Raw LLM Responses

The point is it doesn't replace said labour at this point. In the image AI world there are a bunch of single task AIs that have been well demonstrated to have a similar efficacy to human radiologists, but they can only answer one question - stroke ?y/n that sort of thing. This means that a radiologist still needs to review the images, because missing a tumour or signs of other diseases is not acceptable. There is also the problem that these studies are generally not really representative of the real world. They present an AI or radiologist with images of a specific disease, and images that are totally normal, with none of the normal noise of incidental findings. There isn't the normal accompanying info or previous scans available. If the AI outperformed humans, you could make the argument they improved care, but as it stands they don't seem to add much. I have no doubt that with time and research more general AI will come to be, and these gaps in the research will be filled, but at the mo it isn't there.