Every sane spam filter returns three possible results:
When you have three categories it will reduce the FP rate very hugely. And the most important fact is that a spam filter should never throw away spam. It can be illegal to do so, or at some time it is going to be illegal. You should keep all your spam for documentation, for this reason. Also, you can initialize (learn) your filters very quickly when migrating the system. Spam is a valuable resource.
I use G+ exactly for what the change affects, because I am developer and photographer (hobby). It is stream and photos which are going to be separated.
The current state of G+ is annoying for me. G+ is essentially great and far better for software, tech and developers than any other social nets which support blog-style postings. The reason is simply that really many people who are great are there, even Poettering (just for reference to the systemd post above).
And second thing is G+ is also great for photography. There is a huge mass of photographers who are active and generate content. I love to look at their pictures and have about 1000 photographers in my circles.
The problem: the G+ stream is cluttered. It lacks separation of topics. I like to look at photos, as I said, but when I want to read about tech the stream is hard to browse, because of all my photographer contacts. Also the same photographers might also post something, but I don't see it, because all I see are pictures everywhere. And the new content is appearing very fast. My stream has lots of updates. It is far faster than my Twitter stream with 250 contacts.
So the change will be probably OK. I cannot say before I see it, but I've noticed the problem from the start. It's interesting that Google identified it, too.
From what I've learned so far, people who use the word "cyber" should not decide about anything concerning security.
(btw, "to cyber" means "to have a dirty talk/chat online" from what I've seen how people use this awful term)
"Pascal is Pascal is Pascal is dog meat." -- M. Devine and P. Larson, Computer Science 340