7 ms·

No wonder. If I ask ChatGPT a question, it answers it. If I asked Stack Overflow a question, they made sure I wished I hadn't.

leetrout 3y ago
And people white knight for that behavior left and right. I refuse to interact with the platform after these encounters.
- hardware2win 3y ago
  How is it that I asked like 100 questions in last 5 years and hostile behaviour was visible in maybe 2 or 3, yet people act as if that was the most toxic place on the internet? All I was doing was just: putting effort into my questions
  - pbowyer 3y ago
    Probably because your brain works in the same way as the mods think, and so you ask questions that fit the expected format. Or your questions are focused with one right answer, because IME SO hates any question which assumes - unintentionally or intentionally - there are grey areas or more than one way to do it.
  - pasc1878 3y ago
    The answer here is that "putting effort into my questions" A lot of questions show people have not done that by either reading manuals or searching SO. Why should a answereer put effort into writing a good answer when the questioner has not shown any effort?
    - gilleain 3y ago
      A broader problem is that some people do not really want one answer to one question but basically some tech support to solve their problem. Either they then ask a series of separate questions (not always fun times) or try to engage in the comments (which then get deleted!).
- gilleain 3y ago
  I would not want to white knight for SO, but perhaps grey knight? My understanding of the reasoning behind some of the moderation - flawed as that often is - is to make the questions and answers 'general'. Which means a question that is very specific to one person may be voted to close or voted down. To take an example of another stack exchange site (chemistry.SE), they get a lot of 'homework questions'. Literally some questions are like "I have an exam in an hour PLEASE HELP". The stated policy is _not_ to just answer these questions, but to guide people towards a solution. Very frustrating for the person asking the question - who, you know, has an exam in an hour ... Ultimately there are policies and processes intended to improve the overall quality, and sometimes that is a good thing and sometimes it leads to everyone getting annoyed with the whole thing.
  - xnorswap 3y ago
    > a question that is very specific to one person may be voted to close or voted down That's the opposite of the guidance, or at least the guidance as it was back when I was active on SO. General questions "What's the best way to do X" are shut down. Questions should be specific and about actual problems you have rather than general problems. Solving real problems with real solutions builds up a strong data set, not questions about imagined problems and their answers
    - jalapenos 3y ago
      "The best way" type questions would still create a ballot with the best solution probably rising to the top, for use by others who encounter a similiar situation. People aren't always starting from a very specific problem, but from a vague general one. Ultimately though, Stack Overflow just faced the customer service curse: as volume rises, moderation gets blunter.
      - Arnt 3y ago
        SO doesn't have such ballots for a reason: SO doesn't want opinionated discussions. If you think that choice is wrong, SO is not going to be a good place for you. But there are very many other sites where you can have opinionated discussions. May I ask your SO id? It'll be a link like https://stackoverflow.com/users/1 https://stackoverflow.com/users/1, except with a different number.
znafelrif 3y ago
Doesn't ChatGPT probably regurgitate stackoverflow answers? If stackoverflow doesn't exist, ChatGPT probably also gets worse as it doesn't have the data necessary to be able to answer questions.
- jalapenos 3y ago
  It could interpolate from the original source - the full code of open source codebases and their docs. It'd only get worse in cases where only a human had access to the knowledge or the means of attaining that knowledge.
  - peteradio 3y ago
    Remove stackoverflow from the training and revise your assumptions. ChatGPT is not connecting the dots based on source code and docs.
- gmerc 3y ago
  With full access to github and github issues it does just fine with whatever knowledge it has
- moffkalast 3y ago
  You mean the data that Stackoverflow was given freely by millions of people? They have no more moral claim to it than your local park has to the content of conversations spoken within its limits. Besides, Github, Gitlab, Bitbucket, etc. have far more public data to train on, Stackoverflow only really provides the schema one would want for fine tuning to make sure the answers are in the right general question-code-answer-code format. Once that is trained properly (see: to the understanding level of compression, not just memorization) you only need to update the model on new libraries and languages, which can be trained from repos and docs directly.
  - altdataseller 3y ago
    They provided the platform and the medium by which questions get answered. That’s not free, and expecting to get it for free isn’t realistic
aleph_minus_one 3y ago
My experience differs: - Ask ChatGPT (or a similar AI chat bot) a question, it at best gives some at most semi-true answer. In the best situation, I can immediately see myself where the mistake in the answer is. - Look at Stack Overflow: very often find an answer - In the rare situation that I cannot find an answer on Stack Overflow, and ask a question: get a good answer (often many good answers) after a few hours that let(s) me realize that I have a lot to learn about the respective programming topic, which I will do.
