ChatGPT and DeepSeek in Physics Education: A Narrative and Thematic Literature Review of Pedagogical Implications

This paper presents a narrative and thematic literature review examining the pedagogical implications of integrating two leading large language models, ChatGPT and DeepSeek, into physics education. It investigates how their distinct architectures, interaction styles, and affordances influence conceptual understanding, problem-solving practices, and instructional design. ChatGPT's dialogic and adaptive nature aligns with constructivist and inquiry-based approaches, fostering metacognitive engagement and conceptual change through scaffolded explanations, analogies, and multimodal representations. Conversely, DeepSeek emphasizes computational efficiency, precision, and iterative refinement, making it particularly effective in structured problem-solving contexts and high-stakes learning environments. By synthesizing recent empirical and conceptual studies, the review identifies complementary pedagogical roles for these tools. It emphasizes the importance of strategic teacher mediation to avoid cognitive passivity and ensure meaningful learning. Ethical and practical challenges, including academic integrity, data privacy, algorithmic bias, and teacher professional development, are critically examined. The paper concludes by outlining future research directions for integrating generative AI in physics education, emphasizing the need for theoretically grounded, classroom-based studies that address both pedagogical opportunities and epistemic implications. This review provides a comprehensive foundation for understanding how AI can reshape physics teaching and learning in a rapidly evolving educational landscape.