJer
??? No, AI-transcription doesn't flag as AI-written. This experiment is free to run!
As to writers who think that can't write without it— that's okay, but I don't want to read them.
Those people are wrong. See my reply to Villiam below.
whatever the explanation is, I go to this blog for high quality interactions with humans. this comment section is made worse the more I have to second guess if what I'm reading is AI.
This is encouraging to see.
I must be getting old, because I'm more or less resigned to "age verification" taking over the mainstream internet. I only hope that when this inevitably pushes teens to shady tactics and unmoderated sites, the process teaches them some technical skills and social sense. Social media has become too easy to use, what we need is a new generation of hackers.
Years ago Robin Hanson proposed a more decentralized approach to verifying physical identity https://www.overcomingbias.com/p/reliable-private-human-identityhtml My understanding is that some places with laws are relying on a third-party site, but I don't think there's some standard protocol permitting competing sites.
There's probably a way of checking ID in anonymous way using things like zero knowledge proofs. But this is not the way it's going to be implemented, so that's moot anyway.
"If you will it, it is no dream". Get on it, ZK id developers!
Jeez, that proposal sounds pretty awful. Mandatory RFID tags on all people sending to central databases (yes, technically he gives an ancap solution where there are many companies with different databases, but almost certainly there wouldn't be enough decentralization to matter) is definitely not privacy preserving, no matter what extra idea goes into it. People will change the rules and abuse the system to remove the privacy preserving parts, and all that will be left is "there is an RFID tag on every person".
I like physical IDs where there actually is a need for a method of identification. That way whoever needs the ID needs to go out of their way to ask you. If whoever is asking for your ID puts it in a database, it's probably their own special database, which law enforcement would have to individually subpoena. If you make tracking people a pain in the ass, then you could probably still catch the criminals who actually matter, while making it harder to stoop to enforcement of pettier things.
The more computerized the ID, the worse.
What's going to happen with $current_generation AI hardware in the next year or two?
I'm wondering if we should expect a drop of cheaper used server hardware like ram, graphics cards, and maybe even whole rack units in the next 12 months once data centers start upgrading from hopper (this year?) and Blackwell (next year?)
My understanding is that the $current_generation − 1 and maybe even $current_generation − 2 hardware is still in use.
> My understanding is that the $current_generation − 1 and maybe even $current_generation − 2 hardware is still in use.
Oh, it's much farther back than that. A100's at 40-80gb of vram are still being used in data centers for inference, and there are ~zero on the used market compared to the absolute volume bought and deployed in data centers (~2M or so).
That's 2020, the era of Nvidia 3090 consumer GPU's, and basically 4 generations back from Rubin GPU's coming out in 2026.
Thanks for the insight. I guess I'll have to look at other ways of running models at home
Speaking as somebody who's gone the local model route to test them out, here's my own research results, if it helps:
If you want an easy plug-and-go solution, the Mac M1-M4 workstations have unified memory and ultra high GPU/CPU bandwidth that can run 70B models. An Apple Pro Ultra runs about $7k new and can run a 70B model. The earliest M1 generation goes for about $3500 used and can also run 70B models, just with lower t/s.
People will argue that this money is better spent on getting a regular desktop with two 4090's or 3090's, because you can upgrade the GPU's later, but that's more work to put together a system, and some people like Macs. My theory is you can always sell the Mac workstation and get a real GPU box if you get into it enough that you want to upgrade.
You can also run smaller than 70B models on lesser Mac hardware than Ultras - plenty of people run 7B and 13B models on Macbook Pros and Airs with decent ram (ie 32gb and up, I think max is 128gb in M4 macbook pros, and 96gb in earlier M's).
Here's some different model size benchmarks on an M2 Max for comparison you might find interesting:
https://github.com/ggerganov/llama.cpp/pull/1642
If you want to go the GPU route, used server stuff a couple generations back is getting pretty cheap now - you can get EPYC 7702 boards with 128 PCIE lanes and CPU's and ram included for a couple thousand bucks now on Ebay, and can fit 6-8 GPU's in them with risers and splitters.
If you were a crypto bro and have a bunch of GPU's, you might want to look at Juicelabs' repo, which allows you to do GPU-over-IP:
https://github.com/Juice-Labs/Juice-Labs
The Juice Labs solution appeals to me because I have a ton of GPU's and mining board setups from an old crypto mining operation I put together - all I need to do is put a linux live usb in with the Juice server / client package and I should be able to do a lot with the hardware I have already.
In terms of multi-gpu setups here's the order of things that matter:
> First GPU VRAM (more the better)
> 2nd GPU VRAM (more the better, ideally NVLINK to 1st GPU)
> Ram GB / speed (there are arguments that 2 48gb sticks are actually better at 96GB total vs 4 at 192GB total, because the 2 can run at full DDR5 or DDR4 speed, versus your speed gets discounted in almost all mobos with more than 2 sticks of ram)
> CPU / Ram / GPU bus' bandwidth in gb / s - where M1/M2 macs really shine, with huge bus speeds up to 800 gb / s for the Ultra M2
> CPU speed
> PCIE speed
When you have multi-gpu's doing inference the main talking they do back and forth is sending hidden weight updates to each other, which is actually a very small amount of bandwidth, so PCIE bandwidth / speed doesn't end up mattering much compared to VRAM and ram.
Dang, thanks for sharing! I need to digest all this info--I just started digging into this stuff after realizing it would be nice to run a model at home and try to get some simple automation set up that is a lot more private than sending over all my data to one of the cloud operators. Not sure how much utility I'll get out of using it but evening setting it up would be a nice exercise in home labbing
usually datacenters swap their hardware, because old hardware requires more electric power per processing, i.e. it's not worth to pay for power for old hardware, when newer hardware exists. iirc they used to update hardware about once per 4 yearss. That was in the past, where 1 FLOP from some CPU was equal to 1 FLOP from some other CPU. With AI loads, this changes, because some GPU have architecture that is much better suited to some models, than other.
