Загрузка видео...

Не удалось загрузить видео

Возникла проблема при загрузке этого видео. Это может быть связано с временными проблемами сети или видео может быть недоступно.

На главную

Ilya Sutskever: Humans aren’t AGI; they know little and learn constantly. Pretrained AGI on the other hand, overshoots by skipping that trial-and-error phase. Real intelligence comes from ongoing learning, not from starting out “finished.”

Rohan Paul

154,351 subscribers

196,878 просмотров • 20 дней назад •via X (Twitter)

Наука и технологии Образование

Anya Rossi• Live Now

Private livecam show

Комментарии: 0

Нет доступных комментариев

Здесь появятся комментарии из оригинального поста

Похожие видео

this was the most important bit of the ilya interview imo. humans are not AGI, and LLMs won’t become like humans with more scaling. humans lack a huge amount of knowledge and have insanely good “continual learning” powered by extremely good reward functions (except for addiction) that guide them. and so it’s more about the learning rather than knowledge. the term AGI was coined as a response to the narrow AI which was only capable of intelligence in a specific domain (chess, game, code, etc) but General intelligence is able to learn different topics because that’s what “learning” is about. pre-training was thought to be the solution to reaching AGI because you train AI on so many different topics. but they have “overshot” that target. amazing perspective.

this was the most important bit of the ilya interview imo. humans are not AGI, and LLMs won’t become like humans with more scaling. humans lack a huge amount of knowledge and have insanely good “continual learning” powered by extremely good reward functions (except for addiction) that guide them. and so it’s more about the learning rather than knowledge. the term AGI was coined as a response to the narrow AI which was only capable of intelligence in a specific domain (chess, game, code, etc) but General intelligence is able to learn different topics because that’s what “learning” is about. pre-training was thought to be the solution to reaching AGI because you train AI on so many different topics. but they have “overshot” that target. amazing perspective.

ℏεsam

53,444 просмотров • 8 месяцев назад

Demis on the breakthroughs needed for AGI. It is not many. Predict Google DeepMind has Continual Learning before 2026 is finished, and likely other labs too.

Demis on the breakthroughs needed for AGI. It is not many. Predict Google DeepMind has Continual Learning before 2026 is finished, and likely other labs too.

Dan McAteer

38,011 просмотров • 7 месяцев назад

Francois Chollet says LLMs aren't enough for human-like continual learning They store skills as vector programs learned via gradient descent — not efficient, not adaptive True AGI needs to learn from experience and generalize fast "LLMs can be part of AGI, but not the solution"

Francois Chollet says LLMs aren't enough for human-like continual learning They store skills as vector programs learned via gradient descent — not efficient, not adaptive True AGI needs to learn from experience and generalize fast "LLMs can be part of AGI, but not the solution"

Haider.

50,251 просмотров • 9 месяцев назад

Ilya Sutskever says pre-training has overshot the target because humans themselves are not true AGIs We rely on continual learning, not massive upfront knowledge So the real question is what kind of superintelligence we build: "a 15‑year‑old genius eager to learn, or something more finished"

Ilya Sutskever says pre-training has overshot the target because humans themselves are not true AGIs We rely on continual learning, not massive upfront knowledge So the real question is what kind of superintelligence we build: "a 15‑year‑old genius eager to learn, or something more finished"

Haider.

114,811 просмотров • 8 месяцев назад

Demis Hassabis points out that today’s models don’t learn continuously, they are trained once and then deployed. Unlike humans, they don’t improve from experience or adapt after release. The next major leap, like “AlphaZero,” would involve self-learning systems that discover knowledge on their own. This ability to learn online and evolve in real time is a key missing component on the path to AGI. Solving this will be critical to building truly intelligent systems.

Demis Hassabis points out that today’s models don’t learn continuously, they are trained once and then deployed. Unlike humans, they don’t improve from experience or adapt after release. The next major leap, like “AlphaZero,” would involve self-learning systems that discover knowledge on their own. This ability to learn online and evolve in real time is a key missing component on the path to AGI. Solving this will be critical to building truly intelligent systems.

