We're graduate students, postdocs, faculty and scientists at the cutting edge of artificial intelligence research.
Berkeley, CA
Joined July 2017
- Tweets1.7K
- Following488
- Followers299K
- Likes737
Berkeley AI Research retweeted
Come check out our poster at #EMNLP2026!
We present ParaStudent, a fine-tuned student simulator that revises code like a real learner, so AI tutors can be evaluated by how students take up their feedback before deployment 💬
📍Hall H, Sun 16:00–17:30
ParaStudent is accepted to #EMNLP2026 Main Conference!
Since this checkpoint we’ve been asking: what does a user simulator have to get right to evaluate interactive assistants like AI tutors? Turns out the key is how partially users act on advice 👀
🔗 arxiv.org/abs/2507.12674
Berkeley AI Research retweeted
We're excited to introduce AW-0: Agentic World Action Models! 🤖 (0/8)
The first World Action Model that plans in code, predicts the future, and controls the robot—all in one generative model.
AW-0 jointly generates:
🧠 Code - explicit skill calls, object arguments, and control flow
🎥 Future video - predictive visual dynamics
🦾 Robot actions - continuous motor control
All three are generated together through joint denoising, letting semantic plans directly shape both predicted futures and robot behavior, with:
🤝 Hybrid execution — generated programs can also run directly through perception and control APIs
⚡ Just 40 ms of additional inference latency over the base WAM on an RTX 5090
📚 Pretrained on ~3,000 hours of robot + human demonstrations
One model for planning, prediction, and execution.
🌐 Project: agentic-wam.github.io/
📄 Paper: arxiv.org/abs/2610.30136v1
💻 Code: github.com/ZhuoyangLiu2005/a…
🤗 Model: huggingface.co/miniFranka/ag…
🧵 How it works—and what we found ↓
Berkeley AI Research retweeted
I'll be at #EMNLP2026 in Budapest this week! 🎉 🇭🇺
Presenting "RAG over Thinking Traces Can Improve Reasoning Tasks" today at 11 AM
📍Poster Session A, Hall H.
Come say hi! happy to chat about reasoning, RAG, and everything in between!
1/ Thrilled to introduce T³: a corpus for RAG over reasoning tasks, built from thinking traces.
We show that surprisingly RAG can improve reasoning— with the right corpus.
Rag with Transformed Thinking Traces T³ gain by up to 43.9% on AIME 2025-2026.
🔗 arxiv.org/abs/2605.03344 🧵
Berkeley AI Research retweeted
We 'll be hiring very actively at Berkeley!
Assistant/ Associate/ Full Professor – Electrical Engineering & Computer Sciences (Deadline: Dec. 7, 2026 at 11:59 pm Pacific Time). Apply here: aprecruit.berkeley.edu/JPF05…
Assistant Professor – AI for Biology – Electrical Engineering & Computer Sciences and Center for Computational Biology (Deadline: Nov. 9, 2026 at 11:59 pm Pacific Time) Apply here: aprecruit.berkeley.edu/JPF05…
Assistant Professor – Intersection of Statistics and CS – Departments of Electrical Engineering & Computer Sciences and Statistics (Deadline: Nov. 10, 2026 at 11:59 pm Pacific Time). Apply here: aprecruit.berkeley.edu/JPF05…
Assistant/ Associate/ Full Professor – Computational Biomedicine - Center for Computational Biology (Deadline: Nov. 18 2026 at 11:59 pm Pacific Time). Apply here: aprecruit.berkeley.edu/JPF05…
Assistant / Associate Professor – Electronic, Functional, and Quantum Materials – Electrical Engineering & Computer Sciences and Materials Science & Engineering (Deadline: Dec. 7, 2026 at 11:59 pm Pacific Time). Apply here: aprecruit.berkeley.edu/JPF05…
Coming Soon! Assistant/ Associate/ Full Professor - AI and the Physical Sciences -Bakar Institute of Digital Materials for the Planet (BIDMaP) and home department(s) to be determined. (Deadline: Dec. 7, 2026 at 11:59 pm Pacific Time) Apply here: aprecruit.berkeley.edu/JPF05…
Coming Soon! Assistant/ Associate/ Full Professor-in-Residence - Gladstone Institute of Data Science and Biotechnology and home department(s) to be (Deadline: TBD at 11:59 pm Pacific Time) Apply here: aprecruit.berkeley.edu/JPF05…
Berkeley AI Research retweeted
1/🧵 Coding agents can make code faster. Can they also prove it still does the same thing?
Introducing Agentic Verified Optimization 🚀 Agents return faster code plus a Lean proof that it equals the original on every input.
