These days in ML it's more often "a convenient way" of driving CUDA. So I guess there are no "standalone programming languages". It's a silly argument, though.

These days in ML it's more often "a convenient way" of driving CUDA. So I guess there are no "standalone programming languages". It's a silly argument, though. If you have a million lines of Python code, and some of it calls some library that contains lower-level code, does that mean Python is a "scripting language"? I wouldn't say so.