I asking about the description of annoted buttons in the prompt. and I get the below response.
I am getting below output. which is very poor.
“1”: “Gmail link”
“2”: “Images link”
“3”: “Google Search button”
“4”: “I'm Feeling Lucky button”
“5”: “Google Account button”
“6”: “Google apps button”
“7”: “Settings button”
Instead of this if I send the individual cropped annoted images, I get very good result.
Any Idea what needs to be improved in the first prompt.
Unrelated to your prompt or the output, but rather programmatic image analysis in general: you might instantly have better results if either a) the screenshot was taken at a lower resolution, or b) you increased the zoom level of the browser before taking the screenshot.
I’m not sure if either of those are an option for you, but it might help.
did you try things like making it define it and what it thinks the word means before you prompt it? or tell it to search bing for some examples of what It thinks you’re asking it for. then rule out the ones that aren’t right. or tell it to give you another word that means the same thing. it helps if it knows what knowledge to focus on before telling it to do it. its like if I said list different bats it may look for gross bats and not a baseball bats. but if just before I prompted it we were talking about baseball and then I said list bats. it wouldn’t look for ugly gross flying rats with wings. my point is talk to it about the subject for a brief time before prompting. just stuff that has helped me
