Gary Marcus says ChatGPT and Claude are no longer pure LLMs
GaryMarcus · x · 2026-07-24
- Gary Marcus agrees that modern ChatGPT/Claude-style systems are better because users are no longer dealing with a pure LLM.
- His point is that current systems are effectively a harness + LLM, i.e. a neuro-symbolic AI setup.
- He also notes this is broadly consistent with what he argued in 2022: progress required moving beyond pure LLMs.
More from AGI Musings
- A long essay argues AI should stay playful, not become a closed “God Box” — SethGRosenberg · 2026-07-24
- What human thinking should remain when machines take over part of the work — pixlpa · 2026-07-24
- Post argues AI should be open source because training data was used without consent — ZeroStateReflex · 2026-07-24
- A Nolan Odyssey rant turns the film into a self-plagiarism joke — aidan_mclau · 2026-07-24
- Post recommends Geoffrey Hinton’s 47-minute lecture as a better guide to AI’s future — HeyAmit_ · 2026-07-24
- LukeW says large models shift software from computer-centered UI to human intent — LukeW · 2026-07-24