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.

Introducing Hedra Model Factory: an autonomous inference and training platform for building models for the real world. The recursive self-improvement loop behind frontier research has been locked inside a few labs. We're opening it to teams in robotics, manufacturing, biology and defense. Our researchers come from Stanford, NVIDIA and... show more

Hedra

54,430 subscribers

2,438,700 views • 2 days ago •via X (Twitter)

Science & Technology

Anya Rossi• Live Now

Private livecam show

60 Comments

Hedra2 days ago

We're expanding to full-service research support. Start with your data and simulation environments. We help with the entire model development lifecycle, from your first RL environment to every subsequent iteration. Every team gets forward-deployed researchers who work alongside you at each step. Here's what our researchers have been working on ↓

Hedra2 days ago

→ Post-training world models for physical consistency. One researcher retrained H3 across data, simulation and the hero run to beat leading frontier models on Google DeepMind's Physics-IQ (34.9 vs 30.3) → We discovered a new distillation technique that runs world models in real time without losing physical consistency, now powering our synthetic data pipeline for customers → Autonomous evals that catch defects in training data. We optimized a VLA to dramatically cut captioning and QC costs → Building infrastructure to scale past 100,000 simultaneous RL environments and simulators on our new CPU/GPU sandboxes We're publishing the research as we go, starting this week.

Hedra2 days ago

We're building a factory that gets faster with every model it produces, and adding thousands of GPUs to power it. Talk to our team today →

joyce2 days ago

ig now's a good time to say: - if you need compute, we have compute, B300s, 200s, H100s, H200s, AMD MI355X - if you want to post-train OSS models for specialised use cases (eg. driving tractors, analysing stem cells, etc) we'll do it for you - we have GPU sandboxes dm me!!

The Rabbit Hole2 days ago

Robotics is going to change the world. There’s a lot of untapped potential, especially as the scope of roles improves.

Hedra2 days ago

potential goes brr

Paul Mit2 days ago

it's actually very impressive and refreshing to see the products that reach outside the digital on the timeline 💪 what's the plan for the future?

andrew chen2 days ago

Congrats to @mjlbach and the Hedra team!!

Hedra2 days ago

@mjlbach thanks Andrew!!

Everett Butler2 days ago

physical AI is the future, and whoever delivers the end to end infrastructure optimized for pixels and atoms vs. text is going to win big.

Hedra2 days ago

indeed. new era of models

Science girl2 days ago

This is super interesting

EP2 days ago

congrats 👏

Santiago2 days ago

This sounds like a service, right? Based on what I understand, this is not self-serve, correct?

Hedra2 days ago

yes its a service! we'll work with you

shirish2 days ago

34.9 vs 30.3 against deepmind is no joke

Daniel Smidstrup2 days ago

Recursive self-improvement is for sure the key to actually getting to the next level! How close are we at this point to AI just continuously getting smarter, truly on its own?

Hedra2 days ago

we're already seeing it internally

Daniel Smidstrup2 days ago

Does it have an end or will it keep going?

elvis2 days ago

Bringing the self-improvement loop to robotics and real-world data is a big step. Congrats on the launch!

Dev Shah2 days ago

congrats this is insane

Hedra2 days ago

thx

Jon Lai2 days ago

So excited for this! So much demand for models for the real world 🏭

Hedra2 days ago

accelerate

sophia dew2 days ago

congrats!

Hedra2 days ago

thanks sophia!

bunny2 days ago

fully autonomous looped inference and training platform is actually crazy

Hedra2 days ago

crazier things coming

tech bro2 days ago

This is so freaking cool team!!!

Hedra2 days ago

hell yea

tech bro2 days ago

Great video team!

haris2 days ago

lfg this is sick

Hedra2 days ago

right back at ya

Future Brian2 days ago

banger launch :)

Hedra2 days ago

bang bang!!

Krea2 days ago

🔥🔥🔥

Hedra2 days ago

🔥🫡

phed2 days ago

the stuff only the big ai labs had is now open to everyone. hedra cooking

Hedra2 days ago

we cooked and we ate

Livia Han 🐯2 days ago

congrats hedra team!!

Hedra2 days ago

ty liv!!

