Video yükleniyor...

Video Yüklenemedi

Bu video yüklenirken bir sorun oluştu. Bu geçici bir ağ sorunundan kaynaklanıyor olabilir veya video kullanılamıyor olabilir.

Ana Sayfaya Dön

Today's LLMs can't generate token 10 until token 9 exists. Inception CEO Stefano Ermon thinks he's found the winner: diffusion LLMs that generate tokens in parallel. New episode of No Priors, out now.

No Priors

12,502 subscribers

54,919 görüntüleme • 4 gün önce •via X (Twitter)

Eğitim Bilim & Teknoloji

Anya Rossi• Live Now

Private livecam show

3 Yorum

No Priors profil fotoğrafı

No Priors4 gün önce

Full conversation:

Zee Palm Labs profil fotoğrafı

Zee Palm Labs3 gün önce

@_inception_ai @StefanoErmon Parallel token generation could be a huge shift. Really curious to see how diffusion LLMs hold up on reasoning and coherence as they scale.

Nelson profil fotoğrafı

Nelson3 gün önce

@_inception_ai @StefanoErmon Does it generate better reasoning or is the main benefit speed?

Benzer Videolar

Today's autoregressive models generate one token at a time. Mercury 2 generates tokens in parallel. Over 1,000 tok/sec on standard GPUs, at comparable quality to speed-optimized models. Since launch, the community has been showing what diffusion LLMs can unlock. Thanks to the team at Clyep for the breakdown.

Today's autoregressive models generate one token at a time. Mercury 2 generates tokens in parallel. Over 1,000 tok/sec on standard GPUs, at comparable quality to speed-optimized models. Since launch, the community has been showing what diffusion LLMs can unlock. Thanks to the team at Clyep for the breakdown.

Inception

21,231 görüntüleme • 4 ay önce

“Next-token prediction” just got a serious rival 🤯 Ant Group just dropped LLaDA 2.1, and it challenges the dominant paradigm of LLMs. Unlike most models that generate one token at a time, LLaDA 2.1 uses diffusion to generate blocks of text in parallel. Why this changes everything: → Global Planning: It effectively sees the "future" while writing the "past" → Parallel Generation: It generates chunks in parallel, not sequentially, being much faster → Massive Efficiency: 16B MoE architecture that only uses ~1.4B active parameters per step. 100% Open Source.

“Next-token prediction” just got a serious rival 🤯 Ant Group just dropped LLaDA 2.1, and it challenges the dominant paradigm of LLMs. Unlike most models that generate one token at a time, LLaDA 2.1 uses diffusion to generate blocks of text in parallel. Why this changes everything: → Global Planning: It effectively sees the "future" while writing the "past" → Parallel Generation: It generates chunks in parallel, not sequentially, being much faster → Massive Efficiency: 16B MoE architecture that only uses ~1.4B active parameters per step. 100% Open Source.

Alvaro Cintas

31,277 görüntüleme • 7 ay önce

What if language models didn't have to generate one token at a time? Our CEO Stefano Ermon joined TBPN to break down how Mercury 2's diffusion LLM hits 1,000+ tok/s on standard NVIDIA GPUs — and why speed changes the product for coding, voice agents, and search.

What if language models didn't have to generate one token at a time? Our CEO Stefano Ermon joined TBPN to break down how Mercury 2's diffusion LLM hits 1,000+ tok/s on standard NVIDIA GPUs — and why speed changes the product for coding, voice agents, and search.

Inception

27,188 görüntüleme • 6 ay önce

A new Robinhood launchpad by print.world is releasing soon and I found the first token launched on the platform. On their docs page here: It says that its standard token list can be found here: And the very first token created on there is $HERO 0x0b9dfa193e0af4e47c2fee5b12f3a7ac78230e32 And the $GME token featured in the video is this one: 0xba11ef8f0d67579e5afe9efed6081abd39b3e607 Order of tokens from newest to oldest confirms that HERO is first:

