Meta has released Open-Vocabulary Embodied Question Answering (OpenEQA) to test how AI can understand the world.
OpenEQA is designed to provide sensory inputs for AI agents to gather clues from its environment and provide value to humans.
Meta envisions an embodied AI agent acting as the brain of a home robot or smart glasses....
...leveraging sensory modalities like vision to understand its surroundings and communicate in clear, everyday language.
OpenEQA could work in the wild, asking AI agents where users placed an item or if they still had food left in the pantry.
Meta plans to add AI labels to Facebook, Instagram, and Threads.
OpenEQA has over 1,600 "non-templated" question and answer pairs that could represent how a human would interact with AI.