Bingxuan Li

How to produce sounding videos that align with your creative story?🎬📖➡️🔊 Introducing EchoFoley (accepted by #CVPR2026 main conf🎉)! In creative storytelling, sound shapes the story we perceive. Given a silent video, generating audio that matches the story-shaped imagination requires fine-grained control over how each sound is crafted and transformed over time. We formulate such task with event-centric hierarchical control, and propose an effective solution to bring your creative imagination to life. 🪄 Check our paper, code, and data for more details: echofoley.github.io

Excited to share that "Energy-Based Transformers are Scalable Learners and Thinkers" was accepted to #ICLR2026 as an oral! 🎉 I'll be giving the oral this Friday in Brazil, so come watch if you're around :)

Huge thanks to awesome collaborators and mentor @MasterJeongK @qiancheng1231 @xiusi_chen , Eitan Anzenberg, Niran Kundapur, @hengjinlp 🫶

🧵3/ We propose PEARL, a self-evolving agent for time management: - a persistent preference memory (Strategy Hub) - RL recipe that learns from historical decisions over time. PEARL reduces errors by 55% 🚀Agents that do not just reason, they learn you!

Can we trust LLMs to manage our time? ⏰ Our recent work (accepted by #ACL2026 main conf🎉) gives the answer! Our CalendarConflictBench shows that even strong models make up to 35% errors when resolving calendar conflicts on users’ behalf.🥲But no worries! Meet PEARL: We design a RL recipe to train a self-evolving agent that learn and adapt to user preferences over long-horizon for preference-driven decision-making, reducing errors by 55%🚀 Paper: arxiv.org/pdf/2601.11957

Huge thanks to my awesome collaborators and mentors : Yiming Cui, Yicheng He, Shu Zhang Longyin Wen, @wangyiw33973985 , @Yulei_Niu

How to produce sounding videos that align with your creative story?🎬📖➡️🔊 Introducing EchoFoley (accepted by #CVPR2026 main conf🎉)! In creative storytelling, sound shapes the story we perceive. Given a silent video, generating audio that matches the story-shaped imagination requires fine-grained control over how each sound is crafted and transformed over time. We formulate such task with event-centric hierarchical control, and propose an effective solution to bring your creative imagination to life. 🪄 Check our paper, code, and data for more details: echofoley.github.io

🚀Accepted to ACL 2026 Main! Please check it out if you are also interested in the intersection of agent and world model. Happy to chat in San Diego!

Excited to share our paper🥳: Agent Q-Mix, a collaborative multi-agent rl framework for adaptive communication in LLM multi-agent systems.🎉🎉🎉 Agent Q-Mix outperforms Lobster (OpenClaw), LangGraph, Microsoft AutoGen, while using fewer tokens!🚀 Paper: arxiv.org/pdf/2604.00344

Can AI move beyond chatting to actually making biological discoveries? What if AI could think like a microbiologist? 🧪 📣 Excited to introduce Eubiota: a modular agentic AI framework for autonomous discovery in the gut microbiome. 🧬🤖🦠 👉Explore: eubiota.ai The gut microbiome is complex; manual discovery is slow. 🐌 We built Eubiota to think like a researcher: planning experiments, verifying data, and synthesizing results. It’s an open-source system designed to empower the entire community. 🌐 Eubiota demonstrated its capability across end-to-end discovery tasks, with results validated in the lab: ✅ Identified uvr-ruv DNA repair axis for fitness under inflammatory stress by screening ~2,000 genes🧬 ✅ Designed a microbial consortium therapy to reduce inflammation in mice colitis models.🐁 ✅ Uncovered diet-associated anti-inflammatory metabolites & engineered antibiotic cocktails.💊 Eubiota is the "scientific copilot" the field has been waiting for. Let AI be your partner to accelerate discovery! 🚀 Incredible teamwork with @lupantech and all our co-authors! I'm deeply grateful for the guidance from amazing advisors @james_y_zou, @LabSonnenburg, and brilliant mentors KC Huang, @YejinChoinka. Read more here: 🔗biorxiv.org/content/10.648… Code: github.com/lupantech/Eubi… Try out our system in your browser! 🤖app.eubiota.ai/chat #Eubiota #AI4Science #Gut #Microbiome #AgenticAI