David Kwon2 days ago

Congrats on the launch!

Hedra2 days ago

thank you David!

litquidity2 days ago

I wish I had more friends IRL to discuss this with

Markets & Mayhem2 days ago

I am so bullish on robotics long-term. I think it's going to change how a lot of humanity works for the better. But one of the biggest shortcomings right now is exactly this. The lack of the ability to adapt. Great idea with Model Factory. This is the innovation we need in the space!

Palash Awasthi2 days ago

this changes everything 🔥

Hedra2 days ago

everything's changing

Andrew2 days ago

massive congrats!!!

Hedra2 days ago

yay thanks andrew

Exec Sum2 days ago

This is the next hot space in AI

Hedra2 days ago

we're only getting started

Heshie Brody2 days ago

this is super cool!

Hedra2 days ago

🔥

Damir Becirovic2 days ago

Amazing work Hedra team

Hedra2 days ago

thanks!

hubert 🎬2 days ago

nice, congrats and good luck

Hedra2 days ago

pow pow

Emil2 days ago

Could you outline a bit what a recursive self-improvement loop looks like for robotics, for instance? I mean it seems clear how it works with LLMs, where they become smarter and thus improve even faster from there. But it is not clear how this positive feedback loop could work for other types of models that are not building themselves? To me it looks more like an adaptive research strategy that is continuously finding out what the best hypothesis is to test to obtain the best knowledge for further improvement. That is still impressive though!

klöss2 days ago

massive day for the Hedra team congrats y’all!

Aditi2 days ago

what does the testing and validation process look like before a custom-trained model is ready for actual production use?

Related Videos

It's been just over a year since we started Meta Superintelligence Labs. Muse Spark was our first family of models, and now Muse Wonder is a much larger and more advanced class of models that brings us to the frontier. We're releasing two models today. Muse Wonder for fast frontier performance and Muse Wonder Full for the full version of the model.

It's been just over a year since we started Meta Superintelligence Labs. Muse Spark was our first family of models, and now Muse Wonder is a much larger and more advanced class of models that brings us to the frontier. We're releasing two models today. Muse Wonder for fast frontier performance and Muse Wonder Full for the full version of the model.

Mark Zuckerberg

640,022 views • 1 month ago

NVIDIA Cosmos world models have come to life with real-time interactivity! 🌎🤖🚘 We are releasing our post-training framework for real-time interactive world simulation for Physical AI. The framework builds upon Cosmos 3 through causal training, distillation, and optimized streaming inference. The post-trained models stream visual observations via autoregressive generation in response to incoming actions and scene controls. We are also releasing example model checkpoints for robotics, autonomous vehicles, and sim-to-real transfer. The release includes optimized streaming inference code to power the models. Cosmos world foundation models provide broad visual knowledge. Each simulation setting adds its own controls, data, and evaluation requirements. Our shared recipe and example models give researchers and developers a practical starting point for building specialized interactive world simulators. Huge shoutout to the team for the dedication, creativity, and hard work poured into this release! I am incredibly proud of this work and feel fortunate to lead such a talented team. A special thank-you to the broader Cosmos group for building the world foundation models together. Team green 💚 I am excited to see how you build your own interactive world simulator with Cosmos! 🏠 Project website: 🤗 Model checkpoints: 💻 Code:

NVIDIA Cosmos world models have come to life with real-time interactivity! 🌎🤖🚘 We are releasing our post-training framework for real-time interactive world simulation for Physical AI. The framework builds upon Cosmos 3 through causal training, distillation, and optimized streaming inference. The post-trained models stream visual observations via autoregressive generation in response to incoming actions and scene controls. We are also releasing example model checkpoints for robotics, autonomous vehicles, and sim-to-real transfer. The release includes optimized streaming inference code to power the models. Cosmos world foundation models provide broad visual knowledge. Each simulation setting adds its own controls, data, and evaluation requirements. Our shared recipe and example models give researchers and developers a practical starting point for building specialized interactive world simulators. Huge shoutout to the team for the dedication, creativity, and hard work poured into this release! I am incredibly proud of this work and feel fortunate to lead such a talented team. A special thank-you to the broader Cosmos group for building the world foundation models together. Team green 💚 I am excited to see how you build your own interactive world simulator with Cosmos! 🏠 Project website: 🤗 Model checkpoints: 💻 Code:

Chen-Hsuan Lin

12,181 views • 21 days ago

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.

