Stack Overflow's Decline Threatens the AI Models It Once Trained
Stack Overflow has seen a sharp drop in question volume since ChatGPT launched in November 2022, leading to a 28% staff layoff in October 2023. Developers increasingly turn to AI tools like ChatGPT and Copilot for direct answers, bypassing the forum that once served as a primary resource. The concern extends beyond Stack Overflow itself: large language models were heavily trained on its high-quality, human-reviewed archive, and that archive is no longer growing at the same pace. Research published in Nature in 2024 by Shumailov and colleagues demonstrated that models trained on AI-generated content degrade over generations, a phenomenon called model collapse. As new human-contributed data dries up, future AI models risk becoming increasingly reliant on stale or synthetic information, potentially reducing their ability to handle novel or niche programming problems.
This is an AI-generated summary. ShortSingh links to the original source for the complete article.
Discussion (0)
Log in to join the discussion and vote.
Log in