LIVE

Researchers Use Gemini 4 and Multi-Agent Systems to Solve Decades-Old Scientific Open Problems

🤖 AI Models 2 min read 👁 8

Researchers successfully deployed Gemini 4 and multi-agent harnesses like /teamwork and Stellar Colosseum to independently solve long-standing open problems across computer science, quantum computing, and mathematics.

Researchers Use Gemini 4 and Multi-Agent Systems to Solve Decades-Old Scientific Open Problems🤖AI Models

Photo: Hacker News Front Page

Key points

  • Gemini 4 and multi-agent harnesses solved decades-old open problems in math and computer science.
  • Multi-agent systems independently proved and Lean-verified near-optimal sample bounds for adaptive data analysis.
  • New deterministic algorithms improved matrix multiplication, integer multiplication, and optimization runtimes.

A team of researchers has demonstrated significant breakthroughs by utilizing advanced AI models and multi-agent frameworks to tackle complex, unresolved problems in engineering and science. By posing open challenges directly to Gemini 4 alongside multi-agent systems such as /teamwork and Stellar Colosseum, the investigators received end-to-end algorithms, mathematical proofs, and formalizations verified in Lean 4.

Among the notable achievements, a multi-agent harness running on Gemini 4 Argon proved near-optimal sample bounds for adaptive data analysis, effectively resolving a statistical query gap that had persisted since 2014. In quantum computing, a newly developed construction settled a decade-old question regarding n-bit Boolean functions, while other efforts yielded deterministic algorithms for Boolean matrix multiplication and major progress on Tarjan's 1985 Deque Conjecture.

The projects also advanced optimization, randomized linear algebra, and data structures. For instance, campaigns using these AI systems produced a deterministic algorithm for online inverse linear optimization that reduces runtime from exponential to polynomial time, alongside a new $O(n \log n)$ integer multiplication algorithm whose C implementation matches the GNU Multiple Precision Arithmetic Library bit-for-bit up to a billion bits.

What to watch next: Look for further adoption of multi-agent AI harnesses in academic research as scientists increasingly rely on automated verification tools like Lean 4 to validate machine-generated proofs.

Source Hacker News Front Page — research.google/blog/advancing-science-engineering-with-antigravitys-te… ↗

Contact the newsroom

Got a tip, a proposal or a partnership question? Write to us and we will reply shortly.