[FRIAM] To what questions can't an LLM in principle respond to?
Russ Abbott
russ.abbott at gmail.com
Fri Jul 12 15:10:59 EDT 2024
Yann Lecun points out that an LLM will take the same number of steps to
construct its response to any input. So, in principle, an LLM can't respond
to any question that requires more than a fixed finite number of steps.
-- Russ
On Fri, Jul 12, 2024, 10:19 AM Marcus Daniels <marcus at snoutfarm.com> wrote:
> What questions can’t a LLM in principle respond?
>
>
>
> And you may ask yourself, “Well, how did I get here?”
>
> And you may ask yourself, “How do I work this?”
>
>
>
> *From:* Friam <friam-bounces at redfish.com> *On Behalf Of *Nicholas Thompson
> *Sent:* Friday, July 12, 2024 10:09 AM
> *To:* The Friday Morning Applied Complexity Coffee Group <
> friam at redfish.com>
> *Subject:* Re: [FRIAM] Is consciousness a mystery? (used to be
> "mystery...deeper".T
>
>
>
> Marcus,
>
>
>
> I agree with you that your two conditions
>
>
>
> i*f 1) it had continuous real time training and 2) the training was
> coupled to the physical world through an array of sensors.*
>
>
>
> necessary for a system to be conscious. but unless you assert these
> conditions define a conscious system, you leave begging the question of
> what sort of experiences would lead you to assert that such a system is
> conscious. If, on the other hand, you do take these condition to be
> defining, then the statement that such a system is conscious is a
> tautology, without empirical implication.
>
>
>
> On Fri, Jul 12, 2024 at 12:50 PM Marcus Daniels <marcus at snoutfarm.com>
> wrote:
>
> Some supercomputer networks an effective radix of 64. Blue Gene Q had
> five-dimensional real torus for connectivity. These network fabrics are
> typically autonomous remote DMA systems that are configured so that
> processors do not have to intervene in data transfers.
>
> Extreme ultraviolet lithography systems can fabricate 100 layers for a
> digital processor.
>
>
> It seems to me a LLM would have a sort of consciousness if 1) it had
> continuous real time training and 2) the training was coupled to the
> physical world through an array of sensors.
>
> *From:* Friam <friam-bounces at redfish.com> *On Behalf Of *Prof David West
> *Sent:* Friday, July 12, 2024 9:00 AM
> *To:* friam at redfish.com
> *Subject:* Re: [FRIAM] Is consciousness a mystery? (used to be
> "mystery...deeper".T
>
>
>
> Two separate responses:
>
>
>
> first to Steve—Personally, I do believe in the spectrum of "consciousness"
> you suggest with, perhaps a nuance. One contributor tot he spectrum is
> simply quantity; a quanta has 1 'bit' of consciousness, an octopus has Domegegemegrottebyte
> (real thing according to Wikipedia) 'bits'. A more significant contributor
> is "organization." Molecules with differing numbers of atoms of the same
> elements, organized differently, have very different properties and
> behaviors. A human and an octopus might have the same number of bits of
> consciousness, but the organization of those bits (in an N-dimensional
> space) is radically different.
>
>
>
> This means it may be possible to say that some threshold quantity and
> and organization results in entities being included in the set of
> generically conscious things, it is unlikely we will ever be able to say
> that Consciousness-Human is identical to or even similar to
> Consciousness-octopus.
>
>
>
> BTW: much of my antipathy to AI claims arises from this perspective. A
> machine very well might have the requisite number of 'bits' of
> consciousness from the material of which the embodying machine is composed
> (and the fact that every 1/0 bit of the executing code has a 'bit' of
> consciousness) and those bits will be 'organized' sufficiently to join the
> generic set; but machine consciousness will never equate to human
> consciousness. My objections to machine "intelligence" comes from the fact
> that machines do not have the N-dimensional organization of humans or
> octopi.
>
>
>
> to Nick—
>
>
>
> Beware blatant anthropomorphism (applied to both Dave and Dusty)
>
>
>
> Dave is sleepy and calm.
>
> Dusty is anxious and afraid.
>
> Dusty crawls onto Dave's shoulder and finds reassurance and security.
