Tbh, if I had Claude Code, Gemini, and ChatGPT during my PhD, I’d probably have graduated in 1 year instead of 5.5 years. My PhD was ~50% coding, 25% writing/polishing my papers, 25% reading others' papers. AI now accelerates each by at least 10×. Nothing will ever be the same.

Once someone demonstrates something is possible, the community moves so fast. This will be the year of real world RL I think. Looking forward to blog post!

One of those moments everyone had been waiting for, for a very long time! Looking forward to seeing all those cars on the highway turning into robots

Proud of the good work the team has put towards Core! 🚀

Same 👇

First tagged release (0.1.0) of Widgeteer. Widgeteer is like an MPC for your Qt application! An AI Agent can use it to interact with the GUI or to easily build integration tests. Give it a try github.com/AurynRobotics/…

I am sharing the most interesting experiment I have done with Claude Code. "Do whatever pleases you." It creates genuinely interesting and intellectually stimulating stuff. I am enjoying watching this more than scrolling on social media. Command here (add it to your

This is probably one of the most potential impactful innovations I saw in AI recently. si.inc/posts/fdm1/

We’ve developed a memory system for our models that provides both short-term visual memory and long-term semantic memory. Our approach allows us to train robots to perform long and complex tasks, like cleaning up a kitchen or preparing a grilled cheese sandwich from scratch 👇

For a long time, I was skeptical about action-conditioned video prediction for robotics. Many models look impressive, but once you ask them to handle long-horizon manipulation with real physical interaction, things quickly fall apart (e.g., Genie is amazing but mostly focused on

now you are obliging me to add 2D and 3D to #PlotJuggler... I am sorry, you wanted it...

Building is cheap now. Knowing what to build and how is still a fundamental skill yasint.dev/we-might-all-b…

Advanced Machine Intelligence (AMI) is building a new breed of AI systems that understand the world, have persistent memory, can reason and plan, and are controllable and safe. We’ve raised a $1.03B (~€890M) round from global investors who believe in our vision of universally

Unveiling our new startup Advanced Machine Intelligence (AMI Labs). We just completed our seed round: $1.03B / 890M€, one the largest seeds ever, probably the largest for a European company. We're hiring! [the background image is the Veil Nebula - a picture I took from my

Another great article that summarizes something that I have been thinking a lot: "Grief and the AI Split". Many of us feel a sense of grief: skills that took a long time to acquire are now being commoditized by AI. And we see two clear groups of people (thread) 🧵👇

We developed an RL method for fine-tuning our models for precise tasks in just a few hours or even minutes. Instead of training the whole model, we add an “RL token” output to π-0.6, our latest model, which is used by a tiny actor and critic to learn quickly with RL.

Back in Nov we developed Recap and trained π*-06 with RL. Now, we developed a fast *online* RL method that improves π-06 with as little as 15 min of robot data for precise tasks, using "RL tokens" exposed by our model that can be fed into a small actor-critic method.

Video datasets are finally big enough that perhaps we are all better off dropping images

JEPA are finally easy to train end-to-end without any tricks! Excited to introduce LeWorldModel: a stable, end-to-end JEPA that learns world models directly from pixels, no heuristics. 15M params, 1 GPU, and full planning <1 second. 📑: le-wm.github.io

Meet “Roadrunner,” a new 15kg (33 lb) bipedal wheeled robot prototype featuring multi-modal locomotion: side-by-side and inline wheel modes and and stepping. Its symmetric legs articulate at the knee for obstacle avoidance.