A benchmark defines inputs, expected outputs or judging procedures, and an aggregation method. Results are meaningful only within that setup and can be distorted by contamination, prompt choices, or repeated tuning.
An AI benchmark is a standardized set of tasks, data, and scoring rules used to compare model performance.
A benchmark defines inputs, expected outputs or judging procedures, and an aggregation method. Results are meaningful only within that setup and can be distorted by contamination, prompt choices, or repeated tuning.