DeepSeek: Everything you need to know about the AI chatbot app

DeepSeek is making headlines as it skyrockets to popularity, establishing itself at the forefront of AI chatbot technology. This Chinese AI lab has recently dominated the charts on both the Apple App Store and Google Play, sparking discussions among Wall Street analysts and tech experts about the U.S.'s ability to maintain its competitive edge in the AI sector. With the demand for AI chips in question, the story of DeepSeek's rise is both compelling and instructive. Founded by Liang Wenfeng, who co-established High-Flyer Capital Management in 2015, DeepSeek is an offshoot of a hedge fund that employs AI for trading. The lab was launched in 2023, focusing on developing AI tools distinct from its financial roots. With High-Flyer’s backing, DeepSeek quickly built its own data centers for model training. However, U.S. export restrictions have posed challenges, compelling the company to utilize Nvidia H800 chips—a less powerful alternative to the H100 that American firms have access to. DeepSeek’s workforce is notable for its youthful composition, aggressively recruiting AI PhDs from leading Chinese universities as well as individuals lacking formal computer science training to broaden its knowledge base. In November 2023, the company introduced its initial models, including DeepSeek Coder and DeepSeek Chat. The AI community took particular notice when the next-generation DeepSeek-V2 models were released in spring 2024, showcasing superior performance in various benchmarks at a fraction of the operational cost of rival systems. The launch of DeepSeek-V3 in December 2024 further solidified the company's reputation. Internal benchmarks indicated that DeepSeek V3 surpassed both open-source models, such as Meta’s Llama, and proprietary APIs like OpenAI’s GPT-4o in performance. In January, the introduction of the R1 reasoning model, which self-fact-checks, marked a significant advancement in AI reliability, particularly in complex fields like science and mathematics. However, these models face stringent regulations from China’s internet authorities, limiting their responses to sensitive topics. By March, DeepSeek reported over 16.5 million visits, though still lagging far behind ChatGPT's impressive 500 million weekly users. The company has continued to innovate, releasing updates and new models, such as the V3.2-exp designed for efficiency in long-context operations. Despite a host of venture capital interest, DeepSeek has maintained a mysterious business model, offering many of its services at competitive prices or for free, claiming that operational efficiencies underpin their pricing strategy. The success of DeepSeek has drawn attention from industry giants, with its influence noted in Nvidia’s stock fluctuations and responses from OpenAI’s leadership. While Microsoft has integrated DeepSeek into its Azure AI Foundry, concerns over data security and potential propaganda have led to bans from various organizations and governments, including the U.S. Commerce Department and New York state. As the company looks to the future, its trajectory remains uncertain. While advancements in AI models are expected, the increasing scrutiny from the U.S. government regarding foreign influences in technology could pose significant challenges ahead.