LLM Classification as Feature Engineering

A Hacker News discussion argues that using LLMs as classifiers is feature engineering: treat LLM outputs as features and train conventional models (e.g., logistic regression) on them for more stable classification.