{
  "id": 19478,
  "url": "https://arxiv.org/abs/2608.12447v1",
  "title": "Geometric and Behavioral Stratification in Transformer Residual Streams",
  "summary": "Trained transformer models develop privileged bases: coordinate axes whose statistics differ from the rest of the residual stream. But what kind of direction does such a basis select? We investigate the prediction direction, the unembedding direction of the token a model currently predicts, and find that it functions as a content-defined privileged anchor. Measured with respect to this anchor, residual-stream variation is geometrically and behaviorally stratified by proximity to the prediction.",
  "authors": "Nelson Guda",
  "category": "research",
  "topics": "finance-investment",
  "orgs": null,
  "regions": null,
  "published_at": "2026-08-12T17:42:20.000Z",
  "fetched_at": "2026-08-14T05:10:49.168Z",
  "source_slug": "arxiv-cslg",
  "source_name": "arXiv cs.LG",
  "source_homepage": "https://arxiv.org/list/cs.LG/recent",
  "ethics_ai_record_url": "https://ethics.ai/record/19478",
  "original_url": "https://arxiv.org/abs/2608.12447v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}