When Do LLM Preferences Predict Downstream Behavior?

This research investigates the critical disconnect between how Large Language Models express preferences and their actual performance on downstream tasks, re...

Level: advanced

By Katarina Slama and 5 other authors

Category: research