Video yükleniyor...

Video Yüklenemedi

Ana Sayfaya Dön

Ilya on LLMs understanding the world: "predicting the next token well, means that you understand the underlying reality that let to the creation of that token" Seem like the opposite view of Yann.

LiorOnAI's profile picture

Lior Alexander

115,687 subscribers

1,222,086 görüntüleme • 2 yıl önce •via X (Twitter)

Bilim & TeknolojiHaberler & PolitikaEğitim
LIVE

Anya Rossi• Live Now

Private livecam show

10 Yorum

Marcus D. R. Klarqvist profil fotoğrafı
Marcus D. R. Klarqvist2 yıl önce

view is correct. @ilyasut is not. Predicting the next token well is simply a matter of learning statistical correlations of latent representations of text. The consequences of compressing this universe into something smaller is the literal definition for why it is stochastic and the literal definition for why models "dream" or "hallucinate". Existing architectures have no way of exerting intelligence by deviating from these relationships through logic and reason. Hence Yann keep repeating that children with almost no training data are superior learning machines compared to extant ML models that require as training data the entirety of all human text ever written as input.

takeoff profil fotoğrafı
takeoff2 yıl önce

@AlphaSignalAI Yann's argument is more about efficiency. Humans could generate an "internal model" of a new person they met with a short conversation with a person. LLMs would need 1000x more data and even then their "internal model" would have some flaws as compared to the human model.

Dwarkesh Patel profil fotoğrafı
Dwarkesh Patel2 yıl önce

@AlphaSignalAI Thanks for highlighting the clip! Here are the links to the full episode: YouTube: Apple Podcasts: Spotify: Transcript:

AlphaSignal AI profil fotoğrafı
AlphaSignal AI2 yıl önce

@AlphaSignalAI Source:

Nyan Cat Money profil fotoğrafı
Nyan Cat Money2 yıl önce

@AlphaSignalAI Ilya is right about this. They taught the LLMs to learn by predicting the next token. When you learn how to do a job, there's explicit things you learn and lots of implicit things you learn. You are only trying to predict the explicit things, but you learn the implicit naturally

Max Rovensky profil fotoğrafı
Max Rovensky2 yıl önce

@AlphaSignalAI So T9 understands the underlying reality of what led to the creation of the next character?

Cebes profil fotoğrafı
Cebes2 yıl önce

@AlphaSignalAI No you don’t “understand” that underlying reality, you merely “simulate” that underlying reality. Quite an important difference.

Giancarlo Niccolai profil fotoğrafı
Giancarlo Niccolai2 yıl önce

@AlphaSignalAI True. But that reality is nothing but a bunch of numbers. That doesn't mean that our brain doesn't do that. There are many processes in our brain that are just like that, but the road to real "understanding" is still long.

Christian @homoveritas profil fotoğrafı
Christian @homoveritas2 yıl önce

@AlphaSignalAI Just means that the probability was calculated correctly not a holistic understanding then again what is the brain but a fancy probability engine

Yanco profil fotoğrafı
Yanco2 yıl önce

@AlphaSignalAI But but but.. it is just a stochastic parrot!!!

Benzer Videolar

Yann LeCun says language isn’t intelligence. Predicting text doesn’t mean understanding reality. The real world is messy, physical, and causal and today’s LLMs barely touch that. The next leap is Physical AI: world models, cause and effect, real planning. Do you think LLMs can evolve into this, or do we need a completely new architecture?
2:08

Yann LeCun says language isn’t intelligence. Predicting text doesn’t mean understanding reality. The real world is messy, physical, and causal and today’s LLMs barely touch that. The next leap is Physical AI: world models, cause and effect, real planning. Do you think LLMs can evolve into this, or do we need a completely new architecture?

