We're only interested in the AI systems when they communicate human in a human like way. That's where the reinforcement learning with human feedback seems to come in. The more the AI say the stuff that kind of interests us, the more we're impressed by it.