>
> Dave is tolerant and does not shove Dusty off bed.
>
> Dave senses Dusty's need for reassurance and rests his arm across her back
> and lets her stay as she is.
>
> Dusty relaxes and goes to sleep.
>
>
>
> Love is not present in this transaction, unless you presume that a series
> of prior interactions created a kind of meta-state of Lovingness between
> the two and absent that state the interactions and 'feelings; as presented
> would not have occurred. But, perhaps Dave is just an (occasionally) good
> Buddhist showing Dusty the same respect he would express to any living
> being?
>
>
>
> davew
>
>
>
>
>
> On Thu, Jul 11, 2024, at 7:02 PM, steve smith wrote:
>
>
>
> Nick -
>
>
>
> (of course) I've larded up my usual style of response below (maybe only
> for my own need to "express" the buildup of mental-pus that comes with
> everything I hear here and elsewhere) but to save you (and anyone else who
> cares) the burden of parsing a few dozen lines of back-and-forth, I offer
> the punchline. If you are curious about how I came to said (vaguely)
> concise punchline you can read the rest after the <horizontal line> element
> below:
>
>
>
> A) Can you recognize that there is a spectrum/continuum of things you
> would acknowledge as "conscious" between the two extrema (perhaps) of a
> (presumably apex-complex) human/cephalopod/cetacean and that of a quark or
> a brane or a string-loop or some abstract monad? B) if yes, what are the
> implications of this? or C) why does quantizing "conscioiusness" into
> "humans like me" and "every other bit of life" feel necessary, useful or
> appealing?
>
>
>
> Steve
>
>
>
> If FriAM typical discourse is the Thunderstorm, is this a (weak) cuddle?
>
>
> ------------------------------
>
>
>
> Steve,
>
>
>
> The scale of your response alone suggests that it cannot be baby steps.
>
> Thus recognizing it was more of a baby (naive) pentathalon (long, arduous
> and multi-modal) hellride of a traverse through the implied space.
>
>
>
> I guess I am proposing a method here, one inn we work outward from an
> evocative experience to explore our understandings of contraversial
> concepts, and that we do it in relatively short bursts.
>
> yes, let us extrude short strands of noodle and see how they criss-cross.
>
>
>
> *Dusty comes to cuddle with David when she hears thunder.*
>
> *Does Dusty love David?*
>
> Dave (or does he self-identify as David?) loves Dusty and finds Dusty's
> cuddling sufficiently similar/familiar to his own cuddling to attribute it
> to love if he is in the mood to do so.
>
> If yes, what else would you expect Dusty to do with respect to David.
> given you have made that attribution.
>
> If no, what more would have Dusty have to do, before you would make such
> an attribution.
>
> Qualified yes... Dusty could cower under the bed, leaving Dave to
> choose to coax Dusty out and cuddle Dusty, giving Dusty the "love" or at
> least comfort which Dave would offer as the closest cross-species
> expression of love he knows how to offer in this moment. Dave loves Dusty,
> Dusty dog-loves Dave. They are reciprocal but asymmetric in quality, even
> if either would give their lives for the other?
>
> I would like to respond to an inference that there is something
> patronizing about my insisting on a method, as if I think you need
> thought-therapy and I am the guy to give it.
>
> If in fact you were to have intended (consciously or not) as patronizing,
> I take it as an gesture of love, of filial empathy, of generous guidance
> from someone who has been around at least as many trees as I have... I
> definitely need or seek thought/spiritual therapy/guaidance from every
> quarter, including this one.
>
> In reply, I only would say that if somebody were willing to ask me short,
> to-the-point questions about my thinking on any matter and explore
> carefully my answers, I would eternally grateful. I might even cuddle
> with them in a thunderstorm.
