{"id":97816,"date":"2026-09-19T19:30:12","date_gmt":"2026-09-19T16:30:12","guid":{"rendered":"https:\/\/twodots.gr\/?p=97816"},"modified":"2026-09-19T19:30:14","modified_gmt":"2026-09-19T16:30:14","slug":"lacl-gui-gui-agents-poiotita-diadromis","status":"publish","type":"post","link":"https:\/\/twodots.gr\/bg\/lacl-gui-gui-agents-poiotita-diadromis\/","title":{"rendered":"\u03a0\u03ad\u03c1\u03b1 \u03b1\u03c0\u03cc \u03c4\u03b7\u03bd \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b1 \u03ba\u03b1\u03b9 \u03c4\u03b7\u03bd \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b1: \u03c0\u03ce\u03c2 \u03bf\u03b9 GUI agents \u03bc\u03b1\u03b8\u03b1\u03af\u03bd\u03bf\u03c5\u03bd \u03b1\u03c0\u03cc \u03c4\u03b7 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ae"},"content":{"rendered":"<div class=\"td-article-lede\">\n<p><strong>\u0391\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7 \u03c0\u03c1\u03ce\u03c4\u03b1:<\/strong> \u03b7 LACL-GUI \u03b4\u03b5\u03af\u03c7\u03bd\u03b5\u03b9 \u03cc\u03c4\u03b9 \u03b3\u03b9\u03b1 \u03c4\u03b7\u03bd \u03b5\u03ba\u03c0\u03b1\u03af\u03b4\u03b5\u03c5\u03c3\u03b7 GUI agents \u03b4\u03b5\u03bd \u03b1\u03c1\u03ba\u03b5\u03af \u03bd\u03b1 \u03b3\u03bd\u03c9\u03c1\u03af\u03b6\u03bf\u03c5\u03bc\u03b5 \u03b1\u03bd \u03bc\u03b9\u03b1 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 \u03c4\u03b5\u03bb\u03b5\u03af\u03c9\u03c3\u03b5. \u0394\u03cd\u03bf \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b5\u03c2 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03ad\u03c7\u03bf\u03c5\u03bd \u03c4\u03b5\u03c1\u03ac\u03c3\u03c4\u03b9\u03b1 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03ac \u03c3\u03b5 \u03c0\u03b5\u03c1\u03b9\u03c4\u03c4\u03ac \u03ba\u03bb\u03b9\u03ba, latency \u03ba\u03b1\u03b9 \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2, \u03b5\u03bd\u03ce \u03b4\u03cd\u03bf \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b5\u03c2 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b1\u03c0\u03ad\u03c7\u03bf\u03c5\u03bd \u03b1\u03c0\u03cc \u03c4\u03bf\u03bd \u03c3\u03c4\u03cc\u03c7\u03bf \u03ba\u03b1\u03c4\u03ac \u03ad\u03bd\u03b1 \u03b2\u03ae\u03bc\u03b1 \u03ae \u03ba\u03b1\u03c4\u03ac \u03bf\u03bb\u03cc\u03ba\u03bb\u03b7\u03c1\u03b7 \u03c4\u03b7 \u03b4\u03b9\u03b1\u03b4\u03b9\u03ba\u03b1\u03c3\u03af\u03b1.<\/p>\n<p>\u0397 \u03bc\u03ad\u03b8\u03bf\u03b4\u03bf\u03c2 \u03c0\u03c1\u03bf\u03c3\u03b8\u03ad\u03c4\u03b5\u03b9 \u03b4\u03cd\u03bf \u03c3\u03ae\u03bc\u03b1\u03c4\u03b1 \u03c0\u03ac\u03bd\u03c9 \u03b1\u03c0\u03cc \u03c4\u03bf \u03b4\u03c5\u03b1\u03b4\u03b9\u03ba\u03cc outcome: \u03c0\u03c1\u03bf\u03c4\u03b9\u03bc\u03ac \u03c4\u03b9\u03c2 \u03c0\u03b9\u03bf \u03c3\u03cd\u03bd\u03c4\u03bf\u03bc\u03b5\u03c2 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b5\u03c2 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ad\u03c2 \u03c4\u03b7\u03c2 \u03af\u03b4\u03b9\u03b1\u03c2 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1\u03c2 \u03ba\u03b1\u03b9 \u03b4\u03b9\u03b1\u03ba\u03c1\u03af\u03bd\u03b5\u03b9 \u03c4\u03b9\u03c2 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b5\u03c2 \u03b1\u03c0\u03cc \u03c4\u03bf \u03c3\u03b7\u03bc\u03b5\u03af\u03bf \u03cc\u03c0\u03bf\u03c5 \u03b1\u03c0\u03bf\u03ba\u03bb\u03af\u03bd\u03bf\u03c5\u03bd \u03b1\u03c0\u03cc \u03bc\u03b9\u03b1 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b7 \u03b1\u03bd\u03b1\u03c6\u03bf\u03c1\u03ac. \u0393\u03b9\u03b1 product teams, \u03c4\u03bf \u03c7\u03c1\u03ae\u03c3\u03b9\u03bc\u03bf \u03c3\u03c5\u03bc\u03c0\u03ad\u03c1\u03b1\u03c3\u03bc\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03c3\u03b1\u03c6\u03ad\u03c2: completion rate, \u03c0\u03bf\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ae\u03c2 \u03ba\u03b1\u03b9 \u03b1\u03c3\u03c6\u03ac\u03bb\u03b5\u03b9\u03b1 \u03b5\u03c0\u03b1\u03bb\u03ae\u03b8\u03b5\u03c5\u03c3\u03b7\u03c2 \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03bc\u03b5\u03c4\u03c1\u03ce\u03bd\u03c4\u03b1\u03b9 \u03c7\u03c9\u03c1\u03b9\u03c3\u03c4\u03ac.