31–40 of 1001 posts
There are still airgapped places in the world where transferring information to offsite LLMs is expressly forbidden, but the offline LLMs available perform so terribly that they’re not worth using. An SO type application can be immensely helpful for engineering teams working in these environments.
And if you're making an internal-only site, it doesn't really need to be name-brand SO.
Not a big surprise once LLMs came along: stack overflow developed some pretty unpleasant traits over time. Everything from legitimate questions being closed for no good reason (or being labeled a duplicate even though they often weren’t), out of date answers that never get updated as tech changes, to a generally toxic and condescending culture amongst the top answerers. For all their flaws, LLMs are so much better.
Now with LLMs, I can't remember the last time I visited StackOverflow.
Only ever asked one question and I tried to answer more than a handful but never really clicked with the site.
I do wonder if it would have faired better under the original ownership before it was sold in 2021-06-02.
They will no doubt blame this on AI, somehow (ChatGPT release: late 2022, decline start: mid 2020), instead of the toxicity of the community and the site's goals of being a knowledgebase instead of a QA site despite the design. PS - This comment is closed as a [duplicate] of this comment: https://news.ycombinator.com/item?id=46482620
At the same time, I think there was another factor: at some point, the corpus of answered questions has grown to a point where you no longer needed to ask, because by default, Google would get you to the answer page. LLMs were just a cherry on top.
Looking back at my Stack Exchange/Stack Overflow (never really got the difference) history, my earlier, more general programming questions from when I just started are all no-brainers for any LLM.
Not a big surprise once LLMs came along: stack overflow developed some pretty unpleasant traits over time. Everything from legitimate questions being closed for no good reason (or being labeled a duplicate even though they often weren’t), out of date answers that never get updated as tech changes, to a generally toxic and condescending culture amongst the top answerers. For all their flaws, LLMs are so much better.
Not a big surprise once LLMs came along: stack overflow developed some pretty unpleasant traits over time. Everything from legitimate questions being closed for no good reason (or being labeled a duplicate even though they often weren’t), out of date answers that never get updated as tech changes, to a generally toxic and condescending culture amongst the top answerers. For all their flaws, LLMs are so much better.
Now that StackOverflow has been killed (in part) by LLMs, how will we train future models? Will public GitHub repos be enough? Precise troubleshooting data is getting rare, GitHub issues are the last place where it lives nowadays.
They would just use documentation. I know there is some synthesis they would lose in the training process but I’m often sending Claude through the context7 MCP to learn documentation for packages that didn’t exist, and it nearly always solves the problem for me.
Software quality is now generally a bit better than it was in 2010, but that need is ultimately still there.