Re: [PLUG] ChatGPPT

[Rich Freeman via plug <plug@lists.phillylinux.org>]

On Wed, Dec 18, 2024 at 8:01 AM Casey Bralla via plug
<plug@lists.phillylinux.org> wrote:
>
> I just read about a new AI search site that is unique because it sites
> its sources.
> https://www.theregister.com/2024/12/16/opinion_column_perplexity_vs_google/?td=rt-3a.
>
> Not sure if this site is any good yet, but I like the "scholarly" idea
> of siting sources.

From a technology standpoint I'm kinda curious how they can marry
sources to the determinations.  I feel like the nature of LLMs would
make such associations kinda loose.

I feel like what you're going to get is an LLM answer cross-indexed to
a generic search engine result.  I tried it out, and one thing I
noticed is that the results were mostly indexed to news articles -
often opinion/advocacy-oriented ones.  (I asked about court cases, and
got links to sites talking about them.)  The results were also kinda
slanted by what is in the recent news, when my question was more
open-ended.

I feel like it has a bit of a recency bias as well.  Maybe that is a
result of the web having one.

Here is an interesting example.  I asked "What is the clearest
evidence of the factuality of the general theory of relativity?"

One of the answers was: "Gravitational time dilation: Precise atomic
clocks have verified that time runs slower in stronger gravitational
fields, as predicted by the theory."  So far so good.  However, the
first reference for that one was:
https://backreaction.blogspot.com/2019/08/how-we-know-that-einsteins-general.html

Now, the author of that blog is somebody I follow who has a lot of
great stuff, and she certainly wasn't trying to "debunk" general
relativity, but that article is about why general relativity isn't
complete, and doesn't talk about the atomic clock experiments at all.
A blog like that certainly wouldn't be a primary source though it
might be fine as a secondary one if it cited sources and actually was
relevant.

(The second citation for that answer was to britannica.com and I'd
consider it a great citation for the purposes of something like this.)

I don't think the concept is bad.  I just don't know how possible it
is to make it work at least with how LLMs work today.  I don't have
any confidence that the linked articles actually are the "reason" the
results were the way they are.

-- 
Rich
___________________________________________________________________________
Philadelphia Linux Users Group         --        http://www.phillylinux.org
Announcements - http://lists.phillylinux.org/mailman/listinfo/plug-announce
General Discussion  --   http://lists.phillylinux.org/mailman/listinfo/plug