The Mirage of Calibrated Confidence: Trajectory-Independence of Verbalized Confidence in Vision-Language Models
It is shown that confidence is insufficiently sensitive to what the reasoning trajectory actually contains, and that calibration training can paradoxically worsen this disconnect, and proposed TGS-Bench is a model-agnostic suite spanning 10 benchmarks with controlled good/bad trajectory pairs.
Jisoo Yang, Jaeho Han, T. Pham et al.
· 0 citations