{
  "id": 18688,
  "url": "https://arxiv.org/abs/2608.10537v1",
  "title": "Measuring Semantic Abstractness of SAE Features via Nonlocality",
  "summary": "Sparse autoencoders (SAEs) have helped uncover mechanistic explanations for LLM behaviours such as reasoning, jailbreaking etc., via understanding the corresponding task-relevant and causally effective features. To evaluate such mechanistic explanations, downstream studies must distinguish surface lexical features from genuinely high-level ones. However, neither an autointerp-based semantic description nor causal steering utility fully resolves the abstraction level of a feature. To this end, we",
  "authors": "Chuqiao Lin, Shivaji Sondhi, Xiao-Liang Qi",
  "category": "research",
  "topics": "safety-alignment",
  "orgs": null,
  "regions": null,
  "published_at": "2026-08-11T06:19:48.000Z",
  "fetched_at": "2026-08-12T05:10:43.828Z",
  "source_slug": "x-arxiv-red-teaming-query",
  "source_name": "arXiv red teaming query",
  "source_homepage": "https://arxiv.org/a/redteam",
  "ethics_ai_record_url": "https://ethics.ai/record/18688",
  "original_url": "https://arxiv.org/abs/2608.10537v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}