Skip to content

Comment on Anthropic researchers: Thinking longer makes models dumber

Comments

The problem is that they're calling it "reasoning" when these are nothing more than repeated calls to the same probabilistic guessing machine with different stacked prompts.

That's not reasoning. We've all seen people go off on wild guessing sprees, too.

AboutSource Built by g1lg1l

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