Steering Along Manifolds to Control Neural Networks

{
  "doi": "10.48550/arxiv.2605.05115",
  "pdf": "https://arxiv.org/pdf/2605.05115",
  "url": "https://arxiv.org/abs/2605.05115",
  "tags": [
    "representation-steering",
    "manifold-geometry",
    "feature-steering",
    "Llama-3",
    "cyclic-concepts",
    "mechanistic-interpretability",
    "goodfire"
  ],
  "year": 2026,
  "saved": "2026-05-14",
  "title": "Steering Along Manifolds to Control Neural Networks",
  "venue": "arXiv preprint",
  "status": "full-text-saved",
  "landing": "https://www.goodfire.ai/research/manifold-steering",
  "arxiv_id": 2605.05115,
  "published": "2026-05-07",
  "enrichment": {
    "is_stale": true
  },
  "affiliation": "Goodfire (+ Stanford / collaborators)",
  "openalex_id": "W7160544910",
  "openalex_year": 2026,
  "openalex_enriched_at": 1778977722,
  "openalex_match_title": "Manifold Steering Reveals the Shared Geometry of Neural Network Representation and Behavior",
  "openalex_cited_by_count": 0
}