Xinyan Velocity Yu

on leave from #NLProc PhD @usc, bsms @uwcse | Previously @AIatMeta @Microsoft @Pinterest | Doing random walks on the west coast | Opinions are my own.

[Xinyan Velocity Yu @XinyanVYu]

#ARR and yes I am talking about ARR.

[Xinyan Velocity Yu @XinyanVYu]

What kind of review is that for a benchmark paper, they want to know if your one benchmark can wrap up the evaluation problem for the field, single-handedly cover every task type that exists in the space, and if it can't, that's treated as a flaw serious enough to reject over?😅

[Xinyan Velocity Yu @XinyanVYu]

@jieyuzhao11 🫡🥲

[Xinyan Velocity Yu @XinyanVYu]

Missed #ICML2026 and #ACL2026, but never missed a chance to grind ARR reviews 😎-> 🥲.

[Xinyan Velocity Yu @XinyanVYu]

@zouharvi 🫡

[Xinyan Velocity Yu @XinyanVYu]

@alisawuffles Alisa!! So excited for you and this next chapter. We’ll definitely miss you in Seattle 🥹!

[Zhaofeng Wu @zhaofeng_wu]

Introducing ><former

Most transformers are rectangles◻️: every layer has the same width

But is that optimal?🤔

We propose variable-width transformers that have different widths across layers, improving loss while cutting compute & KV cache size 🧵

[Harvey Yiyun Fu @harveyiyun]

Seeing👀 is not reasoning🤔?

Introducing our new blog post. We tested 8 open VLMs on 9 VQA benchmarks, and found that a lot of "visual" accuracy isn't visual at all. Many benchmarks can be solved without ever looking at the image, and some VLMs reason better from text than from pixels.
[1/n]

[Xinyan Velocity Yu @XinyanVYu]

@johntzwei Congrats Dr. Johnny Wei!!!!

[Deqing Fu @DeqingFu]

New paper: Convergent Evolution: How Different Language Models Learn Similar Number Representations.

Language models, classical word embeddings, and even raw token frequencies all develop the same Fourier features for numbers. But only some develop the underlying structure. 🧵

[Zhaofeng Wu @zhaofeng_wu]

Excited to share our new work from Meta MSL 🔥

LLMs write great Python/C++ but struggle with uncommon languages. Data scarcity is the bottleneck ⌛ Can we leverage cross-PL transfer to overcome this? Yes ✅

A new method to unlock cross-PL transfer 🧵 https://t.co/8EK05gMgCl

[Xinyan Velocity Yu @XinyanVYu]

@clara__meister I also broke it lately... It is telling me all its thinking process, and I had to terminate it because it just kept going. 🥲

[William Yijiang Li @Williamiumli]

🚨 Announcing our #CVPR2026 the 2nd Workshop on Multi-Modal Reasoning for Agentic Intelligence @CVPR @CVPRConf.

🚀 Call for Papers!!!
We accept regular papers limited to eight pages in the CVPR style.

Timeline:
📅 Submission open: February 1th 2026, 23:59 AoE Time
📣 Submission deadline: March 15th 2026, 23:59 AoE Time

Submit now: https://t.co/2LEDAV1Yzh

🌐 Learn more at our Website: https://t.co/aOPc7GXA9G

Kudos to our amazing co-organizers @9LdROhjZE56jSh9 , @zixianma02 , Anda Epure, @kamath_amita , @MahtabBg , @alexttoshev , @RanjayKrishna , @lucy_x_shi , @_anniechen_ , @philiptorr

[Xinyan Velocity Yu @XinyanVYu]

I might be the last one to the show#NeurIPS2025 Finally almost done with TA duties, heading to San Diego right now. DM is open! Happy to chat about multilingual NLP/retrieval/reasoning/multimodal/any random stuff, or to grab boba and coffee! 🫡

[Zhaofeng Wu @zhaofeng_wu]

I’ll be at #neurips2025! I’m currently on the industry job market for LLM roles. DM/email me if you want to chat!

[Xinyan Velocity Yu @XinyanVYu]

@jifan_zhang @BhattGantavya @rdnowak @SharonYixuanLi @LihongLi20 Congrats Jifan!!

[Zhaofeng Wu @zhaofeng_wu]

I'll be in Vienna at #acl2025! Please lmk if you'd like to chat 🙂

[Xinyan Velocity Yu @XinyanVYu]

@AkariAsai @uwcse @LTIatCMU @mldcmu Congratulations Akari!! 🎉🥳👏🎊🙌🍾🥂

[Kaiser Sun @KaiserWhoLearns]

What happens when an LLM is asked to use information that contradicts its knowledge? We explore knowledge conflict in a new preprint📑
TLDR: Performance drops, and this could affect the overall performance of LLMs in model-based evaluation.📑🧵⬇️ 1/8
#NLProc #LLM #AIResearch

[Xinyan Velocity Yu @XinyanVYu]

@danikhan632 @huggingface it is down for me too ;(