ChatGPT has been taking the world by storm this past week. I've experimented with using it for everything from re-wording and simplifying long blog posts to crafting shorter mastodon posts (follow me at indieweb.social/@thomasreggi). It's been a great help in making my text more coherent and easy to understand. I've also, like many programmers, used it to generate code snippets and even answer some coding questions.
Next, I tried using ChatGPT on stack overflow, pasting some of my questions directly into the chat to see if it could provide any useful answers. While it did provide some correct answers, it also produced some incorrect or non-functioning code, which wasn't immediately apparent until I tried running it.
As an example, I asked ChatGPT the question "Node path relative same directory missing './' prefix" and while I was checking the output, I received a similar answer from a user on stack overflow. The summary and code snippet of the answers are different, but the answers themselves are the same.
Could this all be just a coincidence? It's hard to say for sure. There has been a lot of talk about using software to detect generated text, but we may not be there yet. Today, StackOverflow announced that it will be banning answers generated by ChatGPT: Temporary policy: ChatGPT is banned.
According to StackOverflow, the main issue with ChatGPT is that although its answers have a high rate of being incorrect, they often appear to be correct and are very easy to produce. This makes it difficult for users to determine which answers are reliable and which are not.
Here are the two answers side by side:
My ChatGPT answer:
The user-contributed answer:
The post has since been deleted, and the user has been banned. I'm curious if this is because they had started to spam more and more of these incorrect answers throughout the site.
What do you think about ChatGPT? Has it helped you learn anything new, such as fears, dreams, hopes, or goals? Comment below!
Top comments (5)
Interesting to see how this evolves. I think we'll all be following closely.
I just tried ChatGPT-3 and it is really cool. Amazing really. dev.to/codingtomusic/playing-aroun...
Clearly agree that those answer shouldn't be posted. If as a user you want AI answer, ask the AI yourself, but if I'm asking on so, It's because i want answer from someone that did experiment with that subject.
Maybe after a bit of evolution, ai generated answer could be suggested when writing the question with a clear disclaimer of its provenance, but still not posted.
I still feel unreliable when it comes to non-general questions.
While the data used to train ChayGPT are outdated, it was able to understand why stack overflow could have banned it.
This is a funny yet amazing chat I made with it, regarding stack overflow ban.
wp.me/peqBxK-b