FunSearch

An evolutionary program synthesisprogram synthesis, n.: Getting a computer to write a program automatically from a description of what the program should do — a logical specification, a few input–output examples or a request in plain language. method from Google DeepMind in which a pretrained language modellarge language model, n.: A computer program trained on enormous amounts of text to predict the words that come next. At vast scale, that simple skill lets it chat, answer questions, translate, summarize and write code. serves as the mutation operator. The user supplies an evaluation function and a program skeleton; FunSearch keeps a database of candidate programs, split into islands to preserve diversity, repeatedly prompts the model with a few high-scoring programs and asks for a better one, runs and scores each new program automatically, and adds successful ones back to the database. Automatic scoring filters out incorrect or hallucinated programs, so reported results can be checked, and since the product is a short program, a person can often read it and see why it works. FunSearch found cap sets in eight dimensions larger than any previously known and heuristics for online bin packing that beat standard ones. Its successor, AlphaEvolveAlphaEvolve, prop. n.: A Google DeepMind system that combines AI language models with an evolution-like search to invent and improve computer programs, finding new results in mathematics and computing. (2025), evolves larger codebases with stronger models.