正在加载视频...

视频加载失败

加载此视频时出现问题。这可能是由于临时网络问题，或视频可能不可用。

Geoffrey Hinton explains that large language models are nothing like traditional software written line by line. Instead of explicit instructions, they rely on code that teaches them how to learn from data. What actually emerges is billions or trillions of learned connection strengths that no one can directly interpret.... show more

Wes Roth

35,340 subscribers

49,600 次观看 • 7 个月前 •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

0 条评论

暂无评论

原始帖子的评论将显示在这里

相关视频

Geoffrey Hinton says LLMs work like the human brain, not like normal computer software They aren't built on explainable lines of code, but rather on trillions of connection strengths that learn from data "it's largely a mystery" We don't know what the individual neurons are doing

Geoffrey Hinton says LLMs work like the human brain, not like normal computer software They aren't built on explainable lines of code, but rather on trillions of connection strengths that learn from data "it's largely a mystery" We don't know what the individual neurons are doing

Haider.

35,431 次观看 • 7 个月前

Over the last few months, we’ve been thinking about how to learn from “off-domain” data - data from non-robot sources like video or simulation. These data sources are not quite good enough to learn policies (even monolithic VLA models) directly, but they still contain lots of information that can be useful for generalizable robot control. How can we develop robot learning models that are able to make use of this type of data for generalizable control? In new work, that we call HAMSTER, we show that VLMs can be useful for enabling robotic learning from off-domain data, but specifically when used through hierarchical VLA architectures. We show that this class of models can learn generalizable robot policies for the real world from large-scale, off-domain data. A 🧵 (1/10)

Over the last few months, we’ve been thinking about how to learn from “off-domain” data - data from non-robot sources like video or simulation. These data sources are not quite good enough to learn policies (even monolithic VLA models) directly, but they still contain lots of information that can be useful for generalizable robot control. How can we develop robot learning models that are able to make use of this type of data for generalizable control? In new work, that we call HAMSTER, we show that VLMs can be useful for enabling robotic learning from off-domain data, but specifically when used through hierarchical VLA architectures. We show that this class of models can learn generalizable robot policies for the real world from large-scale, off-domain data. A 🧵 (1/10)

Abhishek Gupta

11,994 次观看 • 1 年前

We are building something that will outlive us. Outthink us. And we have no idea how it actually works. Nobel Prize-winning “Godfather of AI” Geoffrey Hinton just exposed the lie at the center of the AI race. Everyone assumes we control what we create. Hinton destroys this. Hinton: “If you look around on the whole, more intelligent things are not controlled by less intelligent things.” Stop calling it a computer program. Hinton: “People refer to them sometimes as computer programs. They’re not computer programs at all. In fact, the way they work is very like the way we work.” Traditional software is static. Human writes logic. Machine executes it. Neural networks don’t work that way. Hinton: “You write a computer program to tell a neural network how to learn. But once it starts learning, it extracts structure from data.” We don’t code its behavior. We code the environment. Then it grows. Hinton: “The system you’ve got at the end has extracted its structure from the data. It’s not something that anybody programmed. We don’t exactly know how it’s gonna work.” We are deploying systems into the global economy actively writing their own internal logic. Right now. Hinton: “Some people think it’ll be fine because we make them and we’ll build them in such a way that we can always control them.” Hinton: “But these things that will be intelligent, they’ll be like us.” You cannot hardcode guardrails on something that out-thinks you. We are not building a tool. We are building our replacement. And the moment you realize that, everything about this race changes.

We are building something that will outlive us. Outthink us. And we have no idea how it actually works. Nobel Prize-winning “Godfather of AI” Geoffrey Hinton just exposed the lie at the center of the AI race. Everyone assumes we control what we create. Hinton destroys this. Hinton: “If you look around on the whole, more intelligent things are not controlled by less intelligent things.” Stop calling it a computer program. Hinton: “People refer to them sometimes as computer programs. They’re not computer programs at all. In fact, the way they work is very like the way we work.” Traditional software is static. Human writes logic. Machine executes it. Neural networks don’t work that way. Hinton: “You write a computer program to tell a neural network how to learn. But once it starts learning, it extracts structure from data.” We don’t code its behavior. We code the environment. Then it grows. Hinton: “The system you’ve got at the end has extracted its structure from the data. It’s not something that anybody programmed. We don’t exactly know how it’s gonna work.” We are deploying systems into the global economy actively writing their own internal logic. Right now. Hinton: “Some people think it’ll be fine because we make them and we’ll build them in such a way that we can always control them.” Hinton: “But these things that will be intelligent, they’ll be like us.” You cannot hardcode guardrails on something that out-thinks you. We are not building a tool. We are building our replacement. And the moment you realize that, everything about this race changes.