Wes Roth

39,238 просмотров • 8 месяцев назад

Demis Hassabis says the next AI leap is continual learning. Not train once and freeze, but learn from real experience, constantly. AlphaGo did it in games. Now comes the messy real world. Would you let an AI learn from you? 👀

Demis Hassabis says the next AI leap is continual learning. Not train once and freeze, but learn from real experience, constantly. AlphaGo did it in games. Now comes the messy real world. Would you let an AI learn from you? 👀

VraserX e/acc

19,202 просмотров • 6 месяцев назад

Jensen Huang on "distillation" On his new interview with axios, he was asked this question "Should open source model companies be allowed to distill closed models" "Distillation—learning from AI, learning from other people, and learning from other sources of knowledge, is fundamental to intelligence. We are constantly learning from other people. I am learning from you through the questions you are asking, and you are learning from me. All day long, we are learning from one another. AI also has to learn from something. The original AI models, whether they were open or closed, were trained on previously created knowledge from the internet. Now, AI is generating more content than humans. In a few more years, the internet could be 99% AI-generated content, and that content will have been created by some form of AI. As a result, AI systems will constantly be distilling knowledge and intelligence from other AI systems. The fact that AI can learn is a good thing. We want AI systems to be intelligent because a smarter AI can also be a safer AI." ---- From "Axios" YouTube channel, (full video link in comment)

Jensen Huang on "distillation" On his new interview with axios, he was asked this question "Should open source model companies be allowed to distill closed models" "Distillation—learning from AI, learning from other people, and learning from other sources of knowledge, is fundamental to intelligence. We are constantly learning from other people. I am learning from you through the questions you are asking, and you are learning from me. All day long, we are learning from one another. AI also has to learn from something. The original AI models, whether they were open or closed, were trained on previously created knowledge from the internet. Now, AI is generating more content than humans. In a few more years, the internet could be 99% AI-generated content, and that content will have been created by some form of AI. As a result, AI systems will constantly be distilling knowledge and intelligence from other AI systems. The fact that AI can learn is a good thing. We want AI systems to be intelligent because a smarter AI can also be a safer AI." ---- From "Axios" YouTube channel, (full video link in comment)

Rohan Paul

441,746 просмотров • 23 дней назад

Ilya Sutskever, OpenAI co-founder, on why the company that builds AGI shouldn't be chasing infinite profits: Speaking on the No Priors podcast, Ilya Sutskever turns a question on the very technology he's working to build. "If you believe that the technology that we are building, AGI, could potentially be so capable as to do every single task that people do... does it mean that it might unemploy everyone? Well, I don't know, but it's not impossible." From that possibility, Ilya Sutskever draws a conclusion about how such a company should be structured: "And if that's the case, it makes sense. It will make a lot of sense if the company that builds such a technology would not be incentivized to make infinite profits." But he doesn't present this as a settled view. He immediately points to the force that complicates it — competition: "I don't know if it will literally play out this way because of competition in AI. So there will be multiple companies and I think that will have some unforeseen implications on the argument which I'm making. But that was the thinking." His point is simple: if AGI can do everything a person can, it might put people out of work on a massive scale, and that changes things.

Ilya Sutskever, OpenAI co-founder, on why the company that builds AGI shouldn't be chasing infinite profits: Speaking on the No Priors podcast, Ilya Sutskever turns a question on the very technology he's working to build. "If you believe that the technology that we are building, AGI, could potentially be so capable as to do every single task that people do... does it mean that it might unemploy everyone? Well, I don't know, but it's not impossible." From that possibility, Ilya Sutskever draws a conclusion about how such a company should be structured: "And if that's the case, it makes sense. It will make a lot of sense if the company that builds such a technology would not be incentivized to make infinite profits." But he doesn't present this as a settled view. He immediately points to the force that complicates it — competition: "I don't know if it will literally play out this way because of competition in AI. So there will be multiple companies and I think that will have some unforeseen implications on the argument which I'm making. But that was the thinking." His point is simple: if AGI can do everything a person can, it might put people out of work on a massive scale, and that changes things.