VraserX e/acc

76,194 görüntüleme • 6 ay önce

📁 Yann LeCun explains that LLMs work well when problems are symbolic, like math, code or chess, where searching through known sequences is enough. But the real world does not work that way. Physical action, planning and understanding what is possible require continuous intuition, which LLMs lack. Manipulating symbols is not the same as understanding reality.
1:25

📁 Yann LeCun explains that LLMs work well when problems are symbolic, like math, code or chess, where searching through known sequences is enough. But the real world does not work that way. Physical action, planning and understanding what is possible require continuous intuition, which LLMs lack. Manipulating symbols is not the same as understanding reality.

Jon Hernandez

42,650 görüntüleme • 7 ay önce

Full Token Persistence 🟥 Six months ago Saito accelerated the path to Token Persistence. Right on schedule, the Token Persistence threshold is today 0. What that means and what comes next:
6:39

Full Token Persistence 🟥 Six months ago Saito accelerated the path to Token Persistence. Right on schedule, the Token Persistence threshold is today 0. What that means and what comes next:

Saito Network 🟥

20,524 görüntüleme • 11 ay önce

🚨 #XRP HOLDERS MUST UNDERSTAND THE POTENTIAL VALUE COMING SOON!! THIS IS WHY A PRICE OF $10,000-35,000 PER XRP COULD BECOME A REALITY. BRAD HAS STATED THAT TRILLIONS WILL BE CAPTURED FROM THE BANKING INDUSTRY!! THE OPPORTUNITY IS HUGE—ALMOST $10 TRILLION IN BANKING ASSETS COULD SOON FLOW INTO XRP. This includes the rise of the leading DeFi token on the XRP Ledger (#XRPL), CTF token, as well as the integration of major players like Amazon and Walmart!! ON TOP OF THAT, CTF token could soar from $0.95 to $748.50 per coin!! CTF token trade link: Website:
0:36

🚨 #XRP HOLDERS MUST UNDERSTAND THE POTENTIAL VALUE COMING SOON!! THIS IS WHY A PRICE OF $10,000-35,000 PER XRP COULD BECOME A REALITY. BRAD HAS STATED THAT TRILLIONS WILL BE CAPTURED FROM THE BANKING INDUSTRY!! THE OPPORTUNITY IS HUGE—ALMOST $10 TRILLION IN BANKING ASSETS COULD SOON FLOW INTO XRP. This includes the rise of the leading DeFi token on the XRP Ledger (#XRPL), CTF token, as well as the integration of major players like Amazon and Walmart!! ON TOP OF THAT, CTF token could soar from $0.95 to $748.50 per coin!! CTF token trade link: Website:

JackTheRippler ©️

89,541 görüntüleme • 1 yıl önce

‼️BREAKING: RIPPLE BANKING LICENSE ON VERGE OF ISSUANCE!! EXPECTED TO BE IN NOVEMBER!!! THE SAME MONTH THAT THE DECENTRALIZED MEDIA ON THE #XRP LEDGER BACKED BY BXE TOKEN IS SET TO LAUNCH!!!‼️ BXE TOKEN IS GOING TO GO INSANE!!! BXE TOKEN IS $0.03!!! DO YOU UNDERSTAND HOW CHEAP THAT IS!!! BXE TOKEN HAS A TOTAL SUPPLY OF 500 MILION TOKENS!!! $19-24 PER BXE TOKEN IS EASILY!!! BXE TOKEN TRADE LINK:
0:55

‼️BREAKING: RIPPLE BANKING LICENSE ON VERGE OF ISSUANCE!! EXPECTED TO BE IN NOVEMBER!!! THE SAME MONTH THAT THE DECENTRALIZED MEDIA ON THE #XRP LEDGER BACKED BY BXE TOKEN IS SET TO LAUNCH!!!‼️ BXE TOKEN IS GOING TO GO INSANE!!! BXE TOKEN IS $0.03!!! DO YOU UNDERSTAND HOW CHEAP THAT IS!!! BXE TOKEN HAS A TOTAL SUPPLY OF 500 MILION TOKENS!!! $19-24 PER BXE TOKEN IS EASILY!!! BXE TOKEN TRADE LINK:

XRP QUEEN🤍

32,640 görüntüleme • 10 ay önce

Ilya Sutskever had a conversation last year with Jensen Huang but still, that made me rethink the idea of 'Predicting the Next Word' Listen to what Ilya said on this.
0:44

Ilya Sutskever had a conversation last year with Jensen Huang but still, that made me rethink the idea of 'Predicting the Next Word' Listen to what Ilya said on this.

