No dataset. I'm using two types of prompts to direct the LLM on outputs, an overall system prompt that applies to all the outputs and eight style prompts corresponding to each of the eight cringe categories.