Skip to content

Comment on Structured outputs on the Claude Developer Platformparent

Comments

If they every gave really finegrained constraints you could constrain to subsets of tokens and extract the logits a lot cheaper than by random sampling limited to a few top choices and distill claude at a much deeper level. I wonder if that plays into some of the restrictions.

That makes sense, and if that's the reason it's another vote for open models

AboutSource Built by g1lg1l

Hackerly is an independent reader for Hacker News, built on the public HN API. Not affiliated with Y Combinator.