I assume you mean what works for what problem. I think the problem is these are moving so fast that what didn't work yesterday has been coerced into being the

I assume you mean what works for what problem. I think the problem is these are moving so fast that what didn't work yesterday has been coerced into being the best option for, say, image classification today. Hell, ANN itself was widely the last choice for ML problems until a couple of years back and now it seems like the first choice go-to.