Yao Qin
@YaoQin_UCSB
Followers
4K
Following
969
Media
6
Statuses
111
Assistant Professor @UCSB; Research Scientist @Google DeepMind; PhD @UCSD
Santa Barbara, CA
Joined November 2015
Andong (@andong_1997) and Kenan (@KenanTang) did an awesome job digging into the prompt sensitivity of LLMs — turns out it’s mostly about evaluation artifacts, not a fundamental flaw in the models! 🚀👏
Are LLMs really so prompt-sensitive? 🤔 🚨 Thrilled to share our EMNLP 2025 main conference paper! Prompt sensitivity has long been seen as a core weakness of LLMs—where tiny wording changes flip benchmark results. Our study finds: much of this effect stems from evaluation
1
0
20
Come to join us today at our EMNLP poster “Flaw or Artifact? Rethinking Prompt Sensitivity in Evaluating LLMs”, presented at Session 4, Wed, November 5, 14:30–16:00! I will be in person here and share more research details on this line of project 😃
Are LLMs really so prompt-sensitive? 🤔 🚨 Thrilled to share our EMNLP 2025 main conference paper! Prompt sensitivity has long been seen as a core weakness of LLMs—where tiny wording changes flip benchmark results. Our study finds: much of this effect stems from evaluation
0
2
13
Come to join us at REAL AI @ai_ucsb 😊🤗
The future of scientific discoveries lies in this synergy: human expertise guiding AI🤖, and AI augmenting human expertise🧪. Interested in #AI4Science? 👉My perspective as a #publicvoices fellow for the @opedproject w/ figure from @HaewonJeong00 & @YaoQin_UCSB . Link below.
0
2
14
When I first started working with LLMs, I tried to reproduce prior results—but often the original prompts were missing. Writing my own prompts would swing results wildly. Looking closer, I saw much variance came from evaluation, not the models. 🔍That insight led to this paper.
Are LLMs really so prompt-sensitive? 🤔 🚨 Thrilled to share our EMNLP 2025 main conference paper! Prompt sensitivity has long been seen as a core weakness of LLMs—where tiny wording changes flip benchmark results. Our study finds: much of this effect stems from evaluation
0
1
2
The era of artificial scientific intelligence is here. As algorithms generate discoveries at scale, what role remains for human scientists?🤔 Thanks @PLOSBiology for publishing my perspective @ai_ucsb @ucsbece @UCSBengineering @ucsantabarbara ! https://t.co/ZUtziZM9BN
journals.plos.org
Can AI become a true scientist? This Perspective explores how new technologies are reshaping scientific discovery, and why human expertise remains essential as we enter a new era of research powered...
3
26
116
Glad to share this work to build connections between these two research areas! Full paper is available at: https://t.co/HD9BqjNhgg.
🧵New survey: Bridging Distribution Shift and AI Safety Distribution shift and AI safety have long been studied in parallel. But how can their insights formally inform each other? We present the first comprehensive, mathematically grounded, and one-to-one aligned treatment. 1/6
0
2
13
Great work by Litian on this Neural Collapse inspired OOD detection work, accepted at CVPR-25! Code is also available at: https://t.co/9rabOdTdbd. Try out to see how Neural Collapse helps robustness!
