The takeaway should be that StackOverflow traffic was already down 50% from its peak before ChatGPT launched. While AI was probably a factor in its demise, the site was already on life support.

So, ChatGPT-like tools are accelerating(/or locking in) the decline of community sites like Stack Overflow, which used to be an indispensable tool for both junior developers to troubleshoot issues and also technology builders to see where users were hitting pain points with your tech (so they could improve it).

A couple of major ramifications from its decline:

#1 (the bigger one): The decline of Stack Overflow-like sites will (imo) degrade or cap the quality of ChatGPT tools themselves on questions pertaining to code post-2022. I doubt that advances like "reasoning" or other AI breakthroughs are going to fully make up for the oncoming draught of quality training data. Sites like SO were a crutch that companies who underinvested in documentation leaned on (their attitude essentially being: "we've done enough, let the coders figure it out amongst themselves, b/c it works"). I doubt companies are going to suddenly realize they need to invest more in solid docs (for both the developers and AI companies). -- While many initially saw this coming (AI killing the web it trained on), now we have pretty dramatic data show that has happened.

#2: questions about new technologies and their shortcomings will be asked in the dark, giving AI companies valuable data that used to otherwise exist in public forums. Among other things, this will make it harder for tech-builders to know what to improve, therefore preventing it from improving as quickly, and keeping people more reliant on AI tools for troubleshooting. This seems to be another example of AI companies _creating_ problems that they are best positioned to "solve".

10 days ago: https://news.ycombinator.com/item?id=43999125