Salvato in:
| Autori principali: | , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2024
|
| Soggetti: | |
| Accesso online: | https://arxiv.org/abs/2412.11867 |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866915183887122432 |
|---|---|
| author | Spies, Alex F. Edwards, William Ivanitskiy, Michael I. Skapars, Adrians Räuker, Tilman Inoue, Katsumi Russo, Alessandra Shanahan, Murray |
| author_facet | Spies, Alex F. Edwards, William Ivanitskiy, Michael I. Skapars, Adrians Räuker, Tilman Inoue, Katsumi Russo, Alessandra Shanahan, Murray |
| contents | Recent studies in interpretability have explored the inner workings of transformer models trained on tasks across various domains, often discovering that these networks naturally develop highly structured representations. When such representations comprehensively reflect the task domain's structure, they are commonly referred to as "World Models" (WMs). In this work, we identify WMs in transformers trained on maze-solving tasks. By using Sparse Autoencoders (SAEs) and analyzing attention patterns, we examine the construction of WMs and demonstrate consistency between SAE feature-based and circuit-based analyses. By subsequently intervening on isolated features to confirm their causal role, we find that it is easier to activate features than to suppress them. Furthermore, we find that models can reason about mazes involving more simultaneously active features than they encountered during training; however, when these same mazes (with greater numbers of connections) are provided to models via input tokens instead, the models fail. Finally, we demonstrate that positional encoding schemes appear to influence how World Models are structured within the model's residual stream. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2412_11867 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Transformers Use Causal World Models in Maze-Solving Tasks Spies, Alex F. Edwards, William Ivanitskiy, Michael I. Skapars, Adrians Räuker, Tilman Inoue, Katsumi Russo, Alessandra Shanahan, Murray Machine Learning Artificial Intelligence I.2 Recent studies in interpretability have explored the inner workings of transformer models trained on tasks across various domains, often discovering that these networks naturally develop highly structured representations. When such representations comprehensively reflect the task domain's structure, they are commonly referred to as "World Models" (WMs). In this work, we identify WMs in transformers trained on maze-solving tasks. By using Sparse Autoencoders (SAEs) and analyzing attention patterns, we examine the construction of WMs and demonstrate consistency between SAE feature-based and circuit-based analyses. By subsequently intervening on isolated features to confirm their causal role, we find that it is easier to activate features than to suppress them. Furthermore, we find that models can reason about mazes involving more simultaneously active features than they encountered during training; however, when these same mazes (with greater numbers of connections) are provided to models via input tokens instead, the models fail. Finally, we demonstrate that positional encoding schemes appear to influence how World Models are structured within the model's residual stream. |
| title | Transformers Use Causal World Models in Maze-Solving Tasks |
| topic | Machine Learning Artificial Intelligence I.2 |
| url | https://arxiv.org/abs/2412.11867 |