Don't classify. Hallucinate!

This post explores a novel technique for LLM (Large Language Model) classification, suggesting that allowing the model to generate plausible but fake entities can enhance its ability to classify by later resolving those fakes into real entities.