ChatGPT uses Grokipedia – Musk's Wikipedia competitor
Grokipedia is an AI-generated encyclopedia, managed by Elon Musk's xAI team. ChatGPT accesses its sometimes crude content.
(Image: PopTika / Shutterstock.com)
xAI published an AI-generated encyclopedia in the style of Wikipedia in October. It is called Grokipedia, named after the chatbot Grok. Content from Grokipedia is now also appearing in responses from ChatGPT and Claude.
The Guardian examined the responses from ChatGPT with version GPT-5.2 and found the striking content. In principle, it is not reprehensible to use various sources, including Grokipedia. However, the encyclopedia had already been noted for inaccuracies and special interpretations. These include that, according to Grokipedia, pornography contributed to the AIDS crisis. Furthermore, derogatory terms for transgender people were allegedly used, and ideological justifications for slavery were written down.
The background to Grokipedia is that Musk accuses Wikipedia of bias, specifically that conservative views are less visible. This is certainly not very surprising, considering that Musk believes that misleading or sexualized chatbot content should simply be viewed with more humor. Anything else is censorship. Recently, numerous people had used Grok to generate partly pornographic images.
AI models can be poisoned
Nevertheless, the found responses show how Grokipedia can also infiltrate other chatbots and information sources. In addition, ChatGPT allegedly did not properly attribute the source. While the responses should clearly originate from Grokipedia, the encyclopedia was not listed as a source.
Videos by heise
Other websites can, of course, also influence the search results and responses of chatbots. Russian propaganda networks, for example, are already exploiting this. Poisoning training data or hiding instructions on websites are fundamental dangers for AI models and users.
(emw)
