LLM shows surprisingly usable calibration classifying abstracts on human subjects

RexDouglass · x · 2026-09-29

Researchers had Jev (an LLM) classify whether abstracts analyze humans, finding its probability calibration is not totally made up — surprisingly useful given low cost, speed, and minimal tuning effort. A practical observation on using LLMs for literature screening tasks.

Original post →

More from Models

Models channel →