Mingyang Zhou

🧵 Decomposing the Delta: What Do Models Actually Learn from Preference Pairs? 1/n 💡 Why do methods like DPO and KTO actually improve reasoning? In standard alignment, we use preference pairs, but we don't fully understand what properties of the data drive downstream gains. We investigate two distinct notions of quality: Generator-level Delta - the capability gap between the models producing the chosen vs. rejected traces Sample-level Delta - The fine-grained quality difference within a single pair (Factuality, Coherence, Precision) 👇

@HaoyiQiu @steeve__huang @UCLA good job!

No costume for discussion session this week, but I did my best to bring the spirit anyway — here’s Labubu in a UCLA + Halloween edition! 👻💙🎃 #UCLA #Halloween #LABUBU

@shi_weiyan Come and visit New York sis

Yesterday I asked what makes you happy. 226 voted--61% said family & friends, like those AI researchers at the panel. Not AI, just the people around us. This resonates with Harvard's 87-year study on happiness. Their conclusion: happiness is not about money/fame/success, it's all about good relationships. I wasn't convinced when first reading it 10 years ago. But over this decade, I've watched this truth play out everywhere--through my own experiences, others' lives. And now, through the AI panel discussion and your poll responses. In this crazy AI time, we're all racing toward some undefined finish line. But maybe the real edge isn't working harder--it's knowing when to close the laptop and have dinner with someone who matters. Maybe that's the answer to AI burnout: remember why we started building AI in the first place--to make life better for the people we care about 💙

@Sylvia_Sparkle @Zhou_Yu_AI Big Con Siyan!

PAPILLON 🦋 has been accepted into NAACL Main Conference!! 🥳🥳🥳✨✨✨ Check out the original PAPILLON thread below to learn more about our approach for privacy preservation during inference-time interactions with LLMs!

@gentaiscool @AyuP_AI Good job man! I wish your name will be the answer for this question soon🤪

Testing Grok2 and I am proud that I was ⁦@AyuP_AI⁩’s student and collaborator 😀 #Nlproc

@xwang_lk Well said and huge respect to these words from a brave junior faculty

Many asked me: "Why take a leave to do startup before tenure?" "Why speak up and challenge senior voices?" "Have you got tenure yet?" My response: Tenure is nothing if it hinders chasing dreams and staying true to my values. I believe tenure is just a byproduct of doing meaningful work, not the end goal of a faculty career.

Ironic to see the racism claim is made on a speech talking about ethics. Good job Professor, I also hope you are an exception.

1/3 Today, an anecdote shared by an invited speaker at #NeurIPS2024 left many Chinese scholars, myself included, feeling uncomfortable. As a community, I believe we should take a moment to reflect on why such remarks in public discourse can be offensive and harmful.

@robinson_n8 congrats!🎉

I hit my 💯 citation milestone today! Huge thank you to everyone who has engaged with my work, there is much more to come 🎉

@billyuchenlin @xai congrats Yuchen

Join @xAI ! 🚀

⭐️ We're going to launch Grassroots Science, a year-long ambitious, massive-scale, fully open-source initiative aimed at developing multilingual LLMs aligned to diverse and inclusive human preferences in Feb 2025. 🌐 Check our website: grassroots.science and be sure to complete the interest form if you'd like to participate. ☄️ This global collaboration was initiated and organized by members of SEACrowd @sealpworkshop, @MasakhaneNLP, @ai4bharat, @RojakNLP, and past collaborators of @BigscienceW, etc. Our mission is to establish an initiative where grassroots can work together in high-profile projects. We are excited to see what we can achieve together! #NLProc #GrassrootsScience Thank you to early initiators who efficiently helped kick off the project! @gentaiscool @yong_zhengxin @ruochenz_ @ljvmiranda @AlhamFikri, including our grassroots community partners @davlanade, @HolyLovenia and @prajdabre1 and all future collaborators!

@xwang_lk @ucsbcs Big congrats!

Life update: Excited to announce that I’ll be joining UCSB CS (@ucsbcs) as an Assistant Professor in Fall 2025—where my research dream began! If you’re interested in PhD research (3+ openings) on multimodal, GenAI, or agents (embodied/digital), apply to UCSB CS by Dec 15! 1/n

On my way to attend EMNLP2024 So exciting to meet with old and new friends~ Come talk to me if you are interested in the full-time and intern opportunity at Capital One LLM Pretraining team. We are hiring! #emnlp #emnlp2024 #CapitalOne

@YiFung10 @steeve__huang @kenchanhp @zhenhailongW @hengjinlp @uiuc_nlp @columbianlp Thanks Yi!

Tonight, I'm flying out to attend #ACL2024NLP and present three posters! [1] Come to Poster Session 2 on August 12 at 5:45PM to discuss our paper 💥Do LVLM Understand Charts? Analyzing and Correcting Factual Errors in Chart Captioning 💥 See you in Bangkok 🇹🇭! 🤗

We analyzed the hallucinations produced by LVLMs on chart captioning tasks and proposed a factual error correction model. Kudos to @steeve__huang and our fantastic collaborators!

Kudos to Steeve for driving hiss amazing work! Please check if out if you want to learn the types of hallucination LVLMs will make when they encounter Chart images!

🥳 Excited to share that our work on "Do LVLMs Understand Charts? Analyzing and Correcting Factual Errors in Chart Captioning" has been accepted as a Findings paper at #ACL2024🇹🇭 ! Many thanks to @steeve__huang and our amazing collaborators~ Check out 📣arxiv.org/pdf/2312.10160

@yong_jae_lee @imhaotian @xueyanzou1 Congrats Prof. Yong Jae and all the friends from UC Davis~

Incredibly proud of my PhD students Haotian Liu @imhaotian, Xueyan Zou @xueyanzou1, Yuheng Li, Utkarsh Ojha who graduated this weekend! This special group moved with me from UC Davis to Wisconsin 3 years ago, helped set up my lab & did amazing research. I'll really miss them😭 1/

@steeve__huang @hengjinlp @haopeng_nlp @hanzhao_ml @JotyShafiq @uiuc_nlp @IllinoisCS @uofigrainger Shout out to Dr Huang! What a journey and good luck in your new chapter of life!

Officially Dr. Huang! 🎓 Thrilled to share that I've successfully defended my PhD thesis. Immensely grateful for my inspiring advisor @hengjinlp, and the guidance of my thesis committee: ChengXiang Zhai, Kathleen McKeown, @haopeng_nlp, @hanzhao_ml, and @JotyShafiq #PhDDone

@shi_weiyan @simon_ycl @jcz12856876 @JiayiZhang0427 Congrats Weiyan! And best luck with the new journey in Boston! Can’t wait to see the upcoming exciting research work from CHATS lab!

💥Hello world! The CHATS lab just got 3x smarter! 💥 Say Hi to my first-ever batch of PhD students🥳🥳🥳 @simon_ycl @jcz12856876 @JiayiZhang0427 – these brave souls signed up for endless research🧐, questionable coffee☕️, nice and cold days in Boston🌇