The Uneven Decline of Collective Knowledge Production: Evidence from Stack Overflow After Generative AI
Source
Myokyung Han, Taegyoon Kim, Jinhyuk Yun, Lanu Kim
Author
Myokyung Han, Taegyoon Kim, Jinhyuk Yun, Lanu Kim
Date
Key takeaways · AI-distilled
The authors analyze more than two million Stack Overflow questions posted between 2020 and 2025, treating the release of ChatGPT-3.5 as a natural shock and tracking two dimensions: question difficulty and how much data exists for a topic.
The drop in easy questions was accompanied by rising code complexity in posted questions, which the authors treat as corroboration that the remaining traffic skews harder.
The two effects interact: the decline in easy questions is concentrated in data-rich domains, while difficult questions increased regardless of how much data a topic had.
The pattern extends beyond Python to other programming languages, and more prevalent languages show sharper shifts, according to the paper.
Why it matters
Public developer Q&A is shifting toward harder, data-scarce questions, which affects what future models and engineers can learn from shared knowledge. It tells you where AI assistants have the most public training material and where they have the least.
articleWhat Users Think of Generative AI: A Cross-Platform NLP Analysis of Trust and Friction in App Store ReviewsMd Jafrin Hossain, Umme Nusrat Jahan, Shouvaggo Sharif Shammo
articleRetrieval-Augmented Generation for Scientific Code UnderstandingAaron Nobile, Andreas Adelmann, Mohsen Sadr
articleHow Researchers Use and Verify AI Coding Assistants: Tasks and Validation Practices in Scientific ProgrammingGabrielle O'Brien, Reed Milewicz, Nasir Eisty