Swift for TensorFlow is a result of first-principles thinking applied to machine learning frameworks, and works quite differently than existing TensorFlow language bindings. Whereas prior solutions are designed within the constraints of what can be achieved by a (typically Python or Lua) library, Swift for TensorFlow is based on the belief that machine learning is important enough to deserve first-class language and compiler support.
I’m not going to lie. This is more or less exactly what I was hoping would happen when Chris Lattner joined the Google Brain team. Much as I love Python, I’ve lamented more that once that Swift wasn’t the language of choice for machine learning. For me, it’s a sweet spot for maintainability, expressiveness and type-safety. Python is awesome for hacking together a prototype in a very short amount of time. Swift (IMO) is better for building a maintainable system, whilst also being efficient to write. Also: I like it when the compiler catches my typos.
Actually adding language level support for machine learning is something else altogether. That could make Swift the language for machine learning, rather than just a language for machine learning.