A new Robinhood launchpad by print.world is releasing soon and I found the first token launched on the platform. On their docs page here: It says that its standard token list can be found here: And the very first token created on there is $HERO 0x0b9dfa193e0af4e47c2fee5b12f3a7ac78230e32 And the $GME token featured in the video is this one: 0xba11ef8f0d67579e5afe9efed6081abd39b3e607 Order of tokens from newest to oldest confirms that HERO is first:

TrenchDiver

22,642 görüntüleme • 2 ay 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:

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

124,677 görüntüleme • 4 ay önce

OpenAI's chairman Brett Taylor on the new Chinese model Kimi K3. the timeline thinks it's a frontier killer. he says every CEO he talks to asks the same two things: am I getting value from my token spend, AND what's my moat in the age of AI. on token efficiency, he says the frontier models still finish the same task in far fewer tokens. worth a listen.

OpenAI's chairman Brett Taylor on the new Chinese model Kimi K3. the timeline thinks it's a frontier killer. he says every CEO he talks to asks the same two things: am I getting value from my token spend, AND what's my moat in the age of AI. on token efficiency, he says the frontier models still finish the same task in far fewer tokens. worth a listen.

Damian Player

471,038 görüntüleme • 2 ay önce

📢 New Spot Listing — DEFI.ssi & MEME.ssi SSI (SoSoValue Indexes) We’re excited to announce that DEFI.ssi and MEME.ssi are now live for spot trading on SoDEX. What are SSI Tokens? SSI (SoSoValue Index) tokens let you gain exposure to an entire crypto sector through a single token — with monthly auto-rebalancing and institutional-grade custody by Cobo/Ceffu. 🔷 DEFI.ssi — 10 DeFi leaders, one token 🟡 MEME.ssi — 10 meme leaders, one token No need to manage 10 different coins yourself. Buy once, and get exposure to the whole sector. 👉 Trade now at

📢 New Spot Listing — DEFI.ssi & MEME.ssi SSI (SoSoValue Indexes) We’re excited to announce that DEFI.ssi and MEME.ssi are now live for spot trading on SoDEX. What are SSI Tokens? SSI (SoSoValue Index) tokens let you gain exposure to an entire crypto sector through a single token — with monthly auto-rebalancing and institutional-grade custody by Cobo/Ceffu. 🔷 DEFI.ssi — 10 DeFi leaders, one token 🟡 MEME.ssi — 10 meme leaders, one token No need to manage 10 different coins yourself. Buy once, and get exposure to the whole sector. 👉 Trade now at

SoDEX

15,562 görüntüleme • 6 ay önce

$Introducing SubQ - a major breakthrough in LLM intelligence. It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA), And the first frontier model with a 12 million token context window which is: - 52x faster than FlashAttention at 1MM tokens - Less than 5% the cost of Opus Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention). Only a small fraction actually matter. Subquadratic finds and focuses only on the ones that do. That's nearly 1,000x less compute and a new way for LLMs to scale.$

Introducing SubQ - a major breakthrough in LLM intelligence. It is the first model built on a fully sub-quadratic sparse-attention architecture (SSA), And the first frontier model with a 12 million token context window which is: - 52x faster than FlashAttention at 1MM tokens - Less than 5% the cost of Opus Transformer-based LLMs waste compute by processing every possible relationship between words (standard attention). Only a small fraction actually matter. Subquadratic finds and focuses only on the ones that do. That's nearly 1,000x less compute and a new way for LLMs to scale.

Alexander Whedon

13,486,299 görüntüleme • 4 ay önce

Andrej Karpathy recently coined the term “vibe coding” to describe how LLMs are getting so good that devs can “give in to the vibes, embrace exponentials, and forget that the code even exists.” In this episode of the Lightcone Podcast, the hosts discuss this new way of programming and what it means for builders in the age of AI.

Andrej Karpathy recently coined the term “vibe coding” to describe how LLMs are getting so good that devs can “give in to the vibes, embrace exponentials, and forget that the code even exists.” In this episode of the Lightcone Podcast, the hosts discuss this new way of programming and what it means for builders in the age of AI.