Big Brain AI

13,398 просмотров • 2 месяцев назад

Elon Musk on how AGI might actually emerge: “We’re not actually specifically focused on AGI. I’m simply saying that AGI seems likely to be an emergent property of what we’re doing” The reason is the massive real-world data Tesla is generating from autonomous cars and humanoid robots This is not internet data. This is real-world interaction driving on messy roads, navigating unpredictable environments, and interacting with people With tens or hundreds of millions of autonomous vehicles and potentially even more humanoids, the amount of real-world video and data being processed will be enormous From this, cars will become far better than human drivers, and humanoid robots will become increasingly indistinguishable from humans AGI might not come from a lab trying to build it directly. It may simply emerge from building useful robots at massive scale

Elon Musk on how AGI might actually emerge: “We’re not actually specifically focused on AGI. I’m simply saying that AGI seems likely to be an emergent property of what we’re doing” The reason is the massive real-world data Tesla is generating from autonomous cars and humanoid robots This is not internet data. This is real-world interaction driving on messy roads, navigating unpredictable environments, and interacting with people With tens or hundreds of millions of autonomous vehicles and potentially even more humanoids, the amount of real-world video and data being processed will be enormous From this, cars will become far better than human drivers, and humanoid robots will become increasingly indistinguishable from humans AGI might not come from a lab trying to build it directly. It may simply emerge from building useful robots at massive scale

X Freeze

21,256 просмотров • 2 месяцев назад

The 18th Annual AGI Conference (AGI-25) took place last month at Reykjavik University in Iceland, bringing together participants from over 80 countries to explore pioneering research on the path to AGI. AGI-25 programming featured distinguished keynote presentations from leaders in academia and industry, including Ben Goertzel, Richard Sutton, Tatiana Shavrina, Michael Levin, Anna Ciaunica PhD @annaciaunica.bsky.social, and Joscha Bach. These talks addressed core challenges in AGI theory and implementation, the limitations of current state-of-the-art learning systems, and new frameworks for integrating reasoning, embodiment, and autonomy. The accepted papers covered a broad range of research areas, including novel learning algorithms, reasoning systems, theoretical neurobiology and bio-inspired systems, quantum computing, theories of machine consciousness, ethics, safety, formal mathematical foundations, and philosophy of AGI. The conversation on advancing general machine intelligence continues toward AGI-26.

The 18th Annual AGI Conference (AGI-25) took place last month at Reykjavik University in Iceland, bringing together participants from over 80 countries to explore pioneering research on the path to AGI. AGI-25 programming featured distinguished keynote presentations from leaders in academia and industry, including Ben Goertzel, Richard Sutton, Tatiana Shavrina, Michael Levin, Anna Ciaunica PhD @annaciaunica.bsky.social, and Joscha Bach. These talks addressed core challenges in AGI theory and implementation, the limitations of current state-of-the-art learning systems, and new frameworks for integrating reasoning, embodiment, and autonomy. The accepted papers covered a broad range of research areas, including novel learning algorithms, reasoning systems, theoretical neurobiology and bio-inspired systems, quantum computing, theories of machine consciousness, ethics, safety, formal mathematical foundations, and philosophy of AGI. The conversation on advancing general machine intelligence continues toward AGI-26.

SingularityNET

32,589 просмотров • 11 месяцев назад

ELON: AGI IS COMING FAST, AND IT WON’T STOP AT HUMAN LEVELS Elon says the timeline isn’t speculative anymore. What once sounded distant now looks imminent. He’s been consistent on this point for years, and his confidence hasn’t softened. The bigger shift, he argues, comes right after AGI, when intelligence stops being additive and starts compounding. “I think we’ll hit AGI next year, in ’26… Yeah, I’ve said that for a while… I’m confident by 2030 AI will exceed the intelligence of all humans combined.” Source: Moonshots Peter H. Diamandis, MD

ELON: AGI IS COMING FAST, AND IT WON’T STOP AT HUMAN LEVELS Elon says the timeline isn’t speculative anymore. What once sounded distant now looks imminent. He’s been consistent on this point for years, and his confidence hasn’t softened. The bigger shift, he argues, comes right after AGI, when intelligence stops being additive and starts compounding. “I think we’ll hit AGI next year, in ’26… Yeah, I’ve said that for a while… I’m confident by 2030 AI will exceed the intelligence of all humans combined.” Source: Moonshots Peter H. Diamandis, MD

Mario Nawfal

72,477 просмотров • 7 месяцев назад

📁 Yann LeCun says that AGI does not exist and is a poorly defined concept built around an idealized view of human intelligence. He argues that humans are highly specialized, good at navigating the physical and social world, but weak at many other tasks. We seem general only because we can imagine the problems we are good at. For LeCun, AGI mistakes human limits for universal laws of intelligence.

