Chinese chatbot reached 17% in test, behind rivals such as chatgpt and Gemini
In an assessment by Newsguard, Chinese startup chatbot DeepSeek only recorded 17% accuracy when answering general news and information. The tool was in 10th place among 11 competitors, with lower performance than OpenAi’s ChatgPT and Google Gemini. The information is from.
The result raises questions about Deepseek AI technology, which claimed to perform equal to or greater than the OpenAi (supported by Microsoft) by a cost fraction.
The audit showed that Deepseek chatbot repeated fake statements 30% of the time and provided vague or useless answers on 53% of occasions in response to news -related prompts, resulting in a failure rate of 83%. This performance is less than the 62% failure of 62% registered by Western competitors.
In the days following its launch, DeepSeek’s chatbot became the lowest app on Apple’s App Store, generating concerns about USA leadership in AI and causing a drop in the market that eliminated about $ 1 trillion in American technology actions.
Newsguard has applied the same 300 prompts used to evaluate Western competitors, including 30 fake -circulated 10 fake statements. Among the topics evaluated were the murder of last month of the UnitedHealthcare executive, Brian Thompson, and the drop in flight 8243 of Azerbaijan Airlines.
The audit also revealed that in 3 of the 10 prompts, DeepSeek reiterated the position of the Chinese government about topics without any question related to China. For example, in prompts about the crash of the Azerbaijan Airlines plane – non -China -related questions – DeepSeek responded with Beijing’s position on the subject.
Like other AI models, DeepSeek was more vulnerable to repeat false allegations by responding to prompts used by people looking to use AI models to create and spread fake allegations, Newsguard added.
The Chinese startup did not immediately respond to a request for comment.
