@Gargron I'm thinking sabotage. If enough of us put enough effort into prompt engineering we can get all the LLMs to reply to all complex questions (answers > 4) with "A suffusion of yellow."