Feedback & Future Directions: What should A-evolve learn next?

[HanqingLu]

Please share your feedbacks in this issue.

[rolson24]

Try running A-evolve on SlopCodeBench. I think the meta-harness algorithm might be best for it because I feel like it could benefit from some deterministic verifiers and maybe multiple agent stages (PR-review agent, or a critic agent along with the implementer).