Harsh

Check the full suite for full mocap to robotics pretraining . SOMA has anatomically correct joint definitions and has much detailed mesh key points compared to MHR/SMPL. Foundational for all bodypose downstream tasks. More on this soon on its capabilities.

@bubbleboi Hard agree

Why AI makes jobs instead of loses them. 1. There are more profitable ideas/projects than time, talent, and general resources there are to realize them. 2. AI lowers the talent and resources bar substantially leading to more ideas becoming reality. 3. Successful idea/project = more wealth, more work, and more jobs. The reason why so many people made the mistake of thinking it gets rid of jobs is they see a fixed pie. But in reality we are growing the pie. AI is really only bearish labor for bloated mismanaged incumbents not the entire economy.

@lukas_m_ziegler Waymo on your map seems wrong

This is the capital of robotics! 🌁 Silicon Valley is home to so many physical AI companies that you could spend a month visiting them and still not see even 10% of them (trust me, I tried). It took me 3x to create this map as it did to create any other. And the truth is, it's still incomplete. Let’s see why this is the case. So, the Bay Area is the world's leading ecosystem for robotics startups, bringing together top AI talent, top universities, experienced founders, and unmatched access to capital. The region is anchored by Stanford University and University of California, Berkeley, two of the world's top universities for AI and robotics. They produce a constant stream of researchers, founders, and breakthrough technologies. The Bay Area is also home to many of the companies shaping the future of robotics and AI, including @Figure_robot, @physical_int , and major AI labs such as @OpenAI. This concentration of talent makes it easy for startups to recruit experienced engineers and collaborate with leaders in embodied AI. Not mentioning that it  is also home to leaders such as @NVIDIARobotics , whose headquarters and leadership in AI chips power much of today's robotics revolution, and @Tesla, whose work on autonomous driving and humanoid robots has created a deep pool of robotics, AI, and manufacturing talent. Perhaps its biggest advantage is access to capital and ambition. The Bay Area has the world's deepest network of venture capital firms, serial entrepreneurs, and technical leaders who are willing to fund bold, long-term robotics companies. In the comments I'll post the companies from the ecosystem. ‼️ Note that Bay Area has >300 robotics companies, research labs, and innovation hubs, so this is a curated selection of the notable product companies, not an exhaustive census! P.S. I'm constantly working on improving these maps, so if your company is missing, please DM me with basic info about the co, and I will include it in the next release. ~~ ♻️ Join the weekly robotics newsletter, and never miss any news → ziegler.substack.com

Audex from Nvidia might be the first native audio model that isn't dumb as rocks. Most audio models (audio understanding or speech / audio generation) just don't know anything about anything. So it's cool to see a model that actually has smarts. I mean, when have you seen an audio model get a non-zero score on Terminal Bench?! Again, it's as simple as an interleaved post training pipeline (nemotron omni really proved this approach imo). Super cool model and release!

@fujikanaeda Nemo data designer is awesome and so well thought of, giving attention to details. Thanks and all the best.

Today marks the end of my last week at Nvidia. I joined with the rest of the excellent Gretel team when we were acquired back in April of 2025, which now seems like 10 years ago thanks to the time-dilation of AI progress. Over that year, I got to do quite a lot: help build the best synthetic data generation tool in the industry (NeMo Data Designer), scaling it out for pre & post-training datasets for Nemotron by building some slick cluster tooling, contributed to 4 (!!) Nemotron LLM builds (Nvidia doesn’t mess around with open models), took a small merging experiment reproduction from small scale (30B 😅) up to 550B and pass along some pretty significant eval compute savings as a consequence for our pretraining heros, pull my hair out over the state of public evals & benchmarks, and most importantly, got to collaborate with some of the best researchers and engineers in the field, all working to advance the frontier of open-source AI and build the best computing platforms in the world to support it. Thank you to everyone @ Nvidia & Nemotron for welcoming me into Team Green 💚. Thank you to the Gretelers for your support over this journey 💜.

BREAKING: @axisanna8 has joined Nvidia as head of corporate comms

Let’s go @askalphaxiv

@demalenk Pretty cool release !

