Google's Cogentic multi-agent system produces novel results on 5 open research problems
Dr_Singularity · x · 2026-10-06
Google Research unveiled Cogentic, a multi-agent harness for automated proof discovery on open research problems, built on Gemini (Gemini 3.1 Pro and an early Gemini 4 Argon).
Key design:
- Instead of a single answer, a population of independent AI provers explores different proof directions in parallel
- Outputs undergo adversarial verification; only confirmed intermediate results enter a persistent "verified ledger" that later rounds build on
With this setup, Cogentic produced novel results on 5 open problems spanning online learning, auction theory, and mechanism design — including O(d) regret online inverse optimization and auto-bidding auction efficiency — each independently verified by domain experts and written up in companion papers.
More from coding & agent
- Teknium catalogs 305 open, hackable devices your AI agent can control and live in — Teknium · 2026-10-06
- ESR: LLMs make closed-source binaries 'transparent as glass', ending secrecy as we know it — nitarshan · 2026-10-06
- Crawler + GPT agent scanned 5,137 jobs for $6.53, found 3 low-competition niches — Arindam_1729 · 2026-10-06
- xAI Cookbook adds 4 Grok SDK examples: screenshot-to-React, X sentiment tracker, AI ad generator — tetsuoai · 2026-10-06
- Power user runs a swarm of personal agents — poke, instinct, muse, dot, Grok bot, pickle and more — sebkrier · 2026-10-06
- omni-macos: A Fully Local Semantic Finder for Apple Silicon, Now Daily-Driver Good — JinaAI_ · 2026-10-06