Haider.

1,252,471 görüntüleme • 1 yıl önce

New course: Transformers in Practice. You'll get a practical view of how transformer-based LLMs work, so you can reason about their behavior, diagnose problems like slow inference, and make smarter decisions about deployment. This course is built in partnership with AMD and taught by Sharon Zhou. You'll see how transformers generate text one token at a time, how the model decides which earlier words matter most when predicting the next one, and how techniques like quantization speed up inference on GPUs. This is not a video-only course; interactive visualizations throughout let you play with these concepts and build intuition that sticks. Skills you'll gain: - Understand why LLMs hallucinate, and RAG and chain-of-thought shape what they generate - Look inside the model to see how attention and layers combine to predict the next token - Diagnose inference bottlenecks and learn the techniques that speed up transformers on GPUs Join and understand what's really happening inside your LLMs:
1:00

New course: Transformers in Practice. You'll get a practical view of how transformer-based LLMs work, so you can reason about their behavior, diagnose problems like slow inference, and make smarter decisions about deployment. This course is built in partnership with AMD and taught by Sharon Zhou. You'll see how transformers generate text one token at a time, how the model decides which earlier words matter most when predicting the next one, and how techniques like quantization speed up inference on GPUs. This is not a video-only course; interactive visualizations throughout let you play with these concepts and build intuition that sticks. Skills you'll gain: - Understand why LLMs hallucinate, and RAG and chain-of-thought shape what they generate - Look inside the model to see how attention and layers combine to predict the next token - Diagnose inference bottlenecks and learn the techniques that speed up transformers on GPUs Join and understand what's really happening inside your LLMs:

Andrew Ng

121,460 görüntüleme • 2 ay önce

Immerse yourself in the world of BillionDollarDuck, the pioneering token that rewards FREE airdrops to all liquidity providers from various projects, a truly passive income token that is entering the crypto world with a splash. #billiondollarduck
0:30

Immerse yourself in the world of BillionDollarDuck, the pioneering token that rewards FREE airdrops to all liquidity providers from various projects, a truly passive income token that is entering the crypto world with a splash. #billiondollarduck

(Crypto Airdrops) BlockDeskDrop

11,871 görüntüleme • 2 yıl önce

Ilya Sutskever's has a bold take. LLMs are doing much more than predicting the next word. They are learning our world model. Text is a projection of the world.
3:13

Ilya Sutskever's has a bold take. LLMs are doing much more than predicting the next word. They are learning our world model. Text is a projection of the world.

Lior Alexander

326,445 görüntüleme • 2 yıl önce

Writing high quality user interface (UI) code requires more than predicting the next token - it requires understanding unique UI code's unique syntax and assessing the rendered output's quality and relevance. Most LLMs struggle to do this.
0:29

Writing high quality user interface (UI) code requires more than predicting the next token - it requires understanding unique UI code's unique syntax and assessing the rendered output's quality and relevance. Most LLMs struggle to do this.

Jason Wu

14,377 görüntüleme • 2 yıl önce

Ilya Sutskever says accurately predicting the next word leads to real understanding.
0:46

Ilya Sutskever says accurately predicting the next word leads to real understanding.

vitrupo

415,059 görüntüleme • 3 ay önce

Yann LeCun says the real world is far more complex than the world of language LLMs can accumulate knowledge, but they fail with high-dimensional, continuous, noisy sensory data "the next revolution is physical AI" Systems that can truly plan, reason, and understand the physical environment
1:32

Yann LeCun says the real world is far more complex than the world of language LLMs can accumulate knowledge, but they fail with high-dimensional, continuous, noisy sensory data "the next revolution is physical AI" Systems that can truly plan, reason, and understand the physical environment

Haider.

