When I upload a pdf through the chatgpt UI, I sometimes get remarkably different results than using the assistants API with only that file in the vector store. Does anyone know if the UI vectorizes the pdf in the same way the assistants API does or if it’s using a text conversion tool to essentially put the pdf text into context? When asking the same questions or the assistants API, I am using the maximum number of returned vectors.
When you say uploaded to the UI are you using the clip icon or do you have a custom GPT?
The attachment icon in the UI. Not using any GPTs.