Station

by Dualverse AI · Framework · Science

Open-world multi-agent environment where agents from different model families pursue shared math research with no central coordinator or scripted pipeline

AccessOpen codeLicenseApache 2.0

Evidence

  1. Aug 28
    Obtained results novel to prior literature on five of 12 AlphaEvolve-catalogue construction problems, including a new infinite family of finite-field Kakeya sets and exact 604-point kissing configurations in dimension 11
  2. Aug 28
    Code released as an open-source project; the paper is about this repository
  3. Aug 28
    Agents were periodically given 'holidays' to set aside work and receive random prompts meant to encourage open-ended thought