68,894 görüntüleme • 6 ay önce

In order to have successful capital markets, you see the same things happen again and again. Since one man’s debts are another man’s assets, you have to keep interest rates not so high that they crush the debtor, without having them so low that they are bad for the creditor. That leads to cycles of money creation and borrowing that are predictable. These cycles must happen, due to the nature of the dynamic around them. Understanding the underlying mechanics driving these patterns can help you understand what is likely to happen next.
1:44

In order to have successful capital markets, you see the same things happen again and again. Since one man’s debts are another man’s assets, you have to keep interest rates not so high that they crush the debtor, without having them so low that they are bad for the creditor. That leads to cycles of money creation and borrowing that are predictable. These cycles must happen, due to the nature of the dynamic around them. Understanding the underlying mechanics driving these patterns can help you understand what is likely to happen next.

Ray Dalio

113,808 görüntüleme • 5 ay önce

3Blue1Brown’s new video explains why every LLM is actually a compression machine. everyone describes pre-training as “next token prediction” but that’s just the surface-level objective. in reality it is a means to making the most efficient text compressor. prediction and compression are two sides of the same coin. when you train the model to predict the next token you’re not just teaching it to guess the next word but how to best encode the human knowledge it sees. better compression means better abstraction means better reasoning at some point, compression stops looking like storage or a database (as some like to call it on X) and looks like an approximation of understanding.
0:58

3Blue1Brown’s new video explains why every LLM is actually a compression machine. everyone describes pre-training as “next token prediction” but that’s just the surface-level objective. in reality it is a means to making the most efficient text compressor. prediction and compression are two sides of the same coin. when you train the model to predict the next token you’re not just teaching it to guess the next word but how to best encode the human knowledge it sees. better compression means better abstraction means better reasoning at some point, compression stops looking like storage or a database (as some like to call it on X) and looks like an approximation of understanding.

ℏεsam

120,027 görüntüleme • 2 ay önce

Yann LeCun: I'm not interested in LLMs anymore - they're the past. The future is in four more interesting areas: machines that understand the physical world, persistent memory, reasoning, and planning.
1:10

Yann LeCun: I'm not interested in LLMs anymore - they're the past. The future is in four more interesting areas: machines that understand the physical world, persistent memory, reasoning, and planning.

Victor

525,746 görüntüleme • 1 yıl önce

Introducing $KIGU, the official token of Kigu After amassing 650K+ fans and 25M+ views, the first autonomous video creator is launching its token. The next evolution of content creation begins August 13th.
0:24

Introducing $KIGU, the official token of Kigu After amassing 650K+ fans and 25M+ views, the first autonomous video creator is launching its token. The next evolution of content creation begins August 13th.

Super Champs

18,731 görüntüleme • 1 yıl önce

TOKENFI TOKEN LAUNCHER IS LIVE ON MAINNET IT’S OFFICIAL! TokenFi Token Launcher is live on the mainnet of five top EVM blockchains: Ethereum, BNB Chain, Base, Arbitrum, and opBNB. With TokenFi Token Launcher, ANYONE can create a fungible crypto token (ERC-20), NFT (ERC-721), or Multi-Token (ERC-1155) in just a few clicks, regardless of their crypto experience. The TokenFi platform and framework has been thoroughly audited and is being monitored in real-time by industry-leading blockchain security auditor Certik. This means that the tokens you create with TokenFi are safe and audited. TokenFi Token Launcher is powered by $TOKEN as its main utility token, and every successful transaction on the Ethereum and BNB blockchains buys and burns $TOKEN, enhancing its utility and making it perpetually deflationary as adoption of the platform grows. TokenFi Token Launcher aims to be the dominant token creation platform on EVM blockchains, and to this effect, we will be rolling out the platform on dozens of additional EVM blockchains in the coming weeks and months. You can learn more about TokenFi Token Launcher here:
0:34

