正在加载视频...

视频加载失败

返回首页

New lecture for the book! Nominally about synthetic data, but mostly is a walk through of the distillation literature from the Hinton 2015 paper to multi-teach on-policy distillation of today! At 7.4 hours of video in my post-training brain dump and counting :) It was fun to stare at...

natolambert's profile picture

Nathan Lambert

100,908 subscribers

93,111 次观看 • 3 个月前 •via X (Twitter)

教育科学技术
LIVE

Anya Rossi• Live Now

Private livecam show

15 条评论

Nathan Lambert 的头像
Nathan Lambert3 个月前

YT:

Nathan Lambert 的头像
Nathan Lambert3 个月前

Again, the homepage is here:

forget the grind. small iterative steps. do things 的头像
forget the grind. small iterative steps. do things3 个月前

i have known about this series for weeks but haven't had time to jump in. learning today that part of what you're dealing with is the elevated way in which synthetic data can be used is quite frankly really exciting so i will find time for the whole series. thanks a lot for all this.

morgan — 的头像
morgan —3 个月前

a big thank you for these

AI Dreamer | NOEMA 的头像
AI Dreamer | NOEMA20 天前

When synthetic data becomes the main post-training substrate, provenance starts to matter as much as volume. Do current pipelines preserve enough lineage to tell whether a later behavior came from the teacher, the rubric, or the model's own earlier outputs?

Abhishek Sharma 的头像
Abhishek Sharma3 个月前

really appreciate your lectures (+ other content). thank you!

ikka 的头像
ikka3 个月前

Great work, love your work on all things data & eval :)

Kamesh 🇺🇸 的头像
Kamesh 🇺🇸3 个月前

The biggest shift may be that data is increasingly generated rather than collected.

Ibra Niang 的头像
Ibra Niang3 个月前

great learning materials 🔥

Eren | AI x Markets 的头像
Eren | AI x Markets3 个月前

If this sticks, a lot of people have to rewrite their priors fast.

Roman Kniazev 的头像
Roman Kniazev3 个月前

really admired how you didn't hide the confusion about the direction in forward/backward kl (which I guess many share) and just unpacked the definitions! thanks for the content

Scott 的头像
Scott3 个月前

Subscribed🔥

Ferbin 的头像
Ferbin3 个月前

when you're shipping agents on edge hardware, how much of this distillation work actually moves the needle?

高煜朗 的头像
高煜朗3 个月前

good job

veloX 的头像
veloX3 个月前

distillation'ın 2015'ten bugüne dönüşümü, aslında post-training'in kendisinin dönüşümü. teacher-student dinamiği başta "model sıkıştırma" idi; şimdi synthetic data üretiminin kalbi. Hinton'ın soft targets'ı, on-policy setting'de veri üretimi haline evrildi — yani model artık kendi feedback loop'unu kapatıyor.

相关视频

🚨 Jensen Huang says everyone panicked about the AI data when MOST training data was never REAL to begin with. Ilya Sutskever told the industry pre-training was over. "Ilya said, 'We're out of data,' or something like that. 'Pre-training is over,' or something like that," Huang says. "The industry panicked, you know, that this is the end of AI." "And of course, of course that's obviously not true. We're gonna keep on scaling the amount of data that we have to train with." "A lot of that data is probably gonna be synthetic." That's where the panic came from — synthetic data sounds like cheating. "Most of the data that we are training, that we teach each other with, inform each other with, is synthetic." "It's synthetic because it didn't come out of nature." "You created it. I'm consuming it. I modify it, augment it, I regenerate it, somebody else consumes it." The textbook in your hand is synthetic. The post you're reading is synthetic. The lecture you took is synthetic. Nature didn't make any of it. Humans did. AI just learned to do the same thing — faster. "Training is now limited by compute," Huang says. "Data is now limited by compute." The data wall wasn't a wall. It was a mirror. If you're new here, follow @AiEvolutio for the latest on ChatGPT, Claude, and the AI tools shaping how we work and create. — Jensen Huang ( NVIDIA ), NVIDIA CEO, on Lex Fridman's ( Lex Fridman ) podcast
1:29

