DeepSeek tests “sparse attention” to slash AI processing costs

DeepSeek, a Chinese AI lab, has released version 3.2 of its model, which explores a technique called "sparse attention" to reduce the processing costs associated with running AI systems. Sparse attention is a method that aims to reduce the computational requirements of attention mechanisms, a key component in many modern AI models. The new release claims to significantly decrease the computational resources needed to run AI models, making them more accessible and cost-effective, particularly for resource-constrained environments. This could have implications for the widespread adoption of AI in various applications, as the reduced processing costs may make it more feasible for organizations and individuals to utilize these technologies. The article highlights the potential benefits of this approach, which could make AI more accessible and widely deployed, especially in situations where computational resources are limited. However, the article does not provide detailed technical information or independent assessments of the claims made by the DeepSeek team.