TOKENFI TOKEN LAUNCHER IS LIVE ON MAINNET IT’S OFFICIAL! TokenFi Token Launcher is live on the mainnet of five top EVM blockchains: Ethereum, BNB Chain, Base, Arbitrum, and opBNB. With TokenFi Token Launcher, ANYONE can create a fungible crypto token (ERC-20), NFT (ERC-721), or Multi-Token (ERC-1155) in just a few clicks, regardless of their crypto experience. The TokenFi platform and framework has been thoroughly audited and is being monitored in real-time by industry-leading blockchain security auditor Certik. This means that the tokens you create with TokenFi are safe and audited. TokenFi Token Launcher is powered by $TOKEN as its main utility token, and every successful transaction on the Ethereum and BNB blockchains buys and burns $TOKEN, enhancing its utility and making it perpetually deflationary as adoption of the platform grows. TokenFi Token Launcher aims to be the dominant token creation platform on EVM blockchains, and to this effect, we will be rolling out the platform on dozens of additional EVM blockchains in the coming weeks and months. You can learn more about TokenFi Token Launcher here:

TokenFi

1,288,915 görüntüleme • 2 yıl önce

Yann LeCun (Yann LeCun ) beautifully explains how the architecture and principles used to train LLMs can not be extended to teach AI the real-world intelligence. In 1 line: LLMs excel where intelligence equals sequence prediction over symbols. Real-world intelligence requires learned world models, abstraction, causality, and action planning under uncertainty, which current next-token training does not provide. He says current LLMs learn by predicting the next token. That objective works very well when the task itself can be reduced to manipulating discrete symbols and sequences. Math, physics problem solving on paper, and coding fit this pattern because success largely comes from searching and composing the right sequences of symbols, equations, or program tokens. With enough data and scale, these models get very good at that kind of structured sequence prediction. Real-world intelligence is different. The physical world is continuous, noisy, uncertain, and high dimensional. To act in it, a system needs internal models that capture objects, dynamics, causality, constraints from the body, and the outcomes of actions over time. Humans and animals build abstract representations from rich sensory streams, then make predictions in that abstract space, not at the raw pixel level. That is why a child can learn intuitive physics, plan multi-step actions, and adapt quickly in new situations with little data. His claim about saturation follows from this gap. Scaling token prediction keeps improving symbol manipulation tasks like math and code, but it hits limits on embodied reasoning and common sense because text alone does not provide the right learning signals for world models. Predicting the next word cannot efficiently teach contact forces, affordances, occlusion, friction, or how actions change the state of the environment. For that, he argues we need architectures that learn abstractions from sensory data and predict futures in abstract latent spaces, then use those predictions to plan actions toward goals with built-in guardrails. --- From 'Pioneer Works' YT Channel (link in comment)
1:13

Yann LeCun (Yann LeCun ) beautifully explains how the architecture and principles used to train LLMs can not be extended to teach AI the real-world intelligence. In 1 line: LLMs excel where intelligence equals sequence prediction over symbols. Real-world intelligence requires learned world models, abstraction, causality, and action planning under uncertainty, which current next-token training does not provide. He says current LLMs learn by predicting the next token. That objective works very well when the task itself can be reduced to manipulating discrete symbols and sequences. Math, physics problem solving on paper, and coding fit this pattern because success largely comes from searching and composing the right sequences of symbols, equations, or program tokens. With enough data and scale, these models get very good at that kind of structured sequence prediction. Real-world intelligence is different. The physical world is continuous, noisy, uncertain, and high dimensional. To act in it, a system needs internal models that capture objects, dynamics, causality, constraints from the body, and the outcomes of actions over time. Humans and animals build abstract representations from rich sensory streams, then make predictions in that abstract space, not at the raw pixel level. That is why a child can learn intuitive physics, plan multi-step actions, and adapt quickly in new situations with little data. His claim about saturation follows from this gap. Scaling token prediction keeps improving symbol manipulation tasks like math and code, but it hits limits on embodied reasoning and common sense because text alone does not provide the right learning signals for world models. Predicting the next word cannot efficiently teach contact forces, affordances, occlusion, friction, or how actions change the state of the environment. For that, he argues we need architectures that learn abstractions from sensory data and predict futures in abstract latent spaces, then use those predictions to plan actions toward goals with built-in guardrails. --- From 'Pioneer Works' YT Channel (link in comment)

