In an unexpected technical irony, a Guardian report revealed that OpenAI’s latest artificial intelligence model, designed for professional use, relies in some of its answers on a competing and controversial encyclopedia owned by Elon Musk.
Article contents:
- Introduction
- The Guardian’s test findings
- Sensitive topics and sources
- What is Grokipedia?
- OpenAI’s response
- The risk of model collapse
- FAQs
Introduction
When OpenAI launched its advanced GPT-5.2 model last December, it described it as the most advanced frontier model for professional business. It was expected that this model would rely on reliable and highly accurate sources. However, recent tests have cast doubt on the credibility of the model’s source-selection mechanism, showing that it draws information from a competing platform that has long faced criticism for bias and inaccuracy.
The Guardian’s test findings
According to the report published by The Guardian, tests showed that the GPT-5.2 model cited Grokipedia as a source of information. This online encyclopedia is operated by xAI, a company owned by billionaire Elon Musk. Surprisingly, the two companies are in fierce competition, yet OpenAI’s search algorithms apparently did not exclude competitor content from their database.
Sensitive topics and sources
The citations focused on specific, controversial topics. For example, ChatGPT used Grokipedia as a source for claims regarding the Iranian government’s relationship with the telecommunications company MTN Irancell. It also relied on it for answers regarding British historian Richard Evans, who served as an expert witness in Holocaust denier David Irving’s libel trial. Interestingly, the model avoided using Grokipedia when asked about other topics such as media bias against Donald Trump, suggesting an inconsistent pattern in source selection.
What is Grokipedia?
Grokipedia is Elon Musk’s attempt to create an alternative to traditional encyclopedias, relying heavily on AI generation. The platform has faced widespread criticism since its launch, with studies noting that it included citations from neo-Nazi forums and relied on sources described by US research as “questionable” and “problematic.” A leading model like GPT-5.2 relying on such a source raises serious questions about quality standards.
OpenAI’s response
In response to The Guardian’s report, OpenAI attempted to defend its model, explaining that it searches the web for “a wide range of publicly available sources and viewpoints.” The company emphasized that it applies “safety filters” to reduce the risk of links associated with severe harm appearing, but did not clarify why Grokipedia was not categorized as a source to be cautious of.
The risk of model collapse
This incident highlights a troubling phenomenon known as “model collapse.” When AI models begin training on or citing content generated by another AI (especially if it is less accurate), the quality of information can degrade over time, and facts become blurred. This closed loop of automatically generated information could pollute the global information environment.
FAQs
Question: What is the problem with using Grokipedia as a source?
Answer: Because it is an AI-generated platform that has faced criticism for containing inaccurate information and citations from extremist sources.
Question: Does this affect ChatGPT’s accuracy?
Answer: Yes, relying on unreliable sources reduces the accuracy of answers, especially in sensitive historical and political topics.
Question: How did OpenAI respond to this report?
Answer: It stated that it searches a wide range of sources and applies safety filters, without mentioning Grokipedia by name as a banned platform.
