Gary Marcus warns Anthropic bulls: Lean-verified math wins won't generalize

GaryMarcus · x · 2026-11-22

Quoting Liv Boeree's musing that the Riemann hypothesis now influences a trillion-dollar AI company's valuation, Gary Marcus issues an investor warning: anyone who doesn't grasp the difference between verifiable and open-ended problems — and assumes Anthropic's Riemann progress, resting on 17 million lines of Lean code, generalizes to everyday busy problems without such formal scaffolding — "deserves what they get" if they lose money on Anthropic.

Core claim: AI breakthroughs in formally verifiable domains (math, code) don't automatically transfer to open-ended real-world tasks lacking verification signals.

Original post →

More from AGI Musings

AGI Musings channel →