A lesser-known AI research facility in China has caused a stir across Silicon Valley with its recent advancements in artificial intelligence. Their new models, known as DeepSeek, have shown superior performance to America’s leading AI systems while being more cost-effective and requiring less advanced hardware. DeepSeek’s introduction of a freely available, open-source large-language model in late December, which was developed in just two months with an investment under $6 million, has sparked concerns over the potential decline of America’s forefront position in the global AI race. This situation has also sparked debates about the extensive investments by major tech companies in AI research and infrastructure. In comparative tests, DeepSeek’s model excelled over competitors like Meta’s Llama 3.1, OpenAI’s GPT-4o, and Anthropic’s Claude Sonnet 3.5, demonstrating higher accuracy in areas such as complex problem-solving, mathematics, and programming. CNBC’s Deirdre Bosa brings this story to light, including her comprehensive interview with Perplexity CEO Aravind Srinivas.

Video

Video Description

DeepSeek, a relatively obscure AI lab from China, has recently made headlines by releasing AI models that outperform those developed in the US, despite being created more economically and with less powerful hardware. The lab released a free, open-source large-language model in late December, claiming it took only two months and less than $6 million to develop. These new advancements raise serious questions about whether the US can maintain its leadership in AI, and challenge the significant expenditures by large tech firms on AI development and data centers. In several third-party benchmark tests, DeepSeek’s model surpassed Meta’s Llama 3.1, OpenAI’s GPT-4o, and Anthropic’s Claude Sonnet 3.5 in tasks ranging from complex problem-solving to mathematics and coding. The video also features a full interview with Perplexity CEO Aravind Srinivas, conducted by CNBC’s Deirdre Bosa.