According to MIT Technology Review, a new AI model released by China's DeepSeek last week uses new technology that has the potential to significantly enhance AI's "memory" capability. This is an optical character recognition (OCR) model that works by extracting text from images and converting it into machine-readable text. This is the same technology used by scanning applications, photo text translation, and many assistive tools. According to the paper and some early reviews, DeepSeek's new model performs comparably to top models in key benchmark tests. The main innovation of this model lies in its information processing method - particularly the way information is stored and retrieved. Improving AI models' "memory" capability can reduce the computational resources required for operation, thus lowering the growing carbon footprint of AI itself.
Image source: Internet
Original: www.toutiao.com/article/1847654171317386/
Statement: This article represents the views of the author himself.