Karpathy mentioned entropy collapse in LLMs, where they stop doing interestin...

Karpathy mentioned entropy collapse in LLMs, where they stop doing interesting things because they just don’t have interesting things in their training data

tempting to think they just need randomness, but i think that would make it worse

they need memory & experience

i tend to prefer hiring people that are on their second career bc i think it solves some of the same entropy collapse that happens in LLMs

entropy is the opposite of randomness, it’s just a new source of dense signal

meh, by “entropy” i mean “low entropy”, i.e. the opposite of what i said. just vibe with it