hugetim·7 か月前·議論Contrary to what the leaderboard lists as the human score, their technical paper implies a human baseline of ~48%.
hugetim·7 か月前·議論Am I crazy to think these models have actually surpassed human performance on ARC 2? https://www.lesswrong.com/posts/DX3EmhmwZjTYp9PBf/ai-perform...