I was experimenting with visual chatgpt for describing an image. First impressed with the results and details then sadly realised if you ask for more details than you should, it just fills the gaps with false information :( #chatgpt #virtualchatgpt github.com/microsoft/visu…
1
0
1
186
0
Capabilities are limited with the blip image captioning and visual question answering which is good for some cases but not for all. + If you already know what to ask to which model, you can isolate the related classes and prevent openai query x 3 (most cases) and lower costs.