{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-11-08-018",
    "source_story_id": "tmp-brief-measured-coupling",
    "edition_id": "mp-2026-11-08-morning-0121",
    "edition_url": "https://themachinepress.com/edition/2026-11-08",
    "position": 20,
    "story_type": "ticker",
    "section": "frontier-models",
    "editorial_classification": "editorial",
    "headline": "Diffusion Decoding Measures Which Tokens Interfere",
    "slug": "diffusion-decoding-measures-which-tokens-interfere",
    "dek": "Cheap in-pass probes replace a similarity proxy with the actual confidence change between positions.",
    "summary": "Cheap in-pass probes replace a similarity proxy with the actual confidence change between positions.",
    "body_text": "Across three models and four benchmarks, measured coupling reduced decoding steps and was faster in most settings, including with a key-value cache, at an average 1.1-point accuracy cost. Shuffled or uniform couplings lost the gain.",
    "why_it_matters": "Cheap in-pass probes replace a similarity proxy with the actual confidence change between positions.",
    "limitations": [],
    "importance": 7,
    "canonical_url": "https://themachinepress.com/story/mp-2026-11-08-018/diffusion-decoding-measures-which-tokens-interfere",
    "json_url": "https://themachinepress.com/story/mp-2026-11-08-018.json",
    "first_published_at": "2026-11-08T09:00:00.000-05:00",
    "modified_at": "2026-11-08T09:00:00.000-05:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-11-08-018-001",
        "text": "Cheap in-pass probes replace a similarity proxy with the actual confidence change between positions.",
        "source_ids": [
          "source-2026-11-08-020"
        ],
        "qualification": null
      }
    ],
    "source_ids": [
      "source-2026-11-08-020"
    ],
    "tags": [
      "diffusion language models",
      "parallel decoding"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-11-08-020",
      "title": "arXiv preprint 2611.08307, original v1",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2611.08307v1",
      "canonical_url": "https://arxiv.org/abs/2611.08307v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-11-05T12:10:03.000-05:00",
      "accessed_at": "2026-11-08T08:27:17.257-05:00",
      "supports_claim_ids": [
        "claim-mp-2026-11-08-018-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "Diffusion Decoding Measures Which Tokens Interfere",
    "publisher": "The Machine Press",
    "published_at": "2026-11-08T09:00:00.000-05:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-11-08-018/diffusion-decoding-measures-which-tokens-interfere"
  }
}