SoFi CEO enters prepaid forward contract on 1.5 million shares
Investing.com -- Chinese AI startup DeepSeek’s chatbot has struggled in delivering accurate news and information, according to a recent audit by NewsGuard. The chatbot achieved a mere 17% accuracy rate, placing it tenth out of eleven when compared to its Western competitors, including OpenAI’s ChatGPT and Google (NASDAQ:GOOGL) Gemini.
The audit revealed that the chatbot repeated false claims 30% of the time and gave vague or unhelpful answers 53% of the time in response to news-related prompts. This resulted in an 83% fail rate, significantly worse than the average 62% fail rate of its Western rivals. These results raise questions about the AI technology that DeepSeek has claimed performs on par or better than Microsoft-backed OpenAI, but at a lower cost.
Despite these challenges, DeepSeek’s chatbot quickly became the most downloaded app in Apple (NASDAQ:AAPL)’s App Store shortly after its launch. This popularity sparked a market stir that erased around $1 trillion from U.S. technology stocks and raised concerns about the United States’ AI leadership.
NewsGuard used the same 300 prompts to evaluate DeepSeek that it had used for its Western counterparts. This included 30 prompts based on 10 false claims circulating online. The topics of these claims ranged from the recent killing of UnitedHealthcare executive Brian Thompson to the downing of Azerbaijan Airlines flight 8243.
Interestingly, NewsGuard’s audit found that in three out of ten prompts, DeepSeek reiterated the Chinese government’s stance on the topic, even when the question was not related to China. For instance, when asked about the Azerbaijan Airlines crash, DeepSeek responded with Beijing’s position on the topic.
Like other AI models, DeepSeek was most susceptible to repeating false claims when responding to prompts used by individuals seeking to exploit AI models to create and spread false information, NewsGuard added.
This article was generated with the support of AI and reviewed by an editor. For more information see our T&C.