{"id":53225,"date":"2026-08-31T06:21:18","date_gmt":"2026-08-30T21:21:18","guid":{"rendered":"https:\/\/overcentral.com\/ja\/?p=53225"},"modified":"2026-08-31T06:21:18","modified_gmt":"2026-08-30T21:21:18","slug":"evo-harness-rl-meta-ai-53225","status":"publish","type":"post","link":"https:\/\/overcentral.com\/ja\/evo-harness-rl-meta-ai-53225\/","title":{"rendered":"\u6210\u529f\u738796.9%\u3001Meta\u306e8B\u30e2\u30c7\u30eb\u304cClaude Opus 4.5\u306b\u5339\u6575"},"content":{"rendered":"

Meta AI<\/a>\uff08\u30e1\u30bfAI\uff09\u3068\u30a4\u30ea\u30ce\u30a4\u5927\u5b66\u30a2\u30fc\u30d0\u30ca\u30fb\u30b7\u30e3\u30f3\u30da\u30fc\u30f3\u6821\u306e\u7814\u7a76\u30c1\u30fc\u30e0\u306f\u3001AI\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u306e\u9577\u671f\u30bf\u30b9\u30af\u5b9f\u884c\u80fd\u529b\u3092\u98db\u8e8d\u7684\u306b\u5411\u4e0a\u3055\u305b\u308b\u65b0\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u300cEvoHarness-RL\u300d<\/strong>\u3092\u767a\u8868\u3057\u305f\u3002\u3053\u306e\u624b\u6cd5\u3092\u7528\u3044\u308b\u3053\u3068\u3067\u3001\u30d1\u30e9\u30e1\u30fc\u30bf\u657080\u5104\uff088B\uff09\u306e\u5c0f\u578b\u30e2\u30c7\u30eb\u300cQwen3-8B\u300d\u304c\u3001\u5e73\u5747\u6210\u529f\u738796.9%<\/strong>\u3092\u9054\u6210\u3002\u3053\u308c\u306f\u3001\u540c\u5206\u91ce\u3067\u6700\u9ad8\u5cf0\u3068\u3055\u308c\u308bClaude Opus<\/a> 4.5\uff08\u6210\u529f\u738796.4%\uff09<\/strong>\u306b\u5339\u6575\u3059\u308b\u6027\u80fd\u3067\u3042\u308a\u3001\u8a08\u7b97\u30b3\u30b9\u30c8\u306e\u9762\u3067\u5927\u304d\u306a\u512a\u4f4d\u6027\u3092\u793a\u3059\u7d50\u679c\u3068\u306a\u3063\u305f\u3002<\/p>\n

EvoHarness-RL\u306f\u3001AI\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u304c\u5916\u90e8\u74b0\u5883\u304b\u3089\u5f97\u305f\u60c5\u5831\u3092\u3069\u306e\u3088\u3046\u306b\u8aad\u307f\u53d6\u308a\u3001\u66f4\u65b0\u3057\u3001\u7d71\u5408\u3059\u308b\u304b\u3068\u3044\u3046\u6839\u672c\u7684\u306a\u554f\u984c\u306b\u53d6\u308a\u7d44\u3080\u3002\u5f93\u6765\u306e\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u306f\u3001\u4eba\u9593\u306e\u958b\u767a\u8005\u304c\u4e8b\u524d\u306b\u5b9a\u7fa9\u3057\u305f\u30eb\u30fc\u30eb\u3084\u624b\u9806\u306b\u5f93\u3063\u3066\u52d5\u4f5c\u3059\u308b\u305f\u3081\u3001\u771f\u306e\u81ea\u5f8b\u6027\u3092\u6301\u305f\u306a\u3044\u3002\u8907\u96d1\u306a\u4f01\u696d\u5411\u3051\u30ef\u30fc\u30af\u30d5\u30ed\u30fc\u3001\u4f8b\u3048\u3070\u9867\u5ba2\u30c7\u30fc\u30bf\u306e\u5927\u898f\u6a21\u79fb\u884c\u3068\u3044\u3063\u305f\u9577\u6642\u9593\u306b\u308f\u305f\u308b\u30bf\u30b9\u30af\u3067\u306f\u3001\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u306f\u5185\u90e8\u306e\u30b3\u30f3\u30c6\u30ad\u30b9\u30c8\u30a6\u30a3\u30f3\u30c9\u30a6\u3060\u3051\u306b\u4f9d\u5b58\u3067\u304d\u305a\u3001\u5b9f\u884c\u57fa\u76e4\uff08\u30cf\u30fc\u30cd\u30b9\uff09\u3078\u306e\u4f9d\u5b58\u5ea6\u304c\u9ad8\u307e\u308b\u3002<\/p>\n

