The Anatomy of Uncertainty in LLMs

Aditya Taparia, Ransalu Senanayake, Kowshik Thopalli, Vivek Sivaraman Narayanaswamy

ICBINB Workshop at the International Conference on Learning Representations (ICLR), 2026

picture_as_pdf PDF

Summary

We decompose language model uncertainty into three semantic sources — input ambiguity from underspecified prompts, knowledge gaps from insufficient parametric evidence, and decoding randomness from stochastic sampling — and show that their relative contributions shift across model sizes and tasks, with implications for reliability assessment and hallucination detection.

BibTeX

@inproceedings{taparia2026anatomy,
    title={The Anatomy of Uncertainty in LLMs},
    author={Taparia, Aditya and Senanayake, Ransalu and Thopalli, Kowshik and Narayanaswamy, Vivek Sivaraman},
    booktitle={ICBINB Workshop at the International Conference on Learning Representations (ICLR)},
    year={2026}
}