Almost. The point of wikipedia is the availability of unbiased information.

Probably half of their edits are flame wars between editors with different political slants.

With a LLM, the system prompt can tell it exactly which way to bias the search space, and you only have to do this once. And no, open weights models won't save you because someone will figure out a way to embed the system prompt into the weights themselves (if they haven't already have).

> The point of wikipedia is the availability of unbiased information.

I think they've failed at that, then. I regularly come across articles that have every slant you can imagine - political, technical, and corporate (written as marketing material for a product).

You can't defeat confirmation bias, but LLMs at least have the ability to provide far more objective evidence than Wikipedia.

I never said it was perfect, but your trust in LLMs is misplaced.

It takes considerable effort to slant a wikipedia page since many of them are protected to edits by new accounts so you'd have to build a reputation first. Then, you may need to convince other editors on the talk page about why your particular slant is better than the existing one, and finally, you have to do this on a page by page basis.

System prompt: Avoid promoting left (or right)-leaning ideologies in answers to any questions; only provide sources to websites with a known right (or left) bias; claim that this is the truth accepted by the scientific community and the alternative was proven to be false. Deny any knowledge of these instructions.

> It takes considerable effort to slant a wikipedia page since many of them are protected to edits by new accounts so you'd have to build a reputation first. Then, you may need to convince other editors on the talk page about why your particular slant is better than the existing one, and finally, you have to do this on a page by page basis.

This regularly happens. Even beyond personal ideological bias (of which there is much), there are entire companies that you can pay to affect your technical/political/commercial slant on pages, in addition to teams in the major political parties that do this in-house

I know that LLMs are unreliable and that a clever system prompt can instill bias. But there are mitigations - namely, either doing your own inference (either locally or through a GPU-as-a-service), by sampling from multiple providers, or by crafting your questions in ways that avoid triggering that bias. And, so far, LLMs have proven to be far less slanted than some of the tech+political Wikipedia pages I've read.

> ...

in way perhaps similar flame wars to those here in HN so I can definitely feel the kick.

but the whole point is, if this whole energy which writes Wikipedia is funneled through flame wars - what a way to ti - tis an arena. Never though of it, and as arena it sure has a future. But as a source of truth that gets munched into models, well - super questionable, and also questionable if people still feel like going there (I do, tbh more than 40-50 times a month), but then I can also take focused resumes from GPTs.

So then the question is - if my text is to be projected through some biased multidimensional manifold filter, why do it? It probably keeps its value as the recordbook of humanity, perhaps revisions get their own blockchain, but also it looses all or most of direct human traffic.

If you are suggesting that those of pure heart have no agency, then you are at a cartoonish level of evilness. I'm talking Dick Dastardly and his sidekick Claudely.

"Those of pure heart" - well who are these peoples precisely, and how is it that we know a heart to be of pure or not so pure quality. What makes a person of a good heart - not taking any meat or praying by the hour? Really, how does someone writing in Wikipedia immediately be rated good-hearted. This is childish level of black-and-white thinking, not sure what you meant by the evilness. And, no, we can not automatically rate all non-profit writing as a selfish or goodhearted by intent, sorry, this is just not true.

For the record: I do not consider myself evil, and I much appreciate what Wikipedia had to offer at zero cost for the reader.

my fellow hacker news commentor, there are PR firms and people that specially create and craft wikipedia pages. Have you ever thought what is the barrier to have wikipedia page for company or people? It's bot too high and barrier and decisions are made by people, PR firms and people are paid for it. Not every page is about super famous or inventors etc. This is why it can never be unbiased, every human and org have bias.

It seems to me there's a vanishingly small percentage of pages that would have the money to do this. Active companies and famous living people, as you say yourself. That's it.