So long as perceived LLM skill is still "spiky" - e.g. within a domain, still showing relatively high variation in ability (often depending on the task or user, to be fair), people will continue to dismiss it