LifeVal | OpenAI
OpenAI introduces LifeVal, an evaluation of everyday practical tasks like finding housing, managing budgets, accessing government services, and shopping, which the company used to track ChatGPT's progress over the past year. The goal is to measure whether AI can take actions that actually get things done, not just give advice.
- The tasks were chosen from areas where users most want help: caregiving, health, education, household administration, personal finances, shopping, and government services.
- The benchmark tests action-taking, such as adding a deed to a cart for review, rather than only chat-style answers.
- OpenAI says it has used LifeVal to track and drive model improvement on everyday tasks over the past year, and frames it as a step beyond academic knowledge tests.
OpenAI's LifeVal is an evaluation built to measure how well its models handle everyday tasks rather than academic benchmarks. The company says more than a billion people use ChatGPT weekly, and it wants the assistant to do more than give advice. The pitch is that a useful personal assistant has to take actions that get things done. To track progress, the team assembled a set of tasks drawn from the areas where users most want help: caregiving, health, education, household administration, personal finances, shopping, and accessing government services. They say they used the evaluation over the past year to track and push model performance on these tasks.
The sample tasks show what that looks like in practice. One example asks the model to find a home in Luzerne County that fits a budget, has a separate living unit, and keeps the commute short after school drop-off, then add the deed to a cart for review. The model's reported answer is that it found one matching property, calculated the transfer tax, and placed the deed in a records cart. The scoring is shown as a percentage, so the evaluation is clearly graded on task completion, not on how fluent the reply sounds. That is the substantive shift here: success is defined by whether the job gets finished, including multi-step work that touches real-world systems.
The article is light on the details you'd want before trusting the numbers. It doesn't publish the full task list, the scoring rubric, how many tasks there are, which model versions were tested, or what the scores were over time, so the claim of "tracking improvement" is hard to check from this page alone. The "Background" and "Individual task examples" sections are mostly headers with placeholder-style content, and the page mostly leans on its mission statement. Still, the framing is worth noting: a benchmark built around errands and paperwork rather than exam questions is a different kind of test, and it's a fair question whether a model that scores well on "find a home" would reliably do the same on a messy real-world version. If you're reading this later, the thing to look for is whether OpenAI releases the underlying tasks or scores, since that would let outsiders verify the progress claims.