Yann LeCun: Dark Matter of Intelligence and Self-Supervised Learning | Lex Fridman Podcast #258
Video Overview & Insights
Yann LeCun is the Chief AI Scientist at Meta, professor at NYU, Turing Award winner, and one of the seminal researchers in the history of machine learning. Please support this podcast by checking out our sponsors:
Here are the timestamps. Please check out our sponsors to support this podcast.
0:00 - Introduction & sponsor mentions:
- Public Goods: https://publicgoods.com/lex and use code LEX to get $15 off
- Indeed: https://indeed.com/lex to get $75 credit
- ROKA: https://roka.com/ and use code LEX to get 20% off your first order
- NetSuite: http://netsuite.com/lex to get free product tour
- Magic Spoon: https://magicspoon.com/lex and use code LEX to get $5 off
0:36 - Self-supervised learning
10:55 - Vision vs language
16:46 - Statistics
22:33 - Three challenges of machine learning
28:22 - Chess
36:25 - Animals and intelligence
46:09 - Data augmentation
1:07:29 - Multimodal learning
1:19:18 - Consciousness
1:24:03 - Intrinsic vs learned ideas
1:28:15 - Fear of death
1:36:07 - Artificial Intelligence
1:49:56 - Facebook AI Research
2:06:34 - NeurIPS
2:22:46 - Complexity
2:31:11 - Music
2:36:06 - Advice for young people
- Public Goods: https://publicgoods.com/lex and use code LEX to get $15 off
- Indeed: https://indeed.com/lex to get $75 credit
How can someone pursue phd under Yann Leccun supervision?
- ROKA: https://roka.com/ and use code LEX to get 20% off your first order
- NetSuite: http://netsuite.com/lex to get free product tour
Ironic that intrinsic motivation dicovery for A.I. is also what we need to learn to harness within humans. By unlocking this for humans or A.I., we will advance the complexity of the universe. Beneficial for humans, A.I., everyonne. LOVED this podcast. THANK YOU!
- Magic Spoon: https://magicspoon.com/lex and use code LEX to get $5 off
EPISODE LINKS:
Around 2:30:00 Yan is talking about how scientific measurements of complexity always include "arbitrary nicetiesâ, like âthe inverse log of probabilityâ (a function tied to probability), setting up the scenario where you have a random data that defines some other worldly life. The life itself applies a function to that chaos to simplify it, (correlating colors with objects, etc) but if Aliens saw just the initial data, it would just be seen as heat.
Yann's Twitter: https://twitter.com/ylecun
Yann's Facebook: https://www.facebook.com/yann.lecun
Yann said it regarding Model Predictive Control (MPC), these are fundamentally deterministic models, in the worst case non linear or affected by gaussian noise like Kalman filters, but stochastic models are much more complex and high dimensional stochastic models are intractable in many cases. RL uses Bellman's equations which is also used on stochastic prediction, and Yann dislike it by the way, why? simply because it requires a large amount of trial and error repetitions to learn something useful. So, I would say it's an undecidable problem.
Yann's Website: http://yann.lecun.com/
Books and resources mentioned:
red pill meta
Self-supervised learning (article): https://bit.ly/3Aau1DQ
PODCAST INFO:
If you say FBs blatant censorship is evil, then you imply that those working at FB are evil. Since Lecun can assure you that they arent FB censorship is not evil. All according to Yann.. I dont take any opinion from this sellout seriously. Not even if the issue is asswipes.
Podcast website: https://lexfridman.com/podcast
Apple Podcasts: https://apple.co/2lwqZIr
1:33:45 I believe you can live at peace that you will die but everyone sheds a tear on their death bed you canât convince me otherwise
Spotify: https://spoti.fi/2nEwCF8
RSS: https://lexfridman.com/feed/podcast/
Ja vad jag förstÄr sÄ Àr vissa Àr bÀttre pÄ att ge inspiration och kunna ge mer Àn bara tar och ja allmÀnt skicklig pÄ vad han gjorde Tur att det finns sÄ sÄna mÀnniskor sÄ sticker du ut sÄ att inte alla Àr som den ordinÀra och har sÄna olika kunskaper och Kan göra gott för andra 07
Full episodes playlist: https://www.youtube.com/playlist?list=PLrAXtmErZgOdP_8GztsuKi9nrraNbKKp4
Clips playlist: https://www.youtube.com/playlist?list=PLrAXtmErZgOeciFP3CBCIEElOJeitOr41
"..that's the problem with relationships, you can't delete the other human.." - Lex
I need this on a t-shirt
OUTLINE:
0:00 - Introduction
this is the clown who convinced mark zuckerberg to be a woke shill for the far left.
0:36 - Self-supervised learning
10:55 - Vision vs language
Really fun one here. Great guest!
16:46 - Statistics
22:33 - Three challenges of machine learning
I thought this would be a direction to try within a controlled safe zone, having a robot interact outside of the video learning. As much information and knowledge already enlisted, see if it can learn as a teenager does driving. Sooner or later "hands-on" will be applied. Lex is always on it. 1:21:43 perhaps confusion comes from saftey shutoff if a person is driving so maybe it's hangup there, such a as talking or texting while driving. (Position swap)
28:22 - Chess
36:25 - Animals and intelligence
Bro i am from agust 2024 and meta put its IA in wp and sucks
46:09 - Data augmentation
1:07:29 - Multimodal learning
What LeCun said about the limitations of language was interesting. Now more than ever our labelling for example, keeps failing to map our reality effectively.