- jalapenos 3y ago
  Pay for GPT 4 and use that, it's much better than 3.5
- JustLurking2022 3y ago
  Very often find an answer - from 10 years ago, and an entirely different version of the language/library - from more recently, but it's been closed as a duplicate - with a specific caveat, but that guy ignored and the question was also closed as a duplicate. - with a wrong answer, that you'd like to post a correct answer for, but you haven't karma farmed enough
Waterluvian 3y ago
I was a bit amazed to see this same antisocial behaviour in the gaming stack exchange. “Obviously you haven’t read the manual (for a 20+ year old game) so I’m closing this question.” Is there something about stackexchange format that accidentally encourages this? I go on various discords (learn piano, javascript, typescript, various games) and the absolute worst treatment I got was nobody replying.
- Arnt 3y ago
  I don't think it's accidental. The format is strongly optimised for producing question/answer pairs that are later searchable by people who have the same-ish problem. Among other things, this means getting rid of unanswered questions that would otherwise clutter the search results.
  - SAI_Peregrinus 3y ago
    Yep. SO wants to be a wiki, but in Q&A format. Its rules are designed not primarily to be a Q&A site, but rather a searchable knowledge base. But the interface is largely designed around making it look like a Q&A site, and this conflicting purpose causes a lot of the negative interactions (IMO).
altdataseller 3y ago
While I’ve had SO mods criticize my questions, I’ve more success getting correct answers to questions from SO than ChatGPT. With SO, I also have the benefit of knowing answers have been mostly vetted by others. If it’s wrong, someone will comment on it usually. Not so with ChatGPT
- jalapenos 3y ago
  GPT 3.5 or 4?
  - beepbooptheory 3y ago
    You know, I haven't heard anybody talk about this yet! I was wondering, is gpt4 better than 3.5, or worse?
    - Tiberium 3y ago
      It's miles ahead of 3.5, the difference is really really big. Mainly because GPT-4 is just so much better at "understanding" and logic, it follows everything you ask for or say very, very closely.
      - beepbooptheory 3y ago
        Incredible! I am sure its amazing to use. That said, I doubt it could, say, answer questions from an LSAT exam and get them right though. That would just be way too unbelievable to me.
        Tiberium 3y ago
        I'm not sure if you're joking or not, because OpenAI actually used LSAT as one of the examples where GPT-4 is much better - https://openai.com/research/gpt-4 https://openai.com/research/gpt-4 https://files.catbox.moe/l5n5gd.jpg https://files.catbox.moe/l5n5gd.jpg for some of the results in their technical report
qwertox 3y ago
> they made sure I wished I hadn't. Which says more about your style of asking than about the SO community.
- wink 3y ago
  Not really. There are dozens of reasons why you would want to do some thing the non-standard way and not "the one recommended" way, but it's mostly impossible to ask that on SO because 3 people will jump in to tell you "no you shouldn't". You could try a "I know I should be doing X but I want/need Y" disclaimer (also often useless) or spend writing half a novel, to only get told "can you make that a shorter repro?" or back to square 1, "you shouldn't do that'.
- allarm 3y ago
  That says more about the fact you don’t have to bother about the style when you’re asking ChatGPT.
  - qwertox 3y ago
    Because it is not resource-constrained.
throw1234651234 3y ago
This is absolutely false. Yes, there is a bit of a learning curve on how to phrase questions on stack overflow, but it mainly comes down to "ask a clear question". SO certainly still has better answers than ChatGPT, and I say this as someone who has been talked-down to on SO before. I understand where they are coming from. At the end of the day, you get nice, clear, tagged questions with proven out answers. Way more than you can say for ChatGPT, which generally gives random trash.
dep_b 3y ago
The odd thing is that I really tried to help people for a while on Stack Overflow but if you answer somebody's question you rarely get an accepted answer or even any reply at all. "Sorry buddy, that didn't work" or anything. Lots of people just dump and abandon their questions. Most of my reputation gained there is because I answered my own answer and ironically - just once - copied a solution from a thread on GitHub and pasted it there.
- xnorswap 3y ago
  Almost all my reputation is from a question [1] that I just happened to see and answer before anyone could close it as "bad quality" or "duplicate". Had I been a minute or two later I suspect it would have been closed as poor quality. I only spotted it by pure chance because I had just wondered the same thing before having a "duh" moment when I realised it was a really simple solution to what sounds complicated (but isn't). So I was able to interpret the "real" question and answer it. But, as you say, never a bit of feedback from the question asker. No acceptance, just abandonment. It just slowly gathers karma, around 10 per month for ~10 years. [1] https://stackoverflow.com/questions/27607516/sql-join-with-composite-primary-key/27607549#27607549 https://stackoverflow.com/questions/27607516/sql-join-with-c...

