Saved in:
| Main Author: | |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | https://arxiv.org/abs/2605.18537 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866913141454012416 |
|---|---|
| author | Modell, Alexander |
| author_facet | Modell, Alexander |
| contents | This paper introduces the Manifold Probe, a supervised method for discovering representation manifolds in superposition. The method generalizes linear regression probes by learning the space of features of a concept that can be linearly predicted from the representations, and then learning the directions used to encode them. We demonstrate the probe on representations of time and space in Llama 2-7b, finding manifolds which linearly represent an interpretable set of features in each case. In the case of time, we show that by steering along the manifold, we can influence the model's completions about the years in which famous songs, movies and books were released, providing evidence that the Manifold Probe can discover manifolds which are causally involved in model behaviour. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2605_18537 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Probing for Representation Manifolds in Superposition Modell, Alexander Machine Learning Artificial Intelligence 68T07 I.2.7 This paper introduces the Manifold Probe, a supervised method for discovering representation manifolds in superposition. The method generalizes linear regression probes by learning the space of features of a concept that can be linearly predicted from the representations, and then learning the directions used to encode them. We demonstrate the probe on representations of time and space in Llama 2-7b, finding manifolds which linearly represent an interpretable set of features in each case. In the case of time, we show that by steering along the manifold, we can influence the model's completions about the years in which famous songs, movies and books were released, providing evidence that the Manifold Probe can discover manifolds which are causally involved in model behaviour. |
| title | Probing for Representation Manifolds in Superposition |
| topic | Machine Learning Artificial Intelligence 68T07 I.2.7 |
| url | https://arxiv.org/abs/2605.18537 |