FunSearch
LLM plus automated evaluator evolves programs; found the largest cap sets in two decades and better bin-packing heuristics.
Evolutionary loop where an LLM proposes program mutations and an evaluator filters hallucinations. Early documented case of an LLM-driven search improving a published open maths result, with human-readable code as output. Direct ancestor of AlphaEvolve.
- Date
- Thursday, 14 December 2023
- Lab
- Google DeepMind
- Kind
- paper
- Access
- research preview
Published in Nature (blog dated 2023-12-14).
Sources
This record was checked against its sources on 6 October 2026. How we check