i think people skeptical of AI's impact on scientific and technological research badly underestimate the *country* part of "a country of geniuses in a datacenter". u really don't actually need each individual instance to be incredibly smart or have fully integrated non-spiky intelligence. just *being able to fan out* over a huge number of problems and hypotheses and put non-trivial focus and effort into each one will absolutely yield a huge quantity of results. humans are currently massively limited by the amount of intellectual labor we have, we spend lots of time focusing on picking which things to further explore because we really don't have the time or people to make mistakes! GLP-1s were just sitting in a cabinet for 30 fucking years. what happens when you have next gen models even start to systematically comb through *existing* clinical trials?

@littmath

(What I wrote is screenshotted below.) pic.twitter.com/MjYEetV1UE

there are something like 400k people globally who can participate in mathematical research, including phd students and across all subdomains / stats / etc. maybe 500k material science researchers. 600k cancer researchers. sure, that's a lot of people. but even now devs spin up hundreds of millions of codex + claude code sessions monthly. what does it look like to have a hundred million researchers trying out every possible open problem in cancer research? everything that's ever been mentioned and not fully investigated or thought through or tested ? what does it look like to pluck all the low hanging fruit at once?