This is part of why I love and am mourning the slow decline of master craftspeople. The collaborations I undertake are in person, in the material world. Matthew Crawford's piece this weekend - a sort of love letter to the mechanics in his life - is a great illustration of how enriching this mode of human cooperative excellence can be. The shift into more and more abstract, mental, online work has eroded what can be a very wholesome universe of relational and productive activity.
I think they’ll always be a human niche for that. It’s the same reason why people buy those handcrafted soap bars for £16.99 when you could get an off-the-shelf one at Poundland - people respect the craftsmanship when everything else is cheap and mass produced.
And if the internet itself becomes the symbolic place for the cheap, mass produced and tasteless… maybe the real world - and real, tangible inventorship and craftsmanship within it - becomes the place to find the valuable, curated and beautiful. And that’s a nice seed for a resurgence :)
I agree that the demand isn't going away, but it takes massive resources and time to develop real expertise in many fields, and that's increasingly non-viable for many kinds of craft. I hope we can save as many from extinction as possible, because once they are gone, they often cannot be recovered.
This is another way in which it feels like AI amplifies the problems of the internet, rather than bringing something totally new. I already felt like I was working hurriedly in the shadows, because of the possibility that some of the billions of people on the internet were doing the same thing as me.
I sometimes make web apps with calculators for games, and I was always worried someone else would get the scoop and be the first one to make some specific calculator before I finished mine. But AI made it worse - last year I was working on a calculator for a brand new game system, and someone else vibe coded the same thing just a few days after release. It's not analogous to a millenium problem because the math was simple and there was no need to copy my work, but there is still an extreme amount of stress these days with any work that has a first-mover advantage.
The last couple of years I’ve been using Confer AI (by the people who make Signal) for any ideas that I really don’t want entering training data somehow. & my project on that platform is called Wallfacer.
I think we can extend the Wallfacer theme to damn near everything online until we get major reforms; the safest way to interact with social media and their algorithms is to not be on them at all. What felt like paranoia to those born pre-9/11 has largely become reality (surveillance, AI, algorithms, etc). Like the books, the only thing we can ever be sure of is our own thoughts and keeping those relatively intact will be a huge advantage moving forward while simultaneously not retreating into a black domain either.
All of the major social media platforms seem to have become largely broadcast-based media now. I left FB because it was truly awful, and my Instagram feed is now completely reels. None of my personal contacts seem to use any of this stuff anymore, and I don't blame them because all of them decided to prioritize influencers and reels in their algorithms over the people we actually know. I browse Reddit, but do not have an account. I am become Wallfacer.
Reminds me of the history of Renaissance math where people that came up with new solutions (like the solution to cubic equations, I think) would have public math battles where they would show they could give the answer without revealing their methods.
I wonder if this will become more common. Zero Knowledge Proofs of proofs!
"Liu Cixin’s The Three-Body Problem trilogy, which is the best sci-fi released in China, or frankly anywhere else, in the past two decades" Glad you feel that way, because I loved it, but I think it's the only sci-fi written in the past two decades I've read. (Well, Seveneves I guess.) Just as I was glad to learn that you were a Dan Simmons fan, because I'd come across him by accident, because he wrote a sci-fi fantasy based on the Iliad, and I love Homer. Have you written, or considered writing, an article on good recent-ish sci-fi?
I would love this too. The Expanse just pips The Three Body Problem for me. I've tried a lot of other stuff and find it wanting, even the prize winners. Erik's recommendations would be nice. Maybe they're already posted somewhere?
I’d appreciate that list too. I’ve been going further and further back in time — recently read the Worm Ouroboros, which is one of the earliest “modern” fantasies and quite worth the read.
Little addendum to the Jacobian conjecture section: I was initially confused upon clicking your link, because it goes to a September 2026 math paper. The details about research having been accidentally shared are hidden in the "Supplemental AI statement of the third author" at page 160. Levent Alpöge responded in a conciliatory tone: https://xcancel.com/__alpoge__/status/2097698960966049876
It's a shame that there's no direct evidence either way for whether this happened -- Wayback Machine didn't archive the accidental upload and we cannot see into the training set of Fable. Makes me wonder if it would be technically feasible to certify what did or did not enter the training set of an LLM... These things are not intuitive, cf. the introduction of watermarking, which *can* be done without affecting output distributions.
Yup, it's tough to track down (I should have indicated it was in the AI statement at the bottom) but I think the author has high enough credibility to take what it's saying seriously. I hadn't seen that reply though. Saying that the human author deserves credit and also just that "I thought it didn't appear" is uh, an interesting response!
