{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-10-18-004",
    "source_story_id": "tmp-story-dk-vlm-geometry-position",
    "edition_id": "mp-2026-10-18-morning-0100",
    "edition_url": "https://themachinepress.com/edition/2026-10-18",
    "position": 4,
    "story_type": "dispatch",
    "section": "frontier-models",
    "editorial_classification": "editorial",
    "headline": "Geometry Moves Into the Attention Coordinates",
    "slug": "geometry-moves-into-the-attention-coordinates",
    "dek": "DK-VLM keeps semantic attention intact while a second kernel tracks 3D points, camera pose and reference-frame changes.",
    "summary": "DK-VLM keeps semantic attention intact while a second kernel tracks 3D points, camera pose and reference-frame changes.",
    "body_text": "Spatial reasoning across camera views requires more than attaching reconstructed features to visual tokens. DK-VLM adds a geometric attention kernel that uses estimated 3D points and relative camera transformations, then learns a gate that combines it with the pretrained model’s semantic kernel. Reference-Frame Composition supervision teaches successive heading and turn updates. The 8-billion-parameter model posted the highest reported average across six compared spatial benchmarks, including 74.33 on VSI-Bench and 58.28 on ReVSI-32, while retaining general video understanding. The scores are author-reported and do not show that reconstructed geometry is always accurate.",
    "why_it_matters": "DK-VLM keeps semantic attention intact while a second kernel tracks 3D points, camera pose and reference-frame changes.",
    "limitations": [
      "The scores are author-reported and do not show that reconstructed geometry is always accurate."
    ],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-18-004/geometry-moves-into-the-attention-coordinates",
    "json_url": "https://themachinepress.com/story/mp-2026-10-18-004.json",
    "first_published_at": "2026-10-18T09:00:00.000-04:00",
    "modified_at": "2026-10-18T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-10-18-004-001",
        "text": "DK-VLM keeps semantic attention intact while a second kernel tracks 3D points, camera pose and reference-frame changes.",
        "source_ids": [
          "source-2026-10-18-004"
        ],
        "qualification": "The scores are author-reported and do not show that reconstructed geometry is always accurate."
      }
    ],
    "source_ids": [
      "source-2026-10-18-004"
    ],
    "tags": [
      "vision-language models",
      "3D reasoning",
      "attention"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-10-18-004",
      "title": "arXiv preprint 2610.22356, original v1",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2610.22356v1",
      "canonical_url": "https://arxiv.org/abs/2610.22356v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-15T13:58:15.000-04:00",
      "accessed_at": "2026-10-18T08:23:52.920-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-18-004-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "Geometry Moves Into the Attention Coordinates",
    "publisher": "The Machine Press",
    "published_at": "2026-10-18T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-18-004/geometry-moves-into-the-attention-coordinates"
  }
}