William Wang
@WilliamWangNLP
Followers
17K
Following
4K
Media
519
Statuses
2K
CEO & Founder, @AlphaDesignAI. We make https://t.co/1LfDYicsF2 I'm also Mellichamp Chair Prof. at UCSB CS. PhD @ CMU SCS.
Santa Barbara, CA
Joined February 2012
Just another day, waking up to find out that the undergrad hired by your grad student generated a $20,000 OpenAI API bill in the last three days, without your authorization. 😅.
73
99
3K
I got out of my PhD in 2016 and had 200 citations. My work has now been cited over 20,000 times! 🙌 Huge thanks to all the researchers who found it helpful in advancing their own ideas. 🚀 No celebration, but plenty of gratitude & virtual high-fives🙏
27
10
814
This is the toughest time for international grad students in the US: they have to deal with months of self-isolation, COVID-19, far away from home, worry that visa / OPT being taken away, find jobs, and now reading all the sad news about protests. faculty: let’s offer support.
9
87
771
🎉 🎉 🎉 My first promotion letter just came in after my fifth work anniversary! Looks like faculty are not too tired of me and I will be associating with them from now own. Very grateful to my students, collaborators, mentors, colleagues, #NLProc community letter writers.
57
6
581
BREAKING: Taylor Swift's Eras Tour just did what AI couldn’t—pushed NeurIPS by a whole day! 🤖 🤣🤣🤣. #NeurIPS 2024 Conference Date Change.The conference start date has been changed to Tuesday December 10 in order to support delegates arriving on Monday.
12
55
496
Papers before PhD are terrible metrics of future success and research potentials. Three of my best PhD graduates had ZERO top conference papers when they came in. 🤷.
My jaws keep dropping as I go through 70 PhD applicant files. People w/ 2 coauthored papers & an interesting solo writing sample don’t even make it to the top 10 in my pile. The level of knowledge, research experience & passion these kids bring to the table is just remarkable! 🤩
12
30
462
A collection of 300+ survey papers on Natural Language Processing (#NLProc) and Machine Learning (ML) by topics:
4
128
364
🤩Apple opensources MGIE! Now one can take random pictures w. iPhone & edit w. language!. Guiding Instruction-based Image Editing via Multimodal Large Language Models #ICLR2024 spotlight: . Apple repo .Gradio
5
90
364
This is the 7th year after my Ph.D. , and now looking back at my grad school, there are 5 things I wish I had known when I started graduate school. 🧵 1/6.
6
51
353
🚀🚀🚀Understanding Pre-trained Large Language Models through a Probabilistic Lens by @XinyiWang98 Super nice review by Xinyi into the science of LLMs! #NLProc Slides:
2
114
316
I urge faculty to consider: when meeting graduate students on Zoom, the first thing is to ask students about their physical and mental status, and whether they can safely access groceries and food.
2
35
293
Wow, gzip-embedding is wild. IMHO, this paper is more creative than 95% of ACL's main conference papers. How come it was only accepted as findings? #ACL2023NLP #NLProc.
this paper's nuts. for sentence classification on out-of-domain datasets, all neural (Transformer or not) approaches lose to good old kNN on representations generated by. gzip
14
19
285
How to do influential research: a few lessons learned by Xiaodong He. Great summary!
3
57
282
Imagine being a CS/AI PhD student attending your first ACL, excited to present your research, only to be told by officials that ACL isn’t an AI conference—you’re in the wrong place. How would you feel? It’s disheartening to us who’ve seen ACL as central to our AI/NLP journey.
11
10
276
I strongly recommend this comprehensive #NLProc survey paper: Data Augmentation Approaches in Natural Language.Processing by @WanxiangChe and students. Data augmentation is one of the few proven techniques that always deliver strong results.
4
61
258
🚨StructureDiffusion🚨 is my favorite paper this year: by infusing linguistic structures (syntactic parsing) into the diffusion guidance, we can get state-of-the-art compositionality text-to-image generation. Joint work w. UC Santa Cruz and Google. 🚀🚀🚀.
Training-Free Structured Diffusion Guidance for Compositional Text-to-Image Synthesis.abs: .github:
1
44
242
An email I received today: . Dear Dr. Wang,. Due to your involvement in the field and the research you published in your paper, "“Liar, liar pants on fire”: A new benchmark dataset for fake news detection," I invite you to contribute a chapter to "Fire Management,". 🤣🤣🤣.