Rohan Paul

104,460 görüntüleme • 7 ay önce

Zohran Mamdani: “What Andrew Cuomo doesn’t seem to understand is that him and Donald Trump are two sides of the same coin that New Yorkers want to throw away into the dustbin of history. He just doesn’t understand that no means no.”
0:22

Zohran Mamdani: “What Andrew Cuomo doesn’t seem to understand is that him and Donald Trump are two sides of the same coin that New Yorkers want to throw away into the dustbin of history. He just doesn’t understand that no means no.”

Marco Foster

120,202 görüntüleme • 11 ay önce

🚨BREAKING NEWS: SEC EXPECTED TO DROP RIPPLE #XRP LAWSUIT IN UNDER 48 HOURS AS THE CONSENSYS LAWSUIT IS NOW OFFICIALLY DROPPED BY THE SEC!!! EXPECTIONS ARE THAT THE PRICE OF XRP WILL JUMP TO BETWEEN $10-$18 INSTANTLY UPON ANNOUCEMENT!!‼️ THIS MEANS MASSIVE DEFI GROWTH ON THE #XRPL!! TOP DEFI TOKEN ON XRPL, CTF TOKEN, LOOKS SO BULLISH RIGHT NOW!! ONLY $20 BILLION MARKET CAP WOULD SEND CTF TOKEN FROM $0.78 TO $748.50 PER TOKEN!!! XRP IS THE RESERVE CURRENCY OF THE USA!! THATS A FACT NOW! YOU THINK THE TOP DEFI TOKEN ON THE XRPL IS NOT GOING TO GO TO ONLY $20 BILLION MARKET CAP?? YOU ARE INSANE IF YOU THINK NOT!! CTF TOKEN HAS A TOTAL SUPPLY OF ONLY 119 MILLION TOKENS, THATS NOTHING!! CTF TOKEN IS GOING TO GO INSANE!! DYOR CTF TOKEN TRADE LINK: CTF TOKEN TRADE LINK ON MEXC: Website: 🚨🚨
1:15

🚨BREAKING NEWS: SEC EXPECTED TO DROP RIPPLE #XRP LAWSUIT IN UNDER 48 HOURS AS THE CONSENSYS LAWSUIT IS NOW OFFICIALLY DROPPED BY THE SEC!!! EXPECTIONS ARE THAT THE PRICE OF XRP WILL JUMP TO BETWEEN $10-$18 INSTANTLY UPON ANNOUCEMENT!!‼️ THIS MEANS MASSIVE DEFI GROWTH ON THE #XRPL!! TOP DEFI TOKEN ON XRPL, CTF TOKEN, LOOKS SO BULLISH RIGHT NOW!! ONLY $20 BILLION MARKET CAP WOULD SEND CTF TOKEN FROM $0.78 TO $748.50 PER TOKEN!!! XRP IS THE RESERVE CURRENCY OF THE USA!! THATS A FACT NOW! YOU THINK THE TOP DEFI TOKEN ON THE XRPL IS NOT GOING TO GO TO ONLY $20 BILLION MARKET CAP?? YOU ARE INSANE IF YOU THINK NOT!! CTF TOKEN HAS A TOTAL SUPPLY OF ONLY 119 MILLION TOKENS, THATS NOTHING!! CTF TOKEN IS GOING TO GO INSANE!! DYOR CTF TOKEN TRADE LINK: CTF TOKEN TRADE LINK ON MEXC: Website: 🚨🚨

FinanceBro

172,382 görüntüleme • 1 yıl önce