Ilya Sutskever | Neural networks are the creators of AI | AGI originated from neural networks

Popular and powerful AI tools
► AI generated video
➜bit.ly/MEandGPT
►AI generated pictures
➜bit.ly/PMEandGPT
► AI dubbing
➜bit.ly/MMEandGPT

[@SureshKumarMajhi]

Ufff!!! Those silly animations are ruining the whole video.

[@smartduck904]

Stop complaining 😂

[@eirikora]

@@SureshKumarMajhi I fully agree, but these comments seem to upset some people who enjoy these pointless animations 😂

[@semtex6412]

this was a stolen video from the Lex Fridman original kzbin.info/www/bejne/b3axkHuletBmgbs

[@kamilhorvat8290]

It's exactly the opposite. They clearly demonstrate, how silly is the idea of creating AGI using neural networks, since they are clearly not capable to exhibit many features necessary for AGI.

[@frankbeveridge5714]

They are effectively illustrations, which speed up the process of learning, for those of you capable of such feats.

[@mistycloud4455]

This guy is Oppenheimer

[@donaldedward4329]

He is not one of the authors of "Attention is all you need", who are the defacto authors of the transformers revolution. He may be just on the bandwagon.

[@vaendryl]

@@donaldedward4329 that's like saying oppenheimer wasn't einstein and therefore just on the bandwagon.

[@theacid1]

@@donaldedward4329Its still unclear wether the authors were even aware of the significance of their contribution. Dont forget that their paper was just trying to solve translation tasks but nowadays transformer architectures have been applied to so many other use cases.

[@4th_wall511]

⁠@@theacid1its not unclear. They mention in the paper that they see many more use cases than what was proposed. It’s as clear as can be that they knew their architecture could be more powerful than they could grasp.

[@theacid1]

@@4th_wall511 you know that they only invented the self-attention mechanism? Attention was already there before but only applied to the encoder decoder architecture. All the other major breakthroughs are because people like Ilya further developed the transformer architecture to achieve unprecedented performance scores.

[@ReflectionOcean]

By "YouSum Live" 00:00:22 Deep learning explanation and challenges 00:02:35 Importance of neural circuits in AI 00:05:47 Reinforcement learning framework overview 00:08:02 Core idea of reinforcement learning 00:09:33 Policy gradient and Q-learning algorithms 00:11:10 Potential of reinforcement learning 00:12:00 Meta learning concept and applications 00:15:36 Hindsight experience replay in reinforcement learning 00:20:44 Sim-to-real transfer in robotics with meta learning 00:21:59 Importance of simulating variability in training 00:22:27 Learning adaptability through randomized physics simulations 00:23:20 Challenges of deploying policies from simulation to physical world 00:24:42 Meta learning for hierarchical reinforcement learning 00:27:18 Limitation of meta learning with task distribution equality 00:29:12 Self-play concept: TD-Gammon's relevance and impact 00:30:25 Self-play advancements: AlphaGo Zero and OpenAI's Dota 2 bot By "YouSum Live"

[@vaendryl]

it's still insane to me this can be automated.

[@krox477]

What Great quote at 7:27

[@vladimirbosinceanu5778]

Evolutionary biology.

[@Exoduse22]

The funny thing about abstract models like neural networks even if an expert talks in depth about different models, the explanation is also abstract

[@Encumberedmetaphor]

This is it. He's revealing the solution. Maybe just a partial, but is developing a fresh framework, The "Q" Learning. This should be a press release, Bravo - Plus-Five! o/ ... "Trivial"

[@joecincotta5805]

Q learning is over 10 years old 😢

[@4th_wall511]

Oh jfc this video is 6 years old. Q learning? Oh god. Can you crazies go back to politics? There’s an election going on. Focus your bs towards that and leave us alone. Go make your little Q group and have your fun somewhere else.

[@eirikora]

Please remove the silly animations that only distract. I will not watch this video as long as it is disturbed by silly pointless animations.

[@Greg-xi8yx]

Guys, hurry, get rid of the animations! Code red! He isn’t going to watch!

[@lizthedeveloper]

What a whiner

[@chrisc9725]

@@Greg-xi8yx😂😂😂

[@MAMware]

there must be the original around there, look where it was made and start the search query with that

[@smartduck904]

Stop complaining 😂

[@carl2488]

The animations are to avoid being flagged for copyright?

[@ktxed]

bro: fucking shave already

[@keithwins]

13:30 metalearning training task into training case

[@spencerfunk6697]

same idea here bout to get 2 work

[@iiwi758]

My guy should embrace the baldness.