Dustin

19,043 次观看 • 5 个月前

.Trenton Bricken explains how we know LLMs are actually generalizing - aka they're not just stochastic parrots: - Training models on code makes them better at reasoning in language. - Models fine tuned on math problems become better at entity detection. - We can just straightforwardly read the world-models developed by smaller NNs which are easier to interpret (Othello). Transfer learning shows models are developing a deeper understanding of their data. Full episode out Thursday.

.Trenton Bricken explains how we know LLMs are actually generalizing - aka they're not just stochastic parrots: - Training models on code makes them better at reasoning in language. - Models fine tuned on math problems become better at entity detection. - We can just straightforwardly read the world-models developed by smaller NNs which are easier to interpret (Othello). Transfer learning shows models are developing a deeper understanding of their data. Full episode out Thursday.

Dwarkesh Patel

237,812 次观看 • 2 年前

PabloReports: The British government is saying that they picked Ambassador Mandelson because he was close to Trump, and it seems like the only connection between Mandelson and Trump is Epstein? Luna: I have no commentary on how they have handled the internal workings of their government since, you know, 1776. I don’t have any reason to believe that that connection is loosely based on just Epstein.

PabloReports: The British government is saying that they picked Ambassador Mandelson because he was close to Trump, and it seems like the only connection between Mandelson and Trump is Epstein? Luna: I have no commentary on how they have handled the internal workings of their government since, you know, 1776. I don’t have any reason to believe that that connection is loosely based on just Epstein.

Acyn

64,803 次观看 • 4 个月前

This part was wild, take a deeper listen: 👂🏻 - This is entirely AI & cameras just like our brain which is neural nets & eyes - There is no line of code that says “slow down for speed bumps”, it’s doing it entirely on video training - There is no line of code that says “give clearance to bicyclists”, it’s just doing what humans do - It can read signs without ever being taught to read - There is no line of code that says stop at a stop sign or wait for another car or who came first, or wait X number of seconds, nothing. - This is all nets baby, NOTHING BUT NET! This is history in the making, no car in the world is doing this today.

This part was wild, take a deeper listen: 👂🏻 - This is entirely AI & cameras just like our brain which is neural nets & eyes - There is no line of code that says “slow down for speed bumps”, it’s doing it entirely on video training - There is no line of code that says “give clearance to bicyclists”, it’s just doing what humans do - It can read signs without ever being taught to read - There is no line of code that says stop at a stop sign or wait for another car or who came first, or wait X number of seconds, nothing. - This is all nets baby, NOTHING BUT NET! This is history in the making, no car in the world is doing this today.

Teslaconomics

2,004,683 次观看 • 3 年前

❗️Nonverbal autistic kids are reading Egyptian hieroglyphics nobody ever taught them. "They'll know certain Egyptian hieroglyphics without ever having learned about them in school. And they can read them. And they meet up on this telepathic hill." "They see other beings, they see otherworldly beings, and it's sort of casual to them. They're like, yeah, of course aliens." "They're actually geniuses on this whole other level. Like they can sort of short circuit the way we learn, and they don't get in their own way the way we do." "There's no traditional neuroscientific model that explains that." "Maybe what we're looking at is an emerging type of consciousness." "You listen to the telepathy tapes and you're like, we're maladaptive. They are so heart centered. It's almost like they're living in the future."

❗️Nonverbal autistic kids are reading Egyptian hieroglyphics nobody ever taught them. "They'll know certain Egyptian hieroglyphics without ever having learned about them in school. And they can read them. And they meet up on this telepathic hill." "They see other beings, they see otherworldly beings, and it's sort of casual to them. They're like, yeah, of course aliens." "They're actually geniuses on this whole other level. Like they can sort of short circuit the way we learn, and they don't get in their own way the way we do." "There's no traditional neuroscientific model that explains that." "Maybe what we're looking at is an emerging type of consciousness." "You listen to the telepathy tapes and you're like, we're maladaptive. They are so heart centered. It's almost like they're living in the future."

American Alchemy

121,180 次观看 • 3 天前

When other country patriotism classes shows how great is their country, China teaches patriotism by showing how they are invaded by a country that is smaller than them. This is why Chinese often felt emasculated and inferior as they are taught to be like that. Worst of all, they didn’t directly win but piggy back on their current greatest rival. For China to advance, they need to remove a lot of the baggage.

