Here's the problem:
The real thing that all of this discussion is circling around without actually saying aloud is "discovery is a thing that we want to happen to our content." It's the one part of the traditional social media architecture that is both intuitive and effective.
Consider. You are a person. You create content. You know other people. They know you. Because they know you, they see your content and either like it or not. There is no question of whether or not that when you create that content it will get seen. The question is whether or not it will get liked.
Now, consider the situation on Cake. You are a person. You want to create content. That content is about something – and you want that content to be seen. You have no guarantee that when you conceive of the post, there will be an audience which will read it. The best you can do is to look at the taxonomy as it stands, look at the number of followers for any given topic, and pick one or more that you can justify to yourself as being related to what you wanted to write the first place. Alternately, you can look at the taxonomy as it stands, look at the number of followers for any given topic, and write to target that audience.
Frankly, I hate writing in the latter mode. It feels cheap, it feels dirty, and it feels like I'm writing advertising copy. But there are writers and a lot of bloggers who "make their living" (emotionally or financially) by chasing the Dragon of writing just anything to any group of people they can find to advertise to.
Which brings us to discovery. On Cake, that is driven by two primary interfaces: firstly, search, the use thereof being extremely obvious. Secondly, finding content in topics that you already follow which have links to topics at the bottom to topics you may not already follow.
For myself, I actually prefer this structure of discovery to a lot of potential options online. I don't have to know any given individual, I don't have to judge any given individual; I just figure out what I want to learn about/be involved with and start digging.
But this is terrible for people who feel like they need to write for people to immediately consume. I just sit down and write whatever I feel like at any given time, slap whatever topics feel onto it, and shove it out the door into the world. That's how I do things. They want to know who's going to see this, when it's going to be seen, and if there is a real demand for it.
I am not sure that the twain shall ever be met, realistically.
"For you" is supposed to be one of the major mechanisms for discovery on Cake, particularly when sorted by "interesting," but it doesn't feel like people trust the mechanism to deliver their content to people who will in fact find it interesting. This may be a problem in perception, and it may do a fantastic job. This may be a problem in fact and it doesn't do so great a job. It's very difficult to tell with the current rate of content creation on the platform.
The "fishing for topics" behavior is probably one of the contributing factors for unnatural topic proliferation. People want to cover the broadest number of potential ways for people to see the content and absent a pressure which makes adding more tags more expensive in some sense, they have no reason not to.
Frankly, I suspect this will always be a real issue given the design of Cake as topic-first rather than creator-first. The fear that they won't get read is a big deal.
The Cake rule against "self-promotion" doesn't help, in this sense. I have five or six different places where my longform content may get published and it seems silly to repurpose my text by cut-and-paste from, for example, the steem blockchain to Cake, or a link to one of my professionally published articles to Cake if I would like to have the conversation about it on this platform. Add to that the difficulty of moving formatted content over (which ties to my whole Markdown thing), and it means that Cake has a few things going on which are understandably inserting some confusion into the creators who would like to do some stuff here.
Suggested topics for a given post? That might be okay, but it doesn't solve the original problem that the poster actually possesses which is trying to find an audience to target. There is nothing to say that NLP might suggest perfectly valid, reasonable topics which no one reads, because the article best fits into those topics.
It's a real situation.