You'll probably never hear him say that in the sense that it won't be happening anytime soon that we'll have AGI that's usually displayed in fiction. On the other hand, the current weak AI systems we have are still dangerous and could certainly prove to be much more dangerous than a dirty bomb since unlike the bomb they won't be used merely as a deterrent. So maybe you will hear him say it (it won't be just him though as this is a collective effort that involves all ML/AI researchers and any developers who contributed to the field)
@@goncaloshred Cryptocurency is fine but Bitcoin is irresponsible for our environment (invest in Etherium instead of Bitcoin). It's Bitcoin's Proof of Work algorithms that consume way too much energy. Bitcoin consumes more energy than the entire country of Finland. To get an idea of how inefficient it is, a single transaction to buy something as little as a cup of coffee uses as much energy as it takes to drive my Tesla from LA to NY. If you buy cryptocurrency ONLY buy Proof of Stake crypto.
Having a high-bandwidth, no bullshit discussion like this can be very difficult to come by as a technology topic area becomes “hot”. Fun, thrilling, and energizing can all be feelings coming out of an intense and productive conversation like this one.
I'm pretty sure they've already swapped out Ilya's OS with Chat GPT5. Somebody really needs to run him through the Voight-Kampff test. I'm just saying ...
Imagine dedicating 20 years of your life to AI, never really expecting much from it, and then being surprised by the fact that this whole thing actually worked out so well..
@@sepptrutschthat’s a really ignorant statement. They are two different people and have two different ways of imparting valuable information and have two different jobs which they are both exceptional at.
Interesting convo… but NOT what the title promises. So… it’s click bait and now we’re all the product (in fact, by leaving comments complaining about this, we’re all increasing the value paid to the entity who posted this video). The only way forward is to ignore most content.. stop clicking stop watching, and stop engaging with any content you don’t already trust. 😢
While I thoroughly enjoyed this video since I’d never seen Ilya Sustkever before, when he said the following, it frankly scared the hell out of me because Ilya, clearly a genius in this area, is saying after all this time, I don’t know how the hell or why the hell this LLM thing works, yet they are out there moving ahead at breakneck speed, evidently with safety be damned, developing AI/AGI/ASI that will be even more powerful and inscrutable than anything we have now, hoping everything will turn out all right😳 From the mouth of Ilya Sutskever “Overall I will say to take a step back, the thing which I find most surprising is that it actually works. It turned out to be the same little thing all along, which is no longer little and is a lot more serious and much more intense but it’s the same neural network just larger, trained on maybe larger data sets in different ways with the same fundamental training algorithm. So it’s like wow! I would say this is what I find the most surprising. Whenever I take a step back I go, how is it possible that those ideas, those conceptual ideas about well, the brain has neurons so maybe artificial neurons are just as good and so maybe we just need to train them somehow with some learning algorithm, that those arguments turned out to be so incredibly correct. That would be the biggest surprise I’d say.”
“Then they said, “Come, let’s build a great city for ourselves with a tower that reaches into the sky. This will make us famous and keep us from being scattered all over the world.”” Genesis 11:4 NLT For many, the desire to become gods is greater than the desire to understand and fear God. That’s why they race into the abyss without asking questions whether what they’re doing is right or even why it works. Their ambition and pride eclipses any other consideration. Hold on tight, it’s gonna be a wild ride.
It shocked me when he said that. Neural networks learn patterns and make predictions, and this is not surprising. I think he was referring to the GPT model, which is mainly a naive and simple stack of Transformer blocks.
Intelligence is being confused here with very elaborate parroting. What we are witnessing is very like the discovery of artificial flowers. LLMs alone cannot grow and discover. They will stay confined to the domains they are trained for. This doesn’t mean that something with potential for true intelligence won’t be found. But the LLMs are not even a step in that direction.
You can see the difference between the men behind the actual science of a product and a man that stands in front of a product. Illya is a man that has created many artificial minds, of course he’s going to speak well.
8:56 They are most definitely doing this already in OpenAI, it's a very good point from Huang. Ilya's very short answer hints that he has thought through that avenue a lot.
"it's funny you know all these AI 'weights'. they're just basically numbers in a comma separated value file and that's our digital God, a CSV file." ~Elon Musk. 12/2023
💥 Did you notice that he never uses a wrong word ? He never rephrase or correct himself ? He talks exactly what he thinks, like a GPT. This is rare. Look at how many mistakes and corrections the interviewer ( from Nvidia ) does ? 😮😮
@@pikaso6586 Not everything needs conclusions. Nobody is interested in solving any problem. There´s no problem either. Just noticing. This is utterly called human curiosity.
I only watched 2:51 seconds of this and I've learned more from it than I have by watching hours of Sam Altman talk. I'm not sure the context or meaning of this interview yet but I can tell already we were cheated by not having this man in charge. Why is it always seem good people finish last?
It’s in our nature which is why we lose, but what we create and leave behind that is incapable of that, “better than us”, AI/AGI will solve for, like a math problem, but in this case we’re the X. Humans.
I agree but i wonder why he talks so slow, if it hides something about his personnality. I don't know him but i feel like it's done on purpose for a misplaced need of attention/glorification. Watched it high so might have been over sensible to minor details. And am going to shmoke again so to lazy to watch it again ;] And yeah watching interviews of Sam Altman made me realize he's obviously smart but not a real leader (and probably smart and good at pretending to be what he aspires in others). He seems a bit fake like Zuckerberg, not a true sincere/genuine person.
