“We found that if you ask the LLM, surprisingly it always says that I'm 100% confident about my reasoning.” Chirag Agarwal examines the (un)reliability of chain-of-thought reasoning, highlighting issues in faithfulness, uncertainty & hallucination.

Thrilled to announce that our five-month-old lab got three papers accepted at #NAACL2025! 1. Operationalizing Right to Data Protection: shorturl.at/IPfxn 2. Analyzing Memorization in LLMs through the Lens of Model Attribution: shorturl.at/1OBdf 3. Impact of

NOW OPEN: Applications for post-doctoral researchers to enrich UVA’s expertise in climate research. Could this be you!? Positions and more online - environment.virginia.edu/g2c-fellows

Presenting "The Multilingual Mind", a survey of multilingual reasoning in language models for a deep dive into how current language models reason across languages. Our survey comprehensively reviews existing methods that leverage LLMs for multilingual reasoning, outlining

The LaCross AI Institute awarded the 2025 Fellowships in AI Research (FAIR) to four members of UVA faculty, including School of Data Science Assistant Professor Chirag Agarwal. To read more about his accomplishment, click here: bit.ly/4gRCcce

LLMs may hallucinate non-factual stmts when QA-ing. So why not ask them to reference facts from the input while answering questions? It works! 🌟 Highlighted Chain of Thought prompting (HoT) enables LLMs to highlight facts in the input & then write a fact-grounded answer. 1/n

Google DeepMind India is hiring for research scientist role in multicultural & multimodal modeling. Strong candidates with proven research experience are encouraged to apply I shall be at #icassp2025 Hyderabad on Apr 8, happy to meet and chat, pls DM job-boards.greenhouse.io/deepmind/jobs/…

I’m at #NAACL2025 to present our latest work: "Analyzing Memorization in LLMs through the lens of Model Attribution" - We dive into which parts of the transformer architecture are responsible for memorizing training data 📄 Paper: arxiv.org/abs/2501.05078 🧵👇

Thank you for summarizing this work, Rohan Paul 🙏

ICML Conference is around the corner! Are you presenting any papers or hot takes in Trustworthy ML? Share your work in this thread and we’ll retweet! 🚀

Congrats LOGML Summer School for running an in-person grad ML summer school at Imperial College London and opening up research opportunities for students worldwide! 👏 logml.ai Thanks to organizers, Valentina Giunchiglia, and mentors Guadalupe Gonzalez Chirag Agarwal Ruthie Johnson Yasha Ektefaie

Looking forward to yet another exciting workshop this year at NeurIPS!! Consider submitting your latest work at the intersection of AI safety and regulation.

Had a great time interacting with the students from IIIT Hyderabad, discussing the (un)reliability of CoT reasoning and multimodal explainability. Thank you for the invite, Ponnurangam Kumaraguru “PK”!

⏰ ONLY 9 DAYS LEFT! ⏰ Submit to 3rd Regulatable ML Workshop NeurIPS Conference Call for Papers: regulatableml.github.io/cfp/ Also seeking reviewers! Fill out our form: forms.gle/JxHmzEZuwKgtSd… Don't wait - submit now! #CallForPapers #CallForReviewers #NeurIPS2025 #MLSafety #TechPolicy

Only 3 days remain before the submission deadline. We look forward to your contributions!