Lost in Context: Addressing Context Anxiety in Large Language Models

Ifueko Igbinedion, Jillian Ross, Etienne Ricardez, Sertac Karaman, Eric So
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:49485-49501, 2026.

Abstract

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.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-igbinedion26a, title = {Lost in Context: Addressing Context Anxiety in Large Language Models}, author = {Igbinedion, Ifueko and Ross, Jillian and Ricardez, Etienne and Karaman, Sertac and So, Eric}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {49485--49501}, year = {2026}, editor = {Zhang, Tong and Dudik, Miroslav and Jaggi, Martin and Agarwal, Alekh and Li, Sharon and Schuurmans, Dale and Zhu, Jerry and Berkenkamp, Felix and Dong, Hanze and Bietti, Alberto}, volume = {306}, series = {Proceedings of Machine Learning Research}, month = {06--11 Jul}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v306/main/assets/igbinedion26a/igbinedion26a.pdf}, url = {https://proceedings.mlr.press/v306/igbinedion26a.html}, abstract = {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.} }
Endnote
%0 Conference Paper %T Lost in Context: Addressing Context Anxiety in Large Language Models %A Ifueko Igbinedion %A Jillian Ross %A Etienne Ricardez %A Sertac Karaman %A Eric So %B Proceedings of the 43rd International Conference on Machine Learning %C Proceedings of Machine Learning Research %D 2026 %E Tong Zhang %E Miroslav Dudik %E Martin Jaggi %E Alekh Agarwal %E Sharon Li %E Dale Schuurmans %E Jerry Zhu %E Felix Berkenkamp %E Hanze Dong %E Alberto Bietti %F pmlr-v306-igbinedion26a %I PMLR %P 49485--49501 %U https://proceedings.mlr.press/v306/igbinedion26a.html %V 306 %X 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.
APA
Igbinedion, I., Ross, J., Ricardez, E., Karaman, S. & So, E.. (2026). Lost in Context: Addressing Context Anxiety in Large Language Models. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:49485-49501 Available from https://proceedings.mlr.press/v306/igbinedion26a.html.

Related Material