Yuxuan LU

🚀 Exciting to share CHI'25 LBW paper UXAgent: An LLM-Agent-Based Usability Testing Framework for Web Design 🖥️ 📜 ArXiv: arxiv.org/abs/2502.12561 Collaboration with @BingshengY @JIRIGESI @TobyJLi @dakuowang

@xwang_lk 左眼跳财 要发财了 右眼跳灾 封建迷信

Rejections are normal in academia, even for top researchers. To protect your mental health, practice 精神胜利法 ("spiritual victory method") with a deliberate double standard: ✅ Accepted: celebrate. you deserve it! ❌ Rejected: reviews are noisy; just bad luck. Try again.

Claude Opus 4.6 struggles to spell 'Assertion', especially when asked in Chinese:

🦆 I turned 32 Intel N100 mini PCs into a DeepSeek-V3.2 685B inference cluster. No NVLink. No InfiniBand. Just 1 RTX 5090, 32 old online-judge nodes, and Ethernet. The trick: 🧠 keep Attention on the GPU 🦆 run MoE on the N100 cluster ⚡ design fp9 format so AVX2-only tiny cores can do MoE GEMV Result: 11 input tokens, 2037 output tokens 16.171 tok/s decode ~89% MTP1 acceptance Technical report, currently in Chinese: judgeduck.ai/duck-llm/ *Personal hobby project, NOT related to any organization* ---------- 🦆 我把 32 台 Intel N100 小主机改造成了一个 DeepSeek-V3.2 685B 推理集群。 没有 NVLink。 没有 InfiniBand。 只有 1 张 RTX 5090、32 台原本用于在线评测的 N100、和普通以太网。 关键想法： 🧠 Attention 留在 GPU 🦆 MoE 层交给 N100 集群 ⚡ 自定义 fp9 数据格式，让 AVX2 小核也能跑 MoE GEMV 最终： 11 tokens 输入，2037 tokens 输出 16.171 tok/s decode MTP1 接受率约 89% 技术报告： judgeduck.ai/duck-llm/ *个人娱乐项目，与任何组织均无关联*

As Khoury College's faculty and student researchers descend on Barcelona this week, they're asking how technology — particularly AI — can fulfill users' needs, enhance their abilities, and amplify their humanity. Read more: bit.ly/4miAO6t

We have 4 main conference paper published at @aclmeeting — thanks all collaborators! See you in San Diego!

How far would GPT suggest to drive to the car wash instead of WALKING there? Here is the plot:

cannot use tool

oops

@AlhamFikri But isn’t OI wiki in Chinese?

Getting ready for ICPC. Apparently, they don't have the 25-page limit for the cheatsheet anymore, as long as it's printed. So I asked my students to send their cheatsheet My student proceeds to send a 3000-page cheatsheet Bruh😂

@AlhamFikri Exactly I used to have half a suitcase of printed material lol

Northeastern University@NeurIPS 2025, in sunny San Diego ❤️🤩☀️

I'll be presenting our demo at Exhibition hall B, Amazon booth tomorrow 11:30am-12:00pm!

求和！求和！理解万岁！

news.northeastern.edu/2025/08/05/ai-… One more deliverable from our fruitful collaboration with Prof. @YYingXu. Thanks @JiajuChen13 and @leoleoasd for leading the hard work!

#EMNLP2025 AC is upset I didn’t disclose the shopping platform we collected our data from (which is our affiliation). Do they not understand how double-blind review works?

@iavins I once put string “false” into a Boolean column that was casted to true

sorry not sorry but you gotta know this cursed SQLite fact too

@windx0303 50% of the work being "useful" is good enough to me 🤣

This basically applies to every conference's Appendix and Rebuttal situation now: In order to get one paper in, you write two papers.

So nowadays people even use LLM to respond to rebuttals...

Tenured at @Northeastern ! 🎉 Endless thanks to my mentors, collaborators, students, colleagues, and friends, couldn’t have done it without you. ❤️ Excited for what’s ahead. Collaboration Matters.

@JentseHuang try biorxiv

After 20 days of deep-thinking (on-hold), arXiv thinks that I should transfer to biology