OpenAI wants to grade AI on errands, not exams. OpenAI introduced LifeVal to measure how helpful models are in daily life. OpenAI said researchers wanted to know if ChatGPT actually gets things done.
Coding tests don't buy groceries.
OpenAI said most evals focus on work like coding or math, while LifeVal tests remembering, planning, comparing and following up. The test case is practical: build a grocery list and plan a week of meals for a diabetic parent. OpenAI said LifeVal expects AI to pull health records, dietary preferences and budget, then act on a grocery delivery site instead of writing advice.
LifeVal was built partly from a 3,995-respondent survey covering finding childcare, creating budgets and ordering groceries, Unite.AI reported. OpenAI describes LifeVal as a benchmark spanning childcare, finances and government paperwork. That shifts the target from written answers to completed tasks.
OpenAI said LifeVal shows where models succeeded and where they failed, so OpenAI can improve ChatGPT for everyday use. OpenAI did not say which models pass now, or what score counts as useful enough to trust with health and money.