<\/p>\n<\/div>\n<div class=\"td-article-toc\">\n<div class=\"td-toc-title\">\u0421\u044a\u0434\u044a\u0440\u0436\u0430\u043d\u0438\u0435<\/div>\n<ul>\n<li><a href=\"#outcome-den-arkei\">\u0393\u03b9\u03b1\u03c4\u03af \u03c4\u03bf \u03c4\u03b5\u03bb\u03b9\u03ba\u03cc \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1 \u03b4\u03b5\u03bd \u03b1\u03c1\u03ba\u03b5\u03af<\/a><\/li>\n<li><a href=\"#grpo-real-orio\">\u03a4\u03bf \u03cc\u03c1\u03b9\u03bf \u03c4\u03c9\u03bd GRPO \u03ba\u03b1\u03b9 REAL<\/a><\/li>\n<li><a href=\"#epityximenes-diadromes\">\u03a0\u03ce\u03c2 \u03b1\u03be\u03b9\u03bf\u03bb\u03bf\u03b3\u03bf\u03cd\u03bd\u03c4\u03b1\u03b9 \u03bf\u03b9 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b5\u03c2 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ad\u03c2<\/a><\/li>\n<li><a href=\"#kaliteri-apotyxia\">\u03a4\u03b9 \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u00ab\u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b7 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b1\u00bb<\/a><\/li>\n<li><a href=\"#ypodomi-lacl\">\u0397 \u03c5\u03c0\u03bf\u03b4\u03bf\u03bc\u03ae \u03c0\u03af\u03c3\u03c9 \u03b1\u03c0\u03cc \u03c4\u03b7 LACL-GUI<\/a><\/li>\n<li><a href=\"#osworld-apotelesmata\">\u03a4\u03b9 \u03ad\u03b4\u03b5\u03b9\u03be\u03b1\u03bd \u03bf\u03b9 222 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b5\u03c2 \u03c4\u03bf\u03c5 OSWorld<\/a><\/li>\n<li><a href=\"#ligotera-vimata\">\u039b\u03b9\u03b3\u03cc\u03c4\u03b5\u03c1\u03b1 \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1, \u03b1\u03bb\u03bb\u03ac \u03cc\u03c7\u03b9 \u03c0\u03b1\u03bd\u03c4\u03bf\u03cd<\/a><\/li>\n<li><a href=\"#gimp-paradeigma\">\u03a4\u03bf \u03c0\u03b1\u03c1\u03ac\u03b4\u03b5\u03b9\u03b3\u03bc\u03b1 \u03c4\u03bf\u03c5 Gimp \u03ba\u03ac\u03bd\u03b5\u03b9 \u03c4\u03b7 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03ac \u03bf\u03c1\u03b1\u03c4\u03ae<\/a><\/li>\n<li><a href=\"#production-metafora\">\u0391\u03c0\u03cc \u03c4\u03bf paper \u03c3\u03b5 production GUI agents<\/a><\/li>\n<li><a href=\"#elegxoi-ylopoiisis\">\u0395\u03c0\u03c4\u03ac \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf\u03b9 \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03b9\u03ba\u03cc rollout<\/a><\/li>\n<li><a href=\"#oria-meletis\">\u038c\u03c1\u03b9\u03b1 \u03c0\u03bf\u03c5 \u03b4\u03b5\u03bd \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03ba\u03c1\u03cd\u03b2\u03bf\u03bd\u03c4\u03b1\u03b9<\/a><\/li>\n<li><a href=\"#symperasma\">\u03a4\u03bf \u03c3\u03c9\u03c3\u03c4\u03cc \u03c3\u03c5\u03bc\u03c0\u03ad\u03c1\u03b1\u03c3\u03bc\u03b1<\/a><\/li>\n<\/ul>\n<\/div>\n<h2 id=\"outcome-den-arkei\">\u0393\u03b9\u03b1\u03c4\u03af \u03c4\u03bf \u03c4\u03b5\u03bb\u03b9\u03ba\u03cc \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1 \u03b4\u03b5\u03bd \u03b1\u03c1\u03ba\u03b5\u03af<\/h2>\n<p>\u039f\u03b9 GUI agents \u03b5\u03ba\u03c4\u03b5\u03bb\u03bf\u03cd\u03bd \u03b1\u03ba\u03bf\u03bb\u03bf\u03c5\u03b8\u03af\u03b5\u03c2 \u03b5\u03bd\u03b5\u03c1\u03b3\u03b5\u03b9\u03ce\u03bd \u03c3\u03b5 web, desktop \u03ae mobile \u03c0\u03b5\u03c1\u03b9\u03b2\u03ac\u03bb\u03bb\u03bf\u03bd\u03c4\u03b1, \u03c3\u03c5\u03bd\u03b4\u03c5\u03ac\u03b6\u03bf\u03bd\u03c4\u03b1\u03c2 screenshots, \u03c0\u03c1\u03cc\u03c3\u03c6\u03b1\u03c4\u03bf \u03b9\u03c3\u03c4\u03bf\u03c1\u03b9\u03ba\u03cc, reasoning \u03ba\u03b1\u03b9 \u03b4\u03bf\u03bc\u03b7\u03bc\u03ad\u03bd\u03b5\u03c2 \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b5\u03c2. \u03a3\u03b5 \u03c0\u03bf\u03bb\u03bb\u03ac \u03c3\u03c5\u03c3\u03c4\u03ae\u03bc\u03b1\u03c4\u03b1 Reinforcement Learning with Verifiable Rewards, \u03b7 \u03b1\u03bd\u03c4\u03b1\u03bc\u03bf\u03b9\u03b2\u03ae \u03b4\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03bc\u03cc\u03bd\u03bf \u03c3\u03c4\u03bf \u03c4\u03ad\u03bb\u03bf\u03c2: 1 \u03cc\u03c4\u03b1\u03bd \u03b7 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 \u03bf\u03bb\u03bf\u03ba\u03bb\u03b7\u03c1\u03c9\u03b8\u03b5\u03af \u03ba\u03b1\u03b9 0 \u03cc\u03c4\u03b1\u03bd \u03b1\u03c0\u03bf\u03c4\u03cd\u03c7\u03b5\u03b9. \u039a\u03ac\u03b8\u03b5 \u03b2\u03ae\u03bc\u03b1 \u03c4\u03b7\u03c2 \u03af\u03b4\u03b9\u03b1\u03c2 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ae\u03c2 \u03ba\u03bb\u03b7\u03c1\u03bf\u03bd\u03bf\u03bc\u03b5\u03af \u03ad\u03c4\u03c3\u03b9 \u03c4\u03bf \u03af\u03b4\u03b9\u03bf outcome.<\/p>\n<p>\u03a4\u03bf \u03c3\u03ae\u03bc\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03b1\u03c0\u03bb\u03cc \u03ba\u03b1\u03b9 \u03b5\u03c0\u03b1\u03bb\u03b7\u03b8\u03b5\u03cd\u03c3\u03b9\u03bc\u03bf, \u03b1\u03bb\u03bb\u03ac \u03c3\u03c5\u03bc\u03c0\u03b9\u03ad\u03b6\u03b5\u03b9 \u03c3\u03b7\u03bc\u03b1\u03bd\u03c4\u03b9\u03ba\u03ad\u03c2 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03ad\u03c2. \u0388\u03bd\u03b1\u03c2 agent \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03c0\u03b5\u03c4\u03cd\u03c7\u03b5\u03b9 \u03bc\u03b5 \u03ad\u03be\u03b9 \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b5\u03c2 \u03ba\u03b1\u03b9 \u03ad\u03bd\u03b1\u03c2 \u03b4\u03b5\u03cd\u03c4\u03b5\u03c1\u03bf\u03c2 \u03bc\u03b5 \u03b5\u03af\u03ba\u03bf\u03c3\u03b9 \u03b5\u03c0\u03c4\u03ac, \u03ad\u03c0\u03b5\u03b9\u03c4\u03b1 \u03b1\u03c0\u03cc \u03b5\u03c0\u03b1\u03bd\u03b1\u03bb\u03ae\u03c8\u03b5\u03b9\u03c2 \u03ba\u03b1\u03b9 \u03bb\u03b1\u03bd\u03b8\u03b1\u03c3\u03bc\u03ad\u03bd\u03b1 \u03ba\u03bb\u03b9\u03ba. \u0391\u03bd\u03c4\u03af\u03c3\u03c4\u03bf\u03b9\u03c7\u03b1, \u03bc\u03af\u03b1 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b1 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03c3\u03c4\u03b1\u03bc\u03b1\u03c4\u03ae\u03c3\u03b5\u03b9 \u03bb\u03af\u03b3\u03bf \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03c4\u03b7\u03bd \u03b5\u03c0\u03b9\u03b2\u03b5\u03b2\u03b1\u03af\u03c9\u03c3\u03b7 \u03ba\u03b1\u03b9 \u03bc\u03af\u03b1 \u03ac\u03bb\u03bb\u03b7 \u03bd\u03b1 \u03b1\u03bd\u03bf\u03af\u03be\u03b5\u03b9 \u03bb\u03ac\u03b8\u03bf\u03c2 \u03bc\u03b5\u03bd\u03bf\u03cd \u03b1\u03c0\u03cc \u03c4\u03bf \u03c0\u03c1\u03ce\u03c4\u03bf \u03b2\u03ae\u03bc\u03b1. \u0393\u03b9\u03b1 \u03bc\u03b9\u03b1 \u03b5\u03c0\u03b9\u03c7\u03b5\u03af\u03c1\u03b7\u03c3\u03b7, \u03b1\u03c5\u03c4\u03ad\u03c2 \u03bf\u03b9 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ad\u03c2 \u03b4\u03b5\u03bd \u03ad\u03c7\u03bf\u03c5\u03bd \u03c4\u03bf \u03af\u03b4\u03b9\u03bf \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2 \u03bf\u03cd\u03c4\u03b5 \u03c4\u03bf\u03bd \u03af\u03b4\u03b9\u03bf \u03ba\u03af\u03bd\u03b4\u03c5\u03bd\u03bf.