The researcher focuses on the specific challenges of computer science education, where generative models are used to write and fix code, solve academic tasks, and explain solutions. This makes it difficult to assess whether students have independently mastered the material based on final works.
The proposed decision tree considers several factors: the type of task, the student's level of preparation, the possibility of automatic verification, and time constraints. For instance, the researcher suggests stricter limits on AI use for students in their early years of study.
The author argues that a total ban on neural networks is impractical because professional developers are already expected to be proficient in AI tools. Consequently, for senior students, the researcher proposes integrating AI into assignments and separately evaluating how meaningfully the students interact with these tools.
To ensure the same level of knowledge, the author suggests that some checkpoints must be conducted without AI assistance, such as through oral exams or tasks performed in a controlled environment. This allows instructors to teach the use of modern tools while verifying actual skills.
The author notes that some of the methodologies were tested on a small sample, and their effectiveness across different courses, universities, and in the long term needs further verification.