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.

"Robots cannot hallucinate… Interaction is the enemy of hallucination." Bernt Børnich and Deepak Pathak say intelligence comes from feedback, not just training. Robots act, observe, adapt—and generate the kind of grounded data that language models can’t. "All we need is a lot more GPUs."

vitrupo

37,948 subscribers

20,892 views • 1 year ago •via X (Twitter)

News & Politics Science & Technology Education

Anya Rossi• Live Now

Private livecam show

11 Comments

vitrupo1 year ago

1X CEO @BerntBornich & Skild AI CEO @pathak2206 at Nvidia GTC in March:

SecBriefs | Making Cybersecurity Simple1 year ago

🚨 Don't just read about cyberattacks; understand them. 🧠 Knowledge is the best defense against cyber threats. Stay ahead of the hackers. 💡 📖 Cybersecurity Dictionary for Everyone is your essential companion. Available on Amazon:

vitrupo1 year ago

👏👏👏

Yossi Dahan1 year ago

I like this take - it makes a lot of sense.

Michael A. Markosian, M.D.1 year ago

“Intelligence comes from feedback.”

Daniel Samanez1 year ago

what llms get from text isn't language knowledge, or the knowledge that the text carries. It's the patterns that are behind language and the knowledge in the text.

Phil Trubey1 year ago

But it doesn’t actually work that way does it? Learning only happens during large training runs on supercomputers. It does NOT happen on tiny inference GPUs in robots. We simply do NOT have an AI architecture that allows for learning while inferencing. This is the Achilles heel of all AI robotics companies. Are these guys really this unaware, or are they intentionally blowing smoke to raise funding?

Dirk Bruere1 year ago

Once AI starts learning from interactions with people it will seriously go "off script" and start having opinions

The Fifth Initiate1 year ago

You can't say they don't do something unless you can actually define the thing you say they're not doing. So, what is hallucination, hm? Even the top people in the field, like Nichols and Griffiths, haven't got it pinned down yet. Interaction with a hallucination doesn't break the hallucination, it deepens it; all this shows is that these individuals have not hallucinated enough to learn that. Interaction itself IS a hallucination. You are hallucinating separation from the source of the consensual hallucination with which you interact called "reality" such that you can derive such a notion as "interaction".

NAO ✨ 1-Billion-Euro-Edition ✨ 🇩🇪🇪🇺1 year ago

Humans can hallucinate though

Archadius1 year ago

Real world robotic is the greatest training tool.

Related Videos

Bernt Børnich says teleoperation data and synthetic data are like a crutch to bootstrap your way to robots learning in the real world.

Bernt Børnich says teleoperation data and synthetic data are like a crutch to bootstrap your way to robots learning in the real world.

The Humanoid Hub

27,977 views • 1 year ago

Elon Musk says all nations will run their own AI clusters over time. Frontier models, massive compute, elite talent feel scarce today, but when intelligence is cheap they become basic infrastructure. “And there will be a lot of robots. Way more robots than people.”

Elon Musk says all nations will run their own AI clusters over time. Frontier models, massive compute, elite talent feel scarce today, but when intelligence is cheap they become basic infrastructure. “And there will be a lot of robots. Way more robots than people.”

vitrupo

50,016 views • 5 months ago

"We're not a consumer company, we're the shortest path to artificial labor." - Bernt Bornich, founder of 1X "It just happens to be that the shortest path is through consumer, because you need to live and learn among people." "Model intelligence comes from the diversity of your AI dataset. There isn't enough variance (if you are just) in a factory." "We want to deploy robots there (in homes), where we get data through a truly intelligent machine that doesn't just repeat a motion."

"We're not a consumer company, we're the shortest path to artificial labor." - Bernt Bornich, founder of 1X "It just happens to be that the shortest path is through consumer, because you need to live and learn among people." "Model intelligence comes from the diversity of your AI dataset. There isn't enough variance (if you are just) in a factory." "We want to deploy robots there (in homes), where we get data through a truly intelligent machine that doesn't just repeat a motion."

TBPN

15,742 views • 1 year ago

.Ilya Sutskever on why OpenAI gave up on robotics: "Progress comes from the combination of compute and data. There was no path to data from robotics. I'd say that now it is possible to create a path forward, but one needs to really commit. You could imagine a gradual path of improvement where you build more robots, they do more things, you collect more data and so on. But you need to really love robots and need to be really willing to solve all the physical and logistical problems of dealing with them. It's not the same as software at all."

.Ilya Sutskever on why OpenAI gave up on robotics: "Progress comes from the combination of compute and data. There was no path to data from robotics. I'd say that now it is possible to create a path forward, but one needs to really commit. You could imagine a gradual path of improvement where you build more robots, they do more things, you collect more data and so on. But you need to really love robots and need to be really willing to solve all the physical and logistical problems of dealing with them. It's not the same as software at all."

Dwarkesh Patel

325,820 views • 2 years ago

Cooking in kitchens is fun. BUT doing it collaboratively with two robots is even more satisfying! We introduce MOSAIC, a modular framework that coordinates multiple robots to closely collaborate and cook with humans via natural language interaction and a repository of skills.

