That this question is even asked shows the vast dangers of the premature deployment of these generative AI technologies, and the utter lack of meaningful education of the public about their risks. Disclaimers on the user interfaces are routinely ignored. This is all a disaster waiting to happen.
@lauren Seeing all the people here and elsewhere saying "Sure, ChatGPT includes false information but we should use it anyway because it's easy" is both the most revealing and most discouraging thing I've ever learned about humanity.
@lauren I’m very intelligent too and I slightly disagree. I’ve seen users — e.g., my own children when they were little — ask Google “what are good games for kids” or the like, not knowing Google’s rigid search rules. (BTW, I tried always to supervise my own children, but I could never ever explain all the rules of Google search.) Contrariwise, ChatGPT gives such users exactly what they ask for (see below). I’m not saying either is better, but it’s certainly arguable that neither has had enough ”meaningful education of the public about their risks.” https://chat.openai.com/share/030d50fa-bf1f-4b58-9848-cd7053b325dc
@detreville The problem is that (for example) ChatGPT can throw in outright lies -- fabricated results -- and worse -- while insisting they are accurate, in a manner vastly worse than anything that happens on traditional Google Search, because going to look at linked sites provides context. ChatGPT and its like discourages looking at the source sites. And when bad information is mixed in with good information, it contaminates the entire mix. The perfect misinformation witch's brew.
@jjsylvia@detreville But there's an enormous difference. The pages are separate, not amalgamated into a compelling complex single answer that appears to obviate the need to check the sources that comprise the answer, even if you had the means to do so. This means that the GAI answers co-mingle multiple effectively invisible sources, that may be correct, incorrect, or (worse of all I'd say) a mixture of the two. And bad information mixed in such a way with good information is the worst possible kind of misinformation, because it literally makes it impossible to tell the accurate from the incorrect, including purposeful disinformation.
@lauren@jjsylvia But that’s the world we live in, with bad information mixed with good information in the worst possible ways. Very intelligent people like us can vet our sources if we have time, or in professional settings where it’s important — especially when we’re already familiar with the subject area — but I’m pretty sure most people don’t do this most of the time, on the web or elsewhere. Maybe the right pervasive AI could help us….
@detreville@jjsylvia Let's get real. Most people aren't going to vet GAI responses. Even if every link to every aspect of the answer is present, most people won't bother. TOTALLY different than a SERP and going to sites were you see full context. And of course, these GAI systems are scraping content from sites and the creators of that content won't even usually get back the clicks that they get from SERPs now. How are they supposed to survive? It's a train wreck for everyone but the GAI firms. Disastrous.
@detreville@jjsylvia I have seen nothing definitive regarding that question. In fact, I've seen nobody mention it (as described in my blog post) but me, though I'm sure others must have. But currently there is no specification in the robots exclusion protocol to cover this situation, as far as I know, anyway. After all, AI spiders can claim that they are not Indexing at all, as the term is conventionally used today.
@lauren@jjsylvia I’ve read other discussions of this point on Reddit and elsewhere. I’ve also read that GPT-3 was trained using Common Crawl (https://en.wikipedia.org/wiki/Common_Crawl), which certainly does obey robots.txt, or so I’ve read. Of course, I read all this on The Net of a Million Lies (per Vernor Vinge) and we haven’t even begun to ask who’s paying me to say all this! Or how much I get, and why so little. Or whether I used ChatGPT et al. to compose this very message (as I did, in very small part).
If the models’ trainers are obeying robots.txt files, then isn’t it ever so slightly wrong to say the source data were used “without the explicit permission of the creators of that data or the sites hosting it.” Or perhaps web-crawling should have been opt-in this whole time? That would have avoided all these problems and so many more.
And all the complaints I read, that people are making money on all this, start sounding a little like manufactured moral outrage.
BTW, The Net of a Million Lies also tells me you’ve recently changed occupations and gender (see below or above or wherever). Congratulations!
Finally, I’ve used information and services from over a dozen websites to compose this message, by which I imagine I might somehow benefit. Where should I send the check?
@detreville@jjsylvia I assume you read that blog post I noted. That pretty well lays out my thinking on this. Sites have been willing to permit Search because they got something in return -- clicks and views. They had no reason usually to block Search in the general case. But when GAI takes and gives nothing in return in most cases, this will change attitudes, and there needs to be a way to specify whether you want your content taken by GAI systems who will make it their own in ways that starve you of views.
@detreville@jjsylvia I haven't changed anything. That's some cartoonist middle initial R whose been around forever. A simple click on my profile would save you any confusion. Jeez.
@lauren@jjsylvia (I should add that I’ve followed Lauren online since about 1980, and it feels weird to publicly disagree with him. There’s always a good chance that I’m wrong, even when I think that he is. We could even both be wrong….)
@detreville@jjsylvia I may be wrong. In fact, this is the kind of situation where I hope I'm wrong. But I fear that I'm correct. Now if we're both wrong, that will make for an interesting situation for sure!
Yes - the wording used in this lawyer’s explanation of why his filings included fictitious citations is full of a sense of betrayal that the magic genie he’d found to do his job for him was actually not trustworthy:
“… your affiant has never used ChatGPT… prior to this occurrence and therefore was unaware of the possibility that its content could be false.”
@lauren A close colleague of mine told me that he asked ChatGPT for a very basic formular in our field of work. Something everyone and their cousin should be able to answer when being woken in the middle of the night.
Add comment