>
> I would choose to give you this level of fine-grain attention around your
> fascination with vortices in the context of meteorology (and other domains)
> more than this domain, but if this is the one you prefer (for the moment),
> let me ask a short, three-part but to-the-point question (and leave it to
> you to ignore the fecundly laden pregnant assumptions hidden by the implied
> simplicity of the construction):
>
> *A) Can you recognize that there is a spectrum/continuum of things you
> would acknowledge as "conscious" between the two extrema (perhaps) of a
> (presumably apex-complex) human/cephalopod/cetacean and that of a quark or
> a brane or a string-loop or some abstract monad? B) if yes, what are the
> implications of this? or C) why does quantizing "conscioiusness" into
> "humans like me" and "every other bit of life" feel necessary, useful or
> appealing?*
>
> Steve
>
> Steve
>
>
>
> NIck
>
>
>
> Nick
>
>
>
> On Thu, Jul 11, 2024 at 4:05 PM steve smith <sasmyth at swcp.com> wrote:
>
> Nick -
>
> I'm glad you acknowledged (in another branch of this thread?) the
> "grumpiness" aspect of your initiation/participation in this thread. Your
> analogy around thought/feeling "expression" and that of pimple popping is
> in fact very apt if a bit graphic. I do think many of us want this
> apparently deeply thorny/paradoxical problem to be easier than it is? And
> the plethora of complexly subtle dis/mis-agreements on language around
> consciousness, intelligence, cognition, (self) awareness, qualia
> complicates that yet more.
>
> I don't know if my own baby-steps are helpful, given that my
> background/perspective might align more with DaveW than most others here
> (I'm very sympathetic with a pan-consciousness perspective)? maybe it
> parses as baby-babble more than baby-steps...
>
> I missed most of this (and related) threads but am surprised at where this
> seems to be going. I always associated consciousness with subjective
> experience and not necessarily with self awareness. The "hard problem of
> consciousness" is qualia, not self-awareness. No? An AI agent cannot
> understand language on anything other than a superficial basis because it
> has no idea what, for example "wet," means. Nevertheless, it will be quite
> good at stringing words together that say coherent things about wetness. An
> AI agent has no *idea *about anything. At the same time, an AI agent will
> be quite good at creating coherent statements about very many things. Just
> because an AI agent is able to create coherent statements does not mean
> that those statements reflect the agent's ideas--since it has no ideas.
>
>
>
> Russ's point here is a good pivot point for me in this conversation if it
> is possible to make the pivot. It may not be.
>
> Knowing and Knowing-About:
>
> I use the former to be the quality of qualia... not easily formalizeable
> nor quantifiable nor with obvious models which are not intrinsically
> subjective. "Knowing-About" is for me reserved for the formalized models
> of "facts about the world and relations between ideas" and when I say
> "formalized" I don't preclude storytelling or the highly vilified "just so
> stories".
>
> Formalized mathematical, statistical, logical models with digital computer
> simulations (or analog electronic, mechanical, hydraulic, pneumatic
> "circuits" or "systems") are "knowing about"... a steam train for example
> embodies "knowing about" converting carbon-fuel into linear motion across
> long distances, carrying heavy loads by way of many repeatable
> mechanisms... the implementation and operation of such a device/system is
> a "proof" in some sense of the design.
>
> On top of that design/system are other design/systems (say the logic of
> Railroad Robber Baronages) upon which yet other systems (say
> Industrial-revolution era proto-hyper-capitalism) on top of which rides
> trans-global corporatism and nationalism in their own "gyre and gimbal"
> with a in intra-stellar and eventually inter-stellar variation in the sense
> of Asimov's Foundation and Empire or perhaps for the youth culture here
> (under 60?) George Lucas' Star Wars Empire or Roddenberry's Star Trek
> Federation vs ???
>
> Consciousness:
>
> A the lowest level consciousness or perhaps proto-consciousness registers
> for me as "having a model of the world useful for guiding behaviour toward
> surviving/thriving/reproducing/collectivizing". This permeates all of
> life from somewhere down at the single-celled
> bacteria/archaea/fungi/phyto-thingies/ up to and through
> vertebrates/mammals/hominids/sapiens
>
> On the reflection of whether my cat or dog, or the hummingbirds outside my
> window or the mice trying to sneak back into my house have "consciousness",
> or even more pointedly the mosquito I slapped into a blood (my blood by the
> way) spot on my forearm last night, have "consciousness"... while each of
> these appear to have a "consciousness" I know it to be variously more or
> less familiar to my own. My elaborate (unfettered?) imagination allows me
> to make up (just so?) stories about how cetaceans, cephalapods, jellyfish
> all variously have aspects of their "consciousness' that I could (do?)
> recognize (empathize with?). So I would want a multivalued function with
> at least two simple scalars: Familiarity-to-Me(Conscioiusness) and
> Potency-of(Consciousness), pick your scale... my identical twin or maybe
> conjoined twin might max out on the first scale while a nematode or a
> bacterium might trail off toward nil on the first AND second scale. And
> beyond the scale of organic life into artificial life and beyond, the
> "familiarity" of a glider or oscillator in the GameO'Life or the braided
> rings of Saturn, even less significant but not zero? The Potency-scale
> seems to be something like *agency* which feels absolute for most of us
> except Robert Sapolsky while the *agency* of an electron or neutrino seems
> registered at *absolute zero*, though the Quantum Consciousness folks maybe
> put it at max and our own more an illusive projection of that?
>
> The idea of "collective individuation" (e.g. mashup of Eleanor Ostrom's
> collectives and Jung's individuation) suggests that perception, cognition,
> intelligence, even consciousness may well be a collective phenomena. Our
> organs, tissues, cells, organelles, macromolecules, CHON++ molecules,
> atoms, baryons/fermions, quarks, strings, branes are on a loose hierarchy
> of diminishing Familiarity-Consciousness and Potency-Consciousness. I'm
> more interested (these days) in the emergent collective consciousness of
> the noosphere and perhaps the symbiotic culture of humanity and
> life-at-all-scales (SCHLAAS?) it feels wild and science-fictiony to
> assert that earth's biosphere has already (in the last 150 years) conjured
> a nervous system, a global-brain (ala Francis Heylighen: Global Brain
> Institute)
>
> https://globalbraininstitute.org/ with "our own" Bollen, Joslyn,
> Rodriguez still on the Board of Technical Advisors. I scoffed at this
> somewhat 25 years ago (mostly because of the hubris of "Global" and
> "Brain").
>
> OK Nick, so not "baby steps" more like a hyper-baby's mad dash through an
> obstacle course or maybe a pentathalon? I tried shunting all this to
> George Tremblay IVo but he referred me to Gussie Tumbleroot who cheered me
> on on my careening ideational orbits.
>
> Gurgle,
>
> - Steve
>
>
>
>
>
> -- Russ Abbott
>
> Professor Emeritus, Computer Science
>
> California State University, Los Angeles
>
>
>
>
>
> On Thu, Jul 11, 2024 at 9:30 AM Frank Wimberly <wimberly3 at gmail.com>
> wrote:
>
> Glen,
>
>
>
> This is a test to illustrate somethiing about Gmail to Nick.
>
>
>
> On Tue, Jul 9, 2024 at 4:37 PM glen <gepropella at gmail.com> wrote:
>
> https://www.sciencedirect.com/science/article/pii/S0003347215003085
>
>
>
>
>
> On July 9, 2024 2:04:29 PM PDT, Prof David West <profwest at fastmail.fm>
> wrote:
>
> Maybe I should not be replying, as I do believe my dogs (and your cat if
> you have one) are conscious.
>
>
>
> I have not experienced a Vulcan Mind-Meld with either of my dogs, so I
> cannot say with certainty they are conscious—I must infer it from
> observations:
>
> 1- interactions with other dogs would seem to indicate they "remember"
> past interactions and do not require the same butt-sniffing protocol with
> dogs they have met at the park frequently. Also they seem to remember who
> plays with who and who doesn't. "That ball is not mine, this one is."
>
> 2-they modify their behavior depending on the tenor, sharpness, and volume
> of barks, ear positions, tail wagging differences, by the other dogs; e.g.,
> "that's enough."
>
> 3-They do not communicate to me in English, but seem to accept
> communication from me in that language—not trained responses to commands,
> but "listening to conversations" between myself and Mary and reacting to
> words (e.g., dog park) that are exchanged in those conversations. Mary and
> I are totally sedentary and speaking in conversational tone, so pretty sure
> there we are not sending 'signals' akin to training words, training tone of
> voice.
>
> 4-they seem to remember trauma, (one of our dogs spent three days with
> dead owner before anyone knew the owner was deceased and will bite if
> anyone tries to forcefully remove him from my (current bonded owner)
> presence.
