Temporal Blindness in Multi-Turn LLM Agents: Misaligned Tool Use vs. Human Time Perception

TL;DR

The paper identifies temporal blindness as a fundamental limitation in multi-turn LLM agents lacking real-world time awareness between turns. It introduces TicToc-v1, a diverse benchmark of 700+ trajectories across 34 time-sensitive scenarios, with explicit timestamps and human preferences to evaluate alignment of tool-use decisions with human time perception. Experiments across twelve models reveal that, absent time signals, alignment is only modestly above chance, and timestamp augmentation yields only small gains; prompting-based alignment strategies offer limited improvements. The findings motivate targeted post-training alignment approaches to better align LLM tool use with human temporal perception in dynamically evolving environments.

Abstract

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.

Figures (10)

• Figure 1: Illustrative examples showing the liability of temporal blindness in multi-turn LLM agents, in comparison to human. The first row shows the case of over-reliance, where the model displays over-confidence in outdated context, resulting in erroneous outputs. The second row shows the case of under-reliance, where the model displays excessive caution through repeated tool calls, resulting in unnecessary delays and latency.
• Figure 2: Normalized preference alignment rate of all models with and without timestamps. Without timestamps, models perform only slightly above random (max alignment marginally exceeding 60%). With timestamps, larger commercial models improve modestly, peaking just over 65%.
• Figure 3: Model attempt rates for both prefer-Tool and prefer-noTool cases. Without timestamps, models show varying tool-use biases; with timestamps, attempt rates rise for both classes, indicating limited alignment with human-like temporal awareness.
• Figure 4: Normalized preference alignment rate of Qwen-3-8B with and without long CoT reasoning, under settings of both with and without timestamp. Long CoT reasoning shows no improvement in tool-use alignment with human time perception.
• Figure 5: Prompt used to explicitly align models' tool-use decisions with human expectations. Few-shot example rules are used to illustrate when tool calls are appropriate or unnecessary depending on elapsed time, providing models with explicit guidance. Note that the scenarios mentioned in the prompt do not overlap with our coverage in TicToc-v1.