I just had a Claude episode. Instead of trying to fix the bug, it edited the data to hide the bug in the sample run. This kind of BS behavior is not rare. Absolutely, if you do not understand every bit of what's going on, you end up with a pile of BS.