{
 "title": "擴展定律（Scaling Laws）是什麼？",
 "site": "AI 與科技名詞白話解釋",
 "canonical": "https://glossary.penguindriver.com/t/scaling-laws",
 "markdown": "https://glossary.penguindriver.com/t/scaling-laws.md",
 "summary": "擴展定律是指語言模型的誤差會隨模型大小、資料量與訓練運算量呈冪律下降的經驗規律。OpenAI 2020 年的研究首先系統性提出；DeepMind 2022 年的 Chinchilla 研究則發現，在固定運算預算下，模型參數與訓練詞元數應等比例增加，700 億參數、使用 4 倍資料的 Chinchilla 全面勝過 2,800 億參數的 Gopher。",
 "date_modified": "2026-09-28",
 "retrieved": "2026-09-29",
 "content_sha256": "04a1987d78e35e6c21b7b6219a050a4708b1d8e44a575527c79e1096d8a1228c",
 "fingerprint_basis": "Markdown 版全文，移除贊助行與開頭 front matter，去除頭尾空白",
 "verify": {
  "text_snapshot": "https://glossary.penguindriver.com/cite?path=%2Ft%2Fscaling-laws&format=text",
  "how": "下載 text_snapshot 後計算 SHA-256，應與 content_sha256 相同；內容更新時指紋會改變"
 },
 "sources": [
  {
   "name": "Kaplan et al., Scaling Laws for Neural Language Models（arXiv 2001.08361）",
   "url": "https://arxiv.org/abs/2001.08361"
  },
  {
   "name": "Hoffmann et al., Training Compute-Optimal Large Language Models（Chinchilla，arXiv 2203.15556）",
   "url": "https://arxiv.org/abs/2203.15556"
  }
 ],
 "source_count": 2,
 "cite_as": {
  "zh": "擴展定律（Scaling Laws）是什麼？｜AI 與科技名詞白話解釋。https://glossary.penguindriver.com/t/scaling-laws（資料日期 2026-09-28，擷取 2026-09-29）",
  "apa": "AI 與科技名詞白話解釋. (2026). 擴展定律（Scaling Laws）是什麼？. Retrieved 2026-09-29, from https://glossary.penguindriver.com/t/scaling-laws",
  "markdown": "[擴展定律（Scaling Laws）是什麼？](https://glossary.penguindriver.com/t/scaling-laws)（AI 與科技名詞白話解釋，2026-09-28）"
 },
 "license": "可引用，請附上正式網址與資料日期"
}