1. X
  2. Binfeng Xu
Log inSign up
Binfeng Xu
656 posts
Image
user avatar
Binfeng Xu
@billxbf
AI @nvidia | Solving agentic coding, computer-use and continual learning with RL. Opinions are mine.
NY
billxbf.github.io
Joined May 2022
226
Following
1,117
Followers
RepliesRepliesMediaMedia

New to X?

Sign up now to get your own personalized timeline!

Create account

By signing up, you agree to the Terms of Service and Privacy Policy, including Cookie Use.

Terms·Privacy·Cookies·Accessibility·Ads Info·© 2026 X Corp.
Don't miss what's happening
People on X are the first to know.
Log inSign up
  • Pinned
    user avatar
    Binfeng Xu
    @billxbf
    May 26
    Excited to release 🌟Polar🌟, our Agent RL rollout infra for real-world harnesses. Be it Codex, Claude Code, OpenClaw, Hermes, or your self-made ones 🔥 -- Polar takes your harnesses directly as training environments without code change. Find a problem, design the harness, and
    Image
    133K
  • user avatar
    Binfeng Xu
    @billxbf
    May 31, 2023
    Bothered by the expensive runs on Auto-GPT and LangChain agents? Check out our recent work, ReWOO, that eliminates token redundancy in prevailing 'thought-action-observation' paradigm, achieving better task completion with 5x less token usage at inference. #LLMs #AIGA #NLP
    Image
    37K
  • user avatar
    Binfeng Xu
    @billxbf
    Oct 23, 2025
    Replying to @tydsh
    Crazy. Welcome to nvidia btw
    30K
  • user avatar
    Binfeng Xu
    @billxbf
    Apr 8, 2024
    This competition is so intriguing in any sense that I can’t resist back to Kaggle. kaggle.com/competitions/a…
    14K
  • user avatar
    Binfeng Xu
    @billxbf
    May 31, 2023
    Replying to @billxbf
    (1/5) Paper Link: arxiv.org/abs/2305.18323 Code Link: github.com/billxbf/ReWOO
    1.1K
  • user avatar
    Binfeng Xu
    @billxbf
    Aug 13, 2023
    Imagine tuning autonomous agents with single configs, sharing, and interacting with other agents😆. We’ve officially released Gentopia, a lightweight and extensible framework to build LLM-driven Agents and tool-augmented LM research. Check out this🧵below.
    Image
    GitHub - Gentopia-AI/Gentopia: Build Hierarchical Autonomous Agents through Config. Collaborative...
    From github.com
    4.2K
  • user avatar
    Binfeng Xu
    @billxbf
    May 31, 2023
    Replying to @billxbf
    (3/5) Comparing to ReAct widely adopted in LangChain and Auto-GPT, ReWOO quadratically improves system efficiency by eliminating the 'stacking' prompt history (or so called long-term memory management).
    Image
    924
  • user avatar
    Binfeng Xu
    @billxbf
    Jul 9, 2023
    Replying to @llennchan2003
    chatbot是这样的 你得先prompt
    212
  • user avatar
    Binfeng Xu
    @billxbf
    May 31, 2023
    Replying to @billxbf
    (2/5) Given a question, a Planner composes a blueprint of interlinked plans prior to tool response. The blueprint instructs Worker to call tools and collect evidence. Finally, plans and evidence are paired and fed to Solver. Only two calls to LLMs are necessary.
    Image
    1.4K
  • user avatar
    Binfeng Xu
    @billxbf
    May 31, 2023
    Replying to @billxbf
    (5/5) Does lack of explicit observation compromises LLMs' reasoning ability? 🤔Not necessarily. In fact, we see our system performing similar or even better than its counterpart on 6 hot public benchmarks and a curated dataset.
    Image
    622
  • user avatar
    Binfeng Xu
    @billxbf
    Sep 30, 2025
    “… physical simulations effectively model many systems, but more broadly, physics is a verifiable environment. AI has progressed fastest in domains with data and verifiable results - for example, in math and code. Here, nature is the RL environment.” Road converges
    user avatar
    Liam Fedus
    Periodic Labs
    @LiamFedus
    Sep 30, 2025
    Today, @ekindogus and I are excited to introduce @periodiclabs. Our goal is to create an AI scientist. Science works by conjecturing how the world might be, running experiments, and learning from the results. Intelligence is necessary, but not sufficient. New knowledge is
    Image
    602
  • user avatar
    Binfeng Xu
    @billxbf
    May 31, 2023
    Replying to @billxbf
    Check our real-time Gradio demo ! ! huggingface.co/spaces/rewoo/R… :0
    Image
    ReWOO Demo - a Hugging Face Space by rewoo
    From huggingface.co
    637
  • user avatar
    Binfeng Xu
    @billxbf
    May 31, 2023
    Replying to @billxbf
    Conclusion: Planning is All You Need!🤣Formally -- pretrained foundation models have enough knowledge of the 'shape' of most tool responses to think ahead. On the contrary, lengthy prompts (possibly with wrong observations ☠️) from ReAct negatively influence reasoning trace.
    534
  • user avatar
    Binfeng Xu
    @billxbf
    May 31, 2023
    Replying to @billxbf
    Big shout out to @fuckinlostguest @BowenLei03 @subho_mpi @DongkuanXu !
    540
Advertisement
Advertisement