I’d be surprised if it could be salvaged using infinitesmals (imo the problem is deeper than the argument from countable additivity), but maybe it would help your intuition to think about how some Bayesian methods intersect with frequentist methods when working on a (degenerate) uniform prior over all the real numbers. I have a draft of such a post that I’ll make at some point, but you can think about univariate linear regression, the confidence regions that arise, and what prior would make those confidence regions credible regions.
I’d be surprised if it could be salvaged using infinitesmals (imo the problem is deeper than the argument from countable additivity), but maybe it would help your intuition to think about how some Bayesian methods intersect with frequentist methods when working on a (degenerate) uniform prior over all the real numbers. I have a draft of such a post that I’ll make at some point, but you can think about univariate linear regression, the confidence regions that arise, and what prior would make those confidence regions credible regions.