[flatline]

Interesting. Most ML models presumably operate over IEEE double precision floating point values. These have a couple interesting properties like INF and NAN, but not infinitesimals.