LifeVal
OpenAI introduces LifeVal, an evaluation benchmark for everyday tasks, showing model performance improvements over the past year.Summary
OpenAI's mission is to ensure that artificial general intelligence benefits all of humanity. With over one billion weekly ChatGPT users, OpenAI aims to build AI that helps with practical everyday challenges like childcare, budgeting, and meal planning. To be a truly helpful personal assistant, ChatGPT should take actions that get things done. OpenAI built an evaluation called LifeVal to measure model performance on everyday tasks. They identified valued tasks through large time-use studies and a demographically representative survey of 3,995 respondents, covering areas such as caregiving, health, education, household administration, personal finances, shopping, and government services. Examples show high success rates, such as 93.1% for enrolling in kindergarten and 100% for finding a home. Over the past year, models have become faster and more capable in browser use, computer use, and connectors, moving ChatGPT towards an agentic personal assistant. LifeVal aims to measure and improve model progress on what matters in everyday people's lives.
(Source：OpenAI)