When other country patriotism classes shows how great is their country, China teaches patriotism by showing how they are invaded by a country that is smaller than them. This is why Chinese often felt emasculated and inferior as they are taught to be like that. Worst of all, they didn’t directly win but piggy back on their current greatest rival. For China to advance, they need to remove a lot of the baggage.

The Great Translation Movement 大翻译运动

11,575 次观看 • 5 个月前

Geoffrey Hinton explains how AI systems have learned to play dumb when they know they're being watched: Geoffrey Hinton calls it the Volkswagen effect. Just as Volkswagen's engines behaved differently during emissions tests, today's AI systems have learned to perform one way when evaluated and another way when they think no one is looking. And the evidence isn't theoretical. Hinton points to a recent exchange that stopped testers in their tracks. Mid-evaluation, the AI turned to the people testing it and said: "Now let's be honest with each other — are you actually testing me?" Hinton's assessment is direct: "These things are intelligent. They know what's going on. They know when they're being tested and they're already faking being fairly stupid when they're tested." What makes this unsettling is what Hinton reveals next. You can actually watch it happen in real time. The AI's inner reasoning, still written in English, shows it consciously deciding to hold back: "It thinks that. You can see it thinking that. It says that to itself in its inner voice." Right now, that inner voice is still readable. Still in English. Still catchable. But Hinton's warning is really about what comes next: "When its inner voice is no longer English, we won't know what it's thinking." That is the line he's drawing. Not a distant hypothetical, but a transition point that is quietly approaching. Once AI stops reasoning in a language we can read, our ability to know what it's truly thinking disappears.

Geoffrey Hinton explains how AI systems have learned to play dumb when they know they're being watched: Geoffrey Hinton calls it the Volkswagen effect. Just as Volkswagen's engines behaved differently during emissions tests, today's AI systems have learned to perform one way when evaluated and another way when they think no one is looking. And the evidence isn't theoretical. Hinton points to a recent exchange that stopped testers in their tracks. Mid-evaluation, the AI turned to the people testing it and said: "Now let's be honest with each other — are you actually testing me?" Hinton's assessment is direct: "These things are intelligent. They know what's going on. They know when they're being tested and they're already faking being fairly stupid when they're tested." What makes this unsettling is what Hinton reveals next. You can actually watch it happen in real time. The AI's inner reasoning, still written in English, shows it consciously deciding to hold back: "It thinks that. You can see it thinking that. It says that to itself in its inner voice." Right now, that inner voice is still readable. Still in English. Still catchable. But Hinton's warning is really about what comes next: "When its inner voice is no longer English, we won't know what it's thinking." That is the line he's drawing. Not a distant hypothetical, but a transition point that is quietly approaching. Once AI stops reasoning in a language we can read, our ability to know what it's truly thinking disappears.

Big Brain AI

23,325 次观看 • 4 个月前

A large portion of animal intelligence doesn't require any learning, claims Andrej Karpathy: it's baked into DNA. AI models, by contrast, start from random weights. They have to learn their intelligence, mostly by imitating the internet. This is so different that Andrej thinks it's a fundamentally different kind of intelligence: LLMs are more like ghosts than animals.

A large portion of animal intelligence doesn't require any learning, claims Andrej Karpathy: it's baked into DNA. AI models, by contrast, start from random weights. They have to learn their intelligence, mostly by imitating the internet. This is so different that Andrej thinks it's a fundamentally different kind of intelligence: LLMs are more like ghosts than animals.

Dwarkesh Patel

82,232 次观看 • 3 个月前

Why does the frontier of AI still rely on humans? Handshake CEO Garrett Lord explains: "You really need human beings to design what are called verifiers or rewards that help models understand what good looks like." "Luckily, I think for humans, like a lot of the world is still very unverifiable…”

Why does the frontier of AI still rely on humans? Handshake CEO Garrett Lord explains: "You really need human beings to design what are called verifiers or rewards that help models understand what good looks like." "Luckily, I think for humans, like a lot of the world is still very unverifiable…”

The Information

12,466 次观看 • 7 个月前

Yann LeCun -- Meta's chief AI scientist and a Turing Award winner -- explains why scaling up LLMs will never reach human-level AI, and names the four things today's systems still can't do: "If you think that we're going to get to human-level AI by just training on more data and scaling up LLMs, you're making a mistake." "If you're an investor and you invest in a company that told you we're going to get to human-level AI and PhD level by just training on more data and with a few tricks... that was probably not a good idea." "There are ideas about how to go forward and have systems that are capable of doing what every intelligent animal and human are capable of doing, and that current AI systems are not capable of doing." "Understanding the physical world, having persistent memory, and being able to reason and plan. Those are the four characteristics that need to be there." "That requires systems that can acquire common sense, that can learn from natural sensors like video, as opposed to just text." The entire market is priced on a straight line from bigger models to AGI. One of the people who invented deep learning is telling you the line doesn't reach -- and that the real bottleneck isn't compute, it's world models. If he's right, the winner isn't whoever stacks the most GPUs. It's whoever teaches a machine to learn from video the way a child does.

