Posts on X are crediting GPT-5.6 with solving a decades-old Erdős problem about the asymptotic density of 2-primitive sets, based on posts amplified by Sebastien Bubeck and Andrew Curran, who added that it allegedly happened with "no harness" in the standard ChatGPT interface. So far, the public evidence is limited to those X posts and the linked post they point to, not independent reporting or a paper readers can inspect outside X.