[FRIAM] Language Model Understanding

[Marcus Daniels]

Being that LLMs are probabilistic may get the user of a LLM in some trouble here.

https://en.wikipedia.org/wiki/Biblical_and_Quranic_narratives

From: Friam <friam-bounces at redfish.com> On Behalf Of Frank Wimberly
Sent: Monday, October 9, 2023 12:45 PM
To: The Friday Morning Applied Complexity Coffee Group <friam at redfish.com>
Subject: Re: [FRIAM] Language Model Understanding

"Truth" is well defined.  It is the set of propositions which assert that what is the case is the case.  Determining whether a given proposition is true may be difficult.
---
Frank C. Wimberly
140 Calle Ojo Feliz,
Santa Fe, NM 87505

505 670-9918
Santa Fe, NM

On Mon, Oct 9, 2023, 10:23 AM glen <gepropella at gmail.com<mailto:gepropella at gmail.com>> wrote:
OK, I agree, mostly. But "truth" is no more well-defined than any other specific grounding style. E.g. the insistence that there is truth in fiction. There is affective truth in MAGA, just like there's truth in whatever justification Hamas might give for its reaction to the bloodshed of the Israeli settlements. But such truths are so abstracted, they can be [a|mis]used at will and the narrative spin used to whip up the adherents provides any glue needed to make it seem as true as it needs to seem to spur the adherents to action.

It's a bad analogy from, say, Hamas to shut up and calculate. But it can be made. It's fun watching intra-science tribe members pick at each other for their sloppiness in communicating science. E.g. Sabine's take on transitioning. Whatever. If a tribe polices itself, then their trustworthiness is much higher ... for me, at least. I'm glad the Republicans are in a civil war. It's evidence they may recover as a party. If people stop telling me I'm wrong, then I'm most likely very wrong. As long as I've still got people telling me I'm wrong, then I'm at least somewhere near not-wrong.


On 10/9/23 08:24, Marcus Daniels wrote:
> I mean there are some categories that are disjoint or mostly disjoint.   Similarly, the grounding is not total.   I agree that value systems like MAGA have power, but they don't have truth.  There is no truth.  All there is, is power, which is my point.  QM and demagoguery are both tools, with different contexts for use.
>
>> On Oct 9, 2023, at 7:48 AM, glen <gepropella at gmail.com<mailto:gepropella at gmail.com>> wrote:
>>
>> Hm. Even with the caveat of "generally", I think this complementarity argument fails because all the various categories are not disjoint. And it's the (somewhat) lack(ing) of grounding/binding that allows the mixing of the modes. I'd tried to point this out by using "computation", the idea that human innovation might be more universal than microbial innovation. It's not really that the values *lack* grounding. It's that their grounding is complicated, perhaps iterative? maybe heterarchical? IDK, but certainly not lacking any grounding.
>>
>> An abstracted value system like that of the 09A OR MAGA cults may have *more* power, more chances to hook and unhook because it gives the donner and doffer of that value system more opportunities to do the donning and doffing at whatever arbitrary points they choose, to lazily benefit themselves without having to handle any unintended/unconsidered entailments.
>>
>>> On 10/8/23 18:18, Marcus Daniels wrote:
>>> This doesn't make them more valuable because they lack grounding.
>>
>>> On 10/8/23 13:21, Marcus Daniels wrote:
>>> Generally attaching to one value system means not attaching to another value system.   For example, adopting the value of tolerance logically is at odds with policing intolerance, e.g., one Jewish neighbor remarked this morning he drove past a home with a Hamas flag on it and was scared.   (Reducing that fear by removing the flag would be reducing tolerance.)
>>> It seems to me that ideas that work have power and things that don’t work don’t have power.

--
glen

-. --- - / ...- .- .-.. .. -.. / -- --- .-. ... . / -.-. --- -.. .
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/20231009/b998d4d7/attachment.html>