[FRIAM] Datasets as Experience

[Marcus Daniels]

I am a little surprised that gaslighting / mansplaining would be so prevalent in the media sources used to train chatGPT.   Cold-blooded gaslighting occurs with some people, but in my experience disagreements amongst people don’t cycle that way.   People will get mad or disengage from conversation.   Maybe there is a simpler explanation why chatGPT fails in this way?

From: Friam <friam-bounces at redfish.com> On Behalf Of Steve Smith
Sent: Thursday, February 9, 2023 12:00 PM
To: friam at redfish.com
Subject: Re: [FRIAM] Datasets as Experience




I've been in long-standing confusion about the meaning of "ethics". And almost without fail, if/when I say that to a group of people, particularly men, everyone jumps in and explains to me what they think it means.

<let me try a little of my own mansplaining> As for "mansplaining", I find that there is a different mode (I find it different/complementary to proper mansplaining) which is to report the mental scattering that a particular word/phrase/concept generates in the receiver.  A collective free-associative exploration (worst case is random tangenting?  word-salad from a salad shooter instead of a nicely laid out salad bar or well constructed Cobb?).

Maybe it is all on the same spectrum.  When tightly focused and (intended to be) coherent, it *becomes* mansplaining I think?   I am pretty sure that the concept (if not precisely the term) was first popularized in Rebecca Solnit's Men Explain things to me<https://en.wikipedia.org/wiki/Men_Explain_Things_to_Me>.

She described being introduced to a man at a party who had just read her  book (River of Shadows<https://en.wikipedia.org/wiki/River_of_Shadows>) on Muybridge's early stop-action photo work and motion studies.   The hostess was apparently fairly clear to the man that he was being introduced to the *author* of the book he was so taken with, but instead of hearing that I saw Solnit as his next victim to tell *all about* the subject of the book that he just read that she *wrote*.   As she tells it, the conversation ended without him ever twigging to the fact that when she tried to interrupt him, it was to tell him that *she* was the author (and would be happy to talk with him about the subject but probably didn't need a lecture on the content of her own book).   I don't know if her failing to clue him in was her own passive aggressive trick or if she really couldn't get a word in edgewise.  I've seen both things happen...   but her essay on this really moved me.  Her work/voice in general has been a great thing to/for me... FWIW.
Of course, each of their explanations is different and often pairwise incommensurate. So, if they're sedate, by the end of the conversation, I can convince most people *they* don't know what "ethics" means, either. Add to that the implicit question of whether non-humans have ethics and the file metaphor (from paper to bits on disk to orchestrated bits on multiple disks to in-context learning modified bits on multiple disks), then that sentence is all over the map of possible meanings. That was supposed to be the point of my remark ... in the context of DaveW's question about the semantics of LLM workflows.

I don't know if I'm coming around to what it is you mean when you talk about communication being an illusion, but it is a much more comfortable concept now than it was the first time I heard you say it.   Maybe you are getting through to me?  Is that communication?

I'd ask what *co*-munnication and *commune*-ication might mean if not this highly-technical   transfer of mental-emotional-states-between-entities-via-serialization-and-tokenization  ?   I tend to think of "communication" more as the process of seeking/building resonance in many modes across many entities...  though this is probably not a definition most here want to use...

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://redfish.com/pipermail/friam_redfish.com/attachments/20230209/1d7218e6/attachment.html>