Lex Fridman Podcast #258 – Yann LeCun: Dark Matter of Intelligence and Self-Supervised Learning
192 snips
Jan 22, 2022 
Yann LeCun, Chief AI Scientist at Meta and Turing Award winner, dives into the fascinating world of self-supervised learning. He discusses how this approach mimics human learning, distinguishing it from traditional methods. LeCun explores the complexities of machine intelligence, emphasizing the blend of causal reasoning and background knowledge. The conversation also touches on the evolution of intelligence across species, the philosophical implications of AI and mortality, and the future of human-machine interaction, making for an enlightening dialogue on the nature of knowledge and learning.
AI Snips
Chapters
Transcript
Episode notes
Dark Matter of Intelligence
- Self-supervised learning is like the dark matter of intelligence, crucial but unseen.
- Humans and animals learn efficiently through observation, unlike current AI.
Background Knowledge
- Background knowledge, built through observation, is key for quick learning.
- Humans learn intuitive physics by observing the world, unlike tabula rasa AI.
Self-Supervised Learning Signal
- Self-supervised learning extracts truth from the world without explicit human labels.
- Predicting future events in video or filling gaps in text/images helps build world models.