Peter Hebert

We developed an RL method for fine-tuning our models for precise tasks in just a few hours or even minutes. Instead of training the whole model, we add an “RL token” output to π-0.6, our latest model, which is used by a tiny actor and critic to learn quickly with RL.

Earlier this month we hosted our first ever Health x Intelligence Summit led by our @deenashakir. From @ChelseaClinton on translating AI into real-world health impact at global scale, to whether AI agents are ready to prescribe. Below are just a few highlights from a thought-provoking day. ⚕️ @Cigna used an AI voice agent to call 100,000 unvaccinated members during flu season — 55% engagement, and 4,000 flu shot commitments. @nikillinit unpacked it live with @Cigna, Stellarus & Amazon Pharmacy. ⚕️Within days of the summit, @medra_ai told @Forbes' @annatonger their 100-robot facility was set to spend an entire weekend running lights-out experiments on an AI-designed antibody. This kind of scientific infrastructure simply didn't exist five years ago. ⚕️ @johnbrownstein reported that @BostonChildrens now has 45 AI agents running across the hospital — saving 50,000 hours and ~$5M in labor, always with a human in the loop. Watch the full Youtube playlist 👇

A breakthrough in real-time video generation. As a research preview developed with @NVIDIA and shared at @NVIDIAGTC this week, we trained a new real-time video model running on Vera Rubin. HD videos generate instantly, with time-to-first-frame under 100ms. Unlocking an entirely new creative paradigm and bolstering the foundations of our General World Model, GWM-1. Real-time generation opens a fundamentally different design space for video models and world simulation. We're investing in co-designing our models alongside advances in hardware to keep pushing this frontier.

We raised a $27M Series A to replace the spreadsheets and human duct tape behind $100 trillion in global assets. Fund administration is the invisible backbone of private equity and venture capital - and it’s broken. Why? Financial data is scattered, stale, and locked inside legacy providers. Books take forever to close. Basic questions about your own fund take days to answer. So we rebuilt the general ledger, waterfall engine, investor portal, and portfolio management from scratch. One single source of truth for your firm. Our AI agents read emails, propose journal entries, and extract portfolio updates in seconds. Our CPAs review every output. Today, we administer $15 billion in assets - and we’re just getting started. Every fund CFO keeps getting asked: how will you adopt AI? Now you have an answer. Run your firm in real-time with @hanoverpark. –- Excited to partner with Jake Saper at @emergencecap @peterjhebert at Lux, @chadbyers/@pratyushbuddiga at Susa and CFOs at the largest private equity firms to forge this future.

Hey that’s us!

Lux family companies @physical_int (embodied intelligence) + @medra_ai (building physical AI scientists)

@EdLudlow looks miserable!

And… breathe

US Army awards Anduril Industries a contract with a total value of as much as $20 billion to buy the defense startup’s software, hardware and services bloomberg.com/news/articles/…

. You thought I was joking about Reindustrializing America. I was not. Tune in March 20th for the the start - not the end - of the largest private Marshall plan ever devised. Alabama and F4 is the start. Winning the Production war is the end.

We've been working to bring something big to The Shoals, and now, it's time. On March 20th, we'll be launching our newest facility - Factory 4 - in Alabama. Expect: • 2.25M sq ft of manufacturing • 1,000+ jobs created • $2B in Private/Public investment • Remarks from @SenKatieBritt, @SenTuberville, @RepMikeRogersAL, @Robert_Aderholt and other senior officials • A private concert from our favorite rock band of all time, local BBQ, and the best entertainment Alabama has to offer Building for America is worth celebrating. Apply for an invite here, spots limited → HadrianAlabama.com

The Anduril 250 is in 100 days. Start your engines.

EXCLUSIVE: AI startup Sandbar has raised $23M in Series A funding for its note-taking smart ring, Stream, launching widely this summer. I spoke to CEO Mina Fahmi (@minafahmi) about his vision for a voice interface for your thoughts -- and for working with AI agents. "The thing that gets me excited is the promise of technology as an augmentation of the individual, that places the human in control,” he tells @UpstartsMediaCo.

Introducing, Runway Characters. Real-time intelligent avatars that turn the internet into a conversation. Deployable anywhere via the Runway API, Runway Characters can be customized in any way across every style. All with the ability to embed bespoke knowledge banks, custom voices and instructions. Start integrating Runway Characters directly into your apps, websites, products and services today. Available now at the link below.

I'm extremely excited to announce that we've successfully inferenced π0.5 on our excavator! We've collected a massive corpus of real-world data with natural language labels from operators in the industry and are using it to create some really cool policies. Here's our first demo of it successfully completing a task with just 200 trajectories. More on the way :) Read our blog post: labs.actor/research/vla-e… @physical_int

Oil

I've known Cameron for 36 years and when he locks in on something, get out of the way. I asked him if he's excited to have founded a unicorn (as first check alongside @Lux_Capital, I am 😃). He characteristically pivoted to a product update he's excited about and a customer they just partnered with. When you meet the @Nominal_io team, you see how special this culture is. They lead with their heart. @CameronLMcCord is intense about the mission and generous with the people around him. Now it is increasingly clear that he's building a generational company. All systems Nominal 🚀

Thank you to @Lux_Capital for believing in me, and in Nominal, before others understood. We work hard every day to make that a great decision. I consider myself part of the Lux family: I'm proud to watch them grow and am energized by the early bets they've placed in companies redefining the physical world. At Nominal, we're lucky to call many of these hardware organizations customers, and there are more that will be soon. In many ways, the Lux schoolhouse was where some of the earliest theses of Nominal were first pressure-tested. When @brycestrauss and I met with Josh in Lux's New York office in 2022 and pitched him on our idea to help engineers test hardware like software, the vision and ambition was clear, but the details uncertain, and we knew the entrepreneurial journey would be winding. He told us he believed we'd figure it out and he'd bet on us to succeed if we were building literally anything – even "an ice cream shop." 3.5 years and a $B company later, I look back and smile at that conversation. Here's to many more years of partnership – thanks for believing in us. To @wolfejosh, @peterjhebert, @Farshchi, Shaq, @graceisford, and many more.

Ambition is timeless. Tools are not. Today we are announcing an $80M acceleration round at a $1B valuation led by @foundersfund, with participation from @sequoia @generalcatalyst @Lux_Capital @RedGlass @lightspeedvp Avenir, and @Haystack. In the last year, we grew revenue 7x, added some of the world’s most serious hardware enterprises, and built over twice the new product surface area than in our prior two years combined. The most ambitious and fastest-moving hardware teams choose Nominal, from aerospace to energy, and automotive to defense, because it helps them learn faster and scale more efficiently. Our mission is to bring that advantage to every team building in the physical world. For those not yet using Nominal, we hope to have the opportunity to earn your trust and work hard for you and your company soon. All systems Nominal. 🚀 Thank you @business and @EdLudlow for sharing out story. And @traestephens and @zebulgar for doubling down. Read more about our announcement here: bloomberg.com/news/articles/…

To this end, we developed Multi-Scale Embodied Memory (MEM). The key idea: use different modalities to represent memory at different time scales. 📹 For short horizon memory, we developed an efficient video encoder that lets the model remember fine-grained details about its recent interactions. 📜 For long horizon memory, we train the model to summarize events in text, allowing it to remember events for up to 15 min.

Without memory, robots can struggle in many real-world situations, like failing the same way over and over again, getting stuck, or losing track of time. But we can’t just cram minutes or hours of video into our model’s context – that would quickly become intractable.