Temporal Blindness in Multi-Turn LLM Agents: Misaligned Tool Use vs. Human Time Perception
This addresses a critical liability for users of multi-turn LLM agents in dynamic environments, where misaligned tool use can reduce efficiency and reliability, though the findings are incremental as they highlight a known limitation with limited solutions.
The paper tackles the problem of temporal blindness in multi-turn LLM agents, where they fail to account for real-world time elapsed between messages, leading to misaligned tool use decisions; the result shows that without time information, models perform only slightly better than random (top alignment rate just over 60%), and adding timestamps provides modest improvement (peaking at around 65%).
Large language model agents are increasingly used in multi-turn conversational settings to interact with and execute tasks in dynamic environments. However, a key limitation is their temporal blindness: they, by default, operate with a stationary context, failing to account for the real-world time elapsed between messages. This becomes a critical liability when an agent must decide whether to invoke a tool based on how much time has passed since the last observation. Without temporal awareness, agents often either over-rely on previous context (skipping necessary tool calls), or under-rely on it (unnecessarily repeating tool calls). To study this challenge, we introduce TicToc-v1, a test set of multi-turn user-agent trajectories across 34 scenarios with varying time sensitivity. Each trajectory ends with a user question, where the need for a tool call depends on the amount of time elapsed since the last message. To give LLMs temporal context, we augment dialogue messages with explicit timestamps, bridging the gap between static dialogue and evolving environments. We then collected human preferences for these samples, creating two subsets: one where humans preferred relying on the previous observation (prefer-noTool), and another where they preferred a new tool call (prefer-Tool). We evaluated how well LLM tool-calling decisions align with human preferences under varying time intervals on TicToc-v1. Our analysis show that without time information, most models perform only slightly better than random, with the top alignment rate being just over 60%. While adding timestamps leads to a slight improvement, particularly for larger models, the improvement is modest, peaking at around 65%. We also show that naive, prompt-based alignment have limited effectiveness. Our findings highlight the need for specific post-training alignment to align multi-turn LLM tool use with human temporal perception.