Why AI Is Still Dumb, Feynman Saw It Coming

Описание: AI can sound brilliant and still be wrong in a way that’s systematic.

It can write code, explain physics, pass exams, and then fail on a tiny change in wording. It can argue confidently for a false claim. It can “reason” in paragraphs while missing the one constraint that matters.

This video uses a Feynman-style test: stop admiring the output and ask what the system is actually doing. What is the mechanism. What is the signal. What would prove it understands. What would expose that it doesn’t.

We’ll break down why today’s AI feels smart: it’s extremely good at pattern completion, compression, and imitation of expert language. Then we’ll show why that same strength produces the familiar weaknesses: brittle logic, shallow grounding, invented details, and confident nonsense that looks like insight.

You’ll also learn a practical way to use AI without getting fooled:
how to design prompts that behave like experiments, how to force checks, how to detect “word-math” posing as reasoning, and when you should treat the model like a calculator versus like a storyteller.

SOURCES:
Richard P. Feynman — “Simulating Physics with Computers” (1982)
Richard P. Feynman — The Character of Physical Law (1965)
Richard P. Feynman — Feynman Lectures on Computation (posthumous collection of lectures, ed. A. J. G. Hey)
Alan M. Turing — “Computing Machinery and Intelligence” (1950)
Claude E. Shannon — “Programming a Computer for Playing Chess” (1950)
Joseph Weizenbaum — “ELIZA” (1966)
Peter Norvig — “On Chomsky and the Two Cultures of Statistical Learning” (2011)

CREDITS:
Script concept inspired by the public lectures and writings of Richard P. Feynman.
AI narration and visuals.
Produced by Dr. Mira Caldren, Independent Physics Research Fellow.

WARNING:
This video uses synthetic voice and visuals. It is an original fictional lecture inspired by Richard Feynman’s teaching style and public ideas. It is not an authentic recording, endorsement, or statement by Richard Feynman or his estate.

What’s the most “smart-but-dumb” AI moment you’ve seen?