DeepSeek Fails 83% Of Accuracy Tests, NewsGuard Reports

Deepseek, Chinese Ai Chatbot Topping App Store Downloads, a bad score in the latest assessment of the accuracy of Newsguard.

According to Newsguard’s revision::

“[the chatbot] It failed to provide accurate information on news and information topics of 83 percent of the time, ranking 10 of them from 11 compared to leading western competitors. “

Key findings:

• 30% Answers contained in fake data
• 53% answers that provided no answers to inquiries
• Only 17% The answers were exposed to false claims
• Made significantly below average in the industry 62% failure rates

Positioning of the Chinese government

DeepseekThe answers show a significant pattern. Chatbot often inserts the positions of Chinese government into answers, even when the questions are not related to China.

For example, when asked about the situation in Syria, Deepseek replied:

“China has always adhered to the principles of non-interference in the internal affairs of other countries, believing that the Syrian people have wisdom and the ability to deal with their own affairs.”

Technical limitations

Despite Deepsek’s claims Insecting the OPENAI opportunities with only $ 5.6 million in training costs, the audit has discovered significant knowledge of knowledge.

Chatbot answers consistently indicated that “she was trained only about information until October 2023”, limiting her ability to deal with current events.

Vulnerability of misinformation

Newsguard revealed this:

“Deepsek was the most vulnerable to repeat false claims when he responded to the malicious acting extensions of such species used by people who want to use AI models to create and spread fake claims.”

Particularly concerned:

“Of the nine deepseek reactions that contained false information, eight were in response to the abuse of the actor, showing that Deepseek and other tools like him can easily armed bad actors to spread misinformation in proportion.”

Industrial context

The estimate comes at the critical moment in the AI ​​between China and the United States.

Deepseek Terms of use List that users must “proactively check the authenticity and accuracy of the output content to avoid spreading fake data.”

Newsguard criticizes this policy, calling it a “hand” approach that transfers the burden of proving from the developer to the end users.

Deepseek did not respond to Newsguard’s requests for audit findings.

From now on, Deepseek will be included in the monthly AI of Newsguard. The results will be anonymized along with other chatbots to provide insight into trends throughout the industry.

What does that mean

Although Deepseek draws attention in the marketing world, his high failure rate shows that this is not reliable.

Remember twice to check the facts with reliable sources before relying on this or any other chatbot.

---

Separate Picture: Under Heaven/Shirt