Core Insights - The forum at the 2025 World Artificial Intelligence Conference (WAIC) highlighted the challenges of reliability in specialized applications of large models, emphasizing that "hallucination" is an inherent cost of intelligence rather than a flaw to be eliminated [1][4] - Ant Group's Ant Mi-Suan announced the open-sourcing of the High-Order Program (HOP) framework, aiming to enhance the reliability of large models in professional applications through a combination of intelligent and engineering systems [2][4] Group 1: Reliability Challenges - Professor Chen Chun from Zhejiang University pointed out that while current technology has reached 80% to 90% of the state-of-the-art (SOTA), there remains a significant gap in reliability for specialized fields like healthcare, industry, and finance [1] - Chen emphasized that eliminating all "hallucinations" would reduce large models to mere retrieval tools, suggesting that breakthroughs in reliability should focus on constructing an engineering assurance framework [1][4] Group 2: HOP Framework Implementation - The HOP framework aims to transform complex standard operating procedures (SOPs) into executable processes and code, enabling intelligent orchestration and automated execution in financial risk control [3] - By integrating HOP technology, the modeling cycle can be shortened while ensuring high precision, thus reducing processing costs and alleviating the shortage of specialized talent [3] Group 3: Industry Applications and Comparisons - The analogy of large models to the electric motor system of new energy vehicles was made, indicating that while the intelligence engine is crucial, the reliability assurance relies on the control system [4] - Both Chen and Wei Tao highlighted that "hallucinations" should not be viewed solely as negative elements; rather, they can reflect a certain level of intelligence development and should be managed through a combination of intelligent and engineering approaches [4]

Core Viewpoint - Ant Group's Ant Misuan announced the open-source High-Order Program model trust application technology framework to address reliability issues in large models for specialized applications, aiming to accelerate their large-scale application in professional fields [1][2]. Group 1: Reliability Challenges - The primary challenge for specialized applications of large models is reliability, with current top technologies achieving only 80% to 90% accuracy, which is insufficient for fields like healthcare, industry, and finance [1]. - The term "hallucination" is used to describe the reliability issue, but it is viewed as an inevitable cost of intelligence rather than a problem to be eliminated entirely [1]. Group 2: Engineering and Intelligence Integration - The chairman of Ant Misuan, Wei Tao, emphasized that overcoming reliability issues should not rely solely on perfect large models but rather on the integration of intelligent and engineering systems [2]. - The High-Order Program framework draws from successful human engineering management experiences, incorporating mechanisms like standard operating procedures and checklists to ensure reliability even in the presence of hallucination issues [2]. Group 3: Application and Future Potential - The High-Order Program framework has been preliminarily applied in various industries, including financial risk control, network intrusion detection, and medical billing, showing significant improvements in reliability and timeliness [2]. - Wei Tao believes that solving the reliability problem will lead to the emergence of new "Killer Apps" for large models, highlighting the importance of engineering and intelligence fusion for reliable applications [3].