github.com
Detecting Out-of-Distribution through the Lens of Neural Collapse (CVPR 2025) - litianliu/NCI-OOD
Our #CVPR2025 paper is out! Inspired by Neural Collapse, we show that OOD samples lie far from the origin and class weights—tying together many prior methods. Huge thanks to @YaoQin_UCSB, @liuziwei7, and @JingkangY for the collaboration & OpenOOD support! https://t.co/g0zzhGMw0E
0
1
11
Kenan has lead this amazing work for precise, iterative and customizable image editing. Go to our paper or website: https://t.co/R3G8SOg2DQ to learn more about it! 😃
Excited to introduce SPICE, a novel image editing framework that supports precise and local editing. In the example shown below, we iteratively edit the area around a fridge for various tasks. Paper: https://t.co/1DWPB9B5dV Code: https://t.co/tXpNHktLwt
0
3
15
🚨Only 1 day to go! 🚨 Join us at AIM-FM: Advancements In Medical Foundation Models workshop at NeurIPS 2024! 📅 When: December 14th, 2024, 8:20 a.m. PST 📍 Where: East Ballroom A, B We will be exploring the transformative potential of Medical Foundation Models (MFMs) in smart
0
1
11
Can't believe it's a year already! Incredibly proud of what this team has accomplished 🧠🌟 @emilyjacobs @caitaylo @russpoldrack @amykooz @crntozlu @joshbuck @kbcasaletto @bucklr01 @PaLab_UCSD Suzanne Baker @susanna_carmona @MagdaMartinezGa @AdeleMyersPhD @louisacornelis
One year ago, we launched the Ann S. Bowers Women’s Brain Health Initiative and what a year it’s been! From groundbreaking research to building a community of motivated and brilliant people, we’re proud to champion women’s brain health for generations to come. Here’s to many...
0
11
29
I'll be at @NeurIPSConf next week and would love to catch up in person! 🎉My lab at UCSB is hiring PhD students working on AI Safety and AI for Healthcare. If you're passionate about these areas, welcome to apply to UCSB and happy to chat if you're attending NeurIPS!
3
21
122
Posting a call for help: does anyone know of a good way to simultaneously treat both POTS and Ménière’s disease? Please contact me if you’re either a clinician with experience doing this or a patient who has found a good solution. Context in thread
142
327
1K
Great work and congratulations 👏 @WilliamWangNLP
🚀Introducing ChipAgents: the World's First AI Agent for Chip Design and Verification. Get ready to supercharge your workflow and accelerate your time-to-market! 💻⚡
0
0
5
Welcome to submit to our AIM-FM Workshop at NeurIPS 2024 to advance the medical foundation models 😀
0
3
14
Join us for this exciting postdoc opportunity and work on AI for science✌️😊
We are recruiting postdocs for 2024/25 @ai_ucsb ! You want to build next-gen artificial scientific intelligence?🤖 Apply to the UCSB Real AI Initiative! Deadline: Sept 15 Competitive salary Prime location for your future office 🌊 @YaoQin_UCSB @HaewonJeong00 @UofCalifornia
0
8
67
If you’re unfamiliar with long COVID, this thread from @SalvMattera about his experiences makes it clear why it’s so important to fund research to better understand, treat, and cure this disease
My case is mild compared to many other people. I have friends who are younger and healthier than me, and have been hit even harder. Many of them can no longer work. Last week, @BernieSanders introduced a bill to fund long COVID research. It is vital that it be passed.
3
73
274
🥰 Super excited to share this new work on benchmarking LLMs for carbohydrate estimation, which is a huge daily burden that every patient with diabetes needs to deal with multiple times every day. 👏👍Proud of my students for starting to investigate the potential of LLMs in
2
9
38
Excited to share this new work on degradation in the chain of diffusion done by our amazing students, co-advised with @HaewonJeong00 😊!
🏞️Ever wondered what happens if we iteratively finetune a diffusion model on its own outputs? SEVERE DEGRADATION! 🚨 Our latest research uncovers this phenomenon and introduces ReDiFine, a novel approach to ensure sustainability. 📜Preprint: https://t.co/ngTC50FuAN 🧵1/5
0
4
13
The 3rd AdvML-Frontiers Workshop (@AdvMLFrontiers
https://t.co/bYfJa1DeM4) is set for #NeurIPS 2024 (@NeurIPSConf)! This year, we're delving into the expansion of the trustworthy AI landscape, especially in large multi-modal systems. @trustworthy_ml
@llm_sec🚀 We're now
2
8
22