This article experimentally explores the way in which human agents learn how to process and manage new information. In an abstract setting, players should perform an everyday task: selecting information, making generalizations, distinguishing contexts. The tendency to generalize is common to all participants, but in a different way. Best players have a stringer tendency to generalise rules. A high score is, in fact, associated with low entropy for mistakes, that is with a tendency to repeat the same mistakes over and over. Though the repetition of mistakes might be considered a failure to properly employ feedback or a bias, it may instead turn out as a viable and successful procedure. This result is connected to the literature on learning.