Pinned
Xinya Du
237 posts
Xinya Du
@Xinya16
Joined January 2016
We compiled a list of open-source LLMs’ model and data usage restrictions and licenses. It helped with my search for an LLM to include in our grant proposal. -- guess which one(s) we finally included :). It is added to "The Practical Guides for LLMs" github.com/Mooler0410/LLM…
GIF
Should we choose to use LLMs or smaller finetuned models in practical use cases? Take a look at our survey arxiv.org/abs/2304.13712 , which covers NLU tasks, Generation tasks, Knowledge-intensive tasks, abilities regarding scaling, some miscellaneous and real-world tasks.- Thank you, Heng! Very happy to join Erik Jonsson School of Engineering and Computer Science (@UTDJonsson) at UTD (@UT_Dallas) Fall 2022 in a "warm" and super exciting city.
- Just arrived at Singapore🇸🇬 for #EMNLP2023. Excited for reconnecting with colleagues/friends. Say hi if you’d like to chat about (multimodal) alignment/evaluation/hallucinations, interactive QA, IE, etc. I'm hiring fully funded PhD students. Pls reach out if interested.
- Check our recent work on building better LLM-based evaluators!Evaluating LLMs automatically via a ``strong’’ LLM (e.g. GPT-4) has inherent problems like self-enhancement. Inspired by insights from psychology and education, we propose peer eval (peer rank and peer discussion) to make LLMs better evaluators in this work. 🏅🥈🥉
- An insightful ACL 2022 paper based on research of Cornell's awesome undergraduates (advised by @clairecardie and me). ---- We built an automatic error analysis tool to gauge progress in IE since its inception "30 years ago", vs. four systems from the MUC-4 (1992) evaluation😀.ift.tt/81x7PHE Automatic Error Analysis for Document-level Information Extraction. (arXiv:2209.07442v1 [cs.CL]) #NLProc
- Nice to see Ozan (also a student of @clairecardie) as a member of this project! I started the NLP journey on opinion extraction with Bi-RNN back in 2014, which was based on his solid C++ implementation "entirely from stretch".BloombergGPT is a new LLM for finance. It's a 50 billion parameter language model trained on financial data. Claims the largest domain-specific dataset yet with 363 billion tokens... further augmented with 345 billion tokens from general purpose arxiv.org/abs/2303.17564…
- Replying to @kaiwei_chang @jieyuzhao11 and 12 others@ZhiyuChen4 is joining UTD. 👏
- Replying to @Xinya16Also, very excited to be a colleague with many awesome faculties in the department -- @irvlutd @rishiyer @davidyoung8906, etc.
- Replying to @Xinya16Thank you so much for my advisors (@clairecardie @elgreco_winter) and many collaborators @srush_nlp @scottyih @ABosselut @LuhengH (including many others that are not on Twitter)😄!
- Replying to @Xinya16Contact me (email: [email protected]; xinyadu.github.io) if you are interested in PhD/MS/Internship opportunities or research collaborations.
- Congrats and welcome, Zhiyu!I will be joining @UT_Dallas as an Assistant Professor in @UTDJonsson in Fall 2024. Deep thanks to my advisors @WilliamWangNLP, Xifeng Yan, collaborators, and friends for supporting me as always. Currently, I'm working as a postdoc in @S3DatCMU. Look forward to the new journey!
- Replying to @_julianmichael_I recall our brief discussion on this project, and now I'm excited to have a look at the preprint :)
