Tensor brains are not robustly trained. Ones can simply mess up with their condition numbers without typical memory address attacks as in hackers.
In these ways, we can interact without adding another mode :)
On 24 Jul BE 2566, at 08:08, Albretch Mueller via Corpora corpora@list.elra.info wrote:
On 7/23/23, Peratham Wiriyathammabhum peratham.bkk@gmail.com wrote:
It’s simply something called “distributional semantics” if you want to know.
https://en.wikipedia.org/wiki/Distributional_semantics
I know about such thing, but, quite honestly, I understand neither its mathematical grounding nor the possible reach of its pragmatism. Protagonism aside, I have had personal experiences with such things which I found abysmally stupid. I took two poems I wrote, one in German when I was in my early 20's and one in English later in life:
https://hsymbolicus.wordpress.com/category/gedichter/ (Pyramiden)
https://hsymbolicus.wordpress.com/category/poems/ (lies ...)
The reaction of ChatGPT was hilarious (so cunningly weird that I thought "the government" was once again messing with me). It couldn't even identify the language of the German poem (which I forced a bit for poetic purposes and "hey! it might have even 'learned' its share by now, no?" ;-)). It could not detect that both poems were written by the same writer and what kind of person would author such poems (it was just "stochastically parroting" "lies ..." stanzas). Someone who told me he worked on Latent Semantic Analysis said to me "my poem had broken his algorithm". I had shared my poems. I don't know what he meant.
I can certainly share the kinds of anxieties I had in mind when I wrote "lies ...", but I would love to hear first what Frau "Distributional Semantics", Herr "Latent Semantic Analysis", ... have to say about them.
lbrtchx _______________________________________________ Corpora mailing list -- corpora@list.elra.info https://list.elra.info/mailman3/postorius/lists/corpora.list.elra.info/ To unsubscribe send an email to corpora-leave@list.elra.info