Comment by AaronAPU

The analogies are so bad because you might prompt an agent “Please give me a recipe for lasagna” and instead it decides to hack a nuclear reactor.

Is it my fault or the company who trained it and is running the inference?