logoalt Hacker News

vascotoday at 11:24 AM0 repliesview on HN

Most data in the training set of most reasoning models is crap I guess.