The Russian neural network YandexGPT 2 took several versions of the Unified State Exam (EGE) in literature, successfully 'enrolled' in the university by passing the minimum threshold score. Her overall result was close to the average score of Russian students. It scored 55 points compared to the average of 64 points for students in 2022.
For testing, real versions of the USE in literature from the training exams in 2021-2023 were used. The YandexGPT database used for training did not contain answers to them.
As noted in the press service of 'Yandex', this is the first case in Russia when a generative neural network passed the EGE with various types of tasks requiring knowledge of the text, its linguistic and cultural context.
YandexGPT 2 successfully coped with the tests, analysis of the work and writing an essay on a given topic to assess erudition, writing style and creative abilities. All her answers were entered into the forms and sent to the official EGE experts in literature for checking.
Students prepare carefully for the USE in advance and delve deeply into the specifics of the exam. The neural network didn't do that, its approach is similar to the situation when an adult solves tasks, relying only on accumulated knowledge. The results of YandexGPT 2 show its erudition and ability to analyze information.
Director of the Moscow Center for Continuous Mathematical Education Ivan Yaschenko
Earlier, the developers of the domestic neural network have already taught it to retell videos in Russian and to retell large print texts in thesis.