Sparse Feature Coactivation Reveals Causal Semantic Modules in Large Language Models
Fuente:
arXiv
Salvato in:
| Autori principali: | , , , , , , , |
|---|---|
| Natura: | Preprint |
| Pubblicazione: |
2025
|
| Soggetti: | |
| Accesso online: | |
| Tags: |
Aggiungi Tag
Nessun Tag, puoi essere il primo ad aggiungerne!!
|
| _version_ | 1866915943925415936 |
|---|---|
| author | Deng, Ruixuan Hu, Xiaoyang Gilberti, Miles Storks, Shane Taxali, Aman Angstadt, Mike Sripada, Chandra Chai, Joyce |
| author_facet | Deng, Ruixuan Hu, Xiaoyang Gilberti, Miles Storks, Shane Taxali, Aman Angstadt, Mike Sripada, Chandra Chai, Joyce |
| contents | We identify semantically coherent, context-consistent network components in large language models (LLMs) using coactivation of sparse autoencoder (SAE) features collected from just a handful of prompts. Focusing on concept-relation prediction tasks, we show that ablating these components for concepts (e.g., countries and words) and relations (e.g., capital city and translation language) changes model outputs in predictable ways, while amplifying these components induces counterfactual responses. Notably, composing relation and concept components yields compound counterfactual outputs. Further analysis reveals that while most concept components emerge from the very first layer, more abstract relation components are concentrated in later layers. Lastly, we show that extracted components more comprehensively capture concepts and relations than individual features while maintaining specificity. Overall, our findings suggest a modular organization of knowledge and advance methods for efficient, targeted LLM manipulation. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2506_18141 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Sparse Feature Coactivation Reveals Causal Semantic Modules in Large Language Models Deng, Ruixuan Hu, Xiaoyang Gilberti, Miles Storks, Shane Taxali, Aman Angstadt, Mike Sripada, Chandra Chai, Joyce Computation and Language Artificial Intelligence We identify semantically coherent, context-consistent network components in large language models (LLMs) using coactivation of sparse autoencoder (SAE) features collected from just a handful of prompts. Focusing on concept-relation prediction tasks, we show that ablating these components for concepts (e.g., countries and words) and relations (e.g., capital city and translation language) changes model outputs in predictable ways, while amplifying these components induces counterfactual responses. Notably, composing relation and concept components yields compound counterfactual outputs. Further analysis reveals that while most concept components emerge from the very first layer, more abstract relation components are concentrated in later layers. Lastly, we show that extracted components more comprehensively capture concepts and relations than individual features while maintaining specificity. Overall, our findings suggest a modular organization of knowledge and advance methods for efficient, targeted LLM manipulation. |
| title | Sparse Feature Coactivation Reveals Causal Semantic Modules in Large Language Models |
| topic | Computation and Language Artificial Intelligence |
| url | https://arxiv.org/abs/2506.18141 |