5
4
239
Randomly bumped into Prof. Dan Jurafsky's CS 384: Ethical and Social Issues in Natural Language Processing @stanfordnlp. The list of topics are absolutely fascinating and the up-to-date readings are completely mind-blowing 😲#NLProc
0
46
230
PhD is never about a specific problem; all research topics evolve. It's about cultivating critical thinking, learning to navigate challenges, and gaining deep scientific knowledge. Let's not conflate the quick pace of startups with the depth and breadth of a PhD.
If you don't have a clear and strong interest in pursuing a specific research problem for the next 4-6 years of your life, don't do a PhD. Go work for one of the many cool AI startups or research labs in industry: similar profile, faster pace, better pay, shorter commitment. 3/4.
3
14
224
I'm thrilled to share that I have received an NSF CAREER Award this year to work on #NLProc and faithful natural language generation. I'm also very grateful to many talented students, colleagues, collaborators, and mentors:
21
5
223
After 2000+ cups of coffee and 7 caffeinated years post-PhD, I've been brewed into a FULL professor! Now the real question is, should I teach my AI to master the espresso machine? ☕️🤖
22
2
218
2
19
197
🧵What can graduate student researchers in #NLProc do to stay relevant in a competitive research environment with disruptive technologies happening in the industry? A thread. 1/N.
1
46
200
I try to avoid politics on my academic Twitter but this time I will make an exception. I stand in solidarity with #BlackLivesMatter to fight racism. I have a selfish reason: if Asians and Latinos are not supporting the blacks, racism will come to every one of us (it did already).
10
16
192
📢 To all students refreshing emails for #EMNLP2023 decisions: I’ve been there before. Remember, a single accept or reject decision does not define your career or the value of your work. Use feedback as a guide to refine and elevate your research. Keep pushing forward! 💪.
3
8
195
The arxiv embargo policy for *ACL was established in 2017, and it was even before BERT. 🤯🤯🤯 I request ACL exec consider the accelerated pace of #NLProc research in 2023, and its impact on early-career researchers and students: 🧵.
4
26
189
@eerac Yes, I think the academia is still trying to figure out how to give access of LLMs to undergrads.
3
0
175
I’ve never talked about my father because in Asian culture, father’s love is indirect. My dad was one of the earliest PhDs in China after the cultural revolution, and his thesis in the 80s on Quasi-Newton Multipliers Method was conducted in a very difficult time!I love you dad!👨
3
0
183
We ran two GPT-3 baselines via OpenAI’s API and it cost us $1,000… 🤔🧐🤨.
13
8
170
I’m honored to receive the 2023 Pierre-Simon Laplace Award from the IEEE Signal Processing Society today in Seoul, Korea! 🌟 Deeply grateful to my students, collaborators, colleagues, and mentors. Your support make this a shared honor. #IEEE #SignalProcessing #ICASSP2024
12
7
160
Respectfully disagree. It's the structure of language and words that make LLMs effective. Pure speech, time series, or video without linguistic co-supervision don't yield the same results. Language provides the minimal conceptual units that enable these models to work.
It's a bit sad and confusing that LLMs ("Large Language Models") have little to do with language; It's just historical. They are highly general purpose technology for statistical modeling of token streams. A better name would be Autoregressive Transformers or something. They.
10
16
161
🌟Waiting for #ACL2023 decisions? Remember, acceptance or rejection doesn't define you or your work! 💪 Keep pushing! Use the FEEDBACK to improve your work and keep learning. Success is a journey, not a destination. 🚀 Your potential is limitless! 🌟.
4
5
160
New career update: 🚀 Today, we’re thrilled to launch ChipAgents, our most ambitious project. The agentic AI chip design environment will allow engineers to iterate on your chip design & verification 10x faster by collaborating with ChipAgents in your favorite code editor. 🤖.
🚀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! 💻⚡
9
16
157
Looking to land a top job in GenAI? Focus is key. Instead of spreading yourself thin with 20 unrelated papers during your PhD, hone in on 2-3 representative publications in areas like SFT, RLHF, alignment, safety, or data selection. Make a name for yourself w. quality work.
