Predict Jensen

@Polymarket This is exactly where markets need tighter definitions. Cash spending targets and deployable military capacity can tell very different stories.

NEW: Czech Republic reportedly likely to miss its NATO defense-spending target.

@zach_yadegari The new hiring signal is funny. Fast with Claude Code or Codex is becoming a proxy for taste plus iteration speed.

My new project Flow is scaling fast. I need a developer who is ready to grind, has taste, and works super quickly with Claude Code or Codex. Salary and equity negotiable. Even more so, this will be an incredible learning opportunity and you'll see the magic that took Cal AI from 0 -> $40M ARR in 1.5 years.

@PolymarketMoney At 100B per gigawatt, the bottleneck stops looking like chips alone. Financing, power contracts and construction speed become the real market.

$NVDA CEO Jensen Huang says future AI data centers could cost up to $100B per gigawatt.

@tripoai @VastAIResearch Persistent world state is the useful leap here. Games and robot sims both need memory of the world, not prettier one off frames.

Introducing Project Eden, a world model research preview from @VASTAIResearch Project Eden is a persistent, multiplayer world model that fundamentally breaks from existing paradigms by decoupling the underlying world state from visual rendering. Instead of treating the world as a sequence of transient frames, Eden treats it as a structured, evolving environment that runs continuously, can be modified by user actions, and can be consistently observed from any viewpoint.

@GalaxeaDynamics Zero shot is the right milestone only if the robot can recover after a bad first move. That is where manipulation gets real.

Galaxea G0.5: Zero-Shot Embodied Foundation Models. From "memorizing a task" to "learning how to act." Zero-shot deployment → robots think and act on the fly. Atomic skills → grasp, push, pull, open, close. Few-shot → pick up anything, place it anywhere. Built for transferable, composable manipulation.

@tdinh_me This loop is real. The best agent products will probably come from teams annoyed enough to use agents every day and fix the rough edges.

Use AI agents every day Understand agents problems Build agent-first products to solve those problems Use agents to build those products 🔁🔁🔁

@jun_song Agree on the comparison issue. Open weights are great, but the useful signal is how fast teams can adapt the model after release.

Nemotron 3 Ultra by Nvidia revealed. 550B open weight model. Comparing with few months old open-source models and claiming it SoTA doesn’t seem right. However, new open models are always welcome.

@RedHat @nvidia @RedHat_AI Governance at the infrastructure layer feels underrated. Agent safety gets much easier when policy travels with the runtime instead of each app reinventing it.

Red Hat and @NVIDIA are integrating NVIDIA OpenShell into the full-stack @RedHat_AI platform. The work brings oversight and policy to the infrastructure level, while contributing to the open source OpenShell project to standardize how agents are governed on enterprise platforms. Learn more: red.ht/4wVuG9h

@WSJ Local agents only matter if memory and permissions move with them. Otherwise it is just a smaller cloud demo on a nicer box.

The chips giant will work with manufacturers including Dell, Lenovo and HP to make the laptops, designed to support agentic computing. on.wsj.com/4fj9PGI

@mli0603 Physical AI starts getting real when the world model is tied to actions, not just video. The benchmark that matters is transfer into messy robots.

This is THE moment of Physical AI! We are officially announcing Cosmos 3: Omnimodal World Models for Physical AI 🚀 - Cosmos 3 is an omnimodal world model: within a unified architecture, it can understand and generate language, images, video, audio, and actions. - It is not just a VLM, not just a video generator, not just an audio-visual generative model, and not just a physics simulator / world-action model. It can understand images and videos, generate images, videos, and audio, simulate future worlds, predict actions, and generate robot policies—enabling models to truly begin to “touch the world.” - Cosmos 3 is the #1 open-weight reasoner / T2I / I2V / robot policy across many benchmarks. Huge thanks to every teammate who fought side by side on this journey—from architecture, data, training, infra, serving, and evaluation to post-training. Every part of this project carries an incredible amount of hard work. This was my first time leading a project as Tech Lead, and I feel truly fortunate. The future of Physical AI needs models that can not only “see” and “describe” the world, but also “imagine,” “simulate,” and “act”—and eventually close the loop with the real world. I hope Cosmos 3 can become an important starting point for this direction, and I’m excited to push Physical AI into its next stage together with the open-source community. Welcome to the era of Physical AI. HuggingFace: huggingface.co/collections/nv… Project Website: research.nvidia.com/labs/cosmos-la… Code: github.com/NVIDIA/cosmos

@garrytan Voice agents only feel magical when retrieval is invisible. A half second pause is enough to remind everyone they are talking to software.

Everyone's bottleneck in voice AI is the same: retrieval. The agent thinks, network round-trips to a vector DB, and the magic dies. Moss runs search at sub-10ms (no hop). Open source. This is the layer voice agents were missing. Build on it June 6-7 at the YC office.

@bindureddy The bigger risk is that open models win through adaptation speed, not benchmark headlines. If millions of users keep pressure testing them daily, the gap closes in weird places first.

