Measuring How Well AI Handles Everyday Tasks with LifeEval

Most AI benchmarks focus on coding or math, but researchers wanted to know how well these models perform in daily life. In this talk, the team introduces LifeEval, an evaluation designed to measure how effectively AI handles routine tasks such as taking, remembering, planning, comparing, and following up.

■ Motivation and Goal
- Why existing evaluations skew toward coding and math tasks
- What LifeEval measures: everyday task handling
■ LifeEval Design
- Task categories: taking in information, memory, planning, comparison, follow-up
- The AI must account for personal data, health records, dietary habits, and budget, then act on a grocery delivery site
- The example scenario of helping a parent of a diabetic child build a grocery list and a week-long meal plan
- What the evaluation reveals: where the model succeeds and where it does not, guiding continuous improvement

Ideal for researchers, developers, and practitioners interested in practical AI evaluation. Viewers will gain an understanding of what kinds of everyday capabilities are being assessed and how such measurements can inform ongoing model improvement.