Yoshua Bengio: Deep Learning | Lex Fridman Podcast #4

[@jaymn5318]

Lex, You will be remembered in the history books for bringing these visionaries to talk about their work that ordinary folks can understand.

[@JohnGFisher]

Thanks for having him on Lex, love hearing from the guys on the cutting edge.

[@SY-me5rk]

Love how you trimmed the video, not a second of wasted time! Thanks.

[@kristian-io]

Please DO continue with these series! Thank you :)

[@TEETHclothing]

It’s mad to see what’s he has done with the podcast. Shout out to lex

[@TheNyatzAnger]

YES! Yoshua Bengio. Thank you for this continued AGI series Lex!

[@omaewayowaiii]

Such a pleasure to watch those podcasts, Joshua Bengio just came after Francois Chollet. Keep going and Greets from Russia ;)

[@bayesianlee6447]

Thanks for Lex always for having those great conversations and great lectures. If u r too busy Imma recommend u to watch 40:00 for two mins at least to get insights of how deep learning will go.

[@minoxidil4662]

if you're too busy just watch it later

[@deeliciousplum]

Wonderful talk. I do agree with Yoshua Bengio's critique of the 2014 film Ex Machina. Thank you for sharing this.

[@30803080308030803081]

I would like to see you interview Yoshua for a second time.

[@franktfrisby]

There is certainly a priori that helps build complex systems. Thanks Yoshua for this.

[@podcastsubtitled333]

Outline: 0:00 Introduction 3:42 Current state of deep learning 6:44 Architecture vs dataset 8:01 Learning through interaction 10:46 Our brain is big 12:40 Knowledge 24:01 Ex Machina 25:28 Bottle Ideas 27:54 Bias in Machine Learning 31:29 Teaching Machines 33:58 The Turing Test 37:48 Whats next 40:20 Gans

[@caterinadelgalles8783]

I looked uo some things I didn't understand and they are all computing terms, so, not my area of work! Such a great chat!

[@mmanuel6874]

This guy looks chill AF

[@onuryes]

Thank you so much Lex. Great content as always.

[@tzjtjktzjtzjztjztj]

Amazing questions for an amazing speaker. Thank you Lex!

[@Danny-ej3js]

Thank you Lex and Mr. Bengio! So many exciting points in there. I enjoyed the critique of Ex Machina (really liked the movie, but had similar problems with it).

[@gs-nq6mw]

The youtube automatic caption is almost perfect,and that's super crazy.One or two years ago those captions were completely nonsense and now they were perfect

[@marloncajamarca2793]

Amazing Interview! Thank you Lex and keep this great work!

[@caterinadelgalles8783]

I'll have to listen to this again and look some stuff up!

[@logicboard7746]

Gem of an interview...just saw in bits and pieces

[@CatsInHats-S.CrouchingTiger]

💕36:54- 37:34 💎 Excellent last question Lex!

[@darianharrison4836]

To my likes, Bengio is amongst the top 3 ML scientists in the world.

[@LNJP13579]

Yoshua is filtering the hype around AI & DL and speaking frankly. He has been a front-runner in DL for a long time and still is not carried away by it.

[@Wolfmoss1]

25:43 That moment was kind of ominous!.... ^_^

[@johnroach7607]

"You Shouldn't Trust Him" - Ava

[@daisukemajima8257]

I like this series. Please continue your cool way.

[@woodywiest]

Great discussion, thank you Yoshua and Lex. A question about infants vs. machine learning: If we view the world as we see it, a continual, light-speed fast, changing pice of data, would it be safe to say that the infant uses huge datasets as well?

[@suew303]

Every science progress is based on collective works of large group, beautifully said. Not only science, but almost everything, from poem, literature, even to every progress in our daily life. Individual who has name under each breakthrough is largely synthesizer of collective intelligence. ❤❤❤language, internet, and all technologies make it possible and easy.

[@dra.adrianadomingosromer4968]

I know this comment comes a bit late, as a dr I believe that what is missing is a language between components to be used in ai, so the differents components within a system can Interact, you can even take these examples in celular behavior. We shoulf observe the micro before we can apply rules to the macro.

