An infinite sequence of algorithms converging on arbitrarily good polynomial error reduction? Fair enough, I certainly can’t rule that out at this stage.
But I don’t understand your last point: how can you pay only logarithmically for using a finer grid?
The post had a concrete complexity measure, which pays logarithmically for a finer grid (that is, doubling the size of the universe is the same as adding one more bit of complexity). The point is, you can only afford to pay logarithmically in the size of the universe (if you want known physical theories to have good complexity as compared to stupid explanations for our observations). Making the grid twice as fine is just making the universe twice as large, so you only pay 1 more bit: the extra bit needed to describe the larger size of the universe. If you disagree with this then you probably disagree fundamentally with my approach. That is obviously valid; I don’t really like my approach that much. But alternative approaches, like the speed prior, seem much worse to me.
Oh, sorry, yes, when your cost measure is complexity, then a finer grid incurs at most a logarithmic penalty, I agree. I also agreed the speed prior is a much worse approach—I would go so far as to say it is flat-out falsified by the observed extreme computational cost of physics.
An infinite sequence of algorithms converging on arbitrarily good polynomial error reduction? Fair enough, I certainly can’t rule that out at this stage.
But I don’t understand your last point: how can you pay only logarithmically for using a finer grid?
The post had a concrete complexity measure, which pays logarithmically for a finer grid (that is, doubling the size of the universe is the same as adding one more bit of complexity). The point is, you can only afford to pay logarithmically in the size of the universe (if you want known physical theories to have good complexity as compared to stupid explanations for our observations). Making the grid twice as fine is just making the universe twice as large, so you only pay 1 more bit: the extra bit needed to describe the larger size of the universe. If you disagree with this then you probably disagree fundamentally with my approach. That is obviously valid; I don’t really like my approach that much. But alternative approaches, like the speed prior, seem much worse to me.
Oh, sorry, yes, when your cost measure is complexity, then a finer grid incurs at most a logarithmic penalty, I agree. I also agreed the speed prior is a much worse approach—I would go so far as to say it is flat-out falsified by the observed extreme computational cost of physics.