Elon Musk launches Grokipedia, a Wikipedia competitor. What could go wrong?

Grokipedia, Elon Musk's attempt to create an alternative to Wikipedia, is now online. Initial analysis suggests the site was run by Musk's xAI and fact-checked by Grok, the company right-wing AI assistant – is already a kind of self-sustaining nuclear response from misinformation.

Most importantly, Grokipedia represents another front in Musk's war on wokeness, and another example of Musk taking something that works – in this case Wikipedia – creating a broken version of it and declaring the battle won.

If Musk gets his way and Grokipedia actually becomes a real competitor to Wikipedia, the average internet user will have a problem. We've already seen how Musk can use his wealth and power to transform a platform, X. into a misinformation machine. Creating a repository for this misinformation that could train xAI's model or even competing AI models will inevitably accelerate its spread.

It's not just that Grokipedia might be bad. It could make things worse for the rest of the web.

Grokipedia appears to use Wikipedia as its primary source, but inserts some far-right politics and conspiracy theories on specific topics before presenting the information as fact. There are currently no photos and no links, making the whole thing look a bit like the results of a chatbot prompt, which it actually is. Grokipedia is also about seven times smaller than Wikipedia. But that's just version v0.1 and Musk says“Version 1.0 will be 10x better.”

I was quite surprised that there wasn't an article about “apartheid”, but if you searched for “white genocide theory” – one of Musk's ideological obsessions and… the center of many awkward Grok rants Earlier this year you'll find an article lamenting the tendency of science to “relegate the theory to the status of a fringe conspiracy, despite the observable data on population trends.” Wikipedia, for what it's worth, refers to this theory as a conspiracy theory in the title of his article.

To understand Grokipedia, you have to know its origins, which can be traced back a tweet from President Donald Trump's AI czarVenture capitalist and long-time Elon friend David Sacks. The Sept. 29 tweet said, in part: “Wikipedia is hopelessly biased. An army of left-wing activists maintains the BIOS and fights against sensible fixes.”

It really feels like Sacks is tweeting directly at Musk, who has been stepping up his criticism of Wikipedia all year. Last Christmas Eve, Musk he told his followers to “Stop donating to Wokepedia,” claiming that the organization spends too much on diversity, equity and inclusion. musk called Wikipedia “an extension of old media propaganda” and announced that xAI would create Grokipedia in response to Sacks' tweet.

The Blurry JPEG Theory of the Internet

When I heard about the launch of Grokipedia, the name immediately came to mind “The blurry JPEG” piece that the New Yorker published in 2023. Written by science fiction author Ted Chiang, the article does an excellent job of explaining the then-unknown concept of large language models, how they generate synthetic text based on real writing, and whether they can accurately convey real knowledge.

The blurry JPEG he's talking about refers to the problem with uploading an image to the Internet that requires compression; Download the lower resolution version; and do this over and over again. Over time, the image becomes unrecognizable because so much information is lost when a copy is copied.

This has been the case with information on the Internet since its inception. And in some ways, it's this idea of ​​downloading, remixing, and redistributing content that makes the web so entertaining. Blogging, which is what got me and many others into journalism, often boils down to reading what's happening online, digesting the ideas, and repackaging them for a specific audience, sometimes with a slant and usually in a post that's shorter than the source material. Tweeting, an offshoot of blogging, compressed these posts even further, but the medium retained the fundamental goal of democratizing and accelerating the spread of knowledge and ideas online. In its simplest form, Wikipedia does this too.

But inevitably, as with JPEGs or sheets of paper sent through old-fashioned Xerox machines, making copies blurs out certain details, often ones that seem less important. Compression makes it easier to share the data, but makes it harder to find its way back to the original source.

This seems to be happening with Grokipedia. It's not exactly clear how xAI created it, but Matteo Wong did offers a theory over on the Atlantic. The world's richest man bought Twitter and welcomed far-right voices on the platform. “He then fed this trove of conspiracy theories, hate and memes into an AI model that was already designed not to shy away from controversial or even hateful views,” Wong writes. “Eventually, Musk used this AI model to write an anti-woke encyclopedia.”

In other words, there was People involved in building Grokipedia, but it was probably mostly Musk. It's like he's uploading his anger, downloading the responses from his far-right followers, and re-uploading them into an AI that organizes the ideas into an encyclopedia: Grokipedia. In contrast, Wikipedia is not perfect and, especially due to its open platform, is full of misinformation at all times, but there is a human-centered system that takes care of it.

What terrifies me is the idea that the fuzzy JPEG analogy, while worrying, misses the point. In the months following ChatGPT's launch, we didn't know whether this technology would do more good than bad. Now, with the advent of AI slop and sites like Grokipedia, we're seeing a lot of bad things. It seems inevitable that generative AI and its many offshoots, including AI-generated encyclopedias, will reproduce the contents of the Internet—and, to some extent, knowledge itself—in ways that are lower in resolution, lower in quality, and less sharp. slut is just an example.

What I'm really worried about is what happens when this bastard is weaponized and trained for a specific purpose – such as radicalizing a larger portion of the online population – and starts destroying the integrity of institutions like Wikipedia that are dedicated to preserving knowledge on the troubled Internet.

Elon Musk won't make a better Wikipedia. But he has trained many bots to reduce people's trust in Wikipedia. The blurrier Musk's version of reality becomes, the more dangerous it becomes.

A version of this story was also published in the User Friendly newsletter. Register here So you don't miss the next one!