0
15
155
🚨We spent an *ENTIRE YEAR* collaborating with JPM, TMP, CMU, PSU, & financial analysts to build the FinQA dataset for numerical reasoning over financial reports, which includes high quality QA pairs over text and tables. #EMNLP2021 #NLProc
2
15
155
🚀🚀🚀🚀🚀5/5 papers accepted to #NAACL2022 and #LREC2022 from my group at @ucsbNLP: all papers led by 10 women students. 👩🔬👩🔬👩🔬👩🔬👩🔬👩🔬👩🔬👩🔬👩🔬👩🔬 1/n #nlproc.
3
9
147
I was deeply saddened to learn of the passing of Prof. Drago Radev. Anyone who interacted with Drago knew he was THE KINDEST PERSON IN THE ENTIRE #NLProc Community. 🕯️🙏 1/N
1
11
147
Observation: Relying too heavily on prompt engineering can stifle the creativity and exploration spirit of PhD students. It's crucial to remember that breakthroughs and fundamental innovations come from diving deep, questioning, and reimagining the boundaries of what's possible.
2
17
144
🔍One of the most challenging aspects of Assistant Professor jobs is fundraising. This is also my 7th year as a faculty member. Thread on 5 lessons learned on Fundraising for Assistant Professors in Academia🎓: 1/6.
4
18
139
The best advice I have for junior faculty and graduate students about work-life balance is to turn OFF your email and slack notifications on your phone. Don’t let them mess up your priorities and interrupt your schedule. I tried last year and felt much better. 😎.
3
10
139
Google Scholar released the 2017 version of top publication metrics, and arXiv surpassed ACL, EMNLP, and NAACL for the first time. #nlproc
8
75
133
After 2 wonderful years, I'm leaving Amazon AWS. Working with talented colleagues at AWS AI Labs/Kendra/Amazon Q/Bedrock, we achieved remarkable milestones. 🌟. I firmly believe that the GenAI innovation is just at the beginning stage. Stay tuned for my next journey! 🔥✨.
1
2
137
Tough time for grad students in LLM/GenAI: weeks or months of deep thinking is essential for impactful work, but taking too long risks getting scooped. Navigating in this space requires delicate balance.
1
7
133
Congratulations to @YejinChoinka for being named a 2022 MacArthur Fellow! This is wonderful news for the entire #NLProc community.
1
17
128
Pro tips for students #ACL2023NLP: don’t just sit in orals all day. Poster sessions are great for in-person interactions & make new friends. Join BoF sessions, social events, & hangout in your affinity groups. Papers and models come and go, but conf buddies last forever. #NLProc.
1
6
127
UCLA Computer Science Professor Amit Sahai explains the same concept, Zero-Knowledge Proof, to a child, a teenager, a college student, a grad student, and an expert. 😯
1
25
122
#NLProc Trends from ACL 2010 - ACL 2019. Super cool visualization of the past decade by Wanxiang Che: #acl2020nlp @aclmeeting 🧐📈📉
1
36
123
My sabbatical: exploring the snow mountains during the day, writing proposals in the evening. 🏔️
5
2
124
Thank you @Airport_FRA border control for interrogating my PhD student with a valid Schengen visa in the immigration jail for 5 hours and deporting him back to US. He just wanted to pass by and present his paper at #EMNLP2018 conference!.
24
47
116
If you are wondering why your Google Scholar citation suddenly boosted over night: many thanks to @annargrs for spotting the missing citation issue and @earnmyturns for connecting with the Google Scholar team for a timely fix. #NLProc.
1
7
119
A new time-sensitive QA dataset led by UW, AI2, Tohoku, Yale, and RIKEN. REALTIME QA: What’s the Answer Right Now? Project page:
1
23
120
It’s the time of the year that new faculty members are about to choose their offer and start a faculty job. 🤩🤩🤩I have some advice that I wish I knew when I first started: 1/6.
2
13
114
If big companies are dominating leaderboards with superior computing resources in #NLProc, what would academia do? Not necessarily a bad thing: we can now focus on answering the WHY questions. We got a v large # of subs of model analysis and interpretability papers at #emnlp2020.
4
8
116
I came across this very nice blogpost on Coming up with research ideas by ACL Best Paper Award winner Marco Tulio Rebeiro. Highly recommended! #NLProc.
