Stack Overflow questions per year

As AI papers have multiplied on arXiv, Stack Overflow questions have declined, a negative correlation of -0.978 that documents the exact moment AI began eating its parent: the research that promised to make machines intelligent also made the human Q&A platform less necessary. The paper describes the model, the model answers the question, and Stack Overflow loses the traffic. The research creates the technology that destroys the community.

AI papers grew from about 5,000 to over 40,000 per year between 2015 and 2023. Stack Overflow questions declined from about 12 million to about 5 million. The correlation contains a real causal mechanism: AI code assistants (Copilot, ChatGPT) built on the research published in those papers directly reduced the need for human-to-human programming help on Stack Overflow. The papers created the models, the models answered the questions, and the platform declined.

Nine years of more AI papers and fewer Stack Overflow questions is one of the most genuinely causal correlations on this site: the research published the models, the models displaced the platform, and the chart documents the technological substitution in real time. The paper is published, the assistant answers, and the human expert is no longer needed. The stack overflows. The questions do not.