This is an impactful read; at the risk of hyperbole, what is being described may, if it hasn’t already, impact the engine of our evolutionary trajectory. Homo sapiens hunted by Homo faber in the “Dark Forest” as the “Wallfacers” hunker down, like our diminutive, primordial, mammalian ancestors did, while waiting for the meteor which brought down the dinosaurs. Thank you for an incisive, important and timely essay.
Cue "math is math" meme. Ultimately, it's good to be scooped for scientific progress, it gives you a next platform and a new outlook. (Yes, even when "My name in Dnepropetrovsk is cursed".)
Are you aware of Stricker & friends' new platform, Dark Forest Operating System? https://dfos.com/ - interested to see if people move into it in numbers or not. I like it a lot, but we're so used to network effects, I wonder if it'll manage to maintain itself. I really hope it does, since it's one of the few real alternatives I've seen.
Oof. I'm an artist and this bugs me. Oh sure, there are thiefs out in the world but at least you could put something out, make a bit of money before you find bad copies of whatever it was you are doing out in the world. For awhile people thought well maybe if I do hardcopy of whatever idea because AI is online, yes, that should work. No.
Because now Anthropic and others are plundering old media, scraping them and destroying the old media. And you can't get fast enough to get faster than the machine. All we've done is hand the keys to the thieves. Tinfoil hat time: that this was the plan all along, to kill creativity.
I loved reading this, but wanted to present a different view. I think private and verifiable compute hardware is also following its own "Moore's law" silently. Eventually, efforts such as Venice AI or Darkbloom (that offer provably private computation), for example, would command a premium over non-private AI and it would get priced in. Academics would continue talking about ideas publicly and present unfinished work, but the presented work would be of a much grander scale that is not easily solvable by AI - when none of us are peasants, we can all compose symphonies. I might be missing something here or being super-naive, but I do not think culture ever becomes a dark forest since any developments like AI would get priced in, so to speak. That is, if you are presenting an idea that you are worried might leak into AI and lead to an immediate discovery - it is better to just use a private AI and solve that problem yourself, and focus your energy on presenting ideas about more real world frictions and implementation. Even super abstracted domains such as pure mathematics would focus on different ways of understanding a result, or presenting proofs of results more elegantly (to fellow humans) rather than generating theorems and proofs. That is again something that AIs might get good at and open up the next frontier of abstraction that needs good understanding.
“Even super abstracted domains such as pure mathematics would focus on different ways of understanding a result, or presenting proofs of results more elegantly (to fellow humans) rather than generating theorems and proofs.”
This is the crux of Tao’s ICM keynote [1] starting around 33:00. Paraphrasing him: proofs used to be scarce, and when an important result was proved, many people would drop what they were doing to read it; in the age of AI, proofs are becoming abundant, and mathematicians are becoming more selective about what’s worth ‘digesting’ – analogous to food scarcity / abundance.
What strikes me is that a sci-fi novel about alien civilisations has become the best vocabulary we have for what's happening to human culture in 2026. Sci-fi didn't predict this exactly. It handed us the words before we needed them. I'm torn on the conclusion, though. As a marketer, I've watched ideas grow because they were shared early and imperfectly. If everyone becomes a wallfacer, I wonder if we lose the very messiness that made culture worth hunting in the first place.
Here’s another consequence of the dark forest: we may lose the ability to observe how knowledge develops.
Imagine two researchers arriving at the same discovery. One spent years testing, failing, revising, and learning from others. The other generated the solution in minutes.
If we evaluate only the final product, their contributions may look identical. But what happens when the next problem has no known solution?
We may be creating a world where outcomes become increasingly visible while the processes that produce human expertise become increasingly invisible.
How will we recognize expertise if we stop seeing how it’s built?
I think the concern is legitimate. But there's an important distinction worth making: I don't use AI as a substitute for my own research. I use it as a tool to explore hypotheses that would be much harder to formalize and test on my own.
That leaves us with a more difficult question: what matters more for human culture, protecting the origins of individual ideas, or preserving the free exchange of them?
If everyone becomes a wallfacer, we risk losing something essential to discovery itself: the collision of different ways of thinking.
This is part of why I love and am mourning the slow decline of master craftspeople. The collaborations I undertake are in person, in the material world. Matthew Crawford's piece this weekend - a sort of love letter to the mechanics in his life - is a great illustration of how enriching this mode of human cooperative excellence can be. The shift into more and more abstract, mental, online work has eroded what can be a very wholesome universe of relational and productive activity.
