And yet!!

Hannibal famously got elephants through the alps more than 2000 years ago and no defender has ever learned this lesson

We’re watching a shift where AI simulates the cohesion and reinforcement that used to require an entire online community. A single person can now build a complete parallel reality with nothing more than a prompt window. Which seems bad.

So here's what we have: An AI-powered encyclopedia that heavily leans on a volunteer effort that Musk has repeatedly attacked, which selectively edits a subset of topics to recast based with an undisclosed set of instructions. For those pages, citations to conspiracy sites and hate groups go up.

After a week of reviewing the site-wide data and many entries in detail, I think Grokipedia has made significant editorial decisions around sourcing and topic treatment that lend credence to @matteowong.bsky.social's argument in The Atlantic that it is “the next step in Musk’s propaganda machine.”

Grokipedia cites domains with a super low quality score (0.0 to 0.2) *seven times* more than Wikipedia.

We used the Perennial Sources list maintained by EN-Wiki as well as a domain quality score compiled by Lin et al. to approximate the quality of sources cited by Grokipedia v Wikipedia.

Grokipedia includes 2.7 million citations to "generally unreliable", "blacklisted" or "deprecated" sources

The pilfering, however, is selective. On articles in Wikipedia's "controversial topics" bucket, the differences are far greater.

And that's where Grokipedia disproportionately adds low-quality sources, including Stormfront and InfoWars.