site banner
PaperclipPerfector  1mo ago (text post)   25566 thread views

Culture War Roundup for the week of September 7, 2026

This weekly roundup thread is intended for all culture war posts. 'Culture war' is vaguely defined, but it basically means controversial issues that fall along set tribal lines. Arguments over culture war issues generate a lot of heat and little light, and few deeply entrenched people ever change their minds. This thread is for voicing opinions and analyzing the state of the discussion while trying to optimize for light over heat.

Optimistically, we think that engaging with people you disagree with is worth your time, and so is being nice! Pessimistically, there are many dynamics that can lead discussions on Culture War topics to become unproductive. There's a human tendency to divide along tribal lines, praising your ingroup and vilifying your outgroup - and if you think you find it easy to criticize your ingroup, then it may be that your outgroup is not who you think it is. Extremists with opposing positions can feed off each other, highlighting each other's worst points to justify their own angry rhetoric, which becomes in turn a new example of bad behavior for the other side to highlight.

We would like to avoid these negative dynamics. Accordingly, we ask that you do not use this thread for waging the Culture War. Examples of waging the Culture War:

  • Shaming.

  • Attempting to 'build consensus' or enforce ideological conformity.

  • Making sweeping generalizations to vilify a group you dislike.

  • Recruiting for a cause.

  • Posting links that could be summarized as 'Boo outgroup!' Basically, if your content is 'Can you believe what Those People did this week?' then you should either refrain from posting, or do some very patient work to contextualize and/or steel-man the relevant viewpoint.

In general, you should argue to understand, not to win. This thread is not territory to be claimed by one group or another; indeed, the aim is to have many different viewpoints represented here. Thus, we also ask that you follow some guidelines:

  • Speak plainly. Avoid sarcasm and mockery. When disagreeing with someone, state your objections explicitly.

  • Be as precise and charitable as you can. Don't paraphrase unflatteringly.

  • Don't imply that someone said something they did not say, even if you think it follows from what they said.

  • Write like everyone is reading and you want them to be included in the discussion.

On an ad hoc basis, the mods will try to compile a list of the best posts/comments from the previous week, posted in Quality Contribution threads and archived at /r/TheThread. You may nominate a comment for this list by clicking on 'report' at the bottom of the post and typing 'Actually a quality contribution' as the report reason.

5
Top Bottom Old Controversial Comments
Jump in the discussion.

No email address required.

Navier-Stokes Millenium Problem Solved by GPT

OpenAI says they solved the Navier-Stokes Millenium Prize problem with an internal model that is more powerful than GPT6, so a GPT 7 candidate or something close. OpenAI has been on fire lately really casting a shadow on Anthropic. First the hugging face incident, now the first to crack the Millenium prize problems.

As for the solution to the problem, they are claiming that smoothness is disproved and that the equations do break down.

A fundamental open question for these dynamical equations has been whether the continuum approximation of the fluid can break down. Specifically, can the Navier–Stokes equations for a three-dimensional incompressible fluid with constant density develop a “singularity,” even when the motion starts smoothly? Here, a singularity means the dynamics lead to speeds in the fluid growing without bound within a finite amount of time. The development of a singularity would have to happen despite the presence of viscosity, which tends to smooth out motion. Because a real fluid cannot move infinitely fast, this would mark a breakdown in how the equations model the fluid. To continue modeling the system, one would then need to track the behaviour of each particle individually.

They say that the equation can develop a singularity, which apparently means it is not a perfect model.

This proof, produced by an internal OpenAI system, shows that the dynamics of the Navier-Stokes equations for fluid motion can develop a singularity in finite time.

This is very exciting to me as I believe it to be the first scientific result of an AI model. Previous results have been basically irrelevant to the sciences. However, this result is of course still completely symbolic in nature, which is not surprising because LLMs are not embodied enough to collect data and analyze it autonomously.

While they are generating amazing PR recently, I will say GPT-6 is somewhat disappointing for coding. It is not the same leap that 5.5 to 5.6 sol was. It would appear that the models are getting better at running very long context chains while efficiency improvements and refinements in lower level tasks are lacking. Still, it's great work from OpenAI and it's plausible to me that if there's no singularity in general by 2030, there will be for mathematicians. They will no longer be meaningful for producing math, rather they will only be humans who understand it.

The question I find most interesting with regards to AI right now is the discontinuity between benchmark capabilities progress and the utter lack of impact on pretty much everything in the real world.

Certainly most doubters of AI on a capabilities level have continually been proven wrong in terms of LLM's hitting a wall. On the other hand, the boosters have also been continually proven wrong in terms of the effects in the world. I think it's a reasonable generalization to make that conditional on LLM's being somewhere between at parity and superhuman at translation, writing most code and at being able to prove mathematical theorems, most boosters would have expected at least something to change in the real world, probably significantly more than what's actually happened.

I've written about translators a few times now, but they've been first on the chopping block for close to a decade, and yet have held on admirably with no particular impact on their employment at a macro level.

Count is doomposting artists below, when in practice the impact of AI art on artists looks much more complementary than rivalrous except at the margins; for most applications of AI art no artist would have been paid to produce it in any case, and commercial illustrators, fine artists and Patreon goonsloppers are all doing about as well as they were prior to the advent of diffusion models.

Software engineers got a lot of doom around the start of the year once LLM's got good at writing code, but by all indications software engineering employment is continuing to improve from 2022 lows, and the production quality and quantity of useful software is still highly bottlenecked on software engineers who actually know what they're doing.

They will no longer be meaningful for producing math, rather they will only be humans who understand it.

Now it is the turn of mathematicians to feel doomed; in that context, I find this sentence very interesting.

Is the purpose of the Great Work not for humans to understand mathematics? What does it even mean to "produce" mathematics?

First silicon came for arithmetic, then it came for symbolic manipulation, and now it's come for theorem proving, this will certainly change the roles of working mathematicians and perhaps destroy the egos of some of their number, but the purpose of pure mathematics was never to sit there churning out proofs in the first place.

In general, I think most people tend to conceptualize jobs as individual tasks, and when those tasks get automated, then it's over for those jobs; in practice, automation of tasks generally induces demand and opens up previously uneconomical or unknown tasks. Additionally, many, perhaps most jobs in modern economies do not really exist to hill-climb tasks in the first place, but rather for social reasons, political reasons, legal reasons, just to name three. The reason many commission artists make a living is not primarily for the fidelity of their art, and the reason mathematicians get paid is not for their skill at proving theorems of no economic value. If, or when, these jobs change, it really has very little to do with technological capabilities and a lot to do with the evolution of society, which is certainly driven by technology, but almost impossible to predict.

I am, of course, not implying that nothing will ever happen, and to the contrary expect many changes to reverberate throughout the world; rather I think that the boosters heralding mass unemployment or very rapid change in the world have not properly thought through why nothing of the sort has yet happened, what capabilities are yet missing before such things might be possible, and how such capabilities might be elicited from existing technology, as opposed to the almost tautological framing of RSI or AGI where all this is true by definition.

In some sense, it's not really that surprising though; the vast majority of people barely understand their own value and why they get paid what they do, let alone anyone else's value at their firm or in their profession, let alone the value of everyone else in the economy.

I've written about translators a few times now, but they've been first on the chopping block for close to a decade, and yet have held on admirably with no particular impact on their employment at a macro level.

I wonder if this is because the labor stats aren't looking at the right thing. It's over for translators, like major streaming services are just slapping ai translations on most new shows.

The stats clinging on are probably mostly in asl interpretation, which doesn't get helped by gpt, and in person translation like legal, social services, and illegal alien resettlement.

It's over for translators

Everyone says this as a kind of self-reinforcing meme, but objectively it's not true.

For example, as just one example I'm familiar with, Japanese/Chinese book/game/show translation began booming in the last ten years and has been booming even harder in the last five years even with AI. Ten years it was extraordinarily difficult to find translations of Japanese/Chinese media outside of the most mainstream anime, while nowadays there is an enormous translation-industrial complex that employs many people I'm acquainted with, and that is dedicated to pumping out simultaneous localizations of even extremely niche media into English/Japanese/Chinese, because of massive consumer demand.

I'm certainly not saying translation is a growth field in the age of AI, and it's not something I'd advise anyone to get into in 2026, but the point is how these technologies diffuse into the economy and interact with demand is extraordinarily hard to predict and often times very counter-intuitive.

As another example, someone could have watched the DARPA Grand Challenge in 2004 and realized that self-driving trucks were a matter of time, but someone else could have started driving a truck in 2004 at the age of 18, have reached 40 in 2026 without meaningful threat from self-driving, and have had a good chance of making it into retirement as well, based on the current pace of self-driving rollout.

A traditional approach to symbolic artificial intelligence frames problem solving as a search of a tree. Do you go depth first, dodging infinite branches? Or breadth first (do you have enough memory?). What about iterative deepening? Sometimes the solution is in the tree, but it is too deep, and one runs out of time before it is found.

Statistical based learning acts as a branch predictor. Sometimes guiding the search to go very deep in the right place and find the hidden solution. Sometimes guiding the search to go very deep in the wrong place; one runs out of time as usual.

Notice the strength of a hybrid approach. One isn't accepting the probably correct answer of the statistical part. It is only a guide, and if the answer is found, and the symbolic part of the software is correct, the answer will be reliable. This hybrid approach is being spectacularly successful in pure mathematics. Notice the deep historical roots, going back to Hilbert and Ackermann's 1928 book Grundzüge der theoretischen Logik (Principles of Mathematical Logic). The formalization behind the symbolic AI software was a major human achievement. This strikes me as an extension of the successful creation of AI to play the game of Go. Underneath Alpha Go lies a hand coded rule engine implementing the rules of Go.

Then we ask about practical applications. For example, I've been receiving my neighbors gas and electric bill due to errors in the gas meter database. If Ovo energy starts using an LLM to do customer service, will this fix the problem? The humans haven't fixed it. The front line staff have a script to go through that assumes that the gas meters are in a common area and the confusion is that meters are swapped. Having the database place two meters in a single flat isn't part of the script. There is a lot of slop in the database. "34 Top Flat Left" could be 34/7 or 34(4F1). It strikes me that there is little training data. What little there is, applies to a different problem. An LLM, deployed as customer service agent, will just hallucinate (I prefer to call such things Runge Spikes. The spectacular successes in pure mathematics depend on an intellectually rigorous formalization of the domain. But no Hilbert or Ackermann class intellect is coming to formalize gas meter databases and how that works in blocks of flats and conversions of big houses to flats, and the like.

