« Back to publications

Better Scores, Worse Grounding: Hidden Regressions after Fine-Tuning in Dialogue Fact Verification