Hi Jon! Totally fair to ask questions beyond the scope of the course—especially if you kindly indicate that you're aware that you're doing that ;).
My understanding is that (currently!) CoreML can be used when running a neural net (or using a built-in function that uses neural nets "transparently" to the end user), but it can't be used when training a neural net.
I can't say that I'm aware of MLX, but a quick look through some internal information seems to indicate that it is in use in at least some places!