On Sun, Dec 25, 2022 at 1:00 AM Anders Wennersten
<mail@anderswennersten.se> wrote:
> For me the only question is of Google come first (who has better knowledge how to interface
> backend knowledge repositories the Wikipedia will become) or if chatGPT will learn this

No speech interface as far as I can tell, but FYI, there now is at
least one search engine that already integrates a language model based
chatbot into search: https://you.com/, which has the backing from
Salesforce founder & billionaire Marc Benioff (a bit more:
https://www.protocol.com/you-dot-com-benioff). Unlike ChatGPT, it
tries to directly cite web sources. When that source is Wikipedia,
you'll note it's basically rewriting/summarizing the Wikipedia
article. I don't know if it uses GPT underneath or its own language
model; Salesforce has certainly funded the creation of models of its
own.

When I asked You.com if it uses GPT-3, it said yes. When I asked it to
provide a source, it generated a URL that does not exist.

I also observed other failure modes, such as combining multiple
persons with the same name into one, or giving directly contradictory
answers to the same question being asked repeatedly. All of these
failure modes are characteristic of language models, which are a bit
like pinball machines in that they will generate results
nondeterministically from the training data.

Of course, this is the technology as it exists today, and even with
those limitations in mind it can prove useful (though it seems
irresponsible to market it as part of a search engine in its current
form).

Warmly,
Erik
_______________________________________________
Wikimedia-l mailing list -- wikimedia-l@lists.wikimedia.org, guidelines at: https://meta.wikimedia.org/wiki/Mailing_lists/Guidelines and https://meta.wikimedia.org/wiki/Wikimedia-l
Public archives at https://lists.wikimedia.org/hyperkitty/list/wikimedia-l@lists.wikimedia.org/message/FAGXGQAMO6DHAANORKLHEY6DKHEL33UW/
To unsubscribe send an email to wikimedia-l-leave@lists.wikimedia.org