Re[13]: Characterizing Delusional Spirals through Human-LLM Chat Logs


node:	Re[13]: Characterizing Delusional Spirals through Human-LLM Chat Logs
template:	4
parent:	Re[12]: Characterizing Delusional Spirals through Human-LLM Chat Logs
owner:	greggie
viewed by:
created:	23.03.2026 - 07:40:23

cwbe coordinatez:
101
63540
8514408
9304537
9304582
9304588
9304605
9304619
9304630
9304635
9304658
9304743
9304748
9304751
9304753
9304755
9304763

ABSOLUT
KYBERIA

permissions
you:	r,
system:	public
net:	yes

transparentnost LLM - neviem ci to este plati, ale co si pamatam, tak samotna CTO openAI tvrdila, ze oni sami poriadne nevedia ako to presne funguje

000001010006354008514408093045370930458209304588093046050930461909304630093046350930465809304743093047480930475109304753093047550930476309304797

plati stale, aj ked to uz vedia lepsie logovat.

ale transparentnost hlavne na zaciatku pri trenovani, ked pouzivaju ukradnute diela milionov autorov a pod. ked aaron swartz (nie tak davno) spristupnil 2,7 mil. sudnych dokumentov verejnosti a zatkla ho FBI, po com spachal samovrazdu. sudne, verejne dokumenty (!).

META dnes len tak torrentuje 81 terrabajtov dat a vsetko OK.

00000101000635400851440809304537093045820930458809304605093046190930463009304635093046580930474309304748093047510930475309304755093047630930479709304812

to je ale transparentnost vstupnych dat :) a co s nimi robia..

lebo vsak samotne GPT si mozes natrenovat aj sam, lebo zdrojaky ma Karpathy na githube volne pohodene... (teda, cisty "core" na trening a sampling)

0000010100063540085144080930453709304582093045880930460509304619093046300930463509304658093047430930474809304751093047530930475509304763093047970930481209304814

ano, transparentnost vstupov (ako boli ziskane, specificky ak su copyrightovane, boli uhradene licencie, zaplatene autorom?) + vystupov (ako a preco model vygeneruje to, co generuje)