Angana Borah

Do LLMs exhibit belief-based biases like humans do? We examine the presence of belief congruence in multi-agent systems and find that LLMs exhibit even stronger belief congruence than humans. Paper: arxiv.org/pdf/2503.02016 Code: github.com/MichiganNLP/Be… 🧵(1/n)

🎙️ Michigan AI Director Prof. @radamihalcea highlights the transformative role that @UMich is playing in #AI — from groundbreaking research to real-world applications. At Michigan AI, we’re not just advancing AI — we’re shaping its future. youtube.com/shorts/X5rw2-I…

Excited to share our #ACL2025NLP Findings paper Rethinking Table Instruction Tuning. ⚠️ Table LLMs fail at out-of-domain tasks & lose general skills? Our fix: 🔥 Less data + smarter tuning = Better Table LLMs! We show: 🔍 Hyperparameters (e.g., learning rate) make/break…

New RAG-empowered tool for cancer care prep. Helps patients go from uninformed to visit-ready - by guiding knowledge, value reflection, and visit question preparation, integrated with the Ottawa Patient Decision Guide 👇 Demo arxiv.org/abs/2507.04026 #CancerCare #RAG #AI4Health

🏛️Can we trust #LLMs to value democracy? 🎉Check our paper "Democratic or Authoritarian? Probing a New Dimension of Political Biases in LLMs". We find Llama4 has a stronger authoritarian bias, and all models show more bias in Mandarin than in English📜arxiv.org/abs/2506.12758…

🔥 Excited to introduce ManyICLBench (ACL 2025) 🧐 Do many-shot ICL tasks evaluate LCLMs' ability to retrieve the most similar examples or learn from many examples? We carefully analyzed numerous tasks and categorized them. 📄 Paper: arxiv.org/abs/2411.07130 #ACL2025

🔍LLMs now give medical diagnoses, legal advice, and even tackle scientific problems. ❓Your LLM sounds smart. But what if it’s just good at faking expertise? 🚀We built ExpertLongBench to find out. 📉And the results? They revealed several concerns.👇 🔗 huggingface.co/spaces/launch/…

Really excited about our recent large collaboration work on NLP for Social Good. The work stems from our discussions at the @NLP4PosImpact Workshop at #EMNLP2024 @emnlpmeeting. Thanks to all our awesome collaborators, workshop attendees and all supporters!

Those of you in Ann Arbor: why not do “AI & dinner” tonight? With the flurry of AI tools coming out any day, chances are that your (and your data’s) privacy are on your mind. Join us for a conversation with top experts on AI & Privacy—Profs. @RoyaEnsafi and Florian Schaub ⬇️

Very proud of all of Do June’s accomplishments, all the way to his successful PhD defense this week! 🎓👏 He not only developed new NLP methods to address key challenges in counselor training, but also saw them through to real-world deployment, directly supporting counselors in…

🎙️WEDNESDAY at #AAMAS2025, don't miss Prof. @radamihalcea's keynote, focused on exploring the evolving relationship between #AI agents and human behavior @AAMASconf aamas2025.org/index.php/conf…

Check out the BRIGHTER dataset -- Human-Annotated Textual Emotion Recognition for 28 Languages. I hope to see all the amazing collaborators in Vienna #ACL2025