Introducing Eubiota: A multi-agent AI framework for autonomous discovery in the human microbiome. 🧬🤖🧫 👇 Explore the platform: eubiota.ai Eubiota doesn’t just chat: it plans, uses tools, verifies evidence, and drives end-to-end discovery—from hypothesis to wet-lab validation. Eubiota achieved 87.7% accuracy on mechanistic reasoning (vs. GPT-5.1 77.3%). But we went further. We used Eubiota to drive 4 discoveries with experimental validation: ✅ Gene Discovery: Identified the uvr-ruv stress axis by screening 1,945 genes & 10K papers in hours (on 2 GPUs) 🧬⚡ ✅ Therapeutics: Designed a microbial therapy that reduced colitis inflammation 🦠💊 ✅ Antibiotics: Engineered a cocktail that kills pathogens but spares commensals 🎯🛡️ ✅ Metabolites: Discovered novel anti-inflammatory molecules from large human data 🥗🧪 📄 Paper: biorxiv.org/content/10.648… 💻 Code: github.com/lupantech/Eubi… Try the live app to start your own discovery: 🎮 App: app.eubiota.ai Huge thanks to fantastic co-lead @YifanGao15, our incredible PIs @james_y_zou @LabSonnenburg, stellar advisors Kerwyn Casey Huang, @YejinChoinka, and the excellent team! 👏 #Eubiota #Microbiome #Inflammation #AI4Sci #AgenticAI #Agent

𝗧𝗵𝗶𝗻𝗸 𝘄𝗶𝗱𝗲𝗿. 𝗧𝗵𝗶𝗻𝗸 𝘀𝗵𝗼𝗿𝘁𝗲𝗿. 🚀 𝗜𝗻𝘁𝗿𝗼𝗱𝘂𝗰𝗶𝗻𝗴 𝗠𝘂𝗹𝘁𝗶𝗽𝗹𝗲𝘅 𝗧𝗵𝗶𝗻𝗸𝗶𝗻𝗴: token-wise branch-and-merge reasoning for LLMs. 💸 Discrete CoT is costly. 🎛️ Existing continuous tokens often clash with 𝗼𝗻-𝗽𝗼𝗹𝗶𝗰𝘆 𝗥𝗟 𝗲𝘅𝗽𝗹𝗼𝗿𝗮𝘁𝗶𝗼𝗻. 🎥 𝗠𝘂𝗹𝘁𝗶𝗽𝗹𝗲𝘅 𝗧𝗵𝗶𝗻𝗸𝗶𝗻𝗴, a sampling-based continuous reasoning paradigm:

🚨We introduce Multiplex Thinking: Humans think in high-bandwidth “soft” internal representations, while LLMs must commit token-by-token. Prior work adds continuous thought tokens to widen the channel, but often sacrifices the stochasticity that RL-style training needs. Multiplex Thinking keeps both: - hard-sample multiple tokens per step (stochastic, RL-friendly) - soft-merge them into a continuous state (higher bandwidth) Bonus: sampling run in parallel ⇒ near-zero overhead + shorter reasoning. We’re releasing everything: paper+model+code. This project is led by my talented first-year PhD student @tyao923 at Penn (@CIS_Penn @PennEngineers) with collaborators at MSR (@donglixp @yaruhao @qx_dong @furu_wei) Paper: arxiv.org/abs/2601.08808 Blog: gmlr-penn.github.io/Multiplex-Thin… Code: github.com/GMLR-Penn/Mult… Model: huggingface.co/Multiplex-Thin…

📣 CFP: ICLR 2026 Workshop — Lifelong Agent: Learning • Aligning • Evolving (Apr 26/27, Rio de Janeiro | full-day). We’re building a forum for agents that improve over time!

🔮 Can a world model (simulator) give today’s AI agents foresight? We tested “world model as a tool”… and found it often doesn’t help—sometimes it hurts. Check our newest paper here: arxiv.org/pdf/2601.03905… #AIagents #WorldModel #ToolUse

Introducing Large Video Planner (LVP-14B) — a robot foundation model that actually generalizes. LVP is built on video gen, not VLA. As my final work at @MIT, LVP has all its eval tasks proposed by third parties as a maximum stress test, but it excels!🤗 boyuan.space/large-video-pl…

Have you ever done a dense grid search over neural network hyperparameters? Like a *really dense* grid search? It looks like this (!!). Blueish colors correspond to hyperparameters for which training converges, redish colors to hyperparameters for which training diverges.

Want to use LLMs to build your websites? They may not be as good as shown by prior benchmarks! Introducing🔥FronTalk🔥that evaluates multi-turn multi-modal front-end coding - Users propose new evolving requests in multiple turns, including visual instructions🖼️in the form of sketches and annotations. 🤔Can LLMs/VLMs still faithfully follow instructions? Check our project! frontalk-benchmark.github.io