
A Researcher Says GPT-5.6 Pro Cracked a 30-Year-Old Math Conjecture After He Told It to 'Do a Breakthrough'
Dmitry Rybin says OpenAI's model produced a counterexample that disproves the long-open Dinitz–Garg–Goemans conjecture — using a four-prompt chat that mostly amounted to 'do a breakthrough' and 'now finish it.' The catch: the result hasn't been independently verified.
Read →






















