Gary Marcus on whether frontier LLMs can solve open math problems without symbolic harnesses
GaryMarcus · x · 2026-10-07
- A user asked Gary Marcus to clarify: given his accounting of LLM capabilities, could a frontier LLM solve an open math problem without a Lean / code / symbolic harness?
- Marcus called it "an actually intelligent question" and answered below, implying he believes LLMs alone cannot.
- A representative exchange in the ongoing debate over raw LLM capability vs tool-augmented reasoning.
More from AGI Musings
- AI Agents Formalize Major Proof in Lean in a Week for ~$135K, Producing 600K Lines — geoffreyirving · 2026-10-07
- Ex-OpenAI/Anthropic researcher: automated AI researcher by 2027-28, humans far out of the loop — rohanpaul_ai · 2026-10-07
- Researcher slams Wired's 'ridiculous' coverage of a landmark math breakthrough — Afinetheorem · 2026-10-07
- "It hasn't even started": AI is just clearing brute-forceable math problems, argues researcher — eigenron · 2026-10-07
- AI law scholar on the Malfoid debate: some things are best left untouched — technollama · 2026-10-07
- 80% of the past 3 years' math progress happened in a single day — aaron_lou · 2026-10-07