YSK: In addition to Reddit, StackOverflow is on Strike
meta.stackoverflow.com
Why you should know: StackOverflow is facing a mod strike in a similar way as Reddit's mod strike. They are doing this in response to StackOverflow's failure to address it's promises and provide moderation tools
You are viewing a single comment
Unpopular opinion: for a beginner, ChatGPT gives way better answers than stackoverflow users. The advantage of ChatGPT is that I can command it to dumb it down. Stackoverflow users are used to answer in a language that resembles the language in documentations. They are dry, abstract, lack good examples to the point that the "foobar" shit triggers an immediate defensive reaction in my brain and are phrased for people who already understood a concept but need to refresh their knowledge. Their core problem, as is tradition in any IT field, is that they lack the empathy to understand the viewpoint of someone who understands less of something than they do. It's like asking someone to teach you reading and getting a poem with the advice to just read it as an answer.
I can circumvent that via ChatGPT by asking it to ELI5. Also, I get an answer instantly, am not discouraged to ask further questions and not advised to read a link where a solution is offered in an equally difficult language.
People are saying that using ChatGPT doesn't give accurate information and fails to convey important concepts, but I feel it's actually the other way around. Since there is ChatGPT, I'm making way more progress than before.
I understand that users don't want AI answers, but I also don't get why anyone would want that on this platform. You can just, you know, use AI directly.
That's not unpopular. But there is a problem. ChatGPT can answer your questions mostly because it was trained on the posts and answers of sites like StackOverflow.
If people abandon SO and similar forums then the quality of ChatGPTs answers will go down too.
Especially with something like programming. It's always changing. Next year there will be new versions of C++ and python. There will be new JS frameworks as always. It doesn't stand still.
And without new discussions about new problems, there's nowhere for ChatGPT to learn about them.
Haven't though about that, you're right.
I wish my students would care about the concepts and try to understand the answers instead of just blindly copying and pasting ill-fitting code (and then wondering why it only kinda works...).
As a former student now practicing engineer this habit never gets broken. All of us accept cargo cult computing to one extent or another. It sucks.
Usually the engineers with the least tolerance for it do better but only in the long run. In the short run they are yelled at for holding back projects.
I think that one issue with using AI to help you solve programming problems is that sometimes it will wholesale make things up. Of course, people can do that too, which is why communities of coders can vote on the best answer. I say, more power to you, using the tools that work for you. Just be cautious.
I think that one issue with using AI to help you solve programming problems is that sometimes it will wholesale make things up. Of course, people can do that too, which is why communities of coders can vote on the best answer. I say, more power to you, using the tools that work for you. Just be cautious.
ChatGPT is incredible for middle ground developers like myself. I understand the goal I'm trying to achieve, and I understand the general process of how to do it. I can ask very granular, specific questions to ChatGPT and it will spit out some code that will get me close to what I need.
If I was a complete novice, I think ChatGPT would make me too dependent on using it for answers.
That seems like a totally valid use case. I occasionally will outline some very specific requirements and have AI generate the code, which just saves a lot of time typing, versus it generating it entirely on its own. And I still go through all the code and verify that it's good. It's just a tool that can be used to make your job easier.
Totally. The other day I had to test a csv/xls upload tool. I wanted to make sure that no matter what configuration an asshole user had for phone numbers, it would strip everything out so it would be a valid integer for my database.
I told chatgpt to make me a csv with 20 rows, 6 columns with xyz headers, and to give me an assortment of different phone number formats. Took 10 seconds.
You're storing phone numbers as integers?
Here to echo the same. I thought using AI to assist me in coding would just make me lazy and learn nothing, but turns out I actually learn more than ever since it's much faster, more polite and patient, and the semantics are usually more catered to my needs and self explanatory than the average answers I find elsewhere.
It's great for writing snippets and creating basic frameworks. However, it definitely makes a lot of mistakes which I doubt a total beginner can spot, especially if the error lies in logic and not syntax.
Works great only as a tool for now, but chances are AI will probably surpass human coders sooner than we think.
I played around with ChatGPT for programming for a few hours a while back.
It is far better at explaining code in plain language than pretty much any human I've seen, atleast online. It's absolute dogshit st writing anything but the most basic of code, but it does do a good job explaining.
Programmers are shit at communicating.
I've found that it gives me a decent skeleton of something that I can then apply to my actual problem, but not much more, and it usually comes with some pretty big mistakes. I was trying to learn Z80 assembly and it gave me a good idea of how my code should generally look, but I did end up having to rewrite a whole bunch of it before I could actually execute anything.
I think there's a sweet spot for how many other resources are out there. JavaScript GPT answers are pretty good. But when you get to a less popular language like Elixir not so much
I've yet to get a useful answer out of chatgpt for a technical question. It's good for fluffing up emails, but I haven't been super impressed with any use case I've tried for it.
When I’ve used it for decently complex programming questions I’ve found it often likes to make up functions and libraries. It’ll be like just use this reasonable sounding function from this library, and I look it up and the library does not have that functionality at all. Over and over!
Well it's a large language model that generates text probabilistically. It's trained on vast amounts of data, so it's expert at sounding like a skilled programmer, but there's absolutely no reason at all for the results to be useful code. It will sound like useful code and look like useful code, and it will be on the right topic, and that might well be enough, but it might not.