We investigate the internal behavior of Transformer-based Large Language Models (LLMs) when they generate factually incorrect text. We propose modeling factual queries as Constraint Satisfaction ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results
Feedback