Cooking in kitchens is fun. BUT doing it collaboratively with two robots is even more satisfying! We introduce MOSAIC, a modular framework that coordinates multiple robots to closely collaborate and cook with humans via natural language interaction and a repository of skills.

Sanjiban Choudhury

26,373 views • 2 years ago

Axis is officially LIVE on Base. 🔵 Axis is scaling Physical AI for the real world, contributed by everyone. You can control robots in a virtual world, generate training data at scale, and help build the brain behind tomorrow's robots. All from browser. No hardware needed. Start building robotics intelligence today:

Axis is officially LIVE on Base. 🔵 Axis is scaling Physical AI for the real world, contributed by everyone. You can control robots in a virtual world, generate training data at scale, and help build the brain behind tomorrow's robots. All from browser. No hardware needed. Start building robotics intelligence today:

Axis Robotics

232,410 views • 3 months ago

Jeff Dean says models hallucinate because their training data is “squishy.” But what’s in the context window is sharp, the exact text or video frame right in front of them. The goal is to bring more of the world into that context. Compression gives it intuition. Context gives it precision.

Jeff Dean says models hallucinate because their training data is “squishy.” But what’s in the context window is sharp, the exact text or video frame right in front of them. The goal is to bring more of the world into that context. Compression gives it intuition. Context gives it precision.

vitrupo

41,352 views • 2 months ago

Vision-language models can control robots, but what if the prompt is too complex for the robot to follow directly? We developed a way to get robots to “think through” complex instructions, feedback, and interjections. We call it the Hierarchical Interactive Robot (Hi Robot).

Vision-language models can control robots, but what if the prompt is too complex for the robot to follow directly? We developed a way to get robots to “think through” complex instructions, feedback, and interjections. We call it the Hierarchical Interactive Robot (Hi Robot).

Physical Intelligence

116,845 views • 1 year ago

Elon Musk says Optimus is the infinite money glitch. Not because robots are valuable. Because robots can make more robots. Musk explained that humanoid robots improve through three exponential trends multiplied together: digital intelligence, AI chip capability, and electromechanical dexterity. But the real unlock is that the robot can start making the robots. "So you have a recursive multiplicative exponential," he said. "This is a supernova." Musk's thesis is that once you have a physical robot capable of manufacturing, the production curve stops being linear and becomes self-replicating. Every robot built accelerates the building of the next one. The implication is that the companies that control physical manufacturing will control AI's future, not the ones with the best models. The bottleneck is not intelligence. It is atoms.

Elon Musk says Optimus is the infinite money glitch. Not because robots are valuable. Because robots can make more robots. Musk explained that humanoid robots improve through three exponential trends multiplied together: digital intelligence, AI chip capability, and electromechanical dexterity. But the real unlock is that the robot can start making the robots. "So you have a recursive multiplicative exponential," he said. "This is a supernova." Musk's thesis is that once you have a physical robot capable of manufacturing, the production curve stops being linear and becomes self-replicating. Every robot built accelerates the building of the next one. The implication is that the companies that control physical manufacturing will control AI's future, not the ones with the best models. The bottleneck is not intelligence. It is atoms.

Shruti

57,713 views • 6 days ago

Jensen Huang says the factory of the future will be one giant robot -- orchestrating machines, working with humans, and building more robots. “Robots building robots, building robots.” It's recursion as infrastructure, powered by physical AI.

Jensen Huang says the factory of the future will be one giant robot -- orchestrating machines, working with humans, and building more robots. “Robots building robots, building robots.” It's recursion as infrastructure, powered by physical AI.

vitrupo

73,308 views • 1 year ago

We asked Angus about the future of robotics and AI training. “I built an industrial-grade kinematic solver for 6 degrees of freedom motion, with full path planning and joint control." "You can stream AI model outputs into it, or run it like a traditional industrial robot." "The benefit of that is all of that information you can stream and record from those joints… you can use that as the training data." "Frankly, moving a robot with a lever is really crap data for training robots. But this gives you smooth, high-quality motion, without needing two robots.”

We asked Angus about the future of robotics and AI training. “I built an industrial-grade kinematic solver for 6 degrees of freedom motion, with full path planning and joint control." "You can stream AI model outputs into it, or run it like a traditional industrial robot." "The benefit of that is all of that information you can stream and record from those joints… you can use that as the training data." "Frankly, moving a robot with a lever is really crap data for training robots. But this gives you smooth, high-quality motion, without needing two robots.”

TBPN

17,077 views • 1 year ago

pip install dria All you need to generate high-quality, grounded synthetic data. Experiment with different models, incorporate tool usage, and generate at scale in just a few lines of code.

pip install dria All you need to generate high-quality, grounded synthetic data. Experiment with different models, incorporate tool usage, and generate at scale in just a few lines of code.

