China's DeepSeek has taken the artificial intelligence world by storm, outperforming major players such as Meta, OpenAI, and Anthropic in third-party benchmark tests. The lab's large-language model, released in late December, demonstrated superior accuracy over Meta's Llama 3.1, OpenAI's GPT-4, and Anthropic's Claude Sonnet 3.5. Built in just two months and costing under $6 million, DeepSeek's achievements have sent shockwaves through Silicon Valley, raising concerns about America's waning dominance in AI technology.
DeepSeek, a lab emerging from the Chinese hedge fund High-Flyer Quant, has maneuvered through stringent U.S. semiconductor restrictions. These restrictions have barred China from accessing the most advanced chips, such as Nvidia's H100s. Despite these limitations, DeepSeek built its model using Nvidia's reduced-capability H800s. This innovative approach not only allowed them to circumvent the restrictions but also led to the creation of a highly efficient and effective AI model.
"Because they had to figure out work-arounds, they actually ended up building something a lot more efficient." – Perplexity CEO Aravind Srinivas
The open-source model from DeepSeek can tackle complex problem-solving, math, and coding tasks with remarkable precision. Additionally, DeepSeek released a reasoning model called r1 that has outshone OpenAI's o1 in several third-party tests. This development comes at a time when other tech giants are rapidly advancing their AI capabilities. ByteDance, TikTok's parent company, recently claimed that its updated model surpasses OpenAI's o1 in key benchmark tests.
"They can take a really good, big model and use a process called distillation," – Benchmark General Partner Chetan Puttagunta
These advancements have prompted questions about the necessity of massive investments by big tech companies in AI models and data centers. As DeepSeek's cost-effective approach gains traction, it challenges the traditional narrative that significant financial resources are required to develop competitive AI technologies.
The founder of DeepSeek, Liang WenFeng, remains relatively unknown but has achieved significant breakthroughs with his team. Their success highlights how necessity can drive innovation and efficiency.
"Necessity is the mother of invention." – Perplexity CEO Aravind Srinivas