Definition
Plain language
A Google DeepMind system that works alongside research mathematicians, drafting proofs, critiquing its own attempts, and helping a human spot the way through.
As stated in the literature
A hierarchical multi-agent research-math system that refines the question with the user, runs parallel literature/computation/proof-search workstreams, and uses adversarial reviewer agents; scored ~48% on FrontierMath Tier 4 and contributed to resolving open problems including a Kourovka Notebook conjecture.
Why it matters: It matters because it shifts AI from answering textbook questions to actively helping push the frontier of open mathematical research alongside humans.
For example, a mathematician stuck on an unsolved group-theory problem could hand it to the system, which drafts candidate proofs, has its own reviewer agents poke holes in them, and surfaces a promising line of attack.
Heard on the show
“It's called "AI Co-Mathematician: Accelerating Mathematicians with Agentic AI," and we're recording one day later.”Episode 029 — Why Forty-Eight Percent on FrontierMath Isn't the Real Story in DeepMind's New Math Paper