🚨 Jensen Huang says everyone panicked about the AI data when MOST training data was never REAL to begin with. Ilya Sutskever told the industry pre-training was over. "Ilya said, 'We're out of data,' or something like that. 'Pre-training is over,' or something like that," Huang says. "The industry panicked, you know, that this is the end of AI." "And of course, of course that's obviously not true. We're gonna keep on scaling the amount of data that we have to train with." "A lot of that data is probably gonna be synthetic." That's where the panic came from — synthetic data sounds like cheating. "Most of the data that we are training, that we teach each other with, inform each other with, is synthetic." "It's synthetic because it didn't come out of nature." "You created it. I'm consuming it. I modify it, augment it, I regenerate it, somebody else consumes it." The textbook in your hand is synthetic. The post you're reading is synthetic. The lecture you took is synthetic. Nature didn't make any of it. Humans did. AI just learned to do the same thing — faster. "Training is now limited by compute," Huang says. "Data is now limited by compute." The data wall wasn't a wall. It was a mirror. If you're new here, follow @AiEvolutio for the latest on ChatGPT, Claude, and the AI tools shaping how we work and create. — Jensen Huang ( NVIDIA ), NVIDIA CEO, on Lex Fridman's ( Lex Fridman ) podcast

AI Evolution

15,565 次观看 • 3 个月前

I had an awesome time interviewing idan shenfeld and Jonas Hübotter from MIT and ETH Zurich about self-distillation. this very promising post-training paradigm where the model acts as its own teacher by conditioning on environment feedback or demonstrations. we cover the SDPO algo for reinforcement learning with rich feedback and SDFT for continual learning without forgetting along with many applications. we dig into how it works, why it's simpler and faster than GRPO, and where this is already showing up in production systems. table of content: 0:00 - what is self distillation 2:50 - idan (MIT) and jonas (ETH Zurich) introduction and motivation 18:40 - different perspective of on-policy self-distillation (presentation) 36:00 - metacognition and specificity in self-distillation 37:24 - very long hard task and self-distillation 42:00 - continual learning with self-distillation (presentation) 1:16:50 - what is next in this research direction? 1:20:00 - is there any experience with subjective feedbacks? 1:22:50 - quality vs number of feedbacks? 1:26:40 - what setting would self-distillation struggle vs GRPO? the slides were super crisp really cool of them to share! enjoy my guys 🌹
1:31:19

I had an awesome time interviewing idan shenfeld and Jonas Hübotter from MIT and ETH Zurich about self-distillation. this very promising post-training paradigm where the model acts as its own teacher by conditioning on environment feedback or demonstrations. we cover the SDPO algo for reinforcement learning with rich feedback and SDFT for continual learning without forgetting along with many applications. we dig into how it works, why it's simpler and faster than GRPO, and where this is already showing up in production systems. table of content: 0:00 - what is self distillation 2:50 - idan (MIT) and jonas (ETH Zurich) introduction and motivation 18:40 - different perspective of on-policy self-distillation (presentation) 36:00 - metacognition and specificity in self-distillation 37:24 - very long hard task and self-distillation 42:00 - continual learning with self-distillation (presentation) 1:16:50 - what is next in this research direction? 1:20:00 - is there any experience with subjective feedbacks? 1:22:50 - quality vs number of feedbacks? 1:26:40 - what setting would self-distillation struggle vs GRPO? the slides were super crisp really cool of them to share! enjoy my guys 🌹

Yacine Mahdid

13,109 次观看 • 4 个月前

What happens when robot world models learn from human experience at scale? 🤔 DreamDojo from NVIDIA Research is a generalist robot world model pretrained on 44K hours of egocentric human videos and then post-trained on robot data to generalize across new objects and environments. After distillation, it runs at 10 FPS for live teleoperation, policy evaluation, and model-based planning. Read the ICML paper to learn more 📄
0:32

What happens when robot world models learn from human experience at scale? 🤔 DreamDojo from NVIDIA Research is a generalist robot world model pretrained on 44K hours of egocentric human videos and then post-trained on robot data to generalize across new objects and environments. After distillation, it runs at 10 FPS for live teleoperation, policy evaluation, and model-based planning. Read the ICML paper to learn more 📄

NVIDIA Robotics

23,137 次观看 • 2 个月前

Sequoia's Dean Meyer: "If distillation is real, the question is, how do we address the asymmetry?" "Western companies aren't going to set up 35,000 proxy accounts to steal from Anthropic." "There *is* a structural asymmetry, and it's something we have to think about." "The gap between closed and open source has closed a lot." "And the key conversation online, over the past week, was distillation the reason for this? Or is there some other, mysterious reason?" "The truth is, from the conversations we've had, is that it's somewhere in the middle." "Whether it's 20% or 40%, it doesn't really matter as much. But it's meaningful that we know it's happening." "We know from the labs that there's 10s of millions of queries through these large-scale proxy distillation accounts." "We also know, if you speak to anyone who's leading in synthetic data gen on pre- or post-training at the labs, they will tell you that even if you do hardcore KYC, and then just get access to the output (let's assume you just get the output from the prompt), it is still extremely valuable." "That is 100% true. We know distillation is real."
3:21

