Comment by simiones

It might get it right some of the time, but it will be necessarily wrong some of the time. It's also very possible that for many (possibly even most) TMs, the most efficient algorithm for predicting the output is that TM itself.