lol why record a video of the guy and not the screen? we cant see what he is talking about. Claud 3 can read prescriptions directly from the image but the hand writing recognition is not that great. It could be improved if paired with a proper hand writing recognition model. We are waiting for the release of cheapsest model haiku which is priced to compete with GPT3.5 but when it drops it should greatly improve this kind of work.. one suggested workflow would be to take the image, run hand writing recongnition on it, then pass the image and the raw OCR results to the multimodal LLM for it to cleanup and format the data.
