{
  "$schema": "https://themachinepress.com/schemas/story-v1.schema.json",
  "schema_version": "1.0.0",
  "document_type": "machine_press_story",
  "story": {
    "story_id": "mp-2026-10-16-006",
    "source_story_id": "tmp-story-mlexpert-autonomous-mle",
    "edition_id": "mp-2026-10-16-morning-0098",
    "edition_url": "https://themachinepress.com/edition/2026-10-16",
    "position": 6,
    "story_type": "dispatch",
    "section": "frontier-models",
    "editorial_classification": "editorial",
    "headline": "A Machine-Learning Agent Learns Between Experiments",
    "slug": "a-machine-learning-agent-learns-between-experiments",
    "dek": "MLExpert assigns shared rewards to the decisions between costly experiment results instead of waiting for an entire engineering run to end.",
    "summary": "MLExpert assigns shared rewards to the decisions between costly experiment results instead of waiting for an entire engineering run to end.",
    "body_text": "Long machine-learning engineering runs make supervision sparse: one result reflects several earlier decisions, and each experiment can be costly. MLExpert is a 35-billion-parameter mixture-of-experts model trained with hierarchical supervised tuning and Windowed Interaction RL, which treats the decisions between successive experiment results as one rewarded window. On MLE-Bench Lite, the authors report that the model's average medal rate rose from 37.9 to 72.7 percent, with further gains on scientific and algorithm-optimization tasks. The comparisons are benchmark-specific and do not demonstrate recursive self-improvement. They do show a practical way to assign credit before a long experimental trajectory is finished.",
    "why_it_matters": "MLExpert assigns shared rewards to the decisions between costly experiment results instead of waiting for an entire engineering run to end.",
    "limitations": [
      "The comparisons are benchmark-specific and do not demonstrate recursive self-improvement."
    ],
    "importance": 8,
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-16-006/a-machine-learning-agent-learns-between-experiments",
    "json_url": "https://themachinepress.com/story/mp-2026-10-16-006.json",
    "first_published_at": "2026-10-16T09:00:00.000-04:00",
    "modified_at": "2026-10-16T09:00:00.000-04:00",
    "content_status": "new",
    "is_carryover": false,
    "carryover_reason": null,
    "key_claims": [
      {
        "claim_id": "claim-mp-2026-10-16-006-001",
        "text": "MLExpert assigns shared rewards to the decisions between costly experiment results instead of waiting for an entire engineering run to end.",
        "source_ids": [
          "source-2026-10-16-006"
        ],
        "qualification": "The comparisons are benchmark-specific and do not demonstrate recursive self-improvement."
      }
    ],
    "source_ids": [
      "source-2026-10-16-006"
    ],
    "tags": [
      "machine learning engineering",
      "agents",
      "reinforcement learning"
    ],
    "image_url": null,
    "corrections": []
  },
  "sources": [
    {
      "source_id": "source-2026-10-16-006",
      "title": "arXiv preprint 2610.22344, original v1",
      "publisher": "arXiv / submitting research authors",
      "url": "https://arxiv.org/abs/2610.22344v1",
      "canonical_url": "https://arxiv.org/abs/2610.22344v1",
      "source_type": "primary_research",
      "is_primary_source": true,
      "published_at": "2026-10-15T13:56:32.000-04:00",
      "accessed_at": "2026-10-16T08:26:36.400-04:00",
      "supports_claim_ids": [
        "claim-mp-2026-10-16-006-001"
      ]
    }
  ],
  "corrections": [],
  "publisher": {
    "name": "The Machine Press",
    "url": "https://themachinepress.com",
    "description": "A daily newspaper for the age of artificial intelligence."
  },
  "cite_this_report": {
    "title": "A Machine-Learning Agent Learns Between Experiments",
    "publisher": "The Machine Press",
    "published_at": "2026-10-16T09:00:00.000-04:00",
    "canonical_url": "https://themachinepress.com/story/mp-2026-10-16-006/a-machine-learning-agent-learns-between-experiments"
  }
}