That's because the latest count is on the order of 3M views while the scale of the graph is 1B views, or 0.003% of the full scale. It's dramatic but not wrong if you believe SO's own data. Take a look at the raw data linked in the article.
That just has got to be wrong. I refuse to believe that we had a significant proportion of the world population viewing StackOverFlow answers in 2011, either the counting is wrong or the dating or aggregation is wrong. It simply cannot be true. I know a lot of HN readers live in the tech-mindset-bubble, but it simply was never the case that any answers on stack overflow rivaled GangNam styles popularity at its peak breakthrough.
This is number of views, not number of unique users. It also likely includes views by bots. Notice how the viewership drops dramatically around 2014 once you could start downloading the database instead of scraping the site.
I did and that raw data query appears to be counting votes. Is there something odd about the SO schema that I don't understand?