Naming things with LLMs

The default output of any model, asked for names, is a puddle of made-up compound words. The workaround: constrain the search.

Feed it real references, real constraints, and real rejection criteria. Ask it to explain, not just generate. Then throw away 90% of what it produces without guilt.

You're not asking the model to name the thing. You're asking it to widen your search so you can find the name yourself.