• Due to ongoing issues caused by Search, it has been temporarily disabled
  • Please excuse the mess, we're moving the furniture and restructuring the forum categories
  • You may need to edit your signatures.

    When we moved to Xenfora some of the signature options didn't come over. In the old software signatures were limited by a character limit, on Xenfora there are more options and there is a character number and number of lines limit. I've set maximum number of lines to 4 and unlimited characters.

ChatGPT

Earlier in the thread I posted about the idea of combining ChatGPT with WolframAlpha to make something capable of answering questions with mathematical components more accurately than just GPT alone.

Scott Aaronson recently had a published a paper in which he does just that. Here's his discussion from his blog:
https://scottaaronson.blog/?p=7460


Click through to see the example problems.


There's more discussion of takeaways at the link.

A LLM as a "base application" with addon modules to give specific functionality sounds like the right way to go.

But that doesn't seem to be what AI makers (and fans) "want"; Open AI for instance really seems to want ChatGPT to eventually be all things for all people, so much so that they expect people will actually want to pay more for a pared-down and narrowed-scope version of it, a la the GPT Store.
 
(Haven't really checked this thread in a while, so this may have been discussed already. Any case, this is completely obvious, just an observation, nothing much to discuss really.)

In another thread, while reading the output when a query was put to Bing, it occurred to me that given how these thingies tend to "hallucinate", it would be cool if, like Wikipedia, answers came equipped with exhaustive footnotes. So that one can, if so inclined, check the sources for these answers, and do that very quickly and without having to waste time "researching" by trawling the internet.

And then it occurred to me that I've not myself actually tried out these AI thingies myself. Yes, yes, very very easily done, I know, but I'd never ever bothered, and only ever relied on what others have said, whether here, or elsewhere, in forming my views about AI.

So I went to Bing, and put that same question there, and right there, it does come equipped with that set of footnotes I was talking about. So that's ...very cool!

(Like I said, very obvious observation, to anyone who's actually taken the trouble to observe. I hadn't, so far, which is why this came as a revelation to me. So Bing gives you answers, and also gives you full-on links of where it's sourced its answer from, so that you can go check and validate if you want to, what more can you possibly ask for? Very cool.)
 
You will also notice that the answers Bing produces answers that are taken verbatim from the links in the footnotes.
 
You will also notice that the answers Bing produces answers that are taken verbatim from the links in the footnotes.
Indeed. I was the one who posted the extract that Chanakya is referring to, and I edited out the footnotes when I posted it.

I am finding Copilot (aka Bing AI) a good search engine for these natural-language type questions. I asked it the other day what the origin of the term "grauniad" (referring to The Guardian) was, and in a few seconds it told me, which was interesting. It's one of those thinks that just seems to work.
 
I am finding Copilot (aka Bing AI) a good search engine for these natural-language type questions. I asked it the other day what the origin of the term "grauniad" (referring to The Guardian) was, and in a few seconds it told me, which was interesting. It's one of those thinks that just seems to work.
I once tried to find the name of a stone I have in my tiny mineral collection: a brown, glassy stone with what looks like copper filings inside. Bing came up with tons of wrong suggestions, even when I tried to narrow the search with more lengthy descriptions, and telling it to exclude its previous suggestions. But Bing just regurgitated the old suggestions, so I gave up, and used Google and my own eyes to sift through pictures of stones until I found the right one: sunstone.

Not too much intelligence apparent here. I often have the feeling that ChatGPT does better, but i didn’t try it here.
 
You will also notice that the answers Bing produces answers that are taken verbatim from the links in the footnotes.


Does it? (I didn't check the links either, merely noted, with approval, that they're there!) ...But doesn't that kind of ...I don't know, wasn't that the whole point, that separates this lot of AI from vanilla-Google say, that it would provide original text, as well as original art and songs and whatever else?

But of course, as a search engine, this is way better, providing verbatim answers. Less chance of "hallucination", in that case --- although I suppose the possibility of wrong answers still remains. And again, that makes it that much easier to cross-check, since you needn't even read the linked articles, you can simply CNTRL-F there, and read only the the relevant portions.

(To an extent Google does that already. Many's the time I've simply looked simply at the thing it puts on top, without bothering to actual open the link. And that of course is simply taken from what the Google algo found most the most promising entry. So Bing goes one step beyond, I suppose, and mixes from more than one source.)


eta: Or does it? Maybe Bing-AI too simply regurgates from what its algo decides is the likeliest entry? Except at more length than Google's initial page does?

...In any case, my take --- completely fallible, given how little I've fiddled with these things --- my own take is that Google's still the way to go, for regular searches. AI's simply a curiosity at this stage, something to fool around with if you're that way inclined, basically have some fun, is all. Although no doubt one of these days it'll become better than vanilla-Google, sure (at which point vanilla-Google will itself cease to be, I suppose, and instead go the AI route itself --- or maybe provide both options, whatever).
 
Last edited:
I once tried to find the name of a stone I have in my tiny mineral collection: a brown, glassy stone with what looks like copper filings inside. Bing came up with tons of wrong suggestions, even when I tried to narrow the search with more lengthy descriptions, and telling it to exclude its previous suggestions. But Bing just regurgitated the old suggestions, so I gave up, and used Google and my own eyes to sift through pictures of stones until I found the right one: sunstone.

Not too much intelligence apparent here. I often have the feeling that ChatGPT does better, but i didn’t try it here.
It depends on the training data. Maybe it didn't get a lot of deep-level geology information. The fact that you had to sift through pictures of stones rather than easily being able to find an adequate description suggests that the question you were asking was rather technical. I don't think it's reasonable to expect an AI to be an expert on every subject ever. Not yet, anyway.
 
Back
Top Bottom