[hellotheretoday]

A lot of times when I ask chatgpt for references on something it sends me dead links or links to something that do not have anything to do with what it claims to be citing. I point that out and then it sends me more of the same. I would not be surprised if they were already working on this to improve that aspect to convince people who actually bother to check sources. I also wonder how many people don’t bother to check and are convinced of its potentially flawed perspectives simply because it delivers some citations that are completely irrelevant

It reminds me of a time I met some kook who was arguing the merits of this dumb bullshit they bought off instagram. It was $800 and claimed to cure anxiety with magnetic power. They sent a word document provided by the company and it was just a bunch of random studies about transcranial direct current stimulation, which is a real thing with some evidence, but was completely unrelated and is based on electrical currents and not magnetic woo woo bullshit.

[sshine]

That's very backwards. When I ask Kagi's Assistant for references, it provides the search queries that it crawled before answering. That behavior is independent of what LLM is being used, but the reference output format may vary depending on the LLM.

[pjc50]

The current situation is "The Triumph of the Woo" (Riefenstahl passim). There's a lot of such people, they vote, and their money has ended up in an industry which makes donations to ensure that it can continue to scam people.