Excited to share that our Multi-Object Hallucination paper will be presented at #NeurIPS2024! Looking forward to seeing you in Vancouver!

🚀 Excited to share our latest paper: “Learning 3D Representations from Procedural 3D Programs” We explore self-supervised learning of 3D representations using procedurally generated shapes, with no reliance on human-designed 3D datasets. We found that Self-supervised 3D

(1/10) Excited to share one of the most elegant works I’ve been working on: Parallelizing Linear Transformers with the Delta Rule over Sequence Length! 🎉 📄 Published at NeurIPS ‘24 📍 Catch my poster in person: NeurIPS East Exhibit Hall A-C #2009 🗓️ Fri, Dec 13 | 4:30–7:30 p.m

I am at Vancouver this week🇨🇦! I’ll present our work at #NeurIPS2024 this Friday, Dec 13 4:30-7:30 pm at East Exhibition Hall A-C #4303. Feel free to drop by and chat if you are interested. I’m also actively looking for PhD opportunities for Fall 2025. Would love to connect with

10 short videos about LLM infrastructure to help you appreciate Pages 12-18 of the DeepSeek-v3 paper (arxiv.org/abs/2412.19437) 🧵 youtube.com/watch?v=76gulN…

📢 Join us for the ACL Mentorship Session on Zoom! Session Link: mentorship.aclweb.org/scheduleAsk Questions: app.sli.do/event/7kyB9E9h… Mentors: • Furong Huang (UMD Department of Computer Science) • Xiang Yue (Language Technologies Institute | @CarnegieMellon) • Songlin Yang (MIT CSAIL) • Martin Ziqiao Ma (Computer Science and Engineering at Michigan) • Oana Ignat 👩‍💻🎓📚🇷🇴🌍 (Santa Clara Univ)

🚀 Announcing ASAP: asap-seminar.github.io! A fully virtual seminar bridging theory, algorithms, and systems to tackle fundamental challenges in Transformers. Co-organized by Simran Arora Xinyu Yang Han Guo Our first speaker: Alex Wang on Test-time Regression

We release Magma, the first VLM that explores the vital interplay between multimodal understanding and actions across both virtual and physical worlds—not only controlling your laptop but also directing your robot! 🥳 Magma bridges verbal and spatial-temporal intelligence.

I convinced Songlin (Songlin Yang) of Flash Linear Attention fame to give me a personal tutorial on the current state of attention. Video to come, (it's like 2hr!)

Excited about the tutorial on "The Lifecycle of Knowledge in LLMs: Memorization, Editing, and Beyond" with Zoey Sha Li Yuji Zhang Chi Han Heng Ji . Slides/Video(upcoming): llmknowledgelifecycle.github.io/AAAI2025_Tutor… Time: Feb 26 8:30-12:30 Location: Room 116 Zoom: underline.io/events/487/ses…

Today is the day! Welcome to our 2nd workshop on Knowledgeable Foundation Models in Room 112. Come and talk with these wonderful speakers Eduard Hovy Wenpeng_Yin Eric Wong Lianhui Qin Li "Harry" Zhang Huajie Shao@WM ! Special thanks to our organizers Zoey Sha Li Mor Geva Xiaozhi Wang

⚡️ Excited to announce Fast3R: 3D reconstruction of 1000+ images in a single forward pass! Fast3R achieves 251 FPS at its peak. 🔥 Try the demo with your images or video! 🔗 Website: fast3r-3d.github.io 🎮 Demo: fast3r.ngrok.app #CVPR2025 #3D AI at Meta

📜New Paper: on the surprising effectiveness of multilingual chain of thought (originally discovered w/ Mirac Suzgun and Google DeepMind folks in ICLR23). Linguistic diversity in prompts helps solve problems in low-resource languages, even if there's no alphabetical overlap!

Chi Han Chi Han ‘s recent deep analysis on the hidden computational mechanisms behind LLM position generalization shows: Attention logits ≈ positional pattern + semantic importance (0.959 correlation!) arxiv.org/abs/2503.13305

Vision-Language Models (VLMs) can describe the environment, but can they refer within it? Our findings reveal a critical gap: VLMs fall short of pragmatic optimality. We identify 3 key failures of pragmatic competence in referring expression generation with VLMs: (1) cannot

We are extremely excited to announce mCLM, a Modular Chemical Language Model that is friendly to automatable block-based chemistry and mimics bilingual speakers by “code-switching” between functional molecular modules and natural language descriptions of the functions. 1/2

Does RL truly expand a model’s reasoning🧠capabilities? Contrary to recent claims, the answer is yes—if you push RL training long enough! Introducing ProRL 😎, a novel training recipe that scales RL to >2k steps, empowering the world’s leading 1.5B reasoning model💥and offering

SAB3R: Semantic-Augmented Backbone in 3D Reconstruction Xuweiyi Chen, Tian XIA, Si.X, Jianing “Jed” Yang @ CVPR, Joyce Chai, Zezhou Cheng tl;dr: MASt3R+distillation->open-vocabulary segmentation+3D reconstruction arxiv.org/abs/2506.02112

(1/5) Super excited to release our new paper on Reinforcement Learning: "Self-Aligned Reward: Towards Effective and Efficient Reasoners"! Preprint: arxiv.org/pdf/2509.05489

🚀 Introducing UserRL: a new framework to train agents that truly assist users through proactive interaction, not just chase static benchmarking scores. 📄 Paper: arxiv.org/pdf/2509.19736 💻 Code: github.com/SalesforceAIRe…