AI评测转向现实任务执行，OpenAI推出LifeVal衡量个人助理能力 近日，OpenAI发布LifeVal基准测试，用于衡量AI在照护、健康、教育、家庭管理、个人财务及政府服务等日常生活场景中的表现

AI evaluation turns to real-life task execution, OpenAI launches LifeVal to measure the capabilities of personal assistants Recently, OpenAI released the LifeVal benchmark test to measure the performance of AI in daily life scenarios such as care, health, education, home management, personal finance and government services. Different from traditional evaluations that focus on knowledge questions and answers and individual abilities, LifeVal focuses on whether AI can advance tasks to an executable and deliverable state, and evaluates its ability to solve practical problems. The benchmark is based on the American Time Use Survey and is based on 3,995 respondents. Over the past year, OpenAI has continued to use LifeVal to track model capabilities. Public test cases show that AI scored 93.1% in screening kindergarten tasks for children, 98.0% in organizing flood prevention volunteers, and 90.4% in formulating meal plans for diabetics. OpenAI also emphasized that LifeVal is currently mainly used as an internal evaluation tool and has not yet fully opened the task set and scoring code to the outside world, so the relevant test results still need to be interpreted with caution. The launch of LifeVal reflects that OpenAI is further extending the focus of model evaluation from information understanding and content generation to the planning and execution of complex real-life tasks, promoting the evolution of AI from a conversation tool to a personal assistant with practical action capabilities. via AI News (author: AI Base)