logoalt Hacker News

ChadMoranyesterday at 9:28 PM0 repliesview on HN

My hunch is that much of the model tuning to make it more effective has been for its internal thinking prose. That leaks out into its external writing prose.