leetrout 3y ago

And people white knight for that behavior left and right. I refuse to interact with the platform after these encounters.

hardware2win 3y ago
How is it that I asked like 100 questions in last 5 years and hostile behaviour was visible in maybe 2 or 3, yet people act as if that was the most toxic place on the internet? All I was doing was just: putting effort into my questions
- pbowyer 3y ago
  Probably because your brain works in the same way as the mods think, and so you ask questions that fit the expected format. Or your questions are focused with one right answer, because IME SO hates any question which assumes - unintentionally or intentionally - there are grey areas or more than one way to do it.
- pasc1878 3y ago
  The answer here is that "putting effort into my questions" A lot of questions show people have not done that by either reading manuals or searching SO. Why should a answereer put effort into writing a good answer when the questioner has not shown any effort?
  - gilleain 3y ago
    A broader problem is that some people do not really want one answer to one question but basically some tech support to solve their problem. Either they then ask a series of separate questions (not always fun times) or try to engage in the comments (which then get deleted!).
gilleain 3y ago
I would not want to white knight for SO, but perhaps grey knight? My understanding of the reasoning behind some of the moderation - flawed as that often is - is to make the questions and answers 'general'. Which means a question that is very specific to one person may be voted to close or voted down. To take an example of another stack exchange site (chemistry.SE), they get a lot of 'homework questions'. Literally some questions are like "I have an exam in an hour PLEASE HELP". The stated policy is _not_ to just answer these questions, but to guide people towards a solution. Very frustrating for the person asking the question - who, you know, has an exam in an hour ... Ultimately there are policies and processes intended to improve the overall quality, and sometimes that is a good thing and sometimes it leads to everyone getting annoyed with the whole thing.
- xnorswap 3y ago
  > a question that is very specific to one person may be voted to close or voted down That's the opposite of the guidance, or at least the guidance as it was back when I was active on SO. General questions "What's the best way to do X" are shut down. Questions should be specific and about actual problems you have rather than general problems. Solving real problems with real solutions builds up a strong data set, not questions about imagined problems and their answers
  - jalapenos 3y ago
    "The best way" type questions would still create a ballot with the best solution probably rising to the top, for use by others who encounter a similiar situation. People aren't always starting from a very specific problem, but from a vague general one. Ultimately though, Stack Overflow just faced the customer service curse: as volume rises, moderation gets blunter.
    - Arnt 3y ago
      SO doesn't have such ballots for a reason: SO doesn't want opinionated discussions. If you think that choice is wrong, SO is not going to be a good place for you. But there are very many other sites where you can have opinionated discussions. May I ask your SO id? It'll be a link like https://stackoverflow.com/users/1 https://stackoverflow.com/users/1, except with a different number.

znafelrif 3y ago

Doesn't ChatGPT probably regurgitate stackoverflow answers? If stackoverflow doesn't exist, ChatGPT probably also gets worse as it doesn't have the data necessary to be able to answer questions.

jalapenos 3y ago
It could interpolate from the original source - the full code of open source codebases and their docs. It'd only get worse in cases where only a human had access to the knowledge or the means of attaining that knowledge.
- peteradio 3y ago
  Remove stackoverflow from the training and revise your assumptions. ChatGPT is not connecting the dots based on source code and docs.
gmerc 3y ago
With full access to github and github issues it does just fine with whatever knowledge it has
moffkalast 3y ago
You mean the data that Stackoverflow was given freely by millions of people? They have no more moral claim to it than your local park has to the content of conversations spoken within its limits. Besides, Github, Gitlab, Bitbucket, etc. have far more public data to train on, Stackoverflow only really provides the schema one would want for fine tuning to make sure the answers are in the right general question-code-answer-code format. Once that is trained properly (see: to the understanding level of compression, not just memorization) you only need to update the model on new libraries and languages, which can be trained from repos and docs directly.
- altdataseller 3y ago
  They provided the platform and the medium by which questions get answered. That’s not free, and expecting to get it for free isn’t realistic

aleph_minus_one 3y ago

My experience differs: - Ask ChatGPT (or a similar AI chat bot) a question, it at best gives some at most semi-true answer. In the best situation, I can immediately see myself where the mistake in the answer is. - Look at Stack Overflow: very often find an answer - In the rare situation that I cannot find an answer on Stack Overflow, and ask a question: get a good answer (often many good answers) after a few hours that let(s) me realize that I have a lot to learn about the respective programming topic, which I will do.

jalapenos 3y ago
Pay for GPT 4 and use that, it's much better than 3.5
JustLurking2022 3y ago
Very often find an answer - from 10 years ago, and an entirely different version of the language/library - from more recently, but it's been closed as a duplicate - with a specific caveat, but that guy ignored and the question was also closed as a duplicate. - with a wrong answer, that you'd like to post a correct answer for, but you haven't karma farmed enough

Waterluvian 3y ago

I was a bit amazed to see this same antisocial behaviour in the gaming stack exchange. “Obviously you haven’t read the manual (for a 20+ year old game) so I’m closing this question.” Is there something about stackexchange format that accidentally encourages this? I go on various discords (learn piano, javascript, typescript, various games) and the absolute worst treatment I got was nobody replying.