\u3053\u306e\u30cf\u30fc\u30cd\u30b9\u306f\u3001\u30b5\u30fc\u30d0\u30fc\u30ed\u30b0\u306a\u3069\u306e\u5b9f\u884c\u30d5\u30a3\u30fc\u30c9\u30d0\u30c3\u30af\u3092\u63d0\u4f9b\u3057\u3001\u52d5\u7684\u306aAPI\u63a5\u7d9a\u306e\u72b6\u614b\u3092\u6b63\u78ba\u306b\u628a\u63e1\u3059\u308b\u52a9\u3051\u3068\u306a\u308b\u3002\u307e\u305f\u3001\u5b8c\u4e86\u6e08\u307f\u304a\u3088\u3073\u672a\u51e6\u7406\u306e\u30b5\u30d6\u30b4\u30fc\u30eb\u3092\u7ba1\u7406\u3059\u308b\u72b6\u614b\u30c8\u30e9\u30c3\u30ab\u30fc\u3084\u5236\u5fa1\u30d5\u30ed\u30fc\u6a5f\u69cb\u3082\u63d0\u4f9b\u3059\u308b\u3002\u3057\u304b\u3057\u3001\u5f93\u6765\u306e\u30a2\u30d7\u30ed\u30fc\u30c1\u3067\u306f\u3001\u3053\u308c\u3089\u306e\u4ed5\u7d44\u307f\u306f\u624b\u52d5\u3067\u30b3\u30fc\u30c9\u5316\u3055\u308c\u305f\u56fa\u5b9a\u30ed\u30b8\u30c3\u30af\u306b\u4f9d\u5b58\u3057\u3066\u304a\u308a\u3001\u30e2\u30c7\u30eb\u304c\u30a2\u30c3\u30d7\u30b0\u30ec\u30fc\u30c9\u3055\u308c\u308b\u305f\u3073\u306b\u518d\u8abf\u6574\u3068\u30c7\u30d0\u30c3\u30b0\u306e\u9577\u671f\u30b5\u30a4\u30af\u30eb\u304c\u767a\u751f\u3059\u308b\u3068\u3044\u3046\u8ab2\u984c\u304c\u3042\u3063\u305f\u3002<\/p>\n

EvoHarness-RL\u3068\u306f\u4f55\u304b\u2014\u2014BPE\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306e\u6838\u5fc3<\/h2>\n

EvoHarness-RL\u304c\u5f93\u6765\u624b\u6cd5\u3068\u6c7a\u5b9a\u7684\u306b\u7570\u306a\u308b\u70b9\u306f\u3001\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u306b\u300c\u30cf\u30fc\u30cd\u30b9\u306e\u6700\u9069\u306a\u4f7f\u3044\u65b9\u300d\u3092\u5f37\u5316\u5b66\u7fd2\u3067\u6559\u3048\u8fbc\u3080<\/strong>\u70b9\u306b\u3042\u308b\u3002\u7814\u7a76\u30c1\u30fc\u30e0\u306f\u3001\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u306e\u5916\u90e8\u30b5\u30dd\u30fc\u30c8\u30b7\u30b9\u30c6\u30e0\u3092\u300cBelief\uff08\u4fe1\u5ff5\uff09\u300d\u300cProgress\uff08\u9032\u6357\uff09\u300d\u300cExperience\uff08\u7d4c\u9a13\uff09\u300d\u3068\u3044\u30463\u3064\u306e\u6a5f\u80fd\u9818\u57df\u306b\u7d71\u5408\u3057\u305f\u7d71\u4e00\u30a4\u30f3\u30bf\u30fc\u30d5\u30a7\u30fc\u30b9\u300cBPE\u300d<\/strong>\u3092\u8a2d\u8a08\u3057\u305f\u3002<\/p>\n

