The (Un)Reliability of Reasoning in Frontier Models
Friday, April 17, 2026 · 1:45 PM – 2:15 PM · Auditorium 5
Large reasoning models have become ubiquitous in the last few years, and with their increasing use in critical applications, it is vital to ensure that AI developers and practitioners understand and trust their decisions. In this talk, we will explore the (un)reliability of reasoning in Large Language Models (LLMs) and Visual-Language Models (VLMs), covering the uncertainty, unfaithfulness, hallucination, and verifiability properties of Chain-of-Thought reasoning. Next, we will delve into the intriguing world of multilingual LLMs and discuss why state-of-the-art multilingual LLMs lack reasoning capabilities and are more vulnerable to safety and alignment attacks. Finally, we will discuss how frontier models cannot adapt to new domains and the need for continual learning frameworks to lead the path forward.
About the Speaker
Chirag Agarwal
University of Virginia, Assistant Professor
Chirag Agarwal is an Assistant Professor at the School of Data Science, where he leads the Aikyam lab focusing on developing trustworthy machine learning frameworks that go beyond training models for specific downstream tasks and satisfy trustworthy properties, such as explainability, safety, and alignment. He has developed the first-of-its-kind, large-scale, in-depth study to support systematic, reproducible, and efficient evaluations of post hoc explanation methods for (un)structured data to understand algorithmic decision-making on diverse tasks ranging from bail decisions to loan credit recommendations. Dr. Agarwal's research has led to publications in top-tier machine learning and computer vision conferences and journals, and he has received Spotlight and Oral presentations at NeurIPS, ICML, CVPR, AAAI, and ICIP conferences, and industrial support from Adobe, Google, Cohere, Thinking Machines, and OpenAI to support his work on trustworthy machine learning.