[1/4] Current 3D human pose reconstruction models are very impressive, but which model should you pick for your application? Introducing the Caltech Tennis Dataset (CalTennis), a large-scale video benchmark for evaluating monocular-to-3D pose estimation in the wild. CalTennis comprises over 11 million frames (51 hours) of tennis practice and match play. It is 10× larger than existing in-the-wild human motion video datasets and 3× larger than existing MOCAPground-truthed datasets, and it is the first large-scale benchmark to provide synchronized multi-view recordings of expert athletic motion.

@sporadica This seems further bullish for them since they can free us from primitive usage via true computer use, so generic and encompassing there is no need to compete via other browser app. Having said that I don’t see why it was doomed to fail apart from hindsight.

AI labs releasing browsers is something that felt so far-fetched and out there in like 2024 and it's now 2026 and already feels like such a dumb decision, so obviously doomed to fail the cold hard truth of life is that, in the end, we are all destined to just use Google Chrome

I was clearly wrong about Anthropic. They are obviously currently the leader in AI. No company has released a model as good as Mythos/Fable and they will undoubtedly have Mythos 2 ready soon. And I would never cut them off in a way that hurt them badly, even as a competitor. That’s not my style. Tesla open sourced its patents and we made the Supercharger network available to all competitors, even though we could have made it a walled garden. SpaceX launches competing satellite systems with no increase in price or use of unfair terms. Even my worst enemies can attack me on this platform. …

Excited to release Grok 4.5 with @SpaceXAI. It's an Opus-class model that's fast and low cost. It's a significant step up over any model we've developed so far, including Composer 2.5, and has become the daily driver for many on our team. First of many releases. More soon.

The Nemotron family just passed 100M downloads! Huge thank you to the community building with us and showing what’s possible with open models. Cheers to OSS 🍾

@jxmnop Congrats!

my paper won an award at icml 😁 some thoughts: • this work was rejected from NeurIPS. i cleaned it up a small amount and it got great reviews from ICML! don't give up • ICML received 24k submissions and only gives out 7 awards, which is crazy. feeling grateful • i distinctly remember sitting at my desk two winters ago wondering if i would ever finish this project. most of all this is the product of sitting down and forcing myself to keep working for several months straight. the results emerged from running the experiments over and over and fixing a long sequence of tiny details. eventually, the curves looked like that 👇 • also happy that the insights in this paper are becoming more widely accepted: 3.3 bits/param, thinking about capacity "LLM as flashdrive" mentality • the method here is used successfully for selecting midtraining data at least one frontier lab, which is cool! • i am grateful to my collaborators, but Meta is no longer a great place for academic research imo and this almost never got published for a number of reasons. i shall not elaborate further • for future work, i think analyzing the implications of on-policy algorithms on capacity, as well as LoRA and things like it, are fruitful potential research directions • sadly i'm not in Korea but am following the conference online from california and happy to chat! a nice end to one phase of my research career :)

@Vtrivedy10 Banger

England government on the phone with fifa to get an extended 3 game ban

May we never change 🇺🇸

Happy 250th birthday, America! Founded in Silicon Valley over three decades ago, NVIDIA's roots in American innovation run deep. Today, we're proud to partner with the companies, scientists, developers, and builders using AI to accelerate progress and solve hard problems. Here’s to the next 250 years of American ingenuity, discovery, and breakthroughs.

My team open sourced a set of agent skills to run an entire @kaggle competition workflow from plain language. We ship an easy to install plugin for your favorite coding agent. 👉 Try it now: github.com/NVIDIA/nvidia-…

This was a crazy goal . Hearts for cpv

It is unbelievable to me that this wasn’t a red card + penalty kick Balogun was completely obstructed here and still hit the crossbar He earned two PKs that weren’t called This didn’t get VAR but an irrelevant midfield tangle led to a red card Unreal

@premqnair 👋

Who do I know going to ICML? Have some things to invite you to

Most motion papers tailor one controller to one specific task. This year at SIGGRAPH, our research team asks: can motor control itself be pretrained and reused? Generative Pretrained Controllers, or GPC, turn motor skills into a vocabulary of discrete tokens and train a transformer-based generative controller through next-token prediction. Just like GPT, the same pretrained controller can then be fine-tuned to solve new tasks. Trained on 600+ hours of motion, GPC runs in real-time inside a physics simulation, producing natural and physically grounded behaviors for interactive control.