Gary Marcus: pure LLM scaling won't reach AGI, but adding symbolic layers is necessary

GaryMarcus · x · 2026-09-04

Asked whether recent progress changed his view, Gary Marcus reiterates that pure scaling of LLMs cannot lead to AGI. However, he argues most recent advancements aren't about scaling at all — they add deterministic, symbolic machinery on top of LLMs. Even that won't suffice for AGI, he says, but it is necessary (see his article 'Next Decade in AI').

Original post →

More from AGI Musings

AGI Musings channel →