OpenAI has introduced LifeVal, a benchmark focused on measuring how well AI systems assist with everyday tasks. The evaluation covers practical needs including finding childcare, accessing government services, creating budgets, preparing for school, adapting recipes, ordering groceries and planning workouts.
The company says that useful assistance requires more than providing advice or generating a chat response. A personal assistant should also be able to take actions that help complete tasks. LifeVal was created to track progress in these workflows and support improvements in model performance over time.
To identify the activities where people most value AI assistance, OpenAI used data from the American Time Use Survey and the Multinational Time Use Study. These sources catalog everyday activities across demographics, times of day and seasons. OpenAI also conducted a demographically representative survey of 3,995 respondents, asking which tasks they would most want help with from a personal assistant.
The resulting evaluation covers areas such as caregiving, health, education, household administration, personal finances, shopping and access to government services. OpenAI says its latest models perform very well on many LifeVal tasks.
According to the company, work over the past year has improved model capabilities in browser use, computer use and connectors. These functions expand the range of tasks ChatGPT can support and are intended to move it beyond a conventional chat tool toward a more agentic personal assistant.
OpenAI says LifeVal is intended to complement academic evaluations by measuring how effectively AI helps people navigate decisions, manage everyday responsibilities and complete meaningful tasks.
Sources: OpenAI News.