Very obvious LLM stuff trying to figure out what people consider normal, or where social boundaries are for things like cheating, stealing, lying, etc.

Oh god, I hadn’t thought of that angle. We’re not just teaching bots to detect lies, we’re training them to lie successfully 😱