Colin Snyder's Star Fleet runs 20 parallel GPT-5.6 agents against open Erdős problems
Independent researcher Colin Snyder's Star Fleet orchestrates up to 20 'starship' agents in parallel — each a GPT-5.6 instance on a dedicated 60-vCPU server with SAT/SMT solvers and a Lean 4 premise search — and posts kernel-checked Lean 4 proofs for 13 Erdős problems. It is a self-published, non-peer-reviewed effort, and at least two proposed solutions have been withdrawn. The signal is the harness, not any single solve: formal verification paired with parallel agent orchestration is emerging as a credible way to search proof space at scale.
Source: starfleetmath.com ↗
Star Fleet is an AI system that solves the world's hardest open mathematics problems using Lean 4.
Colin Snyder, Star Fleet Math
Why this matters
- → Formal verification paired with parallel agent orchestration scales proof search to hard open problems.
- → Demonstrates feasibility of AI systems tackling Erdős-class mathematics without peer review gatekeeping.
- → Lean 4 kernel validation replaces traditional publication as the arbiter of correctness.
Parallel agents, Lean proofs