China’s AI Chatbot DeepSeek: Innovation or Hype?

A closer look at DeepSeek’s impact on AI innovation and its global implications.

DeepSeek, a new AI chatbot app launched by a Chinese tech firm last week, has created ripples across global markets and raised questions about the future of AI leadership. Despite the buzz, experts are debating whether the app is as groundbreaking as it appears. Here’s what we know so far.

How DeepSeek Works

At first glance, DeepSeek operates like any other AI chatbot. Users can type in questions on various topics, and the app generates detailed, neutral responses. Much like its US counterparts, OpenAI’s ChatGPT or Google’s Gemini, DeepSeek refuses to take sides on subjective issues. For instance, it won’t declare a preference between Pepsi and Coke or rate former US President Donald Trump’s leadership.

When asked to compare itself with ChatGPT, DeepSeek avoided direct judgement, instead listing the pros and cons of both platforms in strikingly similar language to ChatGPT’s own evaluations. Its training data cuts off in October 2023, but the app offers access to current information, whereas the web version does not—a feature reminiscent of earlier iterations of ChatGPT.

Limitations and Censorship

One significant difference lies in DeepSeek’s censorship. The app avoids answering questions about politically sensitive topics in China, such as the 1989 Tiananmen Square protests. When asked about the events of June 4, 1989, DeepSeek responds with vague deflections like, “I’m sorry, I cannot answer that question. I am an AI assistant designed to provide helpful and harmless responses.” In some instances, partial answers disappear mid-response, replaced with suggestions to “talk about something else.”

By contrast, US-developed ChatGPT provides detailed accounts of Tiananmen Square and other controversial topics. According to Kayla Blomquist, a researcher at the Oxford Internet Institute, while the Chinese government has taken a “relatively hands-off” approach to DeepSeek so far, the recent surge in government investment in AI signals potential changes in oversight.

Performance and Technology

DeepSeek’s functionality is not without flaws. It occasionally struggles under the weight of viral demand, resulting in slower performance. However, it is built on cost-effective yet capable foundations, using Nvidia’s older-generation chips, Meta’s open-source Llama model, and Alibaba’s Qwen architecture.

For everyday users, the chatbot offers a comparable experience to existing AI platforms. As Blomquist explains, it’s like solving a mathematical problem to eight decimal places instead of 32—it may not be perfect, but for most people, it’s sufficient.

A Threat to US AI Dominance?

DeepSeek’s emergence challenges the business models of leading US AI firms, which rely on heavy resource investments to maintain cutting-edge performance. By leveraging more efficient computational methods, China’s DeepSeek may indicate a shift in how AI models are developed globally.

“This could mark a significant shift in AI development strategies,” said Blomquist. “It remains to be seen whether this approach will redefine the industry, but the next few months will be crucial.”

The Road Ahead

While DeepSeek is not without its limitations, its rapid rise signals that the global AI landscape is changing. Whether it can maintain momentum—and compete with US giants like OpenAI and Google—will depend on how it evolves in the coming months. For now, the app remains a fascinating case study in balancing innovation with the constraints of state oversight and limited resources.