Hello Team,
I am working on a coursework which has a graph in it. I find it difficult to attach the graph on Chatgpt. Kindly assist
Hi,
you cannot use images on the current GPT Models, only text Input. So you might have to describe the content of the Image to provide a information basis for your reply. Maybe a image to text tool could help you? BLIP - a Hugging Face Space by Salesforce this can describe the content of a picture to you, I however do not know if this is viable for graphs…
1- Using mermid or similar syntaxes
for example,
graph LR
A[Square Rect] -- Link text --> B((Circle))
A --> C(Round Rect)
B --> D{Rhombus}
C --> D
2: Using embedding : language do have shape and colors as the vector representation of a cat and a dog is different. (this is the advance topic but embadding sometimes is helpful.)
Hi could you please elaborate a bit on how to use embeddings for this use case? This is the dev forum so even if its more advanced I’d really appreciate if you would take the time to do so 
What is hard for me to grasp is how to convert the graph the user is talking of into a form where I can pass this into the embedding?
It is an advanced topic as I do not have enough knowledge about it.
Embedding is helpful when you want to compare or classify. because the result of feature projection is not the same, I guess it will be difficult to get the exact inputs from the image.
The idea is basically to convert the image into vector space using clip and to differentiate it based on the position of the vector space.
Openai is working on this problem and has the solution in gpt4.
1 Like
I like your enthusiasm but in this case embeddings are really not a solution. Embeddings work so that you build your vectors from the text (and then in a next step visualizations) but the other way around it doesn’t work.
Please let me specify: What you proposed with clip would be a good first step in general, but in the context of this usecase I don’t see the benefit.
For example the above shown example from the link you’ve send shows that we can classify but do not get deeper insights in the picture. This would be required if you want to use the graph to answer the coursework.