I think the problem is that there is much more emphasis on the technology stack and trendy patterns than there is on the actual product. When I build software, I build it for a purpose and it is designed for maintainability. I'm not doing it just to play around with a new toy. But it seems most of the attention goes towards toys.
I swear, most of the ML projects I see right now are basically, "I totally figured out how to setup TensorFlow and stumbled on a large set of pre-existing training data." The resulting project looks unusable as a product or component of a product. And given the difficulty in setting up the environment and finding training sets, I have some doubts as to replicability or even likelihood anyone will try to replicate it. It almost feels like you could create completely fake results and nobody would question it.
Not meant to be a shot at the ML field, but more at the technology "community" that consumes this stuff. It's like magazine articles vs books. Everyone reading articles about books, literary critique, discussing authors and genres. Few people actually reading books. Fewer still writing them.
I swear, most of the ML projects I see right now are basically, "I totally figured out how to setup TensorFlow and stumbled on a large set of pre-existing training data." The resulting project looks unusable as a product or component of a product. And given the difficulty in setting up the environment and finding training sets, I have some doubts as to replicability or even likelihood anyone will try to replicate it. It almost feels like you could create completely fake results and nobody would question it.
Not meant to be a shot at the ML field, but more at the technology "community" that consumes this stuff. It's like magazine articles vs books. Everyone reading articles about books, literary critique, discussing authors and genres. Few people actually reading books. Fewer still writing them.