
Smart Until It's Dumb: Hallucinations

AI is Smart Until It's Dumb: Why LLM Will Fail When You Least Expect It - Emmanuel Maggiori
Topic 5 of 19
Published Nov 10, 2025 · Clip 13:54–17:38 (3:44)
Emmanuel explains the thesis of Smart Until It's Dumb: AI is impressive until it makes a mistake a human would never make—what's now called a hallucination (he originally called it an 'epic mistake'). He cites self-driving cars as the casualty of this flaw: Cruise has more remote operators than cars, and Waymo intervenes manually every three to five miles. He explains that machine learning learns by repetition and interpolation from examples, so it fails on out-of-distribution situations where humans excel thanks to internalized world models. This same flaw hampers enterprise adoption and serious coding use.