DeepSeek's Latest AI Accelerates Race for Large-Scale Models

Chinese artificial intelligence (AI) company DeepSeek has entered the world model race by unveiling its next-generation AI model, 'V4,' capable of processing long prompts. V4 features a novel design that efficiently handles significantly longer texts than previous generations. This is considered a crucial advancement in the competition to develop 'world models,' which are essential for AI to understand and interact with the complex physical world.

The V4 model, despite being open-source, demonstrates performance comparable to the closed-source models of leading players like Anthropic, OpenAI, and Google. Furthermore, as the first release optimized for Huawei's Ascend chips, it carries the significance of testing China's indigenous chip ecosystem.

Alongside DeepSeek V4's launch, professors like Fei-Fei Li of Stanford University and founders like Yann LeCun of AMI Labs argue that world models are the key to overcoming the limitations of LLMs and realizing AI's potential in the field of robotics.