Clear+Vivid with Alan Alda

Gary Marcus: Is AI mostly hype?

Mar 24, 2026
Gary Marcus, cognitive scientist and author who critiqued modern AI, offers a skeptical take on large language models. He discusses where LLMs fall short, the need for neurosymbolic approaches and durable world models, economic and security risks of unchecked AI access, and concerns about funding and long-term research. Short, pointed reflections on AI’s future and pitfalls.
Ask episode
AI Snips
Chapters
Books
Transcript
Episode notes
INSIGHT

AlphaFold Shows Where AI Truly Helps Science

  • Gary Marcus values scientific AI applications over chatbot hype.
  • He praises AlphaFold as a focused model that predicts 3D protein structure and calls it the best AI contribution to science.
INSIGHT

LLMs Are Fluent Predictors Not Meaningful Thinkers

  • Marcus argues LLMs merely predict next words and lack genuine understanding.
  • He says that statistical mimicry leads to hallucinations, making LLMs unreliable for tasks requiring factual trust.
ANECDOTE

ChatGPT Wrote A Wrong Biography For Harry Shearer

  • Gary Marcus recounts ChatGPT writing a false biography of Harry Shearer as British when he was born in Los Angeles.
  • He uses this to show how easily LLMs hallucinate basic factual details despite available sources.
Get the Snipd Podcast app to discover more snips from this episode
Get the app