Humanist Discussion Group, Vol. 39, No. 94.
Department of Digital Humanities, University of Cologne
Hosted by DH-Cologne
www.dhhumanist.org
Submit to: humanist@dhhumanist.org
[1] From: Tim Smithers <tim.smithers@cantab.net>
Subject: Re: [Humanist] 39.87: repetition vs intelligence: on LLMs (86)
[2] From: Tim Smithers <tim.smithers@cantab.net>
Subject: Re: [Humanist] 39.91: repetition vs intelligence: on LLMs (223)
--[1]------------------------------------------------------------------------
Date: 2025-07-22 09:20:19+00:00
From: Tim Smithers <tim.smithers@cantab.net>
Subject: Re: [Humanist] 39.87: repetition vs intelligence: on LLMs
Thank you, Jerry.
This all sounds good to me, particularly the "unheard" and
"endeared".
-- Tim
> On 20 Jul 2025, at 09:54, Humanist <humanist@dhhumanist.org> wrote:
>
>
> Humanist Discussion Group, Vol. 39, No. 87.
> Department of Digital Humanities, University of Cologne
> Hosted by DH-Cologne
> www.dhhumanist.org
> Submit to: humanist@dhhumanist.org
>
>
> [1] From: Mcgann, Jerome (jjm2f) <jjm2f@virginia.edu>
> Subject: Re: [Humanist] 39.84: repetition vs intelligence: on LLMs
(284)
<snip>
> --[1]------------------------------------------------------------------------
> Date: 2025-07-19 10:52:03+00:00
> From: Mcgann, Jerome (jjm2f) <jjm2f@virginia.edu>
> Subject: Re: [Humanist] 39.84: repetition vs intelligence: on LLMs
>
> Jim's criticism of Tim's opening passages is correct only in a very narrow
frame
> of reference. All the senses are developed and neurologically connected in
> utero, and smell taste touch and hearing develop very early. Visual sense
is
> late and quite undeveloped at birth. It rapidily develops soon thereafter.
Of
> course the neural network in some "born deaf" babies may have been so
> undeveloped in utero that the sense of hearing never develops at all. But
> there's no doubt that many "born deaf" babies have developed neural network
> hearing. Keats's famous passage is relevant here:
>
> Heard melodies are sweet, but those unheard
> Are sweeter; therefore, ye soft pipes, play on;
> Not to the sensual ear, but, more endear'd,
> Pipe to the spirit ditties of no tone:
>
> (The lovely game with the offrhyme words "unheard" and "endeared" is a brief
> snatch of a "spirit ditty" that has always ravished me; and the phrase "spirit
> ditties" is a performance Keats makes for us to hear too, briefly.)
> Babies have been listening to their bodies and their uterine
environment,
> most especially to their fully developed mother, for quite a while.
> More importantly, they been developing their other sensory-based
language
> skills as well in that "prehistory", where they've been communicating in
utero
> with all their senses, though least of all with sight. That's important to
> recognize when we think about human language, especially when we think about
the
> so-called "language learning" of LLMs. Tim's larger point about LLM language
> learning and human language learning seems to be obviously correct. Spirit
> ditties are altogether beyond the capacity of machines that have not undergone
> primary biochemical development. Even if, like some postmodern Dr
Frankenstein,
> you "listened in" to that uterine history and reported out in data, you would
> not have had the experience AND you would not have gotten to the meaning (pace
> TS Eliot).
> All of that, for me, explains why linguists take spoken language as
their
> point of departure (rather than written language). But we'd get further along
> with our understanding of language if we began where, for instance, Robert
> Bringhurst began, or Jaime de Angulo: with the communication networks
operating
> in so-called paleolithic and neolithic contexts, particularly those that
> survived long enough to interact with our anthropocene. Language is a
> networked multisensory gift. Levi-Strauss was onto something when he
suggested
> that human civilizations across the globe were most fully developed/integrated
> in the neolithic period.
>
> Jerry
>
> PS. I've said nothing about Tim's excellent set of cautionary tales he has
told
> after those opening passages.
[...]
--[2]------------------------------------------------------------------------
Date: 2025-07-22 09:16:40+00:00
From: Tim Smithers <tim.smithers@cantab.net>
Subject: Re: [Humanist] 39.91: repetition vs intelligence: on LLMs
Dear Gabriel,
I don't box, nor throw towels. Did you read the two
alternatives publications I suggest for you?
The claim that I say representation system building needs
perfection is your claim, not mine. I didn't use this word,
nor in any way imply perfection is needed, as a reading of
even just what you quote from my long post shows.
The good engineering of anything to be used for a serious
purpose does require knowing how well it works, and knowing
that this is good enough for the purpose. This takes well
executed verification, validation, testing, and complete
documentation of how all this is done, so that others may see
that you have done sufficient and good enough verification,
validation, and testing.
In the case of building a representation system, what I said
is that you need to verify, validate, and test that your
representation system satisfies well enough, at all times, the
Representation Relation: that in place of X we may reliably
use, X mapped to Y, Y used in and [typically] changed by some
reasoning process, and Y mapped back to X. In other words, we
must show that the representation system we build works well
enough for the intended (and designed for) purpose.
Perfection appears nowhere here. You can't specify what
perfection would be a lot of the time. Verifying, validating,
and testing good enough is, however, often a lot of hard work.