Y Combinator

828,368 görüntüleme • 1 yıl önce

What is a "token factory"—and why is China building them? China's first city-level green token factory is taking shape in Changzhou, where renewable energy will power the production of 60 trillion AI tokens each year. Unlike traditional factories, token factories generate the digital building blocks that AI models use to understand and create text. As AI demand continues to surge, China is combining clean energy with next-generation computing to build a more efficient and sustainable AI infrastructure.

What is a "token factory"—and why is China building them? China's first city-level green token factory is taking shape in Changzhou, where renewable energy will power the production of 60 trillion AI tokens each year. Unlike traditional factories, token factories generate the digital building blocks that AI models use to understand and create text. As AI demand continues to surge, China is combining clean energy with next-generation computing to build a more efficient and sustainable AI infrastructure.

Hi,GBA

59,590 görüntüleme • 2 ay önce

‼️‼️‼️BOOOOOOOOOOOOOOM‼️‼️ SWIFT PROCESSES $1.25 QUADRILLION TRANSACTIONS YEARLY!! IF #XRP JUST GETS 10% OF THAT SUPPLY, IT WOULD PUT XRP MARKET CAP AT $125 TRILLION!!! THATS $2250 PER XRP!!! THE MARKET SENTIMENT NEEDS TO BE ONBOARD FIRST!!! THATS WHY THE LARGEST DECENTRALIZED MEDIA PROGRAM LAUNCHED ON THE XRP LEDGER BACKED BY BXE TOKEN!! JOURNALIST NOW CAN EARN ON XRP LEDGER!!!! NO MORE USD!! JOURNLIAST IMMEDIATELY GET PAID IN BXE TOKEN RIGHT ON XRP LEDGER!!! WHICH BY THE WAY ALSO MEANS MASSIVE BURNS OF XRP!!! BXE TOKEN AS A MASSIVE BURN!! 10 MILLION BXE TOKENS ON DECEMBER 1ST!!! FROM AN ALREADY LOW SUPPLY OF 500 MILLION TOKENS!! BXE TOKEN TRADE LINK ON XMAGNETIC: BXE TOKEN ON MEXC:

‼️‼️‼️BOOOOOOOOOOOOOOM‼️‼️ SWIFT PROCESSES $1.25 QUADRILLION TRANSACTIONS YEARLY!! IF #XRP JUST GETS 10% OF THAT SUPPLY, IT WOULD PUT XRP MARKET CAP AT $125 TRILLION!!! THATS $2250 PER XRP!!! THE MARKET SENTIMENT NEEDS TO BE ONBOARD FIRST!!! THATS WHY THE LARGEST DECENTRALIZED MEDIA PROGRAM LAUNCHED ON THE XRP LEDGER BACKED BY BXE TOKEN!! JOURNALIST NOW CAN EARN ON XRP LEDGER!!!! NO MORE USD!! JOURNLIAST IMMEDIATELY GET PAID IN BXE TOKEN RIGHT ON XRP LEDGER!!! WHICH BY THE WAY ALSO MEANS MASSIVE BURNS OF XRP!!! BXE TOKEN AS A MASSIVE BURN!! 10 MILLION BXE TOKENS ON DECEMBER 1ST!!! FROM AN ALREADY LOW SUPPLY OF 500 MILLION TOKENS!! BXE TOKEN TRADE LINK ON XMAGNETIC: BXE TOKEN ON MEXC:

FinanceBro

116,005 görüntüleme • 10 ay önce

Andrew Tate announcing his new token $TRW lets you receive percentage of profits from his business. “If you own 10% of the token, you own 10% of TRW which is $2M a month. You will be able to stake it based on the profit TRW generates.” In theory - that sounds amazing. Tokens that pass through revenues generated by the protocol/business directly to holders? That’s how every ‘governance’ token should work. But there’s a reason none of them do it...it paints a giant target on your back for the SEC. He either makes 9 figs or gets a 9 fig legal battle with the US government. Bold strategy.