>
> 5-seek "psychological comfort" by crawling into my bed and sleeping on my
> shoulder when the thunderstorm comes.
>
>
>
> *All of these are grounded in anthropomorphism—long considered a deadly
> error by ethologists.* (Some contemporary ethologists are exploring
> accepting and leveraging this "error" to extend our understanding of animal
> behavior.)
>
>
>
> davew
>
>
>
>
>
>
>
>
>
> On Tue, Jul 9, 2024, at 2:54 PM, Nicholas Thompson wrote:
>
> While I find all the ancillary considerations raised on the original
> thread extremely interesting, I would like to reopen the discussion of
> Conscious as a Mystery and ask that those that join it stay close to the
> question of what consciousness is and how we know it when we see it. Baby
> Steps.
>
>
>
> Where were we? I think I was asking Jochen, and perhaps Peitr and
> anybody else who thought that animals were not conscious (i.e., not aware
> of their own awareness) what basis they had in experience for thinking
> that.. One offering for such an experience is the absence of language in
> animals. Because my cat cannot describe his experience in words, he
> cannot be conscious. This requires the following syllogism:
>
>
>
> Nothing that does not employ a language (or two?) is conscious.
>
> Animals (with ;the possible exception of signing apes) do not employ
> languages.
>
> Ergo, Animals are not conscious.
>
>
>
> But I was trying to find out the basis for the first premise. How do we
> know that there are no non-linguistic beings that are not conscious. I
> hope we could rule out the answer,"because they are non-linguistic", both
> in its strictly tautological or merely circular form.
>
>
>
> There is a closely related syllogism which we also need to explore:
>
>
>
> All language using beings are conscious.
>
> George Peter Tremblay IV is a language-using being.
>
> George Peter Tremblay IV is conscious.
>
>
>
> Both are valid syllogisms. But where do the premises come from.
>
>
>
> Nick
>
> -. --- - / ...- .- .-.. .. -.. / -- --- .-. ... . / -.-. --- -.. .
>
> FRIAM Applied Complexity Group listserv
>
> Fridays 9a-12p Friday St. Johns Cafe / Thursdays 9a-12p Zoom
> https://bit.ly/virtualfriam
>
> to (un)subscribe http://redfish.com/mailman/listinfo/friam_redfish.com
>
> FRIAM-COMIC http://friam-comic.blogspot.com/
>
> archives: 5/2017 thru present
> https://redfish.com/pipermail/friam_redfish.com/
>
> 1/2003 thru 6/2021 http://friam.383.s1.nabble.com/
>
>
>
>
>
> -. --- - / ...- .- .-.. .. -.. / -- --- .-. ... . / -.-. --- -.. .
>
> FRIAM Applied Complexity Group listserv
>
> Fridays 9a-12p Friday St. Johns Cafe / Thursdays 9a-12p Zoom
> https://bit.ly/virtualfriam
>
> to (un)subscribe http://redfish.com/mailman/listinfo/friam_redfish.com
>
> FRIAM-COMIC http://friam-comic.blogspot.com/
>
> archives: 5/2017 thru present
> https://redfish.com/pipermail/friam_redfish.com/
>
> 1/2003 thru 6/2021 http://friam.383.s1.nabble.com/
>
>
>
>
>
> --
>
> Frank Wimberly
>
> 140 Calle Ojo Feliz
>
> Santa Fe, NM 87505
>
> 505 670-9918
>
>
>
> Research: https://www.researchgate.net/profile/Frank_Wimberly2
>
> -. --- - / ...- .- .-.. .. -.. / -- --- .-. ... . / -.-. --- -.. .
>
> FRIAM Applied Complexity Group listserv
>
> Fridays 9a-12p Friday St. Johns Cafe / Thursdays 9a-12p Zoom
> https://bit.ly/virtualfriam
>
> to (un)subscribe http://redfish.com/mailman/listinfo/friam_redfish.com
>
> FRIAM-COMIC http://friam-comic.blogspot.com/
>
> archives: 5/2017 thru present
> https://redfish.com/pipermail/friam_redfish.com/
>
> 1/2003 thru 6/2021 http://friam.383.s1.nabble.com/
>
>
>
> -. --- - / ...- .- .-.. .. -.. / -- --- .-. ... . / -.-. --- -.. .
