Sooner or later someone is going to figure out how to do active training on AI models. It's the holy grail of AI before AGI. This would allow you to do base training on a small set of very high quality data, and then let the model actively decide what it wants to train on going forward or let it "forget" what it wants to unlearn.