Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

ARC is a noble endeavour but mistakes visual/spatial reasoning for reasoning and thus fails.


No, I don't think it does. I think that the ideas in a system that could solve this type of problem would be highly generalisable to other tasks.


thankfully we can just wait and see here. concretely, I predict time from first multimodal llm that can reliably read a chessboard and analogue clock without finetuning (obviously not reasoning) until ARC is solved is <4 months




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: