This image was from some medium post : https://miro.medium.com/v2/resize:fit:828/format:webp/1*pGfGy22G0aSbpBqtrM_u-g.png
and it was the perfect depiction of what goes on.
I'm actualy done trying prompting. The article said to ask chatgpt how many cushions are in a photo and the answer is wrong. Current vision models convert images to text before reasoning. So this is completey useless when asking chatgpt how to improve photoshop lighting because i have been doing that.
Apparently a paper by Berkely and MIT said so.
Edited: Article is https://medium.com/1st-division/the-great-ai-lie-i-see-what-you-mean-365b9476ab03
[link] [comments]