Enhancing Education with ChatGPT 4o and Microsoft Copilot: A Review of Opportunities, Challenges, and Student Perspectives on LLM-Based Text-to-Image Generation Models

Abstract

This review examined the educational potential and challenges of integrating large language model (LLM)-based text-to-image generation tools using the enhanced OpenAI ChatGPT 4o Image Generation model as a central case study. While the examples were primarily drawn from ChatGPT 4o, the insights and findings broadly applied to other text-to-image models across platforms. Through an extensive analysis of interdisciplinary literature and classroom practices, this review identified how text-to-image generation supported various applications. These included fostering creative storytelling, enhancing curriculum design, visualizing abstract STEM and medical concepts, reconstructing historical settings, supporting language acquisition, and promoting inclusivity in special education. Educators leveraged these tools to generate customized instructional materials. At the same time, students engaged with them to visualize concepts, develop richer descriptive language, and explore global cultures through personalized, image-driven learning experiences. Moreover, the review also revealed challenges, including critical technical, ethical, and pedagogical challenges. Technical issues included inconsistent image accuracy, prompt sensitivity, and resource access disparities. Ethical concerns involved algorithmic bias, potential misinformation, content filtering, and intellectual property rights. Pedagogically, educators needed to ensure alignment with learning objectives, assess AI-assisted student outputs effectively, and avoid overdependence on automation. In addition, this study incorporated qualitative data from an 8-week classroom program conducted with primary school students who used Microsoft Copilot to generate images from text prompts. The findings highlighted students' high engagement, growing descriptive vocabulary, increased cultural awareness, and emerging critical understanding of AI's limitations and biases. Students reported a sense of agency and creativity in crafting prompts, collaborated with peers to refine their outputs, and demonstrated early-stage digital literacy skills. These real-world classroom insights provided grounded evidence of how thoughtfully implemented text-to-image tools could enhance educational outcomes.

Ähnliche Arbeiten