Why is frequentism bad because it only gives certainty for infinite samples, but complexity is good despite being non-computable?
It's two sides of the same coin -- computable uncertainty va non-computable certainty.
It's two sides of the same coin -- computable uncertainty va non-computable certainty.
In the case of Kolmogorov complexity, a reasonable takeaway might be to use the shortest program that we're able to find, even if it's not the shortest overall.