clem

/

gemini

like 90

I'm not 100% sure if this is a fair comparison as ChatGPT 4 can use tools (like it did in this case), but the solution adjustment is better in my opinion since the step where the student made a mistake is more explicit

@bri25yu In your GPT-4V sample, are you using one single image as the entire prompt? I guess with text prompts, you will get better answers from GPT-4V.

Yup, just a single screenshot of the Gemini technical report input

I think that is harder than direct text prompting. Here is the result with similar prompting as Gemini.

Oh sorry @taesiri , in my earlier post I meant that the ChatGPT 4 response is better than the response included in the Gemini report haha