it can be a useful reference but I would not trust it by itself
at least, when it comes to the marketed ML model stuff. There's a lot to be said for correlated, properly sorted datapoints on *everything*, like an index for whatever you'd like. But people keep insisting it should smash the datapoints together itself like that's meaningful. It's not.
In regards to general usage outside of that, under controlled and specific situations machine learning/vision has accuracy people don't, memory and information-retention that people don't. Automating low-effort routine tasks so you can do the more important stuff is valuable, and I will forever hate the cryptobros and silicon valley cash-grab types for shoehorning it in to "we're going to automate the Important Stuff because you don't like paying people for it!"