Comment by thephyber
4 days ago
Stackoverflow visits fell[1] off a cliff since GPT became popular.
Google is getting destroyed by the chatbot workflow because it is no longer the start of a browser session and clickthrus (the things that earn the high sponsored link rates) are falling as more users get their queries answered faster with less effort.
StackOverflow has been dying a slow death since longer than before ChatGPT. Sure, ChatGPT is helping to accelerate it. The real data (leave aside social/community for a moment) issue with SO.com: Many answers don't age well. So, you have an answer from 8 years ago with 65 upvotes, but now the lang/lib was updated in 2023. A newer, more relevant answer is waaaaaaay down and only has one upvote. Personal note: I still pine for the old days when Joel Spolsky and Jeff Atwood were at the helm. They really knew how to build and sustain a vibrant community.
Are they, though? Inaccurate info is pretty common from LLMs.
Inaccurate info exists everywhere. StackOverflow contains inaccurate, outdated, incomplete info. Caveat Emptor wherever you are.
LLMs are like a knife. It is a tool that can hurt you if you misuse it, but it also has the capability to save LOTS to time if you use it well.
A knife's function is deterministic. LLMs are not.
They routinely misinterpret the information they've ingested and confidently spit out incorrect statements. Worse - they confidently spit out incorrect statements in ways we cannot anticipate.
This isn't comparable to a person. This isn't comparable to human intelligence. This isn't a problem that can be handwaved away by saying "people are sometimes wrong too!"