Someone should probably inform the White House's "AI & Crypto Czar" that no one is forcing AI companies to train their models on Wikipedia

You would think the obvious solution to "the volunteer-powered project we all train our AI models on for free isn't adequately twisting reality to our political views" would be "... and so we stopped training on it" and not "... and so we will force the volunteers to bend to our will"

#Wikipedia #AI

@molly0xfff I wonder why there is no good “unbiased” (of course they mean biased towards *their* world view) knowledge source that they could train their models on.

Maybe, because – just like in the case of universities – there is not so much “activism” in play here, but rather a natural tendency of things that are done properly with at least moderately academic methods to appear somehow liberal, slightly leftish … which to them then is “radical left”?

@HeptaSean @molly0xfff There is already a "Conservapedia" which is "written from a self-described American conservative and fundamentalist Christian point of view" so they could always go train their LLMs on that. Maybe their problem isn't that the thing they want doesn't exist, but more that they want the thing that other people want to not exist https://en.wikipedia.org/wiki/Conservapedia
Conservapedia - Wikipedia

@dmarti I've read that in other parts of this thread.
Could also be that the key word here is *good* knowledge source.
Just because they *try* to build a “conservative” encyclopedia doesn't mean that it's comprehensive enough to get a usable LLM out of it.
(Also I will never understand why the respectable conservatives that once existed do not do more against their wing of society being drawn into *this*.)

@HeptaSean
Resonates - I also wonder about this: “Also I will never understand why the respectable conservatives that once existed do not do more against their wing of society being drawn into *this*.” 🤔

@dmarti

@HeptaSean Yes, on many subjects the approved "conservative" text about an issue is a mismatch with what a usable LLM is going to be expected to output. Wikipedia is going to be more likely to get paraphrased versions of multiple points of view, which seems more useful as training text.
@HeptaSean @dmarti There have never been respectable conservatives. I remember when they were screaming about invisible Communists. Their most respected genius, William F Buckley jr, was just a KKK member with a Northeastern accent and an LLM level vocabulary.