DeepSeek’s Low-Budget Model Raises Questions About Regulation, Viability And AI Power
DeepSeek is preparing to launch its V4 model, following the earlier release of its V3 model, which had a notable impact on market activities. The V3 model, developed at a cost of under $6 million using lower-powered Nvidia chips, contrasted significantly…
DeepSeek is preparing to launch its V4 model, following the earlier release of its V3 model, which had a notable impact on market activities. The V3 model, developed at a cost of under $6 million using lower-powered Nvidia chips, contrasted significantly with the extensive investments made by major U.S. technology firms, such as Amazon, Microsoft, Meta, and Google, which have collectively allocated hundreds of billions of dollars toward AI development.
Potential Use of Restricted U.S. Chips
It has been reported that DeepSeek's new AI model was trained using Nvidia's advanced Blackwell chips. These chips are subject to U.S. export controls, which prohibit their shipment to China. The U.S. government has expressed concerns regarding the potential circumvention of these controls, suggesting that DeepSeek may have removed technical indicators to obscure the use of American AI chips.
Anthropic has claimed that DeepSeek, along with other labs, has engaged in large-scale campaigns to extract capabilities from its Claude model, allegedly breaching terms of service and regional access restrictions. This process, known as distillation, involves leveraging existing models to quickly acquire advanced capabilities at reduced cost and time.
Ahlert highlighted concerns regarding the effectiveness of hardware-based export controls, noting that chips often move through various intermediaries. He emphasized the importance of considering computational access, data, and talent as key factors in AI regulation.
DeepSeek is preparing to launch its V4 model, following the earlier release of its V3 model, which had a notable impact on market activities.
Collin Hogue-Spears, Senior Director at Black Duck Software
Hogue-Spears pointed out the challenges in enforcing export controls, suggesting that the focus should be on evaluating AI models based on performance and governance rather than pre-release speculation.
Thomas Randall, Research Director at Info-Tech Research Group
Randall noted that the AI industry must address transparency issues related to training data practices. He suggested that geopolitical factors often influence how accusations against companies like DeepSeek are perceived.
Aamir Qutub, Founder & CEO, Enterprise Monkey
Qutub emphasized the need for regulatory conversations to shift from hardware restrictions to accountability for model outputs. He highlighted the rapid diffusion of AI capabilities and the importance of adaptable AI strategies for businesses.
Based on reporting by techround.co.uk.