Kevin L

Co-founder @ Dreamscale Labs

[Kevin L @KDog112358]

@jerrydjin @DreamscaleLabs @ycombinator Forever my day one ❤️
You're next my friend :)

[Kevin L @KDog112358]

@DreamscaleLabs is joining @ycombinator

A question I keep coming back to is: what does it actually mean to be intelligent?

To me, intelligence is most interesting when you don’t already know the answer.

You encounter something you haven’t seen before. You look at what you have seen, find the pieces that seem relevant, and synthesise your best guess at what to do next.

If we want robots capable of doing that across the enormous complexity of the physical world, why should we expect all of their intelligence to fit inside a computer the size of a tissue box?

We don’t think a robot’s physical form should define the limits of its intelligence. As physical AI models become more capable, they are also becoming more computationally demanding. Today, robot developers are forced to trade model capability against latency, power, and increasingly expensive onboard compute.

At Dreamscale, we’re building real-time cloud inference for robotics: infrastructure that lets robots run frontier models remotely, over low-latency connections. We’re raising the ceiling on robotic intelligence from the computer we can fit inside a robot to the most powerful computers we know how to build.

And by the greatest stroke of luck, I’m building this with @AntoineNeedGPUs, @pentestduck and @eternaI_entropy, a group of people who seem to share one useful defect: none of us are particularly interested in the easy path.

We argue, build, break things, and prove ourselves wrong. Every now and again, we discover that we might be right. I wouldn’t trade that for an easier problem.

The bet we’re making is simple: robot brains belong in the cloud.

We intend to prove it.

[Kevin L @KDog112358]

@pentestduck @DreamscaleLabs @ycombinator Time to put robot brains in the cloud together 🤖

[Antoine Nguyen @AntoineNeedGPUs]

Robot brains can (and should) live in the cloud 🐑 ⚡

OpenAI's Astra showed us truely competent spatial reasoning arises out of enormous scale and data.

Skyrocketing dram costs, onboard energy limits and most painfully the need for generality afforded only by scaling are forcing us off board onto datacenter grade compute.

So we're now announcing @DreamscaleLabs , an applied research company engineering this future into a possible reality, gutting the latency and amping up the reliability necessary for safe remote operation.

Onwards to Physical AGI 🚀

@pentestduck @KDog112358 @eternaI_entropy

[Dreamscale Labs (YC F26) @DreamscaleLabs]

@DreamscaleLabs is joining @ycombinator to build the fastest cloud inference platform for robotics!

[Kevin L @KDog112358]

A robot’s intelligence shouldn’t be capped by the GPU bolted to its chassis.

We’re starting to see the first signs that robotics is becoming a true compute-scaling problem:

• @GeneralistAI’s GEN-1.5 and @SkildAI’s S1 are showing early signs of physical in-context learning — robots adapting to entirely new tasks from a single demonstration.

• @GoogleDeepMind’s Gemini Robotics shows embodied reasoning improving as you give the model more test-time compute.

• New systems like τ₀-VLA are explicitly spending additional inference compute on planning, world-model rollouts and difficult decisions.

This changes the cloud robotics argument.

The question is no longer simply whether a capable model can fit onboard.

Truly intelligent robots will demand truly capable hardware. And we believe cloud compute will let robots access levels of intelligence that would be impossible to permanently provision inside every chassis.

To lay out our thesis, we wrote about why we believe robot brains will increasingly live in the cloud:

https://t.co/rqc4OHD4sb

[Kevin L @KDog112358]

How long until FelonyBench saturates? 🤔

[Meryem Arik @MeryemArik9]

Pictured: Google Deepmind CEO Demis Hassabis steps down to become chair

[Kevin L @KDog112358]

Chatbot intelligence lives in the cloud.

Why not robot intelligence too?

[Chris Yoo @pentestduck]

Cloud inference will never be fast enough for robotics?

Here's MolmoAct2 VLA model running over cloud on an SO-101 arm.

Smooth motions despite the DC being 1000km away.

[Kevin L @KDog112358]

Feeling the vibes 🤖

[Chris Yoo @pentestduck]

The 4 Aussie Physical AI builders have landed in SF! 🇦🇺 🤖

And of course we had to bring our robot arms with us.

If you're a robotics founder, builder, researcher, or even a tinkerer excited about the future of generalist AI model inference for robots, let's chat!

[Kevin L @KDog112358]

Model predictive control is the backbone of any world-based model control approach.

Extending the horizon of "good predictions" will inevitably come from hard baking physical constraints into these models so this sort of dynamics based regularisation is a great step forward!

[Haldun Balim @haldunbalim1]

❌Diffusion planners are missing a fundamental ingredient of classical control: an explicit model of the system dynamics.

📍We address this gap by introducing Model Predictive Diffuser, a new diffusion planning framework with a novel sampling strategy alternating between a planner and a dynamics model.

✅Achieving better performance, greater adaptability, improved sample efficiency and more!