LLMs exhibit temporal flattening with substantially reduced semantic and cognitive-emotional drift compared to humans, allowing 94% accurate distinction from variability patterns alone.
Zefang Liu and Yinzhu Quan
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
fields
cs.CL 2representative citing papers
No single temporal tokenization strategy is best for all event data; performance depends on matching the tokenizer to the statistical shape of the data.
citing papers explorer
-
Temporal Flattening in LLM-Generated Text: Comparing Human and LLM Writing Trajectories
LLMs exhibit temporal flattening with substantially reduced semantic and cognitive-emotional drift compared to humans, allowing 94% accurate distinction from variability patterns alone.
-
Temporal Tokenization Strategies for Event Sequence Modeling with Large Language Models
No single temporal tokenization strategy is best for all event data; performance depends on matching the tokenizer to the statistical shape of the data.