Amid accelerating globalization and digital transformation, traditional translation practices are increasingly challenged by limitations in efficiency, semantic depth, domain adaptability, and multimodal integration. Large language models (LLMs) have emerged as a disruptive force in intelligent translation, offering advanced capabilities in contextual understanding, semantic reasoning, multilingual generalization, and adaptive learning from human feedback. This study examines how LLMs transcend the constraints of conventional neural machine translation (NMT) and drive a systematic reconfiguration of the translation workflow—from intelligent pre-translation planning and context-aware generation to post-editing and quality assurance. It further explores the evolving translation ecosystem, highlighting shifts in technological infrastructures, professional roles, service paradigms, and data governance. In addressing these transformations, the study examines how these shifts fundamentally impact translation education and training, analyzing pedagogical adaptations including curriculum redesign, competency reconstruction, and the integration of human-machine collaborative learning approaches. The research outlines the trajectory of human–AI collaboration in translation, emphasizing the redefinition of translator competencies, the restructuring of service value models, and the emergence of standardized data ethics. While acknowledging limitations in LLM capabilities, particularly in processing low-resource languages and highly creative texts, this study provides both theoretical foundations for intelligent translation transformation and practical frameworks for educational adaptation, exploring future development trends and collaborative paradigms that will define human-machine translation in the digital humanities context.