\u3053\u306eBPE\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306b\u304a\u3044\u3066\u3001\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u306f4\u3064\u306e\u30e1\u30bf\u30a2\u30af\u30b7\u30e7\u30f3\uff08track\u3001commit\u3001recall\u3001note\uff09\u3092\u7528\u3044\u3066\u3001\u74b0\u5883\u306e\u8ffd\u8de1\u3001\u30ef\u30fc\u30af\u30d5\u30ed\u30fc\u306e\u66f4\u65b0\u3001\u904e\u53bb\u306e\u6226\u7565\u306e\u60f3\u8d77\u3001\u65b0\u305f\u306a\u77e5\u898b\u306e\u8a18\u9332\u3092\u5b9f\u884c\u3059\u308b\u3002\u3053\u308c\u306b\u3088\u308a\u3001\u8907\u96d1\u3067\u30c9\u30e1\u30a4\u30f3\u56fa\u6709\u306eAPI\u3092\u610f\u8b58\u3059\u308b\u3053\u3068\u306a\u304f\u3001\u5916\u90e8\u72b6\u614b\u3092\u7ba1\u7406\u3067\u304d\u308b\u3088\u3046\u306b\u306a\u308b\u3002<\/p>\n

\u8ad6\u6587\u306e\u5171\u540c\u57f7\u7b46\u8005\u3067\u3042\u308bXuying Ning\u6c0f\u306f\u3001\u3053\u306e\u30a2\u30d7\u30ed\u30fc\u30c1\u306e\u91cd\u8981\u6027\u3092\u6b21\u306e\u3088\u3046\u306b\u8aac\u660e\u3059\u308b\u3002\u300c\u5f93\u6765\u306e\u8ffd\u8a18\u578b\u30e1\u30e2\u30ea\u306f\u3001\u3088\u308a\u591a\u304f\u306e\u30b3\u30f3\u30c6\u30ad\u30b9\u30c8\u304c\u5e38\u306b\u6709\u76ca\u3067\u3042\u308b\u3068\u3044\u3046\u524d\u63d0\u306b\u7acb\u3063\u3066\u3044\u307e\u3059\u3002\u3057\u304b\u3057\u3001\u9577\u6642\u9593\u306e\u30bf\u30b9\u30af\u306b\u304a\u3044\u3066\u3001\u30e1\u30e2\u30ea\u306b\u306f\u53e4\u3044\u7d50\u8ad6\u3084\u5931\u6557\u3057\u305f\u8a66\u884c\u3001\u3042\u308b\u3044\u306f\u3082\u306f\u3084\u95a2\u9023\u6027\u306e\u306a\u3044\u60c5\u5831\u304c\u84c4\u7a4d\u3055\u308c\u3001\u3080\u3057\u308d\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u306e\u63a8\u8ad6\u3092\u52a3\u5316\u3055\u305b\u308b\u53ef\u80fd\u6027\u304c\u3042\u308a\u307e\u3059\u3002\u9577\u671f\u7684\u306a\u30bf\u30b9\u30af\u3092\u51e6\u7406\u3059\u308b\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u306b\u306f\u3001\u60c5\u5831\u3092\u66f4\u65b0\u3057\u3001\u5727\u7e2e\u3057\u3001\u7f6e\u304d\u63db\u3048\u308b\u3053\u3068\u304c\u3067\u304d\u308b\u52d5\u7684\u306a\u30e1\u30e2\u30ea\u304c\u5fc5\u8981\u306a\u306e\u3067\u3059\u300d<\/p>\n

2\u6bb5\u968e\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u3067\u5b9f\u73fe\u3057\u305f\u300c\u30b3\u30b9\u30c8\u3092\u610f\u8b58\u3057\u305f\u5b66\u7fd2\u300d<\/h2>\n

EvoHarness-RL\u306e\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u306f\u30012\u6bb5\u968e\u306e\u30d7\u30ed\u30bb\u30b9\u3067\u69cb\u6210\u3055\u308c\u308b\u3002\u7b2c\u4e00\u6bb5\u968e\u3067\u306f\u3001\u6559\u5e2b\u3042\u308a\u5b66\u7fd2\u306b\u3088\u308b\u30cf\u30fc\u30cd\u30b9\u306e\u5fae\u8abf\u6574\uff08Supervised Harness Fine-Tuning\uff09\u3092\u5b9f\u65bd\u3002\u30d9\u30fc\u30b9\u30e2\u30c7\u30eb\u306f\u3001\u8907\u96d1\u306a\u30a4\u30f3\u30bf\u30e9\u30af\u30b7\u30e7\u30f3\u30ed\u30b0\u304b\u3089\u6709\u7528\u306a\u60c5\u5831\u3092\u62bd\u51fa\u3057\u3001BPE\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306b\u69cb\u9020\u5316\u3059\u308b\u65b9\u6cd5\u3092\u5b66\u7fd2\u3059\u308b\u3002<\/p>\n

