{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-08-02-013",
    "source_story_id": "tmp-story-kaist-modality-adaptive-decoding",
    "edition_id": "mp-2026-08-02-morning-0024",
    "edition_url": "https://themachinepress.com/edition/2026-08-02",
    "position": 13,
    "story_type": "dispatch",
    "section": "frontier-models",
    "editorial_classification": "editorial",
    "headline": "The Model Learned When a Sensor Was Lying",
    "slug": "the-model-learned-when-a-sensor-was-lying",
    "dek": "A modality-adaptive decoder reduced cross-sensor hallucinations when cameras, lidar and other inputs degraded at night or in smoke.",
    "summary": "A modality-adaptive decoder reduced cross-sensor hallucinations when cameras, lidar and other inputs degraded at night or in smoke.",
    "body_text": "KAIST researchers combined a sensor-understanding benchmark with modality-adaptive decoding that adjusts how much a multimodal model trusts each incoming channel. Tests included conditions such as darkness, fog, smoke and corrupted sensory inputs. The method reduced cross-modal hallucinations without retraining the full model and is intended for systems that fuse cameras with other sensors. Benchmark improvements do not establish safe autonomous operation in open-world conditions.",
    "why_it_matters": "A modality-adaptive decoder reduced cross-sensor hallucinations when cameras, lidar and other inputs degraded at night or in smoke.",
    "limitations": [
      "Benchmark improvements do not establish safe autonomous operation in open-world conditions."
    ],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-02-013/the-model-learned-when-a-sensor-was-lying",
    "json_url": "https://themachinepress.com/story/mp-2026-08-02-013.json",
    "first_published_at": "2026-08-02T09:00:00.000-04:00",
    "modified_at": "2026-08-02T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-08-02-013-001",
        "text": "A modality-adaptive decoder reduced cross-sensor hallucinations when cameras, lidar and other inputs degraded at night or in smoke.",
        "source_ids": [
          "source-2026-08-02-013"
        ],
        "qualification": "Benchmark improvements do not establish safe autonomous operation in open-world conditions."
      }
    ],
    "source_ids": [
      "source-2026-08-02-013"
    ],
    "tags": [
      "multimodal AI",
      "sensor fusion",
      "hallucinations",
      "autonomous systems"
    ],
    "image_url": "https://themachinepress.com/issues/2026-08-02/sensor-ai-file-image.webp",
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-08-02-013",
      "title": "IEEE Transactions on Image Processing: diverse sensor understanding",
      "publisher": "IEEE Transactions on Image Processing",
      "url": "https://doi.org/10.1109/TIP.2026.3707796",
      "canonical_url": "https://doi.org/10.1109/TIP.2026.3707796",
      "source_type": "secondary_reporting",
      "is_primary_source": false,
      "published_at": null,
      "accessed_at": "2026-08-02T08:28:19.398-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-08-02-013-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "The Model Learned When a Sensor Was Lying",
    "publisher": "The Machine Press",
    "published_at": "2026-08-02T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-08-02-013/the-model-learned-when-a-sensor-was-lying"
  }
}