it's obvious that the feedback loop into the model brings value.. without it.. it's running without reward and will start doing whatever it can to just provide a response.. even if it hallucinates.. risk assessment and feedback of such risk will play a massive part in AI agents