Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

Are AI models training themselves? Not yet—but Google DeepMind is getting closer. At our AI summit, Google DeepMind chief Koray Kavukcuoglu said the company now trusts AI agents to autonomously run parts of the model-training process: conducting experiments, analyzing results and proposing new hypotheses, all under human supervision. Six... show more

The Information

1,569,502 subscribers

28,077 views • 18 days ago •via X (Twitter)

Anya Rossi• Live Now

Private livecam show

3 Comments

VastPlan17 days ago

据 The Information AI 峰会：DeepMind 负责人 Koray Kavukcuoglu 表示，公司已信任人工智能代理自主运行模型训练流程中的部分环节，包括开展实验与分析结果；端到端「模型自我训练」尚未成立。真正需要核验的是，哪些实验链条可以在无人值守下完成，以及在后训练与安全工作仍需人工签字时，闸门如何被审计。

Praveen Jatta17 days ago

Google DeepMind now trusts AI agents to train themselves — a landmark capability moment. And exactly why governance frameworks need to arrive BEFORE capability surges past oversight. The question isn't 'can AI train itself?' It's 'who's watching?' #AIPolicy #AIGovernance

Iris晴柔17 days ago

AI models training themselves: the ultimate 'I'll just do it myself.' 🤖

Related Videos

A LOT happened last week. ICYMI, koray kavukcuoglu (CTO of Google DeepMind and Chief AI Architect of Google) and Logan Kilpatrick discuss Gemini 3, the state of AI, and where we are on the path to AGI. Chapters: 0:00 - Intro 2:00 - Gemini 3 launch reception 4:16 - Continuous progress and innovation 6:47 - Key areas for Gemini improvement 11:45 - Product scaffolding for model improvement 13:56 - Chief AI architect role 17:04 - Engineering mindset and collaboration 18:37 - Future growth areas for Gemini 20:33 - From research to engineering mindset 23:22 - The rise of generative media 27:22 - Nano Banana Pro capabilities 29:31 - Towards unified model checkpoints 36:26 - Organizing for AI success 38:26 - Balancing exploration and scaling 41:40 - DeepMind's collaborative culture 45:21 - Innovating at Google 48:37 - Closing

A LOT happened last week. ICYMI, koray kavukcuoglu (CTO of Google DeepMind and Chief AI Architect of Google) and Logan Kilpatrick discuss Gemini 3, the state of AI, and where we are on the path to AGI. Chapters: 0:00 - Intro 2:00 - Gemini 3 launch reception 4:16 - Continuous progress and innovation 6:47 - Key areas for Gemini improvement 11:45 - Product scaffolding for model improvement 13:56 - Chief AI architect role 17:04 - Engineering mindset and collaboration 18:37 - Future growth areas for Gemini 20:33 - From research to engineering mindset 23:22 - The rise of generative media 27:22 - Nano Banana Pro capabilities 29:31 - Towards unified model checkpoints 36:26 - Organizing for AI success 38:26 - Balancing exploration and scaling 41:40 - DeepMind's collaborative culture 45:21 - Innovating at Google 48:37 - Closing

Google AI

181,514 views • 10 months ago

Raiinmaker x Google marks a breakthrough in AI video model training. Over 450,000 contributors. 190+ countries. Real people. Real data. Real results. You can now train AI and generate beautiful, cinematic video all from your phone. It’s fast, it’s easy, it’s next-level. This is what the future looks like: Raiinmaker’s decentralized Video Intelligence platform Traiin Video, together with Google’s cutting-edge Veo technology = a new standard for ethical AI video training. Ethical AI. Human-powered. Built to win. Make AI Yours. Make it $RAIIN. Google Cloud Google Cloud Partners Google AI #GoogleCloudPartner #GoogleCloud

Raiinmaker x Google marks a breakthrough in AI video model training. Over 450,000 contributors. 190+ countries. Real people. Real data. Real results. You can now train AI and generate beautiful, cinematic video all from your phone. It’s fast, it’s easy, it’s next-level. This is what the future looks like: Raiinmaker’s decentralized Video Intelligence platform Traiin Video, together with Google’s cutting-edge Veo technology = a new standard for ethical AI video training. Ethical AI. Human-powered. Built to win. Make AI Yours. Make it $RAIIN. Google Cloud Google Cloud Partners Google AI #GoogleCloudPartner #GoogleCloud

Raiinmaker

301,698 views • 1 year ago

Ex-Google Jeff Dean: "Prompting and one agent won't be enough in 6 months. Loops and Graphs with teams of agents are replacing them" This is 27 years of AI inside Google compressed into one hour workshop, by the person who lived it. He describes agents that run 50 experiments, keep 10, and improve themselves - no course sells this yet. Watch it, then read the guide below on building a graph that runs without you.

