Reproduce “Evolution through Large Models” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Joel Lehman, Jonathan Gordon, Shawn Jain, Kamal Ndousse, Cathy Yeh, Kenneth O. Stanley
Evolution Through Large Models (ELM) integrates large language models with genetic programming to generate computer programs, employing LLMs as intelligent mutation operators. This approach achieves higher mutation success rates than traditional methods and enables an invention pipeline to create diverse, functional solutions in domains lacking initial training data.
Key insights
- Large language models can serve as effective, intelligent mutation operators, significantly increasing the success rate of code modifications over traditional random genetic programming.
- Bootstrapping an "invention pipeline" with ELM allows for the creation of domain-specific code generators, even without initial training data, enabling conditional design based on environmental cues.
- Fine-tuning the diff model with successful mutations discovered during evolution further enhances its ability to explore the solution space and discover diverse designs.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.