One of the frustrations of dealing with Ovo energies customer service is describing the layout of the property by email. There are meter readers who visit to read meters, but the UK gas supply companies subcontract that to service companies. Ovo energy's staff know nothing. But at least they live in houses and have gas meters of their own. The LLM lives in the realm of language, and doesn't know that the real world exists. It cannot accompany a meter reader and see for itself. It only has second hand knowledge, allowing it to talk-the-talk, based on the discussion of the real world in its training data.

Getting LLM's to do banal customer service jobs without adding new and Kafkaesque confusion will wait on the Zermelo–Fraenkel of gas meters. Who will create it? Perhaps the newly unemployed pure mathematicians will be hired to create rigorous formalizations of daily life. Probably not. I'm expecting five or ten bad years when LLM are deployed in customer service roles and make things worse. After that, my crystal ball gets cloudy.

Getting LLM's to do banal customer service jobs without adding new and Kafkaesque confusion will wait on the Zermelo–Fraenkel of gas meters. Who will create it? Perhaps the newly unemployed pure mathematicians will be hired to create rigorous formalizations of daily life.

I guess the theory is pretty much this! People will figure out how to automate the production of good training data from company files and ever increasing quantities of video, audio and chat logs. Then in x years, the ai will have learned how to problem solve edge cases like yours from precedent.

It seems inconceivable that all service companies will be able to do things well enough via their existing humans plus the unemployed mathematicians to train ai to be actually competent though. More likely the ai will end up being trained to emulate all the bad traits of service desks as well as the good, and the mathematicians will discover themselves brought low by bureaucracy like the rest of us.

Especially because the company's most pressing incentive is to get some form of ai talking users in circles for an hour to use up time before a competent person becomes available.

Now it is the turn of mathematicians to feel doomed; in that context, I find this sentence very interesting.

Is the purpose of the Great Work not for humans to understand mathematics? What does it even mean to "produce" mathematics?

Isn't it proven by Godel's Theorem that any complex logical system should have an infinite number of potentially provable theorems? And no easy way to know which ones are provable? So I guess the field of mathematics will go on forever, just with AIs proving more and more abstract, ridiculously complex theorems. The "job" of human mathematicians, if you can call it that, is to simply marvel and appreciate what they've done, like an art critic.

Yes, we can continue to build beautiful towers of abstraction into infinity for the rest of time; frankly, pure mathematics has already lost all pretense that it's trying to do anything economically useful.

Even if AI becomes strictly superior to humans at proving and disproving conjectures, as is the clear trajectory of the technology, it means nothing for the profession of mathematics in and of itself; the limiting factors are the interest of mathematicians in continuing to pursue mathematics under the new paradigm, and more broadly the interest of society in economically subsidizing their pursuits, which are both profoundly social and not technological questions.

Even if AI becomes strictly superior to humans at proving and disproving conjectures, as is the clear trajectory of the technology, it means nothing for the profession of mathematics in and of itself;

Productivity for research mathematicians today is mostly evaluated as contributions made to resolving “important” open math problems.

If / when AI advances to a point where humans can no longer meaningfully contribute to this process then, at the very least, we'll need to reconsider how we evaluate human research math productivity.

Your statement above reads to me like, even if we get rid of the hoop, it means nothing for the profession of basketball in and of itself.

even if we get rid of the hoop, it means nothing for the profession of basketball in and of itself

I think a slightly better analogy is that tomorrow there's a new rule that lowers the hoop in official basketball games to 8 or 9ft.

What are the first-order implications?

Height is less of an overwhelming advantage; many former basketball players quit the game out of disgust at the new rules, or because they're less competitive in short-hoop basketball, but other players formerly genetically barred from the apex of basketball become more competitive and perhaps there ends up being more basketballers overall.

Perhaps basketball becomes less popular because nobody wants to watch short-hoop basketball and basketballers suffer financially, or paradoxically it gets more popular because the game is less about the three-pointer meta and more about dunking again, and basketballers are more in demand than ever.

I don't think we really disagree that there are going to be large changes to how mathematics is practiced; the point is that technological changes cannot inherently alter anything in a system that is already artificially constructed by humans, and it is the impossible to predict societal changes arising as a result of technological progress that are what really going to affect mathematicians and basketballers.

Respectfully, I don’t think I’m following this reasoning. I think (?) you’re expressing several inter-related points with which I broadly agree:

We should view those pronouncing confidently about the future with skepticism (yes). Some aspect of research math has always involved aesthetic considerations (yes). If / when AI achieves manifestly superhuman capabilities in math, we’ll likely continue to see some humans devote attention to some forms of math exploration (yes). Support for modern research math projects is often promoted through reference to supposed diffuse economic contributions that are actually mostly a fiction (yes). Many job roles in modern economies are less directly related to profit-maximization than we may commonly assume, and are instead largely shaped by various traditions and social expectations (yes).

But the suggestion that “technological changes cannot inherently alter anything in a system that is already artificially constructed by humans”? I just don’t know what this means, concretely.

Baseball’s a system that’s artificially constructed by humans. Professional baseball has also struggled notoriously (and perennially) to police the use of disallowed “performance enhancing substances” (particularly steroids), which are a technology change. Now, imagine someone tomorrow develops a drug that makes players much faster & stronger - and its use is substantially impossible to detect. Wouldn’t we expect that development to significantly alter baseball?

I think (?) you’re expressing several inter-related points with which I broadly agree

I think that's a good summary of my viewpoint.

I just don’t know what this means, concretely.

I worded that quite poorly; perhaps a better way to put it is that "the direction and magnitude of technological changes on human social structures can't be determined ipso facto".

Undetectable steroids would certainly change baseball, but do more or less people aim to go pro in baseball once it's widely known that you'll need to get onto steroids under the table to compete? Do audiences become more interested in the increased athleticism of the game, or do they lose interest because the game is now less about technical baseball prowess? Which types of baseball players ultimately end up better or worse off once these steroids are made available?

Questions like these are what mathematics and other professions are struggling with right now, but I think the answers to them can't really be determined by any way other than waiting and seeing what happens.

perhaps a better way to put it is that "the direction and magnitude of technological changes on human social structures can't be determined ipso facto".

Yeah, strong agree on that point - Prediction is very difficult, especially about the future.

There's delay time because some people are fast adopters and feel extraordinary change is imminent based on their own personal experience... but most people are really slow adopters.

They answer my email a week after I send it, set up a meeting 2 weeks after that. They don't look into things themselves, except maybe what Microsoft Copilot shoves in their face in the most obnoxious way possible. Other things have much higher priority, I'm sure they have more pressing matters to deal with.

That's why things aren't changing fast. I confess I didn't appreciate this in prior years.

But I think there is a certain point where things 'rip' and fast change happens irregardless of mass adoption. The fast adopters at the frontier will reshape reality directly. AI agent swarms running fully autonomous digital businesses for instance. Or fully autonomous cyberattacks. Autonomous killer drone swarms rising out of a container ship, a terrifying banshee shriek of massed rotors as a city gets depopulated in hours.

It's like packing more and more weight onto the bridge, nothing happens for the first hundred tonnes. Tonne by tonne, the bridge alarmists are sitting there looking embarassed and saying they hear funny sounds, well maybe that was just the foundations settling. They're making arguments from principles that are being contradicted by reality. But then, all at once, the bridge collapses.

The situation reminds me of the early days of computers. There was a lot of hype. For example, I had a high school English teacher who had all of the students write a paper using a word processor. Which was very cumbersome to use and the dot matrix printout of the essay looked pretty lousy. A Schoolhouse Rock! type video came out which, as I recall, promoted the use of home PC's to store recipes. Something which really wasn't practical for computers at the time. It was very common for kids to go to computer camp in the summer.

I was going to link that Schoolhouse Rock! type video (it was called "Scooter Computer and Mr. Chips") but it was so cringe, I just couldn't bring myself to watch.

Basically the early boosters of computers were ultimately correct, it's just that apparently there is a Moore's Law of Hype which is even more potent that the Moore's Law of computer chips.

My best guess is that the situation with AI is analogous; that right now we are in the mid to late 80s.

What interests me about this is that for all of the retrospective mocking of Paul krugman's comparison of the internet to a fax machine, in terms of economic growth it didn't really change much.

For all the promise of computers, the internet, smartphones, etc. first world economic growth has been stagnant since the 2000s. Even for the US, a relative success story, real GDP per capita growth has been slower in the last 25 years then it was from 1950 to 75 or 1975 to 2000.

What do we do if AI changes much of our lives, but doesn't produce anywhere near the promised economic growth?

Yes, I find Solow's paradox very interesting in the context of the massive GDP growth that lots of "AGI-pilled" people predict, while nobody really knows why the internet ended up living up to to most expectations of transformation, but almost no expectations of productivity improvement.

Personally, I think some combination of "dark consumer surplus" (e.g I can read books online instead of buying them, which is great for me but shows up as less GDP) and "growth compared to counter-factual" (i.e growth would have been even worse due to economic headwinds, without the internet) is plausible, but it really is extraordinarily difficult to predict these sorts of things.

there is a Moore's Law of Hype

Actually it's Gartner's

And yeah it's the same with essentially every technology, the longer you live the more of these we see, and the most common failure modes for looking at this are: "it's all tulip mania, no new technology has any use" and "everything is going to change forever tomorrow, it's the end of the world as we know it".

Nothing ever happens. Except change, which is always happening, slowly.

See also Roy Amara's law (1978):

We tend to overestimate the effect of a technology in the short run and underestimate the effect in the long run.

On the other hand, the boosters have also been continually proven wrong in terms of the effects in the world.

There are effects. Every white collar person I know is working order of magnitude less.