I think they’ll always be a human niche for that. It’s the same reason why people buy those handcrafted soap bars for £16.99 when you could get an off-the-shelf one at Poundland - people respect the craftsmanship when everything else is cheap and mass produced.
And if the internet itself becomes the symbolic place for the cheap, mass produced and tasteless… maybe the real world - and real, tangible inventorship and craftsmanship within it - becomes the place to find the valuable, curated and beautiful. And that’s a nice seed for a resurgence :)
I agree that the demand isn't going away, but it takes massive resources and time to develop real expertise in many fields, and that's increasingly non-viable for many kinds of craft. I hope we can save as many from extinction as possible, because once they are gone, they often cannot be recovered.
Sage, I’m intrigued by the Matthew Crawford piece you mentioned. Could you share a link with us?
Here you go: https://mcrawford.substack.com/p/ode-to-a-master-mechanic
This is another way in which it feels like AI amplifies the problems of the internet, rather than bringing something totally new. I already felt like I was working hurriedly in the shadows, because of the possibility that some of the billions of people on the internet were doing the same thing as me.
I sometimes make web apps with calculators for games, and I was always worried someone else would get the scoop and be the first one to make some specific calculator before I finished mine. But AI made it worse - last year I was working on a calculator for a brand new game system, and someone else vibe coded the same thing just a few days after release. It's not analogous to a millenium problem because the math was simple and there was no need to copy my work, but there is still an extreme amount of stress these days with any work that has a first-mover advantage.
The last couple of years I’ve been using Confer AI (by the people who make Signal) for any ideas that I really don’t want entering training data somehow. & my project on that platform is called Wallfacer.
prescient!
I think we can extend the Wallfacer theme to damn near everything online until we get major reforms; the safest way to interact with social media and their algorithms is to not be on them at all. What felt like paranoia to those born pre-9/11 has largely become reality (surveillance, AI, algorithms, etc). Like the books, the only thing we can ever be sure of is our own thoughts and keeping those relatively intact will be a huge advantage moving forward while simultaneously not retreating into a black domain either.
All of the major social media platforms seem to have become largely broadcast-based media now. I left FB because it was truly awful, and my Instagram feed is now completely reels. None of my personal contacts seem to use any of this stuff anymore, and I don't blame them because all of them decided to prioritize influencers and reels in their algorithms over the people we actually know. I browse Reddit, but do not have an account. I am become Wallfacer.
Reminds me of the history of Renaissance math where people that came up with new solutions (like the solution to cubic equations, I think) would have public math battles where they would show they could give the answer without revealing their methods.
I wonder if this will become more common. Zero Knowledge Proofs of proofs!
Perhaps universities may want to start negotiating accounts for researchers in fields with high marketing potential to get ZDR (Zero Data Retention)
OT.
"Liu Cixin’s The Three-Body Problem trilogy, which is the best sci-fi released in China, or frankly anywhere else, in the past two decades" Glad you feel that way, because I loved it, but I think it's the only sci-fi written in the past two decades I've read. (Well, Seveneves I guess.) Just as I was glad to learn that you were a Dan Simmons fan, because I'd come across him by accident, because he wrote a sci-fi fantasy based on the Iliad, and I love Homer. Have you written, or considered writing, an article on good recent-ish sci-fi?
I would love this too. The Expanse just pips The Three Body Problem for me. I've tried a lot of other stuff and find it wanting, even the prize winners. Erik's recommendations would be nice. Maybe they're already posted somewhere?
I’d appreciate that list too. I’ve been going further and further back in time — recently read the Worm Ouroboros, which is one of the earliest “modern” fantasies and quite worth the read.
Little addendum to the Jacobian conjecture section: I was initially confused upon clicking your link, because it goes to a September 2026 math paper. The details about research having been accidentally shared are hidden in the "Supplemental AI statement of the third author" at page 160. Levent Alpöge responded in a conciliatory tone: https://xcancel.com/__alpoge__/status/2097698960966049876
It's a shame that there's no direct evidence either way for whether this happened -- Wayback Machine didn't archive the accidental upload and we cannot see into the training set of Fable. Makes me wonder if it would be technically feasible to certify what did or did not enter the training set of an LLM... These things are not intuitive, cf. the introduction of watermarking, which *can* be done without affecting output distributions.
