Probing LLMs Instead of Generating

This article describes a technique for using LLMs as classifiers by extracting hidden states rather than generating text. The...