Evolutionary search over algorithms and proofs: Gemini models propose changes to a program, automated evaluators score them, and the strongest candidates seed the next generation (DeepMind). In this corpus it first appears as Terence Tao’s collaborator on an open Erdős problem; it then discovers an activation function that triples ReLU, the point at which the search loop turns on the components of the models that drive it, and establishes new lower bounds for five Ramsey numbers.
AISystem
AlphaEvolve
DeepMind's Gemini-powered evolutionary coding agent, which pairs model-proposed program changes with automated evaluators to discover algorithms, bounds and training components.
Sources
Attribution and reporting linked by this record; links do not imply independent verification.
- Welcome to December 13, 2025 (theinnermostloop.substack.com)
Development timeline
Newest reporting dates first. Reporting dates may differ from event dates.
Knowledge graphopen in browser →
Developed by
Cite or correct this record
Suggest a correction — opens a GitHub draft for you to review and submit.