Image quality when upload images to AI models in Duck.ai (not good for documents)
I have been using Duck.ai to help explain or translate some text documents. I noticed that some of the responses it gives are full of hallucinations. I expect some hallucinations when using AI, but the amount is alarming. I figured out that the images I uploaded (or taken as photos in the DDG app on iOS or Duck.ai’s website in Safari) get compressed so massively before being sent to the AI models, that the models cannot read the text (the image quality is too poor for OCR). So the models can see some headings in a document, and would start making up the rest. I have tried multiple documents, and the image quality isn’t good enough for models to read documents written in regular fonts (using regular sizes, maybe 12pt?) on letter (A4) sized paper. I haven’t tried PDF uploads, but that isn’t my workload.
Has anyone else noticed this? Does anyone know of a fix or workaround?