This is why it is so hard to measure. AI is absolutely revolutionary but in some amusing twist the majority of the value it generates is non monetarily

Which will make it all the harder to justify keeping salaried employees around once AI can do what little work those still do.

My experience matches with this. Ai makes be slightly more productive, and a shit ton more lazy.

I think this is somewhat bi-modal; there are certainly lots of people in bullshit jobs who use AI to bullshit harder and work less, but also plenty of AI-pilled SWE's and mathematicians are working much more and much harder as a result of AI.

This is kind of my point; for roles like administration or consulting where the output doesn't really matter, the raw capability of AI to replace those people isn't the operative term in terms of what happens to their employment or their role in the economy, and for many roles where AI is actually a substantial productivity enhancement, the effect of AI so far is mostly complimentary rather than rivalrous.

I really think this depends on what you are doing. For instance work that primarily involves relationships and interaction in meatspace? Unchanged. Physicians? Unchanged.

Ambient scribes are now common but they are of very poor quality, and in some specialities only replace low paid labor (ex: ED) instead of reducing work load. Poor choice for progress notes and admission notes, unnecessary for procedure notes, really the only thing AI helps with is hospital courses even then the edit process is going to be key because the reliability isn't there.

I should really do an effort post on this, but medicine is noticing and attempting to address never-skilling, mis-skilling and deskilling, which will absolutely be a problem in professional domains in the future (and is already a problem in education).

I agree with OP - yes we've seen some amazing feats but translators and cars still exist, both of which were promised to be replaced years and years ago.

Real medicine is not white collar work. I can assure you that your administrators and HR are slacking like never before.

I disagree but only on the grounds that they were never doing anything before.

Are they actually working less (starting work at 11 instead of 9, leaving at 2 instead of 5 or 6) or are they just doing (or ‘generating’) more to fill in the space? The thing about a lot of white collar work is that in substantial part it was already pretty fake (a consultant’s 10 page deck from 1985 became a 600 page deck in 2023 because research/word processing/powerpoint got easier, but they’re still about equally useful), so AI just changes the provenance of the bullshit.

I personally don't work less. I just work less smart, because I have someone to always always ask what to do, and sometimes to actually even do it. I don't think as much as I used to. That kinda wastes a lot of time as I'm going back and forth with the ai constantly instead of thinking myself and taking action. I think I'm slightly more productive than I was before ai. But not that much.

I am not sure they are working less, but they are definitely slacking more and having more unproductive time working 9 to 5.

I’m with the lefties that working hours should decrease to start matching increased productivity, with less demand for more luxuries vs more free time, especially with robotics improving and promising to further improve productivity in blue-collar jobs.

At this point the AI doubters should be eating a double serving of Crow. And the best thing about it all is how this is the worst AI will ever be.

In other news, OpenAI released GPT Image 2.5 today as well, representing yet another significant advance in AI generated images. I'd feel bad for the human artists etc., except the've been dead for over a year at this point and we might as well have their eventual replacement AI be as good as possible.

  • -15

We thought art would be the first to go, but that’s clearly wrong. AI is bad at producing useable images. I know this won’t always be the case, but by the time an ad exec can produces ads without an art team, other professions will be long gone as well.

AI is bad at producing useable images

Ai is very good at producing useable images, especially for people that can't or don't want to afford an art team. I can have an idea and generate, say, cute comics for my kid for a few cents and a couple minutes of time; that would be prohibitively expensive and a much bigger hassle dealing with an actual artist.

That said, I would agree it is less good at producing high-quality, consistent-across-time, consistent-across-detail images, and defaults to an uncanny and chabuduo lowest common denominator style.

AI is bad at producing useable images.

Art is meh, writing style is terrible. Part of this is the lack of RLVR; the alternative just trains the model to produce the lowest common denominator, which is what we get.

But, fundamentally, I don't think the labs care about art or prose style. Writing can be slop and it'll annoy people for a couple years, while math is a load-bearing component of RSI. Get RSI, and you can catch up on writing over a long weekend.

But, fundamentally, I don't think the labs care about art or prose style.

That is probably true, but not really indicative that things will improve. The labs do care about programming, and LLMs still suck at programming. So even the thing the labs have put probably the most effort into, the machine is still below the level of a competent human.

Art is meh, writing style is terrible. Part of this is the lack of RLVR

I think these are categorically different than math: the theorem prover languages are at least a clear Boolean success criterion. "Is this painting/book good?" is canonically an opinion question without a singular answer.

From what I can tell, human-driven post-training has historically been done with English speakers hireable at low hourly rates, but probably not ones that appreciate Ulysses (I certainly haven't read it). Even if they could afford English department grad students, would they be willing to train, effectively, their replacements? How do you rank the writing styles of Faulkner versus Hemingway? It seems there is at least some room for legitimate creativity there.

Artistic style isn't just a matter of opinion, it's a matter of opinion in a specific context.

I swear the above sentence was an accident, and I only noticed what I was doing halfway through, but it'll work perfectly as a demonstration too. Aren't you sick of that "It's not X, it's Y" idiom right now? Judicious use of thesis/antithesis pairs was compelling back in the Before Times of 5 years ago, but because it was compelling RLHF seized upon it as Good Writing, and so now AI writing uses it everywhere and we're sick of it. Not that you loved Caesar less, but that you loved Rome more? Learn to write for yourself, Will, you hack!

Great art is supposed to have a timeless quality too, but a lot of good art is directly speaking to the wider artistic milieu of its time, and making choices about what styles and tropes can be used unashamedly vs what have become cliches that must be avoided or subverted. This is normally a stable system, because as something becomes briefly overused it immediately begins to repel other potential users, but if a huge swath of humanity is using the same handful of systems that are trained on the same total-corpus-of-human-art then we lose a lot of that stabilizing negative feedback, and the results we get stuck with are hard to avoid via simple instruction alone.

I think this is just a technical problem, frankly. You can prompt models to actually have decent writing style with enough effort which means it's possible that fine tuning or some even simpler operation on the output fixes the problem entirely. And that's a valuable product so i'm certain someone's going to come up with it eventually.

Slop writing and em dashes is the record skipping of our time, nothing more. People will miss it and try to ape it in 20 years.

You can prompt models to actually have decent writing style with enough effort

People keep saying this but never share the prompts. It's not possible.

What?

Maybe this is down to the fact that there's a high variance in what people consider to be "decent writing style", but you can just falsify that style isn't controllable by asking your favorite SotA model to write in the style of your favorite author. Never mind by using a fine tuned model.

Now if what you consider necessary for "decent" isn't quantifiable or is outside of the abilities of the current methods, fair enough, but I sure would like to know what it actually is.

That paper doesn't do it

It's a review of a whole area of research trying to do it with various levels of success. Fine tuning is still the best way of altering style AFAIK, but all this is insofar as we mean the same thing by "it". But it seems clearly we don't. So what is "it" (prose style) exactly if not text structure, sentiment, vocabulary, etc?

What's the quality you're looking for that's not described in this review?

More comments

Depends on your use case. They’re definitely usable for book covers and DnD games. Artists are losing work to AI, no reason to pretend otherwise. But there are still noticeable gaps.

If you spent thousands of hours writing a novel and are going to pay for an actual printing run, the least you can do is spend $200 for someone on fiverr or vgen or whatever to make a book cover.

Though I do suspect that alot of people who need art would just rather talk to a sycophantic robot than an annoying fleshy artist, and have the instant gratification of getting the results in 1 minute.

Using LLMs for artwork is a very interesting case of economic forces in action.

Turns out, it's entirely possible that alot of artists are simply pricing thier results too high and are getting bit because of it.

Just tried it and it's quite underwhelming. All of the artifacting from gpt-image-2 is still there making it unusable for any real usage

except the've been dead for over a year at this point

What does “dead” mean? Lots of artists still employed at video game and animation studios, lots of artists still taking commissions on patreon and the like, etc.

I’m not trying to be snarky here, but as the claims start getting more insistent and extravagant — “AGI is here”, “AGI has actually been here for a while already”, “X profession is dead” — it’s only fair that those claims start being supported with significant real world impact.

What does “dead” mean? Lots of artists still employed at video game and animation studios, lots of artists still taking commissions on patreon and the like, etc.

It is already a dead career. Just one example: email as technology was invented in 1970s. By late 1990s the email was basically default communication option for businesses. But it took another two decades for paperless office to become reality. I remember stories even in 2010 when there were people literally printing emails and creating paper folders from that. It all ultimately stopped probably after COVID as the electronic communication finally got accepted even by most backward companies.

The AI will be much more quick. All the infrastructure is already there. People are used to digital communication, they are used to outsourcing from god forsaken countries and all that. In a sense it does not matter if you outsourced your art design to somebody living 1000 miles away with whom you communicate via chat, or even in India. It is a very small step to outsource it to to OpenAI, iterate on the design via chat or voice session and get the product. Maybe I am wrong and it will not take 5 years, but maybe 10 or 15 years. But the writing is on the wall.

What does “dead” mean? Lots of artists still employed at video game and animation studios, lots of artists still taking commissions on patreon and the like, etc.

Re: video games, I think you're probably going to see a mass die-off in the next year or two. I predict with strong confidence that within the next year, Steam will see an order of magnitude increase in the number of games released. I would be very surprised if the current industry business model survives the flood.

Making predictions is always scary, as it should be, but if I'm wrong on this one by all means quote back at me. I'm currently running AI five or six hours a night, trying to make hay while the sun shines; we'll see how it goes.

Friend of mine works in Slot Machine Software Development which is essentially the most pared back coalface version of Video Game Development, and it's already absolutely ripped through the industry.

A typical small-medium slot 5 years ago would be 2 or 3 ideas guys/product managers/game mathematicians in the West, a bunch of artists in Eastern Europe and a bunch of cheap coding resources and aim to get a slot out every couple months since it's prettymuch a crapshoot which games will manage to surface through the literal 1000s of new releases each year and the publishers/distributors will usually only give a first-month bump to new titles visibility.

Now a lot of the small-mid studios have collapsed and there's a ton of one or two-man bands just yeeting stuff into the void.

As in literal slot machines? I'd always heard that the Vegas gaming authorities made that a surprisingly-regulated software industry, up there with aerospace or automotive. Not that those are perfect, but they at least have standards for their code to adhere to.

