Doctors w/o GPT-4: 73.7%
Doctors with GPT-4: 76.3%
but GPT-4 alone: 92.1% 
Not looking good for human doctors. It was a timed test, so maybe some anxiety caused wrong answers, but the control group was allowed all their usual online resources except an LLM. The study says that the doctors were selected from various institutions, but didn't disclose where. First, machine learning models are great when you present them with nicely packaged clinical vignettes. The trick is to tease out the salient elements. But secondly, that doesn't exonerate the human doctors: they deliberately chose to ignore the right diagnosis in favor of something else. This dispels the notion that AI can help clinicians be better doctors.
Thirdly, it foreshadows the future of medicine – that diagnosticians will be replaced by human nurses who are paid to create clinical vignettes for a model to parse. Real geniuses will go into research and industry, and eschew clinical medicine, also largely because it will pay far less than their earning potential.
Yikes.Spruce Pine is one of the only places in the world to mine high-purity quartz.The mineral is an essential ingredient of chips in countless products, including medical devices, solar panels, cellphones and the chips powering the latest tech craze: artificial intelligence. A few weeks of shutdown is not the end of the world, Conway tells Axios. However, longer than that could put the industry into "another crisis." The semiconductor industry would need to find alternatives.