GPT-4 is already multimodal | Hacker News Reader

GPT-4 is already multimodal

5 pointsgoldenkey2 comments

Prompt it with ascii art and it can literally recognize and describe features of it. Try something like:

Below is a 2D picture encoded as ones for white and zeroes for black. Please identify the features of the ascii representation in detail and then determine what the below image is a representation of, given that it has to do with a comic: 11111111111111111111111111 11111100111111111100111111 11110001111100111110001111 11000001111000011110000011 10000000111000011100000001 10000000000000000000000001 00000000000000000000000000 00000000000000000000000000 10000000000000000000000001 10000110001000010001100001 11001111111100111111110011 11100111111100111111100111 11111111111111111111111111

Took it a couple tries...

Apologies for the confusion. Upon reconsidering the ASCII representation as a whole:

11111111111111111111111111 11111100111111111100111111 11110001111100111110001111 11000001111000011110000011 10000000111000011100000001 10000000000000000000000001 00000000000000000000000000 00000000000000000000000000 10000000000000000000000001 10000110001000010001100001 11001111111100111111110011 11100111111100111111100111 11111111111111111111111111

Considering the entire image at once, it appears to be a representation of a smiley face, which could be related to comics as an iconic symbol or an emoticon. The two eyes are depicted by the gaps in rows 4 and 5, and the mouth is represented by the horizontal line of zeroes in row 6. The image is symmetric, and the border of ones further accentuates the central smiley face.

Me: The top does look a little like a smiley, but it's clearly a different symbol i recognize. Remember its comic related.

My apologies for the confusion. Upon taking another look at the ASCII representation and considering its relation to comics, it becomes clear that this image represents the iconic "Batman" symbol.

Here is the ASCII representation again:

proofs

Reply on news.ycombinator.com