@0.314

birds eye view of what’s happening and what’s about to happen in ai x hard science: ai is making really impressive progress in pure math (quasi riemann), theoretical cs (eth), theoretical physics (yang-mills), and foundational biochem (the alphafolds of the world). what it appears to be doing is a smart brute force search of the entire corpus that our civilization has built so far, and piecing together existing results. it’s clearing out a massive historic backlog. what i think is going to happen soon (2 yrs?) is we are going to finish all the low hanging fruit before hitting a wall. for 2 reasons: 1) we still don’t have einstein-level intelligence (aka the einstein test): einstein did not merely build on top of existing theorems and experimental results. he challenged the very axioms, eg the speed of light is constant and space-time is flexible. it’s unclear to me if the current trajectory of ai will get us there. maybe it still ends up being a search problem, but the set of possible axioms is literally infinite whereas our existing corpus is finite so this seems like a much harder search problem. 2) we are advancing theory much faster than empirical science as we are going to be bottlenecked by the speed and cost of experiments in the physical world. take bio for example, yes ai is going to help find targets and molecules more efficiently, but we are always going to have to run clinical trials where the binding constraints are lack of participants, regulations, and time itself for longitudinal analysis. in physics we already have a half century of beautiful but non-falsifiable theory and this will get worse. tl;dr: the wall is coming in ai x science. after that, progress is gated by two things no amount of compute fixes: einstein-level intelligence that questions the axioms themselves, and a physical world that runs at its own speed.