StackOverflow, the de facto standard when it comes to asking and looking up answers to coding questions (and one of the big reasons why AIs are so good at answering your coding questions), is faltering.
Since ChatGPT launced: Nov 2022 (108,563), it's had 82,997 less questions (3.25x less; -76.5%).
Questions are down by more than three-quarters – that’s bad news not just for StackOverflow, but for the makers of LLMs, as it means there is just no new material to train AIs on. Now, most public AI models like ChatGPT or Claude are continuously trained on their users' questions and the answers the AI gives, but StackOverflow provided something much more important: answers. And not just any answers, but answers ranked by their users based on how accurate and good they were.
I guess OpenAI, Anthropic, Google, et al. will have to get creative…
