Evolutionary Selection

A population of candidate programs is produced by an ensemble of differently configured LLM samplers in parallel; a separate evaluator scores them; the fittest survive and are mutated into the next generation, gradient-free search in solution space.

Properties

family: Generative Exploration
maturity: emerging
diagram: evolution
whyMultiAgent: Requires a population of candidate solutions plus an evaluator that is structurally separate from the generators, so no variant grades itself. Selection pressure external to any individual generator explores more broadly than one model's beam search, and diversity is drawn from multiple differently configured LLM samplers (a fast model for high-volume generation, a stronger model for occasional higher-quality suggestions) sampling from an island-structured population database, not from temperature samples of a single model alone.
mast: [Cat 3]
useCases: [Prompt optimization][Design exploration][Algorithm discovery]
source: AlphaEvolve · arXiv:2506.13131

Provenance

Source: https://arxiv.org/abs/2506.13131

Evolutionary Selection

Properties

family: Generative Exploration
maturity: emerging
diagram: evolution
whyMultiAgent: Requires a population of candidate solutions plus an evaluator that is structurally separate from the generators, so no variant grades itself. Selection pressure external to any individual generator explores more broadly than one model's beam search, and diversity is drawn from multiple differently configured LLM samplers (a fast model for high-volume generation, a stronger model for occasional higher-quality suggestions) sampling from an island-structured population database, not from temperature samples of a single model alone.
mast: [Cat 3]
useCases: [Prompt optimization][Design exploration][Algorithm discovery]
source: AlphaEvolve · arXiv:2506.13131

Provenance

Source: https://arxiv.org/abs/2506.13131