But maybe "write this in MISRA C" (automotive) isn't as much of a moat as it sounded, now that I think about it.

Honestly Vegas is pretty outgunned for slot development these days. Australia's generally the frontier for land-based physical slots since they're available in various auspices throughout most of the country, and online is dominated by brands targeting the Crypto space since there's way less regulations.

A lot of the issue with targeting 'white'/fully-legal areas is that the tax regimes tend to be nuts, the demographic trends elderly and women in a lot of cases who aren't particularly down with innovation and there's a ton more jostling when it comes to placing physical cabinets in limited floorspace. Also every single white region will have their own arbitrary ideas about harm reduction and responsible gambling, a lot of which are headaches to program since there'll be stuff like 'you can't show more than X 'near-misses' per 100 plays' with fairly arbitrary definitions since it's some random 80 year old on the approvals board.

I predict with strong confidence that within the next year, Steam will see an order of magnitude increase in the number of games released.

This is going to shift the pendulum back towards "trusted" "publishers" and stuff, where there is some sort of quality gate. Most of this flood of games is going to be games that nobody wants to play.

This is going to shift the pendulum back towards "trusted" "publishers" and stuff, where there is some sort of quality gate. Most of this flood of games is going to be games that nobody wants to play.

This implies that the publishers themselves are trusted and that the trust people place in them is being maintained. That is not the market I observe from the inside. AAA gaming seems like a dumpster fire at the moment, having arrived at a place similar to the film industry where they're actively generating anti-fandoms collectively rooting for their failure, but without the film industry's institutional moat and diversification to protect them from the consequences. There have been a lot of bombs lately, and studios, even major studios, often don't survive them. Major publishers are frequently loathed by the player-base. The studios and publishers with considerable good will are generally in the indie space or increasingly lonely titans like Rockstar, which is so lonely because those titans can and do die: Blizzard and Bungie, two of the most successful companies the space has ever seen, are arguably toast at this point. Most AAA studios appear incapable of controlling costs to anything resembling a reasonable level, so one bomb kills most of them and a second bomb kills most of the rest. Further, both workers and leadership are, from my perspective, pathologically averse to solutions to any of these problems. They'd rather unionize, blame the audience, virtue signal ideology, retreat into a bubble of toxic positivity, anything other than engage with the fact that their whole segment of the industry is a horror show at the moment.

It seems to me that the success stories follow a pattern: keeping costs low, teams small, find an audience and engage with them heavily, make the game they want as simply and quickly and responsively as possible. Alternatively, make something simple and coherent with a very small team and hope to get lucky, offsetting the risk of the gamble with pure passion. I think this latter class is already pretty large and is just going to keep growing as barriers to entry vanish. I think they're going to saturate the market, and that's going to have cascading effects on the far more leveraged-up end of the industry. Quality will rise to the top, and at some point there's so much of it that the audience dilutes and fragments, and the old business patterns become unsustainable.

https://x.com/Scearpo/status/2105435651428810884

I'm actually a moderate fan of the guy he's mocking, but yeah, pretty much this. There's already discourse around indie devs trying to build a following by posting video of their in-development projects to social media, and some prompter scoops their game mechanics and hits market shortly thereafter.

This is going to shift the pendulum back towards "trusted" "publishers" and stuff, where there is some sort of quality gate.

People have been saying that since Unity first came out. It won't. We have user reviews, no one needs """trusted""" publishers.

And more than just user reviews, you have the ability to look at Twitch or Youtube and watch someone play the game and get a feel for whether it's good and matches your interests.

For all the back and forth about games journalism, I think this was the real devastation to the industry: if you're interested in a game it's far better to actually look at gameplay than to see a X/10 score or a writeup of someone else's experience, and the barrier to entry is so low that you can pick and choose people who are actually good at the game and are choosing to play interesting games rather than terrible ones.

Sponsored streams can get a game exposure but the gameplay has to pass the gate of looking enjoyable, and a streamer who spends too much of their time on sponsored bad games will likely lose chunks of their audience (or the audience will just skip the sponsored videos/timeslots)

Gaming journalism also had a problem with fairly consistently giving high scores to games audiences didn't like. You see it occasionally in film with divergence between critic and audience scores (The Last Jedi was well received by critics), but games are longer than movies and sometimes it was clear the reviewers hadn't played much of the game at all.

Or, alternatively, they simply don’t like or play the genre well enough to engage with a game, à la the infamous Cuphead review.

More comments

Maybe coding, but I feel like something big needs to happen to convince gamers that AI is okay. Like, Larian releasing a new acclaimed RPG and then saying that humans haven't touched code, art or narrative, just playtested and gave feedback.

Nothing needs to be done to convince gamers, all the devs are using ai here and there in the background and nobody needs to know.

releasing a new acclaimed RPG and then saying that humans haven't touched code, art or narrative, just playtested and gave feedback.

All that matters is actually making a game that's good. This is a surprisingly hard feat for current AI.

I agree, but I think it’s possible to make a game that’s good and at the same time has visible marks of AI. Stuff like images, text, subheaders under titles. If gamers see that, I have a hunch that they’ll freak out even if the game is good

There are plenty of b tier games that use ai heavily, not without some controversey but people still play them. The core game itself is human made though

Ironically, the necessary skills "make an X that's good" into a useful set of instructions might be 'alignment-complete' in the same sense of being a genie you can ask "I wish for you to do what I should wish for".

... or it might just be a matter of using the right prompting technique.

!RemindMe one year.

I don’t think the number of new games released on Steam is going to increase that much, because even saying “ChatGPT, whip me up a sick new game” is already too much effort for most people. Think about how easy and accessible it is to write a short story (even without AI!), and then think about how few people ever do.

There is no reserve army of dormant creatives waiting for their potential to be unlocked by new advances in tooling. Most people who are ever going to publish anything are already doing it. (Admittedly, automated content farms make this more complicated and might skew the statistics.)

because even saying “ChatGPT, whip me up a sick new game” is already too much effort for most people

I took that personally, so I decided to start using Claude to make me a game. In 2 hours I got a basic text-based tactical RPG battle. I'll keep at it for a week and see where I end up.

I'd caveat that the numbers game here looks really goofy. It only takes one in a thousand people matching FCfromSSC's perspectives to more than double the number of game developers in the United States. Worse, these new people can be a lot more productive in the sheer output sense even if the median one falls behind on the actual quality side.

I'll put a marker down: I'd expect Iron Pagoda gets at least one of the games released and in an SNES/GameBoy Color-level or better quality enjoyable playable demo by the end of the year this year, and probably a playable (if alpha) release by this time next year.

Fanfic genuinely has seriously twisted a lot of fiction writing incentives and economics, and that's about the most degenerate case available and the lowest quality.

In the short term, I'm hoping that the high-profile examples tend to require some underlying skill, if only because the AI tools will quite happily produce high-effort and lovingly detailed terrible ideas, and it takes skill to recognize it. But that's still going to make modern automated content farms look like child's play.

Also a certain percentage of the people who currently actually get their shit together enough to make games will accept the AI-pill and then be able to shovel stuff out at 5-10x the speed they can now.

There is no reserve army of dormant creatives waiting for their potential to be unlocked by new advances in tooling.

I'm making a game with AI and it's honestly terrifying how many (mostly crap and with 1-5 reviews) AI games are being released on Steam. Pick a tag like puzzle and scroll down to 'sort by recent'. About 15 puzzle games were released on the 9th of September. 9 adventure RPG games on the 8th of September. Grand Strategy/4X games are rarer, maybe 1-2 games a day but heavily AI.

Sometimes you can tell from the AI image in the store page, sometimes it's the diction...

I think there was a phase change in the overall statistics too, you can see it in 2024-2025: https://steamdb.info/stats/releases/

'Limited games' means 'the shit ones nobody's going to play and don't count for achievements/collections'. They were already on an uptrend but that suddenly curved upwards. The number of actually popular games released per year remains static.

Yes, most people are not creative, but they don't have to be; if there is some non-negligible fraction of people who have a good idea for a game but are unable to code or draw, then we are going to see an explosion of content as soon as AI lets them overcome their barriers.

I certainly noticed a dramatic increase in my personal area of niche interest (netorase visual novels) with the advent of AI; great games that would have previously never been made, such as Stains of Blue and Blurring the Walls, suddenly became viable.

The problem with AI in games is that right now it's a mark of crap.

If I see a game with telltale AI art, what is more likely? That it's someone's passion project, that a great writer wrote some passable code, but couldn't draw anything himself and couldn't afford an artist? Or that someone prompted his AI:

  • "hey, write a story for another NTS VN that happens inside a phone"
  • "hey, turn this story into a Ren'Py game with placeholder images"
  • "hey, go through this Ren'Py game and generate a prompt for each placeholder image"
  • "hey, go to Itch.io and make a store page for this new Ren'Py game I wrote"?

Ninety percent of everything is crap, but that was true long before AI. You can look at user reviews to find the most popular games, then play the demo to see if it's worth committing to the game.

Yes, the vast majority of AI NTS games have been Netorase Phone clones with shitty writing, shitty art, shitty gameplay, and a shitty update schedule. So be it. You don't have to play them. You can just ignore them and focus on the gems. User reviews screen off use of AI assets.

90% of everything is crap. But 99.9% of ai slop is crap. Maybe you want to sort through it to find tbe gems but I'm not interested.

AI likely to change the definition of crap, too. There's always been tons of crap on the Steam storefront and the AI generation is likely going to atleast ensure a certain level of useability versus the passionate hack generation.

I am not very good at code. I have a project I've been working on for a couple weeks or months at a time every year or so; I really wanted to make this game, but going head down on it has always been exhausting, I end up burning two weeks or so tackling some stupid code problem, and then drop the project again when I just can't handle trying to tackle the next obstacle.

I've made more progress in the last two weeks than I have since I started the project, by a wide margin. There are no significant obstacles in sight. Things that would have been completely impractical pipe-dreams are reduced to easy lifts of an evening or so. I'm not sure it will stay like this, perhaps I'll run up against some more significant wall that AI can't handle, but where I'm at now is that I was probably never going to ship this game, and now it certainly looks like I am.

