Characterizing LLM Abstention Behavior in Science QA with Context Perturbations
Oct 1, 2024·
,,·
1 min read
Bingbing Wen
Bill Howe
Lucy Lu Wang
Abstract
We characterize how large language models abstain from answering science questions when presented with context perturbations, providing insights into model uncertainty and reliability.
Type
Publication
EMNLP 2024 Findings
We characterize how large language models abstain from answering science questions when presented with context perturbations, providing insights into model uncertainty and reliability in scientific reasoning tasks.
