Measuring help with daily responsibilities
OpenAI described LifeVal on November 2: an evaluation, or set of tasks used to test AI, focused on everyday responsibilities. Time-use studies and a 3,995-person survey informed task selection. [1]
OpenAI shows best-recorded GPT-6 Astra scores, including 100% examples. We have not reproduced the tests or verified their full method. [1]
Our take
The useful test is whether the assistant preserves the details that matter to you. For a routine supply order, those might be the items, quantities, cost, and delivery address. Ask it to prepare the order, then check those details before approving a purchase. A convincing answer or a perfect selected example does not replace that review.