\u7b2c\u4e8c\u6bb5\u968e\u3067\u306f\u3001\u300c\u30b3\u30b9\u30c8\u3092\u8003\u616e\u3057\u305f\u5f37\u5316\u5b66\u7fd2\uff08Cost-Aware Reinforcement Learning\uff09\u300d<\/strong>\u3092\u9069\u7528\u3059\u308b\u3002\u30e1\u30e2\u30ea\u3078\u306e\u554f\u3044\u5408\u308f\u305b\u3084\u30c8\u30e9\u30c3\u30ab\u30fc\u306e\u66f4\u65b0\u306f\u3001\u6642\u9593\u3068\u8a08\u7b97\u30c8\u30fc\u30af\u30f3\u3092\u6d88\u8cbb\u3059\u308b\u3002\u305d\u306e\u305f\u3081\u3001\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u306f\u3059\u3079\u3066\u306e\u30b9\u30c6\u30c3\u30d7\u3067\u76f2\u76ee\u7684\u306b\u5916\u90e8\u30c4\u30fc\u30eb\u3092\u53c2\u7167\u3059\u308b\u308f\u3051\u306b\u306f\u3044\u304b\u306a\u3044\u3002\u3053\u306e\u30d5\u30a7\u30fc\u30ba\u3067\u306f\u3001\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u304c\u5916\u90e8\u72b6\u614b\u3078\u306e\u30a2\u30af\u30bb\u30b9\u304c\u4e88\u7b97\u30b3\u30b9\u30c8\u306b\u898b\u5408\u3046\u304b\u3069\u3046\u304b\u3092\u5224\u65ad\u3059\u308b\u80fd\u529b\u3092\u7372\u5f97\u3059\u308b\u3002\u3053\u308c\u306b\u3088\u308a\u3001\u30c4\u30fc\u30eb\u306e\u4f7f\u7528\u304c\u56fa\u5b9a\u5316\u3055\u308c\u305f\u30d7\u30ed\u30f3\u30d7\u30c8\u304b\u3089\u3001\u5b66\u7fd2\u3055\u308c\u305f\u5b9f\u884c\u6642\u884c\u52d5\u3078\u3068\u5909\u63db\u3055\u308c\u308b\u3002<\/p>\n

\u5b9f\u9a13\u7d50\u679c\u2014\u20148B\u30e2\u30c7\u30eb\u304cClaude Opus 4.5\u306b\u5339\u6575<\/h2>\n

\u7814\u7a76\u30c1\u30fc\u30e0\u306f\u3001\u30c6\u30ad\u30b9\u30c8\u30d9\u30fc\u30b9\u306e\u30de\u30eb\u30c1\u30b9\u30c6\u30c3\u30d7\u30bf\u30b9\u30af\u74b0\u5883\u300cALFWorld\u300d\u3092\u7528\u3044\u3066\u8a55\u4fa1\u3092\u5b9f\u65bd\u3002\u30d9\u30fc\u30b9\u30e2\u30c7\u30eb\u306b\u306fQwen3-8B\u3092\u63a1\u7528\u3057\u3001\u4ee5\u4e0b\u306e\u6bd4\u8f03\u5bfe\u8c61\u3068\u5bfe\u5cd9\u3055\u305b\u305f\u3002<\/p>\n

    \n
  • \u30d5\u30ed\u30f3\u30c6\u30a3\u30a2\u30e2\u30c7\u30eb\uff1a<\/strong>Claude Opus 4.5\u3001GPT-4.1\u3001GPT-5<\/li>\n
  • \u9759\u7684\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\uff1a<\/strong>ReAct\u3001ExpeL\u3001ReasoningBank<\/li>\n
  • \u5b66\u7fd2\u53ef\u80fd\u306a\u624b\u6cd5\uff1a<\/strong>\u6a19\u6e96GRPO\u3001SkillOS\u3001SkillRL<\/li>\n<\/ul>\n

