Po-Shen Loh on Tao's blog: AI will create more jobs than humans, forcing AI progress to slow

Short run, yeah, maybe. Long run those jobs get done by AI too. It's like chimpanzees looking at human society and thinking about how many chimps it must take to run it.

Honestly, we don't know. You might be right but right now it's more likely you're not, and no country is going to bet its research community and decades of world leadership on that hunch.

We're not building an AI society though. We're building tools to serve human society, that's the whole point.

Math is special because proofs are verifiable, safe, and need no physical experiments. You can grind parallel search with RLVR. Try that with surgery and see how far it gets.

The chimp analogy falls apart because chimps can't communicate with us. Better one: look at all the engineers, doctors, scientists and writers, and ask how many mediocre politicians it takes to keep it running. Somehow it keeps running.

Here's the original post: https://poshenloh.com/posts/20260919-math-ai The Tao version is just way more readable for some reason.

Been thinking about this a while: maybe grasping complex phenomena needs an external system, not a human knowledge system. Newton distilled falling objects, planets, collisions and artillery into F=ma, and even that doesn't hold everywhere, it just holds usefully.

Why do we need any job at all? Read Bullshit Jobs by Graeber. Plenty of made-up positions existed long before AI, so AI isn't changing much here.

Jobs are how a society decides who gets what. That's the eternal question, and it doesn't go away just because a model can solve a proof.

This reads like a student in algebra class asking when they'll ever use this in real life.

The Hugging Face hack is doing a lot of work in that argument and I don't buy it. Big labs want bombastic demos before an IPO, and a barely-constrained-superweapon story is exactly the regulatory capture pitch. That escape hatch looks trained in, not emergent.

Then how come OpenAI keeps hiding how far the HuggingFace incident went, and we learn about new victims from third parties every week? 4D chess, or they just want others to find them?

Pure math has been overvalued for decades. Now that we don't need humans for it, time for pure mathematicians to pack it in.

Who says we don't need humans for it? You think non-mathematicians are going to sit there steering agents toward solving pure math problems?

People talk like there's a finite amount of math and then the party ends. Every problem you solve opens ten more, it's a fractal. AI is just a better flashlight, and sooner or later some incident gets governments to switch off the GPUs entirely.

Probably infinite math problems but only finitely many interesting ones.

Maybe this is good. More people get to do world-class math with these tools, and pure math is one of the most elitist, hostile fields around, with people whose self-esteem is welded to being able to do what others can't. That said, AI is overhyped here. It's a super violin played by master violinists, not an autonomous orchestra.

Is pure math really that elitist and hostile? Sounds hyperbolic.