I really want to share that some friends earlier tonight explained to me how and why GPT-like generative text LLMs are fundamentally unethical, so now I finally feel like I get (and share) the animus. I just didn't really put all the facts together before.
Now my point of view is that the models' basis in drawing in the entire content of the internet (more or less) extracts the collective creative labor of billions, which is then blended up and repackaged as a technological innovation, in order to launder the source of this labor and thereby extract its value efficiently.
Even models used by Dolly v2 ultimately derive from earlier GPT models of questionable provenance.
@est an LLM is just a clever statistical model... and presumably, the folks who created the source texts already got their due. if they hadn't, that's a problem for the context in which that text was initially produced, not for LLM training context, unless you're alleging some kind of contract violation for some subset of the text
