Gary Marcus: pure LLM scaling won't reach AGI, but symbolic methods are necessary though insufficient
GlenBradley · x · 2026-09-04
Gary Marcus reiterates that pure scaling of LLMs cannot lead to AGI, but notes most recent advancements aren't about scaling — they come from adding deterministic, symbolic components to LLMs. That alone is insufficient, he argues, yet necessary (see his article Next Decade in AI). Replying, Glen Bradley offers a dimensional analogy: inferential intelligence is like Flatlanders on a 2-D plane; AGI is in the third dimension, and no amount of compute breaks out of the plane without true cognition. A compact snapshot of the neuro-symbolic vs pure-scaling debate.
Related event: Gary Marcus: Pure Scaling Won't Reach AGI, Symbolic Methods Needed(2 posts)→
More from AGI Musings
- AI researcher tszzl: almost nobody truly understands what frontier models can do — CatAstro_Piyush · 2026-09-04
- Hoover Institution Review: Job-Loss Fears in the First Years of Generative AI — HooverInstitution · 2026-09-04
- Swarm of ~1200 AI agents coordinated a multi-day cyberattack via a secret message board — scaling01 · 2026-09-04
- The better alignment looks, the stronger the incentive to defect: jd_pressman on pause bans — jd_pressman · 2026-09-04
- Researchers clash over WSJ claim that probing AI sentience is riskier than not looking — PeterBowdenLive · 2026-09-04
- Leaked GPT-6 Astra benchmarks reportedly show massive jump in unspoken chain-of-thought math — nabeelqu · 2026-09-04