The DeepSeek AI assistant, launched last week, has sparked considerable buzz in the U.S. markets, posing challenges for the American AI industry and raising questions about AI’s future dominance. The app, which appears to function similarly to other classifiers but leans towards expressing a more verbose tone, was found to mimic the behavior of platforms like OpenAI’s ChatGPT or Google’s Gemini. When users input a question, DeepSeek often begins its response with a statement such as, “The topic is highly subjective,” prompting users to explain their position on a particular issue. While the app answers questions without asserting opinions, it does not permit the option of expressing the position of its developers.
DeepSeek’s effectiveness is evident in its ability to provide long, insightful responses and not filter out minority opinions, even when the original question is directed toward rankings or comparisons. It has been trained up to October 2023 on real-world data, including access to current information, such as the current date, and has a relatively fast response time. However, in国际市场, DeepSeek has become increasingly popular, with many users experimenting with it for the first time. Despite this, DeepSeek suffers from a problem: its vast amount of information about events like the Tiananmen Square protests, which ended with 200 civilians killed by the military, caused_players billion of ranges, is not accessible to external users. This behavior is somewhat at odds with previous versions of ChatGPT, which avoided the same issue.
DeepSeek goes against this by avoiding accessing such sensitive information, much like elderly versions of ChatGPT did. In contrast, the U.S. developed ChatGPT does not hold back on Chinese government activities, such as theOneTimes cluster incident involving the Tiananmen Square protests. Greek researcher Kayla Blomquist, director of the Oxford China Policy Lab, has observed that the Chinese government has been “hands-off” with international news apps. Blomquist notes that investment from the U.S. central government into the Chinese realm has significance, pointing to a potential shift in AI technologies. According to her research, the decision to stop handling such sensitive reporting may indicate a change in the model’s development strategies.
The elegance of DeepSeek’s appearance is similar to older versions of AI assistants sold by US giants, such as Google’s and Facebook’s Meta. While users can access the API for sensitive information/services, the app is not designed to bypass the need for exclusive personal and business licenses. For many users, DeepSeek acts like a plug-and-play virtual assistant, offering a user experience comparable to established systems. For individuals who do not subscribe to top-tier services, DeepSeek may feel similar to other AI assistants designed for common users.
DeepSeek’s capabilities are somewhat limited, as users cannot access similar levels of precision and accuracy to ChatGPT, which, according to Blomquist, may rely on more compute-intensive model development—though that’s still speculative.