Calibrated Is Not Informative: What a Fall in Expected Calibration Error Shows About LLM Confidence, Why Verbalised Scores Fail on Resolution as Well as Level, and the Recalibration Baseline the Ensembling Results Omit

APA

Mahendrakar, P. (2026). Calibrated Is Not Informative: What a Fall in Expected Calibration Error Shows About LLM Confidence, Why Verbalised Scores Fail on Resolution as Well as Level, and the Recalibration Baseline the Ensembling Results Omit [Preprint]. Life of Research. https://doi.org/10.5281/zenodo.23606814

Copy APA

BibTeX

@misc{lor1332026,
  title        = {Calibrated Is Not Informative: What a Fall in Expected Calibration Error Shows About LLM Confidence, Why Verbalised Scores Fail on Resolution as Well as Level, and the Recalibration Baseline the Ensembling Results Omit},
  author       = {Pranay Mahendrakar},
  year         = {2026},
  publisher    = {Life of Research},
  doi          = {10.5281/zenodo.23606814},
  url          = {https://lifeofresearch.com/preprints/133},
  keywords     = {large language models, confidence estimation, calibration, expected calibration error, failure prediction, verbalized confidence, selective prediction, evaluation methodology},
  note         = {LOR-00133}
}

Copy BibTeX More on the record page