I must use PyTorch for ML, but badly wish I could use TypeScript instead. Its type system demolishes anything else I've used. Writing anything from SPI communications, to server logic, to rich GUIs is a breeze. Then I work in Python and have to fight `Any`s, ambiguous arguments, more verbose code, and bizarrely slow run times. It makes me sad.
While I agree with you, Python is getting better in this regard, as is mypy.
Making historical interfaces well-typed is another kettle of fish. For instance, Python's range() is actually two overrides with incompatible signatures. I bet PyTorch has more of such examples.