📁 Yann LeCun says that AGI does not exist and is a poorly defined concept built around an idealized view of human intelligence. He argues that humans are highly specialized, good at navigating the physical and social world, but weak at many other tasks. We seem general only because we can imagine the problems we are good at. For LeCun, AGI mistakes human limits for universal laws of intelligence.

Jon Hernandez

28,104 просмотров • 8 месяцев назад

📁 Ben Goertzel, AGI researcher and founder of SingularityNET, says human level AGI will arrive earlier than expected. Not in 2029, but between 2026 and 2028. That moment will trigger massive job obsolescence, not smoothly, but unevenly and chaotically across industries. Psychological and institutional resistance will not stop it, only distort the pattern. The real inflection comes next. A human level AGI can improve itself through software and accelerate further by designing its own hardware.

📁 Ben Goertzel, AGI researcher and founder of SingularityNET, says human level AGI will arrive earlier than expected. Not in 2029, but between 2026 and 2028. That moment will trigger massive job obsolescence, not smoothly, but unevenly and chaotically across industries. Psychological and institutional resistance will not stop it, only distort the pattern. The real inflection comes next. A human level AGI can improve itself through software and accelerate further by designing its own hardware.

Jon Hernandez

21,716 просмотров • 6 месяцев назад

Google DeepMind say true AGI will “reason, adapt, and learn continuously,” and they estimate it’s about 5-10 years away. Meta’s chief scientist Yann LeCun likewise views continual learning as a key pillar of human level AI and is optimistic about achieving it by around 2030 Currently, large language models only improve via offline updates (fine-tuning or reinforcement learning from human feedback) rather than meta learning or modular networks that update themselves gradually. Thankfully All major research teams (OpenAI, DeepMind, Anthropic, Meta, etc.) are actively working on these directions, aiming for models that continuously learn from mistakes as they interact a capability they believe will emerge as we approach true AGI later this decade 2029

Google DeepMind say true AGI will “reason, adapt, and learn continuously,” and they estimate it’s about 5-10 years away. Meta’s chief scientist Yann LeCun likewise views continual learning as a key pillar of human level AI and is optimistic about achieving it by around 2030 Currently, large language models only improve via offline updates (fine-tuning or reinforcement learning from human feedback) rather than meta learning or modular networks that update themselves gradually. Thankfully All major research teams (OpenAI, DeepMind, Anthropic, Meta, etc.) are actively working on these directions, aiming for models that continuously learn from mistakes as they interact a capability they believe will emerge as we approach true AGI later this decade 2029

Chris

284,206 просмотров • 10 месяцев назад

Lots of people think it's obvious that humans learn by imitation, starting with babies imitating their parents. Richard Sutton thinks this is completely wrong. Sutton basically thinks direct imitation plays almost no role in animal learning, and at most a small role in human intelligence. When one animal ends up doing the same thing as another, he thinks this is the result of RL-style trial-and-error, not direct imitation. And if imitation isn't really a kind of learning, then training LLMs to imitate humans isn't really Bitter Lesson–pilled.

Lots of people think it's obvious that humans learn by imitation, starting with babies imitating their parents. Richard Sutton thinks this is completely wrong. Sutton basically thinks direct imitation plays almost no role in animal learning, and at most a small role in human intelligence. When one animal ends up doing the same thing as another, he thinks this is the result of RL-style trial-and-error, not direct imitation. And if imitation isn't really a kind of learning, then training LLMs to imitate humans isn't really Bitter Lesson–pilled.

