Interesting: https://x.com/KLieret/status/2029219763423986030

would be neat to add this to OpenHands Index

I was going to say Zora's recent position paper had a similar finding, and I just realized that Kilian is talking about the same paper! I wonder what would happen if we model this objective somehow during RL - basically among the patches that solved the issue correctly, give higher reward to ones that solve it in fewest lines/hunks possible. In general as well, I think a lot of people complain about code bloat with agents.