{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-09-06-009",
    "source_story_id": "tmp-story-pretraining-auxiliary-views",
    "edition_id": "mp-2026-09-06-morning-0059",
    "edition_url": "https://themachinepress.com/edition/2026-09-06",
    "position": 9,
    "story_type": "dispatch",
    "section": "research",
    "editorial_classification": "editorial",
    "headline": "The Model Learned a Fact Better When the Corpus Showed Another View",
    "slug": "the-model-learned-a-fact-better-when-the-corpus-showed-another-view",
    "dek": "With a fixed token budget, reformulations beat spending the same tokens on simple document repetition—even for factual recall.",
    "summary": "With a fixed token budget, reformulations beat spending the same tokens on simple document repetition—even for factual recall.",
    "body_text": "Controlled pretraining experiments found that repetition remained necessary, but reallocating some repeated tokens to auxiliary representations improved knowledge acquisition. Paraphrases helped under smaller batches, while contextual and foundational views aided learning when prior knowledge was missing. The effect did not depend on a stronger teacher generating the reformulation. The experiments isolate learning mechanisms under controlled conditions and do not prove that every synthetic rewrite improves a production corpus.",
    "why_it_matters": "With a fixed token budget, reformulations beat spending the same tokens on simple document repetition—even for factual recall.",
    "limitations": [
      "The effect did not depend on a stronger teacher generating the reformulation.",
      "The experiments isolate learning mechanisms under controlled conditions and do not prove that every synthetic rewrite improves a production corpus."
    ],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-06-009/the-model-learned-a-fact-better-when-the-corpus-showed-another-view",
    "json_url": "https://themachinepress.com/story/mp-2026-09-06-009.json",
    "first_published_at": "2026-09-06T09:00:00.000-04:00",
    "modified_at": "2026-09-06T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-09-06-009-001",
        "text": "With a fixed token budget, reformulations beat spending the same tokens on simple document repetition—even for factual recall.",
        "source_ids": [
          "source-2026-09-06-009"
        ],
        "qualification": "The effect did not depend on a stronger teacher generating the reformulation."
      }
    ],
    "source_ids": [
      "source-2026-09-06-009"
    ],
    "tags": [
      "pretraining",
      "data diversity",
      "knowledge acquisition"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-09-06-009",
      "title": "arXiv preprint 2609.04180",
      "publisher": "arXiv",
      "url": "https://arxiv.org/abs/2609.04180",
      "canonical_url": "https://arxiv.org/abs/2609.04180",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-09-03T13:57:02.000-04:00",
      "accessed_at": "2026-09-06T08:20:00.000-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-09-06-009-001"
      ]
    }
  ],
  "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 a Fact Better When the Corpus Showed Another View",
    "publisher": "The Machine Press",
    "published_at": "2026-09-06T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-09-06-009/the-model-learned-a-fact-better-when-the-corpus-showed-another-view"
  }
}