Implement ideas from Google's AlphaEvolve

[nileshtrivedi]

AlphaEvolve paper

Here is an open-source replication of AlphaEvolve: https://github.com/codelion/openevolve

I do not think it is a significantly different approach from DGM. They do evolutionary search over functions/programs whereas DGM evolves the coding agent. It should be possible to replicate AlphaEvolve within DGM's current architecture with some refactoring.

[SHA888]

Hi @nileshtrivedi,

Thank you for opening this issue and sharing the AlphaEvolve paper and implementation. I find the idea of integrating evolutionary search over functions/programs into DGM very interesting.

@jennyzzt @ShengranHu, as the maintainers of DGM, I'd be interested in working on this enhancement. The concept of combining DGM's agent evolution with AlphaEvolve's approach to program evolution could lead to some powerful synergies.

Would you be open to me taking this on? I'd be happy to:

1. Create a detailed proposal for the integration
2. Outline a development plan with clear milestones
3. Implement the changes while maintaining DGM's core architecture

Looking forward to your thoughts!

Best regards,

[BradKML]

You guys might want to start collaborating on OpenEvolve side of the equation algorithmicsuperintelligence/openevolve#197 algorithmicsuperintelligence/openevolve#53

The problem with ANY repo is the decision to either (a) kept the code updated but not break out of the scrope of the related paper OR (b) turn itself into a more comprehensive library but do things that might be different from the paper

SakanaAI scientist is open to Discussion for v1 but not v2, specifically to be more industrial and paper-compliant https://github.com/SakanaAI/AI-Scientist/discussions