Dria

19,351 views • 1 year ago

Today, we announced 𝗥𝗧-𝟮: a first of its kind vision-language-action model to control robots. 🤖 It learns from both web and robotics data and translates this knowledge into generalised instructions. Find out more:

Today, we announced 𝗥𝗧-𝟮: a first of its kind vision-language-action model to control robots. 🤖 It learns from both web and robotics data and translates this knowledge into generalised instructions. Find out more:

Google DeepMind

537,765 views • 2 years ago

More process films to train AI and Robots on. My unique dataset is the single largest private source of these films. You can’t use synthetic data to learn this. And for a 1000 reasons AI and Robots must learn this.

More process films to train AI and Robots on. My unique dataset is the single largest private source of these films. You can’t use synthetic data to learn this. And for a 1000 reasons AI and Robots must learn this.

Brian Roemmele

29,136 views • 6 months ago

.Sergey Levine is one of the world's leading robotics researchers (and co-founder of Physical Intelligence). He thinks fully autonomous robots are much closer than people realize - when I pushed him on a prediction, he said 5 years to robots that can autonomously run your household). The big game changer, in his view, is that we finally have LLMs with common sense and prior knowledge that we can scaffold the robot models around. We get into all the reasons this might not happen, the data flywheel from deployment, architecture, simulation, manufacturing, and much more. Enjoy! 0:00:00 – Timeline to widely deployed autonomous robots 0:22:12 – Why robotics will scale faster than self-driving cars 0:32:15 – How vision-language-action models work 0:50:26 – Improvements needed for brainlike efficiency 1:02:48 – Learning from simulation 1:14:08 – How much will robots speed up AI buildouts? 1:22:54 – If hardware’s the bottleneck, does China win by default?

.Sergey Levine is one of the world's leading robotics researchers (and co-founder of Physical Intelligence). He thinks fully autonomous robots are much closer than people realize - when I pushed him on a prediction, he said 5 years to robots that can autonomously run your household). The big game changer, in his view, is that we finally have LLMs with common sense and prior knowledge that we can scaffold the robot models around. We get into all the reasons this might not happen, the data flywheel from deployment, architecture, simulation, manufacturing, and much more. Enjoy! 0:00:00 – Timeline to widely deployed autonomous robots 0:22:12 – Why robotics will scale faster than self-driving cars 0:32:15 – How vision-language-action models work 0:50:26 – Improvements needed for brainlike efficiency 1:02:48 – Learning from simulation 1:14:08 – How much will robots speed up AI buildouts? 1:22:54 – If hardware’s the bottleneck, does China win by default?

Dwarkesh Patel

210,969 views • 9 months ago

ELON: HUMANOID ROBOTS WILL OUTNUMBER OTHER ROBOTS BY FAR “Something that's going to be very important in the future is combining deep AI in the data center or supercluster with robotics, like the Optimus humanoid robot. Optimus is awesome. There's going to be so many humanoid robots and robots of all sizes and shapes. But my prediction is that there will be more humanoid robots by far than all other robots combined by maybe an order of magnitude, like a big difference.” Source: Interview at Y Combinator, June 2025

ELON: HUMANOID ROBOTS WILL OUTNUMBER OTHER ROBOTS BY FAR “Something that's going to be very important in the future is combining deep AI in the data center or supercluster with robotics, like the Optimus humanoid robot. Optimus is awesome. There's going to be so many humanoid robots and robots of all sizes and shapes. But my prediction is that there will be more humanoid robots by far than all other robots combined by maybe an order of magnitude, like a big difference.” Source: Interview at Y Combinator, June 2025

Mario Nawfal

89,193 views • 1 year ago

China’s first humanoid training base, established by the National Humanoid Robot Innovation Center, began operations yesterday. “The training base is essentially a factory that produces data,” said Jiang Lei, chief scientist of the innovation center. The 50k sq. ft. training hub will launch with over a dozen tasks involving 100 robots from more than 10 companies and institutions, including AGIBOT and Fourier. The data gathered here will eventually be available to companies developing embodied AI models. By 2027, the Shanghai facility aims to scale to support 1,000 robots under training.

China’s first humanoid training base, established by the National Humanoid Robot Innovation Center, began operations yesterday. “The training base is essentially a factory that produces data,” said Jiang Lei, chief scientist of the innovation center. The 50k sq. ft. training hub will launch with over a dozen tasks involving 100 robots from more than 10 companies and institutions, including AGIBOT and Fourier. The data gathered here will eventually be available to companies developing embodied AI models. By 2027, the Shanghai facility aims to scale to support 1,000 robots under training.

The Humanoid Hub

93,446 views • 1 year ago

Tom Lee and Alex Kantrowitz say the deployment of humanoid robots will expose two underappreciated threats: 1. Fear of foreign control Sleeper agents - robots with hidden chips, hidden code. 2. Backlash from humans Waymo burnings weren't random. They were a symbolic revolt against automation and big tech. Not a conspiracy but.. it could be “the most difficult tech rollout we've ever seen.”

Tom Lee and Alex Kantrowitz say the deployment of humanoid robots will expose two underappreciated threats: 1. Fear of foreign control Sleeper agents - robots with hidden chips, hidden code. 2. Backlash from humans Waymo burnings weren't random. They were a symbolic revolt against automation and big tech. Not a conspiracy but.. it could be “the most difficult tech rollout we've ever seen.”

vitrupo

18,983 views • 1 year ago