As perhaps a few Elon Musk fans may remember, he got really annoyed at Wikipedia last year and tasked his second-rate LLM, Grok, with recreating it as “Grokipedia.” Grokipedia, launched nearly a year ago, basically starts by forking Wikipedia and then having its AI “generate” more details and more stories. It’s… not very good.
And apparently, it hasn’t been updating. And it seems like almost no one noticed.
Over at Lawfare, they put out an article showing that Grokipedia basically stopped accepting updates back in April.
At first, we wondered whether the AI was avoiding sensitive topics. To test that possibility, we submitted an uncomplicated factual update: SpaceX has recently launched its initial public offering (IPO). Grokipedia had not added this extensively sourced, indisputable fact to the SpaceX page. We observed that other users had suggested this edit as well; their requests were also “in review,” some since the day after the June 12 IPO.
We then examined the most popular pages, reasoning that if anything moved through the queue, we would likely see it there. Because Grokipedia provides view counts on individual pages but offers no site-wide ranking leaderboard, we approximated one using the site’s own search-suggestion (typeahead) feature, which returns, for any queried word, the most-viewed pages whose titles contain it. For example, if you type “The ” into the search bar, the first two suggestions are “The Beatles” and “Alexander the Great,” both of which have over 4.7 million views. Using the 10,000 most common words in Grokipedia page titles (obtained from the 5.9 million pages listed in the site map) gave us over 300,000 pages. Like many websites, its individual pages have a steep popularity curve. High-traffic entries on topics such as Elon Musk, ChatGPT, Donald Trump, Taylor Swift, World War II, and Bitcoin draw millions of views, while millions of minor pages sit in a zero-visit long tail.
But neither popular nor unpopular pages seemed to be updating. On ChatGPT’s entry, 12 edits submitted on April 24 were approved the same day, and every edit submitted afterward—May, June, and July—remains “in review.” The same pattern holds across topics and traffic levels; the largest political entries on the site (across both parties) and a second-division football club are both stalled, which is difficult to reconcile with a content- or topic-specific explanation.
In our subsequent analysis of 34,519 pages with at least one suggested edit in our sample, containing a total of 225,496 recommended edits, we found no accepted or rejected corrections dated within the past three months.
You’d think someone would have noticed sooner.
But, as the article notes, it sure looks like the actual human users of Grokipedia quickly dwindled as well:
Across the pages examined, submissions from human users continue at a reduced volume—averaging 216per week post-April.
Other tools that monitored activity on Grokipedia apparently died much earlier:
A public feed that once showed the editing process at grokipedia.com/live—a stream that the Tow Center had scraped to assemble its dataset—stopped functioning between mid-January and early March. A Wayback Machine capture from Jan. 12 shows the feed fully operational, with a running count of approved edits, while a capture from March 5 returns an error page.
These are all signs of a project that’s basically flatlined. For all the hype Grokipedia received as the antidote to Wikipedia, it doesn’t seem to have gained any traction. And, the fact that it basically broke months ago seems to have been noticed by almost no one other than… the dozen or so people out there trying and failing to edit Grokipedia:
While more than half of all contributors suggested only a single edit, a tiny cohort of 13 power users accounts for 42.6 percent of all human edit requests (nearly 40,000 edits). The most prolific contributor submitted over 8,000 corrections across 4,000 pages.
One self-described frequent contributor reported in mid-June that the review system had been stuck for more than 50 days. That places the onset in late April, consistent with our data. He also wrote on X that someone he identified as an xAI team member had acknowledged the complaint but could not provide a status update.
For these power contributors, Grokipedia went from a platform with rapid review times to a black hole.
The article notes that Elon hasn’t mentioned the site since February — barely four months after it launched in October. That silence has now stretched on for about six months and counting.
Even worse, Grokipedia’s logging system appears to have broken as well:
A mass rewrite of the encyclopedia seems to have happened on March 14. Because user suggestions are anchored to specific text selections (“Highlighted sections”), the rewrite appears to have broken the anchors. Grokipedia’s logging system retroactively reclassified previously accepted edits as rejected, attaching the error message, “Highlighted section not found.” However, in several cases that we reviewed, the textual changes appear to have been incorporated into the articles anyway. For example, the 10 launch-week edits to the entry for the actress Prunella Scales are recorded as approved in the Tow archive; nine of the edits now display as rejected on the live site, despite their contents appearing to have been incorporated into the article.
The edit log, in other words, is not entirely stable or reliable. This is potentially confusing for users who made suggestions and might have seen their valid contribution accepted with a “rejected” message nonetheless. It is also not ideal from an auditing standpoint.
It’s entirely possible that someone at xAI (or SpaceX or wherever the checks get cut these days) will flip the server back on and get things started again, but given how much the media hyped up Grokipedia when it launched as a potential “Wikipedia killer,” shouldn’t at least some of them acknowledge what a total failure it has been?