- DeepSeek (Liang Wenpeng)'s Expansion of Influence in the AI Market
- 70 Times Cheaper Innovation Compared to GPT-4 Turbo
[Unblock Media] Chinese AI company DeepSeek is rapidly expanding its influence in the AI market. Originally a quantitative trading firm, DeepSeek recognized the potential of AI technology and established an AI research center in 2023, shifting its focus to AI development. Particularly, the DeepSeek V2 model offers high cost-effectiveness as an open-source model, with an inference cost of just 1 yuan per million tokens—70 times cheaper than GPT-4 Turbo. This cost innovation is prompting major Chinese corporations such as ByteDance, Tencent, Baidu, and Alibaba to lower their API prices, influencing the market.
DeepSeek's founder Liang Wenpeng(梁文锋), who established the quant hedge fund High-Flier in 2015 to develop machine learning-based trading algorithms for the financial market, recognized in 2023 that AI technology could revolutionize various industries beyond finance and focused on AI development by creating the AI research lab.
The AI model from DeepSeek incorporates MLA (Multi-head Latent Attention) and MoE (Mixture of Experts) techniques, overcoming the limitations of existing algorithms and enabling more precise and efficient data analysis.
DeepSeek maximizes cost-efficiency through its innovative AI architecture. For instance, it reduced AI model training costs by 30%, generating over $500,000 in additional profit. The company also adopted a strategy of using NVIDIA's H800 GPUs to train models, thereby reducing computational costs and maximizing performance. This approach supports DeepSeek's pursuit of sustainable innovation in the AI market beyond mere technological development.
DeepSeek maintains a techno-idealist attitude, prioritizing 'right and wrong' over 'profit and loss.' Founder Liang Wenpeng emphasizes the philosophy of "having crazy ambition and being insanely truthful," focusing on the transparency and ethics of AI technology development. To practice this, DeepSeek focuses on research and tech development, opting to open-source its AI models instead of commercializing them. Moreover, it strictly adheres to a data ethics code, rejecting illegally collected data and using only validated data for AI model development. An example project excluded illegal data usage and utilized ethically verified data for AI model development, showcasing practical ethical implementation.
DeepSeek's innovative technology development and open-source strategy are rapidly changing the competitive landscape of the AI market. Through cost efficiency and continuous R&D, DeepSeek is providing new strategic directions to Chinese AI enterprises and gaining attention in the global AI market.
It will be interesting to see how DeepSeek continues to lead AI technology advancements and market changes in the future.