Gary Marcus: Riemann was human-AI teamwork on a bespoke solver, not AGI
GaryMarcus · x · 2026-11-22
Gary Marcus pushes back on the widespread misreading of the Riemann hypothesis result: AI did not solve it alone. A top mathematician with AI expertise drove the entire process in collaboration with AI.
Key points:
- The proof relied on a special-purpose symbolic AI solver with 16 million lines of code, built over years by hundreds of people — of little value for open-ended problems in biology, business, or robotics.
- He likens it to Deep Blue beating Kasparov in 1997: impressive brute force in a narrow domain that didn't generalize, consistent with his 2024 prediction that verifiable methods help math more than other fields.
- It does not mean AGI has arrived, and the market agrees — $ANT barely moved after hours.
- His bottom line: current AI remains insufficiently trustworthy for many applications and not profitable enough to justify the massive investment.
Related event: Gary Marcus: Riemann Proof Was Human-Symbolic AI Teamwork, Not Solo LLM(8 posts)→
More from AGI Musings
- Lenny's late-2026 tech job market report: PMs winning, engineering stalled — bibryam · 2026-11-22
- PauseAI founder on BBC: race for superintelligence risks human extinction — birchlse · 2026-11-22
- Blogger predicts uncontacted tribes' pristine minds may become a scarce resource in the AI age — jjvincent · 2026-11-22
- AI is rebuilding the firm: Anthropic hits ~$1.5T with 6,000 staff vs Meta's 67,000 — inductionheads · 2026-11-22
- Toby Ord debate: it's unclear whether humans or AI are the jagged intelligence — tobyordoxford · 2026-11-22
- LLMs inherit knowledge like humans do—we're all born into an informational niche — dioscuri · 2026-11-22