Not really. In fact it might suggest something I'm specifically more worried about. Datasets that we use in research aren't really appropriate in production. They have a lot of biases that we don't exactly care about in research but you do in production that can also get you into a lot of political and cultural trouble. So really if they are going to just use public datasets and not create their own then I expect a substantially low performance, potential trouble ahead, and I'm concerned about who is running their machine learning operations.
Being in the ML community I have a lot of criticisms of it. There are far too many people, especially in production, that think "just throw a deep neural net at it and it'll work." There is far more to it than that. We see a lot of it[0]
Not really. In fact it might suggest something I'm specifically more worried about. Datasets that we use in research aren't really appropriate in production. They have a lot of biases that we don't exactly care about in research but you do in production that can also get you into a lot of political and cultural trouble. So really if they are going to just use public datasets and not create their own then I expect a substantially low performance, potential trouble ahead, and I'm concerned about who is running their machine learning operations.
Appreciate the detail here. Given your relevant experience sounds like something that the devs need to address.
Being in the ML community I have a lot of criticisms of it. There are far too many people, especially in production, that think "just throw a deep neural net at it and it'll work." There is far more to it than that. We see a lot of it[0]
[0] https://news.ycombinator.com/item?id=28252634
4 replies →