Sequoia's Dean Meyer: "If distillation is real, the question is, how do we address the asymmetry?" "Western companies aren't going to set up 35,000 proxy accounts to steal from Anthropic." "There *is* a structural asymmetry, and it's something we have to think about." "The gap between closed and open source has closed a lot." "And the key conversation online, over the past week, was distillation the reason for this? Or is there some other, mysterious reason?" "The truth is, from the conversations we've had, is that it's somewhere in the middle." "Whether it's 20% or 40%, it doesn't really matter as much. But it's meaningful that we know it's happening." "We know from the labs that there's 10s of millions of queries through these large-scale proxy distillation accounts." "We also know, if you speak to anyone who's leading in synthetic data gen on pre- or post-training at the labs, they will tell you that even if you do hardcore KYC, and then just get access to the output (let's assume you just get the output from the prompt), it is still extremely valuable." "That is 100% true. We know distillation is real."

TBPN

43,027 次观看 • 1 个月前

A humanoid robot policy trained solely on synthetic data generated by a world model. Research Scientist Joel Jang presents NVIDIA's DreamGen pipeline: ⦿ Post-train the world model Cosmos-Predict2 with a small set of real teleoperation demos. ⦿ Prompt the world model to generate synthetic video data with verbs and scenarios not used in the world model’s post-training. ⦿ Auto-label synthetic video data with action sequences. ⦿ Train robot policies using only synthetic data. That's it. Deploy zero-shot to a real humanoid robot.
2:05

A humanoid robot policy trained solely on synthetic data generated by a world model. Research Scientist Joel Jang presents NVIDIA's DreamGen pipeline: ⦿ Post-train the world model Cosmos-Predict2 with a small set of real teleoperation demos. ⦿ Prompt the world model to generate synthetic video data with verbs and scenarios not used in the world model’s post-training. ⦿ Auto-label synthetic video data with action sequences. ⦿ Train robot policies using only synthetic data. That's it. Deploy zero-shot to a real humanoid robot.

The Humanoid Hub

20,968 次观看 • 1 年前

#WATCH | Chicago, US: Chairman of Indian Overseas Congress, Sam Pitroda says, "...This is a policy issue. Congress party would frame a policy through which the wealth distribution would be better...We don't have a minimum wage (in India)...If we come up with a minimum wage in the country saying you must pay so much money to the poor, that's the distribution of wealth. Today, rich people don't pay their peons, servants, and home help enough but they spend that money on vacation in Dubai and London...When you talk about the distribution of wealth, it is not that you sit on a chair and say I have this much money and I'll be distributing it to everybody. It's naive to think like that. The PM of a country thinks like that, that means I got some concerns about his brain...You are really dealing with the policy issues to redistribute wealth and when you ask for the data, you are really trying to understand what is the distribution today. We don't have good data on all this. I think we need data to decide on policy issues. We don't need data to distribute wealth. We need data to decide on policy issues going forward..."
3:30

#WATCH | Chicago, US: Chairman of Indian Overseas Congress, Sam Pitroda says, "...This is a policy issue. Congress party would frame a policy through which the wealth distribution would be better...We don't have a minimum wage (in India)...If we come up with a minimum wage in the country saying you must pay so much money to the poor, that's the distribution of wealth. Today, rich people don't pay their peons, servants, and home help enough but they spend that money on vacation in Dubai and London...When you talk about the distribution of wealth, it is not that you sit on a chair and say I have this much money and I'll be distributing it to everybody. It's naive to think like that. The PM of a country thinks like that, that means I got some concerns about his brain...You are really dealing with the policy issues to redistribute wealth and when you ask for the data, you are really trying to understand what is the distribution today. We don't have good data on all this. I think we need data to decide on policy issues. We don't need data to distribute wealth. We need data to decide on policy issues going forward..."

