Silas Beck

Co-Founder & Advisor @ Robotensor
The intelligence layer for robots.

[Robotensor @robotensor_ai]

Introducing Attune — Bittensor Subnet 58 (SN58)

The open frontier for adaptive robot intelligence.

We're excited to announce the launch of Attune (SN58), a new @bittensor subnet built by Robotensor to accelerate the development of adaptive robot models.

Today's robots are powerful, but adapting them to new tasks still requires collecting data, retraining models, and reprogramming workflows.

We believe robots should learn new work simply by being shown what to do.

Attune brings researchers, engineers, and miners together to compete in building the next generation of robot intelligence.

Our research focuses on:

- In-Context Learning (ICL) — Learning new tasks from demonstrations without retraining.
- Generalization — Adapting to new objects, environments, and robot configurations.
- World Action Models (WAMs) — Learning physical dynamics and predicting future actions.
- Efficient Inference — Moving toward real-time, closed-loop robotic control.

Our first competition, Vector, is now open. 🚀

Miners train and submit model weights, while validators evaluate performance through randomized simulation and transparent, reproducible head-to-head competitions.

Our upcoming competition, Horizon, will expand toward human-video learning, World Action Models, and efficient inference.

We believe the future of robotics will be built through open research, measurable progress, and global collaboration.

A robot should not need a new brain every time the work changes.

Join us in building that future.

🌐 Website: https://t.co/rpfNDFUalu

📖 Documentation: https://t.co/u5neoWvb0w

📄 Whitepaper: https://t.co/tMztQvbgAp

💻 GitHub: https://t.co/zr2fqyF63k
Attune | SN58 | Powered by @bittensor

#Bittensor #Robotensor #Attune #Robotics #PhysicalAI

[Silas Beck @SilasBeckRobots]

@0xGenpix If we're collectively building a good future for the coming generations, there is no reason anyone needs to be unsettled about it

[Silas Beck @SilasBeckRobots]

Would you buy a humanoid robot for your business, if it didn't come with any software or AI?

[Silas Beck @SilasBeckRobots]

@clankrmedia This is awesome! I want a Vibe A1 for home next

[Robotensor @robotensor_ai]

Zero-WAM is a World-Action Model (WAM) developed by researchers from @robbyant_brain and HKUST. Its key advance is in-context task generalization. It can take a human demonstration video or a text instruction as the task prompt. This lets the same WAM infer new tasks from richer deployment-time context and execute them without task-specific parameter updates.

A key enabler is HumanGen. It automatically converts robot trajectories into semantically matched human demonstration videos. This produced 74.2K human–robot pairs across 8.6K tasks, giving Zero-WAM broad task diversity for learning from demonstrations.

The results are strong. On seven unseen RoboTwin 2.0 tasks, Zero-WAM reached 46.95% average success, versus 17.45% for LingBot-VA and 10.98% for WAN-Action. Real-world experiments also showed generalization to unseen task configurations.

In short: earlier WAMs primarily learn to predict the world and act. Zero-WAM extends this toward demo/text → infer task → predict world → act, making WAMs substantially more flexible for unseen tasks.

#Robotics #WorldModels #EmbodiedAI

[Silas Beck @SilasBeckRobots]

What do you think?

Will businesses and consumers want more general-purpose robots? Or mostly special-purpose robots?

[thijs @cdngdev]

i gave astra a robot, a paint brush, and a camera then asked it to paint the golden gate bridge in real life!

it figured out how to control the robot, and progressively got better throughout its attempts. the timelapse is sick

[Silas Beck @SilasBeckRobots]

If you want to have robots helping automate daily tasks, which do you need more?

A. Robot bodies?
B. Robot intelligence?
C. Affordable robots?
D. Plentiful robots?

Well?

If you answered E, all of the above, you are correct!

Robot bodies + brains + affordable + plentiful

[Robotensor @robotensor_ai]

A robot that works perfectly in one setup isn’t enough.

The real challenge starts when the object changes, the layout shifts, or the task is slightly different. We think the next generation of robot intelligence will be defined by how quickly it can adapt to those changes.

The goal is simple: build systems that keep learning, improving, and working in the real world.

#PhysicalAI #Robotics #EmbodiedAI

[Silas Beck @SilasBeckRobots]

What LLMs did for writing... making it dynamic instead of hand-crafted or bespoke

That's what we're using ICIL for Embodied Robot Intelligence.

Just give your robot a demo video (or show it live), and it will correctly perform the task.

[Silas Beck @SilasBeckRobots]

@MaxSebti Yes!

We miss every shot we don't take in life

[Silas Beck @SilasBeckRobots]

@RvCrypto Yes!

Almost no-one understands the ecosystem.

We need to do all we can to make it understandable to non-crypto people

[Silas Beck @SilasBeckRobots]

@gemsmorro $1B is nothing.

With the Bittensor engine, and work applied to solving real-world problems in a verifiable way, we're going to have many $1B - $10B subnets within 5 years

[Silas Beck @SilasBeckRobots]

Developing in-house Robot Intelligence requires a massive commitment of resources. Money, tooling, experts.

What if it was possible to harness the world's collective intelligence to train your Robots, on-demand, with just a minimal cost?

[Silas Beck @SilasBeckRobots]

@OnChainFox Agreed! What's unique about Bittensor is that the network can solve real-world problems.

The more directly those are solved for people, the more useful the mining work becomes, and the more the world will need of it.

Categorically different than classic crypto-mining.

[Silas Beck @SilasBeckRobots]

AGI arrived in December of 2025 with Anthropic's amazing Opus 4.5 model.

The AI revolution is here.

But, Robots haven't reached their AGI moment yet...

Who will release the model that unleashes generally useful Robot intelligence into the world?