Yup, it's tough to track down (I should have indicated it was in the AI statement at the bottom) but I think the author has high enough credibility to take what it's saying seriously. I hadn't seen that reply though. Saying that the human author deserves credit and also just that "I thought it didn't appear" is uh, an interesting response!
This is an impactful read; at the risk of hyperbole, what is being described may, if it hasn’t already, impact the engine of our evolutionary trajectory. Homo sapiens hunted by Homo faber in the “Dark Forest” as the “Wallfacers” hunker down, like our diminutive, primordial, mammalian ancestors did, while waiting for the meteor which brought down the dinosaurs. Thank you for an incisive, important and timely essay.
Cue "math is math" meme. Ultimately, it's good to be scooped for scientific progress, it gives you a next platform and a new outlook. (Yes, even when "My name in Dnepropetrovsk is cursed".)
Fretting being 'scooped' in the Dark Forest is still Just Vanity.
Are you aware of Stricker & friends' new platform, Dark Forest Operating System? https://dfos.com/ - interested to see if people move into it in numbers or not. I like it a lot, but we're so used to network effects, I wonder if it'll manage to maintain itself. I really hope it does, since it's one of the few real alternatives I've seen.
Oof. I'm an artist and this bugs me. Oh sure, there are thiefs out in the world but at least you could put something out, make a bit of money before you find bad copies of whatever it was you are doing out in the world. For awhile people thought well maybe if I do hardcopy of whatever idea because AI is online, yes, that should work. No.
Because now Anthropic and others are plundering old media, scraping them and destroying the old media. And you can't get fast enough to get faster than the machine. All we've done is hand the keys to the thieves. Tinfoil hat time: that this was the plan all along, to kill creativity.
I loved reading this, but wanted to present a different view. I think private and verifiable compute hardware is also following its own "Moore's law" silently. Eventually, efforts such as Venice AI or Darkbloom (that offer provably private computation), for example, would command a premium over non-private AI and it would get priced in. Academics would continue talking about ideas publicly and present unfinished work, but the presented work would be of a much grander scale that is not easily solvable by AI - when none of us are peasants, we can all compose symphonies. I might be missing something here or being super-naive, but I do not think culture ever becomes a dark forest since any developments like AI would get priced in, so to speak. That is, if you are presenting an idea that you are worried might leak into AI and lead to an immediate discovery - it is better to just use a private AI and solve that problem yourself, and focus your energy on presenting ideas about more real world frictions and implementation. Even super abstracted domains such as pure mathematics would focus on different ways of understanding a result, or presenting proofs of results more elegantly (to fellow humans) rather than generating theorems and proofs. That is again something that AIs might get good at and open up the next frontier of abstraction that needs good understanding.
“Even super abstracted domains such as pure mathematics would focus on different ways of understanding a result, or presenting proofs of results more elegantly (to fellow humans) rather than generating theorems and proofs.”
This is the crux of Tao’s ICM keynote [1] starting around 33:00. Paraphrasing him: proofs used to be scarce, and when an important result was proved, many people would drop what they were doing to read it; in the age of AI, proofs are becoming abundant, and mathematicians are becoming more selective about what’s worth ‘digesting’ – analogous to food scarcity / abundance.
[1] https://youtu.be/M0--ZH1lOzg
What strikes me is that a sci-fi novel about alien civilisations has become the best vocabulary we have for what's happening to human culture in 2026. Sci-fi didn't predict this exactly. It handed us the words before we needed them. I'm torn on the conclusion, though. As a marketer, I've watched ideas grow because they were shared early and imperfectly. If everyone becomes a wallfacer, I wonder if we lose the very messiness that made culture worth hunting in the first place.
Here’s another consequence of the dark forest: we may lose the ability to observe how knowledge develops.
Imagine two researchers arriving at the same discovery. One spent years testing, failing, revising, and learning from others. The other generated the solution in minutes.
If we evaluate only the final product, their contributions may look identical. But what happens when the next problem has no known solution?
We may be creating a world where outcomes become increasingly visible while the processes that produce human expertise become increasingly invisible.
How will we recognize expertise if we stop seeing how it’s built?
I think the concern is legitimate. But there's an important distinction worth making: I don't use AI as a substitute for my own research. I use it as a tool to explore hypotheses that would be much harder to formalize and test on my own.
That leaves us with a more difficult question: what matters more for human culture, protecting the origins of individual ideas, or preserving the free exchange of them?
If everyone becomes a wallfacer, we risk losing something essential to discovery itself: the collision of different ways of thinking.