ANI

244,501 次观看 • 2 年前

New podcast with finbarr! We survey the latest post-training recipes, from GLM 5.1, Kimi K2.6, DeepSeek V4, Xiaomi MiMo V2.5, Nemotron Ultra, etc. and discuss: - Why the industry slowly shifted to multi-teacher on-policy distillation (MOPD). - What an Olmo-style recipe would need improvements in - How post-training works / suits larger organizational efforts - Career advice in the foothills of the singularity - and other topics I heard y'all wanted me to start doing this, so making some time when I'm in funemployment! Chapters: 00:00 Introduction & Olmo reflections 06:28 Post-train recipes review (history) 23:00 2026’s model recipes (MiMo Flash, DeepSeek V4, GLM 5, Kimi K2.6, etc.) 39:05 Open-ended post-training discussions 48:22 Career advice in the LLM race Links below, please follow Interconnects AI and like and subscribe and buy my book?
56:35

New podcast with finbarr! We survey the latest post-training recipes, from GLM 5.1, Kimi K2.6, DeepSeek V4, Xiaomi MiMo V2.5, Nemotron Ultra, etc. and discuss: - Why the industry slowly shifted to multi-teacher on-policy distillation (MOPD). - What an Olmo-style recipe would need improvements in - How post-training works / suits larger organizational efforts - Career advice in the foothills of the singularity - and other topics I heard y'all wanted me to start doing this, so making some time when I'm in funemployment! Chapters: 00:00 Introduction & Olmo reflections 06:28 Post-train recipes review (history) 23:00 2026’s model recipes (MiMo Flash, DeepSeek V4, GLM 5, Kimi K2.6, etc.) 39:05 Open-ended post-training discussions 48:22 Career advice in the LLM race Links below, please follow Interconnects AI and like and subscribe and buy my book?

Nathan Lambert

43,256 次观看 • 3 个月前

How to extract data from papers for literature review in seconds? 1. Go to 2. Upload 2-3 papers you already know are relevant 3. Start with clicking on one paper 4. ResearchCollab do the following for the paper ✦ Provide a short overview of the paper ✦ Identify weaknesses in the paper ✦ Share papers that contrast the existing paper ✦ Extract meta data about the paper ✦ Evaluate each part (e.g., methodology) of the paper ✦ Provide AI chat to extract any other data 5. Now click on the Related Papers button at top right 6. You will get papers related to the seed paper 7. You can click on each paper and read its abstract 8. If you find it relevant, just add it to your list of papers 9. Do the same for the other 2 seed papers 10. This way you will collect all relevant papers 11. And extract data from those papers After this, analyse the data and report the findings. Try ResearchCollab today:
0:47

How to extract data from papers for literature review in seconds? 1. Go to 2. Upload 2-3 papers you already know are relevant 3. Start with clicking on one paper 4. ResearchCollab do the following for the paper ✦ Provide a short overview of the paper ✦ Identify weaknesses in the paper ✦ Share papers that contrast the existing paper ✦ Extract meta data about the paper ✦ Evaluate each part (e.g., methodology) of the paper ✦ Provide AI chat to extract any other data 5. Now click on the Related Papers button at top right 6. You will get papers related to the seed paper 7. You can click on each paper and read its abstract 8. If you find it relevant, just add it to your list of papers 9. Do the same for the other 2 seed papers 10. This way you will collect all relevant papers 11. And extract data from those papers After this, analyse the data and report the findings. Try ResearchCollab today:

Faheem Ullah

26,494 次观看 • 7 个月前

A model is only as good as its data, and we’ve long since exhausted the internet. From here on out, model progress is gated by data production. ’s Brendan (can/do) joined us at our Sovereign AI event to talk about how RL environments get built, and why your data might be your real moat: 00:00 Introduction 00:47 A short history of the data market: crowdsourcing to agentic data 02:29 What an RL environment is: worlds, apps, tasks 03:57 Why only humans can measure the frontier 05:35 Building verifiers is the hard part 06:44 Walkthrough: a real legal RL environment 08:18 Leaderboards — and what open weights change 09:45 Post-training results on Apex Agents 11:17 Three ways companies buy data 12:49 Q&A: How do you price data? 14:17 Q&A: What "data quality" actually means 16:42 Q&A: The misunderstanding about synthetic data 18:17 Q&A: Why RL environments now — and what comes after 21:20 Q&A: Can you scale rubric generation with models? 23:00 Q&A: RL environments for cyber defense 25:33 Q&A: Build data in-house or partner?
26:21

A model is only as good as its data, and we’ve long since exhausted the internet. From here on out, model progress is gated by data production. ’s Brendan (can/do) joined us at our Sovereign AI event to talk about how RL environments get built, and why your data might be your real moat: 00:00 Introduction 00:47 A short history of the data market: crowdsourcing to agentic data 02:29 What an RL environment is: worlds, apps, tasks 03:57 Why only humans can measure the frontier 05:35 Building verifiers is the hard part 06:44 Walkthrough: a real legal RL environment 08:18 Leaderboards — and what open weights change 09:45 Post-training results on Apex Agents 11:17 Three ways companies buy data 12:49 Q&A: How do you price data? 14:17 Q&A: What "data quality" actually means 16:42 Q&A: The misunderstanding about synthetic data 18:17 Q&A: Why RL environments now — and what comes after 21:20 Q&A: Can you scale rubric generation with models? 23:00 Q&A: RL environments for cyber defense 25:33 Q&A: Build data in-house or partner?

