Hey guys, I am using the new image recognition feature and it generally works very well. I was surprised that a task I thought was easy is instead failing continuously. This is the prompt and the image.
Looking for insights. Any ideas?
“You are an expert analyst of hexagonal game maps, keeping in mind that the hexagonal grid starts from B1 and has a maximum of 4 rows (so max B4) and has 17 columns (so it goes up to R) and they go from top to bottom (i.e., B1 is at the top and descends towards B2/B3 etc.). Give me the coordinates of the blue units you find.”
I don’t have GPT-Vision yet… but I would suspect it doesn’t understand “blue units”… might try “light blue square tokens” or something.
Looks like a really interesting use-case, though!
Thank you for giving it a try. I’ll take a closer look at the positions and let you know. But at first glance, it seems like your result is better than mine.
"It definitely recognized all the units, something that my prompt wasn’t doing. It would always miss one or two.
What surprised me is that even when I gave it another image representing a unit, it still couldn’t provide the correct answer. I’ll start running some tests with your suggestion, thanks again!