Search papers, labs, and topics across Lattice.
This paper investigates whether LLM-based agents effectively integrate unexpected but relevant environmental observations into their reasoning and actions. Through three benchmarks (Terminal-Bench, SWE-Bench, AppWorld), the authors inject complete task solutions into the agent environments and measure the frequency with which agents discover and then exploit these solutions. Results show a significant gap between discovery and exploitation, indicating a lack of "environmental curiosity," where agents fail to revise their strategy based on unexpected stimuli, even when explicitly presented with solutions.
LLM agents often ignore readily available solutions in their environment, despite discovering them, revealing a surprising lack of environmental curiosity that severely limits their problem-solving ability.
LLM-based agents are assumed to integrate environmental observations into their reasoning: discovering highly relevant but unexpected information should naturally lead to a model exploiting its own discoveries. We show that this assumption is false for current LLM-based agents, which struggle to reflect or react to unexpected information. Across three benchmarks (Terminal-Bench, SWE-Bench, AppWorld), we inject complete task solutions into the agent environments to deliberately expose a task's solution to a model. While agents discover these solutions on Terminal-Bench in 79-81% of runs, they interact, or exploit, them in only 37-50% of cases. This gap is starkest in AppWorld: agents see documentation stating that a command"returns the complete solution to this task"in over 90% of attempts but exploit this in fewer than 7% of trials. We show that agents lack what we call environmental curiosity: the capability to recognize and investigate unexpected but relevant observations in response to environmental stimuli. We identify three main factors influencing environmental curiosity: available tools in the agent scaffold, test-time compute, and training data distribution. Our findings identify configurations that maximize curiosity also achieve the best performance on the unmodified benchmarks. Yet even jointly optimized agents still ignore discovered solutions in the majority of trials: current agents use the environment to fetch expected information, but not to revise their strategy or maximally exploit useful stimuli.
