Cycles of Thought: Measuring LLM Confidence through Stable Explanations

Evan Becker, Stefano Soatto

In many high-risk machine learning applications it is essential for a model to indicate when it is uncertain about a prediction. While large language models (LLMs) can reach and ev…