You make a good point! but this is a thorny area. When an article's title is completely generic in a way that conceals what's interesting about it, we sometimes don't revert the title. Such cases are covered by the site guidelines as 'misleading'. Other times we do revert it—for example, when a title is generic for what seems to be a creative reason, we'll often put the generic title back out of respect for the author. Or if it's generic in a way that is playful rather than bland.
There are hybrid cases too, where we'll leave an editorialized title up because of what you call discoverability until the post is well-established on the front page, then change it at that point. Also, we're responsive to feedback, so when a user makes a good case for us having made the wrong edit, we'll change it. When users disagree with each other, though, as happened here, that gets harder.
Btw I wrote about that CUDA title here: https://news.ycombinator.com/item?id=21618991. We'd normally have left the second-last edit up, but I changed it in response to a user complaint. Had it not been midnight on Saturday I'd have just asked the readers whether the macOS detail was really the single important thing in the article or not, and decided based on that.
Your question about moderator motivation doesn't quite land with me though. We're not excluding any of those values, we're just trying to balance them in a good way. What that means is that you're going to get a different answer, often a surprisingly long answer, about specific cases.