    \u7d50\u679c\u306f\u9bae\u660e\u3060\u3002EvoHarness-RL\u3092\u9069\u7528\u3057\u305fQwen3-8B\u306f\u3001\u30d9\u30fc\u30b9\u30e9\u30a4\u30f3\u306eReAct\u3068\u6bd4\u8f03\u3057\u306649.0\u30dd\u30a4\u30f3\u30c8\u3082\u306e\u6539\u5584<\/strong>\u3092\u9054\u6210\u3002\u5b66\u7fd2\u53ef\u80fd\u306a\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3067\u3042\u308bSkillRL\uff0889.9%\uff09\u3084SkillOS\uff0880.2%\uff09\u3092\u5927\u304d\u304f\u4e0a\u56de\u3063\u305f\u3002\u305d\u3057\u3066\u6700\u3082\u6ce8\u76ee\u3059\u3079\u304d\u306f\u3001\u30d1\u30e9\u30e1\u30fc\u30bf\u657080\u5104\u306e\u5c0f\u578b\u30e2\u30c7\u30eb\u304c\u3001Claude Opus 4.5\u306e96.4%\u3068\u3044\u3046\u30b9\u30b3\u30a2\u306b\u5339\u6575\u3059\u308b96.9%\u3092\u8a18\u9332<\/strong>\u3057\u305f\u3053\u3068\u3060\u3002\u3053\u308c\u306f\u3001\u30a8\u30f3\u30bf\u30fc\u30d7\u30e9\u30a4\u30ba\u958b\u767a\u8005\u304c\u8a08\u7b97\u30b3\u30b9\u30c8\u3092\u6700\u9069\u5316\u3059\u308b\u4e0a\u3067\u6975\u3081\u3066\u91cd\u8981\u306a\u610f\u5473\u3092\u6301\u3064\u3002<\/p>\n

    \u3055\u3089\u306b\u3001BPE\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306f\u5f37\u5316\u5b66\u7fd2\u30d5\u30a7\u30fc\u30ba\u3092\u7d4c\u305a\u3068\u3082\u3001\u3059\u3079\u3066\u306e\u30e2\u30c7\u30eb\u898f\u6a21\u3067\u6c4e\u7528\u7684\u306a\u5229\u76ca\u3092\u3082\u305f\u3089\u3059\u3053\u3068\u304c\u78ba\u8a8d\u3055\u308c\u305f\u3002\u7814\u7a76\u8005\u304c\u51cd\u7d50\u72b6\u614b\u306e\u30d5\u30ed\u30f3\u30c6\u30a3\u30a2\u30e2\u30c7\u30eb\u306bBPE\u306e\u30d7\u30ed\u30f3\u30d7\u30c8\u30bf\u30a4\u30e0\u30cf\u30fc\u30cd\u30b9\u3092\u88c5\u5099\u3057\u305f\u3068\u3053\u308d\u3001GPT-4.1\u306e\u6210\u529f\u7387\u306f22.1\u30dd\u30a4\u30f3\u30c8\u3001GPT-5\u306f25.7\u30dd\u30a4\u30f3\u30c8\u305d\u308c\u305e\u308c\u5411\u4e0a\u3057\u305f\u3002<\/p>\n

    \u300c\u30cf\u30fc\u30cd\u30b9\u30a2\u30cb\u30fc\u30ea\u30f3\u30b0\u300d\u2014\u2014\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u304c\u81ea\u5f8b\u7684\u306b\u7372\u5f97\u3059\u308b\u52b9\u7387\u6027<\/h2>\n

    \u5b9f\u9a13\u4e2d\u306b\u89b3\u5bdf\u3055\u308c\u305f\u8208\u5473\u6df1\u3044\u73fe\u8c61\u304c\u3001\u300c\u30cf\u30fc\u30cd\u30b9\u30a2\u30cb\u30fc\u30ea\u30f3\u30b0\uff08Harness Annealing\uff09\u300d\u3060\u3002\u30c8\u30ec\u30fc\u30cb\u30f3\u30b0\u521d\u671f\u306e\u6bb5\u968e\u3067\u306f\u3001AI\u306f\u307b\u307c\u3059\u3079\u3066\u306e\u30b9\u30c6\u30c3\u30d7\u3067Experience\u3068Progress\u30c8\u30e9\u30c3\u30ab\u30fc\u306b\u554f\u3044\u5408\u308f\u305b\u3092\u884c\u3063\u3066\u3044\u305f\u3002\u3057\u304b\u3057\u3001\u30eb\u30fc\u30c6\u30a3\u30f3\u30bf\u30b9\u30af\u3092\u7fd2\u5f97\u3059\u308b\u306b\u3064\u308c\u3066\u3001\u5916\u90e8\u30c4\u30fc\u30eb\u3078\u306e\u4f9d\u5b58\u5ea6\u304c\u80fd\u52d5\u7684\u306b\u4f4e\u4e0b\u3002\u6210\u529f\u30d1\u30bf\u30fc\u30f3\u3092\u76f4\u63a5\u30d1\u30e9\u30e1\u30fc\u30bf\u306b\u57cb\u3081\u8fbc\u3080\u3088\u3046\u306b\u306a\u308b\u3002<\/p>\n

    \u3053\u308c\u3068\u540c\u6642\u306b\u3001\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u306f\u300c\u30cf\u30fc\u30cd\u30b9\u9032\u5316\uff08Harness Evolution\uff09\u300d\u3082\u793a\u3057\u305f\u3002\u3059\u306a\u308f\u3061\u3001\u5358\u7d14\u306a\u30bf\u30b9\u30af\u3067\u306f\u30c4\u30fc\u30eb\u3092\u30d0\u30a4\u30d1\u30b9\u3059\u308b\u4e00\u65b9\u3001\u672a\u77e5\u306e\u74b0\u5883\u3084\u4e88\u671f\u305b\u306c\u969c\u5bb3\u306b\u76f4\u9762\u3057\u305f\u77ac\u9593\u306b\u306f\u3001Belief\u30e2\u30b8\u30e5\u30fc\u30eb\u3068Experience\u30e2\u30b8\u30e5\u30fc\u30eb\u306e\u4f7f\u7528\u3092\u7a4d\u6975\u7684\u306b\u62e1\u5927\u3059\u308b\u3002\u4f8b\u3048\u3070\u3001\u6a19\u6e96\u7684\u306a\u30c7\u30fc\u30bf\u30d9\u30fc\u30b9\u30ec\u30b3\u30fc\u30c9\u306e\u79fb\u884c\u4e2d\u306f\u9ad8\u901f\u306b\u52d5\u4f5c\u3059\u308b\u304c\u3001\u5947\u5999\u306a\u30ec\u30ac\u30b7\u30fcAPI\u30a8\u30f3\u30c9\u30dd\u30a4\u30f3\u30c8\u3084\u8907\u96d1\u306a\u30d0\u30ea\u30c7\u30fc\u30b7\u30e7\u30f3\u30a8\u30e9\u30fc\u306b\u906d\u9047\u3057\u305f\u5834\u5408\u306b\u306f\u3001\u30e9\u30a4\u30d6\u30b5\u30fc\u30d0\u30fc\u30ed\u30b0\u3092\u53d6\u5f97\u3057\u3001\u904e\u53bb\u306e\u30c1\u30b1\u30c3\u30c8\u3092\u691c\u7d22\u3057\u3066\u5b89\u5168\u306b\u89e3\u6c7a\u3059\u308b\u2014\u2014\u3053\u306e\u3088\u3046\u306a\u52d5\u7684\u306a\u884c\u52d5\u6226\u7565\u306e\u9069\u5fdc\u304c\u78ba\u8a8d\u3055\u308c\u305f\u3002<\/p>\n

    \u65e2\u5b58\u30b7\u30b9\u30c6\u30e0\u3078\u306e\u7d71\u5408\u3068\u5b9f\u52d9\u7684\u306a\u5c0e\u5165\u6307\u91dd<\/h2>\n

    \u65b0\u3057\u3044\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u306e\u5c0e\u5165\u306f\u3001\u30a8\u30f3\u30bf\u30fc\u30d7\u30e9\u30a4\u30ba\u306e\u30a8\u30f3\u30b8\u30cb\u30a2\u30ea\u30f3\u30b0\u30c1\u30fc\u30e0\u306b\u3068\u3063\u3066\u6469\u64e6\u3092\u751f\u3080\u53ef\u80fd\u6027\u304c\u3042\u308b\u3002\u3057\u304b\u3057\u3001EvoHarness-RL\u306f\u300c\u74b0\u5883\u30a2\u30c0\u30d7\u30bf\u300d\u3092\u5099\u3048\u3066\u304a\u308a\u3001\u7d44\u7e54\u306e\u65e2\u5b58\u30c4\u30fc\u30eb\u306b\u7279\u5316\u3057\u305f\u5185\u90e8\u5b9f\u88c5\u3092\u7dad\u6301\u3057\u306a\u304c\u3089\u3001\u5b66\u7fd2\u53ef\u80fd\u306a\u30ec\u30a4\u30e4\u30fc\u3092\u5171\u6709\u3059\u308b\u3053\u3068\u3092\u53ef\u80fd\u306b\u3057\u3066\u3044\u308b\u3002<\/p>\n

