The Working Notes: EP2 - How well does the model understand the physical world

Roberto and Marianna, the researchers behind PhysicalRealismBench-U, walk through how the benchmark works, how the dataset was built, and what surprised them most about frontier VLM performance on physical reasoning tasks. Can a VLM tell if an object is behaving physically correctly in a video? The results are more interesting than you'd expect. Learn more about PhysicalRealismBench-U, and submit your model to learn how it fares here: https://link.reka.ai/labs/physical-realism-bench-U #PhysicalAI #AI #RekaLabs #ComputerVision #VLM