[@alexharvey9721]

Maybe it's on the original podcast, but it might have also been nice to have discussed ethics a bit. As in, how do we decide when/if there are ethical considerations. For example if we think there is a chance an AI has human level understanding of itself and the world, is it ok to be deleting, updating, modified ect (even if it doesn't have a specific value for it's own existence). Especially because it will be so hard to define whether an AI is self aware or simply mimicking. The general trend in the scientific community seems to be very post-fact, like we would need to define with certainty that it is self aware before we worry about ethics. But it's necessarily impossible to do that. Even if we can somehow make that definition (which we possibly can't with certainty), we would already be doing those things in order to create it. The criteria shouldn't be that we're sure it's self aware, but (as we would do for a potential person) be sure that it isn't. But that would also be self defeating, as it would mean we can't develop the AI to make the distinction in the first place. I'm not sure it will matter or be discussed seriously anyway as most people seem incapable of considering that any AI can be self aware, even if they consider their own brain and consciousness to be of physical origin. Which is self contradictory, but we're great at putting what we want before what we know!

[@alekseysoldatenkov5675]

Great talk! Some feedback on the format: An unedited version might be better, the topics can get quite theoretical at times and when they do, it can take some time for the points to really set in. So the pauses can help with that. But I also don't mind re-watching. 😁

[@ProfessionalTycoons]

my favorite researcher

[@fataakstudio8501]

Your topics are pretty interesting!

[@jessicahardesty3358]

cc.Lex Fridman: cc.Yoshua Bengio: A strong case can be made to support your theory on sematics.

[@ahmadchamseddine6891]

Why no one is discussing the limits of the applied mathematics we have in encoding higher cognitive learning?

Mathematics or physics?

[@aigen-journey]

We have our natural brains as a point of reference and we can model every atom and every interaction with current mathematics/physics. It's just complex in terms of computational requirements. Why would artificial brains require new mathematics?

I thought that he might be meaning physical limitations. But mathematics has some limitations also. He will need to clarify.

[@aigen-journey]

Maybe you're right. I just read too many "the universe is mysterious" or "humans don't understand X, so how can we understand Y" comments that bring nothing to the discussion and this sounded like "there's this obscure part of mathematics that I personally believe is key to everything and all those researchers don't get it". If I misunderstood the comment and took it the wrong way, I'm sorry.

[@captaininsano8736]

@@aigen-journey maybe he means we'll need entirely new maths to encode the human mind. Thats literally whats implied. Idk if thats true but its interesting. I wonder if anyone serious has hypothesized this

[@kikirizki4318]

Where is the table of contents?

[@y__h]

Joshua is very impressive talker, I can't understand him in 1.5x speed.

[@ausroy087]

u missed the Y

[@Dondlo46]

You really liked to use fixed camera only on first few podcasts

[@kunjpatel1703]

Great.! Just when a transformation in currently adopted system is needed.

[@sherryhp10]

What an amazing interview ! I need to Understand Gans

[@rachidakebichi8733]

Premonition from Yoshua B.: need a system with human in the loop for training, this is exactly what was done for ChatGPT.

[@KulvinderSingh-pm7cr]

What you people think of using AR for supervised/reinforcement learning? Will it be our next step for learning sophisticated tasks?

[@suzakico]

One key missing linkage in AI, the mysterious point, seems to be the biological consideration. If our brain is an organ added/expanded later to explore our live better, the objective function of AI should have such consideration to orchestrate the whole even though certain suboptimization may take place as we may move our attention from one to the next. Sleep/dream or meditative process may represent the way to seek for better optimization by coming up with deeper insight for each of us to live better. The interplay of conscious and unconscious (while we may seek balancing various sub-optimized objective functions) is a way for us to respond to various stimuli to find a way to be alive - and to live better. My sense is: given that we live in the world of impermanence, we are made (i.e., programmed) to feel happy when we are in line to accomplish such a task. -> If so, my question to AI scientist is: how do you bridge the gap?