Andrew Tate announcing his new token $TRW lets you receive percentage of profits from his business. “If you own 10% of the token, you own 10% of TRW which is $2M a month. You will be able to stake it based on the profit TRW generates.” In theory - that sounds amazing. Tokens that pass through revenues generated by the protocol/business directly to holders? That’s how every ‘governance’ token should work. But there’s a reason none of them do it...it paints a giant target on your back for the SEC. He either makes 9 figs or gets a 9 fig legal battle with the US government. Bold strategy.

Luke Martin

1,243,854 görüntüleme • 2 yıl önce

New weekly roundup out now! Santiago R Santos Rob Hadick ＞|＜ We discuss: - State of crypto VC - Are tokens investable? - Dragonfly ＞|＜'s $650M raise - Hyperliquid in DC, Base token launch & more! Timestamps: 00:00 Introduction 00:47 Inside Dragonfly’s $650M Raise 10:57 The State of Crypto VC 30:19 Coinbase Ad 31:03 DAS Plug 31:28 Are Tokens Investable? 40:54 Will Base Launch A Token? 45:23 Hyperliquid’s Policy Center 48:55 Content of The Week

New weekly roundup out now! Santiago R Santos Rob Hadick ＞|＜ We discuss: - State of crypto VC - Are tokens investable? - Dragonfly ＞|＜'s $650M raise - Hyperliquid in DC, Base token launch & more! Timestamps: 00:00 Introduction 00:47 Inside Dragonfly’s $650M Raise 10:57 The State of Crypto VC 30:19 Coinbase Ad 31:03 DAS Plug 31:28 Are Tokens Investable? 40:54 Will Base Launch A Token? 45:23 Hyperliquid’s Policy Center 48:55 Content of The Week

Empire 🟪

12,587 görüntüleme • 7 ay önce

$Lexi AI Fam, Our bot upgrades are now live with enhanced conversational AI features! What is Conversational AI? Conversational AI refers to technologies that enable machines to interact with humans in a natural, human-like manner. This includes understanding and processing natural language, enabling more fluid and intuitive interactions. New Features - Token Trading: Trade tokens by simply typing out the token contract and amount. No need for a fixed syntax! - Multilingual Support: You can now use any language to trade tokens. Check out the attached video demonstrating these new capabilities.

$Lexi AI Fam, Our bot upgrades are now live with enhanced conversational AI features! What is Conversational AI? Conversational AI refers to technologies that enable machines to interact with humans in a natural, human-like manner. This includes understanding and processing natural language, enabling more fluid and intuitive interactions. New Features - Token Trading: Trade tokens by simply typing out the token contract and amount. No need for a fixed syntax! - Multilingual Support: You can now use any language to trade tokens. Check out the attached video demonstrating these new capabilities.

Lexi AI

140,060 görüntüleme • 2 yıl önce

V2 is live. Reimagined from the ground up with a completely redesigned experience and new ways to launch. • Refer friends and earn 10% of the protocol fees they generate • 25% of protocol fees now power automatic daily $AST buybacks & burns • Launch tokens backed by two market positions at once • Build automatic take-profit rules directly into a token More flexible launches. Better incentives. Simpler experience. This is the we’ve been building toward. $AST 0x265b3982ea730748100947f52561a4eab54affff

V2 is live. Reimagined from the ground up with a completely redesigned experience and new ways to launch. • Refer friends and earn 10% of the protocol fees they generate • 25% of protocol fees now power automatic daily $AST buybacks & burns • Launch tokens backed by two market positions at once • Build automatic take-profit rules directly into a token More flexible launches. Better incentives. Simpler experience. This is the we’ve been building toward. $AST 0x265b3982ea730748100947f52561a4eab54affff

ast.fun

23,639 görüntüleme • 8 gün önce

A new episode of Hivemind is out now! This week we discuss the outlook for crypto and equities, token and equity value accrual, and the next AI infrastructure trade. Timestamps: 00:00 Intro & Market Outlook 09:17 Synapse, Options & Hyperliquid 20:10 Grass, Revenue & Token Economics 27:20 The Token vs Equity Debate 44:55 AI Infrastructure & The Next Trade 57:20 Why AI Is Still A Long-Term Bull Market

