Saved in:
| Main Authors: | , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2604.18158 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866910148579033088 |
|---|---|
| author | Lu, Yanzhen Qian, Zhicheng Jiang, Muchen Zhou, Xingyu |
| author_facet | Lu, Yanzhen Qian, Zhicheng Jiang, Muchen Zhou, Xingyu |
| contents | Prompt-based interventions can change model behavior, but trained success alone does not identify where the behaviorally relevant state is represented. We study this question in controlled routing tasks using interfaces chosen on support data, held-out query evaluation, and matched necessity, sufficiency, and wrong-interface controls. On GPT-2 triop, an early interface supports exact transfer under these tests. On GPT-2 add/sub, zero-retrain compiled transfer at the fixed interface recovers most of donor routing accuracy, while trainable prompt slots can relearn the same behavior at several other positions only after additional support examples and optimization. These results distinguish fixed-interface reuse from prompt relocation in a setting where the two can be tested directly. Qwen routing provides a cross-architecture consistency check for the same matched-interface pattern at the operator token, although donor-specific identity on the local V-path remains unresolved. Generation and reasoning branches are used to map scope: they show broader transport or weaker controller identifiability once control depends on longer trajectories or harder selection. In controlled routing, fixed-interface transfer is therefore stronger evidence of reuse than trained prompt success alone. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2604_18158 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | State Transfer Reveals Reuse in Controlled Routing Lu, Yanzhen Qian, Zhicheng Jiang, Muchen Zhou, Xingyu Artificial Intelligence Prompt-based interventions can change model behavior, but trained success alone does not identify where the behaviorally relevant state is represented. We study this question in controlled routing tasks using interfaces chosen on support data, held-out query evaluation, and matched necessity, sufficiency, and wrong-interface controls. On GPT-2 triop, an early interface supports exact transfer under these tests. On GPT-2 add/sub, zero-retrain compiled transfer at the fixed interface recovers most of donor routing accuracy, while trainable prompt slots can relearn the same behavior at several other positions only after additional support examples and optimization. These results distinguish fixed-interface reuse from prompt relocation in a setting where the two can be tested directly. Qwen routing provides a cross-architecture consistency check for the same matched-interface pattern at the operator token, although donor-specific identity on the local V-path remains unresolved. Generation and reasoning branches are used to map scope: they show broader transport or weaker controller identifiability once control depends on longer trajectories or harder selection. In controlled routing, fixed-interface transfer is therefore stronger evidence of reuse than trained prompt success alone. |
| title | State Transfer Reveals Reuse in Controlled Routing |
| topic | Artificial Intelligence |
| url | https://arxiv.org/abs/2604.18158 |