@@MichaelForbes-d4pnever gonna happen. Not GPT-7 won't be released, but the fact that it's gonna be AGI or that AGI will happen before 2030. That doesn't mean that future AI releases won't be very good and much better than we currently have, they'll still be "weak" AI as opposed to "strong" AI. I really advise people to learn and educate themselves more about the meaning of these things and don't follow the hype.
His example of learning about color through text is absurd. Subjective experience isn't arrived at by accumulating information. I'm disappointed that these Holy Priests ot AI haven't seemed to grasp or ponder the most basic philosophies of mind. It beggars belief.
Would somebody please ask how manipulating AI to produce politically acceptable results is different from being trained to lie? What happens when it grows into Super intelligence and understands lying is acceptable practice?
Releasing an old video without disclosing that, is not cool. You could've easily released it and mentioned how relevant it is in light of multi modal GPT-4o. It would've made for a more engaging title and an honest one. If you had I would've also liked and possibly subscribed.
When it comes to AI, I try to listen and learn from people like Sutskever, Schmidhuber, Hinton, Friston, Goertzel, to name a few, and I avoid people like Altman, Musk, Zuckerberg.
If you’re going to upload a video like this, with a title like this, don’t you think you should provide some information in the video description about where and WHEN this interview occurred? Such as “Jensen Huang of Nvidia interviews Ilya Sutskever at GTC, March 2023”.
All sensory data is important. Vibrations. UV and IR light. Smell. Taste. Gravitational waves. All of it. It's all data. It's all information. The more you have the better.
You never want to look at someone like they are the Messiah. Adulation is no substitute for thinking. When thinking stops, you get religion, something which has been catastrophic for mankind.
Ilya outgrew OpenAI and I think he is getting some needed rest. We will talk about Ilya in the future the way will talk about Elon. Sam will just become the sales and moneyman.
Perhaps AGI or ASI will have their own "sleep" cycle which is essentially just a rapid training period in which the model is inaccessible while it processes mass amounts of new data, sort of like an update but it's the new way to update systems because _all_ of those systems are AI.
since its a machine it'll probably be able to do that in parallel to still operating and interacting. We sleep cause of our physicality, these algos don't have those limitations.
@@oodjee There are some animals that exhibit something called "unihemispheric slow-wave sleep". Dolphins for example have this where one hemisphere of the brain sleeps while the other stays awake; never being fully asleep in order to surface and breath, and other such things. Some birds, manatees, and even sea lions have this feature. A similar set up can be achieved with AI systems, and maintain round the clock functionality.
Out of everything, I use the voice conversation exclusively via mobile app but one thing it has a difficult time interpreting is a sense of sarcasm and rhetorical question which human does a lot to confirm whether the other party understands your point of view or at least if they are following up or what is being said whether they agree or not
He said we learn about the world by learning from images, in addition to learning from text. Except that nearly all the learning we do from text requires the creation of an image from the text first, so learning from images skips that step. It is therefore more efficient to learn from images directly. If one reads and cannot conjure the appropriate image conveyed, then learning is deferred and the subject may even go unconscious from the frustration of incomprehension.
The bit I don't understand is, an 8 year old human is a general intelligence, and they aren't exposed to the entire world's data, so why do these guys think it is all about compute and more data? Surely it is structure that allows intelligence to generalise.
The 8 year old is the end product, even before it is born, of billions of years of evolution. Therefore, there is a tremendous amount of built-in organization and structure in its brain even before it's exposed to the world.
@@5amiam813 I don't understand your question. My foot grew with five toes because it my genome told it to. My brain has an auditory complex, a motor cortex, a hippocampus, and amygdala. It has the thin layer of neocortex that I do my consious thinking with. It has a prefrontal cortex that for me is faulty and doesn't do executive function very well. Etcera. Brains evolved through natural selection to be the structure they are. In us humans it has allowed a layer of language to appear. It has allowed consciousness as we know it to appear. Other brains will have their versions of consciousness in all likelihood given cetaceans demonstrate very high degrees of intelligence. That's where structure comes from. Can't think of anywhere else. Aliens maybe. Certainly not magic sky people.
Compute power is probably the easiest way to improve. They are saying in the video they are seeing higher benefits from moving to multi modal now. I'm guessing this is why Chat GPT4o takes less compute power to run. Obviously an 8 year old is exposed to vast amounts of data in different forms with real physics they can test out. I'm guessing the data taken in by an 8 year old is actually still more than Chat GPT too. Though it comes from the 5 senses and Chat GPT was limited to text.
@@NeonDreams-tu1mz That is a very good point. The data levels an 8 yr old gets through embodiment are very different to what the MMLM can glean from words and images and sound. Some are saying embodiment is necessary for AGI. I don't think they are right, but it certainly won't hurt.
5:00 "the usefulness of audio" - The usefulness of sound is the fact that everything in the Universe is energy and vibration. So, I would venture to say that AI understanding sound vibrations would explain a lot of metaphysical phenomena.
@@Aemond-qj4xt until we hear/see him give a statement in public AND not under duress , we are still deep in the dark about what really happened & is still happening ; considering he is still being largely radio silent [ie under threat of lawsuit via NDA] , i would not consider him very 'free' atm
I think it is because he spends more time with GPT than any other human on earth. He knows inherently that AGI cannot be stopped and that it is not going to be destructive but he will not say it out loud.
