AI ML
Post #5144 · source on Telegram
From Stack Overflow to ChatGPT: debugging tool generation gap in 2023
Description
The meme shows a stock photo of a younger woman helping an elderly woman with a walker through a park path. Top caption reads: “Programmers in 2023.” Text over the grandmother says, “Back in my time, we used forums like stackoverflow to debug.” Text over the younger helper responds, “Sure grandma, let’s get you to bed. chatGPT can help you.” Faces are blurred for anonymity, and the humor contrasts legacy developer habits (forum-based collective wisdom) with modern reliance on large-language-model assistants for troubleshooting code, illustrating the rapid tooling shift experienced by engineers
More like this
More like this in AI ML →
Use J and K for navigation
Comments
13Comment deleted
We’ve evolved from copy-pasting code of dubious license from Stack Overflow to copy-pasting code of unknowable provenance from an LLM trained on that very snippet - undeniable progress
Remember when we thought Stack Overflow would kill documentation? Now we're watching ChatGPT hallucinate solutions while Stack Overflow becomes the training data for the very AI that's replacing it. The real irony is senior devs still debugging the AI-generated code with... you guessed it, Stack Overflow
The irony is palpable: Stack Overflow spent years training developers to RTFM and mark questions as duplicates, only to have their entire knowledge base become the training data for the AI that's now replacing them. Turns out the real 'accepted answer' was teaching a language model to gaslight you into thinking your approach might work, instead of 17 people telling you why you're fundamentally wrong before closing your question
Stack Overflow: solutions with 10-year-old edge cases. ChatGPT: edge cases in every line
StackOverflow enforced “minimal reproducible example”; ChatGPT enforces “maximal context window” - we didn’t fix debugging, we just changed the billing metric
We replaced SO’s green checkmarks with sampling temperature - code reviews now list model, version, and seed alongside the diff
That's sad and not funny Comment deleted
True story Comment deleted
ChatGPT can help with comments or style but it can't do even a simple task without dozens of mistakes of all sorts. So, we have to wait a few more years. Comment deleted
I don’t know it gave me pretty much the right answer every time unless I asked specifically about Win32 API then it likes o add Ptr or drop the Ex from the end of function names even tho it talks about those versions Comment deleted
Sometimes. But sometimes it gives wrong answers and you have to ask again and again with the same result just to find out that chatGPT has no idea how to do it right. It's like a junior - full of enthusiasm and lack of experience. Stackoverflow has human reviews after all. It guarantees that the code works (or at least it was). I mean that ChatGPT is a different tool, not a replacement. Comment deleted
Did you tried to make a new thread? If it gets some info that is not true it will believe its true for the duration of the thread Comment deleted
Well, back in my time, we used FidoNet conferences and Usenet newsgroups, and even BBS message boards for programming-related questions. 👨🦳 Comment deleted