Javier F. Figueras

Father & Husband || Physical A.I. @AWS || Technology BD & Strategy //Prev. @MagicLeap @Microsoft @Oracle | BoD @BlueMissions || Thoughts are my own

[General Intuition @gen_intuition]

General Intuition has raised another $220M at a $6.2B valuation as we start making our models available. These are an entirely new class of foundation models, trained on billions of action-labeled videos, capable of acting in realtime in never seen before environments.

1/6

[Javier F. Figueras @Javi_Figueras]

Epic

[Nic Cruz Patane @niccruzpatane]

From concept to reality.

@SpaceX has successfully deployed Starlink V3 satellites into Earth’s orbit and made contact with them for the first time.

At scale, one Starship carries 60 V3 satellites, the same network capacity as about 20 Falcon 9 launches.

[Nic Cruz Patane @niccruzpatane]

From concept to reality.

@SpaceX has successfully deployed Starlink V3 satellites into Earth’s orbit and made contact with them for the first time.

At scale, one Starship carries 60 V3 satellites, the same network capacity as about 20 Falcon 9 launches.

[Inter Miami CF @InterMiamiCF]

FREE KICK MAGIC FROM LEO. 🎯✨

[Anto Patrex @antopatrex1]

OpenAI shut down its robotics team in 2021 explicitly because of the data problem, then restarted it later. having the best LLM didn't solve that then.

everyone assuming a frontier VLM solves robotics runs into the exact same wall:

action data is the actual bottleneck. the internet has no torques, contact forces, proprioception, or failure-recovery trajectories. pretraining compounds for the "what," while the "how" has to be physically collected, costs per hour, and doesn't scale like tokens.

embodiment fragmentation breaks the data pool further. physical data is partly tied to specific hands, kinematics, and sensors. cross-embodiment transfer helps, but it isn't free.

evals happen in the real world and take days, not minutes. that blunts the core advantage of frontier AI labs, which is fast scaled iteration.

the long tail is operational. reaching 99.9% reliability in messy sites means dealing with hardware, safety, maintenance, and customer ops. that is a low-margin, atoms-heavy business, a poor fit for a software-margin org.

the model backbone is commoditizing anyway. open VLMs are good-enough starting points. Physical Intelligence, Skild, and others built competitive policies without owning a frontier LLM.

[Javier F. Figueras @Javi_Figueras]

One of our brilliant colleagues at AWS published this comprehensive Physical AI 101 for public consumption and free learning. No gates. Enjoy!

#PhysicalAI #Robotics #Autonomy

https://t.co/lu3GWmd0w1

[Mark Zuckerberg @finkd]

Here's everything I announced at Meta Connect today 👇

[Javier F. Figueras @Javi_Figueras]

https://t.co/0wQJCReqYZ

[Perry Dong @perryadong]

What will be the “RLHF” moment for robotics? What will it take to get robotics to where LLMs are today and beyond?

New blog post with @chelseabfinn sharing some thoughts on the state of RL for frontier robotics models and what's missing 👇

Blog: https://t.co/crBN1jL5HH

[Javier F. Figueras @Javi_Figueras]

https://t.co/P2SgiMBORY

[Oliver Cameron @olivercameron]

I spent 2015-2023 building self-driving cars. It was so painful.

It's now 2026, and we tasked just 1 researcher to take Odyssey-3, our latest world model, and teach it to drive the roads of India with just 20 hours of driving data.

And…it works! This is insane.

[Dwarkesh Patel @dwarkesh_sp]

Visited the lab - was struck both by how wide the search space is for materials synthesis experiments, and also how amenable it is to depth first search, where the design and informativeness of your next experiment improves as you pile up more data from previous runs.

[Tongzhou Mu 🤖🦾🦿 @tongzhou_mu]

Does scaling pre-training on general web video improve a complex manipulation task in real deployment?
We scale model size and pre-training compute, and test on one industrial task.
Yes. The better a pre-trained model predicts web video, the better its post-trained policy. 🧵

[Robotaxi @robotaxi]

Our unboxed process for Cybercab assembles vehicle modules in parallel & frames the car in one step at the end.

This streamlines automation & reduces the line size by half, unlocking much greater production capacity.

The first real revolution in automotive manufacturing in over a century

[Jay Chooi @chooi_jeq]

GPT-6 Astra scored 95% on a robot control task, up from Fable 5.1's 40%, with 6.2x fewer output tokens at 2.3x lower cost. 🧵

[Javier F. Figueras @Javi_Figueras]

🚀🚀🚀

[COATUE @coatuemgmt]

Chart of the day:

GenAI Spend via Cloud Providers has surged 15-20x across every cohort in just 18 months!

[World Labs @theworldlabs]

Introducing Atlas:

The world's first multimodal world model that generates image and video frames with pixel-perfect camera control and reconstructs them in 3D.

Model the world, move the camera, and simulate space & time.

[🇦🇷 Selección Argentina ⭐⭐⭐ @Argentina]

Gracias eternas, Leo. Te vamos a amar hasta el final de nuestros días 🩵🤍🩵

[Dyna Robotics @DynaRobotics]

After so many demos, models, pilots, robots are still struggling to land real deployments with real customers. Until now.

Today we’re excited to share that our robots have successfully crossed the ROI threshold, and Din Tai Fung, one of the highest revenue per location restaurant chain in the US, is rolling out Dyna robots across its extensive restaurant network.

This brings our rollouts across hotels, logistics, data centers, and many other use cases to a fleet that reaches hundreds of robots by the first half of 2027. And we’re just getting started.

It’s been a wild year, and today we’re double clicking on the battlefield stories and sharing a few learnings about scaling robot deployments. We are just scratching the surface.

Read the full blog post: https://t.co/8QVEoQuzBY

[Skild AI @SkildAI]

Introducing S1, our new foundation model that learns from one example.

It can be taught 10-minute long tasks that it has never seen before, from one video prompt without any fine-tuning.

Watch S1 operate in real-time via in-context learning: