OpenAI introduced LifeVal, a benchmark for AI performance on everyday tasks. It covers areas like childcare, finances, and government paperwork. The benchmark draws on time-use surveys and a new 3,995-person study. Example tasks include enrolling in kindergarten, managing flood volunteers, and finding a home. Models scored between 93.1% and 100% on these examples. LifeVal aims to track progress toward AI that helps with practical, real-life needs. OpenAI used it to improve model performance over the past year. The goal is to move ChatGPT beyond chat toward a personal assistant role.

• 3,995 survey respondents
• 93.1% to 100% model scores on example tasks