Скачать или смотреть VoiceOver image description vs GPT4 Vision

While exploring for ways to improve the image descriptions in the RealityCheck apps viewport, I decided to create a tiny app that takes screenshots of the visionOS simulator and sends them for analysis to the gpt4-vision API. After that, I compared the results using the same simulator states between VoiceOver image description and GPT
#Accessibility #visionOS #gpt4 #a11y #macOS #VoiceOver

(caption:  visionOS simulator is open alongside the custom app. VoiceOver is turned on, with the captions also on. First, VoiceOver is asked to describe the image of the home screen by default in the living room by night scene, and then I ask GPT to do the same. After that, I return to the simulator and manually open and navigate to one of the sections of hello world, asking again for a voiceover to describe what it sees. Then, I repeat the same with the custom app)