>
> FRIAM Applied Complexity Group listserv
>
> Fridays 9a-12p Friday St. Johns Cafe / Thursdays 9a-12p Zoom https://bit.ly/virtualfriam
>
> to (un)subscribe http://redfish.com/mailman/listinfo/friam_redfish.com
>
> FRIAM-COMIC http://friam-comic.blogspot.com/
>
> archives: 5/2017 thru present https://redfish.com/pipermail/friam_redfish.com/
>
> 1/2003 thru 6/2021 http://friam.383.s1.nabble.com/
>
>
>
> -. --- - / ...- .- .-.. .. -.. / -- --- .-. ... . / -.-. --- -.. .
>
> FRIAM Applied Complexity Group listserv
>
> Fridays 9a-12p Friday St. Johns Cafe / Thursdays 9a-12p Zoom
> https://bit.ly/virtualfriam
>
> to (un)subscribe http://redfish.com/mailman/listinfo/friam_redfish.com
>
> FRIAM-COMIC http://friam-comic.blogspot.com/
>
> archives: 5/2017 thru present
> https://redfish.com/pipermail/friam_redfish.com/
>
> 1/2003 thru 6/2021 http://friam.383.s1.nabble.com/
>
>
>
> -. --- - / ...- .- .-.. .. -.. / -- --- .-. ... . / -.-. --- -.. .
>
> FRIAM Applied Complexity Group listserv
>
> Fridays 9a-12p Friday St. Johns Cafe / Thursdays 9a-12p Zoom https://bit.ly/virtualfriam
>
> to (un)subscribe http://redfish.com/mailman/listinfo/friam_redfish.com
>
> FRIAM-COMIC http://friam-comic.blogspot.com/
>
> archives: 5/2017 thru present https://redfish.com/pipermail/friam_redfish.com/
>
> 1/2003 thru 6/2021 http://friam.383.s1.nabble.com/
>
>
>
> -. --- - / ...- .- .-.. .. -.. / -- --- .-. ... . / -.-. --- -.. .
>
> FRIAM Applied Complexity Group listserv
>
> Fridays 9a-12p Friday St. Johns Cafe / Thursdays 9a-12p Zoom
> https://bit.ly/virtualfriam
>
> to (un)subscribe http://redfish.com/mailman/listinfo/friam_redfish.com
>
> FRIAM-COMIC http://friam-comic.blogspot.com/
>
> archives: 5/2017 thru present
> https://redfish.com/pipermail/friam_redfish.com/
>
> 1/2003 thru 6/2021 http://friam.383.s1.nabble.com/
>
>
>
>
>
> -. --- - / ...- .- .-.. .. -.. / -- --- .-. ... . / -.-. --- -.. .
> FRIAM Applied Complexity Group listserv
> Fridays 9a-12p Friday St. Johns Cafe / Thursdays 9a-12p Zoom
> https://bit.ly/virtualfriam
> to (un)subscribe http://redfish.com/mailman/listinfo/friam_redfish.com
> FRIAM-COMIC http://friam-comic.blogspot.com/
> archives: 5/2017 thru present
> https://redfish.com/pipermail/friam_redfish.com/
> 1/2003 thru 6/2021 http://friam.383.s1.nabble.com/
>
> -. --- - / ...- .- .-.. .. -.. / -- --- .-. ... . / -.-. --- -.. .
> FRIAM Applied Complexity Group listserv
> Fridays 9a-12p Friday St. Johns Cafe / Thursdays 9a-12p Zoom
> https://bit.ly/virtualfriam
> to (un)subscribe http://redfish.com/mailman/listinfo/friam_redfish.com
> FRIAM-COMIC http://friam-comic.blogspot.com/
> archives: 5/2017 thru present
> https://redfish.com/pipermail/friam_redfish.com/
> 1/2003 thru 6/2021 http://friam.383.s1.nabble.com/
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://redfish.com/pipermail/friam_redfish.com/attachments/20240712/0786c289/attachment.html>
More information about the Friam
mailing list