The original headline is 11 words: "Stack Overflow's decline threatens the quality of future AI training data"

Short summary

Stack Overflow's question volume has declined sharply since ChatGPT's launch, creating a dangerous feedback loop: AI tools trained on Stack Overflow data are causing the platform to empty out, which means future models will train on increasingly stale corpora. Research on model collapse (Shumailov et al., Nature 2024) shows that training on AI-generated output degrades quality over generations. The 'archive gap' is already visible—models handle common pre-2022 questions well but hallucinate on newer libraries and recent framework versions.

• •Stack Overflow traffic and new questions have plummeted since ChatGPT's November 2022 launch, with layoffs following in 2023
• •AI models trained on Stack Overflow's archive face model collapse as human-generated content dries up, per Shumailov et al. research
• •The 'archive gap' means models already struggle with post-2022 libraries and frameworks, and this will worsen as the human pipeline shrinks

Generated with AI, which can make mistakes.