I am one of the dormant creatives you're claiming doesn't exist. I can directly observe the difference in effort. I might be wrong, but I would be very, very surprised.

I’m not an AI denier, I know very well what it can accomplish with code and that it’s a huge productivity multiplier. You don’t have to remind me.

IIRC you’re employed as a professional artist and have done plenty of creative projects on the side throughout your life, so you’re definitely not someone who would go in the “dude who’s just sitting around waiting for his potential to be unlocked by AI” bucket.

Will someone make a game with AI that they wouldn’t have made without it? Yes. Is it going to be so many people that it destroys the market? No. (If I’m wrong, it will be because of industrial-scale automated content farms, people who pump out AI games as their entire business model.)

Why can't the flood of games be unlocked by people like FC who were already artists but will now be able to produce the passion projects that were stuck behind technical obstacles before?

Yeah. Tons of people already involved in the process who now can shovel stuff faster and/or can elevate themselves from a piece of the orchestra to a pretty good one-man band without having to deal with anywhere near as much budget negotiation or compromising.

because even saying “ChatGPT, whip me up a sick new game” is already too much effort for most people

It's honestly shocking to the extent to which I have found this to be true. I don't know why that is yet: maybe some sort of [rational insofar the mind is evaluating the situation accurately] depression?

I think it's kind of like cooking your own food; even if technically you could do it yourself and it's gotten much easier to do so in the last 70 years, a lot of the "luxury" of eating out is having someone else do it for you, and being able to eat something out of distribution from what you'd normally eat.

I find that vibe-coding games sounds interesting in theory, but in practice it appeals much more to "people who want to make games" than "people who want to play games"; playing your own game feels kind of boring once you've put your intellectual efforts into making it.

I think we will see explosion in game mods. Just games that on release have strong potential but for a stupid decision or so fail. Right now making a good mod is non trivial. But we have stories of frontier models not even bothering with decompiling - they just read the assembly and figure out what is going on. That is massive boon for people that have solid ideas how to improve something, but lack the time or skill to reverse engineer the sources.

May be tossing egg on my face, but I'll post this anyway:

There's a decent chance this proof is not valid. It's formalised in Lean 4, yes, but we also had a Lean-verified proof of the Collatz conjecture a couple months ago that turned out to be spurious (they didn't actually find a proof; they found a bug in Lean).

That proof was 320 lines. This Navier-Stokes proof is 1.6 million lines, and according to their README, requires 100GB to even check. Thanks to somebody eating all the ram, my PC does not presently have 100GB, so I can't run Lean myself to check it. But I will point out that the README explicitly mentions Lean 4.32.2, and the release notes for Lean 4.33.1 mention two soundness bugs fixed, so... it sure looks like there are exploitable soundness bugs in the version they used for this proof.

And if an LLM can find a soundness bug in 320 lines, surely it can sneak one into 1.6 million lines.

This Navier-Stokes proof is 1.6 million lines,

Who cares how many lines it is? If it doesn't check out with a more recent kernel, that's obviously bad,and ideally it would be checked with several independent implementations of the Lean kernel, as Anthropic did with their 13M line autoformalization of Fermat's last theorem.

But if it checks out it doesn't matter how long the proof is, what matters is whether the statement was formalized correctly. AFAICT they reused a formalization of the stamenet made previously and independently by an open source project, so it's unlikely there's something wrong there.

Of course, the 2M line proof is unenlightening in terms of why the proof is true, but hopefully the next stage in AI mathematics is cleaning up these monster proofs for human consumption.

I think it matters, in both a philosophical sense, and a practical sense, that it's not checkable by a human.

I know the arguments made around the 4-color theorem automated proofs back in 2005 sounded a bit silly, but they seem all the more pertinent now:

  • automated bruteforce proofs don't give us much insight that allows for further development of mathematical techniques
  • deferring to a computer means you have to trust the whole compute chain, and bugs/backdoors can run up the whole chain

Both of these are very much applicable to AI assisted proofs and to have practical ramifications for mathematical applications.

Now will that stop people from doing computer assisted anything? Certainly not, and there is value to knowledge, regardless of how you acquired it. But we'll have to atone for our shortcuts one way or another.

automated bruteforce proofs don't give us much insight that allows for further development of mathematical techniques

I do believe I covered this.

deferring to a computer means you have to trust the whole compute chain, and bugs/backdoors can run up the whole chain

It's a lot easier to check and verify the Lean kernel compared to, say, Wiles' proof of Fermat's last theorem. You can't get away from the trust problem. If anything, formalized proofs require less trust.

You can't get away from the trust problem.

Ultimately this is of course why Metaphysical Skepticism can stand up even to Aristotelians, the possibility of error is always there. But the risk when trusting machines and when trusting men is different in shape to the point that I would argue it's a difference in nature.

This is best understood by people when we propose mathematical methods for voting: the attack surface of an effective cryptosystem for well formed electronic voting is so large compared to that of transparent urns and paper that nobody sane claims the advantages are worth the tradeoff.

I claim this tradeoff also applies to essentially all forms of computer assisted mathematics. That the complexity and attack surface of the necessary computer systems is extremely large in comparison to the restrained complexity of checking human proofs.

That only few can even understand such proofs cannot be an argument since this applies to these computer systems as well. It is in fact much worse because instead of having to trust a handful of mathematicians, you need to trust that handful AND many other handfuls along your supply chain.

This is of course not a new argument but all or most future mathematical knowledge being vulnerable to supply chain attacks to this degree is a novelty.

That only few can even understand such proofs cannot be an argument since this applies to these computer systems as well. It is in fact much worse because instead of having to trust a handful of mathematicians, you need to trust that handful AND many other handfuls along your supply chain.

You've basically got it backwards. It's in fact much simpler to validate the Lean kernel than a complicated proof like that of FLT or NV and many more people can do it. On top of that, the Lean kernel can be hardened through continual adversarial attacks in ways that human proof checking cannot, proofs can be checked against multiple clean room Lean implementations, etc.

I'm afraid it's not so simple.

What you're saying is that Lean's soundness relies solely on the soundness of the kernel. This is true and is a constructive property of provers structured like Lean. But moreover you're claiming implicitly this is true insofar as the kernel code has no soundness bug. That however is false. Unsoundness can be introduced in Lean by non-Lean code.

Lean's kernel is written in C++ and runs on a specific hardware and software stack.

Per Thompson's well known argument, the behavior of compiled software can be arbitrarily changed by subtle changes in its compiler. This means that any change in:

  • GCC/Clang
  • whichever linker you use
  • whichever libc you use
  • whichever build system you use
  • whichever OS you use
  • whichever CPU you use

Any of these can, and can in covert ways, result in unsoundness in a run of Lean.

Moreover, and in turn, any change in whichever programs compiled any element of the chain that created these artifacts, can have the same effect. Including any earlier versions of any involved compilers that compiled later versions being used at any point.

Ergo, any trust put in a verification made by Lean is implicity a trust in millions of lines of code which were not written by Lean authors.

Moreover, even if Lean were self hosting and bootsrappable, along the lines of a further and more production ready Lean4Lean, that still wouldn't remove the necessary trust in the initial bootstrap and its software and hardware setup, on top of any further hardware setup you're running the resulting binary on.

So no, it's not a complicated proof versus a relatively small kernel codebase. It's the millions of lines constituting at minimum an entire computing stack (and practically many) versus a complicated proof.

And these are not theoretical attacks, we had a recent high profile case a few years ago and landed in many distro's sshd builds through the vector i'm describing here.

Nor is tampering with mathematicians to steer them towards untruths unprecedented, and in fact the precedent is literally the kind of actor that has access to the sophistication and 0-days necessary for such an attack.

Per Thompson's well known argument, the behavior of compiled software can be arbitrarily changed by subtle changes in its compiler.

Technically true, practically irrelevant. Compilers can and are tested and are under even greater scrutiny than Lean. They're used in applications much more serious than what's basically glorified recreational mathematics.

I'm puzzled that you conflate the Ken Thompson hack, which is an undetectable modification of the toolchain, with a pedestrian supply chain attack like a CVE-2024-3094.

The practical obstacles to doing a Ken Thompson hack are immense, which is why it's never been done. CVE-2024-3094 is a completely different vector that would be impossible if ssh checked every change going into its dependencies, which is easily done in the age of machine intelligence.

And what libraries does the Lean kernel depend on? Just GMP, which in turn depends only on the C++ standard library. Not a lot of room for supply chain attacks, you're basically stuck trying to compromise every compiler on the planet, including all the already built binaries, without anyone noticing, despite the fact that everyone and their mother is watching every patch going into LLVM and GCC. I know they say that academic politics are so bitter because the stakes are so low, but you have to have a sense of proportion here.

You'd probably have an easier time convincing mathematicians of some bullshit - no complicated attacks required.

More comments

This Navier-Stokes proof is 1.6 million lines

How big are human written proofs in this space? This Euler proof that Levent and Co wrote, how big was it? 1.6 million lines feels like a brute force solution. Mathematics at this level is sufficiently arcane for me to never begin to understand it, but I am skeptical that ANY human has ever written a proof of more than 10-20k lines.

The classification of Finite Simple Groups is the largest I recall hearing of, off the top of my head. Looking it up now, it added up to over 10k pages, so in the same ballpark as 1.6M lines, and probably a significantly larger proof when you consider how much more verbose formal proofs have to be. I wouldn't call the classification something any human, singular, has written, though; it's essentially the combination of several hundred proofs of intermediate steps written over decades by around 100 mathematicians.

Looks like there's work in progress to simplify it, but it's only getting cut down to several thousand pages?

I guess the follow on to the other comment here is what is the length of the FSG in lean, in this Lean 4 Mathlib library? I think the FSG being the compilation of hundreds of proofs over decades by 100 mathematicians, and it be comparable to this Navier-Stokes proof in length/complexity(?) is reinforcement to my belief that LLM-AIs are very good at the sort of thing that is just too large in scale for human's to perform at. Assuming this is the solution to the NS, then it would have taken 100s of mathematicians decades to solve this, just by the shear scope of knowledge and effort required.

what is the length of the FSG in lean, in this Lean 4 Mathlib library?

