In my recording of an older gentleman it showed two problems
1. He is not a rapid speaker so it kept throwing in paragraph breaks that I have to take a lot of time to "undo". Plus I have to delete "Speaker 1" at the front of each sentence --OR-- lose all the speakers and try to guess the context later. Both are horrid options.
2. He has a slight stutter so everywhere is "I I I' or "and and and". Things which I would have thought AI would be smart enough to remove triples or at least doubles.
3. In REV.COM transcription, I am allowed to type over names like "Speaker1" and it instantly changes it to "Joe" or whatever throughout. This would be useful in a team approach where I was in charge of labeling the speakers and then handing it off to another team member.