I’ve even heard professors warn that reading the wrong book can “cause brain…

I’ve even heard professors warn that reading the wrong book can “cause brain damage” (c)

I’ve been catching myself pretty often starting to notice that the motivations of AI research teams have started to look less like science and more like shipping a new feature that just grabbed one more benchmark. And all of this gets sold by some “peppy” CEO yelling that we’ll all see AGI soon. Feels like something’s off here, and everything’s turned into fashion. In a lot of top labs, in my subjective feel, the spirit of science has started to evaporate. And then I stumbled on an interesting note (https://sugaku.net/content/understanding-the-cultural-divide-between-mathematics-and-ai/) by a mathematician who visited a congress (a meetup) organized specifically for mathematicians; but to me the precedent itself matters — AI really is heavily affecting pretty much all areas of knowledge and science. Fun times!

A bunch of things there that are mentally and “model-wise” close to me and reflect the essence of my work:
- Terence Tao (one of the strongest mathematicians of our time, definitely read about him) wrote a note back in 2007, “Be Patient,” pointing out the fact that it often takes years from the first day you start working on a problem to its solution. Roughly the same way I see it in truly hard products and decisions I’ve arrived at.
- In math, like in physics, discoveries are often named after the people who worked on them. That underlines the importance of the human factor — same as I think about the importance of a team of people in a startup. A lot of folks underestimate this, but finding genuinely awesome people is extremely hard, and that’s a huge part of success. And the most interesting thing is that in math papers, the team is considered equal: in 75% of cases the last names in the title go in alphabetical order, while in other fields (AI) they go in order of significance of contribution.
- Mathematicians often search for the essence of things, whereas AI is just trying to invent something new or impress. Roughly the same way I think about everything — since childhood I was into physics because I wanted to understand how things work, the universe and all that, because it’s insanely cool, like a new world opens up: when an equation isn’t just a bunch of letters and equals signs, it’s what’s all around you.
- And a good question landed at the end: even if a new AI system can prove the Riemann hypothesis, what if the proof is so complex that a human just can’t understand it. Would there be any point in such a proof? What if nobody understands fundamental discoveries at all anymore? Because that hasn’t happened in human history yet. And will we be able to actually use such discoveries for real?