Viktor
nixCraft 🐧The decline of Stackoverflow 😱
dbread@nixCraft who reads #stackoverflow the most?
Chattie
Erik Ellsinger@nixCraft so what data sources will future LLMs use? Will the knowledge base forever be stuck on 2025?
bazkie, an ongoing process@ellsinger @nixCraft the latter, I'm afraid
oddbyte
Vitlöksbjörn
Kaleissin@ellsinger @nixCraft Worse. "The AI doesn't know about this programming language/library/solution so I can't/won't use it." End of change, basically.
Dawid@ellsinger @nixCraft reddit, it's where a lot of traffic from SO has moved
ferricoxide@ellsinger @nixCraft
2015ish. That seems to be what all the search-engine hits point to …probably because SO's points-system seems to significantly hamper the ability to ask new questions or offer answers that are more-relevant to current usages.
2015ish. That seems to be what all the search-engine hits point to …probably because SO's points-system seems to significantly hamper the ability to ask new questions or offer answers that are more-relevant to current usages.
Gustavo@ellsinger @nixCraft As data sources for software, I hope will keep writing documentation. Which doesn't solve LLM issues in any way: after all this LLM paranoia I researched how they handled questions on a library I maintain and noticed they hallucinated A LOT of non-existent methods, so I updated the documentation to make it easier to use and even added some methods those LLMs hallucinated.
I just tried, it still sucks, unless you ask it to check the documentation it will try answering your questions only from the trained dataset: https://chatgpt.com/share/682b4abb-cb70-8004-b9d1-4a2f58108185
enoch_exe_inc@ellsinger @nixCraft Chinese ones, I guess.
Renato Ramonda@ellsinger @nixCraft hopefully, the shit peddlers will have run out of hype and money in a couple of years, and LLMs will remain a niche curiosity.
I'm almost scared about what absurd idiocy will be the NEXT hype train after crypto, NFT/Web3 and "AI". Probably Quantum Anything.
Urzl@nixCraft ChatGPT might be wrong but it doesn't flame me just for being new or wrong myself.
SrGesus
Wombatadon@nixCraft Stackoverflow was frequently wrong too.
Harry Karadimas@nixCraft Doomsday will be when it will be renamed "Stackunderflow" or even "Stackempty"
Daniel Evers
dodothedev🦤💻@nixCraft
Maybe if it weren't so hostile, it would have survived longer.
Maybe if it weren't so hostile, it would have survived longer.
@nixCraft
Been on a decline since 2017 — quite likely due to how much of a pain in the ass it is to ask (or otherwise participate) with the service. Most of the time that a web search provides a hit that links to Stack*, it's for answers that are elderly (typically in the 2013-2017 range of the graph's plateau). Given how quickly/frequently tech, and associated answers, change - in whole or in part (e.g., an answer perfect for Py27 may be considerably less-perfect for Py312) - the bars they've raised to asking or answering questions really reduces the site's utility.
Beyond that, and assuming that this is an actual case where correlation does equal causation, then, at best, LLM-provided answers have only increased the pace of decline rather than been its impetus.
Been on a decline since 2017 — quite likely due to how much of a pain in the ass it is to ask (or otherwise participate) with the service. Most of the time that a web search provides a hit that links to Stack*, it's for answers that are elderly (typically in the 2013-2017 range of the graph's plateau). Given how quickly/frequently tech, and associated answers, change - in whole or in part (e.g., an answer perfect for Py27 may be considerably less-perfect for Py312) - the bars they've raised to asking or answering questions really reduces the site's utility.
Beyond that, and assuming that this is an actual case where correlation does equal causation, then, at best, LLM-provided answers have only increased the pace of decline rather than been its impetus.
modestmax
John
CausticMango@nixCraft Fuck AI, but I won’t mourn the passing of Stackoverflow.
Søren
krubzor
dtomvan
NEO//LIX 🔊@nixCraft Which tells me that the vast majority of questions on StackOverflow were redundant and could have been be avoided if there the search would have actually worked.
Pete Alex Harris🦡🕸️🌲/∞🪐∫@nixCraft
Where's the point on the graph where stackoverflow started publicly embracing AI? May 2024, right?
Eyeballing it, it looks like the steep drop after that last upward bump.
Peter Krefting@nixCraft Everything of value has been asked 🙂
SO is still a treasure trove of well-curated content to find answers to problems, though. But you have to be careful to look at the dates (both to avoid old content, and to avoid AI slop).
Vurbi
DistroWatch@nixCraft I wonder what their traffic is like in comparison? This shows "questions asked" which makes me wonder if maybe most common questions are already asked and show up in search results? This would mean users don't need to ask questions, but might still be visiting the site for answers.
Neromabene@nixCraft Nothing lasts forever. Wonder what comes after LLMs.
greenpepper22@nixCraft didn’t some VC firm also buy StackOverflow a few years ago? I feel like that would also contribute to the decline
Davide
AnthonyCFoxIt is a knowledge base and the downward trend began 6 years ago.
Not to defend AI - boo ~ hiss ~ bad software!!!
I'm being sarcastic, but there are problems with it...
https://infosec.exchange/@AnthonyCFox/114529136371314287
AnthonyCFox (@AnthonyCFox@infosec.exchange)
Attached: 1 image We’ve been anthropomorphizing AI since the beginning. No one questioned it. We should now. It distorts reality. Opens us to manipulation. Shifts power. Erodes responsibility. Fuels hype and fear. And it shapes the AI itself—models tuned to sound human, even when hallucinating. This is a bug. Here's the whitepaper that details it: AI as Exploit → [https://dev.to/anthony\_fox\_aabf9d00159f3/ai-as-exploit-the-weaponization-of-perception-and-authority-1d3k](https://dev.to/anthony_fox_aabf9d00159f3/ai-as-exploit-the-weaponization-of-perception-and-authority-1d3k) @dyn