I found that it mostly works great, but sometimes it gives really wrong results - without any warnings. My professor complained about various errors too (mostly

I found that it mostly works great, but sometimes it gives really wrong results - without any warnings. My professor complained about various errors too (mostly in dynamical systems, some control theory, but the problems seemed to be of a general sort), so it wasn't just a 40 cm error. To me this suggest that the language has plenty of leaky abstractions. My feeling is that in languages like python or julia, if you do something wrong you are more likely to hear about - and understand - it.