Large Language Models (LLMs) such as GPT-4 are rapidly changing how students engage with knowledge in higher education. While their use in tutoring, feedback, and content generation is increasingly visible, their deeper impact on learning processes—especially in fields like Business Intelligence and Digital Economy—deserves closer attention. This paper explores how LLMs shape the way students understand, apply, and question information in data-driven disciplines.

We reflect on three core aspects of LLM-supported learning:
(1) the extent to which students rely on these models for reasoning,
(2) the types of knowledge involved—from technical facts to ethical judgments, and
(3) how actively learners evaluate and interpret the models' responses. Through classroom examples—such as scenario-based simulations and prompt-driven analysis—we illustrate both opportunities (e.g., cross-domain reasoning, iterative thinking) and risks (e.g., overreliance, misinformation, passivity).

Rather than treating LLMs as simple tools or black-box experts, we argue for their use as dialogue partners that support exploration while demanding critical oversight. We suggest teaching strategies—such as comparative analysis, reflection prompts, and justification exercises—that help students stay engaged and in control of their own reasoning. This contribution adds to ongoing discussions about how AI is reshaping learning and calls for practical approaches to keep epistemic agency at the center of educational design.

Keywords: Large Language Models, higher education, business education, AI in learning, epistemic reasoning, student engagement, critical thinking, digital economy, business intelligence.