Ex-Google Jeff Dean: "Prompting and one agent won't be enough in 6 months. Loops and Graphs with teams of agents are replacing them" This is 27 years of AI inside Google compressed into one hour workshop, by the person who lived it. He describes agents that run 50 experiments, keep 10, and improve themselves - no course sells this yet. Watch it, then read the guide below on building a graph that runs without you.

codila

55,003 views • 2 months ago

Google just released a free 1-hour lecture from Chief Scientist Jeff Dean on full AI Engineering. How to go from one LLM to coordinating 100 AI agents: 01:45 - Build an LLM from scratch 17:22 - Learn how to actually use AI models 30:03 - Master prompt engineering 52:35 - Coordinate 100 agents as one human Most people are still learning how to prompt one model. Jeff Dean is showing the full stack: LLMs → Prompts → Agents → Agent Teams → Graphs Prompting is the old workflow. Coordinating agent systems is the new one. This is 27 years of Google AI compressed into one hour. Bookmark and watch it today Then read the full Graph Engineering guide below ↓

Google just released a free 1-hour lecture from Chief Scientist Jeff Dean on full AI Engineering. How to go from one LLM to coordinating 100 AI agents: 01:45 - Build an LLM from scratch 17:22 - Learn how to actually use AI models 30:03 - Master prompt engineering 52:35 - Coordinate 100 agents as one human Most people are still learning how to prompt one model. Jeff Dean is showing the full stack: LLMs → Prompts → Agents → Agent Teams → Graphs Prompting is the old workflow. Coordinating agent systems is the new one. This is 27 years of Google AI compressed into one hour. Bookmark and watch it today Then read the full Graph Engineering guide below ↓

Lunar

102,967 views • 2 months ago

Demis Hassabis, Nobel laureate and CEO of Google DeepMind: "I think one day, maybe we can cure all disease with the help of AI. Within the next decade. I don't see why not." This is not a hype account. It is the man whose AI mapped 200 million protein structures in a single year, work that used to take a lifetime. A new drug normally takes ten years and billions of dollars. He thinks AI cuts that to months, maybe weeks. Musk said your future doctor will be AI. Hassabis is saying the disease itself might not be waiting for you by then. We treated illness as the fixed price of being alive. Two Nobel-level labs are now quietly saying that price was a choice.

Demis Hassabis, Nobel laureate and CEO of Google DeepMind: "I think one day, maybe we can cure all disease with the help of AI. Within the next decade. I don't see why not." This is not a hype account. It is the man whose AI mapped 200 million protein structures in a single year, work that used to take a lifetime. A new drug normally takes ten years and billions of dollars. He thinks AI cuts that to months, maybe weeks. Musk said your future doctor will be AI. Hassabis is saying the disease itself might not be waiting for you by then. We treated illness as the fixed price of being alive. Two Nobel-level labs are now quietly saying that price was a choice.

Casper

26,292 views • 2 months ago

“I used to take the Coromandel Express from Chennai to IIT Kharagpur, and it would pass through Vizag. I remember it as a quiet coastal city. And now, Google is launching a full-stack AI hub there as part of our investment in India. This will bring jobs to Vizag,” said Sundar Pichai, CEO, Google at the India–AI Impact Summit 2026. “I never imagined Vizag would become a Google hub back then,” he added. Sundar Pichai Google India Google

“I used to take the Coromandel Express from Chennai to IIT Kharagpur, and it would pass through Vizag. I remember it as a quiet coastal city. And now, Google is launching a full-stack AI hub there as part of our investment in India. This will bring jobs to Vizag,” said Sundar Pichai, CEO, Google at the India–AI Impact Summit 2026. “I never imagined Vizag would become a Google hub back then,” he added. Sundar Pichai Google India Google