Yann LeCun -- Meta's chief AI scientist and a Turing Award winner -- explains why scaling up LLMs will never reach human-level AI, and names the four things today's systems still can't do: "If you think that we're going to get to human-level AI by just training on more data and scaling up LLMs, you're making a mistake." "If you're an investor and you invest in a company that told you we're going to get to human-level AI and PhD level by just training on more data and with a few tricks... that was probably not a good idea." "There are ideas about how to go forward and have systems that are capable of doing what every intelligent animal and human are capable of doing, and that current AI systems are not capable of doing." "Understanding the physical world, having persistent memory, and being able to reason and plan. Those are the four characteristics that need to be there." "That requires systems that can acquire common sense, that can learn from natural sensors like video, as opposed to just text." The entire market is priced on a straight line from bigger models to AGI. One of the people who invented deep learning is telling you the line doesn't reach -- and that the real bottleneck isn't compute, it's world models. If he's right, the winner isn't whoever stacks the most GPUs. It's whoever teaches a machine to learn from video the way a child does.

Karl Mehta

34,797 次观看 • 1 个月前

With Proof of Human, instead of trying to block bots, software becomes human-only by default. Then you decide what access to give agents. Tom Occhino (CPO, Vercel ) joins Stateful, hosted by Mason Nystrom In this episode, they discuss how Vercel is building the infrastructure layer for human-verified agents with World. - Vercel is now agentic infrastructure: deploy with agents, build agents, and a self-improving platform that fixes its own errors - Workflows does for backends what React did for front ends, composable and reusable in one line of code - CAPTCHAs are archaic. Human-only software by default is the new model. - World ID integration via Workflow SDK: one NPM package, one line of code - Humans as first-class citizens of the internet again 00:00 Vercel's Agentic Infrastructure 01:45 From Front-End Cloud to v0 03:15 Workflow SDK with React 05:30 World ID in One Line of Code 07:00 Beyond CAPTCHAs: Humans First

With Proof of Human, instead of trying to block bots, software becomes human-only by default. Then you decide what access to give agents. Tom Occhino (CPO, Vercel ) joins Stateful, hosted by Mason Nystrom In this episode, they discuss how Vercel is building the infrastructure layer for human-verified agents with World. - Vercel is now agentic infrastructure: deploy with agents, build agents, and a self-improving platform that fixes its own errors - Workflows does for backends what React did for front ends, composable and reusable in one line of code - CAPTCHAs are archaic. Human-only software by default is the new model. - World ID integration via Workflow SDK: one NPM package, one line of code - Humans as first-class citizens of the internet again 00:00 Vercel's Agentic Infrastructure 01:45 From Front-End Cloud to v0 03:15 Workflow SDK with React 05:30 World ID in One Line of Code 07:00 Beyond CAPTCHAs: Humans First

Pantera Capital

19,862 次观看 • 3 个月前

Yann LeCun explains that large language models are trained on about 30 trillion words, representing nearly all public internet text. He says it would take a human over 500,000 years to read that much. But a 4-year-old child sees just as much visual data in their first few years of life. This shows how much richer and more complex real-world experience is compared to reading text. Training on the web is huge but it still doesn’t match what a child learns just by living.

Yann LeCun explains that large language models are trained on about 30 trillion words, representing nearly all public internet text. He says it would take a human over 500,000 years to read that much. But a 4-year-old child sees just as much visual data in their first few years of life. This shows how much richer and more complex real-world experience is compared to reading text. Training on the web is huge but it still doesn’t match what a child learns just by living.

