The rapid integration of Generative Artificial Intelligence (GenAI) into higher education has sparked a profound pedagogical debate: Does generative technology expand students' creative frontiers as an interactive cognitive scaffolding tool, or does it homogenize intellectual exploration and erode originality? This study investigates the dualistic impact of large language models (LLMs) on undergraduate students' divergent thinking, conceptual originality, and domain-specific creative problem-solving within the Vietnamese higher education sector. Employing a mixed-methods sequential explanatory design, we conducted a controlled experimental study (N=480) across three major universities in Hanoi, Da Nang, and Ho Chi Minh City. Participants engaged in standardized creative concept-generation and essay-design tasks under three randomized experimental conditions: (a) Control (unaided cognitive generation), (b) Unconstrained GenAI Access (direct text generation via commercial LLMs), and (c) Structured Scaffolding (prompt-guided iterative dialogue emphasizing critique and divergence). Outputs were evaluated using the Consensual Assessment Technique (CAT) by an expert panel, complemented by Torrance Tests of Creative Thinking (TTCT) metrics (Fluency, Flexibility, Originality, and Elaboration) and computational semantic distance analysis. Quantitative findings revealed an empirical paradox: while unconstrained GenAI access significantly increased task fluency (F2,477=48.21,p<.001,η^(2)=.168) and structural elaboration, it produced a statistically significant reduction in conceptual originality (d=-0.58,p<.001) compared to the unaided control group. Conversely, students operating under structured metacognitive scaffolding achieved superior overall creativity profiles, balancing semantic breadth with genuine novelty. Qualitative prompt-log analyses (n=36) revealed widespread "early-stage design fixation" and passive cognitive offloading among unconstrained users. This paper articulates the Cognitive Anchor-Rebound Model (CARM) and provides actionable instructional frameworks for educators seeking to harness AI as an intellectual collaborator rather than an algorithmic crutch.
Keywords: Generative Artificial Intelligence, Higher Education, Student Creativity, Divergent Thinking, Originality, Cognitive Scaffolding, Design Fixation, Vietnamese Higher Education.