87% More Energy: The Hidden Cost of DeepSeek AI's Reasoning

DeepSeek AI, a new Chinese AI model, has been making waves with its innovative approach to energy efficiency. However, recent tests reveal that its energy consumption might not be as promising as initially thought. When compared to a similar-sized model from Meta, DeepSeek’s smaller version was found to use 87% more energy for generating responses. This is largely due to its chain-of-thought reasoning technique, which produces longer, more detailed answers. The model’s training phase benefits from a mixture of experts technique and improved reinforcement learning, but these efficiencies are offset by the energy-intensive inference phase. The AI community is now concerned that widespread adoption of such models could lead to a significant increase in energy usage, potentially negating any efficiency gains.

Source: www.technologyreview.com