I imagine I’d find it annoying to have what I learn & change into limited by what a dumber version of me understands, are you sure you wouldn’t think similarly?
I think it would be ok, if I just had to make sure the entire span was covered by smooth incremental steps. Note that these ‘checkpoint selves’ wouldn’t necessarily be active all the time, just unpaused for the sake of, for example, translation tasks.
So original self [0] → [1] → [2] → … ->[999] → [1000]. Each one only needs to be understood by the previous, and to understand the next.
I imagine I’d find it annoying to have what I learn & change into limited by what a dumber version of me understands, are you sure you wouldn’t think similarly?
I think it would be ok, if I just had to make sure the entire span was covered by smooth incremental steps. Note that these ‘checkpoint selves’ wouldn’t necessarily be active all the time, just unpaused for the sake of, for example, translation tasks.
So original self [0] → [1] → [2] → … ->[999] → [1000]. Each one only needs to be understood by the previous, and to understand the next.