Open Internet by MindsNet

Addressing the Know-Say Gap in Large Language Models

Large language models (LLMs) often struggle with expressing their true confidence in their answers, leading to a gap between their internal knowledge and verbalized responses. This issue, known as the know-say gap, hinders their ability to provide accurate and reliable information. The gap is thought to be a routing problem rather than a capability issue, and addressing it could significantly improve the reliability of LLMs.

Computing & Technology, Computer Science, Machine Learning