WorldVQA: A New Benchmark to Sharpen Visual Knowledge in Multimodal AI

WorldVQA introduces a groundbreaking benchmark for evaluating how well **Multimodal** **Large Language Models (LLMs)** understand the visual world! This innovative approach meticulously separates knowledge retrieval from reasoning, paving the way for more accurate assessments of these powerful AI systems.

• •WorldVQA specifically tests what an **LLM** memorizes about the visual world.
• •The benchmark covers a broad range of visual entities, from common to rare.
• •It aims to set a new standard for evaluating visual factuality and reducing **Hallucination**.