Tinted Frames: Question Framing Blinds Vision-Language Models

Y	Hacker News new \| ask \| show \| jobs

	Tinted Frames: Question Framing Blinds Vision-Language Models (davidhalladay.github.io)
	4 points by ritwikgupta 85 days ago

1 comments

ritwikgupta 85 days ago

VLMs are selectively blind — they decide how much to look at an image based on question framing (open-ended vs Yes/No vs MCQ), even when the same visual reasoning is required.

link