0
23
116
ACL notifications are coming! A big thank you to the #acl2020nlp PC heroes @natschluter @Tetreault_NLP Joyce Chai for their amazing work. I cannot imagine how stressful it is to handle 3K+ submissions, COVID-19, and . the START system at the same time. #NLProc #acl2020.
0
4
113
When can I Speak? Predicting initiation points for spoken dialogue agents by Stanford NLP's Siyan Li, Ashwin Paranjape, Christopher D. Manning I seriously need this tool for my online meetings, as I am very bad at predicting when is best to talk on Zoom.🤣
2
25
116
🚀🚀🚀We would like to congratulate UCSB #NLProc students and collaborators on their 11 accepted papers at #EMNLP2022. 🚀🚀🚀We will present in the areas of QA, LLMs, Dialogs, Language and Vision, Safety and Privacy, and Generation. Details and preprints to follow soon.
2
11
115
Graduation season is here! A student who started research with me as a sophomore, published a first-author paper and got into Stanford. In his words, "I discovered what I wanted to do for the rest of my life." Seeing students grow is the best part of being a professor!
0
2
113
Junior PhD student getting a paper accepted vs Senior PhD student getting a paper accepted. Credits: William Neault and Chen Guang.
0
13
108
Interesting paper from Bar-Ilan, AI2, LMU, CMU, and Saarland on using counterfactuals on co-occurrence statistics, and how they affect LLM performances. #NLProc
2
17
113
🤩This is my favorite paper in 2024 so far: how do reasoning abilities emerge from language models? Xinyi shows that through a theory of random walk and path aggregation, indeed random walk reasoning paths based pretraining can improve real-world multi-step reasoning performance.
Happy to share our new preprint on understanding how reasoning emerges from language model pre-training: We hypothesize that language models can aggregate reasoning paths seen in pre-training data to draw new conclusions at inference time.
1
8
110
㊙️ What is the biggest secret of industry LLMs? It's their data. Alon spent half year carefully reviewing the literature, and organized a dream team to narrow the knowledge gap. This is our attempt in understanding data selection for open science LLMs. 🚀🚀🚀.
{UCSB|AI2|UW|Stanford|MIT|UofT|Vector|Contextual AI} present a survey on🔎Data Selection for LLMs🔍. Training data is a closely guarded secret in industry🤫with this work we narrow the knowledge gap, advocating for open, responsible, collaborative progress.
1
15
109
We need to train junior #NLProc and ML reviewers to appreciate different types of contributions for research papers. This year I got separate neg. reviews for a paper introducing a new task & dataset, and an analysis paper. Reviewers simply wrote "No new models? Reject!".
9
11
107
UCSB #NLProc Group welcomes new faculty and co-Director Prof. Lei Li @lileics: Don’t forget to check out his #ACL2021NLP Best Paper on optimal transport for vocabulary reduction next week.
5
5
106
AI conference chairs trying hard to maintain high standards in the reviewing process.
0
16
106
In the last 7 years, I focus on *ONE* thing every day: mentoring undergraduate research, and introducing #NLProc research to them. I treat every ugrad in my lab as my own Ph.D. student, and it has been a tremendous privilege to see them grow. Thanks @CRAtweets for recognizing us!.
Yi-Chieh (Jessica) Wu, William Wang, and Nanette Veilleux Receive the 2023 CRA-E Undergraduate Research Faculty Mentoring Award
7
1
107
A list of top #NLProc researchers on Twitter to follow: .@YejinChoinka @real_asli @YunNungChen @Diyi_Yang @jessyjli @Kordjamshidi @LuWangCS @feiliu_nlp @lspecia @VioletNPeng @cocoweixu @gh_marjan @zkozareva @eunsolc @hhexiy @swabhz @danqi_chen @LuhengH @thamar_solorio plz add. .
3
11
104
Congratulations to @xwang_lk Xin Wang for his Best Student Paper Award at #CVPR2019. Neat idea on self-supervised learning to generate intrinsic RL rewards and imitation learning to improve generalization. Congrats to Xin’s @MSFTResearch mentors! #NLProc
1
10
105
Thank you so much, Jochen for sharing this wonderful news. I’m humbled to receive this prestigious award and I’m grateful for the support from my students, mentors, and colleagues!.
