Abstract:

The rapid acceleration of Artificial Intelligence, specifically the transition from static predictive models to autonomous agentic workflows, has introduced a significant "competency drift" in higher education. Traditional, multi-year curriculum review cycles are increasingly inadequate for master programs in rapidly advancing areas, such as Artificial Intelligence, where the life of technical relevance is short, sometimes measured in months. This paper proposes a robust, dual-pronged educational framework that utilizes Large Language Models as collaborative architectural partners to maintain program-wide coherence while updating specific course content.

Our framework, the Conformity-Resistant Curriculum Cycle (CRCC), operates across two distinct academic layers: the macro-level (degree curriculum) and the micro-level (course syllabus). The technical core utilizes a Triangulated Retrieval-Augmented Generation architecture that cross-references three critical data hubs: real-time industry signals from 2025-2026 labor markets, peer-benchmarks from top-tier research universities to ensure academic prestige, and institutional prerequisite constraints to maintain degree-wide logic.

Central to this framework is a rigorous Human-in-the-Loop protocol. This "Human-Sieve" ensures that faculty remain the final arbiters of pedagogical intent, preventing "vocational drift" and mitigating the risks of AI-driven conformity bias, namely the tendency for models to mirror user preferences at the expense of factual accuracy, and the cognitive debt, defined as the blind acceptance of machine-generated structures without theoretical grounding.

We demonstrate the CRCC framework through a detailed case study: the programmatic update of the M.Sc, in Artificial Intelligence at POLITEHNICA Bucharest and the subsequent transformation of one of its core units, "CTI.M.05 Multi-Agent Systems." The system identifies gaps at the degree level (e.g., the missing competency of "agentic reasoning"), and either proposes a new course or cascades these requirements down to the syllabus level, proposing new modules on the topic. These updates are subjected to a discrepancy audit, where the Program Director and Course Instructor justify the retention or removal of core concepts, ensuring that innovation does not come at the cost of foundational rigor.

The evaluation of the CRCC framework follows a hybrid validation model tailored to institutional requirements for curricular uniformity. First, we present expert qualitative validation derived from semi-structured interviews with senior faculty subject matter experts. Second, the paper reports on a cohort analysis, prepared for comparing student competency metrics from the 2026 living syllabus iteration of CTI.M.05 against the 2025 baseline.

Our preliminary findings suggest that while AI provides the computational speed for alignment, the human-in-the-loop provides the necessary steering and pedagogical approach required for modern graduate-level AI education.

Keywords:

Curriculum Innovation, Human-in-the-Loop, LLMs.