Chinese AI startup DeepSeek rivals OpenAI with breakthrough model

DeepSeek limits registrations due to cyber attack

What is DeepSeek?

DeepSeek aims to develop AI technologies similar to OpenAI’s ChatGPT and Google’s Gemini. By 2021, the startup had acquired thousands of computer chips from U.S. chipmaker Nvidia, crucial for creating powerful AI systems.

Who is Liang Wenfeng?

High-Flyer, a top-performing Chinese quantitative hedge fund, founded DeepSeek. For years, High-Flyer stockpiled GPUs and built supercomputers for financial data analysis. In 2023, Liang Wenfeng, who leads both High-Flyer and DeepSeek, redirected the fund’s resources to create DeepSeek. Armed with a master’s degree in computer science, Wenfeng set out to develop cutting-edge AI models, aiming for artificial general intelligence.

Stock Market Reaction

DeepSeek introduced its DeepSeek-V3 model the day after Christmas, matching the capabilities of top chatbots from OpenAI and Google. A research paper revealed DeepSeek achieved this using a fraction of the computer chips typically required. The startup built a powerful model with significantly less funding than previously thought necessary.

Significance of DeepSeek’s Achievement

The AI boom initiated by OpenAI suggested that creating the most powerful AI systems required billions in specialized AI chips, accessible only to tech giants like Microsoft, Google, and Meta. DeepSeek’s engineers, however, needed only about $6 million in raw computing power to train their new system, roughly 10 times less than Meta’s expenditure.

Innovative Technology with Fewer AI Chips

U.S. AI engineers praised DeepSeek’s research paper for outlining clever and impressive methods to build AI technology with fewer chips. Leading AI systems learn by identifying patterns in vast datasets, including text, images, and sounds. DeepSeek described a method to distribute this data analysis across multiple specialized AI models, reducing time and energy lost in data transfer.

Comparison with OpenAI and Google

OpenAI’s latest model, O3, was designed to “reason” through problems in math, science, and computer programming. On January 20, DeepSeek released its reasoning model, DeepSeek R1, which made a significant impact. This development sent U.S. investors into a frenzy. With export controls implemented in October 2022, DeepSeek demonstrated an alternative approach by revamping the foundational structure of AI models and using limited resources more efficiently.

OpenAI launches new AI model, , GPT-4o and desktop version of ChatGPT