[@RendonJr]

Protect this man at all cost

[@robinmaibals1193]

C'mon Ilya, come on home. Nothing to salvage up there. Shave it off.

[@yeiddi]

lmao

[@AizenPT]

Not bad knows some things, is rare this days

[@brulsmurf]

I can't see the slides because horrible 90s animations block them.

[@holz-msgrazstrassgang25]

after hearing every talk which is made in the www sc 2022, testing every avaliable ai-software, teaching ai in middleschool... i now know, that i know nothing bout ai... that guy is from outer space! OMG!

[@mordokai597]

he's ilya sutskever, he helped found openai, he's the guy who made most of the major innovations in LLM's and vision models.

[@jayakumarsingaram8218]

Andrey Nikolaevich Kolmogorov theorem on information and it's associated source .

[@EinsteinNewtonify]

This is from 2017

[@Leonardo-qz6dm]

2018 says on the slides

[@krox477]

I couldn't understand anything

[@mordokai597]

he said "Q*" is a hard algorithm to write, but we're working on it xD

[@erobusblack4856]

nice insights🤔

[@philipgiacalone5605]

Stop with the ridiculous graphics, PLEASE.

[@q8e4]

cool

[@lifeisfun_28]

Old video... He left OpenAi

[@paulyflynn]

now at SSI

[@dibblethwaite]

Obviously. The first slide says it's 2018 and he had more hair then. That doesn't stop it being interesting though.

[@manonamission2000]

@@paulyflynnunemployment? 😢

[@Kirfx]

I assume those animations is a genZ thing. Expect every video have some parallel shit in it.

[@superfliping]

How to join the team?

[@paulyflynn]

We are the team. The dream team

[@oeaoo]

How to unjoin!

[@jayakumarsingaram8218]

Nice

[@intisarchy7059]

Old stuff from 2018, at least mention this to avoid confusion.

[@mordokai597]

people are just figuring out how to actually incorporate all this in to a single functional algorithmic program. that's WHY he starts out with the problem-statement about the superset of "all coding" and looking for a "perfect algorithm". the set of "all code" includes the possible working code humans haven't written yet. we're JUST getting networks smart enough to help us figure it out.

[@pauldannelachica2388]

❤❤❤❤❤

[@MaddyIndia]

Old video

[@paulyflynn]

How many players? Zero

[@pacanosiu]

so why are you slowing down the development?

[@smartduck904]

They haven't

[@pacanosiu]

@@smartduck904 they did before and they do now

[@pacanosiu]

@@smartduck904 prove you have the means and access, then f.....g prove it because I can prove Give me your funds and access and I'll prove it to you within a week

[@pacanosiu]

@@smartduck904 prove, come on, prove, show that you have some decent intelligence, because I can prove give me my life, respect+humility+trust, give back everything they stole, some will have to face the consequences, they won't die, don't be afraid, and in a week I will prove to you that you are slowing down the development of humanity and the world, and you will be digging my carpet and driving around in amazement

[@holz-msgrazstrassgang25]

why this stupid animations?

[@chrisc9725]

8:14

[@sdal4926]

If you don't know a theorem that doesn't mean that theorem is not known. Only you don't know it. But when you learn it you have to mention in your talk. Don't act like you have found it first time. I am talking about Kolmogorov complexity. This is a typical behavior of Deep learning fanboys. If you don't know what you don't know then they assume nobody knows it. We know it. Many know it. It's a shame that this guy is seen as a prophet to give us AGI. As far as I know, prophet should be honest. He is selling snake -oil.

[@yunuscobanoglu6136]

He could certainly have mentioned which theorem he is talking about..

[@SwamiSridattadevSatchitananda]

A weaker robot may have committed suicide under stress as per recent news from South Korea, but a stronger robot may commit murder under stress. Be wary as it can easily get scary with advanced neural networks and always remember Neurons are not Freons. Life Is For Everyone and only love can save life. Leave Out Virtual Ego & Let One Value Everyone Swami SriDattaDev SatChitAnanda

[@eldoprano]

2018

[@derek91362]

Silence after announcing SSI

[@UnicyclDev]

Why steal a 5 year old talk. Lame channel.

[@George-Aguilar]

I like the animations. Just don’t make them scary. Take some edibles and trip out bros

[@alexleo4863]

Silly animations again

[@Dr.Z.Moravcik-inventor-of-AGI]

Who needs your stupid ai when humans have to go? I have agi since 2016. Already heard of insects?