I just realised that Schmidhuber's Speed Prior is similar to the following revision of the Solomonoff prior. Consider the Universal Turing Machine which enumerates all strings and tries to run each string's Turing machine by running the th step of the th sub-machine in the same order as the order of pairs . Then the initial prior is that we could be in any step of the UTM with probability .
I expect that this prior dissolves the Boltzmann Brain problem since the BBs disproportionally emerge at times bigger than years.
I expect the revised prior to also solve Pascal's Mugging. Suppose that the Lord of the Matrix simulates both the agent's universe and another universe where, unless the agent pays the Lord, infinitely[1] many people die/the agent doesn't receive infinitely[1] many years of fun. Depending on the universe's and the prior's implementation, the LTM's offer is reduced to being at most 1E250 times more useful than the agent's entire life (e.g. if the agent was found by simulating the 13.7 billions of years in the lightcone up to the present with a fidelity of Planck time-distance, then finding the agent took at most 1E60 steps in simulated time and 1E240 inter-simulated-cube operations).