Threatening U.S. Dominance?

DeepSeek is a Chinese AI research lab that achieved significant attention in early 2024 for developing a competitive large language model at a fraction of the cost and time compared to major US companies. Their model, DeepSeek v3, was developed for approximately $5.6 million and took about two months to build, while matching or exceeding the performance of some leading models from companies like OpenAI, Anthropic, and Meta on certain benchmarks.

The lab developed their model using less powerful H-800 GPUs rather than the top-tier H-100 chips that were restricted from export to China. They achieved efficiency through innovative approaches to model training, including clever solutions for mixture-of-experts architecture and floating point-8 bit training techniques.

According to available information, DeepSeek emerged from a Chinese hedge fund called High Flyer Quant, which manages about $8 billion in assets. The company's stated mission on their developer site is to "unravel the mystery of AGI with curiosity" and "answer the essential question with long-termism."

DeepSeek released their model as open-source, making it freely available for developers to use and modify. The model has gained attention for its efficiency in terms of both computational resources and operational costs, with inference costs reported to be significantly lower than comparable models.

The company maintains a relatively low profile, with limited public information about its founder Liang Wenfeng or internal operations. Despite several attempts by media to reach someone at DeepSeek, they received no response.



This story is dsiputed by Alexandr Wang, who says DeepSeek has about 50,000 NVIDIA H100s that they can't talk about because of the US export controls that are in place.