DeepSeek, a new AI-chatbot app from China, has taken the tech world by storm since its launch last week, causing ripples across U.S. markets and raising questions about the future of America's AI leadership. Built on a foundation of existing technologies, DeepSeek employs Nvidia chips, albeit older and more cost-effective versions. It utilizes Meta's open-source Llama architecture alongside AliBaba's equivalent, Qwen. Despite its reliance on established platforms, the app's emergence has sparked a debate over innovation and competition in the AI sector.
DeepSeek's launch has highlighted its potential to disrupt the AI landscape, challenging the monetization strategies of leading U.S. firms. According to Kayla Blomquist, a researcher at the Oxford Internet Institute, DeepSeek's presence "absolutely challenges the idea of monetisation strategies that a lot of leading US AI firms have had." This development points to a shifting dynamic in global AI competition.
The app is not without limitations. It self-censors on topics deemed sensitive by Chinese authorities, such as the 1989 Tiananmen Square protests, adhering to regulations that restrict discussion of certain subjects within China. Blomquist notes that "relatively speaking," the Chinese government has been "hands off" with the app, although certain topics remain taboo.
DeepSeek's training data extends up to October 2023, and while the app can access current information, its website version does not have access to real-time updates. Users have noted that DeepSeek bears similarities to established U.S. AI assistants in terms of look and feel. However, it maintains the same accuracy caveats typical of chatbots and often exhibits an overly chatty nature.
In terms of functionality, DeepSeek's responses resemble those of OpenAI's ChatGPT. The app's training on data up to October 2023 allows it to generate informed responses, yet it remains limited by its programming and data constraints.