How language models extrapolate outside the training data: A case study in Textualized Gridworld
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866915049933635584 |
|---|---|
| author | Kim, Doyoung Lee, Jongwon Park, Jinho Seo, Minjoon |
| author_facet | Kim, Doyoung Lee, Jongwon Park, Jinho Seo, Minjoon |
| contents | Language models' ability to extrapolate learned behaviors to novel, more complex environments beyond their training scope is highly unknown. This study introduces a path planning task in a textualized Gridworld to probe language models' extrapolation capabilities. We show that conventional approaches, including next token prediction and Chain of Thought (CoT) finetuning, fail to extrapolate in larger, unseen environments. Inspired by human cognition and dual process theory, we propose cognitive maps for path planning, a novel CoT framework that simulates humanlike mental representations. Our experiments show that cognitive maps not only enhance extrapolation to unseen environments but also exhibit humanlike characteristics through structured mental simulation and rapid adaptation. Our finding that these cognitive maps require specialized training schemes and cannot be induced through simple prompting opens up important questions about developing general-purpose cognitive maps in language models. Our comparison with exploration-based methods further illuminates the complementary strengths of offline planning and online exploration. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2406_15275 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | How language models extrapolate outside the training data: A case study in Textualized Gridworld Kim, Doyoung Lee, Jongwon Park, Jinho Seo, Minjoon Computation and Language Language models' ability to extrapolate learned behaviors to novel, more complex environments beyond their training scope is highly unknown. This study introduces a path planning task in a textualized Gridworld to probe language models' extrapolation capabilities. We show that conventional approaches, including next token prediction and Chain of Thought (CoT) finetuning, fail to extrapolate in larger, unseen environments. Inspired by human cognition and dual process theory, we propose cognitive maps for path planning, a novel CoT framework that simulates humanlike mental representations. Our experiments show that cognitive maps not only enhance extrapolation to unseen environments but also exhibit humanlike characteristics through structured mental simulation and rapid adaptation. Our finding that these cognitive maps require specialized training schemes and cannot be induced through simple prompting opens up important questions about developing general-purpose cognitive maps in language models. Our comparison with exploration-based methods further illuminates the complementary strengths of offline planning and online exploration. |
| title | How language models extrapolate outside the training data: A case study in Textualized Gridworld |
| topic | Computation and Language |
| url | https://arxiv.org/abs/2406.15275 |