Then you had promising projects like Plos, but they sold themselves. They turned into a joke: open access and good IF, but high fees for the author, thus becoming a quick way to get a sub-par paper published "for the points" if your lab can pay the fee. Pay to win, using a gaming term. If you know you have a good paper, you publish on any other (closed) journal with similar IF but cheaper.
The authors who put them up there didn't even plan on publishing in a journal. They just throw their work out there - no peer review, no nothing. Post the link on Twitter and maybe someone in the field will see it and find it useful.
This is especially true in fast-moving and highly applied fields like ML - the fields that are less "big science gang" and more "high intensity corporate R&D warzone".
PLOS One does publish pretty much anything, but that was always meant to be the point: "here's some data, make of it what you will. "