On October 24th, the International Conference on AI for Engineering, a parallel session of the 2025 World Laureates Forum, concluded successfully. Centered on the theme "Engineering Intelligence: Intelligent Transformation for the Future Industrial System", the conference gathered top global scholars and industry experts to jointly outline a new blueprint for the future industry driven by deep integration with artificial intelligence.

Kai Yu, Distinguished Professor at Shanghai Jiao Tong University and Founder & Chief Scientist of AISpeech Technology Co., Ltd., was invited to deliver a keynote speech titled Towards Reliable Task-oriented LLM Agents. He conducted an in-depth analysis of the core pain point facing current large language model (LLM) agents—the "hallucination" issue—and innovatively proposed a new paradigm of "Uncertainty-Aware Reliability Alignment", pointing the way forward for building truly reliable and practical artificial intelligence.

At its core, this paradigm transcends the traditional binary response mode of AI agents—either answering queries or declining them—and adopts a range of non-deterministic actions, including asking clarifying questions, conducting follow-up inquiries, and dynamically switching tools.

Kai Yu led his team in developing the 1+N Reliable Distributed LLM Agent System based on this paradigm. The "1" refers to a central large model responsible for foundational language processing and general knowledge integration, while the "N" represents N vertical domain models optimized for specific industry scenarios. Together, they form flexible and customizable industry-specific large language models, which have been widely applied in fields such as smart mobility, smart IoT and smart office, demonstrating significant advantages in cost efficiency and response speed.