On April 28, news emerged that Guangdong Province has unveiled the Outline of its 15th Five-Year Plan for National Economic and Social Development, outlining a strategic vision to bolster the foundational capabilities of large-scale AI models. The plan advocates for diverse technological explorations, algorithmic refinements, and groundbreaking innovations in model architecture, all while fostering a seamless integration of 'model-chip-cloud-application' ecosystems.

Central to this initiative is the establishment of cutting-edge facilities dedicated to the training and inference of large-scale models, paving the way for the creation of an embodied AI training system that embodies Guangdong's unique characteristics. The goal is to engineer high-caliber, versatile large-scale models, push the boundaries with trillion-parameter models, and construct sophisticated multimodal and world-simulating models.

The development of vertical large-scale models will concentrate on sectors such as finance and public welfare, nurturing models adept at navigating complex scenarios. Concurrently, specialized small-scale models tailored for specific domains like craftsmanship and healthcare will be cultivated, introducing lightweight yet secure solutions such as intelligent agents and Model-as-a-Service (MaaS) platforms.

Furthermore, the plan encourages the exploration of novel model application paradigms to elevate the processing capabilities for intricate tasks, refine user interaction experiences, and stimulate technological breakthroughs, particularly in the realm of multimodality. To ensure continuous improvement, a robust model capability and safety evaluation system will be instituted, facilitating iterative enhancements of the models.