Business
Tech
In a move that has sent shockwaves through Silicon Valley, Chinese startup DeepSeek has unveiled a powerful new open-source AI model, dubbed R1, which is challenging the status quo in the tech industry. Developed on a remarkably modest budget, DeepSeek's R1 is not only impressive in its capabilities but also signals a significant shift in how AI innovation is approached.
The rise of DeepSeek has sparked discussions about whether the US is losing its edge in AI. However, experts like Ali Ghodsi, CEO of Databricks, see this as part of a broader technological transition. "It's a paradigm shift towards reasoning, and that will be much more democratized," Ghodsi notes. This shift emphasizes developing advanced capabilities over merely scaling up model sizes, creating opportunities for smaller startups like DeepSeek to make a mark1.
DeepSeek's technology emerged from a small research lab linked to one of China's top-performing quantitative hedge funds. A research paper from December revealed that an earlier model, DeepSeek-V3, was built for just $5.6 million, a fraction of what competitors like OpenAI spend on similar projects. OpenAI has mentioned that some of its models cost over $100 million each1.
The efficiency and performance of DeepSeek's models have already prompted discussions about cost-cutting at major tech firms. An engineer at Meta, speaking anonymously, suggested that the company will likely examine DeepSeek's techniques to reduce its own AI expenditure. Meta's spokesperson highlighted the importance of open-source models, stating, "We believe open source models are driving a significant shift in the industry, and that’s going to bring the benefits of AI to everyone faster"1.
DeepSeek's R1 and R1-Zero models demonstrate sophisticated reasoning capabilities similar to those of OpenAI and Google's advanced systems. They achieve this by breaking down problems into manageable parts, requiring extensive training to ensure reliable results. A recent paper by DeepSeek researchers details their approach, which includes automated learning methods and skill transfer from larger models to smaller ones1.
Speculation surrounds the hardware used by DeepSeek, particularly given US export controls on advanced chips. While DeepSeek has mentioned using Nvidia A100 chips in the past, which are restricted, Nvidia declined to comment on the specifics. A source estimated that DeepSeek might have used around 50,000 Nvidia chips for its technology1.
The success of DeepSeek's models underscores a trend towards open-source AI development, which is gaining momentum. Clem Delangue, CEO of HuggingFace, had predicted that a Chinese company would lead in AI due to the rapid innovation in open-source models. "This went faster than I thought," he noted1.
As thousands of developers flock to try out DeepSeek's model, the implications are clear: the future of AI may not be about who can build the largest models but about who can innovate most efficiently.
Will Knight, WIRED https://www.wired.com/story/deepseek-executives-reaction-silicon-valley
Image Credit: WIRED Staff, Lam Yik/Getty Images