Chris Paxton

298 Followers
103 Following
44 Posts

Robotics research scientist at Meta AI/FAIR, formerly NVIDIA. Making robots that can solve hard problems with people. All views my own.

he/him

Websitehttps://cpaxton.github.io/
Scholarhttps://scholar.google.com/citations?user=I1mOQpAAAAAJ&hl=en
Twitterhttps://twitter.com/chris_j_paxton

We're now seeing dramatic shifts in the way conflicts between near-peer nations play out because robotics has made “lower sky” air power very accessible for poorer but technically capable nations.

Read more about it. Curious what people think.

I'm writing a sequence of posts about how robotics is changing different industries and parts of society. One of the most obvious ones is… war.

https://open.substack.com/pub/itcanthink/p/war-always-changes?utm_source=share&utm_medium=android&r=3uj5js

War Always Changes

A brief look at how robotic drones have transformed warfare

It Can Think!
People I respect are speculating that OpenAI is pushing back their open-source release due to Kimi K2. It does strike me that this is what Llama 4 was supposed to be; a massive, impressive open-source MoE model that can form the basis for a new generation of agentic AI applications
People Are Using AI to Flirt on Dating Apps—Then Disappointing Their Dates

Some people in the dating scene: people who are using ChatGPT and other AI chatbots to write charming and witty texts...

VICE
Grok at 4 on aider polyglot. It's resoundingly clear there's no "best model" any more, just a best model for you and your use case

What are robot world models?

If you follow the robotics space, you'll have almost certainly heard the term "world model." In the end, these mean using generative ai to build data-driven simulators or planners in order to build general-purpose robots.

More:
https://open.substack.com/pub/itcanthink/p/what-are-robot-world-models?utm_source=share&utm_medium=android&r=3uj5js

What are Robot World Models?

Using generative AI for planning and data-driven simulation

It Can Think!
So I want to enable this llm agent (discord chatbot) to do arbitrary code execution. What's the suggested best practices there? Or is this one of those "don't do that you idiot" kind of things
@varavs yeah I agree
@aethrvmn yeah thats what it looks like
So all it took was about 56 data points per parameter and you can train grandmaster-level chess w/o search. Shows what a force-multiplier search is when it comes to data efficiency, but also how given unlimited data you can solve basically anything with our current architectures