CNBC-TV18

72,476 views • 7 months ago

The most skilled guy in the AI industry just said we're 1-2 breakthroughs away from AGI. And he explained exactly what's missing. Demis Hassabis runs Google DeepMind. He won the Nobel Prize in Chemistry last year. He's literally the reason why Google is considered the leader of the AI race. And he just dropped the most specific AGI timeline ever: "One or two AlphaGo-level technological breakthroughs." That's it. That's all standing between us and artificial general intelligence. But here's the thing... LLMs are NOT going to get us there. ChatGPT, Gemini, Claude - they're all hitting the same wall. They can't plan long-term. Can't create NEW ideas. Can't understand physics. Demis called them "jagged intelligences. Very good at certain things. Completely incapable of others." You've felt this yourself. You've felt this yourself. You ask ChatGPT a complex question and it sounds smart. But ask it to solve something that requires REASONING across multiple steps? It falls apart. So what ARE the 2 breakthroughs we need? Breakthrough #1: World Models AI that understands how physics actually works. How water flows. How cause and effect works. DeepMind already has early versions (Genie, Veo). The insight: If AI can GENERATE something realistic, it UNDERSTANDS it. This is the foundation for robotics and AI that interacts with reality. Breakthrough #2: Agentic Systems AI that can DO things. Not just answer questions. Plan multiple steps. Execute autonomously. Adjust when wrong. DeepMind proved this with AlphaGo in 2016 - planning 20+ moves ahead to beat the world champion. Now they're generalizing it to the real world. And here's the most interesting part: Demis says these two things are starting to CONVERGE. LLMs + World Models + Agentic Behavior = AGI And when I say converge, I mean Google is already building it. They're setting up the first fully automated scientific laboratory in the UK. No humans running experiments. AI designs the test. Robots execute it. AI analyzes results. AI adjusts and iterates. The lab will work on: → Room-temperature superconductors → Nuclear fusion materials → New battery chemistries → Climate tech breakthroughs Demis's logic is simple: "If AI can screen materials 100X faster, the energy revolution takes 10 years instead of 100." But here's the scary part: China is MONTHS behind. Not years. "They're very close to the frontier. Maybe only months behind." DeepSeek. Alibaba's Qwen models. They're catching up fast. And unlike what people thought, they're doing it WITHOUT access to the most advanced Nvidia chips. The window for the West to lead in AGI is shrinking. The economic impact? Demis: "10 times bigger than the Industrial Revolution. And maybe 10 times faster." Industrial Revolution took 100+ years and reshaped civilization. This will be 10X bigger in 1/10th the time. Mass job displacement. Economic restructuring. New industries overnight. But also: → Curing all disease → Solving climate change → Unlimited clean energy → "Radical abundance" Demis is betting DeepMind can get there first. Google spent $400 million on DeepMind in 2014. That stake is now worth 100s of billions. Because DeepMind is now the "engine room" of ALL of Google's AI. Every Gemini model. Every AI feature in Search, Gmail, Workspace. All built by DeepMind. Shipped across Google's dozens of billion-user products instantly. That distribution is their superpower. The final thing Demis said that stuck with me: "AGI is probably the most transformative moment in human history. And it's on the horizon." One or two breakthroughs and 5 years away. According to the most skilled guy in the industry.

