{
  "id": 12354,
  "url": "https://arxiv.org/abs/2607.18570v1",
  "title": "For What Reason? Interpreting Models' Encoding of Causation and Antithesis",
  "summary": "Discourse relations provide document structure, critical to language understanding and enabling language model performance and ethicality. In this work, we investigate how instruction-tuned Transformer models (LLaMA and Mistral) encode discourse relations in English, with a particular focus on the contrasting relations of causation and antithesis. Framing the task as a next-token prediction task and applying a suite of interpretability techniques to test model internals, our findings show that c",
  "authors": "Abhidip Bhattacharyya, Shira Wein",
  "category": "research",
  "topics": "safety-alignment,finance-investment",
  "orgs": "mistral",
  "regions": null,
  "published_at": "2026-07-20T23:05:04.000Z",
  "fetched_at": "2026-07-22T05:10:49.469Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/12354",
  "original_url": "https://arxiv.org/abs/2607.18570v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}