With the discharge of enormous language fashions like GPT-3 and PaLM, massive techs have been experimenting with them for fairly a while now. Lately, Google additionally joined the social gathering as a response to Open AI’s ChatGPT, referred to as the MultiMediaQA, however particularly for answering medical queries.
3 Likes
Right here’s an enormous a part of the “medical AI” drawback, you’ll be able to’t get the best solutions with out the best enter. USMLE questions already provide the proper enter on a shiny silver platter in a clear-cut case with just one clear-cut reply (not fairly the identical as actual sufferers). I’d be extra bullish on utilizing pc imaginative and prescient for non-interventional Radiology/Pathology.
LLMs and different “AI” fashions have actually mildly stunned me by way of the velocity of development. However we’ve additionally seen giant failures and even primarily near vaporware (cough IBM Watson Well being cough).
I’ve induced ChatGPT to bypass the system and provides medical recommendation – it has been totally incorrect although it would sound plausible to my companion (who isn’t in medication) – with a easy Google search doing barely higher than ChatGPT – though I do just like the plausible Radiology reviews it has generated.
3 Likes
As an alternative of normal queries, what if a affected person chart is given to AI?
Suppose we feed the AI the posted bloodwork of members right here, and ask it for the organic age? Then we evaluate it with Levin, and so on.? We will additionally ask the AI what algorithm it primarily based its conclusions on.
I feel that is the important thing subject right here… Given the present trajectory the place do you assume these AI methods might be in 5 or 10 years?
