Preprint
Oct 2026
Can This LLM Annotate My Data? Estimating Zero-Shot LLM Performance
It is found that it is possible to predict LLM performance - to a certain degree - and that contrary to other work which highlights the usefulness of LLM verbalizations of self-confidence, internal log-probabilities are the most predictive signal.
Tong Lin, Grant Schoenebeck, Sabina Tomkins
· 0 citations