undefined | Better HN

0 pointsAlexCoventry2mo ago0 comments

No, there's no training going on, here, as far as I can tell. E.g., they use GPT-5 as their base model. Also, AFAICT from a quick skim/search there's no mention of loss functions or derivatives, FWIW.

0 comments

alextheparrot2mo ago

The derivative being a grad(ient) student sampling scaffolds against evals + qualitative observations: most prompt-based llm papers

j / k navigate · click thread line to collapse

0 pointsAlexCoventry2mo ago0 comments

No, there's no training going on, here, as far as I can tell. E.g., they use GPT-5 as their base model. Also, AFAICT from a quick skim/search there's no mention of loss functions or derivatives, FWIW.

0 comments

alextheparrot2mo ago

The derivative being a grad(ient) student sampling scaffolds against evals + qualitative observations: most prompt-based llm papers

j / k navigate · click thread line to collapse