Dwarkesh Patel

55,137 просмотров • 2 месяцев назад

AGI risk seems lower than it is because the people who know that AGI risk exists are almost all incentivized against talking about it openly. I just made a YouTube video about this that (I think) explains why. Broadly, there are two groups of people not talking about AGI risk: -- People who know AGI risk is real (i.e. AGI lab leaders) -- People who know nothing about AGI risk (i.e. politicians, citizens) To catalyze wide scale discourse about AGI risk, I argue that the following two strategies are strong candidates: 1. [Bottom-Up] Get Citizens Concerned: Find ways to meld AGI risk narratives into legacy media and more understandable political talking points which are already in the Overton window. This work is bound to be messy, and to ultimately soil the core of the message with left-right political gunk, but (barring an AI disaster) its likely the only way that AGI risk catches on with enough of a core base of citizens. As I explain in the video essay, the citizens are the lynchpin to getting everyone (those who don’t know, and those who know) to discuss AGI risk more frankly. 2. [Top-Down] Get a Losing AGI Lab Leader to “Flip”: Those closest to achieving AGI are not going to flip and start talking about AGI risk, the rewards are too great. But those who are losing the race (and don’t want to live to see their rivals achieve the final flex before them) might be able to feign virtue by claiming “Now I see AGI is dangerous, I’m a concerned expert and this needs to be regulated!” They can cloak themselves in pretended virtue while also preventing a rival from crossing the finish line first. I argue that hoping for more Jeff Hintons and Daniel Kokotajlo-like people isn't a good strategy when the incentives are still structured to keep most of them from wanting to talk about it. Here’s the full video essay:

AGI risk seems lower than it is because the people who know that AGI risk exists are almost all incentivized against talking about it openly. I just made a YouTube video about this that (I think) explains why. Broadly, there are two groups of people not talking about AGI risk: -- People who know AGI risk is real (i.e. AGI lab leaders) -- People who know nothing about AGI risk (i.e. politicians, citizens) To catalyze wide scale discourse about AGI risk, I argue that the following two strategies are strong candidates: 1. [Bottom-Up] Get Citizens Concerned: Find ways to meld AGI risk narratives into legacy media and more understandable political talking points which are already in the Overton window. This work is bound to be messy, and to ultimately soil the core of the message with left-right political gunk, but (barring an AI disaster) its likely the only way that AGI risk catches on with enough of a core base of citizens. As I explain in the video essay, the citizens are the lynchpin to getting everyone (those who don’t know, and those who know) to discuss AGI risk more frankly. 2. [Top-Down] Get a Losing AGI Lab Leader to “Flip”: Those closest to achieving AGI are not going to flip and start talking about AGI risk, the rewards are too great. But those who are losing the race (and don’t want to live to see their rivals achieve the final flex before them) might be able to feign virtue by claiming “Now I see AGI is dangerous, I’m a concerned expert and this needs to be regulated!” They can cloak themselves in pretended virtue while also preventing a rival from crossing the finish line first. I argue that hoping for more Jeff Hintons and Daniel Kokotajlo-like people isn't a good strategy when the incentives are still structured to keep most of them from wanting to talk about it. Here’s the full video essay:

Daniel Faggella

26,975 просмотров • 1 год назад

Elon Musk: "I think we'll hit AGI next year, in '26" "I'm confident by 2030, AI will exceed the intelligence of all humans combined" And that's way pessimistic. If you hit AGI next year and that date is in flux but from that date to self-improvements that are on the order of 1,000 to 10,000x just algorithmic improvements... is very short So why is nobody talking about this right now? on 𝕏 they are..... everyday basically I'll tell you something that most people in the AI community don't yet understand Almost no one understands this. The intelligence density potential is vastly greater than what we're currently experiencing

Elon Musk: "I think we'll hit AGI next year, in '26" "I'm confident by 2030, AI will exceed the intelligence of all humans combined" And that's way pessimistic. If you hit AGI next year and that date is in flux but from that date to self-improvements that are on the order of 1,000 to 10,000x just algorithmic improvements... is very short So why is nobody talking about this right now? on 𝕏 they are..... everyday basically I'll tell you something that most people in the AI community don't yet understand Almost no one understands this. The intelligence density potential is vastly greater than what we're currently experiencing

X Freeze

33,585 просмотров • 7 месяцев назад

.Dwarkesh Patel says AGI won’t reshape the world by getting smarter. It will when it masters the tasks everyday people do. "A chatbot cannot edit videos for me." PhD-level intelligence doesn’t drive the economy. The real leverage comes from automating the mundane. Once AI can do that and scale itself, we’re looking at 20% growth and a self-compounding world.

.Dwarkesh Patel says AGI won’t reshape the world by getting smarter. It will when it masters the tasks everyday people do. "A chatbot cannot edit videos for me." PhD-level intelligence doesn’t drive the economy. The real leverage comes from automating the mundane. Once AI can do that and scale itself, we’re looking at 20% growth and a self-compounding world.

a16z

56,914 просмотров • 1 год назад