Are you confident that its responses are accurate? I’ve had it just make up facts or tack made-up facts onto actual facts. It will also fabricate references to research papers.
I’ve have yet to experience any of the domain knowledge I have being incorrect.
Similarly, today I saw this [0] comment and in an attempt to help the user went to ChatGPT.
Prompt:
> User I remember in the 2000s, there was some site using a decimal IP address (as a single number, not dotted quad) that had hacking/crypto puzzles. Something with a "Alice in Wonderland" theme. Does that ring a bell for you?
Response:
> I'm sorry, but as an AI language model, I don't have personal experiences or memories. However, based on your description, it's possible that you might be referring to the website known as "Project Wonderland." It was a website that featured a series of cryptographic puzzles based on the story of Alice in Wonderland, and it used a decimal IP address (203.208.46.147) as its URL. The website was active in the early 2000s and gained a cult following among enthusiasts of cryptography and puzzle-solving. However, the website is no longer active today.
I got really excited to have found an answer until through Google and the Wayback Machine I realised that ChatGPT just made this all up and no such website existed at that URL.
I tried starting a new chat with ChatGPT to ask it about this "Project Wonderland" website and it had no idea what I was talking about.
[0]: https://news.ycombinator.com/item?id=35748714
(I am using ChatGPT 3.5 (March 23, 2023))
The server might even have refused connection if the HTTP "Host: " header wasn't in that format, but as a human, rather than a language model, I'm not sure about that and might be misremembering :)
* 3301 is one of three numbers that had to be added to get the .com url * The Wikipedia page cites someone close to the group with "follow the white rabbit" * Years don't quite match up but given that you only asked if it rang a bell, that is fair enough
What I got back was confidently incorrect, and gave me examples that were neither of the two. (Interestingly, most of the examples were chiasmus instead.) I tried correcting it to get better responses, but it kept apologizing and then giving me the incorrect examples back.
But then I said No, I'm talking about JR in babylon 5
The response was Sheridan was never refereed as JR in Babylon 5. There was no JR in Babylon 5. But because I asked, ChatGPT happily made up things.