A collaborative effort combining AI-driven insights with human researchers yielded a 20-page arXiv paper, Exist­ence of the Core in Approval-Based Committee Elections, detailing the harmonic entropy mechanism and its proof of core existence.

The work shows that a counterexample never exists: the core cannot be empty, implying an absolutely fair committee must exist in any scenario.

GPT-6 Astra, partnering with three researchers, solved a long-standing major breakthrough math problem from the FrontierMath benchmark, marking a historic moment for AI-assisted mathematics.

Astra introduced a novel harmonic-entropy-based voting framework with a polynomial-time local-search algorithm that finds core-compliant committees without exhaustive enumeration.

Dominik Peters described the result as elegant and potentially generalizable, signaling AI-driven discovery of mathematical principles beyond problem-solving.

Key collaborators include Patrick Becker, Matthias Greger, and Dominik Peters; the arXiv link is 2609.11912 and a GitHub repository is provided.

The broader implication is a shift in AI evaluation, viewing AI as a joint researcher rather than merely a tool for solving problems.

Epoch AI has introduced a new status label, “Human + AI,” honoring significant AI-led contribution while recognizing human prompting and collaboration.

The paper asserts that the local optimum under the harmonic-entropy objective guarantees core membership, enabling a practical polynomial-time algorithm to compute an absolutely fair committee.

The core question concerns the existence of a non-empty core in approval-based committee elections, addressing a challenge to fairness that persisted since 2017.