logo
AI Memorized the Internet. That's Not Intelligence. | Yann LeCun
Futurology

80,236 views

992 likes

Chatbots mastered language because language was the easy part, and a house cat still knows things about the world that no machine does. | Subscribe to Futurology: https://bit.ly/subscribe_futurology

The machines that frighten us are very good talkers. They answer in confident paragraphs, so we assume there's a mind in there that will one day want something. But a system built to predict the next word acts without thinking ahead, which makes today's AI intrinsically unsafe but nowhere near smart enough to be dangerous. So the danger isn't the machine. It's who owns it.

In this episode, Yann LeCun argues that the models behind AI should be open and free for anyone to build on, like the software that runs the internet. LeCun, a Turing Award winner and Meta's former chief AI scientist, now chairs AMI Labs, where he's building machines that learn how the world works, not just how to talk about it. To keep AI out of a few hands, he wants one shared model, trained by countries, universities, and companies that each keep their own data. The alternative, he warns, is a world that gets its answers from three or four companies.

Subscribe to Futurology on YouTube:

https://bit.ly/subscribe_futurology

The Futurology Podcast – full episodes on YouTube:

https://www.youtube.com/playlist?list=PLwkiuf1eRbAAHP4p--2IFJbn0CIdaZMva

Apple Podcasts

https://podcasts.apple.com/us/podcast/futurology/id1821718921

Spotify

https://open.spotify.com/show/2I38HvHP6KlXrD5ysfygxk

Chapters

(0:00) Introduction

(1:21) What Is Intelligence?

(9:28) What a World Model Is

(21:18) The Abilities and Limitations of LLMs

(33:46) AI Is Still Built to Do Our Bidding

(37:45) Giving AI Guardrails

(47:31) Why LeCun Believes Current AIs Are Not Dangerous

(56:28) Who Should Regulate AI?

(1:02:28) Should American AI Be Serving the Whole World?

(1:11:17) Project Tapestry: A Global, Open AI

(1:19:10) Politics of Open Models

Mentioned in this Episode

Project Tapestry — the AI Alliance (Project, 2026)

https://thealliance.ai/projects/tapestry

A Path Towards Autonomous Machine Intelligence — Yann LeCun (Position paper, 2022)

https://openreview.net/forum?id=BZ5a1r-kVsf

A Severe Misalignment of AI in Mathematics — Terence Tao and 24 other Fields Medalists (Declaration, 2026)

https://terrytao.wordpress.com/2026/09/11/a-severe-misalignment-of-ai-in-mathematics/

Pessimists Archive

https://pessimistsarchive.org

Where to find Yann LeCun:

Home page: https://yann.lecun.com

AMI Labs: https://amilabs.xyz

X: @ylecun

Show ideas and feedback? Email:

futurology@berggruen.org

Learn more about the Berggruen Institute

https://www.berggruen.org

Follow Futurology!

Instagram: https://www.instagram.com/futurologypod

Twitter/X: https://www.x.com/futurologypod

TikTok: https://www.tiktok.com/@futurologypod

Facebook: https://www.facebook.com/berggrueninst

LinkedIn: https://www.linkedin.com/company/berggrueninst

Bluesky: https://bsky.app/profile/futurologypod.bsky.social

Credits

Executive Producers: Nicolas Berggruen, Nathan Gardels, Nils Gilman, Dawn Nakagawa, & Jason Hoch

Producers: Grant Slater, Alex Gardels, & Nathalia Ramos

Associate Producer: Elissa Mardiney

Theme Music: Marcus Bagala

Audio Engineer: Aaron Bastinelli & Kyle Scott Wilson

Futurology is a production of Studio B and Wavland for the Berggruen Institute in Los Angeles, California.