RESEARCHInvestigateNEXT 12 MONTHS
CaliDist: Calibrating Large Language Models via Behavioral Robustness to Distraction
arXiv cs.CL — Computation and Language
Factual evidence
What the source reports
Researchers introduced CaliDist, a post-hoc LLM calibration method that penalizes confidence scores based on model vulnerability to distraction.
Open source