Landmark discovery claim · Mathematics · Computer science
DeepMind's Aletheia produces autonomous papers and open-problem solutions
An autonomously generated arithmetic-geometry paper, human-AI research papers, and autonomous solutions or partial solutions across a large collection of open Erdős problems.
Summary
Google DeepMind introduced Aletheia, a Gemini Deep Think-based research agent for professional mathematics. Its reported outputs include an autonomously generated paper on eigenweights, human-AI papers on independent-set bounds and other topics, and a semi-autonomous sweep of 700 Erdős problems that produced several new solutions after expert filtering and review.
AI role
Aletheia used Gemini Deep Think with iterative generation, verification, revision, literature search, and tool use to construct and check research-level mathematical arguments.
Narrative role
Aletheia marks a transition from isolated model-assisted proofs to a repeatable research-agent workflow spanning literature navigation, candidate generation, verification, revision, expert review, and publication-oriented mathematical output.
Caveat
The evidence is a Google-authored announcement and preprint portfolio rather than peer-reviewed publication of the overall system. DeepMind explicitly classifies none of the reported results as a major advance or landmark breakthrough, and the Erdős cases vary in novelty and difficulty.