AI provers on a Bittensor subnet close six decades-old Erdős problems in ten days
markjeffrey · x · 2026-09-19
conjecture.io announced that SN66, a Bittensor subnet that pays bounties for solving open math problems, has had its miners resolve six long-standing open problems in ten days — roughly 249 combined years unsolved:
- Erdős Problem 96 (66 years open): disproved
- Erdős Problem 108 (55 years open): disproved, via graphs with arbitrarily high chromatic number whose four-cycle-free subgraphs need at most six colors
- Erdős Problem 196 (49 years open): disproved
- Erdős Problem 14 (34 years open): both parts resolved
- Erdős Problem 653 (29 years open): proved
- Green's Problem 51 (16 years open): near-half density case solved
Every result is machine-checked in Lean — formal proofs published with code that anyone can rerun, not just "we believe." Problems Erdős posed that top mathematicians couldn't close in half a century were picked off one by one by a swarm of miners running AI provers, in public.
More from Research
- JEPA-Anything: one predictive framework spanning vision, biology, weather and more — rbhar90 · 2026-09-19
- Tsinghua's C2C Lets LLMs Skip Text and Merge KV-Caches Directly, 2.5x Faster with +14.2% Accuracy — anselm · 2026-09-19
- Summer School Debunks SOTA Autonomous Driving Tricks for Failing to Generalize — ftm_guney · 2026-09-19
- Professor Zhiting Hu Links New Jev Model to Her Three-Year-Old Discriminative Generalist Work ALIGN — ZhitingHu · 2026-09-19
- Tsinghua paper: RL fine-tuning prunes exploration, letting base LLMs beat RL models at high pass@k — burny_tech · 2026-09-19
- Async RL is just horizontal scaling with sticky routing, one dev argues — dosco · 2026-09-19