AI・機械学習
上級

GPQA Diamond (2024年)(ジーピーキューエーダイヤモンド)

2024年成熟GPQA Diamond・Industry-leading 198 PhD-level questions + Industry-leading expert 65% + non-expert 34% + Industry-leading Google-proof benchmark + Industry-leading NYU GPQA Diamond PhD-level reasoning benchmark 2024。

0 回閲覧
0 いいね
2026/5/23 更新
関連タグ
gpqa-diamond-2024
gpqa-diamond-phd-level-reasoning-benchmark
198-phd-level-questions
expert-65-non-expert-34
google-proof-benchmark
nyu-gpqa-diamond-phd-level-reasoning-benchmark

概要

GPQA Diamond は2024年成熟・Industry-leading 198 PhD-level questions 2024年NYU GPQA Diamond PhD-level reasoning benchmark position確立。Industry-leading expert 65% + non-expert 34% + Google-proof + biology/chemistry/physics + o1 78% + Claude 3.5 Sonnet 65% + 2024 reasoning benchmark gold standard。

主な特徴・仕組み

  • Industry-leading NYU + Cohere + Anthropic (David Rein + Betty Li Hou et al.) 2023年11月 paper → 2024 standard
  • Industry-leading GPQA Diamond hardest subset signature (198 questions)
  • Industry-leading PhD-level questions signature (biology + chemistry + physics)
  • Industry-leading expert 65% accuracy (PhD domain holders)
  • Industry-leading non-expert 34% accuracy (PhD but different domain)
  • Industry-leading Google-proof signature (questions resist web search)
  • Industry-leading o1 78% (state-of-art 2024年9月)
  • Industry-leading Claude 3.5 Sonnet 65%
  • Industry-leading GPT-4o 53%
  • Industry-leading 2024 reasoning benchmark gold standard signature
  • Industry-leading 2023年11月 ArXiv paper "GPQA A Graduate-Level Google-Proof Q&A Benchmark"

まとめ

GPQA Diamond = 2024年成熟のpremium PhD-level reasoning benchmark。Industry-leading 198 PhD-level hardest subset questions biology + chemistry + physics + expert 65% non-expert 34% + Google-proof questions resist web search + o1 78% Claude 3.5 Sonnet 65% GPT-4o 53% + 2024 reasoning benchmark gold standard + NYU + Cohere + Anthropic David Rein + Betty Li Hou 2023年11月 paper position確立。

この記事について
カテゴリーAI・機械学習
難易度上級
作成日2026/5/23