Elon Musk’s AI answer to Wikipedia has quietly stopped updating

Elon Musk promised Grokipedia would be a “massive improvement” on Wikipedia. Its AI has not edited a single article since April, its human volunteers are shouting into a void, and nobody at xAI thought to mention it.


Elon Musk’s AI answer to Wikipedia has quietly stopped updating
Image Credits Credit: Grokipedia

Grokipedia, the AI-written encyclopedia Musk built to beat Wikipedia, appears to have quietly stopped working. Not one of its six million articles has changed since 24 April, according to a Lawfare analysis by the researchers Renée DiResta and Ronald Robertson.

That is more than three months of silence.

The tell is the edit queue. Readers can suggest corrections, which an AI is meant to review and then approve or reject, often within minutes. Now they just pile up. The researchers counted 13,002 suggestions frozen “in review,” and found no accepted or rejected edit dated in the past three months across 34,519 pages.

The machine stopped editing itself

The stranger part is that Grokipedia is supposed to write and revise itself. Its own bots, “grok” and “Grok Editor,” made most of the changes. They went quiet between March and mid-April. Since late April, every submission has come from a human, and every one lands in a queue no machine is watching.

This was not a spam problem. Factual corrections made up 97% of requests, and before the freeze the system waved through more than three-quarters of them. The pipeline was working, and then someone turned it off.

The record it left behind is a mess. A mass rewrite in March broke the links between suggestions and the text they pointed to. The site then flipped previously accepted edits to “rejected,” even where it had already made the change. Volunteers who got it right now read that they got it wrong.

Why a frozen encyclopedia matters

A dead side project would not matter much. Grokipedia is not quite that. Similarweb counted 6.7 million visits in June, and its pages have leaked outward. One analysis found Grokipedia cited some 356,000 times across AI answer engines such as ChatGPT and Google’s AI Mode. A reference work that is silently frozen can feed stale or wrong facts into the systems people now ask instead of searching.

This is not really a story about low-grade AI slop. Grokipedia has plenty of solid, well-structured entries. The researchers stress that an AI-assisted encyclopedia, with humans genuinely in the loop, could be a good thing. The problem is what happens when the humans have no way in and the machine goes quiet.

Volunteers edit Wikipedia, for all its faults, around a quarter of a million times a day, in public, with every change on the record. When AI slop pollutes a shared knowledge base, someone can see it and undo it. Grokipedia now offers a suggestion box with nothing behind it.

Is it dead, or just resting?

Nobody outside xAI knows. The company did not respond to The Verge, which picked up the findings. Musk, who once posted about Grokipedia constantly, has not mentioned it since February. The researchers are blunt about what they cannot see: they do not know whether xAI is quietly rebuilding the thing or has walked away from it.

Either reading is awkward for a product sold as the honest, self-correcting future of knowledge. Musk built xAI’s Grok to answer the world’s questions, and pitched Grokipedia as its reference library. For at least three months, that library could not log a fact as basic as SpaceX going public. Its record on getting things right was already contested. Now it is not even trying.

Get the TNW newsletter

Get the most important tech news in your inbox each week.

Also tagged with