LLM Users Observe Trend Toward Information-Dense Writing and Rich Vocabulary
AI users have highlighted a growing trend across recent open and closed language models toward highly compressed, information-dense prose. Models like GLM are increasingly utilizing sophisticated vocabulary and compact phrasing to convey complex ideas using fewer words. This shift toward hyper-concise writing can enhance efficiency by reducing token usage, but it may compromise immediate readability for average users. Understanding these stylistic shifts enables prompt engineers and developers to better tailor system instructions for their target audience. The observation highlighted outputs incorporating abstract vocabulary—such as using "Baudrillardian flourish" to explain search ranking failures—comparing the style to compressed science fiction prose like William Gibson's. The author speculates that this behavior could be an intentional byproduct of training models for higher token efficiency and semantic precision.
## BACKGROUND
Large language models (LLMs), such as OpenAI's GPT series or Z.ai's GLM models, are pre-trained on massive text datasets and further refined using instruction tuning and post-training alignment techniques. Model creators adjust output characteristics—such as verbosity, tone, and conciseness—to improve utility and operational efficiency. Because processing longer outputs increases computational latency and API costs, optimization often favors concise phrasing.