Raw LLM Responses

These results are being misconstrued. This is not a good look for AI replacing doctors for diagnosis. Out of the thousands of studies published in 7 years on AI for diagnostic imaging, only 14 (!!) actually compared their performance to real doctors. And in those studies they were basically the same. This is not great news for AI because the ways they test it are the best possible environment for it. These systems are usually fed an image and asked one y/n question about it: does this person have disease x? If in the simplest possible case the machine cannot outperform humans then I think we have a long, long way to go before AI ever replaces doctors in reading images. That’s also what the people who wrote the review say, that this should kill a lot of the uncontrollable hype around AI right now. Unfortunately the Guardian has twisted this to create the most “newsworthy” title possible.