So in addition to the electric power argument from the past, datacenters now have an other incentive to update. So expect more hardware updates and not less
> I was hoping to nab some server grade hardware on the cheap
When I nabbed old hardware in the past, it always was via inofficial channels (i.e. the person in charge of recycling gave a hint)
edit: i never got free hardware from datacenters, only from offices. I don't know what datacenters do with their old hardware
Where compute per se is the bottleneck (i.e., not power or rack space), I'd guess the older GPUs et al. would be used for smaller models (e.g., Haiku) in order to reserve the newer hardware for the largest ones.
Alternatively, prioritize higher-margin API traffic for the fastest hardware and let the subsidized subscription users make do.
That makes sense.
And shoot, I was hoping to nab some server grade hardware on the cheap
I've made PDF ebooks of each complete year of ACX.
https://drive.google.com/drive/folders/1lSP8dz1QeagCeiG9Dt1bnGQMS8PZe8QU?usp=sharing
(With permission.)
Thanks. Why not make epub as well for better reading experience on devices
Maybe at a later point.
I did PDF as it's the most universal.
Remind me, I guess.
Where are people hosting and reading medium-length web fiction these days?
Some writers I know of post fiction on Substack, but it really doesn't seem suited for it, especially stories longer than 10k words or so. I'm also not too keen on posting fiction outside of narrow genres on my main blog, or creating a separate Substack. RoyalRoad seems to be geared for very long-form fiction.
You shouldn't copy my example, but I write my web serial on schlaugh.com. That gets you absolutely no audience (schlaugh is absurdly tiny for a social media site), but one thing I really like about it is that my posts are all scheduled to go up at a particular time of day and I don't even need to configure that - it's just inherent to the schlaugh platform, which only shows new content (including private messages, profile updates, etc) once a day tops.
I am posting on Substack. I tried Royal Road, Wattpad, and Vocal, and found no readers and no engagement. On Substack, I was able to build a readership. My work includes short stories, novellas, and full-length novels that I serialize. Substack is not super-well-suited for serializing - you have to do a lot of the work yourself, like the navigation and landing pages. Still, it seems like the place to be if you want your work to be seen.
yes, and substack bounces me off it for any extended reading due to not playing nice with dark mode settings and not having its own dark mode version. I tried to read some of your stuff but couldn't stare at the bright white background for long enough to concentrate!
Yes, sadly, Substack isn't perfect. Dark mode is a pet peeve for many.
i can handle it for short articles, or here at ACX its like a pleasant light blue background which is more tolerable, but the burn your eyes out white background with black text is a no go for me. I know you can change the colors in the code on your browser but that's way too much effort lol
I changed the color scheme to a nice salmon shade. Does this make the site easier to use for reading?
It does! now I have no excuse!
Bavitz, Nost, Farnham, Menherapilled etc. use AO3, I don't think there's any great place to post this stuff really
The fiction tag on LessWrong has lots of good stuff: https://www.lesswrong.com/w/fiction .
A few good short story writers also post directly to their own sites:
- https://alicorn.elcenia.com/stories/stories.shtml
- https://sprague-grundy.github.io/short_stories/
- https://qntm.org/fiction
Related: I really want to read a street wizard/hacker/conman-esque cyberpunk story, but it's about implausibly skilled prompt engineers expertly manipulating the low-tier-AGI overlords in a future surveillance state. You people know the genre I'm getting at. Does this exist, or do I need to write it myself?
There are a few similar works, but this is probably the ur-story:
https://anappendage.blogspot.com/2010/02/autofac-by-phillip-k.html
Of course, the prompt hacking in this version doesn't go nearly as well as it does in yours...
Write it yourself, just in case!
Don't extra-tall people have all kinds of back & joint problems later in life?
If we have the technology to allow adults to change their body type then we probably have the technology to treat back and joint problems a whole lot better.
If I can grow my bones longer (or shorter!) then I can probably repair my spine.
I guess broadly at the centuries-scale you'd expect those medical techniques to improve in parallel. But on the ground year by year, progress seems to be pretty uneven and isolated between disciplines. Today we're still prescribing physical therapy, cortisone injections and finally things like spinal fusions when people have chronic problems that less-invasive treatments don't resolve, which have tradeoffs and aren't 100% reliable. This is a discussion about sci-fi world anyway though, so I suppose we're free to speculate however we like.
I will instead think "snake with the head of a rhino".
I'll bet that will stop being fun really quick for most people -- the males choosing basketball player, the females choosing the rhythmic gymnast look. Right now, people who have one of those favored body conformations are rare and envied. When 75% of us look that way they won't be. On the other hand, there will be a few people choosing to look like Truman Capote, or a red-headed Eleanor Roosebelt, or just weird -- coal black skin, yellow hair, lavender eyes, extra-large hands . . .-- and that's kind of fun.
Probably be a lot of Paul Newman and Lauren Bacall lookalikes running around.
No idea in real time. In cinema time, any day now.
Just in case the autofilter is triggered by someone replying to the original post, here I am replying to this original post. If that "works", I'll reply to all the others so we won't have a stream of context-free slurs and expletives cluttering up the thread.
You didn't even get all seven! What, never seen George Carlin?
What, did the AI bot infer that you were trying to get it in trouble and so prudently refuse to moderate you? A win for automated moderation, a lose for AI alignment....