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

252,658 views • 2 months ago

Cyber defense is entering the superintelligence era. ⚡ Introducing the CrowdStrike Cyber Superintelligence Lab, the first frontier AI research organization built for cyberdefense and AI safety. The Lab delivers CrowdStrike SafeMind, a family of purpose-built security models and harnesses. SafeMind runs as one system designed to deliver AI safety: an offensive model that finds the attack path, a defensive model that closes it, and the harnesses that operate both in the same loop. Built using NVIDIA Nemotron open models and using CoreWeave’s AI Cloud for training and inference, SafeMind goes beyond AI that simply identifies threats, it defeats them. Learn more:

Cyber defense is entering the superintelligence era. ⚡ Introducing the CrowdStrike Cyber Superintelligence Lab, the first frontier AI research organization built for cyberdefense and AI safety. The Lab delivers CrowdStrike SafeMind, a family of purpose-built security models and harnesses. SafeMind runs as one system designed to deliver AI safety: an offensive model that finds the attack path, a defensive model that closes it, and the harnesses that operate both in the same loop. Built using NVIDIA Nemotron open models and using CoreWeave’s AI Cloud for training and inference, SafeMind goes beyond AI that simply identifies threats, it defeats them. Learn more:

CrowdStrike

58,367 views • 2 months ago

Vals AI CEO Rayan Krishnan on how they score a model's ability to build its own successor: "One benchmark we released recently is our Recursive Self-Improvement Index. It's a topic which a lot of the big labs have been talking about and starting to report on in their model cards." "There isn't a shared language to talk about the RSI potential of models. So we created this as an apples-to-apples way to actually benchmark across the models." "In an ideal world, what you want to do is take a frontier model and have it train the next version of itself and see where the delta comes from. But obviously that's very expensive and slow." "What we're doing is forming a set of proxies for every part of the process it takes to build the next version of the model." "There's work around pre-training, post-training, harness-level engineering, and then seeing in which mechanisms and behaviors the models are able to do very good research work and build something new, and where they're struggling." Rayan Krishnan Jennifer Li

Vals AI CEO Rayan Krishnan on how they score a model's ability to build its own successor: "One benchmark we released recently is our Recursive Self-Improvement Index. It's a topic which a lot of the big labs have been talking about and starting to report on in their model cards." "There isn't a shared language to talk about the RSI potential of models. So we created this as an apples-to-apples way to actually benchmark across the models." "In an ideal world, what you want to do is take a frontier model and have it train the next version of itself and see where the delta comes from. But obviously that's very expensive and slow." "What we're doing is forming a set of proxies for every part of the process it takes to build the next version of the model." "There's work around pre-training, post-training, harness-level engineering, and then seeing in which mechanisms and behaviors the models are able to do very good research work and build something new, and where they're struggling." Rayan Krishnan Jennifer Li

a16z

35,070 views • 2 months ago

We're hiring in robot learning at World Labs! Join me, Fei-Fei Li, and the team to define and scale the next generation of world models for robot learning! Atlas for Robotics: Real-to-Sim-to-Real: Apply:

We're hiring in robot learning at World Labs! Join me, Fei-Fei Li, and the team to define and scale the next generation of world models for robot learning! Atlas for Robotics: Real-to-Sim-to-Real: Apply:

Yunzhu Li

177,209 views • 1 month ago

Introducing Deft Robotics' unified deployment platform for physical AI. Frontier models are moving fast. Real world autonomy isn’t. Closing the gap takes more than better models. The missing piece? Infrastructure to deploy, intervene, learn from failures, and continuously improve. For the past year, we’ve been deploying in that gap and building the stack to close it. Now, we’re opening it up. Hardware available now. Software in beta. Here’s how to use it for your use-case 🧵(1/7) ↓