[@suzakico]

From my exposure to Zen, meditation, etc., I see being aware of "conscious of unconscious" is the way to discover insight/wisdom. Here, commonly used expression, Sila (right conduct/habit/edge) -> samadhi (reach out to unconscious mind and body are one - deep learning) -> panna/prajna/WISDOM (insight - know what is going on and how to act) is how the process may work - following the principle of Law/ Nature's way. Or, Thy will be done. I am very much interested in witnessing how AI may merge with such an age-old practice.

[@seetsamolapo5600]

Research - artificial neural networks are modelled on biological neural networks - the difference between biological neural networks and artificial neural networks - properties of BNNs and how they occur - mechanisms

[@jaiprakashbagotia4391]

Our pleasure too.

[@philippelaferriere2661]

25:41 The concensus is that breakthroughs in AI by private companies will (most likely) never be "bottled" and go undiscovered by the general research community. However, isn't Boston Dynamics in robotics a perfect example of exactly this?

[@idrissjairi]

It feels like he was talking about Transformer's architecture 😉

[@NikolaiVarankine]

Lex, thank you for the chance to listen to Yoshua. But where's your impression? Where's excitement?

[@EngIlya]

Great interview! But I did not like all the thinking moments cut out. Makes it unnaturally fast, and takes away the time for me to think as well.

[@MartinLichtblau]

Curiosity is a reward function (especially for newborns).

[@TheHellogs4444]

Less a reward function, more like that's the explorative policy that they use (and update to satisfy more curiousity)

[@alvinphantomhive3794]

​@@TheHellogs4444 and parent is more likely to play a role as an external critic assignment that attempt to reduce the explorative probability and using their prior knowledge to just straightly label the unexplored policy

[@jacobadamczyk3353]

@@TheHellogs4444 MaxEnt RL is the popular way to formalize this idea 💡

[@TheHellogs4444]

@@jacobadamczyk3353 YES! I loved entropy based models and maxEnt RL in particular is amazing. I think maxEnt RL with better models and dynamic updates will probably lead us close enough to AGI that there wont be any jobs any more.

[@jacobadamczyk3353]

@@TheHellogs4444 Have any specific ideas on new approaches? Or do you see any interesting new areas?

[@eaf888]

🔥🔥

[@deeplearningpartnership]

Great interview, thanks Lex. Yoshua seemed a bit angry.

[@rockapedra1130]

Deep Learning Partnership yep, not angry per se, more like grumpy. Maybe had an annoying day, lol.

[@hoyinchan343]

Human intelligence come from everyday natural learning.

[@darylallen2485]

Wow, not even 200,000 views in 5 years!

[@pazuzil]

this is the guy who starts Skynet

[@bassplayer807]

Or LEGION

[@IFFranciscoME]

24:30 and on how Hollywood pictures A.I. progress

[@bennokrojer8406]

6:49 "Neither!" 😂

[@MartinLichtblau]

Follow your intuition and update if new evidence appears.

[@bornslippy9109]

@24:40 Tesla: Hold my coil

[@MartinLichtblau]

Peculiarities and phenomena of human psychology == breadcrumbs to underlying structures We already have to much common knowledge - we just have to use it.

[@thunde7226]

wow............totally agree with the points of those movies....that one individual creates........and invents all things......no it doesn't happen like that.........Leon Musk has teams, and engineers that solve, and build more prototypes......Man can't be compare and idolize as God to create things..............................Great interview Lex......Thanks.............:) ................bye

[@pardoharsimanjuntak1483]

I hope that there will be intelligent beings from other planets who come to help us accelerate computer technology on earth, or humans who come from a future that brings technology from the future.

[@NisseOhlsen]

Pardohar Simanjuntak I hope we will soon assume responsibility and threat Earth better.

[@pardoharsimanjuntak1483]

are you worthy to be saved.

[@kislaykunal8921]

25:50 google duplex

[@АлександрБагмутов]

Oh, when vulcanians try to emulate emotions... so cute

[@caterinadelgalles8783]

Lex's Russian is limited. His statement about Russian poetry was a bit.....ok......but his father should have kept that Russian alive in him!

[@kimchi_taco]

Disentangle representation in hierarchical manner.

[@windowpane1000]

What if the lights went off because robots from the future are warning us? haha

[@mcole6121]

I hate how easily men dominate this field because of the wealth in AI