logoalt Hacker News

htrp • today at 3:55 PM • 0 replies • view on HN

Too bad pretraining hasn't been relevant for model performance since 2025