GPT-5.6 Pro falsifies a 30-year graph theory conjecture as Sol+Codex solves 6 Erdős problems
新智元 · wechat · 2026-07-24
GPT-5.6 Pro and Sol are used to break a 30-year graph theory conjecture and solve six open problems
A new Chinese article claims GPT-5.6 Pro produced a counterexample to the Dinitz-Garg-Goemans conjecture in graph theory, turning a 30-year-open cost-version problem into a false statement. The counterexample uses three terminals with costs structured so that any integral solution must pay at least 60, while a fractional solution reaches 58.
The same piece also says Shouqiao Wang used GPT-5.6 Sol + Codex to solve 6 of 13 attempted open Erdős problems in 5 days, including one that ran for 32 hours. His workflow emphasized:
- selecting problems mathematicians were already actively discussing,
- precisely redefining what counts as a solution,
- and using adversarial agents to attack each candidate proof before finalizing it.
The article frames these cases as evidence that AI is becoming a serious research partner in mathematics, and speculates about the implications for future human-only prizes like the Fields Medal.
Related event: GPT-5.6 Pro Reportedly Refutes 30-Year Graph Theory Conjecture(5 posts)→
More from AGI Musings
- Why So Many AI Researchers Think the Machines Could Kill Everyone — connoraxiotes · 2026-09-11
- jjvincent invokes Terence Tao: ceding exploration to AI means ceding human agency — jjvincent · 2026-09-11
- OpenRouter agents now out-consume humans as AI usage arrives in three waves — AccBalanced · 2026-09-11
- If AI teleports us to solutions, how do underlying fields develop? — jjvincent · 2026-09-11
- Op-ed: the ">10% extinction" narrative is liability evasion — AI is just software, and the vendor is the defendant — gerardsans · 2026-09-11
- AI researchers just saw the power of a single resignation — and still claim there's nothing they can do — birchlse · 2026-09-11