Ken Wenger, Founder and CTO at Squint Cognition, demonstrates the company’s squinting model technology at the 2026 Embedded Vision Summit. Using Skittles and M&Ms, Wenger shows where a standard YOLO object detection model confuses one candy for the other; and does so with high confidence, offering no signal that anything is wrong.
He then runs a squinting model from Squint Vision Studio alongside YOLO. The squinting model predicts when YOLO is about to make a confident mistake, catches it, and corrects the output in real time. The candy is a joke; the failure mode isn’t.