    Ning\u6c0f\u306f\u6b21\u306e\u3088\u3046\u306b\u8ff0\u3079\u308b\u3002\u300cBPE\u306f\u3001\u65e2\u5b58\u306e\u30aa\u30fc\u30b1\u30b9\u30c8\u30ec\u30fc\u30b7\u30e7\u30f3\u30b7\u30b9\u30c6\u30e0\u306b\u7d71\u5408\u3059\u308b\u5927\u304d\u306a\u53ef\u80fd\u6027\u3092\u79d8\u3081\u3066\u3044\u307e\u3059\u3002\u30c1\u30fc\u30e0\u304c\u73fe\u5728\u306e\u30c4\u30fc\u30eb\u3084\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u3092\u7f6e\u304d\u63db\u3048\u308b\u5fc5\u8981\u306f\u5fc5\u305a\u3057\u3082\u3042\u308a\u307e\u305b\u3093\u3002BPE\u306f\u3001\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u304c\u73fe\u5728\u4f55\u3092\u4fe1\u3058\u3066\u3044\u308b\u304b\u3001\u3069\u3053\u307e\u3067\u9032\u6357\u3057\u305f\u304b\u3001\u4f55\u3092\u5b66\u3093\u3060\u304b\u3092\u7d99\u7d9a\u7684\u306b\u6574\u7406\u3059\u308b\u3001\u8ffd\u52a0\u306e\u72b6\u614b\u7ba1\u7406\u30ec\u30a4\u30e4\u30fc\u3068\u3057\u3066\u6a5f\u80fd\u3057\u307e\u3059\u300d<\/p>\n

    \u307e\u305f\u3001\u5c0e\u5165\u306b\u3042\u305f\u3063\u3066\u306f\u3001\u30bf\u30b9\u30af\u306e\u6027\u8cea\u306b\u5fdc\u3058\u305f\u9069\u5207\u306a\u5224\u65ad\u304c\u91cd\u8981\u3060\u3002\u77ed\u671f\u9593\u3067\u5b89\u5b9a\u3057\u305f\u30bf\u30b9\u30af\u306b\u306fReAct\u3084\u6a19\u6e96RAG\u3067\u5341\u5206\u3060\u304c\u3001\u6570\u6642\u9593\u304b\u3089\u6570\u65e5\u3001\u3042\u308b\u3044\u306f\u6570\u9031\u9593\u306b\u308f\u305f\u308b\u8907\u96d1\u306a\u30b7\u30ca\u30ea\u30aa\u306b\u304a\u3044\u3066\u3001BPE\u306f\u771f\u4fa1\u3092\u767a\u63ee\u3059\u308b\u3002\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u304c\u81ea\u8eab\u306e\u5224\u65ad\u3092\u5727\u7e2e\u3057\u3066\u7406\u89e3\u3057\u3001\u904e\u53bb\u306e\u5931\u6557\u3084\u4eba\u9593\u306e\u30d5\u30a3\u30fc\u30c9\u30d0\u30c3\u30af\u304b\u3089\u53cd\u5fa9\u7684\u306b\u6539\u5584\u3059\u308b\u305f\u3081\u306b\u3001Experience\u3078\u306e\u4f9d\u5b58\u304c\u4e0d\u53ef\u6b20\u3068\u306a\u308b\u3002<\/p>\n

