Stack Overflow temporarily bans users from sharing responses generated by ChatGPT, pending a final ruling; the site's mods say most ChatGPT answers are wrong
Stack Overflow, the go-to question-and-answer site for coders and programmers, has temporarily banned users from sharing responses generated by AI chatbot ChatGPT.
The quality concern has since gained supporting evidence: a later review of 517 ChatGPT programming answers found incorrect information in 52% of them. Meanwhile, Stack Overflow’s own survey shows developer use of AI tools rising even as trust in their accuracy falls.
First-order effects
ChatGPT-generated answers are removed from Stack Overflow while the site determines a permanent policy, putting moderators and contributors on the front line of enforcement.
Developers seeking help on Stack Overflow must submit attributable, human-written responses rather than paste chatbot output into the public knowledge base.
Second-order effects
Stack Overflow’s moderation policy makes answer provenance a practical requirement, increasing friction for contributors who use ChatGPT while protecting readers from answers its moderators consider unreliable.
OpenAI’s coding assistant faces a visible quality benchmark in the developer community, reinforced by the later analysis finding misinformation in many programming responses.
Third-order effects
If developers continue adopting AI tools while trusting their answers less, Stack Overflow’s differentiator shifts toward vetted community knowledge rather than raw answer generation.
The site’s later combination of lower question volume and higher revenue from enterprise tools and AI licensing points to a potential decoupling of community activity from the value of its technical corpus.
The trend: Developer knowledge platforms are being pushed to distinguish verified, attributable expertise from AI-generated answers as chatbots become a routine coding tool.
Feedback effects: Google initially relied on the (human-curated) link structure of the internet to derive “importance” signals from the internet graph. As people started using Google instead of links, Google accidentally destroyed the fabric that provided the value.
HUGELY relevant to the debate on misinformation that I have been having with @ylecun. If StackOverflow can't keep up with plausible but incorrect information, what about social media and search engines? What's your plan here, @elonmusk? Cc @sama https://twitter.com/...
I searched for a React question just now, and after I got my answer I tried to get chatGPT to answer me. I wanted to know if the useEffect dependencies array could itself vary in length. Turns out it can't and it will log a warning if you do it. chatGPT kept saying you CAN do it …
I wonder if this is not an issue with the input rather than the output? Even with two humans, wrong question will lead to wrong answer. I ask this, because out of curiosity I had ChatGPT do the first 20 challenges on leetcode and it was successful 100% of the time. https://twitte…
ChatGPT is so much faster and more helpful for figuring out coding stuff than Googling/Stack Overflow. I now always have it open when coding. I do wish it gave some reference URLs to see where the info is coming from so I can verify accuracy.
Wild that an AI is already disrupting an industry (in an incredibly minor way, but still...) Also fascinating that it's providing answers which look right on first glance but which aren't. Going to be a crazy week for AI news methinks. https://twitter.com/...
Use of ChatGPT is now temporarily banned on StackOverflow, too many people use it to generate answers that look correct. https://meta.stackoverflow.com/ ... https://twitter.com/...
We need a term for something beyond “tragedy of the commons” when the extractive platforms try to kill off the sources they depend on to feed their models in the first place. The same thing the gig economy tries to do to workers, and equally short-sighted. https://twitter.com/...
I'm surprised that folks aren't thinking similar things about the consequences of using Copilot (beyond license violations). https://meta.stackoverflow.com/ ...
it's extremely interesting news imo, because a) it's a rare example of concrete harms of large language models in the wild, and b) suggests that some web platforms don't have the tools needed to protect against a flood of AI-generated content https://www.theverge.com/... https://…
StackOverflow has temporarily banned users from posting AI-generated responses from ChatGPT, with mods saying the volume of incorrect but plausible-looking replies was just too great for them to deal with. Details here: https://www.theverge.com/...
it seems very possible that we are now exiting the brief window where a good fraction of all of human knowledge was searchable & instantly available. a window that starts with the invention of the search engine & ends with the invention of large language models.
“People posting GPT answers to stackoverflow to scoop the points without knowing if the answer is actually correct” is one of those problems that's too obvious to even foresee https://meta.stackoverflow.com/ ...
I don't quite understand why people put someone else's question in chatgpt and post the answer to Stack overflow. Do they gain anything from that? Also I don't see how SO is going to be able to enforce that policy. Which is made the real use case for the Tiring test.. https://twi…
ChatGPT, or alternatively known as the Twitter Timeline DoS (TTD), is banned from StackOverflow. More details available here: https://meta.stackoverflow.com/ ...