logoalt Hacker News

whimsicalismyesterday at 9:20 PM0 repliesview on HN

is this not essentially what ARC-AGI-3 is? i agree that in-context/continual learning is somewhere the models are still mostly weak at