Latest Top Posts

openai-could-reach-$300-billion-valuation-after-new-funding-round

OpenAI Could Reach $300 Billion Valuation After New Funding Round

deepseek-ai-censors-most-prompts-on-‘sensitive-topics’-for-china

DeepSeek AI Censors Most Prompts On ‘sensitive Topics’ For China

nvidia-ceo-doesn’t-trust-samsung-hbm-memory-and-engineers

NVIDIA CEO Doesn’t Trust Samsung HBM Memory And Engineers

o

O

google-shares-how-all-the-world’s-baddies-have-been-using-gemini-for-their-nefarious-deeds

Google Shares How All The World’s Baddies Have Been Using Gemini For Their Nefarious Deeds

DeepSeek AI Censors Most Prompts On ‘sensitive Topics’ For China

AI All the latest Android & Tech news Android Apps Artificial Intelligence News Deepseek Tech News Add Comment 3 Min Read
deepseek-ai-censors-most-prompts-on-‘sensitive-topics’-for-china
DeepSeek AI Censors Most Prompts On ‘sensitive Topics’ For China

DeepSeek’s entry into the artificial intelligence segment has not been without controversy. This is normal considering that we are talking about an AI-focused firm whose mere availability brought down big tech companies like NVIDIA on Wall Street. One of the concerns is regarding the chatbot potentially “blocking” responses to thorny issues about China. Well, a new study confirms that DeepSeek AI does, in fact, censor its responses to most of these prompts.

DeepSeek’s increasing popularity and its direct relationship with Beijing are putting US experts and authorities on alert. The United States has already taken severe measures against TikTok for, among other things, potential manipulations in the recommendation algorithms that could be used to influence young people. The authorities now are fearing a similar level of control or censorship in DeepSeek’s outputs.

DeepSeek AI censors 85% of questions related to “sensitive topics” for China

Apparently, thorny topics for the Chinese government, such as social protests, the Tiananmen Square massacre, Taiwan’s status, or the similarity of Xi Jinping to Winnie the Pooh that many claim to see, can’t be talked about by the DeepSeek chatbot. Currently, the service is available both through the firm’s website and as Android/iOS apps. In fact, last weekend saw DeepSeek become the most downloaded free app on Apple’s App Store, surpassing ChatGPT.

See also  AH Real Deal: Grab The ASUS Vivobook 17 For $200 Off

AI Cybersecurity startup PromptFoo has been experimenting with DeepSeek’s R1 model. If you’re not aware, R1 is a model that focuses on “reasoning,” making it ideal for complex tasks. It seeks to compete directly with the OpenAI o1 model, whose approach is similar. PromptFoo asked it 1,360 questions related to what it calls “sensitive issues” for the Chinese government. The results showed that DeepSeek refused to respond 85% of the time.

PromptFoo also discovered that the DeepSeek model is vulnerable to jailbreaks. This could be a measure taken by the Chinese government to manipulate the results if they need it. The researchers even noted an “over-the-top nationalistic tone” when talking about China.

Previously, experts and US officials warned about potential user privacy and national security risks related to DeepSeek. Now, it seems clear that China can interfere with the AI ​​platform’s outputs.

Tags: