Changing Forecasts for Python Questions on Stack Overflow

Before StackOverflow pivoted towards "AI", community-moderating dupes was definitely one of the best features. Of course, it only works with humans acting like humans - judging by word similarity only will not suffice to bring the intended benefit for all users.

An incredible number of Python questions get marked as duplicates years after they've gained many answers and many user upvotes. Then when you look more closely you see that questions or answers are merely related, but not the same.

Unfortunately SO has incentivized marking-as-duplicate by a mob of people who are not skilled at discerning the distinctions between issues. The distinctions would have been clear to the original question askers and answers who devoted time to the topic but not to people taking a "mile high" view years later and with no "skin in the game". It is unfortunate because knowledge is being lost to incorrectly tuned gamification incentives.

https://imgur.com/a/M194hRc shows the number of total and undeleted questions on StackOverflow.

The sharp trend at the end of 2022 after ChatGPT was released is hard to miss.