Launch of Grokipedia: how an alternative to Wikipedia spreads disinformation.

According to Vox: Grokipedia, an initiative by Elon Musk aimed at creating an alternative to Wikipedia, has already launched. Preliminary analyses show that the site, based on xAI and under the control of the right-wing AI assistant Grok, has already begun to function as a self-sufficient response to disinformation.

Grokipedia is essentially another manifestation of Musk's ambitions in his struggle for 'waking consciousness'. This project demonstrates how he transforms successful initiatives like Wikipedia into less effective versions while declaring victory.

If Musk succeeds and Grokipedia becomes a real competitor to Wikipedia, it may create problems for ordinary internet users. We have already seen how Musk can leverage his resources and influence to transform platform X into a mechanism of disinformation. The creation of a single repository of such disinformation that could train xAI models or even competing AI models is likely to accelerate its dissemination.

Talking about Grokipedia is not just about it being bad. It may also worsen the quality of the rest of the internet.

The Path to Grokipedia

Clearly, Grokipedia uses Wikipedia as its primary source but introduces some far-right views and conspiracy theories on certain topics, presenting this information as facts. Currently, Grokipedia lacks illustrations and references, giving the site a resemblance to chatbot query results. Grokipedia is approximately seven times smaller than Wikipedia, but this is only version v0.1, and Musk claims that 'version 1.0 will be ten times better.'

I was surprised that Grokipedia does not have an article on 'apartheid', but if you search for 'gender genocide theory' — one of Musk's favorite discussions — you will find an article lamenting that the academic community allegedly tries to 'move this theory to the fringe of conspiracy theories, despite observed data on demographic changes.' Wikipedia, as far as possible, indicates this theory as a conspiracy theory in its header.

To understand Grokipedia, it is essential to pay attention to its origins — a tweet by David Sachs, the AI chair of President Donald Trump and a longtime friend of Musk. In a tweet dated September 29, he stated that 'Wikipedia is hopelessly biased. An army of leftist activists controls biographies and works against sensible edits.'

It seems Sachs addressed this tweet directly to Musk, who has actively criticized Wikipedia for over a year. Last Christmas, Musk urged his followers to 'Stop donating to Wokepedia,' pointing to the organization's excessive spending on diversity, equity, and inclusion. He called Wikipedia 'a propaganda app of traditional media' and announced that xAI would create Grokipedia in response to Sachs' tweet.

The Blurred JPEG Theory on the Internet

When I learned about the launch of Grokipedia, I immediately recalled the article 'blurred JPEG' published by the New Yorker in 2023. The author, science fiction writer Ted Chiang, detailed how large language models work, creating synthetic text based on real writing, and whether they can reflect genuine knowledge.

Blurred JPEG points to the issue of uploading an image on the internet that requires compression, downloading a lower-resolution version, and then recompressing it. Eventually, the image becomes unrecognizable due to information loss during copying.

This relates to information on the internet for a long time. In a certain sense, this idea of uploading, mixing, and redistributing content has become what makes the web so engaging. Blogging, which was the beginning of journalism for me and many others, often relies on reading what happens online, processing ideas, and reformatting them for a particular audience, sometimes with certain biases. Tweeting, the descendant of blogging, compresses these posts even further, but its primary goal remains: to democratize and speed up the dissemination of knowledge and ideas in the online environment. Wikipedia, in its simplest form, serves the same function.

But, like JPEGs or old-school photocopiers, recopying leads to the blurring of certain details that seemed less important. Compression simplifies data sharing but complicates returns to the original source.

It appears that Grokipedia is already facing this problem. It is unknown how xAI created this resource, but Matteo Wong in the Atlantic offers a theory. The world's richest man bought Twitter and opened the platform to extreme right voices. 'He then used this repository of conspiracy theories, aggression, and memes in an AI model that had already been developed to avoid controversial or even hateful views,' Wong claims. 'Ultimately, Musk used this AI model to create an anti-woke encyclopedia.'

In other words, Grokipedia was created with the involvement of people but largely under Musk's control. It seems he is uploading his outrage, the reactions of his right-wing followers, and then reintegrating this into AI, which organizes ideas into an encyclopedia: Grokipedia. Compared to Wikipedia, which is not perfect and also has its share of disinformation through an open platform, there exists a system that controls content quality.

I am concerned that this blurred JPEG analogy, while alarming, may not be enough to fully capture the situation. A year ago, after the launch of ChatGPT, we could not pinpoint whether this technology would bring more good or harm. Now, with the emergence of AI-generated information and sites like Grokipedia, we observe that negative consequences are becoming increasingly apparent. It seems inevitable that generative AI and its various branches, including AI-generated encyclopedias, will reproduce internet content — and, essentially, knowledge itself — in a worse quality, blurred form.

I am also worried that when this unreliable information begins to be used for a specific purpose — for example, to radicalize a diminished online population — it may undermine trust in institutions that maintain knowledge in troubled web spaces like Wikipedia.

Elon Musk will not create a better Wikipedia. But he has a plethora of bots trained to undermine trust in Wikipedia. And the more blurred his vision of reality becomes, the more dangerous it is for society.