Full podcast episode: th-cam.com/video/5t1vTLU7s40/w-d-xo.html Lex Fridman podcast channel: th-cam.com/users/lexfridman Guest bio: Yann LeCun is the Chief AI Scientist at Meta, professor at NYU, Turing Award winner, and one of the most influential researchers in the history of AI.
Damn. I really don’t understand all this criticism about RL. The “inefficieny” that stems from RL comes from the fact that finding the optimal policy of a Markov Decision Process is a P-Hard problem. So that makes this problem as hard as solving any problem that can be solved efficiently(by a deterministic algorithm). I’m honestly kinda surprised that LeCun would take this stance.
RL is inefficient because it is a random optimizer. Nothing wrong with that, but if you can actually use gradients (first order optimizer) you can be orders of magnitude (100x, 1000x, etc) more efficient.
Honestly compared to Yann's prior statements on RL this is softened. He used to say RL was completely useless. Now he is basically like "It is technically useful, and indispensable at times (as in when planning fails), but is not useful for general use because of its inefficiency." I think this is a much better take.
LeCun's critique of reinforcement learning (RL) offers a refreshing perspective on AI development. His emphasis on efficiency and the use of model predictive control over RL is a fascinating take that challenges conventional wisdom
@@osuf3581 We'll I guess Ill head back to my CS master's program to inform the machine learning faculty to stop referencing his decades of deep learning research and literature in our course material. 🤷♂️ They must have him confused for someone else. I'll also need to stop by Meta to tell them to fire their chief AI scientist, tell all these international organizations to rescind their awards and honors, and all these universities from inviting him to speak. Might also call up the other 2/3 godfathers of AI to disassociate themselves with him and have him stripped from their groundbreaking research. Thanks for the heads up. 🔥
![NISAMANEE - TADA (ท้าดา) [ OFFICIAL MV ]](http://i.ytimg.com/vi/hcQHEPMmJpU/mqdefault.jpg)
![เปิดบ้าน โคตรสวย ป๋องกพล รู้จักตั้งนาน เพิ่งรู้ว่ารวย!!! l [Nickynachat]](http://i.ytimg.com/vi/tmgCF_2HfYI/mqdefault.jpg)
Full podcast episode: th-cam.com/video/5t1vTLU7s40/w-d-xo.html
Lex Fridman podcast channel: th-cam.com/users/lexfridman
Guest bio: Yann LeCun is the Chief AI Scientist at Meta, professor at NYU, Turing Award winner, and one of the most influential researchers in the history of AI.
I thought he meant Real Life is overrated. Was about to agree with that.
😂😂😂
Same
Damn. I really don’t understand all this criticism about RL. The “inefficieny” that stems from RL comes from the fact that finding the optimal policy of a Markov Decision Process is a P-Hard problem. So that makes this problem as hard as solving any problem that can be solved efficiently(by a deterministic algorithm).
I’m honestly kinda surprised that LeCun would take this stance.
RL is inefficient because it is a random optimizer. Nothing wrong with that, but if you can actually use gradients (first order optimizer) you can be orders of magnitude (100x, 1000x, etc) more efficient.
Honestly compared to Yann's prior statements on RL this is softened. He used to say RL was completely useless. Now he is basically like "It is technically useful, and indispensable at times (as in when planning fails), but is not useful for general use because of its inefficiency." I think this is a much better take.
I predict this will be a banger
The popularity of DPO and related algorithms is already casting a shadow on RL
Thought the title was Real Life is overrated 😅
LeCun's critique of reinforcement learning (RL) offers a refreshing perspective on AI development. His emphasis on efficiency and the use of model predictive control over RL is a fascinating take that challenges conventional wisdom
Nice ai generated comment
@@manny3031 AI generated essay paragraph
How to say nothing with many words…
I thought he was gonna talk about why real life is overrated 😆
This Gary Marcus wannabe again? Frequently wrong. Get someone actually competent on the show.
You're joking right? 😅
Do you not know who LeCun is? There aren't many people even AS competent, much less MORE competent.
@@zillyzonka999 - You are uninformed. He's hardly relevant and contradicted by the more qualified
@@osuf3581 We'll I guess Ill head back to my CS master's program to inform the machine learning faculty to stop referencing his decades of deep learning research and literature in our course material.
🤷♂️ They must have him confused for someone else.
I'll also need to stop by Meta to tell them to fire their chief AI scientist, tell all these international organizations to rescind their awards and honors, and all these universities from inviting him to speak.
Might also call up the other 2/3 godfathers of AI to disassociate themselves with him and have him stripped from their groundbreaking research.
Thanks for the heads up. 🔥