<\/p>\n<p>\u0397 \u03b1\u03bd\u03ac\u03b3\u03ba\u03b7 \u03b1\u03c5\u03c4\u03ae \u03c3\u03c5\u03bc\u03c0\u03bb\u03b7\u03c1\u03ce\u03bd\u03b5\u03b9 \u03c4\u03bf <a href=\"https:\/\/twodots.gr\/ai-agents-behavioral-testing\/\">behavioral testing \u03b3\u03b9\u03b1 AI agents<\/a>: \u03c4\u03bf aggregate success rate \u03b4\u03b5\u03bd \u03b4\u03b5\u03af\u03c7\u03bd\u03b5\u03b9 \u03c0\u03bf\u03b9\u03b5\u03c2 \u03c3\u03c5\u03bc\u03c0\u03b5\u03c1\u03b9\u03c6\u03bf\u03c1\u03ad\u03c2 \u03c0\u03b1\u03c1\u03ac\u03b3\u03bf\u03c5\u03bd \u03c4\u03b7\u03bd \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b1, \u03c0\u03cc\u03c3\u03bf \u03b5\u03cd\u03b8\u03c1\u03b1\u03c5\u03c3\u03c4\u03b7 \u03b5\u03af\u03bd\u03b1\u03b9 \u03b7 \u03c0\u03bf\u03c1\u03b5\u03af\u03b1 \u03ae \u03c0\u03cc\u03c4\u03b5 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03b7 \u03c0\u03b1\u03c1\u03ad\u03bc\u03b2\u03b1\u03c3\u03b7.<\/p>\n<div class=\"td-comparison td-comparison-cards td-comparison-cards--horizontal\">\n<div class=\"td-comparison-grid td-comparison-grid--three\">\n<div class=\"td-platform-card\">\n<p class=\"td-comparison-title\">Outcome<\/p>\n<p>\u0391\u03c0\u03b1\u03bd\u03c4\u03ac \u03b1\u03bd \u03bf verifier \u03b5\u03c0\u03b9\u03b2\u03b5\u03b2\u03b1\u03af\u03c9\u03c3\u03b5 \u03c4\u03bf \u03c4\u03b5\u03bb\u03b9\u03ba\u03cc state. \u0395\u03af\u03bd\u03b1\u03b9 \u03b1\u03c0\u03b1\u03c1\u03b1\u03af\u03c4\u03b7\u03c4\u03bf, \u03b1\u03bb\u03bb\u03ac \u03b4\u03b5\u03bd \u03b5\u03be\u03b7\u03b3\u03b5\u03af \u03c0\u03ce\u03c2 \u03ad\u03c6\u03c4\u03b1\u03c3\u03b5 \u03b5\u03ba\u03b5\u03af \u03bf GUI agent.<\/p>\n<div class=\"td-badge-row\"><span class=\"td-badge\">Success<\/span><span class=\"td-badge\">Failure<\/span><\/div>\n<\/div>\n<div class=\"td-platform-card td-platform-card--navy\">\n<p class=\"td-comparison-title\">\u039c\u03ae\u03ba\u03bf\u03c2 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ae\u03c2<\/p>\n<p>\u03a3\u03c5\u03b3\u03ba\u03c1\u03af\u03bd\u03b5\u03b9 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b5\u03c2 \u03c4\u03bf\u03c5 \u03af\u03b4\u03b9\u03bf\u03c5 task \u03ba\u03b1\u03b9 \u03b5\u03bd\u03c4\u03bf\u03c0\u03af\u03b6\u03b5\u03b9 \u03b5\u03c0\u03b1\u03bd\u03b1\u03bb\u03ae\u03c8\u03b5\u03b9\u03c2, \u03ac\u03c3\u03ba\u03bf\u03c0\u03b7 \u03b5\u03be\u03b5\u03c1\u03b5\u03cd\u03bd\u03b7\u03c3\u03b7 \u03ba\u03b1\u03b9 \u03b4\u03b9\u03bf\u03c1\u03b8\u03ce\u03c3\u03b5\u03b9\u03c2 \u03bc\u03b5\u03c4\u03ac \u03b1\u03c0\u03cc \u03bb\u03ac\u03b8\u03bf\u03c2.<\/p>\n<div class=\"td-badge-row\"><span class=\"td-badge\">\u039a\u03cc\u03c3\u03c4\u03bf\u03c2<\/span><span class=\"td-badge\">Latency<\/span><\/div>\n<\/div>\n<div class=\"td-platform-card\">\n<p class=\"td-comparison-title\">\u03a3\u03b7\u03bc\u03b5\u03af\u03bf \u03b1\u03c0\u03cc\u03ba\u03bb\u03b9\u03c3\u03b7\u03c2<\/p>\n<p>\u0394\u03b9\u03b1\u03c7\u03c9\u03c1\u03af\u03b6\u03b5\u03b9 \u03bc\u03b9\u03b1 near-success \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b1 \u03b1\u03c0\u03cc \u03c0\u03bf\u03c1\u03b5\u03af\u03b1 \u03c0\u03bf\u03c5 \u03ad\u03b3\u03b9\u03bd\u03b5 \u03bb\u03b1\u03bd\u03b8\u03b1\u03c3\u03bc\u03ad\u03bd\u03b7 \u03bd\u03c9\u03c1\u03af\u03c2 \u03ba\u03b1\u03b9 \u03c3\u03c5\u03bd\u03ad\u03c7\u03b9\u03c3\u03b5 \u03c7\u03c9\u03c1\u03af\u03c2 \u03b1\u03bd\u03ac\u03ba\u03b1\u03bc\u03c8\u03b7.<\/p>\n<div class=\"td-badge-row\"><span class=\"td-badge\">Divergence<\/span><span class=\"td-badge\">Recovery<\/span><\/div>\n<\/div>\n<\/div>\n<\/div>\n<h2 id=\"grpo-real-orio\">\u03a4\u03bf \u03cc\u03c1\u03b9\u03bf \u03c4\u03c9\u03bd GRPO \u03ba\u03b1\u03b9 REAL<\/h2>\n<p>\u0397 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 \u03be\u03b5\u03ba\u03b9\u03bd\u03ac \u03b1\u03c0\u03cc \u03b3\u03bd\u03c9\u03c3\u03c4\u03ad\u03c2 \u03b1\u03b4\u03c5\u03bd\u03b1\u03bc\u03af\u03b5\u03c2 \u03c4\u03bf\u03c5 Group Relative Policy Optimization. \u039f\u03b9 \u03c3\u03c5\u03b3\u03b3\u03c1\u03b1\u03c6\u03b5\u03af\u03c2 \u03c3\u03c5\u03bd\u03bf\u03c8\u03af\u03b6\u03bf\u03c5\u03bd \u03b4\u03cd\u03bf