The Short Version
Apple published SFI-Bench, a 1,555-question test measuring whether AI understands object function and repair, two years after Google showed similar spatial reasoning at I/O 2024.
Apple built SFI-Bench, a test of 1,555 expert-written questions, to measure whether AI can understand not just where objects are but what they do and how to repair them when broken. Google was already demonstrating this kind of spatial reasoning at its 2024 I/O conference. Apple is now formally measuring a skill its competitor showed off two years ago. Building the ruler is not the same as winning the race.