Aether AI (Causal Intelligence) @AetherLab_AI

A great discussion at the Robotics & World Models Reading Club on where today’s world models begin to fall short in robotics. @huang_biwei shared her perspective on moving from prediction toward models that can support intervention, adaptation, and reasoning when environments change. Thanks to the @saturdayrobotic community for the thoughtful questions and discussion. #AetherAI #Robotics #WorldModels #PhysicalAI

Biwei Huang @huang_biwei

3 days ago

Thank you @saturdayrobotic for having me at the Robotics & World Models Reading Club. I shared why I think today’s LLMs, video generators, and JEPA-style models are important, but still not enough for robotics. A causal world model should not only predict what may happen next.

6 4 44 8K 16

@Ambani_Wessley @huang_biwei Thank you, Wesley. Really appreciate the kind words. This has been years of work for Prof. Huang and the team, and we’re excited to finally share more of what we’ve been building.

Aether AI has raised $20M in seed funding, backed by a group of leading global investors with deep expertise in artificial intelligence and frontier technologies. Founded by Prof. Biwei Huang（@huang_biwei）. We're building Causal World Models for Real World Intelligence. 🔗aetherlabs.ai

@Partnerly_us @huang_biwei Thank you! We really like how you put it. The “why” is exactly the part we care about, especially when AI needs to move from prediction into the real world. Lots ahead to build.

Our founder @huang_biwei joined @Yuancheng for a conversation on world models, robotics, and the direction behind Aether AI. The discussion covers video generation, VLA, WAM, and why robotics requires models that can reason about actions, changes, and physical consequences in the real world. This is closely connected to the work we are doing at Aether AI as we build causal world models for Physical AI. We are also growing our team. Learn more at aetherlabs.ai #AetherAI #Robotics #CausalAI #PhysicalAI #EmbodiedAI #AIResearch

Biwei Huang @huang_biwei

6 days ago

I enjoyed joining @Yuancheng to talk about world models and robotics. The term “world model” is being used in many ways today, from video generation to VLA and WAM. In robotics, the question becomes more concrete. A model needs to understand not only what may happen next, but

0 0 1 5K 0

Last night @huang_biwei joined @GeekPark founder Jack Zhang to make the case for causal world models. The short version: scaling correlation hits a wall in the physical world. Causal structure — knowing why, not just what next — changes the economics of robotics. Our benchmarks vs conventional world models: 25–50% higher success rates, 5–10× fewer samples. Same data, deeper structure. One brain. Many robots. That's the bet. Full conversation → [weixin.qq.com/sph/AuNNpbY1ew]

GeekPark @GeekParkHQ

2 weeks ago

Why Hasn't Embodied AI Had Its GPT Moment? Last night on #GeekParkLive, Jack Zhang sat down with Biwei Huang @huang_biwei, founder of @AetherLab_AI. The question on the table: can next-token prediction ever really understand why things happen? And does that matter for physical

1 1 0 5K 5

Today is the day. Tonight, 9PM Beijing time(06:00 AM PDT), Prof. @huang_biwei goes live with Zhang Peng, Founder and President of @thegeekpark. Topic: Why causal world model is the next AI paradigm? LIVE conversation. No scripts. Straightforward take on where AI is heading. See you in the livestream! #CausalAI #AetherAI #BiweiHuang #GeekPark

Our founder @huang_biwei is doing something different tomorrow night! She'll be in conversation with Zhang Peng, Founder and President of @thegeekpark. Don't miss her unfiltered discussion on why causal world model is the next AI paradigm. Tomorrow, 9 PM Beijing time(06:00 AM PDT). #CausalAI #AetherAI #GeekPark

Biwei Huang @huang_biwei

2 weeks ago

Tomorrow night I'm doing something different. Usually I give structured talks — slides, data, the comfort of a prepared script. Tomorrow on @thegeekpark, I'll just be having a conversation with Zhang Peng, Founder and President of GeekPark. Live. Unfiltered. We'll talk about

3 1 9 2K 1

Aether AI ranked 2nd in the ManipArena Challenge at #CVPR2026! Proud of Junbo Huang and the team for this result. ManipArena was a tough real-robot manipulation challenge with 20 reasoning-intensive tasks. Big thanks to the organizers for putting this together. #ManipArena #Robotics #EmbodiedAI

