Hacker News new | past | comments | ask | show | jobs | submit
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