The large GigaChat MAX language model developed by Sber gave correct answers to 83% of questions in the test that must be passed when obtaining primary accreditation for local general practitioners.
Image source: sber.ru
To successfully pass the test, medical school graduates must answer 70% of the questions correctly. But the GigaChat MAX model did not stop there – it correctly solved 20 situational problems with a norm of at least 17 correct answers, writes Kommersant with reference to the Sber press service. The research project was carried out on the basis of the accreditation center at Sechenov University.
This is not the first success of the Sbera neural network in the field of medicine – in February of this year, the GigaChat model passed the exam for sixth-year students, which is necessary to qualify a young specialist as a general practitioner. Over the past nine months, the neural network has been trained with the participation of several hundred doctors, Sberbank noted.