Aether AI is at @CVPR! Find us at Booth #715 today and tomorrow. We’d love to talk about causal world models, Physical AI, embodied intelligence, and how AI systems can better understand and act in the real world. Stop by and say hi!!! #PhysicalAI #EmbodiedAI #Causality #Robotics #AIforScience #AIHiring #BayAreaJobs #AIResearch #cvpr2026 #MultimodalAI #CausalAI

keyboard_arrow_left Previous keyboard_arrow_right Next

Prof. Biwei Huang shares Aether AI’s core perspective from @CVPR: Physical AI needs Causal World Models — systems that reason about interventions, counterfactuals, and causal structure, not just statistical patterns. #CausalAI #PhysicalAI #CVPR2026 #AetherAI

Biwei Huang @huang_biwei

3 weeks ago

1/4 At @CVPR, I presented a central question: Why do frontier AI models—from GPT to Sora—still struggle with the physical world? My answer: they learn statistical correlation, not causality. Physical AI needs Causal World Models. #CausalAI #PhysicalAI #CVPR2026 #AetherAI

2 3 10 14K 3

The standard response to robot failure is more data. Robot drops the cup? Collect more grasping examples. Robot gets stuck? Add more scenarios to the training set. Models underperform? Scale them up. This works, until it doesn't. Every new environment requires new data. Every new task requires new training. Every new failure requires human intervention. The model gets bigger. The brittleness doesn't always go away. The physical world creates too many variations to cover with examples alone. Liquid level can change. Friction can change. Contact point, force angle, and numerous underlying conditions can all change and combine. More data expands coverage. But coverage alone is not an efficient way to expose the mechanism behind success and failure. The transition in robotics may not simply be from small models to large models. It may be from systems that imitate more trajectories to systems that model underlying mechanisms and adapt decisions as environments change. That's one of the problems we're working on at Aether AI.

🍻 Causal World Model Happy Hour #CVPR | @AetherLab_AI We're hosting a small (~30p), closed-door happy hour near the #CVPR venue — food, drinks, and the kind of unhurried conversation about world models, causality, embodied AI, and the decision brain of Physical AI that doesn't happen on the show floor. This is a invitation-only evening. We're hand-inviting a group of researchers and engineers whose work we follow — and we've opened a limited number of additional seats to apply. Because the room is intentionally small, every RSVP is reviewed. ✨ What we're building Aether AI is building causal world models — a new class of AI systems that understand mechanisms, reason under intervention, and operate reliably in real-world systems. We believe the next leap in AI won't come from simply scaling up to ever-larger models, but from paradigm-level innovation. ✨ Logistics 📅 Friday, June 5, 2026 · 18:00 – 21:00 📍 Walking distance from the CVPR venue, Denver. Exact location shared with approved guests by email. 🍸 Food and drinks will be provided. 🎟️ By approval only. Capacity is limited. You can also find us at Booth #715 any time during the Exhibition! 🎟️ Registration link: luma.com/wi0rtib8 🌐 aetherlabs.ai · 📩 [email protected] #WorldModel #PhysicalAI #EmbodiedAI #Causality #Robotics #AIforScience #AIHiring #BayAreaJobs #AIResearch #CVPR2026

5/ Physics does not adapt to models. Models must adapt to physics. That is why Physical AI needs models of mechanisms, interventions, and consequences. And why we founded #AetherAI: to build causal world models that help agents act reliably, even when environments, constraints, and consequences change.

4/ This is the core difference between imitating a past correlation and modeling the mechanism behind it. A correlation-based system may learn from experience: In scenes that look like this, grasping the handle usually lifts the cup cleanly. But things could be different in new scenes. If the hidden factors change — the cup is fuller, the surface is slipperier, the grasp angle is slightly different — the same action may no longer produce the same result. Instead, a causal model asks: if I apply this force, at this angle, under these physical conditions, what changes next?

1/ ChatGPT makes a mistake. You correct it with words. A robot makes a mistake. It changes the world. A glass falls. Liquid spills. The next attempt starts from a different state. The cost is not just a wrong output. It is a wrong action with real consequences.