Arnt 3y ago
I don't think it's accidental. The format is strongly optimised for producing question/answer pairs that are later searchable by people who have the same-ish problem. Among other things, this means getting rid of unanswered questions that would otherwise clutter the search results.
- SAI_Peregrinus 3y ago
  Yep. SO wants to be a wiki, but in Q&A format. Its rules are designed not primarily to be a Q&A site, but rather a searchable knowledge base. But the interface is largely designed around making it look like a Q&A site, and this conflicting purpose causes a lot of the negative interactions (IMO).

altdataseller 3y ago

While I’ve had SO mods criticize my questions, I’ve more success getting correct answers to questions from SO than ChatGPT. With SO, I also have the benefit of knowing answers have been mostly vetted by others. If it’s wrong, someone will comment on it usually. Not so with ChatGPT

jalapenos 3y ago
GPT 3.5 or 4?
- beepbooptheory 3y ago
  You know, I haven't heard anybody talk about this yet! I was wondering, is gpt4 better than 3.5, or worse?
  - Tiberium 3y ago
    It's miles ahead of 3.5, the difference is really really big. Mainly because GPT-4 is just so much better at "understanding" and logic, it follows everything you ask for or say very, very closely.
    - beepbooptheory 3y ago
      Incredible! I am sure its amazing to use. That said, I doubt it could, say, answer questions from an LSAT exam and get them right though. That would just be way too unbelievable to me.
      - Tiberium 3y ago
        I'm not sure if you're joking or not, because OpenAI actually used LSAT as one of the examples where GPT-4 is much better - https://openai.com/research/gpt-4 https://openai.com/research/gpt-4 https://files.catbox.moe/l5n5gd.jpg https://files.catbox.moe/l5n5gd.jpg for some of the results in their technical report

qwertox 3y ago

> they made sure I wished I hadn't. Which says more about your style of asking than about the SO community.

wink 3y ago
Not really. There are dozens of reasons why you would want to do some thing the non-standard way and not "the one recommended" way, but it's mostly impossible to ask that on SO because 3 people will jump in to tell you "no you shouldn't". You could try a "I know I should be doing X but I want/need Y" disclaimer (also often useless) or spend writing half a novel, to only get told "can you make that a shorter repro?" or back to square 1, "you shouldn't do that'.
allarm 3y ago
That says more about the fact you don’t have to bother about the style when you’re asking ChatGPT.
- qwertox 3y ago
  Because it is not resource-constrained.

throw1234651234 3y ago

This is absolutely false. Yes, there is a bit of a learning curve on how to phrase questions on stack overflow, but it mainly comes down to "ask a clear question". SO certainly still has better answers than ChatGPT, and I say this as someone who has been talked-down to on SO before. I understand where they are coming from. At the end of the day, you get nice, clear, tagged questions with proven out answers. Way more than you can say for ChatGPT, which generally gives random trash.

dep_b 3y ago

The odd thing is that I really tried to help people for a while on Stack Overflow but if you answer somebody's question you rarely get an accepted answer or even any reply at all. "Sorry buddy, that didn't work" or anything. Lots of people just dump and abandon their questions. Most of my reputation gained there is because I answered my own answer and ironically - just once - copied a solution from a thread on GitHub and pasted it there.

xnorswap 3y ago
Almost all my reputation is from a question [1] that I just happened to see and answer before anyone could close it as "bad quality" or "duplicate". Had I been a minute or two later I suspect it would have been closed as poor quality. I only spotted it by pure chance because I had just wondered the same thing before having a "duh" moment when I realised it was a really simple solution to what sounds complicated (but isn't). So I was able to interpret the "real" question and answer it. But, as you say, never a bit of feedback from the question asker. No acceptance, just abandonment. It just slowly gathers karma, around 10 per month for ~10 years. [1] https://stackoverflow.com/questions/27607516/sql-join-with-composite-primary-key/27607549#27607549 https://stackoverflow.com/questions/27607516/sql-join-with-c...