Mon Jan 27 12:51:38 UTC 2025: ## Chinese AI Startup DeepSeek Challenges US Giants with Low-Cost Open-Source Model
**Beijing, China** – A relatively unknown Chinese artificial intelligence (AI) research lab, DeepSeek, is making waves in the global AI industry. The company’s newly launched open-source model, DeepSeek-R1, is not only competing with industry giants like OpenAI’s ChatGPT and Google’s Gemini, but in some areas, outperforming them—all while achieving this with significantly lower costs.
Founded in 2023 by Liang Wenfeng, DeepSeek operates independently, without the backing of major Chinese corporations like Baidu or Alibaba. Liang, who previously worked in the financial industry, diverted resources from a hedge fund to pursue his vision of advancing AI research through scientific discovery.
DeepSeek-R1’s success stems from innovative training techniques. By utilizing technologies like multi-head latent attention (MLA) and mixture-of-experts, DeepSeek reportedly trained its model using only 10% of the resources required for Meta’s Llama model. The model exhibits strong reasoning capabilities and excels in tasks such as mathematics and coding. Furthermore, DeepSeek has made its model and smaller versions available to developers under the MIT license, fostering collaboration and customization.
DeepSeek’s achievement is particularly noteworthy given the US government’s restrictions on advanced chip exports to China in 2022. DeepSeek circumvented these limitations by employing cost-effective training methods including custom communication schemes for efficient data sharing between chips, memory optimization techniques, and a mix-of-models approach.
The open-source nature of DeepSeek-R1 is a significant factor in its challenge to established Western AI companies. By making the technology freely available, DeepSeek is democratizing access to advanced AI and directly challenging the dominance of US-based tech giants. This development has reportedly sparked a renewed focus on AI development in the United States. DeepSeek’s success demonstrates that breakthroughs in AI are not solely dependent on massive investments and access to the most advanced hardware.