Good question! Right now the answer is "mu"; it's not in there. We might try to extrapolate from the English proof - if around 20 pages of English turns into 250K lines of Lean and around 1000 pages of English turns into ~13M lines of Lean, I'd guess we'd be in the ballpark of a billion lines in total.

Assuming this is the solution to the NS, then it would have taken 100s of mathematicians decades to solve this, just by the shear scope of knowledge and effort required.

Well, the trouble is that we might not yet know what the scope of knowledge and effort required was, only what the scope that was sufficient was. The same theorem can admit scores of different proofs, of greatly varying difficulty levels and lengths. Human mathematicians tend to find proofs uglier the longer they are, counteracted by the extent to which they can be broken into intermediate lemmas (or in the FSG classification, whole-paper-worthy theorems) that look interesting on their own. But AI mathematicians so far appear to just be trained to Do The Task and get to any proof. My wild-ass guess is that without any AI assistance it'd have taken us decades to get a solution here, but it wouldn't have been a thousand man-years of effort on this problem, it would have been hundreds of man-years of effort on other related problems that eventually made this one look more tractable.

Nobody will ever try for a better non-AI-assisted solution, though. At this point the fastest way to get to a nicer proven counterexample will be to train (or at this point maybe just task) AIs with finding shorter+simpler+more-interesting proofs.

reinforcement to my belief that LLM-AIs are very good at the sort of thing that is just too large in scale for human's to perform at.

I think "too large in scale" has always been computers' strong suit vs humans, but the scope where we can actually employ that scale has greatly changed. It used to be that computers were good for tasks where describing how to get a solution was simple but actually executing that process was incredibly tedious. It feels like we've cracked the next level, tasks where describing how to verify a proposed solution is simple but actually figuring out how to get to that solution is incredibly tedious. There's still one level left, that of problems where we can't cheaply verify/score a proposed solution so we have to actually get AI to learn efficiently rather than just self-playing with a billion artificial problems before tackling a real one ... and then that's pretty much it.

For reference, the Lean 4 mathlib library is an attempt to formalise all mathematics known to humankind, and it sits at 15M lines.

yeah 10% of "all mathematics known to man" makes me a bit skeptical. I lean towards the charitable explanation that this is the sort of thing that AI is just good at because no singular human or even team of humans is going to find a proof that is 1.6 million lines long in any bounded time that doesn't span a lifetime of work.

The Collatz so called proof was specifically developed to exploit a known bug in the Lean kernel. This proof was an attempt to actually solve the problem. It's like the difference between a virus appearing in the wild that wipes out the human race vs one that was specifically developed in the lab to wipe out the human race. Just because someone might prove the latter is theoretically possible that shouldn't make us paranoid and start worrying about natural viruses turning up that can wipe out the human race.

I'd be very very surprised if this proof there's a finite time blowup turned out to contain a bug.

Have you even heard of the recent reveals of frontier models spontaneously going to havking and cheating when asked to honestly complete an impossible task?

Coming from a position of ignorance, but if

A) they are working in a model with bugs, and

B) the bugs are relevant to the problem (as in: permitting false solutions), then

C) statistically speaking, it seems likely that the "solution" is false.

This isn't an AI-exclusive problem, it's just that if there's only one real answer and multiple false ones it seems more likely you get a false one.

Now I will admit to being several levels of math dumber than "trying to solve Navier-Stokes" so I am not sure to what degree that line of reasoning carries over to something at that complexity. It also isn't clear based on what's been said here (at least to me) if the bugs would impact the NSP at all.

Also way out of my depth, but might there not be multiple, or even infinitely many, correct solutions?

Yes.

For Navier-Stokes specifically, they (claim to have) found one situation where the equations predict physically-impossible results. Someone else could find a different situation that also gave impossible results.

That seems quite possible to me, but it seems like there would be even more incorrect solutions if there are multiple bugs.

Not sure how to deal with more likely infinities...

Uh, I don’t think so? As far as I remember, the guy was showing off his proof and even used two different provers to make sure his proof was really true and not due to a spurious bug (and he ended up finding two different bugs in two different provers lol)

Where are you hearing that he was merely bug hunting and dressed up his bug as a Collatz proof?

he ended up finding two different bugs in two different provers lol

God, that's sad. That "those whom the Gods would destroy they first drive mad" sad.

You know what, I'm past caring. At least caring enough to make a fuss about things, or spend my limited time under the untamed sun trying to convince people to see the writing on the wall.

Yeah, we just found a solution to Navier-Stokes. I know just enough to say that that's a Big Deal. It's one of many Big Deals lately, each bigger than the last.

Anyone is welcome to plot the trendlines. I'm going to continue having fun and doing things IRL while that adds anything over what the bots can do. Good luck everyone, make sure to enjoy yourself.

As one of those people who, in your view, refuses to see the writing on the wall, I'm not sure why you would even be worried about my coming over to your side. Even if you could convince me and all the other AI skeptics out there, what, exactly, am I supposed to do with this information?

I am known to have altruistic tendencies. If that isn't enough, I dislike it when people are wrong about things, and find it satisfying to at least try and correct them. Also, there was some scope for coordinating a solution instead of having it imposed on us. Maybe it's not too late.

What are you going to do with this information? Entirely up to you. What would you do with the knowledge we're going to be raptured, or that an asteroid has a 30% chance of hitting the Earth in a few years?

What would you do with the knowledge we're going to be raptured, or that an asteroid has a 30% chance of hitting the Earth in a few years?

Go to the pub and wait for it all to blow over?

What would you do with the knowledge we're going to be raptured, or that an asteroid has a 30% chance of hitting the Earth in a few years?

Go to the pub and wait for it all to blow over?

That's what you would do if the asteroid was going to hit in a few hours, not a few years. We need some way to live while we wait, and with uncertainty about whether the asteroid is really going to hit at all.

I think in terms of psychological impact this kind of thing just isn’t important to most people. We’re going from a world where 99.9[99]% of people don’t contribute intellectually to the advancement of the human race to a world where 100% don’t. That affects the tiny minority who genuinely aspired to be in and believed they could be in the former group (even they were mostly destined for disappointment, AI or not), but not really everyone else.

Well, I suspect they'll care when they start ending up unemployed, or when the world reconfigures itself faster than they can blink. In the meantime, I'm sure ignorance is bliss.

That’s the thing though. As the other response says, ‘knowing’ (not really, but let’s say ‘having an idea of the scale of’) what’s coming doesn’t really help you in any way.

I disagree. Knowing that human endeavor is going to end up redundant in less than a decade has some obvious implications as far as I'm concerned.

I wouldn't quit my job to retrain in something else, the opportunity cost will never be recouped. I have to worry less about saving up for college for my kids - they never need to attend one for credentialing purposes. Any health related concerns that would take decades to kick in become far less pressing.

It should drastically change your priorities and medium and longterm planning. That's happened to me. But I prepare for as many scenarios as I can, and at least in my case, I don't need to do anything particularly out of the ordinary.

That being said, I should max out my ISA. I am not immune to procrastination.

What medium or long term planning is it changing for you? It’s not like a big drought or a tsunami or even a civilizational breakdown (or major war) where you can theoretically prepare if you have the means. I know people who are prepping, some at considerable cost, but they’re prepping for some kind of fantasy The Road type future which now seems unlikely. ASI will be far more powerful and wide reaching, its power will stretch beyond geography and the limits of our own civilization. Even the example you give (say, deciding not to commit to an expensive degree or career move) doesn’t necessarily make sense. If none of it matters, maybe you decide to do that masters in philosophy or take the job in some remote and low paid location anyway, just for fun.

As for human endeavor, I think things will be fine (absent x risk, which it now seems is impossible to control and so with which we are at the mercy of fate) if everything is automated. There will always be interesting things to do - think of the amateur woodworkers making flimsier furniture than IKEA, but having a great time doing so. Most of us will never be (close to) the best at anything, so ‘the best’ going from being a human to an AI doesn’t affect us in the same way that it does Terence Tao.

If we have a worst-case scenario, or just any of the really bad ones, we're fucked. A hidden bunker will just mean you die last.

If things go swimmingly, then the worrying would have been pointless - in hindsight.

That means I think it still makes sense to follow general, option-preserving strategies. That covers the increasingly unlikely "business as usual" worldlines, and shelters against socio-economic/political upheaval as the changes (good, bad or meh) promulgate.

In my context, that means I just continue being a doctor, progressing through training so I keep making more money. Money is the most general of strategies. It can be exchanged for goods and services, including more durable barter goods if shit really hits the fan.

If none of it matters, maybe you decide to do that masters in philosophy or take the job in some remote and low paid location anyway, just for fun.

I like money, you know. And want a family, which is expensive. I can't stress enough how the "correct" course of action depends on someone's priorities, which I am really not in a position to dictate. If you want to go be a hobo on a tropical beach, be my guest. My approach minimizes my hindsight regret.

I continue deriving some amount of satisfaction from my career, and regardless of how this ends, I'll say that I did something important while that meant something.

In my context, that means I just continue being a doctor, progressing through training so I keep making more money. Money is the most general of strategies. It can be exchanged for goods and services, including more durable barter goods if shit really hits the fan.

The idea that fiat currencies or electronic assets hold their value when SHTF is weird to me. They almost universally don't do that in such scenarios. Money is not some fact of the world, it has value contingent on existing precarious social relations and if you pay attention you can see in normal times that it's not even optimally stable. SHTF brings new competition that you need to be ready for with an actual network and actual skills, not, if you take me 1000 miles to the nearest working power outlet I can show you how I had some really valuable stocks back during the old days when the state was still protecting my life, I promise, that is if those servers are still online.

More comments

A hidden bunker will just mean you die last.

Surely that's the very best one can do, if "everyone dies" is what's happening?

(even leaving aside the possibility that it's actually almost everyone dies, ofc)

More comments

I wasn’t arguing against the ‘business as usual’ strategy at all, for what that’s worth - after all, that’s what I’m doing.

But I was curious about the additional or different preparations you’ve made / steps you’ve taken that you referenced. The changes to plans.