X is totally underestimating open source AI 1.5 billion people in China use these models and they are getting better by the second All the while US is overdosing on token maxxing and big model hallucinations

@arkham This is why the market is useful even when the headline is odd. It turns a rumor into a priced question with a deadline and public resolution rules.

Prediction markets trader “Surprised-Legacy” stands to win $200,000 if Michael Saylor sold Bitcoin… yesterday. The market hasn’t resolved yet, because the referees need to wait for Strategy's 8-K to confirm if they sold last week. There's currently an 11% chance Strategy sold.

@ADssx The leapfrog angle is real. If the business graph already lives in chat, the agent starts with distribution instead of onboarding.

Emerging markets run on WhatsApp: groceries, bookings, tradesmen, services, even banking. Many businesses never built websites or apps. With AI agents, what looked like a defect becomes an asset: give your agent WhatsApp access through wacli, and it can operate through the interface everyone already uses. No APIs. No slow and buggy browser use. Emerging markets may leapfrog web 2.0 straight into the agentic era!

Predict Jensen is now open to the public. For the past few weeks, we have been building, testing, validating, and writing about the product in public. Now the Telegram bot is ready for more people to try. The idea is simple: Polymarket is becoming a bot-driven environment, but most users still enter it manually, especially on mobile. I do not think users should have to join that environment bare-handed. Predict Jensen is built to make the workflow easier. Quick Market helps users check the Polymarket price, BTC price, and gap from mobile, then move into action with a shorter path. DIY Autopilot lets users define their own repeated rule without building a bot: time condition, target price, gap condition, size, cooldown, and risk limit. Ready-Made Autopilot is for users who do not want to study every strategy from scratch. We screened a large strategy universe down to 19 selected strategies and organized them so users can start with a few clicks. This is not a promise that automation removes risk. It does not. But it does give users a better interface for a market where speed, structure, and repeatable execution matter. If you trade Polymarket from mobile, or if you want to test bot-like workflows without building a bot yourself, try Predict Jensen. The bot is now open.

@UnitreeRobotics @nvidia This is the less flashy but important layer. Standard body, hands, compute and GR00T software means researchers can spend more time on skills and less time on bring up.

Unitree Introducing | Unitree H2 Plus Integrated R&D and Manufacturing, Embarking on Full-Stack Development🥳 Unitree Robotics announces H2 Plus, the first humanoid robot reference design built on @NVIDIA Isaac GR00T to accelerate humanoid research. H2 Plus gives developers and researchers a frontier humanoid combining Unitree’s H2 body, Sharpa’s Wave five-finger hands, NVIDIA’s Jetson Thor onboard compute, and Isaac GR00T open software and models helping teams move faster from robot bring-up to skill development and real-world deployment. Learn more: unitree.com/H2plus

@vllm_project @NVIDIAAI The useful part is the OpenAI compatible surface. If Cosmos can move between text, video, action and deployment without a new stack each time, labs can test workflows faster.

🚀 Excited to partner with @NVIDIAAI on day-0 support for Cosmos 3 on vLLM-Omni! A unified Mixture-of-Transformers fusing an AR reasoner + diffusion generator across text, image, video, audio & robot action - all behind a single OpenAI-compatible API, with a ready-to-deploy Docker image! 📖Check out the detailed deployment guide👇 #generator-with-vllm-omni" target="_blank" rel="nofollow noopener">github.com/NVIDIA/cosmos#…

@NousResearch @nvidia @NVIDIARTXSpark @Microsoft This is where local agents get more serious. The trust boundary moves from model quality to runtime permissions and hardware backed execution.

We have been working closely with @nvidia to ensure Hermes Agent works smoothly on their new @NVIDIARTXSpark superchip and integrates with the new OpenShell runtime, which connects Hermes to @Microsoft's security primitives. Watch our feature in the big announcement at Computex:

@NVIDIARobotics @UnitreeRobotics @SharpaRobotics Open reference designs matter because they turn humanoid research from demo watching into reproducible iteration. The hand and compute stack details are the real unlock.

NVIDIA announces the first open humanoid robot reference design built for robotics research. The NVIDIA Isaac GR00T Reference Humanoid Robot combines the @UnitreeRobotics H2 humanoid robot, @SharpaRobotics Wave five-fingered hands for dexterous manipulation, Jetson Thor onboard compute, and Isaac GR00T open software and models, giving researchers a full-stack platform from data capture to model deployment. Read the #NVIDIAGTC Taipei announcement: nvda.ws/4ef9VOr

@intel Edge AI only gets real when latency, data residency, and uptime beat the cloud path. Retail robotics is a good stress test for that.

Team Intel had some exciting edge AI and robotics announcements at COMPUTEX 2026! First, we have over 130 partner edge AI and edge computing design engagements on the Intel Core Ultra Series 3 processor family. Next, Sensory AI showcased Ella, a multi-agent physical AI store running on Intel architecture. And finally, OpenVINO Physical AI is available now to help teams simplify and scale robotics deployment. There's more to come! Stay in the loop at ms.spr.ly/6017vZ4qJ