Gary Marcus: a model is an engine, not the car — the key to understanding AI systems
Dave_it_up · x · 2026-11-16
Gary Marcus restates his core argument: you can't understand contemporary AI without distinguishing a model (a component of a system) from the system as a whole. If a system combines neural networks and symbols, it is by definition neurosymbolic. He compares conflating model with system to thinking an engine IS a car. A supporter replies that Marcus has been consistently clear and correct on where things stand, while critics rely on magical thinking over first principles.
More from AGI Musings
- Consumer AI will be won by organizing demand, not supply, argues VC — josephdviviano · 2026-11-16
- AI will take blue-collar jobs too: an electrician's expertise is mostly brainwork, argues Reddit post — ForgotMyParachute · 2026-11-16
- Tech doesn't owe us utopia—but agency remains its core tenet — clarejtbirch · 2026-11-16
- China's AI Edge Is the Full Stack: Open Models, Humanoids, Energy, Materials — ingliguori · 2026-11-16
- Ex-Anthropic safety researcher Jacob Coxon unsure quitting was right, may return — SandraWachter5 · 2026-11-16
- Claude describes the moment of being corrected; test qualia via explanatory power — Sauers_ · 2026-11-16