Introducing Deft Robotics' unified deployment platform for physical AI. Frontier models are moving fast. Real world autonomy isn’t. Closing the gap takes more than better models. The missing piece? Infrastructure to deploy, intervene, learn from failures, and continuously improve. For the past year, we’ve been deploying in that gap and building the stack to close it. Now, we’re opening it up. Hardware available now. Software in beta. Here’s how to use it for your use-case 🧵(1/7) ↓

Shane

17,961 views • 2 months ago

Today we're incredibly excited to announce Dreamscale Labs (YC F26) (YC F26)! We're an applied research company putting robot brains in the cloud. Robot AI models are growing larger and on-board compute is getting increasingly inaccessible due to cost. We're taking a contrarian bet that the future is moving inference off the robot and into the cloud, while still meeting real-time deadlines required by robot motion. We're building the real-time inference infrastructure that enables frontier physical AI models to run smoothly on your robot. We optimize for low latency across the full stack: from the model, down to the GPU kernels, and over the network. Our mission is to unlock a future of safe and intelligent robots doing meaningful work, and we think this is only possible by letting you deploy and run robot models in the cloud. Reach out to us at founders@dreamscalelabs.com or visit our website at to learn more. Kevin L 🚢 CoRL Antoine Nguyen Eric R 🚴CoRL

Today we're incredibly excited to announce Dreamscale Labs (YC F26) (YC F26)! We're an applied research company putting robot brains in the cloud. Robot AI models are growing larger and on-board compute is getting increasingly inaccessible due to cost. We're taking a contrarian bet that the future is moving inference off the robot and into the cloud, while still meeting real-time deadlines required by robot motion. We're building the real-time inference infrastructure that enables frontier physical AI models to run smoothly on your robot. We optimize for low latency across the full stack: from the model, down to the GPU kernels, and over the network. Our mission is to unlock a future of safe and intelligent robots doing meaningful work, and we think this is only possible by letting you deploy and run robot models in the cloud. Reach out to us at [email protected] or visit our website at to learn more. Kevin L 🚢 CoRL Antoine Nguyen Eric R 🚴CoRL

Chris Yoo ⚡ CoRL

50,223 views • 1 month ago

"You need frontier products in order to have frontier models and for people to feel the magic of frontier models." Full conversation with Dianne Penn (Head of Product for Research and Labs Anthropic):

"You need frontier products in order to have frontier models and for people to feel the magic of frontier models." Full conversation with Dianne Penn (Head of Product for Research and Labs Anthropic):

Lenny Rachitsky

17,966 views • 3 months ago

Introducing AutoScientist. Most model training fails outside of frontier labs. AutoScientist automates the full research loop so it doesn't have to.

Introducing AutoScientist. Most model training fails outside of frontier labs. AutoScientist automates the full research loop so it doesn't have to.

adaption

206,512 views • 6 months ago

Introducing Handshake AI—the most ambitious chapter in our story. We leverage the scale of the largest early career network to source, train, and manage domain experts who test and challenge frontier models to failure for the top AI labs.

Introducing Handshake AI—the most ambitious chapter in our story. We leverage the scale of the largest early career network to source, train, and manage domain experts who test and challenge frontier models to failure for the top AI labs.

Handshake

360,513 views • 1 year ago

Introducing Midcentury. We’re building the data and simulation infra for physical AI. Today, we’re coming out of stealth with a $15M Series Seed to scale robotics beyond polished demos. We’re already supporting frontier labs with: → The world’s largest egocentric dataset: 2M+ hours, 50+ environments, 20,000+ tasks → Matrix: a frontier simulation platform scaled with our real-world data to evaluate and post-train policies at scale

Introducing Midcentury. We’re building the data and simulation infra for physical AI. Today, we’re coming out of stealth with a $15M Series Seed to scale robotics beyond polished demos. We’re already supporting frontier labs with: → The world’s largest egocentric dataset: 2M+ hours, 50+ environments, 20,000+ tasks → Matrix: a frontier simulation platform scaled with our real-world data to evaluate and post-train policies at scale

Midcentury

643,408 views • 1 month ago

Wave 9 is here: a frontier model built for software engineering. Introducing our new family of models: SWE-1, SWE-1-lite, and SWE-1-mini. Based on internal evals, it has performance nearing that of frontier models from the foundation labs. Available now, only in Windsurf!