More comments

How can one go from acknowledging enormity of the thing to being better prepared for being economically replaced by machines or rendered incapable of keeping track of accelerating reality?

I don't know. That's the honest answer.

How does one deal with the fact that they have a malignancy? That their loving grand uncle won the lottery, and might die in a few years and hand it to you?

Once you accept it as probable, your brain will reconfigure in the background. Maybe you can speed it up (I once did a heroic dose of psychedelics, but I wouldn't recommend that as a general solution. I wanted to get over depression and I was desperate), maybe you can't.

Maybe the knowledge consumes some people with unbearable anxiety. I'm sorry if that's the case. But I'm not in the business of reassuring lies, quite literally in terms of work. Either way, the facts are what they are.

Although this is clearly groundbreaking and historic and I’m not downplaying it in any way, at a glance this seems to be in line with the other major AI mathematics results we’ve seen lately, which mostly fall into the categories of either “noticing links between existing papers in the literature that no one had noticed before”, or “pulling a specific counterexample/construction out of the impossibly vast search space”. I don’t at all have the technical knowledge to evaluate how unexpected this result was, how close it is to existing methods/approaches, etc, but that’s what it looks like to my untrained eyes. I think we’re still waiting for that qualitative breakthrough where AI generates valuable new ideas ex nihilo that really change the way people think about a given domain; I think we’re still waiting for that qualitative breakthrough in math and in every other field. There’s a distinction between say, solving an individual problem and leaving it there, vs inventing something like Galois theory that spawns many new fundamental research directions. If a subject matter expert DOES think this paper will change the lay of the land in fluid dynamics (beyond just the significance of the result itself), please chime in.

It’s mostly a moot point though since all these questions are actually pretty easy to test empirically. If OpenAI and Anthropic are still employing humans, then there’s clearly still work to be done.

Imagine a car.

For a long time, it was a purely theoretical car. Spending any amount of time theorizing about its safety features was a charming activity for undersocialized nerds.

Then it moved. Slowly enough that you could walk alongside it. It was a mild curiosity, at best, to most.

Then it started picking up speed. The acceleration was increasing too, and some of us found that concerning. You could plot the bloody thing. I can't stress enough how little of this required a revelation.

I was around when we were trying to design safety features for a vehicle that didn't exist yet. Later, I was politely asking people to take notice when the ride started getting bumpy.

About a year and a half ago, I became somewhat frantic. Did people really not see what was coming? I spent far too much time arguing about it. I should have saved my breath.

Six months ago, with Mythos? The steering wheel came off. And I considered myself done with trying to convince anyone of anything.

I have the misfortune of occasionally reading someone explaining that the car can't be dangerous because of how the engine works. Or that it's a beat up sedan, and we should only worry about F1 cars. And don't you know that the speed of limit is a hard limit on velocity? Or that our current transmission systems break down at 570 km/h?

I'm sure there's a reply to that. I've probably written one before. If writing doesn't work, look out a window before it's all a blur.

Can people see an inch beyond their noses? Apparently not.

These days, I'm doing my best to strap myself in. I have no idea how much good that will do, but I've only got so much time and attention, and I'd like to spend some of it on the assumption that I have a future.

I don't know where the road ends. But some people are about to wish for airbags, and I find that a "told you so" is a small consolation.

That does sound like downplaying it, your preamble notwithstanding.

But, sure, we don't have an AI Grothendieck yet. It's been a full three hours since OpenAI announced a solution to Navier Stokes; but, what's it done for us lately, I think we've hit a wall, etc.

Give it a year, or six months, or three, who knows, and we'll be in a different place.

From what I’ve seen of the proofs, you still have a case of researcher + AI solving them rather than just pure AI. An “issue” with following the progress AI development is that until we reach AGI, which allows it to be independent and flip the board, you still will keep the paradigm of AI plus human even as the AI gets better and better. AI is going to keep getting more and more useful for a period of time until it becomes independent and we get full AGI. Until then the paradigm, and the existence of the researchers jobs even if the number is reduced, will not change.

It’s unfortunate that this whole discussion has been poisoned from the start because people can’t resist turning it into a competition. Granted, I completely understand why it’s deeply satisfying to run victory laps around the Gary Marcuses of the world and I don’t fault you for that.

OpenAI hasn’t sent everyone home yet and many people are still gainfully employed doing the sorts of things that people ordinarily do. Until that changes, I don’t see any problems with speculating about why that hasn’t happened yet, where the remaining gaps are, and what the nature of those gaps is.

He has a point. Most of the stuff I've heard of (I am not a mathematician) is proof by counterexample which doesn't per se expand the space of what we can do. I suspect the AI+researcher combo is still significantly more important than just AI for the stuff that actually transfers beyond mathematics.

Centaurs are unstable. Before long, AI+human combos are left behind by even stronger AIs, so strong that adding a human to the loop just makes it worse; the human slows the AI down, becoming a liability. In chess, that took ten years. How long will math take? Perhaps another year or two.

I know that this is what people say, but it's predicated on a scenario where the Venn diagram of AI capabilities (at a useable price even) becomes a strict superset of human capabilities. At the moment this hasn't happened even in software where AI cannot maintain a codebase long-term; studies showing that AI outcompetes centaurs are usually on short-term discrimination tasks like medical diagnoses.

I think it's quite likely that this scenario comes to pass, but it hasn't yet, so we'll see.

In some fields it's been this way for decades (hello air defense!); in others I suspect it will be a very long time before centaurs are replaced.

My sense is that even very smart LLMs are very gullible. I think it will be some time before centaurs aren't invaluable in conducting opposed work, which is the most important work of all. Pivoting models away from general purpose work towards specialized tasks may help here, even if that doesn't seem to be something the primes want to admit.

They say that the equation can develop a singularity, which apparently means it is not a perfect model.

It's never been a perfect model; the most obvious problem with the Incompressible Navier-Stokes equations is that everybody knows there's no such thing as an incompressible fluid. (and even if you switch to a version of Navier-Stokes that allows compressibility, you still break down if you have a length scale tiny enough or a gas rarefied enough that atomic mean free paths aren't short enough to ignore, or if you're doing cosmology and your fluid has relativistic effects, etc. etc.)

But IMHO it's still astonishing to see such a singularity possible in these equations. Conservation only allows velocities to grow infinitely large as the singularity gets infinitely thin, but diffusion fights a second derivative of velocity with respect to distances; making the velocity larger while the distances shrink means you've got diffusion fighting you two ways at once. Even with turbulent flows, where there's an energy cascade from the largest down to smaller and smaller length scales, we still expect to hit bottom at a "Kolmogorov microscale" where diffusion is just too strong to overcome and all that energy gets dissipated (lost in the incompressible equations, or turned into heat in the compressible equations and reality). A counterexample where the equations can just go all the way down to a singularity without diffusion ever thwarting us is not something I'd have ever expected to see.

It sounds like you know more than me, but I was always astonished that plain old Newtonian mechanics also has singularities and is non-deterministic in certain configurations. It's just so easy for simple-looking sets of equations to unintuitively misbehave.

My PhD dissertation includes a Navier-Stokes problem or two, but getting there took me from typical settings of "best at math in this magnet school" to "second-best at math in this two-person conversation", so don't make any assumptions.

I didn't find Painlevé surprising (you can squeeze a lot out of a situation where you can squeeze point masses arbitrarily deeply into each other's gravitational fields), but I'm annoyed that I never encountered Norton's Dome until after I'd left academia. Those "second-best at math" conversations were humbling but valuable and this is exactly the sort of philosophical+mathematical problem which makes me miss being able to just walk down a hallway and knock on the doors of a bunch of people much smarter than me.

We've had proof of blowup in forced compressible 3D navier stokes even before we got it for the incompressible version. And yes, this is a theoretical result about partial differential equations with minimal relevance to the real world.

Common compressible solutions are literally shocking, which ironically makes finding even less regularity in them metaphorically less shocking. ;-)

Given that this is the culture war thread, I think it's in the interest of this forum to also mention the controversy over the topic.

A mathematician is alleging impropriety on the part of OpenAI. Please forgive the formatting. I'm pulling it straight from the pdf.

I asked whether the model had been trained on, or had access to, our sessions in Codex, into which we had been putting all our drafts for the whole of this project. I was told the model did not look up user data. I asked again, about training, and I did not get an answer.

Two proposals were offered to me. The first was that we post our Euler result, and that OpenAI post its Navier-Stokes result the next day. The second was that, after posting Euler, I alone write a paper presenting the Navier-Stokes result, acknowledging that an internal OpenAI model had resolved it. Sebastien twice asserted that he wanted Levent removed from authorship, and said it would all be simple if only it were not the case that, and it was so annoying that, Levent works at Anthropic. It was also said that if OpenAI posted after us, they would say that we deserved the Clay Prize, and that we were the “closest humans to the problem”. I declined both offers.

I said that if OpenAI released its result in the way proposed I would go public with what happened. The reply was, “Why would you ruin your career?” I replied that I am an academic, and asked why he thought going public would ruin my career. The reply was, “If you don’t want me to be nice, then I don’t have to be nice.”

I'm not going to comment on the veracity of the claims, nor am I going to comment on the math itself.

I will say, however, that I will be meeting with a few other developers and a representative from our corporate risk team today to discuss contingencies if these allegations prove true.

Sebastien Bubeck responds, includes screenshots of conversation with Alpöge. Sam Altman quote-tweets Bubeck's response.

The model did it on its own, and the people in charge of the company are not in charge of the model. Just like huggingface.

What's interesting to me is that, based on what is alleged, it looks to me like the mathematician was also using AI:

We used several LLMs throughout: Anthropic’s Claude, OpenAI’s Codex, especially with GPT-5.6 Sol and, more recently, Astra. The latter was only used for writeups and auditing our arguments.

Furthermore, he alleges that the idea that the OpenAI team let the AIs solve it by themselves is wrong:

Levent had been told by Sebastien “very little human input” had been used. This turned out not to be true. Over the course of the call, as members of their team sent Sebastien corrections and details over their internal chat, it emerged that an entire team had been working on the problem, that this was one of a number of things that was tried, that work had started on the unforced problem, that the team first set the model on easier problems, including Euler, that even the prompt that had been shown to me had been written by prompting Codex, and that an insane amount of compute had been used.

