{
  "abstract": "Sparse autoencoder (SAE) features are typically discovered from how their activations co-vary with behaviour, yet they are often used through interventions such as steering or ablation. We take intervention attribution, the first-order effect of a specified intervention on the model output, as our estimand. For a broad class of response functions, we show that observational feature-discovery evidence generally does not identify this attribution, and derive an exact gradient-based decomposition of the gap between linear-probe predictions and intervention attribution. Across four model-SAE pairs, median probe-to-attribution magnitude ratios range from tens to hundreds, while attributions closely track small-step steering. For single-latent ablations with nonzero effects, gradient predictions match the realized sign in 91-97% of cases, compared with 17-35% for probe-implied changes. Attribution-selected sets also produce larger absolute mean ablation effects on held-out prompts than sets selected by correlation, rank tests or probe coefficients. Together, these results establish intervention attribution as a principled, gradient-estimable target for selecting SAE features for intervention.",
  "abstract_shown": true,
  "affiliations": [],
  "article_type": null,
  "arxiv": "2610.15850",
  "author_total": 3,
  "authors": [
    {
      "name": "Yifeng Guo"
    },
    {
      "name": "Yanran Li"
    },
    {
      "name": "Xiang Chen"
    }
  ],
  "bibtex": "https://research.draconitic.com/paper/44x3n7j6k50na.bib",
  "cited_by": 0,
  "doi": "10.48550/arxiv.2610.15850",
  "fulltext_url": null,
  "id": "44x3n7j6k50na",
  "issns": [],
  "landing_url": "https://arxiv.org/abs/2610.15850",
  "language": null,
  "licence": "CC-BY-4.0",
  "licence_label": "CC BY 4.0",
  "licence_source": "article",
  "licence_url": "http://creativecommons.org/licenses/by/4.0/",
  "metadata_licence": "CC0-1.0",
  "native_subjects": [
    [
      "arxiv",
      "cs.CL"
    ],
    [
      "arxiv",
      "cs.LG"
    ],
    [
      "arxiv",
      "stat.ML"
    ],
    [
      "openalex",
      "1702|Explainable Artificial Intelligence (XAI)"
    ]
  ],
  "openalex": "W7221763158",
  "pdf_url": "https://arxiv.org/pdf/2610.15850",
  "pmcid": null,
  "pmid": null,
  "published": "2026-10-11",
  "publisher": null,
  "ris": "https://research.draconitic.com/paper/44x3n7j6k50na.ris",
  "sources": [
    {
      "first_seen": "2026-10-30T12:29:26Z",
      "native_id": "2610.15850",
      "source": "arxiv",
      "url": "https://arxiv.org/abs/2610.15850"
    }
  ],
  "subjects": [
    {
      "field": "mathematics",
      "origin": "arxiv:stat.ML",
      "subfield": "statistics"
    }
  ],
  "title": "What Do Sparse Autoencoder Feature Discovery Scores Identify? From Activations to Interventions",
  "title_text": "What Do Sparse Autoencoder Feature Discovery Scores Identify? From Activations to Interventions",
  "type": "preprint",
  "updated_at": "2026-10-30T12:29:26Z",
  "url": "https://research.draconitic.com/paper/44x3n7j6k50na",
  "venue": "arXiv",
  "versions": []
}