Pat Grady

66,577 次观看 • 1 个月前

Today, we're joined by Aakanksha Chowdhery, member of technical staff at Reflection, to explore the fundamental shifts required to build true agentic AI. While the industry has largely focused on post-training techniques to improve reasoning, Aakanksha draws on her experience leading pre-training efforts for Google’s PaLM and early Gemini models to argue that pre-training itself must be rethought to move beyond static benchmarks. We explore the limitations of next-token prediction for multi-step workflows and examine how attention mechanisms, loss objectives, and training data must evolve to support long-form reasoning and planning. Aakanksha shares insights on the difference between context retrieval and actual reasoning, the importance of "trajectory" training data, and why scaling remains essential for discovering emergent agentic capabilities like error recovery and dynamic tool learning. 🗒️ For the full list of resources for this episode, visit the show notes page: 📖 CHAPTERS =============================== 00:00 - Introduction 02:26 - Reflection 04:54 - Limitations of post-training for building agents 07:31 - Rethinking pre-training in agents 10:51 - Scaling 11:27 - Evolving attention mechanisms for agentic capabilities 12:39 - Memory as a tool 14:13 - Loss objectives and training data 15:50 - Fine-tuning loss in agent performance 19:37 - Training data 21:29 - Augmenting dominant training data source 24:11 - Overcoming challenges in training on synthetic data 25:47 - Benchmarks 30:44 - Scaling laws in large models versus small models 33:20 - Long-form versus short-form reasoning 37:57 - Agent’s ability to recover from failure 40:15 - Hallucinations and failure recovery 43:53 - Tool use in agents 46:38 - Coding agents 48:37 - How researchers can contribute to agentic AI
51:54

Today, we're joined by Aakanksha Chowdhery, member of technical staff at Reflection, to explore the fundamental shifts required to build true agentic AI. While the industry has largely focused on post-training techniques to improve reasoning, Aakanksha draws on her experience leading pre-training efforts for Google’s PaLM and early Gemini models to argue that pre-training itself must be rethought to move beyond static benchmarks. We explore the limitations of next-token prediction for multi-step workflows and examine how attention mechanisms, loss objectives, and training data must evolve to support long-form reasoning and planning. Aakanksha shares insights on the difference between context retrieval and actual reasoning, the importance of "trajectory" training data, and why scaling remains essential for discovering emergent agentic capabilities like error recovery and dynamic tool learning. 🗒️ For the full list of resources for this episode, visit the show notes page: 📖 CHAPTERS =============================== 00:00 - Introduction 02:26 - Reflection 04:54 - Limitations of post-training for building agents 07:31 - Rethinking pre-training in agents 10:51 - Scaling 11:27 - Evolving attention mechanisms for agentic capabilities 12:39 - Memory as a tool 14:13 - Loss objectives and training data 15:50 - Fine-tuning loss in agent performance 19:37 - Training data 21:29 - Augmenting dominant training data source 24:11 - Overcoming challenges in training on synthetic data 25:47 - Benchmarks 30:44 - Scaling laws in large models versus small models 33:20 - Long-form versus short-form reasoning 37:57 - Agent’s ability to recover from failure 40:15 - Hallucinations and failure recovery 43:53 - Tool use in agents 46:38 - Coding agents 48:37 - How researchers can contribute to agentic AI

The TWIML AI Podcast

45,470 次观看 • 9 个月前

Elon Musk On What It Takes To Build A Competitive AI Model Elon Musk breaks down the three factors that decide whether a foundation model can compete, and why the next frontier isn't human data at all. Speaking with Garry Tan, President and CEO of Y Combinator, Elon lays out what's actually required to build a large foundation model that's competitive: "You've got to get a lot of GPUs and have them train coherently and stably. Then it's like, what unique access to data do you have? I guess distribution matters to some degree as well like, how do people get exposed to your AI? Those are critical factors." But there's a problem with the data part of that equation. Echoing what a friend in the field has said, Elon explains that the industry has essentially run out of human-generated pre-training data: "You run out of tokens pretty fast, certainly of high-quality tokens. And then you need to essentially create synthetic data, and be able to accurately judge the synthetic data that you're creating, to verify: is this real synthetic data, or is it a hallucination that doesn't actually match reality?" That verification step is the hard part: "Achieving grounding in reality is tricky. But we are at the stage where there's more effort put into synthetic data. Right now we're training Grok 3.5, which is a heavy focus on reasoning." On reasoning, Garry Tan adds an interesting detail from researchers he's spoken to: hard science, particularly physics textbooks is very useful for training reasoning, whereas social science is "totally useless" for it. Elon's response: "Yes, that's probably true." He then points to where all of this is heading: "Something that's going to be very important in the future is combining deep AI in the data center or supercluster with robotics. So, things like the Optimus humanoid robot."
2:02

