Cogentic organizes AI as a virtual research team, utilizing an orchestrator to manage provers and a pair of adversarial verifiers to scrutinize every draft. The agents track progress through a shared ledger of verified lemmas and a record of failed attempts to avoid repetition. A process advisor updates instructions based on these logs without providing mathematical opinions, allowing the system to tackle complex academic tasks without human hints.
Wednesday, October 21, 2026
Story
Google Gemini Solves 5 Unsolved Math Problems via Cogentic System
AI19 days ago
A specialized network of AI agents now coordinates task-sharing and verification to handle long-form research problems that typically defeat single prompts. Google Research recently published a paper describing Cogentic, a system running on Gemini that discovered new proofs for five unsolved problems in theoretical computer science. The discoveries, which spanned online learning, auction theory, and mechanism design, were confirmed by domain experts and required between 100 and 1,000 Gemini model calls per problem.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.