Log inSign up
Zihan "Zenus" Wang profile banner
user avatar

Zihan "Zenus" Wang

@wzenus
Reasoning agent / RL / efficiency research @NorthwesternU & @nvidia. Ex @Microsoft @yutori_ai @deepseek_ai @uiuc_nlp @RUC1937.
zenus.me
Joined March 2022
690
Following
23K
Followers
RepliesArticlesMedia
  • Pinned
    user avatar
    Zihan "Zenus" Wang
    @wzenus
    May 29
    🧵 Claude-Opus-4.8 takes you too much tokens - but is this issue general across agents? Do agents know how much they'll spend? Introducing Budget-Aware Agents (BAGEN): We study budget awareness across 4 envs & 5 frontier agents, and find structured failures in most of them.
    Image
  • user avatar
    Zihan "Zenus" Wang
    @wzenus
    Aug 3
    随手写了个让 Claude Code 说人话的 instruction，效果对比如下 (右图为改进后的版本)。 认真手写完才意识到手写能力还是很重要的。毕竟 Claude Code 再怎么写，也写不出一份没有 "CC 味" 的 instruction 🙃
    Image
    Image
  • user avatar
    Zihan "Zenus" Wang
    @wzenus
    Jul 14
    This work is very amazing! Some more thoughts: human sleep to organize memory to condense experience to knowledge and learn world models on how things proceed, so as LLMs do. Also: why do human sleep 8 hours every day, is there such a sleeping compute law in LLMs?
    user avatar
    Ali Behrouz
    @behrouz_ali
    Jul 13
    Moving from conventional ML to continual learning requires revisiting even the fundamental concepts such as “test”/“train” time. LLMs Need Sleep and Dreaming! We introduce a phase, where the model consolidates its fragile short-term memories into stable long-term memories, and
    Image
  • user avatar
    Zihan "Zenus" Wang
    @wzenus
    Jul 9
    Today is the day! 🎯 Failure Modes in Agentic AI (FAGEN) Workshop @ ICML 2026. Nine speakers, 218 submissions, one shared question: what does it mean for an agent to fail, and how do we fix it? Find it. Reproduce it. Trace it. Fix it. 📍 COEX Ballroom 104–105 · all day. See you
    Image