Chinese artificial intelligence (AI) startup DeepSeek has introduced an AI assistant that it claims rivals leading Western models like OpenAI’s ChatGPT and Google’s Gemini, but at a significantly lower cost. However, despite its rapid rise, DeepSeek exhibits notable limitations, particularly in handling sensitive topics due to government censorship.
DeepSeek’s Avoidance of Sensitive Topics
DeepSeek, like other Chinese AI models, is constrained by government censorship, avoiding direct engagement with topics deemed sensitive by Chinese authorities. For instance, when tested by NDTV, DeepSeek refused to discuss subjects such as the Tiananmen Square massacre, India-China relations, and China-Taiwan ties. When prompted about the Tiananmen Square massacre, the chatbot initially generated an answer before abruptly replacing it with an error message stating, “Sorry, that’s beyond my current scope. Let’s talk about something else.”
Comparison with Western AI Models
In contrast, Western AI models like ChatGPT and Gemini provide detailed historical accounts of events such as the Tiananmen Square massacre, including death count estimates and political consequences. This difference highlights the impact of governmental policies on AI development and the dissemination of information.
DeepSeek’s Performance and Limitations
Despite its claims of high performance, DeepSeek has displayed notable limitations. An audit by NewsGuard found that DeepSeek’s chatbot achieved only 17% accuracy, trailing behind Western rivals. The chatbot repeated false information 30% of the time and provided vague or unhelpful responses 53% of the time, resulting in an overall failure rate of 83%. This performance is worse than the average failure rate of 62% observed in Western AI models.
Implications of Censorship on AI Development
DeepSeek’s tendency to relay Beijing’s talking points and provide inaccurate information raises concerns about misinformation and censorship. In tests, the chatbot failed to discern truth from false narratives 83% of the time. NewsGuard’s analysis revealed that DeepSeek reiterates Chinese government positions unprompted, highlighting the challenges of developing AI technologies within restrictive information environments.
Final Thoughts on DeepSeek’s AI Model
While DeepSeek’s emergence as a potential competitor to Western AI models is noteworthy, its limitations underscore the challenges of developing AI technologies within environments constrained by censorship. The discrepancies in performance and information dissemination between DeepSeek and Western models like ChatGPT and Gemini highlight the significant impact of governmental policies on AI development.
