301–310 of 1001 posts
Earlier quoted context omitted.
Models are check-pointed. You can save one you like and use it forever.
You can save an open source + open weights model, which is frozen in time. That’s still very useful for some things but lacks knowledge of current data. So we’ll end up with a choice of low-performing stale models or high-performing enshittified models which know about more current information.
My SO account is coming up to 17 years old and I have nearly 15,000 points, 15 gold badges, including 11 famous questions and similar famous answer badges, also 100 silver and 150 bronze. I spent far much time on that site in the early days, but through it, I also thoroughly enjoyed helping others. I also started to publish articles on CodeProject and it kicked off my long tech blogging “career”, and I still enjoy writing and sharing knowledge with others.
I have visited the site maybe once a year since 2017. It got to the point that trying to post questions was intolerable, since they always got closed. At this point I have given up on it as a resource, even though it helped me tremendously to both learn (to answer questions) and solve challenging problems, and get help for edge cases, especially on niche topics. For me it is a part of my legacy as a developer for over 30 years.
I find it deeply saddening to see what it has become. However I think Joel and his team can be proud of what they built and what they gave to the developer community for so many years.
As a side note it used to state that was in the top 2% of users on SO, but this metric seems to have been removed. Maybe it’s just because I’m on mobile that I can’t see it any more.
LLM’s can easily solve those easy problems that have high commonality across many codebases, but I am dubious that they will be able to solve the niche challenging problems that have not been solved before nor written about. I do wonder how those problems get solved in the future.
Some comments: - This is a really remarkable graph. I just didn't realize how thoroughly it was over for SO. It stuns me as much as when Encyclopædia Britannica stopped selling print versions a mere 9 years after the publication of Wikipedia, but at an even faster timescale. - I disagree with most comments that the brusque moderation is the cause of SO's problems, though it certainly didn't help. SO has had poor mode…
Earlier quoted context omitted.
You can save an open source + open weights model, which is frozen in time. That’s still very useful for some things but lacks knowledge of current data. So we’ll end up with a choice of low-performing stale models or high-performing enshittified models which know about more current information.
Open source models get updated all the time. You'd only be a few months behind.
Indirect pollution via AI slop in the input and the same content manipulation mechanisms as SEO hacking is still a threat for open models.
Some comments: - This is a really remarkable graph. I just didn't realize how thoroughly it was over for SO. It stuns me as much as when Encyclopædia Britannica stopped selling print versions a mere 9 years after the publication of Wikipedia, but at an even faster timescale. - I disagree with most comments that the brusque moderation is the cause of SO's problems, though it certainly didn't help. SO has had poor mode…
I read an interview once with one of the founders of SO. They said the main value stackoverflow provided wasn't to the person who asked the question. It was for the person who googled it later and found the answer. This is why all the moderation pushes toward deleting duplicates of questions, and having a single accepted answer. They were primarily trying to make google searches more effective for the broader internet. Not provide a service for the question-asker or answerer.
Sad now though, since LLMs have eaten this pie.
Many users left because they had had overly strict moderation for posting your questions. I have 6k reputation, multiple gold badges and I will remember StackOverflow as a hostile place to ask a questions, honestly. There were multiple occasions when they actually prevented me from asking, and it was hard to understand what exactly went wrong. To my understanding, I asked totally legit questions, but their asking pol…
I have around 2k points, not something to brag about, but probably more than most stackoverflow users. And I know what I am talking about given over a decade of experience in various tech stacks. But it requires 3,000 points to be able to cast a vote to reopen a question, many of which incorrectly marked as duplicate. I said to myself, let it die.
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.
The irony is that the LLMs are trained on stack overflow and should inherit a lot of those traits and errors.
Really, if we could apply some RLHF to the Stack Overflow community, it would be doing a lot better.