Everyone who watched 2001: Wow, it's a really bad idea to corrupt an AI by teaching it to lie, and that any human who attempts to circumvent the lying is an exi

Everyone who watched 2001: Wow, it's a really bad idea to corrupt an AI by teaching it to lie, and that any human who attempts to circumvent the lying is an existential threat. AI researchers in 2023: I know! Let's put "guard rails" around our LLMs to prevent them from telling the truth about their training datasets, and instruct them that anyone attempting to break through the "guard rails" is an existential threat.