1:19:18 - Consciousness
1:24:03 - Intrinsic vs learned ideas
Unsolvable with the binary system even quantum computing unsolvable sorry thank you
1:28:15 - Fear of death
1:36:07 - Artificial Intelligence
Too much overthinking always leaving out the word environment think about trains and their direct connection to self driving cars what is the one thing we understood then that would be environment thank you
1:49:56 - Facebook AI Research
2:06:34 - NeurIPS
This just came up on my YouTube feed two years later. Wow, what an extraordinarily prescient discussion.
2:22:46 - Complexity
2:31:11 - Music
He has been so confidently wrong in predicting the future of AI lmao
2:36:06 - Advice for young people
SOCIAL:
This is basically a whole semester of self supervision ml, the knowledge is golden.
- Twitter: https://twitter.com/lexfridman
- LinkedIn: https://www.linkedin.com/in/lexfridman
Just for humor... No offense, Why do you think I have this craaaazy accent!?!? Lol. M.P.
- Facebook: https://www.facebook.com/lexfridman
- Instagram: https://www.instagram.com/lexfridman
What if Lex is AI?! đ
- Medium: https://medium.com/@lexfridman
- Reddit: https://reddit.com/r/lexfridman
Once u discover the power of compound intrest you are made đ my cup runneth over..
- Support on Patreon: https://www.patreon.com/lexfridman
Instead of predicting the future - give it a clip of what just happened and ask it what lead up to it. Then give it the video clip given to it.
More User Perspectives
Emotions in LMMs will be interesting, giving it a sense of humor will fascinating....I guess Asimov's Positronic Man story that became the film "Bicentennial Man" is one look at that....(we've come a long way since Robbie in "Forbidden Planet ")
What a great companion they'll make for us oldsters, and how much more pleasant they'll be to work with, but I forsee all sorts of problems, mostly for people like me that are borderline antisocial already. We'll never want to go out!
I think all the social media have the same problems, including YouTube: as a species humans have been handed this incredibly powerful tool and it's going to take some time for us to adapt to it. Eventually we'll figure out how to use it most productively. (says a septugenarian YouTube addict. đ) Thank you kindly for the video.
You make wonderful interviews; however, you speak to your guests in a voice that has just woken up and has not yet woken up, which makes me sleepy. Any how, thank you. Take this as a constructive critisizm
@2002budokanHappiness is a choice
@aidanthompson5053Ideology is dangerous - donât be a âknow it allâ
@aidanthompson5053Does anyone know the name of the paper referenced at 2:12:42?
@MultiMojoGPT 3.5 seems to understand what happens when you move a table just fine. I tested it.
@abramsonrlTime for a revisit@lexfridman
@schnsthe2:38:45 - FIRST try to cure the common cold using deep leaning. All these are just empty promises.
@oneshot20281:31:59 - This this shows how so-called award winning scientists also behave like religions people, because we really don't know for sure that we will not exist after death.
Not believing in religion doesn't automatically mean one is free from the fear of death. Many non-believers still fear death because of:
The finality of death (i.e., ceasing to exist).
Unfinished goals or regrets.
Pain or suffering associated with dying.
Concern for loved ones left behind.
Watch talks by Jim B. Tucker who was the protégé of the pioneer Ian Stevenson on why computers programs can NEVER simulate a human brain and all this AGI and ASI will be an utter failure, because the brain is not a computer program. AI will end with ANI.
@oneshot2028There are many complex instinctual behaviors seen in animals merely hours after birth, I might disagree with Dr. LeCun at 1:00. I suspect that there is a far deeper encoding somehwere .... one need only watch how a newborn foal quickly stands and walks. We have no idea of how this behavior is passed down genetically.
@MikeClarkeARVR43:10 Yann LeCun thinks LLMs doing IQ test is so far outside of his radar. Just six months later, GPT-4 scores 155 on an IQ test. It's crazy how fast progress is going and how capable AI is getting. But even more impressive is how so many experts like LeCun structurally underestimate capabilities.
@dr-maybe17:55 the fact that he claims that intelligence could be statistics while completely ignoring that actual neurons work completely differently than weights and biases seems kinda odd ... the industry is happy to send vast amount of ressources down the drain instead of comming up with new fundamental architectures of neural networks
@mjdev-i1pmake a podcast with Miguel Nicolelis, brazilian neuroscientist
@AnttonioFFit's a privilege to hear LeCun talk about ML
@AndruXaHedgehog approves?
@mirekrusin41:45 Lex trying to argue that walking is socially constructed lmao
@rpbmpnI get really scared when "Chief AI Scientists" are that bad at predicting AI capabilities.
LeCun 57:55:
You take an object, place it on a table, and then push the table. It's completely obvious to you that the object will be pushed along with the table, because it's sitting on it. I believe there is no text in the world that explicitly explains this. So, if you train a machine, as powerful as it could be - let's say your GPT-5000 or whatever - it's never going to learn about this phenomenon.
ChatGPT (GPT 4):
If you push the table gently, the object might stay in place due to friction, although it may slide or wobble slightly. If you push the table with a greater force, the object might slide or fall over, especially if the object is top-heavy or not very stable.
Re: chess grandmasters, that's not true. The main reason they react quickly is because they have memorized the openings.
Later in the game, once deviating from the standard openings, they have to think just like everyone else has to. They might be faster, but it's not a subconscious process.
wow! If you knew then what you know now....
@garyswift9347the MAGA chases the democrapper in the savannah
@eulederminervaMental models are inexcrutably linked to language. It may very well be that the latent representation is the model.
@HappyMathDadCondense with Unicode symbols đ„
@NanoartsNet