Humanist Discussion Group, Vol. 39, No. 28.
        Department of Digital Humanities, University of Cologne
                      Hosted by DH-Cologne
                       www.dhhumanist.org
                Submit to: humanist@dhhumanist.org


    [1]    From: Charles Creegan <ccreegan.lists@gmail.com>
           Subject: RE: [Humanist] 39.21: repetition vs intelligence (41)

    [2]    From: James Rovira <jamesrovira@gmail.com>
           Subject: Re: [Humanist] 39.27: repetition vs intelligence (24)


--[1]------------------------------------------------------------------------
        Date: 2025-05-26 07:21:11+00:00
        From: Charles Creegan <ccreegan.lists@gmail.com>
        Subject: RE: [Humanist] 39.21: repetition vs intelligence

Thank you Tim for the continued efforts to clarify your position on LLMs and
language.

With every clarification on this I still, or increasingly, feel as though a
dichotomy is being imposed on a landscape of token utterance and manipulation
that is more diversified.

My first and rather uncharitable thought was of college freshman (pre-ChatGPT)
responses to essay prompts. The struggle to divide up the wall of reading or
lecture into bits that can be recombined as a response is palpable. Actually
construing sense might not be the goal.

Then consider language learners. Two examples from Beverly Cleary's ur-
kindergartner Ramona:
- "Why do you call that [lamp] a Donzer?  - because it gives a lee light"
- On the first day of school told to "sit there for the present" she waits
expectantly all day for her gift.

And again, academics from other fields who can sort of see how the conversation
is put together, maybe just enough to assemble what looks to them like a
contribution.

All of these may seem to operate at a level of token manipulation without the
reader as translator.

One step further, some fields of token utterance appear to be keeping explicitly
at the token level without even pretending a move to understanding - consider
call center scripting, incidentally an area where chatbots are much on a par
with humans. Trying to interact with such a system as a "reader" is bound to
fail.

Of course we usually grant a certain status to first years or children or long
time Humanist lurkers, from other cues and expectations, that we wouldn't grant
to a LLM. But is that based on their utterances? And if not, doesn't that
violate the spirit of the (popularized) Turing test?

----
Ngā mihi

Charles Creegan
Auckland, NZ

--[2]------------------------------------------------------------------------
        Date: 2025-05-25 20:38:28+00:00
        From: James Rovira <jamesrovira@gmail.com>
        Subject: Re: [Humanist] 39.27: repetition vs intelligence

Gabriel, can you explain exactly how this process works?

"That is not right. Words end up close to one another in vector space not
because they are 'close to each other' in the training text. Rather, they
are close to one another in the vector space if the training process
identifies that they are close in meaning."

Specifically, how does the "training process" identify that words are
"close in meaning"?

And how would the process that you describe generate readable text rather
than, say, a list of words resembling a thesaurus?

Jim R
--
Dr. James Rovira <http://www.jamesrovira.com/>

   - *David Bowie and Romanticism
   <https://jamesrovira.com/2022/09/02/david-bowie-and-romanticism/>*,
   Palgrave Macmillan, 2022
   - *Women in Rock, Women in Romanticism
   <https://www.routledge.com/Women-in-Rock-Women-in-Romanticism-The-
Emancipation-of-Female-Will/Rovira/p/book/9781032069845>*,
   Routledge, 2023


_______________________________________________
Unsubscribe at: http://dhhumanist.org/Restricted
List posts to: humanist@dhhumanist.org
List info and archives at at: http://dhhumanist.org
Listmember interface at: http://dhhumanist.org/Restricted/
Subscribe at: http://dhhumanist.org/membership_form.php