The most skilled guy in the AI industry just said we're 1-2 breakthroughs away from AGI. And he explained exactly what's missing. Demis Hassabis runs Google DeepMind. He won the Nobel Prize in Chemistry last year. He's literally the reason why Google is considered the leader of the AI race. And he just dropped the most specific AGI timeline ever: "One or two AlphaGo-level technological breakthroughs." That's it. That's all standing between us and artificial general intelligence. But here's the thing... LLMs are NOT going to get us there. ChatGPT, Gemini, Claude - they're all hitting the same wall. They can't plan long-term. Can't create NEW ideas. Can't understand physics. Demis called them "jagged intelligences. Very good at certain things. Completely incapable of others." You've felt this yourself. You've felt this yourself. You ask ChatGPT a complex question and it sounds smart. But ask it to solve something that requires REASONING across multiple steps? It falls apart. So what ARE the 2 breakthroughs we need? Breakthrough #1: World Models AI that understands how physics actually works. How water flows. How cause and effect works. DeepMind already has early versions (Genie, Veo). The insight: If AI can GENERATE something realistic, it UNDERSTANDS it. This is the foundation for robotics and AI that interacts with reality. Breakthrough #2: Agentic Systems AI that can DO things. Not just answer questions. Plan multiple steps. Execute autonomously. Adjust when wrong. DeepMind proved this with AlphaGo in 2016 - planning 20+ moves ahead to beat the world champion. Now they're generalizing it to the real world. And here's the most interesting part: Demis says these two things are starting to CONVERGE. LLMs + World Models + Agentic Behavior = AGI And when I say converge, I mean Google is already building it. They're setting up the first fully automated scientific laboratory in the UK. No humans running experiments. AI designs the test. Robots execute it. AI analyzes results. AI adjusts and iterates. The lab will work on: → Room-temperature superconductors → Nuclear fusion materials → New battery chemistries → Climate tech breakthroughs Demis's logic is simple: "If AI can screen materials 100X faster, the energy revolution takes 10 years instead of 100." But here's the scary part: China is MONTHS behind. Not years. "They're very close to the frontier. Maybe only months behind." DeepSeek. Alibaba's Qwen models. They're catching up fast. And unlike what people thought, they're doing it WITHOUT access to the most advanced Nvidia chips. The window for the West to lead in AGI is shrinking. The economic impact? Demis: "10 times bigger than the Industrial Revolution. And maybe 10 times faster." Industrial Revolution took 100+ years and reshaped civilization. This will be 10X bigger in 1/10th the time. Mass job displacement. Economic restructuring. New industries overnight. But also: → Curing all disease → Solving climate change → Unlimited clean energy → "Radical abundance" Demis is betting DeepMind can get there first. Google spent $400 million on DeepMind in 2014. That stake is now worth 100s of billions. Because DeepMind is now the "engine room" of ALL of Google's AI. Every Gemini model. Every AI feature in Search, Gmail, Workspace. All built by DeepMind. Shipped across Google's dozens of billion-user products instantly. That distribution is their superpower. The final thing Demis said that stuck with me: "AGI is probably the most transformative moment in human history. And it's on the horizon." One or two breakthroughs and 5 years away. According to the most skilled guy in the industry.

Ricardo

217,371 views • 8 months ago

Prodigy Research (YC S26) is the frontier AI trading research lab. We’re training the world’s best foundation model for quantitative finance. We've already achieved 140% returns in live trading over our YC batch, while major indices were flat or down. Prodigy’s model beats Claude Fable and GPT-5.6 Sol at autonomous quant research. Our founders are brothers who have worked at Jane Street, trained SoTA foundation models at Google DeepMind and shipped cutting-edge AI to billions of users at Apple. A new era has arrived in the world of quantitative finance. To learn more, contact founders@prodigy.markets

Prodigy Research (YC S26) is the frontier AI trading research lab. We’re training the world’s best foundation model for quantitative finance. We've already achieved 140% returns in live trading over our YC batch, while major indices were flat or down. Prodigy’s model beats Claude Fable and GPT-5.6 Sol at autonomous quant research. Our founders are brothers who have worked at Jane Street, trained SoTA foundation models at Google DeepMind and shipped cutting-edge AI to billions of users at Apple. A new era has arrived in the world of quantitative finance. To learn more, contact [email protected]

Prodigy Research (YC S26)

24,667 views • 1 month ago

$GOOG Koray Kavukcuoglu says he is 100% certain Google DeepMind will get back to the frontier and that nothing else is more important "To put it very bluntly, there's nothing other than being at the frontier that is important for us. Yeah, that's it. All our goals, all our focus is always about that." "And when we are making decisions, when we are making prioritizations in terms of ideas, in terms of what to do and what goals we should follow, it's all about being at the frontier." "And I'm a hundred percent certain that we will be at the frontier." " We have an amazing team. I have so much trust in the team. There's like a great amount of creativity and dedication in the team, and being part of Google, we have amazing, amazing resources available to us. We have the whole full stack that we can, and we are able to optimize to be able to achieve our goals." ________ For a deeper look at Gemini 4 progression across multiple Google exec interviews:

$GOOG Koray Kavukcuoglu says he is 100% certain Google DeepMind will get back to the frontier and that nothing else is more important "To put it very bluntly, there's nothing other than being at the frontier that is important for us. Yeah, that's it. All our goals, all our focus is always about that." "And when we are making decisions, when we are making prioritizations in terms of ideas, in terms of what to do and what goals we should follow, it's all about being at the frontier." "And I'm a hundred percent certain that we will be at the frontier." " We have an amazing team. I have so much trust in the team. There's like a great amount of creativity and dedication in the team, and being part of Google, we have amazing, amazing resources available to us. We have the whole full stack that we can, and we are able to optimize to be able to achieve our goals." ________ For a deeper look at Gemini 4 progression across multiple Google exec interviews:

Fireside Alpha

72,195 views • 23 days ago

How many AI agents work at your company? We now have over 3,258 agents working alongside 1,300 humans. The crazy part is these agents were created by EVERY EMPLOYEE at our company... sales reps, marketers, customer support, product, eng. Literally EVERYONE. BUT I'm most surprised by the adoption and value that MANAGERS are getting from agents. I used to think that every IC would become a manager of agents. Now I think that managers will very likely manage WAY more agents than their ICs combined. And managers' agents will manage their ICs' agents - overseeing them for human-in-the-loop interactions. When creating agents, we use 100% context from all of your activity, files edited, tasks and projects worked on, hierarchy, skills, and role information. We build a user-based context model to make agents as relatable as possible to the specific human that we're building for. This means they truly understand the nuances of the work and what "great" looks like - because great is very much in the eye of the beholder. Great is by definition, subjective. This is also why the human ENGAGEMENT loops are SO vital to agent value. The iteration AFTER the agent is onboarded is where the MAGIC happens. This is just like a manager managing an IC in real life... you're giving feedback. In this case, though, agents learn INSTANTLY, and they retain the knowledge perfectly and indefinitely. Even though I've been pushing AI for years now to everyone in our company, this was the first time we had truly end-to-end AI adoption and retention. This kind of AI adoption is wild. But the value we're realizing is truly INSANE. Super Agents outnumber our humans nearly 3 to 1. What if you could 3X your workforce overnight? Watch this video to see how 👇

How many AI agents work at your company? We now have over 3,258 agents working alongside 1,300 humans. The crazy part is these agents were created by EVERY EMPLOYEE at our company... sales reps, marketers, customer support, product, eng. Literally EVERYONE. BUT I'm most surprised by the adoption and value that MANAGERS are getting from agents. I used to think that every IC would become a manager of agents. Now I think that managers will very likely manage WAY more agents than their ICs combined. And managers' agents will manage their ICs' agents - overseeing them for human-in-the-loop interactions. When creating agents, we use 100% context from all of your activity, files edited, tasks and projects worked on, hierarchy, skills, and role information. We build a user-based context model to make agents as relatable as possible to the specific human that we're building for. This means they truly understand the nuances of the work and what "great" looks like - because great is very much in the eye of the beholder. Great is by definition, subjective. This is also why the human ENGAGEMENT loops are SO vital to agent value. The iteration AFTER the agent is onboarded is where the MAGIC happens. This is just like a manager managing an IC in real life... you're giving feedback. In this case, though, agents learn INSTANTLY, and they retain the knowledge perfectly and indefinitely. Even though I've been pushing AI for years now to everyone in our company, this was the first time we had truly end-to-end AI adoption and retention. This kind of AI adoption is wild. But the value we're realizing is truly INSANE. Super Agents outnumber our humans nearly 3 to 1. What if you could 3X your workforce overnight? Watch this video to see how 👇

Zeb Evans

425,552 views • 8 months ago

Demis Hassabis says AI is taking coding and maths first not because they are easy, but because you can check the answer: Asked at Davos whether he still expected a 50% chance of human-level AI by the end of the decade, he said his timeline had not moved: "Yeah, look, I think I'm still on the same kind of timeline. I think there has been remarkable progress." "But I think some areas of kind of engineering work, coding, or so you could say mathematics, are a little bit easier to see how they would be automated, partly because they're verifiable what the output is." "Some areas of natural science are much harder to do than that. You won't necessarily know if the chemical compound you've built or this prediction about physics is correct." "It may be, you may have to test it experimentally and that will all take longer." Hassabis runs Google DeepMind, which builds in both the coding and the science domains he is ranking here. - Demis Hassabis (Demis Hassabis), CEO of Google DeepMind (Google DeepMind), at the World Economic Forum in Davos, January 2026.