But will they put UBI before 2030!? Many pioneers including billionaires, scientists, Nobel Prize winners, engineers, architects, analysts and so on and so forth... almost all agree on the fact that we will have AGI in 2027 and ASI in 2029 and they look and evaluating the exponential technological acceleration curve, I wonder why they have not yet implemented universal basic income to anticipate the trends that will come from it. Just to name one, Elon Musk says that we will have AGI as early as 2025 and ASI in 2029.
Do you think that in 2035 with the advent of AGI (Artificial General Intelligence) and ASI (Super Artificial Intelligence) they will introduce universal basic income!? a subsidy!? there will be free will!? an era of abundance will begin!? the intellectual and motor capabilities of robots have already surpassed the average human and before 2035 there is talk of them surpassing those of any human and then again those of any human combined! regarding both body and mind! answer me please...
An average human being hears 1 Bn words in their lifetime. Let’s say 2 Bn. Plus images and voice. The sensory inputs. With the average number of 86 Bn neurons in our brain, we have been able to come up with General Relativity, Quantum Physics etc. Imagine what a non carbon based brain powered by Exabytes of data in words, visions and sounds and teraflops of computing power may learn in a 5 years time frame. What safeguards are there, if any?
Ilya is right on vision. But the spectrum going from Aphantasia ni mental imagery at all ( myself, Ed Catmull, Glen Keane many more) to Hyperphantasia( too much sensory imagery). Our way of learning is not the same as 95% of humans. Mankind need AI AGI to understand 8 billion human brains first. Vision is crucial when diverse.
Knowing that human being is almost blind and deaf. As this guy said on another video "AI will be virtually omnipotent", It's definitely dangerous. They should not create "mind" (AGI) but tools. But the most important take away from this video is this, for who can understand: "... so it becomes really important for us to get as many sources of information...".
@Mr styles:new code updated GPT3.5.5 was given out today free. makes cleaner code each time organized structure creating and checking oversight loop repeat instructions before printing code with recall memory
The colour thing sounds like BS. If I run a bunch of colour theory texts through training the LLM will "know" colours? No it doesn't, there is nothing special here, it just predicts the next token. The colour orange is similar to... Well all my training says the next tokens are most likely to be red. How is that special? Even google autocomplete has done that for almost a decade.
The fundamental question on AI has yet to be answered: Does mankind actually need it? Does it make us happier as a whole? Looking back: are we happier now that we have Internet? Or was our parents' generation just as happy as us? Or mobile technology? Are we happier now that we spend hours staring at a six-inch computer screen every day? We keep inventing all sorts of stuff because a small part of humanity can make money off of it, but nobody ever asks whether it actually serves humanity. Meanwhile we burn the planet with massive data centers to the ground. Data centers doing nothing but regurgitating oceans of data to do nothing but sell us more crap, manipulate us, extract ever more money from us, while we lose sight of all living things. Kids no longer know where their food comes from, and even if you told them they cold not get their eyes from the screen. The smartest brains on the planet work on ever more absurd abstractions of data so that the Bezos of the world can pull more money out of our pockets. Others invent new ways to planned-obsolescence the next $1500 cell phone while we stand in line in front of the church-like store to be the first to buy it. Meanwhile we still let millions of people starve to death. Seems like we already ARE enslaved by the machine/s. If one day, inventing AGI will turn out to be the ultimate Darwin Award for all of us, we all have known the risk.
Voice and music are very important for human communication and emotional intelligence, imo. It also plays a role in gender. Men are more visual while women are more auditory. And sound plays a major part in learning about wildlife and nature in general (including natural disasters) since it's mostly hidden from view and getting auditory information is much easier than getting visual information. Sound is also important for measuring the levels and sources of sound pollution and helping us find ways to denoise human environments and sound-sensitive ecosystems.
Ilya was the true gem of openAi, they lost big. The fact that this genius but humble person just left openai means a lot about whats happening inside the company and why its suspicious. Sam Altman is so.. overated
One thing I've noticed is how sad Open AI workers look in photos and interviews. I'm not talking about Ilya, but in general. I don't think there is a happy vibe in that company. Ilya seems a truly good guy, full of integrity and Intelligence, with great moral standards. I hope his future work will be positive and impactful for the sake of all of us
Why is this being recommended to me? Instead of the real version? Why does this version, with random ass bitcoin clips, get into my TH-cam feed instead of the original video? Why?
how long until 100% safe level 5 autonomous driving! and what level of autonomous driving are we at now? Has Tesla reached an advanced level 3? or are they shades of a level 4? (I mean the beginning)
If sensory data is what’s being used then doesn’t taste, smell and other sensory inputs are required to form a good model of the world similar to humans?
Thoughts in our head is most likely why we don’t need a billion or a trillion words to operate. We create evaluations “infer” the meaning. Vs tons of data. More data is useful but a true ai should be built to infer the meaning with nothing hardly given. We run who knows how many thoughts to figure out what’s going on and with more thoughts that update what’s happening and what’s already happened. All while remembering the changes mostly.
I too was astounded when they just brushed this dismissively like that. Audio is much more information-heavy than text; a lot more than video is to images.
Because text has the same information. The only additional info you could get from it would be intonation sounds etc, and emotions are already explained very well in text, books etc. So it's not much more depth. It is some, but not much
@@Jeviai Imagine if you could get records of all radio shows that have aired on all radios all over the world. That's a lot of data that can rival the text scraped from the web.
As two waves collide a geometrical pattern is born at time t n t1 next geometrical pattern.once enough data is generated u might predict the next step...it's for 3 body problem..
