People like to talk about how great a teacher failure is, and how it teaches you valuable lessons... and it's true! Failure does teach you lots of valuable things, including things that you don't learn by succeeding. However, succeeding at something challenging generally teaches you more.

Or to cast it in ML terms... you need examples of both success and failure in your data set, but the "success" data points are more valuable.