MOSCOW, September 12 The YandexGPT 2 neural network scored on the Unified State Exam the score required for admission to a university — its average score was 55 points; This is the first time in Russia that a generative network has passed this exam and completed all the tasks, the Yandex press service said.
«The YandexGPT 2 language model solved several variants of the Unified State Exam in literature. The neural network came close to the average score (63 at the end of 2022) in this subject among Russian schoolchildren who choose literature and specially prepare for the exam. It also exceeded the minimum threshold required for admission to university, — 40 points. Its average score was 55 points. This is the first time in Russia when a generative neural network passed the unified state exam, completing all its tasks, including the creative one,» they reported.
It is noted that specialists received versions of the Unified State Examination in literature, which were used in conducting training exams in 2021-2023, from the Moscow Center for Continuing Mathematical Education and were convinced that the YandexGPT database used for training did not contain answers to them.
«The neural network solved tasks in almost the same way as any schoolchild would do: it generated answers to open questions and selected one or more options from those proposed, if any. After this, the solutions of the language model were entered into the forms and sent for verification to official Unified State Examination experts in literature,” they added.
YandexGPT 2 is a new neural network model that can create and process texts, offer new ideas and takes into account the context of the conversation with the user.