Gary Marcus: Pure LLMs Are Stochastic Parrots; Astra Proves Need for Neurosymbolic AI
GaryMarcus · x · 2026-08-02
Gary Marcus clarified the ongoing confusion around the term "stochastic parrots." He noted that while pure LLMs are essentially stochastic parrots, advanced systems like Google's Astra are almost certainly not pure LLMs.
He reiterated his long-standing position that LLMs will only be part of the ultimate solution, not the entirety of it, emphasizing the necessity of neurosymbolic computation. He argues that if Astra indeed harnesses symbolic tools, it perfectly confirms his conjecture.
More from AGI Musings
- AI Is Nondeterministic and Must Be Verified: Expert Calls for Human-in-the-Loop Audits — alexcovo_eth · 2026-08-02
- LLMs Overturn Math Conjectures, Mathematicians Face Existential Crisis — ryan_t_lowe · 2026-08-02
- Ex-Googler Reveals: Google Had a ChatGPT Rival One Year Before Launch — SIGKITTEN · 2026-08-02
- Developer: AI demos are 'cool' but production systems are far from 'vibe coding' — alexcovo_eth · 2026-08-02
- AI Impact: Is Starting a Theoretical CS PhD Today Still Worth It? — aminkarbasi · 2026-08-02
- Self-Improving Models Will Accelerate AGI, Expected by 2028 — haider1 · 2026-08-02