{
  "id": 3445,
  "url": "https://arxiv.org/abs/2606.07602v1",
  "title": "Sample-Efficient Post-Training for LEGO Spatial-Physics Reasoning",
  "summary": "LLM-based LEGO assembly generation requires both semantic grounding and physical feasibility. We identify a data-induced failure mode, PhysHack, in which the assemblies satisfy physical-validity constraints while producing structures that are geometrically misaligned, semantically inconsistent, or poorly calibrated. To address this challenge, we propose a model-based data selection approach that uses only a small fraction of the training data while improving physically grounded LEGO assembly gen",
  "authors": "Yuhuan Yuan, Zhouliang Yu, Minghao Liu, Weiyang Liu, Ge Lin Kan",
  "category": "research",
  "topics": "safety-alignment",
  "orgs": null,
  "regions": null,
  "published_at": "2026-05-29T09:31:25.000Z",
  "fetched_at": "2026-07-14T16:30:14.371Z",
  "source_slug": "arxiv-ethics",
  "source_name": "arXiv",
  "source_homepage": "https://arxiv.org",
  "ethics_ai_record_url": "https://ethics.ai/record/3445",
  "original_url": "https://arxiv.org/abs/2606.07602v1",
  "evidence_status": "source-only",
  "attribution": "via ethics.ai"
}