Gary Marcus: I praised AlphaProof's hybrid approach, not o1's pure inference compute
GaryMarcus · x · 2026-11-21
Gary Marcus re-shares his own post to push back on critics, accusing them of equivocating about "these systems." He says his comments on the Riemann breakthrough targeted o1's pure inference-time compute, and that he praised neurosymbolic hybrids like AlphaProof and AlphaGeometry back in 2024 — the approach he sees the Riemann solution as evolving from.
More from AGI Musings
- Kevin Hartnett's new book "The Proof in the Code" charts Lean's history transforming math and AI — ipeirotis · 2026-11-21
- Jack Clark: AI speedup in scientific progress will define 2027 and beyond — jackclarkSF · 2026-11-21
- Yann LeCun: AI's biggest risk is power concentration in proprietary systems, not runaway models — ylecun · 2026-11-21
- "If we can't contain AI, why no more incidents?" developer challenges doom narrative — basedjensen · 2026-11-21
- Reddit essay: AI doom is the 21st-century rerun of humanity's age-old power paranoia — Possible-Insect3752 · 2026-11-21
- FT profiles DeepMind's Henry Shevlin on machine consciousness and Nagel's bat — dioscuri · 2026-11-21