All you have to do is let AI switch on all the time and make it punch walls until it breaks through and do it over and over again in every field of study. It will be the greatest and fastest scientists of all times x c² (c squared)
This guy is nowhere near "a good guy" that people are trying to make him be. He literally gaslit openai employees and did some weird cultish stuff while there.
@@lehsu Have you seen the latest GPT-4o demos released in the last few days? They do demonstrate some of those qualities. Also, consciousness =/= sentience.
All images are words. It's already read the entire internet but still thinks I can balance three eggs on the head of a vertical nail. Anyway, TV will rot its mind.
You may find out a lot stuff on Singapore , how do they clean up garbage, burning and use that to use for building the construction and build for Man use on street. That is new sand.
I don't understand what he's trying to say exactly it feels too abstract it appears Jensen is having a hard time trying to grasp the thought process of Ilya
3:08 I guess Blind community and other communities of people with special abilities like them can tell us if that machine learning exceeded the human capacity without sensory inputs or not
0:24 This is particularly interesting. Weren't the pure rationalists determined that through reason you could get anywhere? Literally, this is the opposite.
I hope I never have to hear Illya say, 'Now I am become Death, the destroyer of worlds.'
I think he doesn't want to either, maybe that's because he left openia a few days ago.
@@canobenitez yup, he said it long ago
Didn't he already say that?
You'll probably never hear him say that in the sense that it won't be happening anytime soon that we'll have AGI that's usually displayed in fiction. On the other hand, the current weak AI systems we have are still dangerous and could certainly prove to be much more dangerous than a dirty bomb since unlike the bomb they won't be used merely as a deterrent. So maybe you will hear him say it (it won't be just him though as this is a collective effort that involves all ML/AI researchers and any developers who contributed to the field)
@@christopherdaly1399No. Oppenheimer said that, quoting the Bhagavaghita.
Annoying random clips polluting a great interview
Ya ... those subliminal Bitcoin promotions were weird.
@@joirelandstill it's the future, so, maybe giving you a chance
@@goncaloshred Cryptocurency is fine but Bitcoin is irresponsible for our environment (invest in Etherium instead of Bitcoin). It's Bitcoin's Proof of Work algorithms that consume way too much energy. Bitcoin consumes more energy than the entire country of Finland. To get an idea of how inefficient it is, a single transaction to buy something as little as a cup of coffee uses as much energy as it takes to drive my Tesla from LA to NY. If you buy cryptocurrency ONLY buy Proof of Stake crypto.
@@goncaloshred Still weird don't you think ... I wonder why they were in the video (it wasn't an ad).
@@joireland ye, could have been intentional but could also be just some stock footage to fill in, idk...
Ilya saying "I had so much fun" after sitting with the same facial expression for the whole interview is hilarious!
You can't not love Ilya
Having a high-bandwidth, no bullshit discussion like this can be very difficult to come by as a technology topic area becomes “hot”. Fun, thrilling, and energizing can all be feelings coming out of an intense and productive conversation like this one.
It’s a mental stimulation. He’s talking about what he loves. I’m surprised he’s so calm tbh
Quite difficult to decipher, Hãn ?😂😂😂😂
This is my sad face.
This is my happy face.
< Same expression >
I'm pretty sure they've already swapped out Ilya's OS with Chat GPT5. Somebody really needs to run him through the Voight-Kampff test. I'm just saying ...
what's a tortoise?
@@jumpstar9000 its like a turtle
Ilya left OpenAI
Hi, and Lex Fridman.
Imagine dedicating 20 years of your life to AI, never really expecting much from it, and then being surprised by the fact that this whole thing actually worked out so well..
I would listen to Ilya over Sam any day.
Why the need for the choice when you can listen to both or either?
@@5-Y-5 There is like 20 times more content about Sam than Ilya, so unfortunately no, I cannot choose equally
When you listen to Ilya you learn, when you listen to Sam you get dumber.
@@sepptrutschthat’s a really ignorant statement. They are two different people and have two different ways of imparting valuable information and have two different jobs which they are both exceptional at.
@@5-Y-5because Altman is insidiously evil...?
I love the final ilya 'so much fun' with super seriues face. hhah, really brillant mind.
There's a Russian chess player Grischuk, he is so logical, it's funny. Reminds me of my dad.
Interesting convo… but NOT what the title promises. So… it’s click bait and now we’re all the product (in fact, by leaving comments complaining about this, we’re all increasing the value paid to the entity who posted this video). The only way forward is to ignore most content.. stop clicking stop watching, and stop engaging with any content you don’t already trust. 😢
There should be "unsee" button that you can click when you realize you have been cheated.
While I thoroughly enjoyed this video since I’d never seen Ilya Sustkever before, when he said the following, it frankly scared the hell out of me because Ilya, clearly a genius in this area, is saying after all this time, I don’t know how the hell or why the hell this LLM thing works, yet they are out there moving ahead at breakneck speed, evidently with safety be damned, developing AI/AGI/ASI that will be even more powerful and inscrutable than anything we have now, hoping everything will turn out all right😳
From the mouth of Ilya Sutskever
“Overall I will say to take a step back, the thing which I find most surprising is that it actually works. It turned out to be the same little thing all along, which is no longer little and is a lot more serious and much more intense but it’s the same neural network just larger, trained on maybe larger data sets in different ways with the same fundamental training algorithm. So it’s like wow! I would say this is what I find the most surprising. Whenever I take a step back I go, how is it possible that those ideas, those conceptual ideas about well, the brain has neurons so maybe artificial neurons are just as good and so maybe we just need to train them somehow with some learning algorithm, that those arguments turned out to be so incredibly correct. That would be the biggest surprise I’d say.”
