The field of natural language processing is shifting its focus towards developing more creative large language models. Recent studies have highlighted the limitations of current models in generating truly original and sensible text, with a tendency to default to safe and generic phrasing. Researchers are now exploring new approaches to evaluate and improve creativity in language models, including the use of multi-category creative generation engines and process-oriented studies of authorial creativity. A key finding is that while models can produce fluent text, they often struggle to balance novelty and appropriateness, and may require targeted guidance to counter their tendency to regress to the mean. Noteworthy papers in this area include: Galton's Law of Mediocrity, which formalizes the tendency of large language models to regress to the mean in language. CreAgentive, which presents a novel agent workflow driven multi-category creative generation engine that addresses key limitations of contemporary large language models.