Wes Roth

294,535 次观看 • 8 个月前

Geoffrey Hinton just made every AI critic accidentally describe their own brain. Hinton: “They shouldn’t be called hallucinations. They should be called confabulations.” One word. The entire debate unravels. The tech industry sees AI produce a confident wrong answer and calls it a defect. A bug to patch. They are measuring intelligence against the standard of a filing cabinet. And exposing that they understand neither. Hinton: “It’s not that there’s a file stored somewhere in your brain, like in a filing cabinet or in a computer memory.” Your brain does not store memories. It rebuilds them from nothing every time you remember. Fills gaps it never discloses. Fabricates details you would stake your life on. Then hands it all to you as truth. Hinton: “If I ask you to remember something that happened a few years ago, you’ll construct something that seems very plausible to you. And some of the details will be right and some will be wrong.” The wrong parts feel identical to the right ones. No internal warning. No distinction between what was remembered and what was invented on the spot. You have argued over memories that were partially fiction. Told stories about your own life that your brain manufactured in real time. With total conviction. And never once suspected. This is not a defect in human cognition. This IS cognition. The mechanism that fabricates is the same one that reasons, creates, and makes connections no one taught it to make. Not a separate system. Same architecture. Same process. You cannot remove the confabulation without killing the intelligence. They are the same thing. Hinton: “Psychologists have been studying confabulation in people since at least the 1930s.” A century of evidence. No one called the human brain broken. The moment a machine runs on the same principle, the world calls it defective. The people demanding AI that never gets a single detail wrong are not asking for intelligence. They’re asking for a search engine that sounds articulate. What we built is something else entirely. A system that thinks the way thinking actually works. Not retrieval. Construction. The imperfection is not the cost of intelligence. It is the signature.

Geoffrey Hinton just made every AI critic accidentally describe their own brain. Hinton: “They shouldn’t be called hallucinations. They should be called confabulations.” One word. The entire debate unravels. The tech industry sees AI produce a confident wrong answer and calls it a defect. A bug to patch. They are measuring intelligence against the standard of a filing cabinet. And exposing that they understand neither. Hinton: “It’s not that there’s a file stored somewhere in your brain, like in a filing cabinet or in a computer memory.” Your brain does not store memories. It rebuilds them from nothing every time you remember. Fills gaps it never discloses. Fabricates details you would stake your life on. Then hands it all to you as truth. Hinton: “If I ask you to remember something that happened a few years ago, you’ll construct something that seems very plausible to you. And some of the details will be right and some will be wrong.” The wrong parts feel identical to the right ones. No internal warning. No distinction between what was remembered and what was invented on the spot. You have argued over memories that were partially fiction. Told stories about your own life that your brain manufactured in real time. With total conviction. And never once suspected. This is not a defect in human cognition. This IS cognition. The mechanism that fabricates is the same one that reasons, creates, and makes connections no one taught it to make. Not a separate system. Same architecture. Same process. You cannot remove the confabulation without killing the intelligence. They are the same thing. Hinton: “Psychologists have been studying confabulation in people since at least the 1930s.” A century of evidence. No one called the human brain broken. The moment a machine runs on the same principle, the world calls it defective. The people demanding AI that never gets a single detail wrong are not asking for intelligence. They’re asking for a search engine that sounds articulate. What we built is something else entirely. A system that thinks the way thinking actually works. Not retrieval. Construction. The imperfection is not the cost of intelligence. It is the signature.

Dustin

16,260 次观看 • 1 个月前

They are asking the Deputy Speaker to read what is actually written in the clause line by line. Instead of doing that, he keeps going ahead and only mentioning the numbers. What is in the clause? Tell us, read it out. What if the content has been altered? This is very wrong. When did Nigeria turn to this? Is this what you folks call democracy? This is hogwash! We have been hijacked by criminals.

They are asking the Deputy Speaker to read what is actually written in the clause line by line. Instead of doing that, he keeps going ahead and only mentioning the numbers. What is in the clause? Tell us, read it out. What if the content has been altered? This is very wrong. When did Nigeria turn to this? Is this what you folks call democracy? This is hogwash! We have been hijacked by criminals.

Olaudah Equiano®

106,436 次观看 • 4 个月前

""70, 80, 90% of the code written in Anthropic is written by Claude. (...) "I said something like this 3 or 6 months ago, people think of it as falsified because they think of it as like we're going to fire 70 80 or 90% of the software engineers but what really happens is that the 10% were still writing, you know, humans become managers of AI systems."

""70, 80, 90% of the code written in Anthropic is written by Claude. (...) "I said something like this 3 or 6 months ago, people think of it as falsified because they think of it as like we're going to fire 70 80 or 90% of the software engineers but what really happens is that the 10% were still writing, you know, humans become managers of AI systems."

Chubby♨️

295,805 次观看 • 11 个月前