Tim Bray on Grokipedia

Wondering if the project will get better from the pushback or will just be folded like one of Elon's many ADHD experiments. In a sense, encyclopedias should be easy for LLMs: they are meant to survey and summarize well-documented material rather than contain novel insights; they are often imprecise and muddled already (look at https://en.wikipedia.org/wiki/Binary_tree and see how many conventions coexist without an explanation of their differences; it used to be worse a few years ago); the writing style is pretty much that of GPT-5. But the problem type of "summarize a biased source and try to remove the bias" isn't among the ones I've seen LLMs being tested for, and this is what Elon's project lives and dies by.

If I were doing a project like this, I would hire a few dozen topical experts to go over the WP articles relevant to their fields and comment on their biases rather than waste their time rewriting the articles from scratch. The results can then be published as a study, and can probably be used to shame the WP into cleaning their shit up, without needlessly duplicating the 90% of the work that it has been doing well.

Bray brought up a really good point. The Grokipedia entry on him was several times the length of his Wikipedia entry, not just because Grok's writing style is verbose, but also because it went into exhaustive detail on insignificant parts of his life simply because the sources were online. My own brief browsings of Grokipedia have left me with the same impression. The current iteration of Grokipedia, besides being untrustworthy, wastes a lot of time beating around the bush and, frequently, off into the weeds.

Just as LLM's lack the capacity for basic logic, they also lack the kind of judgment required to pare down a topic to what is of interest to humans. I don't know if this is an insurmountable shortcoming of LLM's, but it certainly seems to be a brick wall for the current bunch.

-------------

The technology to make Grokipedia work isn't there yet. However, my real concern is the problem Grokipedia is intended to solve: Musk wants his own version of Wikipedia, with a political slant of his liking, and without any pesky human authors. He also clearly wants Wikipedia taken down[1]. This is reality control for billionaires.

Perhaps LLM generated encyclopedias could be useful, but what Musk is trying to do makes it absolutely clear that we will need to continue carefully evaluating any sources we use for bias. If Musk wants to reframe the sum of human knowledge because he doesn't like being called out for his sieg heils, only a fool would place any trust in the result.

[1]https://www.lemonde.fr/en/pixels/article/2025/01/29/why-elon...