He uses the example of two babies that are identical (in his example, one in real life and an exact copy in a computer with inputs being identical). He considers both of these babies to really be a single person and thus you aren’t really predicting the second baby using the first, you’re simply ‘retrodicting’ what the baby will do by watching what the baby will do. Compare this answer to my own in the original post.
I do not believe there is no such thing as a deterministic nightmare. It’s just a pseudo-problem. We can stop worrying about this one.