Comment by whizzter

Maybe they realized that they were falling behind too much? To me it seems like user-feedback on bad descisions by the AI once it's trained to a basic level is among the most important signals in tuning the model to perform better.