I have co-authored several papers accepted at top AI/NLP conferences such as ICLR, AAAI, ACL, EMNLP, and EACL. One of my papers, titled “UnNatural Language Inference,” received an Outstanding Paper award at ACL 2021. Furthermore, I co-chaired the workshop on “Novel Ideas in Learning-to-Learn through Interaction” at EMNLP from 2021 to 2023. I also serve as a Meta-Reviewer at ARR-*ACL conferences and regularly review papers for top ML conferences like NeurIPS and ICLR.
Currently, my research interests revolve around leveraging LLMs for real-world applications in language generation and understanding. My primary focus is on identifying systematic failure modes in LLMs that hinder their practical applications, particularly in terms of factual errors and hallucinations. My expertise in addressing these problems encompasses:
1. Developing metrics and diagnostic datasets to pinpoint these problems.
2. Implementing data-driven pipelining approaches, like Retrieval Augmented Generation (RAG) style architectures, to tackle these issues.
3. Training LLMs to utilize tools effectively.
4. Exploring fundamental machine learning questions to create efficient language models for diverse applications.