4 ms·

> An answer on SO is either helpful or not. Who cares how it was written? "Who cares if the diagnosis is done by a medical expert or someone pulling out random

> An answer on SO is either helpful or not. Who cares how it was written? "Who cares if the diagnosis is done by a medical expert or someone pulling out random drugs they tried before? As long as I feel better immediately after taking them, who cares how they were prescribed?" The case on SO is clearly different, as ChatGPT might answer incorrectly or answer with something containing subtle bugs. There's also a good chance that you won't be immediately able to spot those bugs, as, if you were sufficiently knowledgeable in the topic yourself, you would have most likely not asked that question. The case for art is a bit different, as there is no technically correct way to do it, but there is still a value to the way it is created. Would you think the first picture drawn by your child is worth the same as any other bad painting? Would you agree that a perfect copy of the Mona Lisa has equal value to the actual object? If no, it should be pretty easy to see why a painting generated by an AI is different from one created by a human.

danenania 4y ago
"ChatGPT might answer incorrectly or answer with something containing subtle bugs. There's also a good chance that you won't be immediately able to spot those bugs, as, if you were sufficiently knowledgeable in the topic yourself, you would have most likely not asked that question." Playing devil's advocate here, but this also clearly applies to human answers.
- Sebb767 4y ago
  SOs answer quality was never perfect, but there are still a few differences: - Humans actually try to answer the question at hand, whereas ChatGPT is only auto-completing text. This might lead to similar results in a lot of cases, but the goal of ChatGPT is not to produce correct SO answers. - Related, humans will (usually) back off or correct answers if errors are pointed out to them. ChatGPT (at the moment) doesn't. - Humans will (usually) not attempt to answer questions they don't feel like they are qualified to answer. ChatGPT will. I'm not saying that SO is/was perfect as-is, but ChatGPT is - in my opinion - overall not likely to improve SOs quality as of now. Since it is currently banned on SO, it seems like this is not only my opinion.
- Beltalowda 4y ago
  It's a matter of scale and frequency; people were posting significant amount of low-quality nonsense generated by ChatGPT. The average quality of ChatGPT was (is) significantly lower than the average human answer and there were thousands of those answers. As a test I ran some questions through ChatGPT a few weeks ago; both popular ones and just some random ones from the homepage (including some not-very-good questions). In only one case would I say that it was "good enough" to post. All the rest ranged from "mostly correct but with huge omissions" to "used 5 paragraphs to explain what could be one or two sentences" to "this is not even remotely correct in any way, but at a glance it actually looks kinda correct". The "looks kinda correct" can be pretty misleading, because there have been a few answers where I went "wait, is this actually a thing? I didn't know about that!" and when investigating further it turned out it wasn't a thing at all and ChatGPT was just trolling me.

danenania 4y ago

"ChatGPT might answer incorrectly or answer with something containing subtle bugs. There's also a good chance that you won't be immediately able to spot those bugs, as, if you were sufficiently knowledgeable in the topic yourself, you would have most likely not asked that question." Playing devil's advocate here, but this also clearly applies to human answers.

Sebb767 4y ago
SOs answer quality was never perfect, but there are still a few differences: - Humans actually try to answer the question at hand, whereas ChatGPT is only auto-completing text. This might lead to similar results in a lot of cases, but the goal of ChatGPT is not to produce correct SO answers. - Related, humans will (usually) back off or correct answers if errors are pointed out to them. ChatGPT (at the moment) doesn't. - Humans will (usually) not attempt to answer questions they don't feel like they are qualified to answer. ChatGPT will. I'm not saying that SO is/was perfect as-is, but ChatGPT is - in my opinion - overall not likely to improve SOs quality as of now. Since it is currently banned on SO, it seems like this is not only my opinion.
Beltalowda 4y ago
It's a matter of scale and frequency; people were posting significant amount of low-quality nonsense generated by ChatGPT. The average quality of ChatGPT was (is) significantly lower than the average human answer and there were thousands of those answers. As a test I ran some questions through ChatGPT a few weeks ago; both popular ones and just some random ones from the homepage (including some not-very-good questions). In only one case would I say that it was "good enough" to post. All the rest ranged from "mostly correct but with huge omissions" to "used 5 paragraphs to explain what could be one or two sentences" to "this is not even remotely correct in any way, but at a glance it actually looks kinda correct". The "looks kinda correct" can be pretty misleading, because there have been a few answers where I went "wait, is this actually a thing? I didn't know about that!" and when investigating further it turned out it wasn't a thing at all and ChatGPT was just trolling me.