500 prompt variations for semantic routing

In this demo we’ll cover using LLMs for intent classification and a bunch of ablations we ran. We’ll also see if humans are better than LLMs at prompting.