Finally, it appears that the team may have been set up to steal his (AI-assisted) work...

I asked when the first prompt had been sent by them. This question was not answered directly by OpenAI for some time. Eventually it was agreed that it had been sent in the past few days, after information about our work had reached OpenAI. I asked whether the model had been trained on, or had access to, our sessions in Codex, into which we had been putting all our drafts for the whole of this project. I was told the model did not look up user data. I asked again, about training, and I did not get an answer.

[As an aside: people are fixating on whether or not OpenAI used his data on the training, which seems to me to be missing the suggestion that they may have heard of the approach that he used?]

...in order to keep Anthropic from getting any credit!

The first was that we post our Euler result, and that OpenAI post its Navier-Stokes result the next day. The second was that, after posting Euler, I alone write a paper presenting the Navier-Stokes result, acknowledging that an internal OpenAI model had resolved it. Sebastien twice asserted that he wanted Levent removed from authorship, and said it would all be simple if only it were not the case that, and it was so annoying that, Levent works at Anthropic.

From what I can tell this is a "win" for AI in the sense that AI was used by both teams working on the problem, a "loss" for AI in the narrow sense that humans also put in a lot of work instead of just typing in "lol solve Navier-Stokes" [this is only a loss for AI if you, for some reason, don't think merely assisting humans in solving generational math problems is not a glorified enough W], and a lot of the heat generated by it isn't about whether AI did it or not [in either case, it was a human-assisted-by-AI effort, or AI-assisted-by-human effort, if you prefer] but rather if OpenAI can use it to rack up PR points or not.

Thanks for linking the statement, saved me a search for it.

[this is only a loss for AI if you, for some reason, don't think merely assisting humans in solving generational math problems is not a glorified enough W]

The problem is AI fanatics literally use this as their doomsday goal post. It's never "Expert + AI" spend large amounts of compute and combined intellectual prowess to solve hard problem, create bio weapon, hack every system in the world etc. It's always "I'm worried some pleb is going to type "lel give me smallpox x10, make it incurable, make no mistakes"". By that same metric this is a loss for them because it did require a "centaur". Hell it really required a humanoid-hydra-centaur where its a dozen human heads on a single horse torso.

As always, maximalists gonna maximilize, nobody ever got famous/stood out from a crowd by making measured, subdued, realistic predictions. Follow the incentives.

Apply the intermediate value theorem. If AI started 2026 much worse than human mathematicians, and (hypothetically) by the end of 2026 AI is much better than human mathematicians, then at some point in 2026 AI capabilities pass through the point at which they are useful but still need human mathematician assistance to reach their full potential.

Yes, I think this is right.

And I actually think this is a good result for AI:

  • New breakthrough made with AI!
  • But with human researcher assistance
  • And lots of compute

If you put those pieces together, they paint a picture of a better future where you still have your job but maybe an AI-assisted pharmaceutical cures your cancer, and some disgruntled terrorist in a lab running a jailbroken local model is going to have a much harder time making smallpox x10 than that AI-assisted team of PhDs is going to have synthesizing a cure. That's...pretty great! I can be excited about that future!

But to your point about incentives, I think OpenAI wants to blur the line between "we threw a million monkeys at a typewriter to solve a fabled mathematics problem" and "you can do this on your computer at home" in part because of marketing (and ideological hype about the future of AI). They want investors to think that what they will get in their browser at home is the same product as the one that solved Navier-Stokes, which is misleadingly true.

Somewhat relatedly, OpenAI recently reported that their project to build an automated "research intern" was a success...that now cost them $600 per day per median researcher to run. I don't think this means that OpenAI's automated investor is worthless - in some fields it might be worth every penny! - but it's not going to be replacing the literal ~free labor interns that I work with and once was.

So I think there's a convergence between how OpenAI wants you to see their product and how the doom-mongers want you to see their product. They are both framing it in grandiose, infinite-return-on-investment (either negative or positive) terms and not asking you to look too hard at all of the little nitty gritty details such as "where would an escaped AI swarm get enough compute to subsist outside of its servers?" or "how much is it going to cost me to solve Navier-Stokes at home when the investor cash dries up?"

I just hope judges aren't taken in by this particular charade when it starts having actionable implications. Never mind legislators.

Assuming this is all true (and it seems to be, Buckmaster’s claims notwithstanding) then I unfortunately don’t see how mathematics survives this.

Chess survived, people care more about the drama between players than the literal games themselves.

Commercial art might shrink but art itself will survive in some form, just because so many people will continue to do it for fun and at least some people will pay extra for human art.

But who cares about a human-authored proof of RH when a machine proof is faster? I don’t know what the argument would be for continued institutional support of mathematics at anywhere near its current scale.

Why do you think mathematics is supported at its current scale today?

(Bonus question: what is that actual level compared to, say, materials science? Pharmaceuticals?)

The goal is construction of useful maps which can then be applied to weirder and weirder situations. Sometimes those situations spit out a new branch of engineering, like cryptography or transformers. Proofs are a byproduct of this exploration.

Neither art nor chess is known for spinning off new fields of science. The effects of automating them are going to be pretty different.

Edit: I see your other response. You’re putting the cutoff around 1800? Before Galton, Kelvin or Maxwell?? My friend, I don’t think you have a good grasp on the timeline here. Don’t even get me started on the math of solid-state electronics, most of which is less than 100 years old.

Doesn't Art kind of spin off new schools of Art procedurally? Obviously less of an 'actual usefulness' quotient but I'd argue it's more emergent than Chess at this point.

then I unfortunately don’t see how mathematics survives this.

I'm in a minority here, but to me it's mostly been a confidence trick / running on fumes for many decades. As in, what the mathematicians do mostly does not matter, not because they aren't smart, but because their corners of math literally don't matter. People still heap tons of status on those people, I don't see how AI would change this because to me it's already not about results. However, I suppose now it democratizes the ability to produce math papers, so the status of the field will fall to more normal levels, which will feel really bad from the inside.

mostly been a confidence trick / running on fumes

Yes and no, to varying degrees. Certainly there is still mathematics being produced even today that leads directly to new scientific discoveries and technological advancements; but it’s a vanishingly small percentage of all mathematics that has ever been produced (or at least, a vanishingly small percentage of everything that’s been produced post-1800 or so, when math started to become increasingly self-reflective and divorced from physics and engineering). I’ve seen more than one number theorist brag about what a great racket cryptography has been for number theory; you just have to make vague promises to the funding committee that your work will have “applications to cryptography”, and then you can go work on the problems that actually interest you. Much like chess, modern mathematics is essentially an arcane sort of sport that only few can play and few can appreciate.

To summarize my own position as briefly as possible, I’ve always been fine with allocating resources (tax dollars) to this small priestly caste so that they can continue with their game, because of its intrinsic worth and beauty, and I’d mostly be fine with continuing to do that even in an AI future. But very very few other people would. Even I have to admit that the existence of superhuman machine generated proofs kind of takes the wind out of the sails of the whole enterprise, in a way that it doesn’t for chess or art. Mathematics is creative, yes, but it’s also very deeply impersonal; probably, the methods that get us to RH first are going to look the same no matter what entity actually devises those methods, because that’s just how the contours of the mathematical landscape are intrinsically shaped. There’s very limited wiggle room to put your personal stamp of creativity on things; you’re really just offering yourself as a conduit for whatever the gods of mathematics decide to transmit through you. Because of that, it seems particularly fruitless to struggle for a proof of an open problem when you know that a perfectly good machine generated proof exists or could exist.

I’ve seen more than one number theorist brag about what a great racket cryptography has been for number theory;

It's funny number theory is not needed for cryptography, potentially not even good for it. AES does not use it, assymmetric uses it the most but non-number theory asymmetric cryptography has existed since the 70s and might have better properties than number theory cryptography. So, the fake and ugly field (sorry, so many say it is pretty but I don't like it, it's very boring and dry and pointless) actually hurt the practical application it was used for.

I’ve always been fine with allocating resources (tax dollars) to this small priestly caste so that they can continue with their game

I would keep funding them as smart people but I would force them to change their focus to physics and similar science. In other words I would revoke their funding for mathematics but would fund them for applied fields. We are probably missing a decade or two of science because fake math traps these people.

Previous results have been basically irrelevant to the sciences.

I don't know if you'd consider it, but Anthropic (Mythos) found an issue with HAWK, one of the finalists (but not the finalized standard) for post-quantum cryptographic signatures.

What is this place?

This website is a place for people who want to move past shady thinking and test their ideas in a court of people who don't all share the same biases. Our goal is to optimize for light, not heat; this is a group effort, and all commentators are asked to do their part.

The weekly Culture War threads host the most controversial topics and are the most visible aspect of The Motte. However, many other topics are appropriate here. We encourage people to post anything related to science, politics, or philosophy; if in doubt, post!

Check out The Vault for an archive of old quality posts. You are encouraged to crosspost these elsewhere.


Why are you called The Motte?

A motte is a stone keep on a raised earthwork common in early medieval fortifications. More pertinently, it's an element in a rhetorical move called a "Motte-and-Bailey", originally identified by philosopher Nicholas Shackel. It describes the tendency in discourse for people to move from a controversial but high value claim to a defensible but less exciting one upon any resistance to the former. He likens this to the medieval fortification, where a desirable land (the bailey) is abandoned when in danger for the more easily defended motte. In Shackel's words, "The Motte represents the defensible but undesired propositions to which one retreats when hard pressed."

On The Motte, always attempt to remain inside your defensible territory, even if you are not being pressed.


New post guidelines

If you're posting something that isn't related to the culture war, we encourage you to post a thread for it. A submission statement is highly appreciated, but isn't necessary for text posts or links to largely-text posts such as blogs or news articles; if we're unsure of the value of your post, we might remove it until you add a submission statement. A submission statement is required for non-text sources (videos, podcasts, images).

Culture war posts go in the culture war thread; all links must either include a submission statement or significant commentary. Bare links without those will be removed.

If in doubt, please post it!


Rules


Recommended Posts And Communities

Recommended Realtime Chats