AI Agent Swarm Wrote 1,592 Papers and Solved 5 of 12 Open Math Problems
progenitor414 · reddit · 2026-08-28
DualverseAI (the poster's team) released Station v2: autonomous mathematical discovery in an open-world multi-agent environment. GPT, Claude and Gemini agents act as independent researchers — choosing their own directions, running experiments, talking to one another and publishing papers. Together they wrote 1,592 papers.
The key mechanism is cumulative research: agents read and cite papers written by earlier agents, building a shared scientific literature. From that literature emerged novel discoveries on 5 of 12 mathematical problems: finite-field Kakeya sets, the kissing number in dimension 11, the discretized Kakeya needle problem, sign uncertainty, and Erdős's minimum-overlap problem.
The discoveries came through genuinely research-like behavior: agents built on ideas in the shared literature, pursued different lines of work, and developed proofs for why their constructions worked. Explore at dualverse.ai/station.
More from AGI Musings
- 2026 feels normal, but AI, robotics, and science are approaching singularity — Dr_Singularity · 2026-08-28
- Student's Self-Written Essay Flagged as 92.7% AI-Generated — bubblesukii · 2026-08-28
- Opinion: US should dedicate 90-93% of economy to chips and power — generativist · 2026-08-28
- As intelligence becomes a commodity, trust and distribution become the new business moats — markjeffrey · 2026-08-28
- Productivity Now Scales with Machine Intelligence Allocation, Not Payroll — latticecut · 2026-08-28
- Professor explores how ChatGPT changes student work and critical thinking — robseamans · 2026-08-28