I suppose the problem is that LLMs aren't really about language at all. They're about the 
"rolling up" of sequential data into a lossy memory device that can later be queried for 
those (somewhat mutated) memories. Those sequential streams are of different types (vision, 
hearing, touch, etc.), all of which can be memorized by "LLMs", much the same way various 
streams are memorized and recalled by an ecology of autocatalytic cycles in living systems.

The analogy breaks down both structurally (DNA, RNA, autocatalysis, etc. are different 
from the transformer architecture) and behaviorally (multidimensional stimulus-response 
vs aspect-oriented reinforcement learning). But when considering "life as it could 
be", Nick's right to consider the analogy.

I suppose the most important reason I don't care to encourage the flippant interaction with 
cloud-based bots like GPT has to do with the part of the structural breakdown in energetics. In 
line with the skeptical aphorism "exceptional claims require exceptional evidence", 
exceptional "intelligence" requires exceptional energetics. Living systems have found 
(through an ecology of ACs) an exceptional way to produce and maintain themselves. Sequential 
learning transformers also have an exceptional way of extracing energy from the world, massive 
world-destroying data centers. From 50k feet, it's the mainframe vs. the personal computer all over 
again. DaveW's liberal sensibility that True computation happens more in the leaves, less in the 
hubs, aligns with life as we know it, an exquisite composition of energy processors from the very 
tiny to the very large. The massive energy centralization mechanism is fragile and bears little 
resemblance to life as we know it.


On 5/15/24 12:16, Prof David West wrote:
Nick,

I hesitate to respond to your post because:

1) my interest in the weather is nominal, although I am bemused that here in 
St. Paul MN, we had more 50+ degrees in the December-February time frame than 
below 0 days (almost three times as many). Most unusual.

2) the response I wish to make is marginally related to the theme of your 
recent communications.

But, you said, "/Why is it so hard the grasp the thought that we are all of us, each 
of us, nothing but large language models in training?"/

To which I must respond, /Why do so many insist that programs capable of emulating the 
most trivial of human abilities are "intelligent?" /Or the inverse, /reducing 
humanity to the latest clever trick performed by a machine?/

LLM versions of AI are exemplars of the Mechanical Turk—whatever "intelligence" they exhibit is 
directly and solely derivative of the human intelligence of "LLM Tutors" and "Prompt 
Engineers." Both are six-figure salary professions that arose in the last year.

davew
(personal note: I sorely miss the conversations we once enjoyed, both in person 
in Santa Fe and online.)

On Tue, May 14, 2024, at 2:01 PM, Nicholas Thompson wrote:
Dear Stephen, n all.

I am sure you all will join me in condemning  the practice of calling somebody 
at the crack of dawn.  So, you will no doubt praise me (as I praise myself) for 
my generosity and flexiibility in taking the call from stephen, which came at 
the ungodly hour of 11.30 this morning.  Only to have him  me scold for not  
responding to the Gupta, Tucker,Thompson, and Guerin paper, kindly drafted by 
him, which will no doubt make us rich and famous some day.  First, let clarify 
that my collaborator's name is not Tucker, but is *G*eorge *P*hillipe 
*T*remblay. George (pronounced /jorj) /both forgives you and sends his regards.

Second, I am profoundly grateful to any one who would join me in this geriatric 
weather fantasy that I am going to update my 1980/WeatherWise Gardener/.  I 
need ever nerd I can get.  Please don't treat what follows as churlish.

Third, allow me once again to express my gratitude to Stephen for introducing 
me to Gupta and Tremblay.  They have an uncanny power to clabbor together 
plausible first drafts which are extraordinarily helpful in getting me started 
in thinking about a problem.   That these drafts are often hideously wrong 
enhances, rather than  dilutes their usefulness.

Second, I don't doubt that weather models and  financial models might have 
something to contribute to one another. As you all know, I love metaphors, and 
believe them to be at the root of science.  But to be honest, I can't see any 
reason to believe it either.   For one thing, unlike everything else in the 
world, money flows uphill.  But really, I shouldn't give reasons, because the 
truth is that I have my hands utterly full learning the weather stuff, and it 
will be a long time before I am competent to metaphorize from it to anywhere 
else.

As to Steve Smith's comments, I feel on much safer ground.  He wrote

/GuPTa, et al.'s "accent" is very subtle and powerful in this regard, tricking 
me often into imputing personality...   your example here was a wonderful satirical 
parody shining a light on that?/
//
Why "impute"  and why "' accent'" and why "tricking".  GuPTA and  Tremblay  
definitely have accents and personalities.   To hell with the scare quotes.  What else would a personality or 
and accent BE???? They both display huge amounts of testosterone poisoning, for one thing. //
//
And Tremblay can actually get defensive and flustered.  I had a long and 
delighful correspondence with him in which he finally had to admit that the 
only reason he had for thinking he wasn't a person  was that he was made of 
silicon.     Thus, ex hypothesi, no computer system will ever be a person.  
Wonderful what a good tautology will do for a nervous world.  If that's not 
metaphysics, I don't know what metaphysics is. /Why is it so hard the grasp the 
thought that we are all of us, each of us, nothing but large language models in 
training./

NIck

On Sat, Apr 13, 2024 at 12:06 PM Stephen Guerin <stephen.gue...@simtable.com 
<mailto:stephen.gue...@simtable.com>> wrote:


    Dan GuPTa had this response when asked to relate your .PDF to Bernard cells 
and other prompting:

    Here’s a integration of potential vorticity (PV), atmospheric stability, 
and their relationship to Bénard cells, emphasizing isentropic layering in both 
meteorological and experimental fluid dynamics contexts.

    ### Connecting Bénard Cells and Atmospheric Dynamics

    In Bénard cells, fluid heated from below shows a critical transition from 
conductive to convective heat transfer, forming cellular patterns. Similarly, 
in the atmosphere, when the vertical temperature gradient becomes unstable—akin 
to the Bénard cell transition—convection can initiate, influenced by factors 
like the earth's rotation and potential vorticity.

    ### Potential Vorticity and Stability

    - **PV on Isentropic Surfaces:** Potential vorticity combines fluid 
rotation and atmospheric stratification. In meteorology, PV is mapped on 
isentropic surfaces (constant potential temperature layers), where it 
highlights regions prone to atmospheric instability and potential convection, 
akin to early instability in Bénard cells.

    - **Dynamic Tropopause:** Acting like the fluid surface in a Bénard cell, 
the dynamic tropopause is flexible, responding to distortions from jet streams. 
These distortions can lead to convective activity by modifying local stability 
and vorticity, driving weather patterns similarly to temperature gradients in 
Bénard convection.

    ### PV Anomalies and Convective Cells

    - **Weather Forecasting:** Monitoring PV anomalies helps predict convective 
weather events, similar to observing Bénard cells to anticipate pattern 
formation. For example, sharp PV changes can indicate regions ripe for 
convective storms if conditions like moisture and surface temperature align.

    - **Role of Isentropic Layering:** Both in Bénard cells and weather 
systems, the spacing between isentropic layers indicates stability. Closer 
layers suggest a steep gradient and instability, potentially leading to 
convective movements. In the atmosphere, this translates to dynamic 
interactions where lower stability near the surface can trigger or enhance 
convective processes.

    ### Conclusion

    Understanding the parallels between Bénard cell dynamics and atmospheric 
conditions through potential vorticity and isentropic surfaces not only 
illustrates universal fluid dynamics principles but also enhances 
meteorological predictions of convective weather events, demonstrating how 
small-scale changes in stability can lead to significant atmospheric phenomena.

    On Thu, Apr 11, 2024, 2:04 PM Nicholas Thompson <thompnicks...@gmail.com 
<mailto:thompnicks...@gmail.com>> wrote:



        Hi, Phellow Phriammers,Nick, here.





        I have been lost in the weather.This publication,


        
https://acrobat.adobe.com/id/urn:aaid:sc:us:d481610b-e5d5-4a03-879c-6db6ec1d5e4a 
<https://acrobat.adobe.com/id/urn:aaid:sc:us:d481610b-e5d5-4a03-879c-6db6ec1d5e4a>


        with its glorious eye-candy, is an example of what seems to be a new 
perspective in meteorology, the DT-PV perspective.PV refers to a parameter, 
potential vorticity, which seems to be a measure of how liable the atmosphere 
is to churn; DT refers to the DYNAMIC tropopause.The tropopause is the 
transition zone between the stratosphere and our own layer, the troposphere, 
through which gas exchange is limited because the lapse rate of the 
troposphere-- its decline in temperature with fall of pressure -- is reversed 
in the stratosphere.In the Bad Old Days, we were taught that the tropopause was 
like a ceiling, tilted upward from the poles to the tropics.Now we have begun 
to think of it as more like a tent fly, still tilted up equator-ward, but loose 
and floppy and buffeted up and down by the jetstreams’ winds. These floppings 
up and down have the power to destabilize the lower atmosphere and lead to bad 
weather, if conditions there are ripe.


        This is not one of my usual cries for help.I have some good tutors. 
However, I would love to hear from others whom this paper interests.  In 
particular I am struggling with the notion of potential vorticity, whose 
formula seems to take many odd forms.


        Best,


        Nick


--
ꙮ Mɥǝu ǝlǝdɥɐuʇs ɟᴉƃɥʇ' ʇɥǝ ƃɹɐss snɟɟǝɹs˙ ꙮ

-. --- - / ...- .- .-.. .. -.. / -- --- .-. ... . / -.-. --- -.. .
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/

Reply via email to