Matt Bornstein
@BornsteinMatt
This is the right take. Machine learning has *always* been a black box. Darpa even ran a grand challenge once to try and solve this (didn't work). AI models are fundamentally empirical beasts, and need to be measured like a natural phenomenon. See also: why do airplanes stay up
roon@tszzl · Sep 2i agree with this and think that CoT is at best an epiphenomenon of current training methods. it will break (in the future, not now). safety community too often centers thought terminating taboos like “neuralese” and “training on interp”
Open quoted post → 1 8