Bo Ni

🚀 New Survey Alert! 🚀 Excited to share our latest survey on Trustworthy Retrieval-Augmented Generation (RAG) for Large Language Models (LLMs)! 🎉 With the increasing adoption of RAG in AI applications—ranging from medical question answering to legal document analysis—ensuring trustworthiness has become a critical challenge. In this survey, we provide: ✅ A comprehensive roadmap for developing trustworthy RAG systems ✅ A structured discussion on six key pillars: Reliability, Privacy, Safety, Fairness, Explainability, and Accountability ✅ A deep dive into existing challenges, methods, and open problems ✅ A review of evaluation metrics and future research directions This work is a collaboration with Zheyuan (Frank) Liu, Leyao Wang, Yongjia LEI, Yuying Zhao, Xueqi Cheng, Qingkai Zeng, Xin Luna Dong, Yinglong Xia, Krishnaram Kenthapadi, Ryan A. Rossi, Franck Dernoncourt, Mehrab Tanjim, Nesreen K. Ahmed, Xiaorui Liu, Wenqi Fan, Erik Blasch, Yu Wang, Meng Jiang, and Tyler Derr. Huge thanks to my co-authors and advisors for their insights! 📄 Preprint: lnkd.in/dEaWGFTa 🔍 GitHub Reading List: lnkd.in/dXavPb2i We hope this survey serves as a valuable resource for the community! Looking forward to feedback and discussions. Let’s push forward towards more trustworthy AI!

(5/5) This is only one out of numerous comments that perpetuate harmful stereotypes, demonstrating the systematic racism under the veil of a false sense of calm. As a Chinese person, I am hurt and disappointed to learn that such attitudes still exist, but we must face the reality that these issues persist within our community. As a community, like the girl in the stand, we should stand together to call out these injustices, engage in honest conversations, and work towards fostering understanding and empathy.

(4/5) Second, while the comparison between Americans and Chinese might seem fair on the surface, it is fundamentally flawed and unwarranted. Americans, as a group, are not marginalized. On the contrary, they are part of the dominant power structure in global academia and industry. With the privilege, they are less likely to face systemic bias and stereotyping when accused of wrongdoing. Conversely, Chinese individuals are part of a marginalized group that has historically faced suspicion and xenophobia. The two situations are not comparable, and such a suggestion, unfortunately, erases the lived experiences of marginalized groups.

(1/5) I wrote a short opinion piece in response to comments I frequently see in defense of Dr. Picard. I want to share some of my thoughts here.

I am glad to share that our latest work, "Towards Trustworthy Knowledge Graph Reasoning: An Uncertainty Aware Perspective," has been accepted by AAAI 2025! In this paper, we tackle the challenge of uncertainty quantification in KG-LLM (Knowledge Graph-Language Model) frameworks. Our proposed method, UAG (Uncertainty Aware Knowledge-Graph Reasoning), introduces a novel framework integrating conformal prediction to provide theoretical guarantees while balancing prediction size and reliability. Check out our paper below: Paper: arxiv.org/pdf/2410.08985 Code Repository will be publicly available soon! A big thanks to my collaborators Yu Wang (University of Oregon), Lu Cheng (UIC), Erik Blasch (Air Force Research Lab), and Tyler Derr (Vanderbilt University) for their incredible efforts and contributions! Your feedback and thoughts are welcome! Let’s discuss how UAG can be applied to high-stakes scenarios like medical and financial decision-making.