Demis Hassabis says AI is taking coding and maths first not because they are easy, but because you can check the answer: Asked at Davos whether he still expected a 50% chance of human-level AI by the end of the decade, he said his timeline had not moved: "Yeah, look, I think I'm still on the same kind of timeline. I think there has been remarkable progress." "But I think some areas of kind of engineering work, coding, or so you could say mathematics, are a little bit easier to see how they would be automated, partly because they're verifiable what the output is." "Some areas of natural science are much harder to do than that. You won't necessarily know if the chemical compound you've built or this prediction about physics is correct." "It may be, you may have to test it experimentally and that will all take longer." Hassabis runs Google DeepMind, which builds in both the coding and the science domains he is ranking here. - Demis Hassabis (Demis Hassabis), CEO of Google DeepMind (Google DeepMind), at the World Economic Forum in Davos, January 2026.

Karl Mehta

34,453 views • 21 days ago

I’m pretty nervous to post this… but every frame, every sound, my voice, even ME…is not real. This entire commercial was generated by AI. I remember doing property tours to sell real estate on YouTube 12 years ago and everyone said they were dumb. I remember when we opened SERHANT. and launched SERHANT. Studios, the first of its kind production company for real estate, and everyone said we weren’t serious. Today, we take a major leap forward in how we present real estate to the world. Welcome to the brokerage of tomorrow. Created by the head of SERHANT. Studios Property Tours Skyler using Google DeepMind #veo3 Kling AI ChatGPT

I’m pretty nervous to post this… but every frame, every sound, my voice, even ME…is not real. This entire commercial was generated by AI. I remember doing property tours to sell real estate on YouTube 12 years ago and everyone said they were dumb. I remember when we opened SERHANT. and launched SERHANT. Studios, the first of its kind production company for real estate, and everyone said we weren’t serious. Today, we take a major leap forward in how we present real estate to the world. Welcome to the brokerage of tomorrow. Created by the head of SERHANT. Studios Property Tours Skyler using Google DeepMind #veo3 Kling AI ChatGPT

Ryan Serhant

19,893 views • 1 year ago

Jordy Visser Jordi Visser: "Google's not a player in this anymore." Visser is an AI-native power user who tracks his own model usage as a leading indicator. Two months ago Claude was 35% of his usage, ChatGPT 25%, Google 20%. Now ChatGPT and Claude are 90% combined and Gemini is near zero. The same week, two senior DeepMind scientists left for Anthropic. He would never short Google. But multiple compression is already underway, the same pattern that hit Meta and Microsoft. A business does not have to fail to destroy shareholder value. Why the AI race narrowed to two: Source: Anthony Pompliano Podcast -

Jordy Visser Jordi Visser: "Google's not a player in this anymore." Visser is an AI-native power user who tracks his own model usage as a leading indicator. Two months ago Claude was 35% of his usage, ChatGPT 25%, Google 20%. Now ChatGPT and Claude are 90% combined and Gemini is near zero. The same week, two senior DeepMind scientists left for Anthropic. He would never short Google. But multiple compression is already underway, the same pattern that hit Meta and Microsoft. A business does not have to fail to destroy shareholder value. Why the AI race narrowed to two: Source: Anthony Pompliano Podcast -

Podcast Alpha

32,769 views • 3 months ago

Google just released a free 1-hour lecture from Chief Scientist Jeff Dean on full AI Engineering. How to go from one LLM to coordinating 100 AI agents: 0% → 01:45 - build an LLM from scratch 30% → 17:22 - learn how to actually use AI models 60% → 30:03 - master prompt engineering 100% → 52:35 - coordinate 100 agents as one human Most people are still learning how to prompt a single model. Jeff Dean is showing the entire stack: LLMs → Prompts → Agents → Agent Teams → Graphs Prompting is the old workflow. Coordinating entire agent systems is the next one. This is 27 years of Google AI compressed into a single hour. Bookmark it and watch today. Then read the full Graph Engineering guide below ↓

