Hi,
i am trying to test the gpt 4o capabilites on high school math problems , especilly the problems with illustrative figures and geometric shapes and structures in it . the usual way for any human to solve is only to understand the question by looking at the figure.
now i was expecting the model to understand the fighre and solve the problem. but it isnt, why is that.
is there any other better way to do it or ,to do better prompting technique.
If i want to create such assistant what are the tools should i add to it.
please let me know if there is a better way to do this use case.
Hiya,
Welcome to the forum!
All of the models previous to o1 and o1-mini are notoriously bad at math. These to models, on the other hand, get math questions much more accurately.
Until these models are widely available to train and to feed context to, in the meantime, you can specifically ask your GPT to solve the problem using Python and it increases accuracy.
Hi @sekharmuni003 and welcome to the community!
When it comes to geometric shapes, I’ve also struggled using GPT-4o for e.g. grid-like problems. And their Vision API docs actually state these limitations. You can try playing around with size/resolution/fidelity of the image (see here) to see if that improves things. But unfortunately for very nuanced figures, it suffers quite a bit, and this is probably intrinsic to the visual transformer (ViT) it uses under the hood.
thanks guys for your input. ill definitely use them.