Paleonet: Mucking about with ChatGPT
N. MacLeod
n.macleod9 at gmail.com
Sun Oct 8 16:42:20 UTC 2023
I dunno. In a lot of the things I read these days the style and the
substance are pretty lousy. If Chat GPT and its ilk can get halfway
there it's still doing better than a lot of real people.
But I agree, there is much misunderstanding regarding what Large
Language Models and associated software like ChatGPT were developed to
do. Still, it's early days. Who knows what this technology will develop
into in 10, 20 or 50 years? The interesting thing is that very few
expected us to have electronic systems that perform as well as ChatGPT
even 5 years ago. Nevertheless, ChatGPT puts me in mind of the old quip
about the bicycle-riding bear. The bear's act was impressive not because
it rode the bicycle well, but because it was able to ride it at all.
It's long been established that bears can be trained to ride bicycles
and that's an accomplishment of note. But I have yet to see a bear who
can ride one well and I don't expect to see such a bear anytime soon.
Norm MacLeod
> Thomas Richard Holtz <mailto:tholtz at umd.edu>
> 8 October 2023 at 14:38
> Many people have a HUGE misunderstanding of what ChatGPT is. It is not
> a replacement for a Google (or similar) search engines.
>
> It is a human language emulator, and does that pretty well. I can
> duplicate many different styles of human writing (conversation, essay,
> poetry, recipes, scientific papers, etc.) quite accurately. Note that
> it is the STYLE of these things, not the substance! It just makes up
> "facts" to populate the content of these texts that are in the style
> of (and often created by the combination of real elements of)
> comparable components of the thing it is trying to emulate. So it will
> create out of whole cloth (well, out of "0"s and "1"s) references and
> bibliographic information, 'data', and so on. Accuracy is not the
> function; style is.
>
> Furthermore, its corpus of material to draw upon is not up-to-date.
> I'm not sure about the current version, but the one available earlier
> this year only used a database up to 2020 upon which to create its
> simulations.
>
> I was astonished over the last year by people thinking that some of
> its features were new, or significant. I heard reporters saying "You
> can ask it to write you a recipe" or "you can look up medical symptoms
> to figure out your likely condition." Had these people never Googled a
> recipe or medical symptoms (note: not advocating "Dr. Google" here...)
> before??
>
>
>
> --
>
> Thomas R. Holtz, Jr.
> Email: tholtz at umd.edu <mailto:tholtz at umd.edu> Phone: 301-405-4084
> Principal Lecturer, Vertebrate Paleontology
>
> Office: Geology 4106, 8000 Regents Dr., College Park MD 20742
>
> Dept. of Geology, University of Maryland
> http://www.geol.umd.edu/~tholtz/ <http://www.geol.umd.edu/%7Etholtz/>
>
> Phone: 301-405-6965
> Fax: 301-314-9661
>
> Faculty Director, Science & Global Change Program, College Park Scholars
>
> Office: Centreville 1216, 4243 Valley Dr., College Park MD 20742
> http://www.geol.umd.edu/sgc
> Fax: 301-314-9843
>
> Mailing Address: Thomas R. Holtz, Jr.
> Department of Geology
> Building 237, Room 1117
>
> 8000 Regents Drive
> University of Maryland
> College Park, MD 20742-4211 USA
>
>
>
> _______________________________________________
> Paleonet mailing list
> Paleonet at paleonet.org
> http://lists.paleonet.org/mailman/listinfo/paleonet
Dr. Norman MacLeod
Professor
School of Earth Sciences & Engineering,
Nanjing University, 163 Xianlin Avenue, Nanjing
Jiangsu, 210023, P.R. China
mobile (cn) +86 1985 2800 990 | e-mail (cn) NMacLeod at nju.edu.cn
mobile (uk) +44 (0)785 017 1787| e-mail (uk) N.MacLeod9 at gmail.com
webpages: http://paleonet.org/MacLeod/MacLeodCV.pdf (CV)
https://macleod01.online (Courses)
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <https://lists.paleonet.org/pipermail/paleonet/attachments/20231008/1349bfb6/attachment.htm>
-------------- next part --------------
A non-text attachment was scrubbed...
Name: NM16.jpg
Type: image/jpeg
Size: 34667 bytes
Desc: not available
URL: <https://lists.paleonet.org/pipermail/paleonet/attachments/20231008/1349bfb6/attachment.jpg>
More information about the Paleonet
mailing list