Gary Marcus says modern math AIs are hybrids, not pure LLMs
GaryMarcus · x · 2026-07-21
Gary Marcus says the latest attacks on him often miss the technical details, and highlights a key distinction in modern math AI.
- He agrees that hybrid systems have moved the field forward.
- The quoted argument says raw LLMs still suffer from compounding hallucinations on hard math.
- The proposed recipe for progress is a stack of SFT on reasoning traces, RL with process rewards, inference-time search such as MCTS, and executable tools.
- The broader point: today’s stronger math systems are hybrids, not pure transformers.
Related event: Gary Marcus: Pure LLMs Aren't Good at Math, Credit Goes to Hybrid Systems(5 posts)→
More from AGI Musings
- Investor: model version numbers are a mirage, only datacenter margins are real — StewartalsopIII · 2026-10-27
- WSJ Op-Ed: For Anthropic Share Offering Investors, a 10% Extinction Chance May Be a 90% Portfolio Win — davidmanheim · 2026-10-27
- WSJ: for Anthropic investors, a 10% extinction chance may still be a 90% portfolio win — peterwildeford · 2026-10-27
- Simile, the AI human-simulation startup, raises $300M+ at $2B valuation with 85%+ behavior prediction — joon_s_pk · 2026-10-27
- AI images are moldy bread, AI text is spoiled milk, researcher argues — _vaishnavh · 2026-10-27
- Peter Diamandis: revisit the projects you abandoned because you couldn't afford the team — PeterDiamandis · 2026-10-27