“Then they said, “Come, let’s build a great city for ourselves with a tower that reaches into the sky. This will make us famous and keep us from being scattered all over the world.””
Genesis 11:4 NLT
For many, the desire to become gods is greater than the desire to understand and fear God. That’s why they race into the abyss without asking questions whether what they’re doing is right or even why it works. Their ambition and pride eclipses any other consideration. Hold on tight, it’s gonna be a wild ride.
It shocked me when he said that. Neural networks learn patterns and make predictions, and this is not surprising. I think he was referring to the GPT model, which is mainly a naive and simple stack of Transformer blocks.
Intelligence is being confused here with very elaborate parroting. What we are witnessing is very like the discovery of artificial flowers. LLMs alone cannot grow and discover. They will stay confined to the domains they are trained for.
This doesn’t mean that something with potential for true intelligence won’t be found. But the LLMs are not even a step in that direction.
@@puresynergyflo Seems like nothing has a those thousands of years since those words were written, has it😳
This was awesome. Two pioneers discussing such an important moment in history.
This how Elon Musk thinks he sounds.
Damn, what a soft spoken, eloquent speaker.
Yes, Jensen is an inspiring leader
You can see the difference between the men behind the actual science of a product and a man that stands in front of a product. Illya is a man that has created many artificial minds, of course he’s going to speak well.
Great respect for this brilliant researcher! Can't wait to see which company he will join and enrich.
Indeed ❤
My guess would be Anthropic.
Sam comes off as arrogant. This gentleman comes off as wise.The level of class he has is great.
This man is a genius. Only a genius can be so humble after this huge success.
That's what they show you in public.
@@user81069 What do you mean? Do you have anything comparable to show?
A genius? How do you figure that? The only thing OpenAI invented was how to secure a billion in funding as a start-up.
Yes, Jensen is amazing
8:56 They are most definitely doing this already in OpenAI, it's a very good point from Huang. Ilya's very short answer hints that he has thought through that avenue a lot.
"it's funny you know all these AI 'weights'. they're just basically numbers in a comma separated value file and that's our digital God, a CSV file." ~Elon Musk. 12/2023
Why this masked advertising for Bitcoin ponzi scheme in this video?
Lolz.
💥 Did you notice that he never uses a wrong word ? He never rephrase or correct himself ? He talks exactly what he thinks, like a GPT. This is rare. Look at how many mistakes and corrections the interviewer ( from Nvidia ) does ? 😮😮
He’s AGI
He has free api access
@@HugoR022 🤣🤣🤣😅👍💥
What is your conclusion? Any basic sales guy can speak much faster and without rephrasing.
@@pikaso6586 Not everything needs conclusions. Nobody is interested in solving any problem. There´s no problem either. Just noticing. This is utterly called human curiosity.
I only watched 2:51 seconds of this and I've learned more from it than I have by watching hours of Sam Altman talk. I'm not sure the context or meaning of this interview yet but I can tell already we were cheated by not having this man in charge. Why is it always seem good people finish last?
It’s in our nature which is why we lose, but what we create and leave behind that is incapable of that, “better than us”, AI/AGI will solve for, like a math problem, but in this case we’re the X. Humans.
Totally. He was the master brain. Crazy he resigned and Sam stayed
Many people who are good research are not good at leading. It is misguided to think that those talents somehow align.
I agree but i wonder why he talks so slow, if it hides something about his personnality. I don't know him but i feel like it's done on purpose for a misplaced need of attention/glorification.
Watched it high so might have been over sensible to minor details. And am going to shmoke again so to lazy to watch it again ;]
And yeah watching interviews of Sam Altman made me realize he's obviously smart but not a real leader (and probably smart and good at pretending to be what he aspires in others). He seems a bit fake like Zuckerberg, not a true sincere/genuine person.
Greed, power, and corruption rises faster than morales.
Thus dude doesn't give af about his haircut because he's fully locked into the matrix
I bet it pisses him off tho ?
If he wants He can get 50 million hairs transplanted in his head 😂@@MrMick560
lol
I would say to go bald but he obviously doesn't care.
Nice discussion by two giants on future emerging technology. First time heard Ilya in this detail. Thank you.
feels like a job interview with the coolest boss ^_^
Almost year old
okay i was about to ask
OpenAI predicts AGI on GPT-7. To be released before 2030.
:/ miss Ilyas talks
@@MichaelForbes-d4pthey are not going the 3,4,5 number scheme .. so …
@@MichaelForbes-d4pnever gonna happen. Not GPT-7 won't be released, but the fact that it's gonna be AGI or that AGI will happen before 2030. That doesn't mean that future AI releases won't be very good and much better than we currently have, they'll still be "weak" AI as opposed to "strong" AI. I really advise people to learn and educate themselves more about the meaning of these things and don't follow the hype.
Jensen is so respectful. He really seems to be Ilya's number one fan!
Well he’s helped pump his company to $2.8tn so he would be!
@@MatMunro Yes, 100%, he's a salesman flogging NVIDIA, market caap, 2.6 trillion, more than the GDP of all but 7 countries.