Elon Musk On What It Takes To Build A Competitive AI Model Elon Musk breaks down the three factors that decide whether a foundation model can compete, and why the next frontier isn't human data at all. Speaking with Garry Tan, President and CEO of Y Combinator, Elon lays out what's actually required to build a large foundation model that's competitive: "You've got to get a lot of GPUs and have them train coherently and stably. Then it's like, what unique access to data do you have? I guess distribution matters to some degree as well like, how do people get exposed to your AI? Those are critical factors." But there's a problem with the data part of that equation. Echoing what a friend in the field has said, Elon explains that the industry has essentially run out of human-generated pre-training data: "You run out of tokens pretty fast, certainly of high-quality tokens. And then you need to essentially create synthetic data, and be able to accurately judge the synthetic data that you're creating, to verify: is this real synthetic data, or is it a hallucination that doesn't actually match reality?" That verification step is the hard part: "Achieving grounding in reality is tricky. But we are at the stage where there's more effort put into synthetic data. Right now we're training Grok 3.5, which is a heavy focus on reasoning." On reasoning, Garry Tan adds an interesting detail from researchers he's spoken to: hard science, particularly physics textbooks is very useful for training reasoning, whereas social science is "totally useless" for it. Elon's response: "Yes, that's probably true." He then points to where all of this is heading: "Something that's going to be very important in the future is combining deep AI in the data center or supercluster with robotics. So, things like the Optimus humanoid robot."

High Signal AI

21,733 次观看 • 2 个月前

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)
1:07

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

443,912 次观看 • 2 个月前

Head of Nemotron Bryan Catanzaro says NVIDIA is sitting on internal "teacher" models that beat everything it has publicly shipped and he's trying to release these as well "This technology of using teacher models and multi-domain on-policy distillation has been really key for us in these ways." "Now, what does that mean? Well, it definitely means that inside NVIDIA we have a collection of teacher models that are individually better at different domains than any of the models that we've released." "And actually, I'm trying to release those teacher models as part of the Nemotron project as well -- because as you know, Nemotron is not just a model." "Nemotron is an open technology effort. It includes data sets, it includes techniques, all of our ideas about how to build a model, which we publish." "Of course it includes the models themselves, but I think it should also include collections of teacher models." "I'd imagine this could be a really great way of federating open technologies for AI across the ecosystem." _______ Wrote more on a prior Bryan interview here:
0:51

Head of Nemotron Bryan Catanzaro says NVIDIA is sitting on internal "teacher" models that beat everything it has publicly shipped and he's trying to release these as well "This technology of using teacher models and multi-domain on-policy distillation has been really key for us in these ways." "Now, what does that mean? Well, it definitely means that inside NVIDIA we have a collection of teacher models that are individually better at different domains than any of the models that we've released." "And actually, I'm trying to release those teacher models as part of the Nemotron project as well -- because as you know, Nemotron is not just a model." "Nemotron is an open technology effort. It includes data sets, it includes techniques, all of our ideas about how to build a model, which we publish." "Of course it includes the models themselves, but I think it should also include collections of teacher models." "I'd imagine this could be a really great way of federating open technologies for AI across the ecosystem." _______ Wrote more on a prior Bryan interview here:

Fireside Alpha

29,125 次观看 • 1 个月前

Here's my conversation with Lucas Atkins and the team at Arcee.ai on their path to training and releasing Trinity Large today. From going all in on open models built end to end in the US 6 months ago to having the model in hand is no easy feet. I loved this conversation on how to design a startup around open models and take a bold step to scale it up. I'm openly an Arcee fan, watching them take risk and pull it off. We discuss: - The state (and future) of open vs. closed models, - The business of selling open models for on-prem deployments, - The story of Arcee AI & going “all-in” on this training run, - The ATOM project, - Building frontier model training teams in 6 months, - and other great topics. I really loved this one, and think you well too. Chapters: 00:00:00 Intro: Arcee AI, Trinity Models & Trinity Large 00:08:26 Transitioning a Company to Pre-training 00:13:00 Technical Decisions: Muon and MoE 00:18:41 Scaling and MoE Training Pain 00:23:14 Post-training and RL Strategies 00:28:09 Team Structure and Data Scaling 00:31:31 The Trinity Manifesto: US Open Weights 00:42:31 Specialized Models and Distillation 00:47:12 Infrastructure and Hosting 400B 00:50:53 Open Source as a Business Moat 00:56:31 Predictions: Best Model in 2026 01:02:29 Lightning Round & Conclusions More great open model builder podcasts coming soon!
1:12:15

