Writing

Blog

Longer-form notes on selective prediction, uncertainty quantification, evaluation design, and the gap between what models know and what they claim to know.

The blog is the slower companion to the paper archive: fewer entries, more argument, and more room for ideas that are not yet finished.

  1. Why LLMs Need Reject Options

    Selective prediction gives LLM systems a way to trade coverage for reliability instead of treating confidence as one scalar.

    • selective prediction
    • uncertainty
    • LLMs
    • calibration