Listening to the latest podcast, I followed the advice to get an incognito window and look at Stack Overflow as a brand new user:
None of the questions have any votes, only one has an answer, and many have only ever been seen by the author as far as we can tell. (For the moment, let's assume the view number is accurate and current.) Without even reading them, I can guess that these are likely to to be among the least interesting questions on the site.
The very design of the feature optimizes for unanswered (or badly answered), unviewed, new questions. Question score does play a role, but it can be completely negated by 5 answers, one answer with a score of 5, 67 views, or a combination of the three.
Now if you have some favorite and ignored tags, and if you have tag scores, the picture is a little less bleak. When the system gets to know what interests you, it can serve up more posts in tags that are at least potentially interesting. But the less you have interacted with the system, the more random the list appears. Thanks to Sturgeon's Law new users are getting at least 90% crap from this algorithm.
The weird thing is that we know that our good content is in answers, not questions. I get that the idea behind the "interesting" tab is to bring to the surface questions that the user might be interested in answering, not the ones they might be interested in reading. But for a brand new user, is that really what we should be optimizing for? Do we want to show a bunch of junky questions to a person who is likely more interested in asking than in answering?
For users with no tag score and no tag preferences, eliminate questions with no votes and eliminate the criteria related to answers and views. The two criteria would become:
question score +200 × score, up to +1,000 total question last activity date -1 × (seconds / 15)
(Note: it is no longer possible to experiment with the criteria, so it's not possible for me to test this algorithm.)