Whenever there's a reply that is inaccurate by the Genius, being able to rate the reply as such to retrain the language model would help it get better over time. Also being able to highlight particularly good responses as desirable so they happen more often. Being able to add a reason for the rating if desired, would be even better.