gradient pathologies: \u03c3\u03c4\u03b9\u03c2 \u03b8\u03b5\u03c4\u03b9\u03ba\u03ad\u03c2 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ad\u03c2, tokens \u03c5\u03c8\u03b7\u03bb\u03ae\u03c2 \u03c0\u03b9\u03b8\u03b1\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1\u03c2 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03bf\u03c5\u03bd \u03b4\u03c5\u03c3\u03b1\u03bd\u03ac\u03bb\u03bf\u03b3\u03b1 \u03b9\u03c3\u03c7\u03c5\u03c1\u03ad\u03c2 \u03b5\u03bd\u03b7\u03bc\u03b5\u03c1\u03ce\u03c3\u03b5\u03b9\u03c2\u00b7 \u03c3\u03c4\u03b9\u03c2 \u03b1\u03c1\u03bd\u03b7\u03c4\u03b9\u03ba\u03ad\u03c2, \u03bb\u03af\u03b3\u03b1 tokens \u03c5\u03c8\u03b7\u03bb\u03ae\u03c2 \u03c0\u03b9\u03b8\u03b1\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1\u03c2 \u03bc\u03c0\u03bf\u03c1\u03bf\u03cd\u03bd \u03bd\u03b1 \u03ba\u03c5\u03c1\u03b9\u03b1\u03c1\u03c7\u03bf\u03cd\u03bd \u03c3\u03c4\u03b7\u03bd \u03b5\u03bd\u03b7\u03bc\u03ad\u03c1\u03c9\u03c3\u03b7. \u0397 \u03ba\u03b1\u03c4\u03b1\u03bd\u03bf\u03bc\u03ae \u03c4\u03bf\u03c5 gradient \u03b4\u03b5\u03bd \u03b1\u03ba\u03bf\u03bb\u03bf\u03c5\u03b8\u03b5\u03af \u03c0\u03ac\u03bd\u03c4\u03b1 \u03c0\u03b9\u03c3\u03c4\u03ac \u03c4\u03b7\u03bd \u03c0\u03bf\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1 \u03bf\u03bb\u03cc\u03ba\u03bb\u03b7\u03c1\u03b7\u03c2 \u03c4\u03b7\u03c2 trajectory.<\/p>\n<p>\u0397 \u03bc\u03ad\u03b8\u03bf\u03b4\u03bf\u03c2 REAL \u03b1\u03bd\u03c4\u03b9\u03bc\u03b5\u03c4\u03c9\u03c0\u03af\u03b6\u03b5\u03b9 \u03c4\u03bf RLVR \u03c9\u03c2 \u03c0\u03c1\u03cc\u03b2\u03bb\u03b7\u03bc\u03b1 \u03c4\u03b1\u03be\u03b9\u03bd\u03cc\u03bc\u03b7\u03c3\u03b7\u03c2 \u03ba\u03b1\u03b9 \u03c0\u03c1\u03bf\u03c3\u03c6\u03ad\u03c1\u03b5\u03b9 \u03c0\u03b9\u03bf \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03b7, \u03c6\u03c1\u03b1\u03b3\u03bc\u03ad\u03bd\u03b7 \u03ba\u03b1\u03c4\u03b1\u03bd\u03bf\u03bc\u03ae gradient. \u038c\u03bc\u03c9\u03c2 \u03c4\u03bf label \u03bc\u03ad\u03bd\u03b5\u03b9 \u03b4\u03c5\u03b1\u03b4\u03b9\u03ba\u03cc: \u03cc\u03bb\u03b5\u03c2 \u03bf\u03b9 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b5\u03c2 \u03bc\u03c0\u03b1\u03af\u03bd\u03bf\u03c5\u03bd \u03c3\u03c4\u03b7 \u03b8\u03b5\u03c4\u03b9\u03ba\u03ae \u03bf\u03bc\u03ac\u03b4\u03b1 \u03ba\u03b1\u03b9 \u03cc\u03bb\u03b5\u03c2 \u03bf\u03b9 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b5\u03c2 \u03c3\u03c4\u03b7\u03bd \u03b1\u03c1\u03bd\u03b7\u03c4\u03b9\u03ba\u03ae. \u0397 LACL-GUI \u03ba\u03c1\u03b1\u03c4\u03ac \u03b1\u03c5\u03c4\u03cc \u03c4\u03bf \u03b2\u03b1\u03c3\u03b9\u03ba\u03cc \u03cc\u03c1\u03b9\u03bf, \u03b1\u03bb\u03bb\u03ac \u03c0\u03c1\u03bf\u03c3\u03b8\u03ad\u03c4\u03b5\u03b9 \u03c3\u03c7\u03b5\u03c4\u03b9\u03ba\u03ad\u03c2 \u03c0\u03c1\u03bf\u03c4\u03b9\u03bc\u03ae\u03c3\u03b5\u03b9\u03c2 \u03bc\u03ad\u03c3\u03b1 \u03c3\u03b5 \u03ba\u03ac\u03b8\u03b5 \u03bf\u03bc\u03ac\u03b4\u03b1.<\/p>\n<p>\u0397 \u03b4\u03b9\u03ac\u03ba\u03c1\u03b9\u03c3\u03b7 \u03ad\u03c7\u03b5\u03b9 \u03c3\u03b7\u03bc\u03b1\u03c3\u03af\u03b1 \u03b3\u03b9\u03b1 \u03ba\u03ac\u03b8\u03b5 \u03bf\u03bc\u03ac\u03b4\u03b1 \u03c0\u03bf\u03c5 \u03b5\u03be\u03b5\u03c4\u03ac\u03b6\u03b5\u03b9 \u03c4\u03bf <a href=\"https:\/\/twodots.gr\/sparse-reward-ai-hints-critics-teachers\/\">sparse reward \u03c3\u03c4\u03b7\u03bd \u03b5\u03ba\u03c0\u03b1\u03af\u03b4\u03b5\u03c5\u03c3\u03b7 AI<\/a>. \u0388\u03bd\u03b1 \u03b5\u03c0\u03b1\u03bb\u03b7\u03b8\u03b5\u03cd\u03c3\u03b9\u03bc\u03bf \u03c4\u03b5\u03bb\u03b9\u03ba\u03cc score \u03b5\u03af\u03bd\u03b1\u03b9 \u03c7\u03c1\u03ae\u03c3\u03b9\u03bc\u03bf \u03bc\u03cc\u03bd\u03bf \u03cc\u03c4\u03b1\u03bd \u03c4\u03bf learning pipeline \u03b4\u03b9\u03b1\u03c4\u03b7\u03c1\u03b5\u03af \u03b1\u03c1\u03ba\u03b5\u03c4\u03ae \u03c0\u03bb\u03b7\u03c1\u03bf\u03c6\u03bf\u03c1\u03af\u03b1 \u03b3\u03b9\u03b1 \u03c4\u03b9\u03c2 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03ad\u03c2 \u03c0\u03bf\u03b9\u03cc\u03c4\u03b7\u03c4\u03b5\u03c2 \u03c4\u03c9\u03bd rollouts.