Andy Buckleyagbuckley@mastodon.social
Aug 8, 2026, 8:13 AMInteresting on #OpenAI maths results: https://www.scientificamerican.com/article/openais-latest-math-breakthroughs-commit-research-misconduct-experts-say/
Not just the sloppy failures to cite work that the LLMs will 100% have ingested, or the company's prioritisation of hype: their SOP, and right that they get called out for it in research.
But I'm more interested in the observation that the solutions aren't really interesting: "elementary", just at superhuman scale. Like the 4-colour problem, the solution is obtained but we've not really learned anything. Which was the actual point
