Well, I have all the sympathy for the SO people who are stressed out over losing their jobs... but at least a LLM doesn't chew me out when I ask a question that's remotely similar to one asked 7 years ago. And I like the low-latency, interactive rubber ducking experience way better than the slower turnaround on a SO question.
Seems poorly correlated.<p>It seems unlikely that any significant number of developers quit using SO in favor of blindly trusting ChatGPT.<p>More likely - advertising revenue declined in line with the reduced ad spending across the industry, largely due to expectations of economic recession by ad buyers.
<i>Stack Overflow is laying off another 28%</i> (<a href="https://news.ycombinator.com/item?id=37900651">https://news.ycombinator.com/item?id=37900651</a>) (187 points | 2 days ago | 214 comments)<p><i>Stack Overflow announces 28% headcount reduction</i> (<a href="https://news.ycombinator.com/item?id=37898199">https://news.ycombinator.com/item?id=37898199</a>) (118 points | 2 days ago | 171 comments)
I'm thinking that coding is not going to be the most interesting application of current LLMs for a while. Writing code is simultaneously the least time consuming part of product development, and also the most critical. So much of the work is done outside of the IDE, figuring out the problem, deciding on a solution, communicating the solution, etc. Once you do write the code, it's relatively quick and there are enough tools to speed up the process. For AI to truly replace engineers, it needs to be an AGI. I think Stack Overflow should be considering the community component, because as of right now, LLMs are more likely to disrupt creative fields, customer service, and middle management.
> While no chatbot is 100 percent reliable, code has the unique ability to be instantly verified by just testing it in your IDE, which makes it an ideal use case for chatbots.<p>Oh, this is so wrong. Code is not instantly verified by any IDE except for things like syntax errors and compiler warnings. SO critique comments under each answer are so more valuable in that sense, it's not even funny. Every time I give feedback on ChatGPT's proposed answers it tends to go into incredible rabbit holes and endless recursion, reminiscent of WOPR playing tic tac toe.
I used to relay on SO, still do sometimes but most problems seem dated. And with quality of Google search declining, getting to the core of a problem takes too long.<p>I used to relay on GitHub issues but github.com way too slow and quality of search is abysmal.<p>So more and more when in doubts I just read source code. Seems fastest.