> Note that quantum computers are no more powerful than Turing machines; But computers with self-consistent time travel are more powerful. It's perhaps physica

> Note that quantum computers are no more powerful than Turing machines; But computers with self-consistent time travel are more powerful. It's perhaps physically possible to construct one of those, depending on whether we ever end up with a more-correct timeless theory of physics that makes different predictions to existing physics. > Backpropagation through time is also a pretty simple algorithm, which is already implemented on classical computers. I read “backprop through time” as meaning that you have the result of the backprop at the beginning of the process – that's more powerful, if we're using continuous computation. (With discrete computation, you're right that it's no more powerful.)