Position: Stop Anthropomorphizing Intermediate Tokens as Reasoning/Thinking Traces!

Subbarao Kambhampati, Karthik Valmeekam, Siddhant Bhambri, Vardhan Palod, Lucas Paul Saldyt, Kaya Stechly, Soumya Rani Samineni, Durgesh Kalwar, Upasana Biswas
Proceedings of the 43rd International Conference on Machine Learning, PMLR 306:170819-170835, 2026.

Abstract

Intermediate token generation (ITG), where a model produces output before the solution, has become a standard method to improve the performance of language models on reasoning tasks. These intermediate tokens have been called "reasoning traces" or even "thinking traces" – implicitly anthropomorphizing the traces, and implying that these traces resemble steps a human might take when solving a challenging problem, and as such can provide an interpretable window into the operation of the model’s thinking process to the end user. In this position paper, we present evidence that this anthropomorphization isn’t a harmless metaphor, and instead is quite dangerous – it confuses the nature of these models and how to use them effectively, and leads to questionable research. We call on the community to avoid such anthropomorphization of intermediate tokens.

Cite this Paper

BibTeX
@InProceedings{pmlr-v306-kambhampati26a, title = {Position: Stop Anthropomorphizing Intermediate Tokens as {R}easoning/{T}hinking Traces!}, author = {Kambhampati, Subbarao and Valmeekam, Karthik and Bhambri, Siddhant and Palod, Vardhan and Saldyt, Lucas Paul and Stechly, Kaya and Samineni, Soumya Rani and Kalwar, Durgesh and Biswas, Upasana}, booktitle = {Proceedings of the 43rd International Conference on Machine Learning}, pages = {170819--170835}, 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/kambhampati26a/kambhampati26a.pdf}, url = {https://proceedings.mlr.press/v306/kambhampati26a.html}, abstract = {Intermediate token generation (ITG), where a model produces output before the solution, has become a standard method to improve the performance of language models on reasoning tasks. These intermediate tokens have been called "reasoning traces" or even "thinking traces" – implicitly anthropomorphizing the traces, and implying that these traces resemble steps a human might take when solving a challenging problem, and as such can provide an interpretable window into the operation of the model’s thinking process to the end user. In this position paper, we present evidence that this anthropomorphization isn’t a harmless metaphor, and instead is quite dangerous – it confuses the nature of these models and how to use them effectively, and leads to questionable research. We call on the community to avoid such anthropomorphization of intermediate tokens.} }
Endnote
%0 Conference Paper %T Position: Stop Anthropomorphizing Intermediate Tokens as Reasoning/Thinking Traces! %A Subbarao Kambhampati %A Karthik Valmeekam %A Siddhant Bhambri %A Vardhan Palod %A Lucas Paul Saldyt %A Kaya Stechly %A Soumya Rani Samineni %A Durgesh Kalwar %A Upasana Biswas %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-kambhampati26a %I PMLR %P 170819--170835 %U https://proceedings.mlr.press/v306/kambhampati26a.html %V 306 %X Intermediate token generation (ITG), where a model produces output before the solution, has become a standard method to improve the performance of language models on reasoning tasks. These intermediate tokens have been called "reasoning traces" or even "thinking traces" – implicitly anthropomorphizing the traces, and implying that these traces resemble steps a human might take when solving a challenging problem, and as such can provide an interpretable window into the operation of the model’s thinking process to the end user. In this position paper, we present evidence that this anthropomorphization isn’t a harmless metaphor, and instead is quite dangerous – it confuses the nature of these models and how to use them effectively, and leads to questionable research. We call on the community to avoid such anthropomorphization of intermediate tokens.
APA
Kambhampati, S., Valmeekam, K., Bhambri, S., Palod, V., Saldyt, L.P., Stechly, K., Samineni, S.R., Kalwar, D. & Biswas, U.. (2026). Position: Stop Anthropomorphizing Intermediate Tokens as Reasoning/Thinking Traces!. Proceedings of the 43rd International Conference on Machine Learning, in Proceedings of Machine Learning Research 306:170819-170835 Available from https://proceedings.mlr.press/v306/kambhampati26a.html.

Related Material