Check out our new survey paper on data selection for language models. Fantastic work led by Alon Albalak .

🔥Introducing our new #cvpr paper on improving the quality of diffusion generations via efficient particle samplings. This is an amazing work led by Yujian in collaboration with MIT and the MIT-IBM Watson AI Lab.

🛡️New Jailbreak Defenses for LLMs: By harnessing semantic-preserving transformations with randomized smoothing, we have enabled LLMs to defend jailbreaks with minimal impact on their performance for benign tasks. An amazing collaboration between students at UCSB and UPenn.

Join us for our oral presentation on decomposing uncertainty of LLMs!

WHP a.k.a. Who's Harry Potter?⚡️ is a ML method to unlearn knowledge or biases from training data. Yujian Liu, Yang Zhang, #JameelClinic PI Tommi Jaakkola, and Shiyu Chang propose a new way to extend WHP w/ targeted unlearning arxiv.org/pdf/2407.16997

Amazing work by Yujian Liu

Fantastic work by my student Bairu Hou and amazing collaborators at Apple! 🌟 We introduced Instruction-Following Pruning – a dynamic method for pruning LLMs based on prompts, achieving remarkable efficiency and performance.

We reproduced DeepSeek R1-Zero in the CountDown game, and it just works Through RL, the 3B base LM develops self-verification and search abilities all on its own You can experience the Ahah moment yourself for < $30 Code: github.com/Jiayi-Pan/Tiny… Here's what we learned 🧵

In RAG applications, LLMs often reprocess the same database chunks for different queries—leading to high latency and cost from handling massive input tokens. We introduce KVLink, an efficient approach to reuse pre-computed KV caches of retrieved documents, drastically cutting

Our recent work on pruning long chain-of-thought reasoning—simple, effective, and surprisingly powerful. 🧠✂️ Led by my student Bairu Hou , with amazing collaborators!

Elegant algorithm to embed watermarks against piggyback spoofing attacks! 🚀 Led by my amazing student Li An and our great collaborators.

Sad to miss #ICLR2025 this year, but thrilled to see our work led by my student Yujian Liu presented there! Stop by our poster and chat with my amazing collaborators about Prereq-Tune, our new method to improve LLM factuality. 🚀📍#273 | Today (Apr 24), 3–5:30 PM

Sadly missing #NAACL this time, but my student Bairu Hou will present our joint work with MIT NLP and MIT-IBM Watson AI Lab on hallucination detection using a novel probabilistic propagation approach — one of my favorite recent projects from our team.

Our new knowledge tracing method accurately tracks young student understanding — even with just a few student responses. Led by my amazing students Xinyi Gao Qiucheng Wu, with fantastic collaborators! 🌱🧠

On Monday, a United States District Court ruled that training LLMs on copyrighted books constitutes fair use. A number of authors had filed suit against Anthropic for training its models on their books without permission. Just as we allow people to read books and learn from them

Scaling up RL is all the rage right now, I had a chat with a friend about it yesterday. I'm fairly certain RL will continue to yield more intermediate gains, but I also don't expect it to be the full story. RL is basically "hey this happened to go well (/poorly), let me slightly

Sad to miss #ICML2025 this year, but thrilled that my student Bairu Hou will present his exciting work on dynamically pruning LLMs into efficient, task-specific models—done in collaboration with our amazing collaborators from Apple! 🍎✨

Slides for my lecture “LLM Reasoning” at Stanford CS 25: dennyzhou.github.io/LLM-Reasoning-… Key points: 1. Reasoning in LLMs simply means generating a sequence of intermediate tokens before producing the final answer. Whether this resembles human reasoning is irrelevant. The crucial

🚨 SHOCKING: people are unknowingly making their ChatGPT interactions PUBLIC, and they are being indexed by Google (see my test below). My privacy recommendations: When people interact with ChatGPT and use the "Share" feature (for example, to send the conversation to family and

UCSB NLP @ EMNLP 2025 EMNLP 2025! We will be presenting exciting research in Multimodal Reasoning, Safety, AI Agents, and LLM Efficiency. Come meet us in Suzhou this November. Would love to exchange ideas and discuss where the field is headed!🚀 🎉 Huge congrats to our