Here's my conversation with Lucas Atkins and the team at Arcee.ai on their path to training and releasing Trinity Large today. From going all in on open models built end to end in the US 6 months ago to having the model in hand is no easy feet. I loved this conversation on how to design a startup around open models and take a bold step to scale it up. I'm openly an Arcee fan, watching them take risk and pull it off. We discuss: - The state (and future) of open vs. closed models, - The business of selling open models for on-prem deployments, - The story of Arcee AI & going “all-in” on this training run, - The ATOM project, - Building frontier model training teams in 6 months, - and other great topics. I really loved this one, and think you well too. Chapters: 00:00:00 Intro: Arcee AI, Trinity Models & Trinity Large 00:08:26 Transitioning a Company to Pre-training 00:13:00 Technical Decisions: Muon and MoE 00:18:41 Scaling and MoE Training Pain 00:23:14 Post-training and RL Strategies 00:28:09 Team Structure and Data Scaling 00:31:31 The Trinity Manifesto: US Open Weights 00:42:31 Specialized Models and Distillation 00:47:12 Infrastructure and Hosting 400B 00:50:53 Open Source as a Business Moat 00:56:31 Predictions: Best Model in 2026 01:02:29 Lightning Round & Conclusions More great open model builder podcasts coming soon!

Nathan Lambert

26,872 次观看 • 7 个月前

Your AI is only as smart as the context it runs on. Have you built your shared enterprise brain yet? At Databricks Data+AI Summit, I spoke with Prukalpa ✨, co-founder and co-CEO of Atlan, to go deeper on the AI Context Layer in practice. The concept of a shared enterprise brain resonated with me. It is a foundation that every AI agent draws from, learns from, and contributes back to. "The future of organizations are going to be the fastest learning organizations." - Prukalpa ✨ Sankar, co-founder and co-CEO of Atlan Watch the full interview on YouTube: Check out the live demo by Atlan on Building Context for Your AI Agents - #EnterpriseAI #ContextLayer #Data #AI #AtlanPartner Matthias Heidrich
2:03

Your AI is only as smart as the context it runs on. Have you built your shared enterprise brain yet? At Databricks Data+AI Summit, I spoke with Prukalpa ✨, co-founder and co-CEO of Atlan, to go deeper on the AI Context Layer in practice. The concept of a shared enterprise brain resonated with me. It is a foundation that every AI agent draws from, learns from, and contributes back to. "The future of organizations are going to be the fastest learning organizations." - Prukalpa ✨ Sankar, co-founder and co-CEO of Atlan Watch the full interview on YouTube: Check out the live demo by Atlan on Building Context for Your AI Agents - #EnterpriseAI #ContextLayer #Data #AI #AtlanPartner Matthias Heidrich

Helen Yu

10,982 次观看 • 2 个月前

Unreal Engine 5 is becoming the go-to platform for robotics teams, not just as a simulator but as a synthetic data factory powering the next generation of robotics systems🦾 From synthetic data to sim-to-real transfer, read our piece on how UE5 is quietly reshaping how robots learn to see, move, and act in the real world:
0:36

Unreal Engine 5 is becoming the go-to platform for robotics teams, not just as a simulator but as a synthetic data factory powering the next generation of robotics systems🦾 From synthetic data to sim-to-real transfer, read our piece on how UE5 is quietly reshaping how robots learn to see, move, and act in the real world:

Unreal Engine

61,187 次观看 • 5 个月前

Today we're announcing Base Labs, a dedicated research organization focused on advancing open-source AI. We believe in a healthy, open frontier model ecosystem. To enable this, we are working on: - Blue-sky research on continual learning, the science of RL, and how models learn across their full lifecycle, with every experiment and recipe shared openly. - The BaseHub Data Foundry: the highest-quality open RL environments, training data, and real-world benchmarks, built for anyone to train and benchmark on. - Post-post training: taking open-source models and making them better, safer, and more aligned through continual post-training, built on our research, and deployed with our frontier safety stack so organizations can use open-source models with confidence. - Making models cheaper and more performant through our model performance research. This is a mission-driven research effort, not a commercial product. We believe the health of the open-source AI ecosystem matters and that the best way to advance it is to do science in the open. We’re hiring engineers, researchers, and research fellows to advance this mission.
1:05

