The recent surge of interest in computational music creation has been greatly influenced by the advent of large generative models such as ChatGPT and Stable Diffusion. These powerful generative AI models have demonstrated remarkable capabilities, especially in the domain of text and image generation. Spurred by these developments, the music industry has also begun to explore the deployment of large models for music creation, such as MusicLM and Mu-sicGen. However, it's noteworthy that the performance and capabilities of these music-focused generative models have not yet reached the same level of sophistication as their counterparts in text and image generation. The generation of music presents unique challenges, such as capturing the intricate temporal structures, orchestrating an emotional progression, painting a sonic landscape, and managing the sophisticated interplay between various musical elements. The controllability and interactivity of the current AI-based music generation systems is unsatisfactory. In light of these considerations, a critical examination on the evolution of AI-based pop music creation techniques is both timely and essential, particularly from an industry perspective. This paper, drawn from the authors' extensive experience as senior researchers in both industry and academia, provides a comprehensive overview of AI-based music creation techniques and their practical applications in real-world music production. It examines multiple aspects including lyrics generation, melody creation, lyrics-melody matching, arranging, and audio synthesis. The review offers an insight into the evolution and application of AI techniques in actual music production, critically evaluating their advantages, limitations. Furthermore, this paper identifies challenges and potential future directions for the field, with the aspiration of contributing to the development of more intelligent and versatile AI tools that can serve the music industry more effectively.