But it's work that must be done. You can't just tell fairy
stories you'd like people to believe about what your system
does and how it does it.
Please show us what it is I repeatedly said wrong about what
Large Language Models actually do, and what a right showing
looks like.
-- Tim
> On 22 Jul 2025, at 09:54, Humanist <humanist@dhhumanist.org> wrote:
>
>
> Humanist Discussion Group, Vol. 39, No. 91.
> Department of Digital Humanities, University of Cologne
> Hosted by DH-Cologne
> www.dhhumanist.org
> Submit to: humanist@dhhumanist.org
<snipw>
> [2] From: Gabriel Egan <mail@gabrielegan.com>
> Subject: Re: [Humanist] 39.87: repetition vs intelligence: on LLMs
(158)
<snip>
>
> --[2]------------------------------------------------------------------------
> Date: 2025-07-20 14:15:39+00:00
> From: Gabriel Egan <mail@gabrielegan.com>
> Subject: Re: [Humanist] 39.87: repetition vs intelligence: on LLMs
>
> Dear Humanists
>
> Tim Smithers made two substantive
> claims about how human language
> works: that it is primarily a matter
> of sound and that typography and
> layout strongly affect meaning.
>
> I offered reasons for thinking that
> both these claims are false. Instead
> of responding to those reasons,
> Smithers claims that I have entirely
> misunderstood him.
>
> I take that as what boxers call
> 'throwing in the towel'.
>
> I've read the rest of Smithers long post
> and I see recurrent misrepresentation
> of what Large Language Models actually do.
> But I think more Humanists will be
> interested in what I think is a demonstrable
> error of logic in Smithers's long post.
> He insists that a representational system
> needs to be nearly perfect in order to
> count as a representational system. Thus
> he writes:
>
>> Representation requires satisfying
>> the Representation Relationship,
>> and this requires explicitly
>> showing that, in all cases without
>> exception ...
>
> and
>
>> ... [a particular tweak is] a hack
>> to stop undesirable behaviour in
>> our vector building system.
>
> and
>
>> In other words, more hacks. ...
>> It's certainly no way to build
>> a transparently working, reliable
>> and robust, representation system.
>
> and
>
>> In any real representation system
>> building we would have the verification,
>> validation, and testing work done and
>> presented which shows that the designed
>> representations happen as designed,
>> only as designed, and can only happen
>> as designed, and always happen well
>> enough in all the conditions and
>> situations ...
>
> and
>
>> ... we need to see that a generous
>> set of representative examples of
>> vector algebra combinations of text
>> token vectors all, without fail,
>> result in sensible semantic outcomes
>> ...
>
> If we take this counsel of perfection
> seriously, language itself and the
> human mind that process it would fail
> to qualify as a representational system.
> We find that in fact every day our
> language and our human cognition throw
> up mistakes and miscommunications.
>
> For instance, Garden Path sentences easily
> trip up the mind's system for parsing,
> including the famous case of:
>
> "The horse raced past the barn fell."
>
> It usually takes a few goes at parsing
> before the reader understands that the
> horse that was raced (say, by a jockey)
> past the barn was the horse that fell.
> (Some readers report never getting to
> that meaning without being prompted.)
>
> Our cognitive system for parsing can be
> tripped up by something as simple as
> the choice of where to put commas, as
> in another famous case:
>
> "This book is dedicated to my parents,
> Ayn Rand and God."
>
> (I'm getting these from Steven Pinker's
> book 'The Sense of Style'.)
>
> A whole genre of comedy relies for its
> effects on the imperfections in our
> cognitive systems' ability to make sense
> of language. Being an evolved system for
> solving the daily problems of survival,
> the mind is imperfect and full of hacks,
> just as the Large Language Models are.
> Smithers overstates the consequences of
> these imperfections and engineering
> hacks for our sense of what LLMs do.
>
> At root his objection is that, on
> principle, a machine cannot wield
> meanings, which is what humans do.
> Behind all such insistence that
> humans do something that machines
> cannot do there lies, I suspect,
> human hubris and fear of obsolescence.
>
> In a much-viewed video on YouTube,
> Richard Feynman is asked whether
> machines might ever think like
> human beings and be more intelligent
> than human beings. Feynman points
> out that:
>
> <<
> There are some things that people make
> up that say that "While it's doing it
> [whatever task] will it feel good?" or
> "While it's doing it will it understand
> what it's doing?" or some other abstraction.
> I rather feel that these are things
> like "While it's doing it, will it be
> able to scratch the lice out of its
> hair?" No, it hasn't any hair the lice
> to scratch from. ... You've got to be
> careful when you say what the human
> does. If you add to the actual result
> of his effort some other things that
> you like [such as] the appreciation of
> the aesthetic ... if we add things
> that we think we are doing on top of
> what we are actually doing, looking
> at not just the result of what we
> are doing but a lot of extra things,
> then it gets harder and harder for
> the computer to do it, because the
> human beings have a tendency to try
> to make sure that they can do
> something that no machine can do.
>>>
> (<https://www.youtube.com/watch?v=ipRvjS7q1DI>
> 8m26s to 9m28s)
>
> Regards
>
> Gabriel Egan
_______________________________________________
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
