In both his calculation of the nominator and the denominator of his anthropic reasoning. The nominator requires a definition of an "ordinary observer" of which humans supposedly count. But it's not clear what the boundaries of an ordinary observer are in the definition. I don't really have enough time before bedtime to write ten paragraphs on the problem. But to summarize it seems very humanistically chauvinist. (Maybe paste that into Bard or ChatGPT and I'm sure it will know where I'm going with this).

The second problem deals with the assumed boundedness of the number of simulations, which I find problematic. Mainly because Bostrom doesn't seem to be very careful about thinking about how conservation laws will place significant energy constraints on nested simulations.

I don't think a top-level universe lacking conservation laws would be the kind of universe that would permit the minds of physics for life to evolve and build computers with simulations makes any sense, either.

But basically, if you were simulating entire universes within universes, in an ad infinitum nesting as Bostrom suggests, you'd quickly run out of energy to advance the nested simulations in anything resembling an economical way. We're talking about simulating entire universes down to sub-atomic particles, here. There's a real upper bound on the computation-energy budget here.

It turns out, when you take conservation laws into account, we're actually most likely to be living in a universe where no further simulations are possible. And the likelihood we are living in a top-level universe is far higher than in Bostrom (and Musk's) formulation.

That's all assuming we get over the ordinary observer problem.