In the wake of the imminent exit of the Chinese-run short-form video platform TikTok in the United States, another Chinese company is making global waves. DeepSeek, an AI startup, has disrupted the US market and emerged as the most downloaded app on the App Store, according to a report from AP.
Dubbed “China’s ChatGPT,” DeepSeek is an open-source platform built on a fraction of the budget typically required for similar projects in the United States. On Jan. 23, the journal “Nature” noted DeepSeek for its “advanced reasoning model,” which has captured the attention of scientists and users alike.
The company, founded by Liang Wenfeng, introduced its first-generation reasoning models, DeepSeek-R1-Zero and DeepSeek-R1, on Hugging Face, a popular platform for AI experts and enthusiasts. According to DeepSeek, the R1-Zero model, trained through large-scale reinforcement learning (RL) without supervised fine-tuning (SFT), showcases remarkable reasoning abilities.
“With RL, DeepSeek-R1-Zero naturally emerged with numerous powerful and interesting reasoning behaviors,” the company said.
Open source
However, as a first-generation model, DeepSeek-R1-Zero is not without flaws, including issues such as repetitive outputs, language mixing, and poor readability. These challenges spurred the development of DeepSeek-R1, an upgraded version incorporating cold-start data before reinforcement learning.
“DeepSeek-R1 achieves performance comparable to OpenAI-o1 across math, code, and reasoning tasks,” the company claims.
Like OpenAI, DeepSeek has open-sourced both DeepSeek-R1-Zero and DeepSeek-R1, as well as six dense models distilled from DeepSeek-R1, based on Meta’s Llama and Alibaba Cloud’s Qwen.
The launch of DeepSeek has caused ripples in the US AI market, leading to stock declines for major AI players such as Nvidia, Broadcom, and Google’s parent company Alphabet. DeepSeek’s rise signals a shift in the AI landscape, with the Chinese startup challenging established leaders and reshaping the competitive dynamics in the industry.