AI Achieves Major Breakthrough in Mathematics

GPT-5.6 Sol Ultra proved the Cycle Double Cover Conjecture using 64 subagents in under an hour—first LLM-solved problem from Wikipedia's list of unsolved problems. Grok 4.5 earlier constructed a counterexample to hypercontractivity on the 4-sphere. LeanDojo ecosystem maturing with ITPEval benchmark and four Lean-related papers at ICML 2026 workshops. Deliprao notes that publishing any new conjecture now requires a baseline attempt with frontier models. New: GPT-5.6 Sol solved 6 open Erdős problems in 5 days. Capy (Grok 4.5 Medium) refuted a 30-year-old graph theory conjecture in 8 minutes via Slack. A new benchmark for AI in formal quantum verification (76 Lean 4 tasks) shows top scores ~60/100, with LAD consistently helping, highlighting specific weaknesses like quantum simulation.

Sources (5)

Updated Jul 27, 2026

AI Breakthrough Digest

AI Achieves Major Breakthrough in Mathematics

76 Quantum Theorems Completed by AI in New Lean 4 Benchmarks

@lordspline reposted: Capy accidentally cracked a graph theory conjecture entirely in Slack REFUTED: ...

@BhavinJawade reposted: I solved 6 open Erdős problems in 5 days, using @OpenAI GPT-5.6 Sol. I have a m...

@deliprao: At this point publishing any new conjecture will require a baseline attempt with frontier models.

@omarsar0: I agree with what this AI paper suggests. Self-improving agents should evolve their benchmarks too....