<ul><li>Recent studies have found pre-trained language models (PLMs) to suffer from miscalibration, indicating a lack of accuracy in confidence estimates.</li><li>Evaluation methods assuming lower calibration error estimates indicate more reliable predictions may be flawed.</li><li>Fine-tuned PLMs often resort to shortcuts, leading to overconfident predictions that lack generalizability.</li><li>Models with seemingly superior calibration actually have higher levels of non-generalizable decision rules.</li></ul>

The Reliability Paradox: Exploring How Shortcut Learning Undermines Language Model Calibration

Discover more