Temporal Flattening in LLM-Generated Text: Comparing Human and LLM Writing Trajectories

The project addresses a specific gap in LLM evaluation: the absence of temporal evolution in AI-generated personas. While most AI detection focuses on snapshot differences, this project introduces a longitudinal lens. Its defensibility is low (3) because it is primarily a research artifact (paper + dataset) with minimal current traction (0 stars), making it easily reproducible by other academic groups. The frontier risk is low because labs like OpenAI are focused on optimizing immediate output quality rather than mimicking human stylistic drift over years. However, as long-context and persistent-memory architectures (like 'Memory' in ChatGPT) become standard, the 'temporal flattening' observed here may naturally diminish, potentially rendering the specific findings of this dataset obsolete within 1-2 years. The primary value lies in the dataset for training better stylistic detectors or persona-consistent agents.