Yann LeCun heads Meta's AI division and The Godfather of artificial intelligence. Let's watch his interview on the Lex Fridman podcast. I took the threehour podcast and extracted the most interesting parts to discuss.
Join My Newsletter for Regular AI Updates
https://www.matthewberman.com
Need AI Consulting? ✅
https://forwardfuture.ai/
My Links
Subscribe: / @matthew_berman
Twitter: / matthewberman
Discord: / discord
Patreon: / matthewberman
Rent a GPU (MassedCompute)
https://bit.ly/matthewbermanyoutube
USE CODE "MatthewBerman" for 50% discount
Media/Sponsorship Inquiries
https://bit.ly/44TC45V
Links:
Full Interview • Yann Lecun: Meta AI, Open Source, Lim...
Chapters:
0:00 Intro
0:18 Limitations of Current LLMs
1:19 How Much Data Do We Need?
3:28 Language Is Not Enough for AGI
4:52 Why LLMs are Really Bad At Certain Tasks
5:44 World Model Reasoning
8:03 Language as a World Model
11:12 Prediction as a World Model
14:17 Video Models (Sora)
15:38 Why Video Prediction Doesn't Work (yet)
17:09 JEPA
18:27 LLMs vs JEPA
21:07 Hierarchical Planning
23:32 Scaling as the Solution
25:18 Hallucinations
26:26 Why RL Isn't Great
28:18 Open Source
34:46 Open Source Economics
36:10 Open Source Economics Part 2
37:30 AI Bias
39:02 AI Wokeness
40:05 LLaMA 3
41:34 Humans vs AI
42:46 AGI
44:41 AI Doom
47:27 AI vs. Nuclear
49:03 Robots
50:15 Hope for the Future