LLMs mimic human temporal illusions, but may rely on retrieved knowledge

A new research paper explores whether large language models (LLMs) exhibit human-like temporal illusions, which are subjective perceptions of time influenced by context rather than direct duration tracking. The study, "The Machine's Internal Clock: Do LLMs Share Human Temporal Illusions?", used a benchmark of 6,684 narrative pairs to test 14 LLMs and 60 human participants. While humans only showed expected preferences in two out of five illusions, the LLMs succeeded in four, suggesting they may retrieve information from psychology research rather than exhibiting genuine temporal biases. AI

IMPACT Investigates whether LLMs exhibit human-like temporal perception biases, suggesting a divergence in how models process time-based information compared to humans.

RANK_REASON Research paper published on arXiv. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.CL →

• arXiv
• Catherine Bao Bao
• Hugging Face
• LLMs
• The Machine's Internal Clock: Do LLMs Share Human Temporal Illusions?

• paper
• safety

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →