
robotics is an agi problem so we should work on agi

your favorite robots' favorite robot

robotics is an agi problem so we should work on agi

fantastical and undoubtedly propagandic — this will definitely turn the broad population even more against ai than it already is. the characterization of the people even in this trailer is hyperbolic and 1-dimensional for the sake of absurdity
New trailer for Luca Guadagnino's ‘ARTIFICIAL’, starring Andrew Garfield as Sam Altman. The film follows the story involving OpenAI focused on the firing & rehiring of CEO Sam Altman. In theaters on January 8.

coincidentally (or maybe not 👀) both co-taught by the goat @tatsu_hashimoto

tbh these are the only two courses you need if you want to learn modern deep learning / LLMs. i've begun to recommend just these two to everyone who asks me

Robots trained only on teleop make good lab videos. To cross the chasm from lab to real world, Rhoda DVA learns from internet-scale video, then picks up a customer’s workflow with about 10 to 20 hours of task-specific training data. Thanks @BriWessling at @TheRobotReport for the conversation : therobotreport.com/how-rhoda…

is it just me or is the cursor autocomplete model actually crazy good at writing? the things it writes are like completely nonsensical but the actual prose seems a step above the current frontier models

not only are we scaling robotics, we're scaling offices as well! more office space -> more robots -> more evals -> profit
Wonderful first day at our new offices and robotics lab in Sunnyvale! A huge thank you to our facilities and IT team for setting up this amazing new space and making the move seamless!

always love hansheng's work - he brings a very principled and theory-driven approach to improving generative models!
What should a pixel flow model predict? Data, velocity, or others? New paper: AnisoFlow🔥 • A fully learnable anisotropic target • Theory that explains what it learns Results: • ImageNet: FID 1.90 → 1.55 over JiT, no REPA • Pixel T2I without patch artifacts, even in 8-bit


tl;dr: if you want to get accepted to cvpr make sure to juice your model but not your baseline‼️‼️
Having been in the image generation community for this long, I've noticed some tricks that may boost your generation performance at no cost. All the aggressive opinions below are my own, but credit for the findings goes to my amazing friends and collaborators @XiaoyuW9401147 @Weihang404 @lowerbad.🫡

i recreated THE ENTIRE WORLD in minecraft at 1:1 scale. every building on earth, including your house! to do this, i trained a SOTA world-to-minecraft generator on maps data, complete with furnished home interiors! play and check out your neighborhood: worldcraft.gg

just saw a guy type in “incognito” into his browser

this feels spiritually cyberpunk-esque where the rogue ai take over the old web so humans need to make an extremely isolated, walled-off new web
the sheer number of websites now blocking ai agents is astonishing. i often can’t do anything or even basic research cuz the underlying properties all refuse so i have to take screenshots of videos to share with my agents for anything. the agentic web might be even more silo’ed & balkanized than the one that exists today.

Many friends at CoRL worry that frontier models like Astra and Claude will take over robot control. I want to offer a different perspective. I see three routes: VLA: Start with vision-language pretraining, then train on actions. Transfer semantic and spatial understanding into robot capability. WAM: Start with video-generation pretraining, then learn actions. Transfer knowledge of how the world changes into robot capability. Astra/Claude: Bring broader visual, language, and coding capabilities to robot control. Their 3D understanding—and how much of it transfers to manipulation—is striking. My hypothesis is that learning across code and visual domains, such as Blender tasks, contributes to this capability. However, three questions remain for Astra/Claude: 1. Can latency fall enough to support responsive, closed-loop control? 2. How much capability remains with less chain-of-thought reasoning? 3. Can we distill that capability into a smaller, faster model without losing its generality? These are all questions about the cost of turning impressive capability into a practical robot policy. I expect progress on all three, though how much remains uncertain. The same optimism applies to WAMs and VLAs: I expect them to improve with scale. History tells us that the model that turns data and parameters into capability most efficiently will win. Astra and Claude show what’s possible at scale. Which approach gets there most efficiently is still an open question. The competition is just beginning. At Rhoda, we’re betting on WAMs, and our results give us concrete reasons to believe in their data and parameter efficiency [1] [2] [3]. We’re hiring across multiple roles. Join us to build the next generation of robot intelligence.rhoda.ai/career… [1] rhoda.ai/research/direct-vid… [2] x.com/rhodaai/status/2099914… [3] x.com/rhodaai/status/2117663…
At Rhoda, we care deeply about the science of pre-training for robotics. In one of the most rigorous studies of its kind, over thousands of trials and hundreds of hours of robot evaluations, we show how scaling web-video pre-training leads to better real-world robot performance.

all they’ve done is re-org their robotics team into core research. this is expected, unless you’ve been living under a rock
🚨 BREAKING: OpenAI has officially disbanded its robotics team, just five months after relaunching it it’s over

everyone wants to virtue signal but no one wants to be virtuous