The recent advancements in the field of large language models (LLMs) have significantly impacted various domains, particularly in the areas of text generation, detection, and evaluation. A notable trend is the development of more sophisticated methods for detecting AI-generated texts across different domains, addressing the limitations of existing binary classification approaches. Innovations such as multi-level fine-grained detection frameworks and domain-aware fine-tuning techniques are enhancing the accuracy and robustness of detection models, making it feasible to build comprehensive systems that can operate effectively across various contexts. Additionally, there is a growing focus on integrating qualitative rationales into automated essay scoring systems, improving the reliability and interpretability of scoring models. These developments not only advance the technical capabilities of LLMs but also address ethical concerns related to authorship and integrity in academic and creative writing. Furthermore, the exploration of LLMs' stylistic tendencies in poetry generation and the analysis of their rhetorical styles in general writing provide deeper insights into the nuances of AI-generated content, highlighting both their capabilities and limitations. Overall, the field is moving towards more nuanced and context-aware applications of LLMs, balancing innovation with the need for robust detection and evaluation mechanisms.