That isn't really the case; while many neural network implementations make nondeterministic optimizations, floating point arithmetic is in principle entirely deterministic, and it isn't too hard to get a neural network to run deterministically if needed. They are perfectly applicable for lossless compression, which is what is done in this article.

Unix time is just the number of seconds since January 1 1970, isn’t it? How is that base 10, or any other base? If anything, you might argue it’s base 2, since computers generally store integers in binary, but the definition is base-independent afaik.