AI & agents
Google's multi-agent system lands a run of mathematics breakthroughs
- Published
- Source
- Google Research Blog
Summary
Google Research reported that Antigravity's /teamwork multi-agent system, paired with Gemini 4 models, produced results on long-standing open problems in mathematics. It proved information theory's 2013 Courtade–Kumar conjecture in full generality, yielding 50 million lines of Lean 4 proof code, and machine-checked an end-to-end formalization of the Weil conjectures, settled by Pierre Deligne in 1974, in about 1.8 million lines of Lean. The post also claims the first formalization of the Birkhoff conjecture, open since 1927, and opens applications giving 10,000 academic researchers six months of free access to its highest usage tier. All results are as reported by Google and its partner institutions, and confirming that each formalized theorem statement matches the original problem remains the responsibility of human mathematicians.
Why it matters for our work
It shows a research division of labor—agents carrying the detailed proof work while people define problems and make final checks—operating even on top-tier open problems. With these multi-agent research tools now opening to outside researchers, the scope of working with AI is expanding beyond code and documents to proofs themselves.
Translated from the Korean original. Summaries may be translated and edited. Commentary reflects our perspective; forecasts remain the source’s views.