logoalt Hacker News

twoodfin • today at 6:24 PM • 0 replies • view on HN

Right: The humans are training the models on much more than the particulars of the tasks being trained.