Lost in Context: Addressing Context Anxiety in Large Language Models
For developers and users of large language models, this work reveals a previously uncharacterized failure mode that can be addressed through behavioral adjustments rather than model scaling.
The paper identifies and systematically studies 'context anxiety' in frontier reasoning models, where models fail due to premature self-doubt despite possessing necessary capabilities, leading to efficiency losses. It shows that models can learn alternative strategies to mitigate this issue, improving performance without scaling capabilities.
Conventional wisdom suggests that reasoning models fail when problems exceed their capabilities. However, we find that frontier reasoning models sometimes possess the necessary capabilities to solve problems but fail due to premature self-doubt -- a phenomenon informally known as context anxiety. We provide the first systematic study of context anxiety, demonstrating that it arises, in part, from a model's inability to accurately estimate the tokens required to complete a task. We also show that context anxiety leads to material efficiency losses when models operate under perceived constraints. Building on this analysis, we further show that models can learn alternative strategies for solving long-horizon problems without exhibiting context anxiety, suggesting that performance improvements may be achievable not through scaling model capabilities, but by improving models' ability to accurately assess and adapt to their own limitations.