Gary Marcus: AI Struggles to Make Conceptual Leaps in Fuzzier Domains
GaryMarcus · x · 2026-08-10
Responding to Owain Evans' question about whether AI is making advances in harder-to-verify domains like science and humanities, Gary Marcus expressed skepticism.
He noted that while AI excels in verifiable areas like math and hacking, it struggles in domains lacking clear verifiability. He argued that AI advances are often overhyped, and if there were indeed major conceptual leaps in fuzzier domains, the public would likely have already heard about them.
Related event: Scholars Debate AI's Limits in Ambiguous Fields(3 posts)→
More from AGI Musings
- Replit CEO on the 'Self-Driving Company' and Agents as the Brain — amasad · 2026-08-11
- AI infrastructure investment hits 2.8% of US GDP, surpassing the railroad boom — QuintinPope5 · 2026-08-11
- Open Source Catching Up Fast: Is Training Frontier Models Still Worth Millions? — 0xsachi · 2026-08-11
- Prediction: Anthropic Will Split Chat and Agentic Models — natanielruizg · 2026-08-11
- Revisiting 'Stochastic Parrots': Calling an Airplane Flying Aluminum Isn't Useful — dbasch · 2026-08-11
- AI Eats SaaS Margins: Incumbents Trapped by Legacy Cost Structures — signulll · 2026-08-11