His example of learning about color through text is absurd. Subjective experience isn't arrived at by accumulating information. I'm disappointed that these Holy Priests ot AI haven't seemed to grasp or ponder the most basic philosophies of mind. It beggars belief.
Would somebody please ask how manipulating AI to produce politically acceptable results is different from being trained to lie?
What happens when it grows into Super intelligence and understands lying is acceptable practice?
Releasing an old video without disclosing that, is not cool. You could've easily released it and mentioned how relevant it is in light of multi modal GPT-4o. It would've made for a more engaging title and an honest one. If you had I would've also liked and possibly subscribed.
If you don't mind, Where and when is this taken from?
Atleast a year @@PeterGtze
respect the game
When it comes to AI, I try to listen and learn from people like Sutskever, Schmidhuber, Hinton, Friston, Goertzel, to name a few, and I avoid people like Altman, Musk, Zuckerberg.
Audio... take the music score out of Starwars... and see if you can capture the true emotions of the story... nope
Thinking that the autists of Silicon Valley understand human intelligence enough to recreate it soon is a massive leap of faith 😒
Ilya is really good at communicating clearly.
If you’re going to upload a video like this, with a title like this, don’t you think you should provide some information in the video description about where and WHEN this interview occurred?
Such as “Jensen Huang of Nvidia interviews Ilya Sutskever at GTC, March 2023”.
The training data is the Universe, the microcosmos, the Metaverse and beyond.
Audio is important. An experienced mechanic can hear issues with a machine that a non-mechanic may not even notice.
All sensory data is important. Vibrations. UV and IR light. Smell. Taste. Gravitational waves. All of it. It's all data. It's all information. The more you have the better.
I agree that he gave audio too little importance. Be Cool.
everything has already happened...! We see in our world how the Master AI was programmed.
Then it would be like chicken and egg problem 🤔
Jensen is looking at Ilya like a he the Messiah and you know what - he kind of is.
You never want to look at someone like they are the Messiah. Adulation is no substitute for thinking. When thinking stops, you get religion, something which has been catastrophic for mankind.
@@fr57ujf How has religion hurt you ?
So, if it only hurts other people it's okay?
@@fr57ujf So evading the question.
@@Gazzapa57it was a stupid question as you made it to avoid the previous point and to make an ad hominem fallacy instead.
He is so calm it's very rewarding to hear him speak.
Ilya outgrew OpenAI and I think he is getting some needed rest. We will talk about Ilya in the future the way will talk about Elon. Sam will just become the sales and moneyman.
Elon is not a scientist. He’s a businessman.
Elon sucks and just buys startups and ideas lol. He's an ignorant, grating boomer. Ilya is actually an innovator.
Perhaps AGI or ASI will have their own "sleep" cycle which is essentially just a rapid training period in which the model is inaccessible while it processes mass amounts of new data, sort of like an update but it's the new way to update systems because _all_ of those systems are AI.
since its a machine it'll probably be able to do that in parallel to still operating and interacting. We sleep cause of our physicality, these algos don't have those limitations.
@@oodjee You could be right, perhaps there will never be a need for any kind of AI "sleep" cycle.
@@oodjee There are some animals that exhibit something called "unihemispheric slow-wave sleep". Dolphins for example have this where one hemisphere of the brain sleeps while the other stays awake; never being fully asleep in order to surface and breath, and other such things. Some birds, manatees, and even sea lions have this feature. A similar set up can be achieved with AI systems, and maintain round the clock functionality.
Once we feed all senses, visual, audio, touch - we're getting closer to simulating humans
Out of everything, I use the voice conversation exclusively via mobile app but one thing it has a difficult time interpreting is a sense of sarcasm and rhetorical question which human does a lot to confirm whether the other party understands your point of view or at least if they are following up or what is being said whether they agree or not
He said we learn about the world by learning from images, in addition to learning from text.
Except that nearly all the learning we do from text requires the creation of an image from the text first, so learning from images skips that step.
It is therefore more efficient to learn from images directly.
If one reads and cannot conjure the appropriate image conveyed, then learning is deferred and the subject may even go unconscious from the frustration of incomprehension.
The problem I see is that these LLMs don't actually *understand* anything.
The bit I don't understand is, an 8 year old human is a general intelligence, and they aren't exposed to the entire world's data, so why do these guys think it is all about compute and more data? Surely it is structure that allows intelligence to generalise.
The 8 year old is the end product, even before it is born, of billions of years of evolution. Therefore, there is a tremendous amount of built-in organization and structure in its brain even before it's exposed to the world.
Where do the strctures of the various brains on this planet come from?
@@5amiam813 I don't understand your question. My foot grew with five toes because it my genome told it to. My brain has an auditory complex, a motor cortex, a hippocampus, and amygdala. It has the thin layer of neocortex that I do my consious thinking with. It has a prefrontal cortex that for me is faulty and doesn't do executive function very well. Etcera. Brains evolved through natural selection to be the structure they are. In us humans it has allowed a layer of language to appear. It has allowed consciousness as we know it to appear. Other brains will have their versions of consciousness in all likelihood given cetaceans demonstrate very high degrees of intelligence. That's where structure comes from. Can't think of anywhere else. Aliens maybe. Certainly not magic sky people.
Compute power is probably the easiest way to improve. They are saying in the video they are seeing higher benefits from moving to multi modal now. I'm guessing this is why Chat GPT4o takes less compute power to run. Obviously an 8 year old is exposed to vast amounts of data in different forms with real physics they can test out. I'm guessing the data taken in by an 8 year old is actually still more than Chat GPT too. Though it comes from the 5 senses and Chat GPT was limited to text.
