It seems that I can’t go for long without hearing news about OpenAI or Anthropic. Two weeks ago, an X post by an Anthropic ex-employee went viral for calling for a slowdown, estimating a 10% chance of humanity’s extinction.

Yawn. I am so tired of this. I have long ago concluded that these doomdsay predictions are just marketing.¹ I don’t have a problem with marketing in general, but it’s a certain kind that preys on people’s secret desires to see the world end. Even worse, it has given platforms to charlatans such as Eli Yudkowsky. I look forward to the bursting of this bubble and the freeing of people’s mental spaces for more fruitful topics.

Around the same time, OpenAI announced that their latest model had found a solution to the Navier-Stokes Millenium Prize Problem. This is an impressive feat, but again, I think the math community is overreacting (e.g. mathandai.org), primarily because there is no understanding in LLMs.² You still need a mathematician to verify that a solution is correct, LLM-assisted or not. Certainly, the math community will need to adapt, given the dramatic increase in volume of unverified ~code~ proofs, but I do not see the field in any danger.

Where do we go from here? I think that professionals in software, math, and related (finance?) fields should embrace LLMs as a high-leverage tool. It will work very well in reducing the tedium most of the time, but sometimes, it will be faster to actually dig and think through the problem with one’s own mind.

In spite of claims to the contrary, I believe that progress in LLM scaling is slowing down naturally due to natural physical limitations, not any threat of extinction. We will eventually need to find a different approach to artificial intelligence to get past this barrier.