https://www.filosofieinactie.nl/blog/2023/9/5/open-source-large-language-models-an-ethical-reflection (archive version not working) Follow-up wrt openness of LLMs, after the publication of the inteprovincial ethics committee on ChatGPT usage within provincial public sector in NL. At the end mentions the work by Radboud Uni I pointed them to. What are their conclusions / propositions?

Roland Barthes (1915-1980, France, literary critic/theorist) declared the death of the author (in English in 1967 and in French a year later). An author's intentions and biography are not the means to explain definitively what the meaning of a (fictional I think) text is. [[Observator geeft betekenis 20210417124703]] dwz de lezer bepaalt.

Barthes reduceert auteur to de scribent, die niet verder bestaat dan m.b.t. de voortbrenging van de tekst. Het werk staat geheel los van de maker. Kwam het tegen in [[Information edited by Ann Blair]] in lemma over de Reader.

Don't disagree with the notion that readers glean meaning in layers from a text that the author not intended. But thinking about the author's intent is one of those layers. Separating the author from their work entirely is cutting yourself of from one source of potential meaning.

In [[Generative AI detectie doe je met context 20230407085245]] I posit that seeing the author through the text is a neccesity as proof of human creation, not #algogen My point there is that there's only a scriptor and no author who's own meaning, intention and existence becomes visible in a text.

Amplifying dominant perspectives, a feedback loop that ignores all of humanity falling outside the original trainingset, which is impovering itself, while likely also extending the societal inequality that the data represents. Given how such early weaving errors determine the future (see fridges), I don't expect that to change even with more data in the future. The first discrepancy will not be overcome.

Appleton points to the inherent severely limited trainingset and hence perspective that is embedded in LLMs. Most of current human society, of history and future is excluded. This goes back to my take on data and blind faith in using it: [[Data geeft klein deel werkelijkheid slecht weer 20201219122618]] en [[Check data against reality 20201219145507]]

Algogens aren't fixed social entities/identities, but mirrors of the prompts

Algogens are disconnected from reality. And, seems a key point, our own cognition and relation to reality is not just through language (and by extension not just through the language center in our brain): spatial awareness, embodiment, senses, time awareness are all not language. It is overly reductionist to treat intelligence or even humanity as language only.

generative AI as very smart and super incompetent at the same time, which is hard to reconcile. Is this a [[Monstertheorie 20030725114320]] style cultural category challenge? Or is the basic one replacing human cognition?

yes, all generated content is devoid of an author context e.g. It's flat and 2D in that sense, and usually fully self contained no references to actual experiences, experiments or things outside the scope of the immediate text. As I describe https://hypothes.is/a/kpthXCuQEe2TcGOizzoJrQ

generative AI currently follows the pattern of 1 input and 1 output. There's no reason to expect it will stay that way. outputs can scale : if you can generate one text supporting your viewpoint, you can generate 1000 and spread them all as original content. Using those outputs will get more clever.

Mentioned copy writing algogens * Jasper * Wordtune * copy.ai * quillbot * sudowrite * copysmith * moonbeam