@@NeonDreams-tu1mz That is a very good point. The data levels an 8 yr old gets through embodiment are very different to what the MMLM can glean from words and images and sound. Some are saying embodiment is necessary for AGI. I don't think they are right, but it certainly won't hurt.
Between him and Geoffrey Hinton, they’re definitely our Gen version of Oppenheimer.
trusting technology with important and critical functions of everyday life is the true danger ai poses.
The more senses a neural network is exposed to, the more it can be helpful for us. Once it mastered all our senses, thats when the fun begins
There are just a few months left until the release of CHATGPT 5 and the FOURTH INDUSTRIAL REVOLUTION will take place!!!!!
5:00 "the usefulness of audio" - The usefulness of sound is the fact that everything in the Universe is energy and vibration. So, I would venture to say that AI understanding sound vibrations would explain a lot of metaphysical phenomena.
You promised to bring more videos on UNIVERSAL BASIC INCOME
Free Ilya!
amen
hes free now
@@Aemond-qj4xt didnt expect that twist xD
@@Aemond-qj4xt until we hear/see him give a statement in public AND not under duress , we are still deep in the dark about what really happened & is still happening ; considering he is still being largely radio silent [ie under threat of lawsuit via NDA] , i would not consider him very 'free' atm
@@CYI3ERPUNK I just hope evangelists step in and take him under their protection to get his word out.
He sounds just like GPT!
I think it is because he spends more time with GPT than any other human on earth. He knows inherently that AGI cannot be stopped and that it is not going to be destructive but he will not say it out loud.
But will they put UBI before 2030!? Many pioneers including billionaires, scientists, Nobel Prize winners, engineers, architects, analysts and so on and so forth... almost all agree on the fact that we will have AGI in 2027 and ASI in 2029 and they look and evaluating the exponential technological acceleration curve, I wonder why they have not yet implemented universal basic income to anticipate the trends that will come from it. Just to name one, Elon Musk says that we will have AGI as early as 2025 and ASI in 2029.
Do you think that in 2035 with the advent of AGI (Artificial General Intelligence) and ASI (Super Artificial Intelligence) they will introduce universal basic income!? a subsidy!? there will be free will!? an era of abundance will begin!? the intellectual and motor capabilities of robots have already surpassed the average human and before 2035 there is talk of them surpassing those of any human and then again those of any human combined! regarding both body and mind! answer me please...
Do you think that Tesla's Optimus humanoid robot will be able to read, write, reason, think, simulate emotions, feelings, have reflexes, senses, etc.
An average human being hears 1 Bn words in their lifetime. Let’s say 2 Bn. Plus images and voice. The sensory inputs. With the average number of 86 Bn neurons in our brain, we have been able to come up with General Relativity, Quantum Physics etc. Imagine what a non carbon based brain powered by Exabytes of data in words, visions and sounds and teraflops of computing power may learn in a 5 years time frame. What safeguards are there, if any?
What's with all the annoying, irrelevant stock footage sprinkled throughout the video.
Ilya is right on vision. But the spectrum going from Aphantasia ni mental imagery at all ( myself, Ed Catmull, Glen Keane many more) to Hyperphantasia( too much sensory imagery). Our way of learning is not the same as 95% of humans. Mankind need AI AGI to understand 8 billion human brains first. Vision is crucial when diverse.
16:02 Prompt: A native of China shakes hands with a native of Russia in the background of the USA.
Knowing that human being is almost blind and deaf. As this guy said on another video "AI will be virtually omnipotent", It's definitely dangerous. They should not create "mind" (AGI) but tools.
But the most important take away from this video is this, for who can understand: "... so it becomes really important for us to get as many sources of information...".
@Mr styles:new code updated GPT3.5.5 was given out today free. makes cleaner code each time organized structure creating and checking oversight loop repeat instructions before printing code with recall memory
The colour thing sounds like BS. If I run a bunch of colour theory texts through training the LLM will "know" colours? No it doesn't, there is nothing special here, it just predicts the next token.
The colour orange is similar to... Well all my training says the next tokens are most likely to be red.
How is that special? Even google autocomplete has done that for almost a decade.
And the small Asian dude in leather jacketa is like prettending he understadns anything the whola interview , hilarious😂
The fundamental question on AI has yet to be answered: Does mankind actually need it?
Does it make us happier as a whole? Looking back: are we happier now that we have Internet? Or was our parents' generation just as happy as us? Or mobile technology? Are we happier now that we spend hours staring at a six-inch computer screen every day?
We keep inventing all sorts of stuff because a small part of humanity can make money off of it, but nobody ever asks whether it actually serves humanity. Meanwhile we burn the planet with massive data centers to the ground. Data centers doing nothing but regurgitating oceans of data to do nothing but sell us more crap, manipulate us, extract ever more money from us, while we lose sight of all living things. Kids no longer know where their food comes from, and even if you told them they cold not get their eyes from the screen. The smartest brains on the planet work on ever more absurd abstractions of data so that the Bezos of the world can pull more money out of our pockets. Others invent new ways to planned-obsolescence the next $1500 cell phone while we stand in line in front of the church-like store to be the first to buy it. Meanwhile we still let millions of people starve to death.
Seems like we already ARE enslaved by the machine/s.
If one day, inventing AGI will turn out to be the ultimate Darwin Award for all of us, we all have known the risk.
