Generative Public Wiki
I still often find myself going to Wikipedia for information, because even though it's politically biased, woke, systematically pruned of wrongthink, it's a useful jumping off point for LLM questions based on human curation of attention given to specific knowledge and how it's presented.
Venice users can create content for a public wiki, using generative models to write and expand articles, adding and analyzing citations. VVV stakers can allocate toward proposed changes, “upvoting” them until they're accepted.
A lot of the contributions could probably be done by agents who can crawl the existing Wikipedia and rewrite its content.
Log in to comment and vote
Comments2
Silver River
Feb 12, 2025
This also sounds like it could easily be used to manipulate wiki articles to a biased narrative. People, groups and even wealthy nations already do so and it’s hardly just limited to woke but also anti-woke for example how often information is re-written when it comes to genocides. Predominantly wikipedia is also a western inclined narrative due to who has, and has had, more access to digital resources which also means it’s very unlikely to truly represent the truth, same goes for llm’s themselves so as far as this method being unbiased goes…
Either way this sounds like something you’d use the API for and if you wanted an unbiased system prompt than it’s best to avoid political language. “Don’t adhere to woke narratives-” for example would prime the llm potentially actively avoiding woke narratives even when they have an element of truth to them etc. But something like “avoid emotional arguments that focus on morality, stick strictly to information with sourced, reliable and valid primary evidence” would probably have a more unbiased effect while including what most people consider woke.
Orange Goggles
Feb 12, 2025
With a good system prompt for rewriting Wikipedia articles, it could probably be bootstrapped quite quickly. After that, the agent can ask itself "do I know anything about this topic that's not covered yet? Is it covered in another article that's not linked yet?"
Between cooperating/adversarial agents, and human participants, it should be possible to find ground truth.
Fork choice rule slashing would be a good earning incentive for participating. The more agents involved, the more likely that some VVV would get burned by some people sticking to the losing fork.
That would make VVV deflationary through knowledge curation.