This proposal presents the development and application of a theoretical assessment system based on the Confidence-Based Marking (CBM) methodology, utilizing generative artificial intelligence (AI) for the creation of test questions. The system has been implemented in a master's level course, specifically for the evaluation of students' knowledge in exams. The primary goal is to enhance the accuracy of knowledge assessment by incorporating students' confidence levels into their responses, allowing for a more nuanced understanding of their grasp of the material.

Currently, the AI-generated test questions are employed solely for exam purposes, providing a diverse and adaptive range of content that enriches the evaluation process. Surveys were conducted among students to assess their perception of two key aspects: the effectiveness of the CBM methodology in the examination process and the quality and relevance of the AI-generated theoretical questions. Based on the outcomes of these assessments, there is potential to extend this approach towards continuous training using the CBM methodology in the future.

Survey results indicate a positive reception of the CBM methodology, with students emphasizing its role in fostering a deeper and more critical self-assessment of their knowledge during exams. Additionally, the AI-generated questions were rated as appropriate and challenging, supporting high-quality summative assessment. These findings suggest that integrating CBM methodology with generative AI for exams holds promise for future applications in continuous training within higher education.

Keywords: Confidence-Based Marking, CBM, Generative Artificial Intelligence, Higher Education.