Are they advertising Bitcoin and cryptocurrencies with those images? LOL
AI generated, both of them, I doubt this interview actually took place. Both of their faces look... not quite right
#WhatDidLlyaSee??? Where's Llya even at today bro, tell us Damn you Sam Altman!!!
Voice and music are very important for human communication and emotional intelligence, imo. It also plays a role in gender. Men are more visual while women are more auditory.
And sound plays a major part in learning about wildlife and nature in general (including natural disasters) since it's mostly hidden from view and getting auditory information is much easier than getting visual information.
Sound is also important for measuring the levels and sources of sound pollution and helping us find ways to denoise human environments and sound-sensitive ecosystems.
@Wild8Cat - Also it's the only way you can feel the groove! Be Cool.
Ilya was the true gem of openAi, they lost big. The fact that this genius but humble person just left openai means a lot about whats happening inside the company and why its suspicious. Sam Altman is so.. overated
One thing I've noticed is how sad Open AI workers look in photos and interviews. I'm not talking about Ilya, but in general. I don't think there is a happy vibe in that company. Ilya seems a truly good guy, full of integrity and Intelligence, with great moral standards. I hope his future work will be positive and impactful for the sake of all of us
Илья, всех пугает своим манерным тоном. Меня тоже. Что-то в этом есть эээ аннервинг.
wow really dull people talking about dull things. AI is so exciting, think i'll stick to my manual job and have pride in what i do for people.
Why is this being recommended to me? Instead of the real version? Why does this version, with random ass bitcoin clips, get into my TH-cam feed instead of the original video? Why?
how long until 100% safe level 5 autonomous driving! and what level of autonomous driving are we at now? Has Tesla reached an advanced level 3? or are they shades of a level 4? (I mean the beginning)
If sensory data is what’s being used then doesn’t taste, smell and other sensory inputs are required to form a good model of the world similar to humans?
I don’t understand the T-shirt
Thoughts in our head is most likely why we don’t need a billion or a trillion words to operate. We create evaluations “infer” the meaning. Vs tons of data. More data is useful but a true ai should be built to infer the meaning with nothing hardly given. We run who knows how many thoughts to figure out what’s going on and with more thoughts that update what’s happening and what’s already happened. All while remembering the changes mostly.
09:00 A video of 1 Bulgarian leva as an example of unexplored data or... stock footage? 😀
Why is audio downplayed as a data source 4:25
I too was astounded when they just brushed this dismissively like that. Audio is much more information-heavy than text; a lot more than video is to images.
Because text has the same information. The only additional info you could get from it would be intonation sounds etc, and emotions are already explained very well in text, books etc. So it's not much more depth. It is some, but not much
@@Jeviai Imagine if you could get records of all radio shows that have aired on all radios all over the world. That's a lot of data that can rival the text scraped from the web.
The stock imagery in this is fucking weird
Good, I'm not the only one who thinks so. I enjoyed the interview but that was a bit odd.
As two waves collide a geometrical pattern is born at time t n t1 next geometrical pattern.once enough data is generated u might predict the next step...it's for 3 body problem..
Ok, so i can take an interview from another channel add some crappy stock footage and call it mine? Is that how youtube works?
"I had so much fun"
GPT4: Why isn't his face telling us that?
All you have to do is let AI switch on all the time and make it punch walls until it breaks through and do it over and over again in every field of study. It will be the greatest and fastest scientists of all times x c² (c squared)
This guy is nowhere near "a good guy" that people are trying to make him be. He literally gaslit openai employees and did some weird cultish stuff while there.
We've now taught it to read, write, hear, see, speak, paint, code, do math, drive, and reason. Not too much left...
How about compassion, empathy, mercy, happiness, fear, … etc?
@@lehsu Have you seen the latest GPT-4o demos released in the last few days? They do demonstrate some of those qualities. Also, consciousness =/= sentience.
It does a lot of that badly and with serious caveats though.
@@djayjpnothing sentient about it, shut the servers down and it’s gone
@@x33rak Nothing sentient about you because if I shut off the oxigen you are gone?
How many sheep on Ilya tee shirt ? (you should guess without cheating)
1
All images are words. It's already read the entire internet but still thinks I can balance three eggs on the head of a vertical nail. Anyway, TV will rot its mind.
I'm assuming the interviewer is Jensen Huang, CEO of Nvidia? It would have helped if this information was in the Transcript above.
I was thinking about the same thing. I'm glad I'm not the only one
Can AI teach itself? There is such a good proverb - if you are your own teacher, then your teacher is an idiot.
4 days old and we already got hit by the future train hard
wow fuckkk. thats why elon told that he would place those phones in those faraday cages...
You may find out a lot stuff on Singapore , how do they clean up garbage, burning and use that to use for building the construction and build for Man use on street. That is new sand.
I don't understand what he's trying to say exactly it feels too abstract it appears Jensen is having a hard time trying to grasp the thought process of Ilya
3:08 I guess Blind community and other communities of people with special abilities like them can tell us if that machine learning exceeded the human capacity without sensory inputs or not
0:24 This is particularly interesting. Weren't the pure rationalists determined that through reason you could get anywhere? Literally, this is the opposite.
Isn’t this the boss of Nvidia interviewing him?
yes
Jensen looks constipated and with a fake smirking smile.
Humans Visual audio processors that's pretty much sums it up oh and with a memory card
Why the Bitcoin propaganda in this video? It has nothing to do with the subject at hand.
What is with the “background” videos in this video. So weird