Vision-language models are impressive—until you ask them something simple.A recent study shows that state-of-the-art sys...

Vision-language models are impressive—until you ask them something simple.A recent study shows that state-of-the-art systems struggle with basic visual tasks like counting shapes or detecting overlaps, achieving only ~58% accuracy on average—far below human performance So what are they actually “seeing”?AI doesn’t perceive images the way we do. It approximates, infers, guesses. And sometimes, it fails where humans succeed instantly.#AI #ComputerVision #AIRealityCheckhttps://anhnguyen.me/2024/vlms-are-blind/