year: 2022
paper: https://arxiv.org/pdf/2206.08896
website: https://youtu.be/T08wc4xD3KA
code:
connections: evolution, LLM, Kenneth O. Stanley, OpenAI, genetic algorithms
- Argues that genetic programming is fundamentally enhanced by asking LLms to suggest diffs
- Generated hundreds of thousands of diverse “Sodarace” robots coded in python even though this game is absent from LLM training.
This paper pursues the insight that large language models (LLMs) trained to generate code can vastly improve the effectiveness of mutation operators applied to programs in genetic programming (GP). Because such LLMs benefit from training data that includes sequential changes and modifications, they can approximate likely changes that humans would make.