Gary Marcus says AI may solve some math tasks, but still can’t build new theories
GaryMarcus · x · 2026-08-04
Gary Marcus quotes a Terence Tao lecture on AI and math, highlighting two ideas: not every proof is useful, and mathematics is not uniform. He argues that AI may help with some kinds of mathematical tasks, but there is still no evidence it can build new theories.
The retweet also frames this as part of a broader critique of AI hype: the original commenter says Marcus and Ed Zitron are among the best people online for calling out what they see as BS in the Western AI world, warning against people who hype AI in the same way they once hyped Bitcoin and crypto.
Related event: Terence Tao: AI Solves Problems but Lacks New Theories(3 posts)→
More from AGI Musings
- AI labor debate turns to a future where compute determines who wins — Justin_Halford_ · 2026-08-04
- Gary Marcus asks for real counterarguments to OpenAI and Anthropic's math claim — GaryMarcus · 2026-08-04
- A repost says the internet used to feel like wandering, not performing — moonsandhues · 2026-08-04
- Travis Oliphant says AI needs oversight, accountability, and data sovereignty — teoliphant · 2026-08-04
- Reddit debate asks whether stochastic LLMs can really reach AGI in 1–5 years — Reardon-0101 · 2026-08-04
- Speeding up intelligence by 10× could turn a quantitative gain into a qualitative one — metaviv · 2026-08-04