Wave 9 is here: a frontier model built for software engineering. Introducing our new family of models: SWE-1, SWE-1-lite, and SWE-1-mini. Based on internal evals, it has performance nearing that of frontier models from the foundation labs. Available now, only in Windsurf!

Devin Developers

743,706 views • 1 year ago

1/ ChatGPT had the internet. AI for science has no equivalent. Liam Fedus helped create ChatGPT and went on to lead post-training at OpenAI. Now he’s building Periodic Labs because better models alone won’t get us to an AI scientist. We talked about why Periodic is building physical labs and what Liam learned from being inside OpenAI as ChatGPT took off on today's episode of The Frontier.

1/ ChatGPT had the internet. AI for science has no equivalent. Liam Fedus helped create ChatGPT and went on to lead post-training at OpenAI. Now he’s building Periodic Labs because better models alone won’t get us to an AI scientist. We talked about why Periodic is building physical labs and what Liam learned from being inside OpenAI as ChatGPT took off on today's episode of The Frontier.

Brendan (can/do)

28,029 views • 1 month ago

🧬We think biology needs its own world engine.🌏 Introducing LabWorld Factory. Prompt a lab. Generate a world. ⚡ From natural language, to scalable simulated biology labs, for training, benchmarking, and stress-testing AI agents. Not one scene. A growing distribution of labs.

🧬We think biology needs its own world engine.🌏 Introducing LabWorld Factory. Prompt a lab. Generate a world. ⚡ From natural language, to scalable simulated biology labs, for training, benchmarking, and stress-testing AI agents. Not one scene. A growing distribution of labs.

AI4Science Catalyst

70,272 views • 7 months ago

when do we see self-improvement in AI research vs. biology? Liam Fedus, Cofounder Periodic Labs and former lead of post-training OpenAI, on No Priors

when do we see self-improvement in AI research vs. biology? Liam Fedus, Cofounder Periodic Labs and former lead of post-training OpenAI, on No Priors

sarah guo

29,891 views • 7 months ago

Introducing Machine Age Ai Robots need to learn from people who know the work. We’ve been working with robotics companies in stealth, building a network of skilled businesses across the UK, US and China. Now we’re opening it up. Our marketplace connects those businesses with robotics teams looking for real-world training data. Join us:

Introducing Machine Age Ai Robots need to learn from people who know the work. We’ve been working with robotics companies in stealth, building a network of skilled businesses across the UK, US and China. Now we’re opening it up. Our marketplace connects those businesses with robotics teams looking for real-world training data. Join us:

Chris Worsey

469,980 views • 7 days ago

A few labs shape AI, but businesses know what great work looks like. The Trajectory platform from @TrajectoryLabs puts post-training in your hands: train your own model on production data, with training, evals and deployment in one place. That’s a big shift.

A few labs shape AI, but businesses know what great work looks like. The Trajectory platform from @TrajectoryLabs puts post-training in your hands: train your own model on production data, with training, evals and deployment in one place. That’s a big shift.

Out of Context Human Race

31,728 views • 10 days ago

was featured today on NVIDIA AI Nemotron Labs livestream on open model routing via their X pages and YouTube channels. We're an official NVIDIA Inception project running it in production: agent workloads routed between local Nemotron and frontier APIs. Thank you to the NVIDIA AIDev NVIDIA AIInfra and Nemotron Labs team for featuring us. Great session. #NVIDIAInception

was featured today on NVIDIA AI Nemotron Labs livestream on open model routing via their X pages and YouTube channels. We're an official NVIDIA Inception project running it in production: agent workloads routed between local Nemotron and frontier APIs. Thank you to the NVIDIA AIDev NVIDIA AIInfra and Nemotron Labs team for featuring us. Great session. #NVIDIAInception

three.ws

20,001 views • 2 months ago

Introducing Azure AI Foundry Labs: your hub for AI research and experiments from Microsoft. Discover new models, frameworks, and tools to solve complex problems and help shape the future of AI:

Introducing Azure AI Foundry Labs: your hub for AI research and experiments from Microsoft. Discover new models, frameworks, and tools to solve complex problems and help shape the future of AI:

Microsoft Azure

35,401 views • 1 year ago