Gary Marcus slams OpenAI's giant math drop: 'would never pass peer review'
GaryMarcus · x · 2026-10-08
Gary Marcus published a rapid critique of OpenAI's release of new mathematical results from an internal frontier model, arguing the real news is what wasn't disclosed. He notes the report omits the procedure, architecture, whether proofs were one-shot with Lean verification or iterative, failure rates, and training details — making generalizability outside math impossible to assess. He says the system could be a real step toward AGI or just a clever use of Lean and synthetic data in a verifiable domain, and laments that social media debate has become uncritical cheering. The post references complementary remarks from Terence Tao.
Related event: Gary Marcus Slams OpenAI Math Proof Report for Lack of Detail(3 posts)→
More from AGI Musings
- QwQiao likens recent math anomalies to Mercury's orbit: math isn't broken, our knowledge is — CatAstro_Piyush · 2026-10-08
- PwC UK Revenue Falls for First Time in Two Decades; Ex-Consultant Blames Leadership — DavidLinthicum · 2026-10-08
- AI math debate: beyond human-centrism, hoping intelligence keeps expanding its own convex hull — _onionesque · 2026-10-08
- Your Chatbot's First Person Is a UI Choice, Not a Self, Argues AI Voice Debate Post — gerardsans · 2026-10-08
- With AI Running the Information Age, Do 'Idea People' Finally Get Their Shot? — Musenik · 2026-10-08
- "Would math go sterile without humans?" AI debate questions human-centrism in the language of nature — lu_sichu · 2026-10-08