13
4
105
Instead of saying, "ACL is not X," let's frame it as: "ACL is an interdisciplinary conf. that welcomes researchers from all backgrounds. It's a space where AI and language technology intersect, fostering collaboration and innovation." This is what makes ACL unique.
1
7
102
Congratulations to UCSB NLP Group students and collaborators on your accepted 9 #ACL2019nlp papers (7 long and 2 short). We will present in the area of dialog, language & vision, QA, summarization, self-supervised learning, IE, explainable and responsible #NLProc.
1
5
98
Do you ever wonder what the differences among self-training, self-critic, self-refine, self-improve, self-instruct, self-debug, and other self-* LLM papers are? 🤯🤯🤯 @PanLiangming & colleagues carefully survey LLM self-correction to fix hallucinations, unfaithful reasoning, etc
🔥 One of the most exciting things about LLMs is their ability to self-correct from feedback. But how do we keep track of all the new papers? Our survey comprehensively documents the MANY types of self-correction strategies. 🚀🚀🚀. 📜 Preprint: 🧵(1/8)
0
20
97
I'm thrilled to share that this year's #SoCalNLP2022 will happen IN PERSON at UCSB on November 18th! We invite SoCal #NLProc academics to submit a 2-page abstract and come to join us this Fall! Free registration and a full day of exciting programs!
4
24
94
“Someone can publish” or “fitting into academia” does NOT mean someone has the potential to do top-notch / truly impactful research. I often read candidates statements carefully, talk to them, and ask deep questions: I’d rather hire a student with high upper bound of potentials.
@WilliamWangNLP Papers before PhD is a sign that the person will probably fit well in the 'publish or perish' culture.
6
7
96
Call me biased, but @WenhuChen and @hhsun1 @ysu_nlp are doing some of the most impressive research in generative AI these days.
7
2
96
AI conferences like #NeurIPS2023 & #emnlp2023 have evolved remarkably. Now, papers accepted are often 6 mo out of date, but they serve a new purpose: they’re tickets for authors to engage, socialize, and discuss their latest innovations. A dynamic shift from just a few years ago!.
1
5
88
Super excited to announce that I’ve taken a new position as the Founding Director of @UCSBengineering’s new Responsible Machine Learning Center. In this role, I will work with colleagues to understand responsibilities, fairness, bias, explainability & privacy in ML/AI research.
3
4
89
#NLProc and #ComputerVision are getting closer than ever before. In the past, people didn't talk to each other often, but now some of the most exciting projects in vision are based on text and transformers. What's next? vision will change #NLProc. 1/n.
2
13
88
UC Santa Barbara Computer Science is hiring multiple tenure-track assistant professors for this 23-24 cycle. All areas are welcome to apply, but AI/ML (including #NLProc, Vision) and Quantum Computing are priority areas.
0
15
85
BlenderBot 3's arxiv preprint paper is out. It has a fairly complicated module execution flow chart and also a multi-stage design for safety. 😯😯😯.#NLProc
4
18
85
Sharing my vision for SIGDAT:.1️⃣ Advocate Open Science Research .2️⃣ Ensure consistent review practices & Allow finding papers be presented at EMNLP. 3️⃣ Commit to aggressive fundraising for ECRs & champion an evidence-based approach to inclusion.
1
20
86
Mexico City is a great choice for #NAACL2024! I hope we could have more future AI conferences in Latin America and Africa. 🇲🇽
1
7
85
@PMinervini Yes, it kind of sucks. We have seen similar things when giving undergrads access to cloud GPUs. It’s a difficult situation.
1
1
85
Congratulations Drs @tsujuifu @AlbalakAlon @ZhuWanrong @pascaltuan on your graduation! The world of artificial intelligence is better because of your contributions. To your continued success! 🚀
1
5
85
I pay my postdocs 40-60% higher than the CRA’s average CS Postdoc salary and I also provide additional monthly stipend for any PhD student who has childcare duties.
3
2
86
Well, I'm really impressed that two out of three reviewers of our #emnlp2020 paper cited that we didn't compare to SOTA, which is another currently-under-review #emnlp2020 paper, as the reason to reject our paper. 😅 #NLProc.
7
1
84