    \u3053\u306e\u7814\u7a76\u306f\u3001AI\u30aa\u30fc\u30b1\u30b9\u30c8\u30ec\u30fc\u30b7\u30e7\u30f3\u30a8\u30f3\u30b8\u30cb\u30a2\u306b\u3068\u3063\u3066\u3001\u30ef\u30fc\u30af\u30d5\u30ed\u30fc\u30a8\u30f3\u30b8\u30cb\u30a2\u30ea\u30f3\u30b0\u306e\u5b8c\u5168\u306a\u7f6e\u304d\u63db\u3048\u3067\u306f\u306a\u304f\u3001\u300c\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u306e\u884c\u52d5\u3092\u76f4\u63a5\u30b9\u30af\u30ea\u30d7\u30c8\u5316\u3059\u308b\u300d\u304b\u3089\u300c\u3088\u308a\u826f\u3044\u884c\u52d5\u3092\u5b66\u7fd2\u3067\u304d\u308b\u30b7\u30b9\u30c6\u30e0\u3092\u69cb\u7bc9\u3059\u308b\u300d\u3078\u306e\u30d1\u30e9\u30c0\u30a4\u30e0\u30b7\u30d5\u30c8<\/strong>\u3092\u793a\u5506\u3057\u3066\u3044\u308b\u3002EvoHarness-RL\u306f\u3001\u5c0f\u578b\u30e2\u30c7\u30eb\u3067\u3082\u9ad8\u5ea6\u306a\u63a8\u8ad6\u3068\u81ea\u5f8b\u6027\u3092\u5b9f\u73fe\u3067\u304d\u308b\u53ef\u80fd\u6027\u3092\u793a\u3059\u3068\u540c\u6642\u306b\u3001\u30a8\u30f3\u30bf\u30fc\u30d7\u30e9\u30a4\u30baAI\u306e\u5b9f\u88c5\u65b9\u6cd5\u305d\u306e\u3082\u306e\u306b\u5909\u9769\u3092\u4fc3\u3059\u3082\u306e\u3068\u8a00\u3048\u308b\u3060\u308d\u3046\u3002<\/p>\n","protected":false},"excerpt":{"rendered":"

    Meta AI\uff08\u30e1\u30bfAI\uff09\u3068\u30a4\u30ea\u30ce\u30a4\u5927\u5b66\u30a2\u30fc\u30d0\u30ca\u30fb\u30b7\u30e3\u30f3\u30da\u30fc\u30f3\u6821\u306e\u7814\u7a76\u30c1\u30fc\u30e0\u306f\u3001AI\u30a8\u30fc\u30b8\u30a7\u30f3\u30c8\u306e\u9577\u671f\u30bf\u30b9\u30af\u5b9f\u884c\u80fd\u529b\u3092\u98db\u8e8d\u7684\u306b\u5411\u4e0a\u3055\u305b\u308b\u65b0\u30d5\u30ec\u30fc\u30e0\u30ef\u30fc\u30af\u300cEvoHarness-RL\u300d\u3092\u767a\u8868\u3057\u305f\u3002\u3053\u306e\u624b\u6cd5\u3092\u7528\u3044\u308b\u3053\u3068\u3067\u3001\u30d1 […]<\/p>\n","protected":false},"author":1,"featured_media":56189,"comment_status":"closed","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"fifu_image_url":"https:\/\/cards.overcentral.com\/cards\/ja\/53225.png","fifu_image_alt":"\u6210\u529f\u738796.9%\u3001Met","footnotes":""},"categories":[2428],"tags":[],"class_list":["post-53225","post","type-post","status-publish","format-standard","has-post-thumbnail","category-technology"],"fifu_image_url":"https:\/\/cards.overcentral.com\/cards\/ja\/53225.png","fifu_image_alt":"\u6210\u529f\u738796.9%\u3001Met","_links":{"self":[{"href":"https:\/\/overcentral.com\/ja\/wp-json\/wp\/v2\/posts\/53225","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/overcentral.com\/ja\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/overcentral.com\/ja\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/overcentral.com\/ja\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/overcentral.com\/ja\/wp-json\/wp\/v2\/comments?post=53225"}],"version-history":[{"count":1,"href":"https:\/\/overcentral.com\/ja\/wp-json\/wp\/v2\/posts\/53225\/revisions"}],"predecessor-version":[{"id":53226,"href":"https:\/\/overcentral.com\/ja\/wp-json\/wp\/v2\/posts\/53225\/revisions\/53226"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/overcentral.com\/ja\/wp-json\/wp\/v2\/media\/56189"}],"wp:attachment":[{"href":"https:\/\/overcentral.com\/ja\/wp-json\/wp\/v2\/media?parent=53225"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/overcentral.com\/ja\/wp-json\/wp\/v2\/categories?post=53225"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/overcentral.com\/ja\/wp-json\/wp\/v2\/tags?post=53225"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}