Today we're announcing Base Labs, a dedicated research organization focused on advancing open-source AI. We believe in a healthy, open frontier model ecosystem. To enable this, we are working on: - Blue-sky research on continual learning, the science of RL, and how models learn across their full lifecycle, with every experiment and recipe shared openly. - The BaseHub Data Foundry: the highest-quality open RL environments, training data, and real-world benchmarks, built for anyone to train and benchmark on. - Post-post training: taking open-source models and making them better, safer, and more aligned through continual post-training, built on our research, and deployed with our frontier safety stack so organizations can use open-source models with confidence. - Making models cheaper and more performant through our model performance research. This is a mission-driven research effort, not a commercial product. We believe the health of the open-source AI ecosystem matters and that the best way to advance it is to do science in the open. We’re hiring engineers, researchers, and research fellows to advance this mission.

Base Labs

236,375 次观看 • 20 天前

Human generated data has fueled incredible AI progress, but what comes next? 📈 On the latest episode of our podcast, Hannah Fry and David Silver, VP of Reinforcement Learning, talk about how we could move from the era of relying on human data to one where AI could learn for itself. Watch now → 00:00 Introduction 01:50 Era of experience 03:45 AlphaZero 10:19 Move 37 15:20 Reinforcement learning and human feedback 24:30 AlphaProof 29:50 Math Olympiads 35:00 Experience based methods 42:56 Hannah's reflections 44:00 Fan Hui joins
49:30

Human generated data has fueled incredible AI progress, but what comes next? 📈 On the latest episode of our podcast, Hannah Fry and David Silver, VP of Reinforcement Learning, talk about how we could move from the era of relying on human data to one where AI could learn for itself. Watch now → 00:00 Introduction 01:50 Era of experience 03:45 AlphaZero 10:19 Move 37 15:20 Reinforcement learning and human feedback 24:30 AlphaProof 29:50 Math Olympiads 35:00 Experience based methods 42:56 Hannah's reflections 44:00 Fan Hui joins

Google DeepMind

228,029 次观看 • 1 年前

Hot take from my panel on decentralized AI at UC Berkeley: Your data is worthless. The idea that you'll get rich selling your data to train AI is a fantasy. The frontier is high-quality synthetic data, not your dumb tweets. Second hot take: the government censorship of AI has not started yet. It will come soon. And when it does, decentralized AI will matter in the same way that the rise of financial surveillance made Bitcoin matter. Spicy back and forth with Avichal - Electric ϟ Capital and Franklin Bi 👇
33:40

Hot take from my panel on decentralized AI at UC Berkeley: Your data is worthless. The idea that you'll get rich selling your data to train AI is a fantasy. The frontier is high-quality synthetic data, not your dumb tweets. Second hot take: the government censorship of AI has not started yet. It will come soon. And when it does, decentralized AI will matter in the same way that the rise of financial surveillance made Bitcoin matter. Spicy back and forth with Avichal - Electric ϟ Capital and Franklin Bi 👇

Haseeb ＞|＜

81,012 次观看 • 1 年前

0.32 has shipped, and it's a massive release from Rerun. There's a ton of cool new features, and I wanted to highlight 2 in particular 1. OSS Server streaming from disk 2. Dataset review I walk you through them in the video, so take a look. I'll have a much longer blog post next week about the entire pipeline. With 0.32, much of the foundation is set for a unified data layer for physical data, and I'll be getting into the details of it with all that I've built over the past year. This will cover 1. Raw Data Collection 2. Data Ingestion 3. Catalog Registration 4. Query and Review 5. Post Process 6. Training so lots to share
1:54

0.32 has shipped, and it's a massive release from Rerun. There's a ton of cool new features, and I wanted to highlight 2 in particular 1. OSS Server streaming from disk 2. Dataset review I walk you through them in the video, so take a look. I'll have a much longer blog post next week about the entire pipeline. With 0.32, much of the foundation is set for a unified data layer for physical data, and I'll be getting into the details of it with all that I've built over the past year. This will cover 1. Raw Data Collection 2. Data Ingestion 3. Catalog Registration 4. Query and Review 5. Post Process 6. Training so lots to share

Pablo Vela

11,264 次观看 • 4 个月前