Reproduce “The Impact of Response Latency and Task Type on Human-LLM Interaction and Perception” on OpenResearch

About this paper

Felicia Fang-Yi Tan, Moritz A. Messerschmidt, Wen Yin, Oded Nov

This study experimentally investigates how varying response latencies and task types influence user interaction and perception of large language models (LLMs). It demonstrates that moderate delays (9-20 seconds) in LLM responses can enhance perceptions of thoughtfulness and usefulness compared to very short delays (2 seconds), particularly for creative tasks, challenging the conventional HCI assumption that faster is always better.

Key insights

• LLM response latency serves as a tunable design variable, capable of influencing user perceptions of output quality and trust, rather than being solely a technical constraint to minimize.
• Moderate delays (9-20 seconds) can lead to higher perceptions of LLM 'thoughtfulness' and 'usefulness' compared to very short delays (2 seconds), challenging conventional HCI heuristics.
• User interaction behaviors, such as prompt frequency, are primarily shaped by the task type (e.g., Creation tasks elicit more prompts), while subjective quality perceptions are distinctly influenced by the response timing.

Summary by alphaXiv.