Google just released a free 1-hour lecture from Chief Scientist Jeff Dean on full AI Engineering. How to go from one LLM to coordinating 100 AI agents: 0% → 01:45 - build an LLM from scratch 30% → 17:22 - learn how to actually use AI models 60% → 30:03 - master prompt engineering 100% → 52:35 - coordinate 100 agents as one human Most people are still learning how to prompt a single model. Jeff Dean is showing the entire stack: LLMs → Prompts → Agents → Agent Teams → Graphs Prompting is the old workflow. Coordinating entire agent systems is the next one. This is 27 years of Google AI compressed into a single hour. Bookmark it and watch today. Then read the full Graph Engineering guide below ↓

Lunar

25,192 views • 1 month ago

Our animated short film "Dear Upstairs Neighbors" is previewing today at Sundance Film Festival! In creating this film, our Google DeepMind team of Pixar alumni, an Academy Award winner, researchers, and engineers designed new AI capabilities specifically for filmmakers. Here’s how these AI capabilities played a supporting role (pun intended) to human-led creativity: — Custom Training: Veo and Imagen models were trained on the team's original artwork and paintings — Creative control: The team created the story, then AI was used to transform rough animations into stylized videos — Precision editing: The technology allowed the artists to make specific edits without needing to recreate entire shots Learn more about how "Dear Upstairs Neighbors" was created in this behind-the-scenes video with director Connie He 🍿

Our animated short film "Dear Upstairs Neighbors" is previewing today at Sundance Film Festival! In creating this film, our Google DeepMind team of Pixar alumni, an Academy Award winner, researchers, and engineers designed new AI capabilities specifically for filmmakers. Here’s how these AI capabilities played a supporting role (pun intended) to human-led creativity: — Custom Training: Veo and Imagen models were trained on the team's original artwork and paintings — Creative control: The team created the story, then AI was used to transform rough animations into stylized videos — Precision editing: The technology allowed the artists to make specific edits without needing to recreate entire shots Learn more about how "Dear Upstairs Neighbors" was created in this behind-the-scenes video with director Connie He 🍿

Google AI

180,207 views • 8 months ago

$AI agents are about to redefine the internet. The mistake we made with Large Language Models? We let a handful of corporations capture all the value. Action Model is building a different path. By training through our extension, users gain fractional ownership in the Large Action Model, giving them a real stake in the future of AI. When LLMs emerged, the upside flowed to Big Tech. This time, it doesn’t have to. They’re building AI on our data, and keeping the upside for themselves. Community-owned Large Action Model is how we take it back.$

AI agents are about to redefine the internet. The mistake we made with Large Language Models? We let a handful of corporations capture all the value. Action Model is building a different path. By training through our extension, users gain fractional ownership in the Large Action Model, giving them a real stake in the future of AI. When LLMs emerged, the upside flowed to Big Tech. This time, it doesn’t have to. They’re building AI on our data, and keeping the upside for themselves. Community-owned Large Action Model is how we take it back.

Action Model

78,025 views • 7 months ago

Today, we’re excited to announce our $50M Series B, led by Greenfield Partners, with participation from Lightspeed and Notable Capital. 🚀 At Patronus AI, we develop simulations and evals to train and improve AI. The first phase of AI was built on static benchmarks, but that era is over. As agents are used to solve longer and longer tasks, they need to practice in dynamic, living worlds to get better. Simulations are the critical infrastructure powering this next phase. As a company, we’re behind the most influential research and products in AI evaluation, like FinanceBench, Lynx, and Percival. And things have moved at the speed of light since.⚡ We partner with the world's leading frontier AI labs and enterprises, and our revenue has grown more than 15x over the past year. Additionally, today, we’re introducing a preview of the first Digital World Model for AI agent training and simulation: Patronus-DWM. Digital World Models are language diffusion world models that predict realistic environment behaviors and steer agent actions across digital workflows. Just as physical world models predict how objects move through space, we’re developing the equivalent for the digital world: predicting how agents act in digital workflows, then using that to scale the creation of high-quality training data for LLMs. Digital World Models help us push the frontier of ultra long horizon workflows, and unlock a new class of self-improving RL environments. This is our scalable approach to simulating all of the world’s intelligence. The round was also joined by Datadog, Inc., Samsung Ventures, Gokul Rajaram, Factorial Capital, and a large cohort of amazing AI leaders across Anthropic, OpenAI, Google DeepMind, NVIDIA, Recursive, and more.✨ It has been the ride of a lifetime. But we’re just getting started. The best is yet to come. "Do not go gentle into that good night, Rage, rage against the dying of the light" - Dylan Thomas (1954)

Today, we’re excited to announce our $50M Series B, led by Greenfield Partners, with participation from Lightspeed and Notable Capital. 🚀 At Patronus AI, we develop simulations and evals to train and improve AI. The first phase of AI was built on static benchmarks, but that era is over. As agents are used to solve longer and longer tasks, they need to practice in dynamic, living worlds to get better. Simulations are the critical infrastructure powering this next phase. As a company, we’re behind the most influential research and products in AI evaluation, like FinanceBench, Lynx, and Percival. And things have moved at the speed of light since.⚡ We partner with the world's leading frontier AI labs and enterprises, and our revenue has grown more than 15x over the past year. Additionally, today, we’re introducing a preview of the first Digital World Model for AI agent training and simulation: Patronus-DWM. Digital World Models are language diffusion world models that predict realistic environment behaviors and steer agent actions across digital workflows. Just as physical world models predict how objects move through space, we’re developing the equivalent for the digital world: predicting how agents act in digital workflows, then using that to scale the creation of high-quality training data for LLMs. Digital World Models help us push the frontier of ultra long horizon workflows, and unlock a new class of self-improving RL environments. This is our scalable approach to simulating all of the world’s intelligence. The round was also joined by Datadog, Inc., Samsung Ventures, Gokul Rajaram, Factorial Capital, and a large cohort of amazing AI leaders across Anthropic, OpenAI, Google DeepMind, NVIDIA, Recursive, and more.✨ It has been the ride of a lifetime. But we’re just getting started. The best is yet to come. "Do not go gentle into that good night, Rage, rage against the dying of the light" - Dylan Thomas (1954)

PatronusAI

95,899 views • 3 months ago

Demis Hassabis reveals Google's three-stage stack rewrite: AI first, model first, agent first. Bolting agents onto normal software does not scale to billions. "We've got to, we've done a lot of engineering work under the hood in the last year, which you'll see, you see the surface of when we announced things like this." "But actually, you can almost think that whole stack had to be rewritten to be sort of AI first, model first, and now agent first." "So that's quite a big change from like normal software." "And obviously there's all this huge infrastructure that serves billions of users every day at lightning speed and that's all gonna work." "And I think we've made massive improvements on that over the past year. And that's why you can see all the goodness of the AI models flowing through into our product surfaces so quickly and in such compelling ways." Most companies are adding an agent interface to yesterday's software. Demis says Google changed the substrate itself. The moat may not be the model. It may be the infrastructure that lets the model reach billions of users every day without breaking. - Demis Hassabis (Demis Hassabis), CEO of Google DeepMind, with Rowan Cheung at Google I/O 2026

Demis Hassabis reveals Google's three-stage stack rewrite: AI first, model first, agent first. Bolting agents onto normal software does not scale to billions. "We've got to, we've done a lot of engineering work under the hood in the last year, which you'll see, you see the surface of when we announced things like this." "But actually, you can almost think that whole stack had to be rewritten to be sort of AI first, model first, and now agent first." "So that's quite a big change from like normal software." "And obviously there's all this huge infrastructure that serves billions of users every day at lightning speed and that's all gonna work." "And I think we've made massive improvements on that over the past year. And that's why you can see all the goodness of the AI models flowing through into our product surfaces so quickly and in such compelling ways." Most companies are adding an agent interface to yesterday's software. Demis says Google changed the substrate itself. The moat may not be the model. It may be the infrastructure that lets the model reach billions of users every day without breaking. - Demis Hassabis (Demis Hassabis), CEO of Google DeepMind, with Rowan Cheung at Google I/O 2026

Karl Mehta

36,402 views • 2 months ago