It's crazy that people accept "the model logits represent its cognition, to second guess it you'd need a better model" as proof that LLM samplers are optimal. Guys you can calculate policy entropy,…

Posters: jd_pressman. jd_pressman

4 likes recorded by the source corpus.