This might not be entirely true. People were able to fetch urls and data from the internet that couldn't possibly be in any training dataset (in eg posted the same day for the first time).
This actually shows perfectly how convincing ChatGTP can be. The author is too eager to believe the output from ChatGTP, even though this should be a clear indicator something else is happening:
> But ChatGPT has rewritten my article in a different style, what a interesting finding…
The URL conveniently contains the topic of the blog-article, meaning ChatGTP can 'hallucinate' the contents, even without internet connection.
To show this, I tried to follow the same steps as @neonforge, but with a non-existing URL. Unfortunately `lynx` 'didn't work', but curl showed some interesting information.
`curl https://medium.com/@peter/learn-how-to-use-the-magic-of-terr...` returned some html. Instead of a 404 and a <title> element containing just 'Medium' it hallucinated the following title tag: `<title>Learn How to Use the Magic of Terraform Locals (Step-by-Step Guide) | by Peter | Medium</title>`
I'm not saying that ChatGTP definitely has no access to internet, but so far I have not seen any proof or indication that it does.
I don't think this shows at all that ChatGPT can access the internet? It looks like ChatGPT was given a URL with a title and hallucinated a blog post based on that title. The author then brushes it off as 'written in a different style', when it just looks like a totally different article overall (or maybe it's testament to the fact that too many Medium articles are low quality and indistinguishable from the output of a LLM!)
If anything, this blog post is a perfect example about how people can put in whatever they want as an input and take the output as truth, without any rigorous approach about what would count as a true fact about the model.
That article is bunkum. There’s not the slightest reason to believe that it “rewrote his article in a different style”: it hallucinated a completely different article based on the URL. The other “proof” is all clearly wildly inconsistent and hallucinated: it’s a desktop machine, a server, inside Google in I think France, in the USA, in AWS, in Germany, and I think the last couple of “tests” don’t make sense anyway.
It convinced him of what he wanted to believe. He didn’t do a very good job of testing.
> But ChatGPT has rewritten my article in a different style, what a interesting finding…
The URL conveniently contains the topic of the blog-article, meaning ChatGTP can 'hallucinate' the contents, even without internet connection.
To show this, I tried to follow the same steps as @neonforge, but with a non-existing URL. Unfortunately `lynx` 'didn't work', but curl showed some interesting information.
`curl https://medium.com/@peter/learn-how-to-use-the-magic-of-terr...` returned some html. Instead of a 404 and a <title> element containing just 'Medium' it hallucinated the following title tag: `<title>Learn How to Use the Magic of Terraform Locals (Step-by-Step Guide) | by Peter | Medium</title>`
I'm not saying that ChatGTP definitely has no access to internet, but so far I have not seen any proof or indication that it does.