A new episode of Hivemind is out now! This week we discuss the outlook for crypto and equities, token and equity value accrual, and the next AI infrastructure trade. Timestamps: 00:00 Intro & Market Outlook 09:17 Synapse, Options & Hyperliquid 20:10 Grass, Revenue & Token Economics 27:20 The Token vs Equity Debate 44:55 AI Infrastructure & The Next Trade 57:20 Why AI Is Still A Long-Term Bull Market

Delphi Digital

20,472 görüntüleme • 2 ay önce

World Labs co-founders Justin Johnson and Dr. Fei-Fei Li say LLMs use next-token prediction, but spatial intelligence has its own equivalent: Justin: "The soft definition of AI-completeness is there's this fundamental primitive that's an AI task. But if I could solve this AI task in its full, broadest generality, it would solve any intelligence problem." "The classic example in LLMs is that next-token prediction is AI-complete... I think from Ilya: there's a mystery novel, the thing has to read the whole novel, and the final sentence is, 'And the killer was.' Predict the next token. You could basically frame any kind of intelligence task in terms of that." "So clearly next-token prediction is something people believe is AI-complete." "New-view prediction, this primitive that we have in Atlas, especially generative new-view prediction, is also AI-complete." "I want to have a world where Martin is writing a proof of the Riemann hypothesis on the blackboard, and then the camera pans over to the next whiteboard." Fei-Fei: "Evolution had to solve new-viewpoint prediction by making animals move. Nature gave animals eyes, but nature didn't give trees eyes. Why? Because when you move, you see a new viewpoint... We do believe very strongly that next-viewpoint prediction is the equivalent of next-token prediction." Justin Johnson Fei-Fei Li martin_casado

World Labs co-founders Justin Johnson and Dr. Fei-Fei Li say LLMs use next-token prediction, but spatial intelligence has its own equivalent: Justin: "The soft definition of AI-completeness is there's this fundamental primitive that's an AI task. But if I could solve this AI task in its full, broadest generality, it would solve any intelligence problem." "The classic example in LLMs is that next-token prediction is AI-complete... I think from Ilya: there's a mystery novel, the thing has to read the whole novel, and the final sentence is, 'And the killer was.' Predict the next token. You could basically frame any kind of intelligence task in terms of that." "So clearly next-token prediction is something people believe is AI-complete." "New-view prediction, this primitive that we have in Atlas, especially generative new-view prediction, is also AI-complete." "I want to have a world where Martin is writing a proof of the Riemann hypothesis on the blackboard, and then the camera pans over to the next whiteboard." Fei-Fei: "Evolution had to solve new-viewpoint prediction by making animals move. Nature gave animals eyes, but nature didn't give trees eyes. Why? Because when you move, you see a new viewpoint... We do believe very strongly that next-viewpoint prediction is the equivalent of next-token prediction." Justin Johnson Fei-Fei Li martin_casado

a16z

88,239 görüntüleme • 19 gün önce

In this episode of On Rails, Jason Meller (VP of Product at 1Password) talks to Robby Russell about how Rails has become one of the most token-efficient architectures for LLM-assisted development, and why that matters as token costs increasingly shape what’s practical (and profitable) to build. They also dive into security in agentic workflows, an area getting harder to ignore. Jason breaks down what 1Password’s open-source SCAM benchmark reveals about how LLMs handle credentials, where risks emerge when models act autonomously in developer environments, and what teams can do now to stay ahead. Find the full episode here or on Spotify/Apple/YouTube:

In this episode of On Rails, Jason Meller (VP of Product at 1Password) talks to Robby Russell about how Rails has become one of the most token-efficient architectures for LLM-assisted development, and why that matters as token costs increasingly shape what’s practical (and profitable) to build. They also dive into security in agentic workflows, an area getting harder to ignore. Jason breaks down what 1Password’s open-source SCAM benchmark reveals about how LLMs handle credentials, where risks emerge when models act autonomously in developer environments, and what teams can do now to stay ahead. Find the full episode here or on Spotify/Apple/YouTube:

Ruby on Rails

13,035 görüntüleme • 4 ay önce