OpenEvolve: Teaching LLMs to Discover Algorithms Through Evolution

(algorithmicsuperintelligence.ai)

53 points | by codelion 3 days ago ago

9 comments

  • jasonjmcghee 3 days ago ago

    It doesn't mention it in the article, but guessing this is based on / inspired by AlphaEvolve?

    Though I'm not sure the public can access AlphaEvolve yet.

    (https://arxiv.org/abs/2506.13131)

    • gerdesj 3 days ago ago

      If AlphaEvolve is: "a quality-diversity search framework for algorithm discovery" then maybe.

      At the moment I'm mildly skeptical and uncertain of whether to twist or stick.

    • jasonb05 3 days ago ago

      Agreed, not mentioned.

      Nevertheless, I see a link to github for the OpenEvolve project [1] that in turn states:

      > Open-source implementation of AlphaEvolve

      [1] https://github.com/algorithmicsuperintelligence/openevolve

  • DoctorOetker 3 days ago ago

    Very interesting that the LLM weights are co-evolved and reasoning skills improve!

    • viraptor 3 days ago ago

      What do you mean by this? I can't find anything there about modifying the used LLMs and the hosted ones wouldn't be possible to change. Do I misunderstand the convolved part you mentioned?

      • DoctorOetker 3 days ago ago

        you are correct, on re-reading they only evolved the prompts ...

  • N_Lens 3 days ago ago

    Some cool optimisations here: MAP elites, island models to prevent premature convergence & fast rejection of bad candidates.

    What's particularly interesting is the meta level insight: The system discovered scipy.optimize.SLSQP for circle packing - a completely different algorithmic paradigm than it started with. It's genuinely discovering new approaches, not just parameter-tuning.

  • quantbagel 3 days ago ago

    Sakana.ai improved on this by honing in on sample efficiency iirc with shinkaevolve (which is open source and not an ai slop project)