DeepSeek is a new AI startup from China that’s creating quite a buzz in Silicon Valley. Founded in just 2023, this company has developed impressive AI models that perform as well as some of the top chatbots, but at a much lower cost.
DeepSeek’s rise challenges the common belief that developing AI requires massive power and resources. As news of their innovations broke in late January, it sparked concern among investors, particularly for US tech companies and their suppliers.
So, what makes DeepSeek special? The company was started by Liang Wenfeng, who previously led a hedge fund focused on AI. DeepSeek offers open-source AI models, allowing the developer community to improve and expand their technology. Their mobile app quickly climbed to the top of Apple’s download charts in the US after its release in early January.
The DeepSeek app is different from competitors like ChatGPT. It explains its reasoning process before providing an answer to users. The company claims that its R1 model matches the performance of OpenAI’s latest models and allows others to build chatbots using its technology.
Though the specifics aren’t fully disclosed, it seems that training DeepSeek’s models costs considerably less than developing similar models from OpenAI or Meta Platforms. This efficiency raises questions about the need for high-priced AI hardware from suppliers like Nvidia. Meanwhile, US restrictions on technology exports to China seem less effective now, as DeepSeek’s success indicates that Chinese engineers are finding ways to innovate despite these barriers.
DeepSeek’s R1 model has shown strong results in respected benchmarks that test mathematical skills, general knowledge, and question-and-answer capabilities. It even competes well on rankings from institutions like UC Berkeley.
DeepSeek gained international attention when it launched its reasoning model in November 2023. By January of the following year, the app had been downloaded 1.6 million times and ranked first in iPhone app stores across multiple countries, including the US, Canada, and Australia. This success captured the eye of influential investors, with some suggesting it marks a pivotal moment for AI development.
Liang, the founder, has a strong background in electronic engineering and started DeepSeek with modest funding. He believes the main hurdle to progress is not financial support but rather access to advanced technology, which is restricted by the US. His vision is for China to build its own tech ecosystem similar to what Nvidia has created.
In the rapidly evolving AI landscape in China, DeepSeek differentiates itself with an open-source model while major players like Alibaba and Tencent focus on capital and resources. This unique approach has been instrumental in lowering costs for AI development in China, leading to increased competition and price reductions in the industry.
DeepSeek’s achievements might force rivals like OpenAI to reconsider their pricing strategies to remain competitive. Additionally, it raises questions about the current investment strategies of tech giants that are spending billions on AI infrastructure.
As DeepSeek’s technology gains traction, developers worldwide are starting to take notice, potentially accelerating the adoption of AI reasoning models. However, this rapid growth could bring discussions about the need for regulations in AI technology.
While DeepSeek has made impressive strides, it too has limitations. Like other Chinese AI models, it censors certain sensitive topics, avoiding controversial subjects or figures. The sudden interest in DeepSeek has also put pressure on its cloud infrastructure, with the company facing service disruptions as more users join.
Source link
Bloomberg, China, Liang Wenfeng, comparable performance
