CITIC Construction Investment's Wu Chaoze: Large models are continuing to upgrade towards more efficient and reliable directions.

On July 27, at the 2025 World Artificial Intelligence Conference, the Technology, Production, and Finance Innovation Development Forum with the theme "Wisdom in CITIC, Creating New Possibilities Together", Wu Chao, member of the Investment Committee of CITIC Securities and Director of the CITIC Think Tank Expert Committee, stated that large models are continuously being upgraded towards more efficient and reliable directions. She mentioned that AI Agent will be an important direction for AI development in 2025, and large models are transitioning from "problem-solving" to being "multifaceted urban white-collar workers". The emergence of biological communities and multi-agent clusters will further enhance the performance of large models, and enterprises with data advantages and ecological construction capabilities will have greater development potential. "The application of large models is accelerating significantly, surpassing the speed of penetration seen in the previous internet revolution, and highlighting commercial potential." Wu Chao also mentioned that the performance in the multimodal field is impressive, with over 30 model updates released globally in the first half of the year, of which Chinese models account for 75%. Chinese models have significant advantages in specification compatibility, inference speed, effectiveness, and cost-effectiveness, with consumer-end focusing on social entertainment special effects and business-end specializing in marketing material creation, film and television production, leading to significant cost reduction and efficiency improvement effects. Wu Chao believes that the current training method for robot large models has evolved towards end-to-end large models + world models, with increasingly rich modal capabilities. In the future, driven by rapid iteration of large models and cost reduction in the supply chain, the commercialization of embodied intelligence represented by humanoid robots will accelerate.