Netizen Searches for Li Yuanhong with Doubao, Resulting in a Picture of Fan Wei: Experts Warn of Information Hallucinations
Recently, a netizen searching for the historical figure Li Yuanhong using the Doubao search engine received a photoshopped image of actor Fan Wei. This amusing incident reveals the authenticity vulnerabilities in AI's image recognition and information retrieval, with many netizens complaining about the AI's lack of rigor and the entertainment-based errors in searching for serious historical figures, impacting the accuracy of information acquisition.
According to official explanations, the error originated from a misconception in online dissemination: Li Yuanhong and Fan Wei have highly similar appearances. A prank PS image from the casting of "The Founding of a Party" in 2010 was widely circulated, exceeding the reach of genuine historical photos. Some image libraries and encyclopedias mistakenly included it, leading the AI search to prioritize matching this high-profile composite image.
This type of error by AI is essentially a typical manifestation of information hallucination.
Experts point out that large models rely on network data for learning and lack self-checking capabilities. When incorrect information repeatedly appears online, AI is prone to outputting it as a correct result, leading to a situation of "speaking nonsense with a serious face."
Current technology cannot completely eliminate hallucinations, but the error rate can be reduced through optimizing data sources and enhancing retrieval verification.
This incident has also sparked discussions among netizens about the reliability of AI. Some joked about whether similar errors would occur when searching for other historical figures, while others shared experiences of being misled by inaccurate AI information, calling for improved accuracy of AI content.
Currently, relevant platforms are optimizing search logic, prioritizing the promotion of authoritative historical images and reducing interference from online misinformation.