A new position paper published on June 20, 2026, discusses the limitations of the traditional aleatoric and epistemic uncertainty framework in the context of interactive large language model (LLM) agents.
The authors argue that this framework falls short in addressing the complexities involved in clarification seeking during AI interactions.
They propose an alternative approach that emphasizes the importance of underspecification-awareness, which could enhance the effectiveness of LLM agents in understanding user queries.