403 for me - which makes me wonder how anyone else is commenting on the actual content of the link, rather than just recycling general comments. without knowing the details.
Plenty of progress in models that can use tools and search. Would love to see how one of these tool/search-enabled models do at this kind of a task. In my experience, they don't fabricate things anymore, just sometimes occasionally misrepresent the content of citations (put a citation somewhere where it doesn't actually support what is written).
A few days ago I asked GPT 5 for links to news on the Charlotte murder before the story got reported by the mainstream media. It gave me five different links, including AP and Reuters. Every one, five out of five, was a hallucination.
It hallucinated complete documentation to the tech we asked it about just 2 weeks ago. Completely made up documentation with only vague relationship.to how it really works.
I asked GPT-5 for updated literature survey for a paper I was writing with search enabled and explicit asked to use google scholar arxiv etc and yet most papers were non existent and in some cases even pointed to some GitHub repos which were private.
"Appellants counsel has acknowledged that his briefs are replete with fabricated legal authority, which he admits resulted from his reliance on generative AI sources such as ChatGPT, Claude, Gemini, and Grok. Counsel says that he was not previously aware of the problem of AI hallucinations, but he has educated himself about the issue since receiving the OSC."
