3 ms·
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
by AlexCoventry 9mo ago
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.
- alextheparrot 9mo agoThe derivative being a grad(ient) student sampling scaffolds against evals + qualitative observations: most prompt-based llm papers