<\/p>\n<aside class=\"td-article-note\"><strong>\u0394\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03ba\u03b1\u03b8\u03bf\u03bb\u03b9\u03ba\u03cc length penalty:<\/strong> \u03b7 LACL-GUI \u03b4\u03b5\u03bd \u03b5\u03c0\u03b9\u03b2\u03c1\u03b1\u03b2\u03b5\u03cd\u03b5\u03b9 \u03bc\u03b9\u03b1 \u03c3\u03cd\u03bd\u03c4\u03bf\u03bc\u03b7 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b1 \u03c0\u03b5\u03c1\u03b9\u03c3\u03c3\u03cc\u03c4\u03b5\u03c1\u03bf \u03b1\u03c0\u03cc \u03bc\u03b9\u03b1 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b1. \u0397 \u03c3\u03c7\u03b5\u03c4\u03b9\u03ba\u03ae \u03c0\u03c1\u03bf\u03c4\u03af\u03bc\u03b7\u03c3\u03b7 \u03b5\u03c6\u03b1\u03c1\u03bc\u03cc\u03b6\u03b5\u03c4\u03b1\u03b9 \u03bc\u03ad\u03c3\u03b1 \u03c3\u03c4\u03b7\u03bd \u03bf\u03bc\u03ac\u03b4\u03b1 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b9\u03ce\u03bd \u03ae \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03b9\u03ce\u03bd \u03c4\u03b7\u03c2 \u03af\u03b4\u03b9\u03b1\u03c2 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1\u03c2.<\/aside>\n<h2 id=\"epityximenes-diadromes\">\u03a0\u03ce\u03c2 \u03b1\u03be\u03b9\u03bf\u03bb\u03bf\u03b3\u03bf\u03cd\u03bd\u03c4\u03b1\u03b9 \u03bf\u03b9 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b5\u03c2 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ad\u03c2<\/h2>\n<p>\u0393\u03b9\u03b1 \u03c4\u03b9\u03c2 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b5\u03c2 trajectories, \u03c4\u03bf \u03c0\u03c1\u03cc\u03c3\u03b8\u03b5\u03c4\u03bf \u03c3\u03ae\u03bc\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03c4\u03bf \u03bc\u03ae\u03ba\u03bf\u03c2. \u038c\u03c4\u03b1\u03bd \u03c0\u03bf\u03bb\u03bb\u03b1\u03c0\u03bb\u03ad\u03c2 \u03b5\u03ba\u03c4\u03b5\u03bb\u03ad\u03c3\u03b5\u03b9\u03c2 \u03be\u03b5\u03ba\u03b9\u03bd\u03bf\u03cd\u03bd \u03b1\u03c0\u03cc \u03c4\u03bf \u03af\u03b4\u03b9\u03bf state \u03ba\u03b1\u03b9 \u03bf\u03bb\u03bf\u03ba\u03bb\u03b7\u03c1\u03ce\u03bd\u03bf\u03c5\u03bd \u03c4\u03bf\u03bd \u03af\u03b4\u03b9\u03bf \u03c3\u03c4\u03cc\u03c7\u03bf, \u03b7 \u03c3\u03c5\u03bd\u03c4\u03bf\u03bc\u03cc\u03c4\u03b5\u03c1\u03b7 \u03b8\u03b5\u03c9\u03c1\u03b5\u03af\u03c4\u03b1\u03b9 \u03c0\u03c1\u03bf\u03c4\u03b9\u03bc\u03cc\u03c4\u03b5\u03c1\u03b7. \u03a4\u03bf \u03c3\u03ba\u03b5\u03c0\u03c4\u03b9\u03ba\u03cc \u03b5\u03af\u03bd\u03b1\u03b9 \u03cc\u03c4\u03b9 \u03c4\u03b1 \u03b5\u03c0\u03b9\u03c0\u03bb\u03ad\u03bf\u03bd \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1 \u03c3\u03c5\u03c7\u03bd\u03ac \u03c0\u03c1\u03bf\u03ad\u03c1\u03c7\u03bf\u03bd\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03b5\u03be\u03b5\u03c1\u03b5\u03cd\u03bd\u03b7\u03c3\u03b7 \u03c7\u03c9\u03c1\u03af\u03c2 \u03c0\u03c1\u03cc\u03bf\u03b4\u03bf, \u03b5\u03c0\u03b1\u03bd\u03b1\u03bb\u03b1\u03bc\u03b2\u03b1\u03bd\u03cc\u03bc\u03b5\u03bd\u03b1 \u03ba\u03bb\u03b9\u03ba \u03ae \u03bb\u03ac\u03b8\u03b7 \u03b1\u03c0\u03cc \u03c4\u03b1 \u03bf\u03c0\u03bf\u03af\u03b1 \u03bf agent \u03b1\u03bd\u03ad\u03ba\u03b1\u03bc\u03c8\u03b5.