🌐 Blog: aqueduct.verina.io
📄 Paper: arxiv.org/abs/2610.30188
Berkeley AI Research retweeted
We trained an imaginative agent! ✨
Agima: an omni agent that plans by imagining futures.
𝗦𝗙𝗧: Directly generalizes to harder tasks! 𝟮%→𝟰𝟬%, beats GPT-5.6 Sol (𝟮𝟬%).
𝗥𝗟: New ways to imagine emerge! Imagines a second plan only if the first one fails (𝟯𝟮%→𝟭𝟬𝟬%).
Berkeley AI Research retweeted
Leveraging research papers to push the frontier of the LLM evaluation!
Berkeley AI Research retweeted
FrontierCS 2 is our new benchmark for CS research & development.
Each task is based on a paper: we delete the contribution from the released code, and ask agents to outdo the original method, as judged by the authors' chosen metrics. Agents don't have access to the paper or web.
Berkeley AI Research retweeted
1/ 💰Multi-vector retrieval shouldn’t charge you by the token!
🎉 Excited to share FLINT (Fixed-budget Late INTeraction), a late-interaction retriever designed for long, complex queries with up to 89% reduction in index and 23-170× faster retrieval latency.⚡
🎛️ FLINT gives a vector-budget knob to balance storage, latency, and quality.
📄 Paper: arxiv.org/abs/2610.27273
How does it work? 🧵
ColBERT stores and compares one vector per token. FLINT pools every query and every document into a fixed number of vectors, so there’s far less to store and far fewer comparisons at search time.
Berkeley AI Research retweeted
My commentary for @Nature on this nice study - nature.com/articles/s41586-0…. - that develops and tests a personal AI health agent.
Accompanying @NatureNV pointing out the next big step would be to do a RCT for health outcomes nature.com/articles/d41586-0… by @2plus2make5
Berkeley AI Research retweeted
Excited to share TokenGate. By teaching LLMs to conditionally skim context tokens, we achieve up to 11.3x faster prefill and drastically reduce peak memory for long-context tasks!
Berkeley AI Research retweeted
Great work on fast long-context prefill. TokenGate is a trainable module that works with existing models to make long-context inference much more efficient.
Berkeley AI Research retweeted
Game changer for humanoid full-body control training. Now it's superfast (and cheap, just one GPU) to train a single policy on *all* the motions. How come:
(1) World model much faster than regular sim
(2) Backprop through time collects way more signal than policy gradients
We found a way to scale humanoid control that reduces GPU-hours by more than 1000×.
Across 129,514 motions, SuperMimic surpasses the state of the art (128 GPUs, 7 days) with 1 GPU in 3 hours.
Project: supermimic.github.io
Demo: supermimic.github.io/demo/
Berkeley AI Research retweeted
We found a way to scale humanoid control that reduces GPU-hours by more than 1000×.
Across 129,514 motions, SuperMimic surpasses the state of the art (128 GPUs, 7 days) with 1 GPU in 3 hours.
Project: supermimic.github.io
Demo: supermimic.github.io/demo/
Berkeley AI Research retweeted
Estimating empowerment from offline data is useful, because it provides us with an unsupervised objective for learning the relative "value" of states for reaching diverse outcomes. But it's notoriously hard to make this tractable. In our new paper, we present a tractable estimator, as well as some applications!
How should an agent decide where to explore, given unlabeled offline data? 🤖
In our new paper, we show that empowerment, a measure of how much control an agent has at a state, can be estimated from offline data and used to guide exploration!
ishirgarg.github.io/offline-…
(1/7)
Berkeley AI Research retweeted
How should an agent decide where to explore, given unlabeled offline data? 🤖
In our new paper, we show that empowerment, a measure of how much control an agent has at a state, can be estimated from offline data and used to guide exploration!
ishirgarg.github.io/offline-…
(1/7)
Berkeley AI Research retweeted
Turns out that simply extracting policies with supervised learning conditioned on the advantage can get state of the art results in RL, as long as we get a few important details right!
Last year I trained the advantage-conditioned VLA for π*0.6 during my internship @physical_int. Now, I have a much improved recipe.
Introducing ✨Advantage Guidance✨, a performant RL method with stable advantage-conditioned supervised training and contrastive guidance sampling
Berkeley AI Research retweeted
Follow Gemini 4 Argon's report, I implemented their robotics harness.
Gemini did a zero-shot bimanual handover. Now open source 🤖👇
Berkeley AI Research retweeted
❓How can humanoids learn by watching humans?
Naive retargeting can lose crucial contacts: like bracing a hand on a table while bending to pick an object off the floor.
🚀 Introducing VICAR: preserve key contacts from human videos, practice in simulation, deploy on hardware
1/n
