Position: Stop Anthropomorphizing Intermediate Tokens as Reasoning/Thinking Traces!
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , , , , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2025
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866915838727028736 |
|---|---|
| author | Kambhampati, Subbarao Valmeekam, Karthik Bhambri, Siddhant Palod, Vardhan Saldyt, Lucas Stechly, Kaya Samineni, Soumya Rani Kalwar, Durgesh Biswas, Upasana |
| author_facet | Kambhampati, Subbarao Valmeekam, Karthik Bhambri, Siddhant Palod, Vardhan Saldyt, Lucas Stechly, Kaya Samineni, Soumya Rani Kalwar, Durgesh Biswas, Upasana |
| contents | 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 \say{reasoning traces} or even \say{thoughts} -- 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. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2504_09762 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Position: Stop Anthropomorphizing Intermediate Tokens as Reasoning/Thinking Traces! Kambhampati, Subbarao Valmeekam, Karthik Bhambri, Siddhant Palod, Vardhan Saldyt, Lucas Stechly, Kaya Samineni, Soumya Rani Kalwar, Durgesh Biswas, Upasana Artificial Intelligence 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 \say{reasoning traces} or even \say{thoughts} -- 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. |
| title | Position: Stop Anthropomorphizing Intermediate Tokens as Reasoning/Thinking Traces! |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2504.09762 |