Vision foundation models predict hurricane intensity from satellite imagery remarkably well — and pass out-of-distribution tests.
Yet their internal representations collapse exactly where physics matters most.
We call this the Perception–Physics Paradox. New at ICML 2026 🧵
If you are looking for an awesome PhD opportunity in robot learning, world model, and general AI, definitely check this out and work with @liu_yuejiang ! Yuejiang is super insightful and energetic, and always a joy to work with. Don’t miss this opportunity!
Excited to share that I’ll join @NUSComputing as an Assistant Professor in 2027
🏛️ I’ll build LEMA Lab: https://t.co/qzKA7dpA00, study the principles of embodied intelligence, & empower every lab member to thrive
📢 Recruiting 3-6 PhD students in the next application cycles
Introducing CHI-Bench on @huggingface: the world’s first long-horizon healthcare benchmark for AI agents.
75 real healthcare workflows + 20 apps + 200+ MCP tools + 1,290 skills + process / outcome rewards
https://t.co/PKmQ4RiIJY
Any questions, lmk!
Excited to share that I’ll be joining UIUC Statistics and CS (affiliate) @UofIllinois as an Assistant Professor in Fall 2026. Huge thanks to everyone for the support, and I’m looking forward to future collaborations!
I’m recruiting PhD students in AI/ML/Stats (26/27), and also welcome visiting students/interns.
Our research focuses on the next generation of intelligence with causal understanding and actions. Instead of only fitting data, we aim to uncover and leverage the underlying mechanisms behind it, spanning:
• Causality-based Learning / Trustworthy ML
• Foundation models / World Models / Multi-Agent Systems
• Scientific Discovery
We believe truthful understanding is important not only for trustworthiness, but also for efficiency: helping models learn, reason, and plan more effectively, with less noise, hallucination, and reliance on scaling.
More info: https://t.co/OwRJvV1xNO
(1/n) After a few months of work with multiple hospitals, universities and research facilities, today we're open-sourcing CHI-Bench: the first long-horizon benchmark for healthcare AI agents on real clinical and healthcare workflows.
Best frontier agent overall: 28% pass@1.
End-to-end prior authorization: 0%.
A thread on what we found 🧵
1/🧵Can AI agents automate U.S. healthcare workflows end to end given just clinician & insurer apps and operations, medical policy library? Introducing CHI-Bench: 75 long-horizon realistic healthcare workflows × 30 frontier agents. Best agent solves only 28% #AIinHealthcare 👇
I’ll be presenting Ada-Diffuser (https://t.co/SWcKK17Fif, Thu, Apr 23, 10:30 AM–1:00 PM, Pavilion 3, P3-#1808) - a unified framework with both theoretical recoverability guarantees for planning under hidden processes and a practical zig-zag autoregressive diffusion approach for identifying and learning latent factors in RL and imitation learning for world models.
will also present world-action-verifier (https://t.co/JCs0UkXdKI) on Apr 26 and 27 at the RSI and World Models workshops, and co-organize the ICBINB workshop on Apr 27. More 🧵 to come.
Come and say hi!
MBZUAI Machine Learning Winter School 2026: Representation Learning & GenAI (https://t.co/voU5FqSZE3)
on Feb. 9-13, 2026, in Abu Dhabi, UAE.
Application Deadline: Oct. 20, 2025!
Join us for an exciting 5-day program with world-class researchers! Funding available! #MBZUAI
🏹 Job alert: PhD Position on Learning Concepts with Theoretical Guarantees Using Causality and RL in Amsterdam
Join @AmlabUva and cooperate with @tudelft.
📍 Amsterdam 🇳🇱
📅 Apply by: June 15th
https://t.co/8IsrHffl6u
New PhD position at @AmlabUva on learning concepts with theoretical guarantees using #causality and #RL with me, Frans Oliehoek (TU Delft) and @herkevanhoof 💥
Deadline: 15 June
https://t.co/q4dxcRCko5
🥳RLBReW Social🥳 @ RLC (@RL_Conference )
Socials are 🧡 of any conference. We invite you to join us on the evening of August 9 to discuss wacky RL ideas, and find friends and collaborators!
RSVP here: https://t.co/URCGb7nKMq
It is officially less than a week before the workshop begins⌛️
The workshop schedule is posted here: https://t.co/GT8ovpoIPs
A complete list of accepted papers can be found here: https://t.co/fofuB6ts0A
The main issues with the RL framework are the low information content of the reward signal, the task-specificity / lack of generality of reward signals, and the question where reward comes from in the first place (especially for abstract goals where recognizing success itself requires intelligence).
This workshop looks at various ways to do RL without handcrafted rewards, and it looks like they've accepted a bunch of very interesting papers.
⚠️ Final call for submitting your ideas on RL beyond Rewards at the Reinforcement Learning Conference @RL_Conference! Check out our earlier posts for more info and call for papers.