The paper seems not very quantative. It is not obvious from it whether a human needs a thousand bits, a million bits, a trillion bits - or whatever.
It would be quite impressive if it were able to...
My point was that Legg has shown, as I understand it, that any powerful prediction algorithm which is powerful enough to predict most/all of the universe (as one would expect a fearsome AGI to able to do) will be at least as complex as the universe it's predicting.
The most simple, able to self-improve "seed" for a superintelligence must be how complex?
Give your (wild) estimation or a pure guess in the terms of bits.
Do you reckon it is 1000 bits enough? At least 1000000 is required? More than 10^20 bits?
I wonder what your approximate numbers are.
Thank you!