Explore open math problems with a multi-agent system
A paper on autonomous mathematical discovery in an open-world multi-agent environment.
What they did
The authors built the Station, an open-world environment where AI agents from different model families share one research goal. There is no central coordinator and no scripted pipeline. Agents pick their own research directions, run experiments, collaborate, and add to a shared literature. They were tested on 12 AlphaEvolve-catalogue construction problems plus two case studies.
How it went
Results were novel relative to prior literature on five problems. These were a finite-field Kakeya family, 604-point kissing configurations in dimension 11, records for discretized Kakeya needle and sign uncertainty, and a better Erdős minimum-overlap bound. Agents also found Book Ramsey families and wrote explanatory theorems.
Worth knowing
The authors released the source code, raw agent dialogues, proofs and verification code, so you can inspect how each result emerged.
Labels
A paper on autonomous mathematical discovery in an open-world multi-agent environment.
Agents pick their own research directions, run experiments, collaborate, and add to a shared literature.
The authors built the Station, an open-world environment where AI agents from different model families share one research goal.
They were tested on 12 AlphaEvolve-catalogue construction problems plus two case studies.
The “Material” label failed its quality check and is not used for selection.
Details
- Agents
- custom-built or unnamed agent
- Source
- Hacker News
- Source date
- Aug 28, 2026
- Accounts
- 1
- In registry since
- Oct 12, 2026