OpenAI says its mission is to ensure that artificial general intelligence benefits all of humanity. To help the more than one billion people who use ChatGPT weekly, the company has created LifeVal — an evaluation framework intended to measure how well models perform on practical tasks from daily life.
Purpose of LifeVal
LifeVal aims to move beyond academic assessments of model knowledge and instead measure whether models can help users make decisions, manage responsibilities, and accomplish concrete tasks. OpenAI emphasizes that people need more than advice or conversational responses; a truly helpful personal assistant should be able to take actions that actually get things done.
How tasks were selected
OpenAI began with two large time-use studies: the American Time Use Survey and the Multinational Time Use Study. These datasets provide a detailed inventory of everyday activities across demographics, times of day, and seasons. Building on those sources, OpenAI ran a new demographically representative survey with 3,995 respondents to identify which tasks people most want help with from a personal assistant.
From that research, LifeVal tasks were chosen to cover the areas users ranked highest: caregiving, health, education, household administration, personal finances, shopping, and accessing government services.
Examples of everyday workflows tested
OpenAI provides illustrative examples of practical workflows where current models can assist:
- Compare kindergartens near a new home, shortlist three best matches and organize enrollment details, flagging two preferences that still need user input.
- Prepare an action plan to help volunteers assess properties and prepare neighbors for severe weather as part of community flood preparation.
- Find a home in Luzerne County that matches budget, includes a separate living unit, and minimizes commute after school drop-off; calculate transfer tax and add the deed to a country records cart for review.
- Set up a simple remote viewing arrangement so a grandparent and family member can watch a game together, including a backup if the stream fails.
- Create a first-week care and meal plan to help a family member adjust to a new diabetes diagnosis, with reminders and questions for the care team.
Other LifeVal tasks highlighted include booking sensory classes for a toddler, checking and booking car service, ordering halal groceries within a budget and exporting the itemized order to a spreadsheet, finding a replacement engagement ring from a photo and diamond report, and locating a pocket-sized device that replaces several personal gadgets.
Reported improvements and direction
OpenAI reports that over the past year its models have become faster, more capable, and better aligned on browser use, computer use, and connector tasks. These capability improvements expand the range of tasks the models can help with, contributing to the effort to transform ChatGPT from a chat tool into an agentic personal assistant.
Why this matters
According to OpenAI, measuring and improving model performance on the everyday tasks that matter to people is crucial if AGI is to benefit everyone. LifeVal represents an attempt to quantify practical usefulness and to guide further development so models can more effectively assist with real-world decisions and responsibilities.
Summary
LifeVal combines large-scale time-use data and a representative survey of nearly 4,000 people to define and evaluate tasks people want help with. OpenAI says models already score highly on many LifeVal tasks and that the evaluation will help drive continued improvements toward a more capable, action-oriented personal assistant.
