Hacker News Reader

Does it get reinforced for doing an obviously bad or good job?

Yes. Our data feeds include the outcome variables we are optimizing for which helps the algorithms retrain as new patterns emerge.