I don’t necessarily think this is the only way to do decomp with LLMs I suppose.

Show Post

LLM documents program behavior -> other LLM (even e.g. a subagent) writes new code to spec seems pretty clearly clean room? I mean there’s no absolute guarantee, I suppose, but it doesn’t seem like LLMs make this any different provided the original code is not part of the training data.

Show Post

we’re just making stuff up ig

Show Post

like holy fuck nigga

Show Post

Elon is just sticking with the proud historical tradition of fascists desperately wanting to be killed in violent car accidents

Show Post