<\/p>\n<p>\u0397 \u03bc\u03ad\u03b8\u03bf\u03b4\u03bf\u03c2 \u03b5\u03c6\u03b1\u03c1\u03bc\u03cc\u03b6\u03b5\u03b9 \u03bc\u03b9\u03ba\u03c1\u03ae, zero-mean \u03bc\u03b5\u03c4\u03b1\u03b2\u03bf\u03bb\u03ae \u03c3\u03c4\u03b1 logits \u03c4\u03b7\u03c2 \u03b8\u03b5\u03c4\u03b9\u03ba\u03ae\u03c2 \u03bf\u03bc\u03ac\u03b4\u03b1\u03c2. \u039f\u03b9 \u03c3\u03cd\u03bd\u03c4\u03bf\u03bc\u03b5\u03c2 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b5\u03c2 \u03b1\u03c0\u03bf\u03ba\u03c4\u03bf\u03cd\u03bd \u03c3\u03c7\u03b5\u03c4\u03b9\u03ba\u03ac \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b7 \u03b2\u03b1\u03c1\u03cd\u03c4\u03b7\u03c4\u03b1 \u03ba\u03b1\u03b9 \u03bf\u03b9 \u03bc\u03b1\u03ba\u03c1\u03cd\u03c4\u03b5\u03c1\u03b5\u03c2 \u03bc\u03b9\u03ba\u03c1\u03cc\u03c4\u03b5\u03c1\u03b7. \u039c\u03b5 stop-gradient \u03c3\u03c4\u03b1 \u03bf\u03bc\u03b1\u03b4\u03b9\u03ba\u03ac \u03c3\u03c4\u03b1\u03c4\u03b9\u03c3\u03c4\u03b9\u03ba\u03ac, \u03bf\u03b9 \u03c3\u03c5\u03b3\u03b3\u03c1\u03b1\u03c6\u03b5\u03af\u03c2 \u03b1\u03c0\u03bf\u03b4\u03b5\u03b9\u03ba\u03bd\u03cd\u03bf\u03c5\u03bd first-order invariance: \u03b7 objective \u03b4\u03b9\u03b1\u03c4\u03b7\u03c1\u03b5\u03af \u03c4\u03b7 REAL \u03bc\u03ad\u03c7\u03c1\u03b9 \u03cc\u03c1\u03bf\u03c5\u03c2 \u03b4\u03b5\u03cd\u03c4\u03b5\u03c1\u03b7\u03c2 \u03c4\u03ac\u03be\u03b7\u03c2 \u03c9\u03c2 \u03c0\u03c1\u03bf\u03c2 \u03c4\u03bf\u03bd \u03c3\u03c5\u03bd\u03c4\u03b5\u03bb\u03b5\u03c3\u03c4\u03ae perturbation.<\/p>\n<p>\u03a4\u03bf \u00ab\u03c3\u03c5\u03bd\u03c4\u03bf\u03bc\u03cc\u03c4\u03b5\u03c1\u03bf\u00bb \u03b4\u03b5\u03bd \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b1\u03c0\u03bf\u03c3\u03c5\u03bd\u03b4\u03b5\u03b8\u03b5\u03af \u03b1\u03c0\u03cc \u03c4\u03b7\u03bd \u03bf\u03c1\u03b8\u03cc\u03c4\u03b7\u03c4\u03b1. \u03a3\u03b5 workflow \u03c0\u03bf\u03c5 \u03b1\u03c0\u03b1\u03b9\u03c4\u03b5\u03af \u03b5\u03c0\u03b9\u03b2\u03b5\u03b2\u03b1\u03af\u03c9\u03c3\u03b7 \u03b1\u03c0\u03bf\u03b8\u03ad\u03bc\u03b1\u03c4\u03bf\u03c2, \u03c3\u03c5\u03b3\u03ba\u03b1\u03c4\u03ac\u03b8\u03b5\u03c3\u03b7 \u03c0\u03b5\u03bb\u03ac\u03c4\u03b7 \u03ae \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03b4\u03b7\u03bc\u03bf\u03c3\u03af\u03b5\u03c5\u03c3\u03b7, \u03b7 \u03c0\u03b1\u03c1\u03ac\u03bb\u03b5\u03b9\u03c8\u03b7 \u03b5\u03bd\u03cc\u03c2 \u03b2\u03ae\u03bc\u03b1\u03c4\u03bf\u03c2 \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 efficiency. \u0395\u03af\u03bd\u03b1\u03b9 \u03b5\u03bb\u03ac\u03c4\u03c4\u03c9\u03bc\u03b1. \u0397 \u03c3\u03cd\u03b3\u03ba\u03c1\u03b9\u03c3\u03b7 \u03ad\u03c7\u03b5\u03b9 \u03bd\u03cc\u03b7\u03bc\u03b1 \u03bc\u03cc\u03bd\u03bf \u03b1\u03bd\u03ac\u03bc\u03b5\u03c3\u03b1 \u03c3\u03b5 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ad\u03c2 \u03bc\u03b5 \u03af\u03b4\u03b9\u03bf \u03b1\u03c1\u03c7\u03b9\u03ba\u03cc state, \u03af\u03b4\u03b9\u03bf \u03c3\u03c4\u03cc\u03c7\u03bf \u03ba\u03b1\u03b9 \u03b5\u03c0\u03b1\u03bb\u03b7\u03b8\u03b5\u03c5\u03bc\u03ad\u03bd\u03bf outcome.<\/p>\n<h2 id=\"kaliteri-apotyxia\">\u03a4\u03b9 \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u00ab\u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b7 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b1\u00bb<\/h2>\n<p>\u0393\u03b9\u03b1 \u03c4\u03b9\u03c2 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b5\u03c2 trajectories, \u03c4\u03bf \u03c3\u03c5\u03bd\u03bf\u03bb\u03b9\u03ba\u03cc \u03bc\u03ae\u03ba\u03bf\u03c2 \u03b4\u03b5\u03bd \u03b1\u03c1\u03ba\u03b5\u03af. \u039c\u03b9\u03b1 \u03c3\u03cd\u03bd\u03c4\u03bf\u03bc\u03b7 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b1 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03b1\u03c0\u03bb\u03ce\u03c2 \u03bd\u03b1 \u03b5\u03b3\u03ba\u03b1\u03c4\u03ad\u03bb\u03b5\u03b9\u03c8\u03b5 \u03b1\u03c0\u03cc \u03c4\u03b7\u03bd \u03b1\u03c1\u03c7\u03ae. \u0397 LACL-GUI \u03c3\u03c5\u03b3\u03ba\u03c1\u03af\u03bd\u03b5\u03b9 \u03ba\u03ac\u03b8\u03b5 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b1 \u03bc\u03b5 \u03c4\u03b7 \u03c3\u03c5\u03bd\u03c4\u03bf\u03bc\u03cc\u03c4\u03b5\u03c1\u03b7 \u03b4\u03b9\u03b1\u03b8\u03ad\u03c3\u03b9\u03bc\u03b7 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b7 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ae \u03c4\u03b7\u03c2 \u03af\u03b4\u03b9\u03b1\u03c2 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1\u03c2. \u0397 \u03b1\u03bd\u03b1\u03c6\u03bf\u03c1\u03ac \u03c0\u03c1\u03bf\u03ad\u03c1\u03c7\u03b5\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03c4\u03b9\u03c2 \u03c4\u03c1\u03ad\u03c7\u03bf\u03c5\u03c3\u03b5\u03c2 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b5\u03c2 \u03ae \u03b1\u03c0\u03cc Success Buffer \u03c0\u03bf\u03c5 \u03b1\u03c0\u03bf\u03b8\u03b7\u03ba\u03b5\u03cd\u03b5\u03b9 \u03c0\u03c1\u03bf\u03b7\u03b3\u03bf\u03cd\u03bc\u03b5\u03bd\u03b5\u03c2 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b5\u03c2 trajectories.<\/p>\n<p>\u0397 \u03c3\u03cd\u03b3\u03ba\u03c1\u03b9\u03c3\u03b7 \u03b3\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03b2\u03ae\u03bc\u03b1 \u03c0\u03c1\u03bf\u03c2 \u03b2\u03ae\u03bc\u03b1 \u03c3\u03c4\u03b9\u03c2 \u03bf\u03c0\u03c4\u03b9\u03ba\u03ad\u03c2 \u03c0\u03b1\u03c1\u03b1\u03c4\u03b7\u03c1\u03ae\u03c3\u03b5\u03b9\u03c2 \u03bc\u03b5 Structural Similarity Index. \u03a4\u03bf \u03c0\u03c1\u03ce\u03c4\u03bf \u03c3\u03b7\u03bc\u03b5\u03af\u03bf \u03cc\u03c0\u03bf\u03c5 \u03b7 \u03bf\u03bc\u03bf\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1 \u03c0\u03ad\u03c6\u03c4\u03b5\u03b9 \u03ba\u03ac\u03c4\u03c9 \u03b1\u03c0\u03cc \u03c4\u03bf \u03ba\u03b1\u03c4\u03ce\u03c6\u03bb\u03b9 \u03bf\u03c1\u03af\u03b6\u03b5\u03b9 \u03c4\u03bf divergence length. \u03a3\u03c4\u03b7 \u03c3\u03c5\u03bd\u03ad\u03c7\u03b5\u03b9\u03b1 \u03c5\u03c0\u03bf\u03bb\u03bf\u03b3\u03af\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 \u03c4\u03b1 \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1 \u03c0\u03bf\u03c5 \u03b1\u03ba\u03bf\u03bb\u03bf\u03cd\u03b8\u03b7\u03c3\u03b1\u03bd \u03bc\u03b5\u03c4\u03ac \u03c4\u03b7\u03bd \u03b1\u03c0\u03cc\u03ba\u03bb\u03b9\u03c3\u03b7. \u0391\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b5\u03c2 \u03c0\u03bf\u03c5 \u03ad\u03bc\u03b5\u03b9\u03bd\u03b1\u03bd \u03ba\u03bf\u03bd\u03c4\u03ac \u03c3\u03c4\u03b7 \u03c3\u03c9\u03c3\u03c4\u03ae \u03c0\u03bf\u03c1\u03b5\u03af\u03b1 \u03c4\u03b9\u03bc\u03c9\u03c1\u03bf\u03cd\u03bd\u03c4\u03b1\u03b9 \u03b7\u03c0\u03b9\u03cc\u03c4\u03b5\u03c1\u03b1 \u03b1\u03c0\u03cc \u03b5\u03ba\u03c4\u03b5\u03bb\u03ad\u03c3\u03b5\u03b9\u03c2 \u03c0\u03bf\u03c5 \u03b1\u03c0\u03bf\u03ba\u03bb\u03af\u03bd\u03bf\u03c5\u03bd \u03bd\u03c9\u03c1\u03af\u03c2 \u03ba\u03b1\u03b9 \u03c3\u03c5\u03bd\u03b5\u03c7\u03af\u03b6\u03bf\u03c5\u03bd \u03bc\u03b5 \u03c0\u03bf\u03bb\u03bb\u03ad\u03c2 \u03ac\u03c3\u03c4\u03bf\u03c7\u03b5\u03c2 \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b5\u03c2.<\/p>\n<p>\u0391\u03c5\u03c4\u03cc \u03b4\u03b5\u03bd \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u03cc\u03c4\u03b9 \u03b7 \u03bf\u03c0\u03c4\u03b9\u03ba\u03ae \u03bf\u03bc\u03bf\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b1\u03c0\u03bf\u03b4\u03b5\u03b9\u03ba\u03bd\u03cd\u03b5\u03b9 \u03c3\u03c9\u03c3\u03c4\u03ae \u03c0\u03c1\u03cc\u03b8\u03b5\u03c3\u03b7. \u0394\u03cd\u03bf screenshots \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03bc\u03bf\u03b9\u03ac\u03b6\u03bf\u03c5\u03bd \u03b5\u03bd\u03ce \u03c4\u03bf \u03c5\u03c0\u03bf\u03ba\u03b5\u03af\u03bc\u03b5\u03bd\u03bf state \u03b4\u03b9\u03b1\u03c6\u03ad\u03c1\u03b5\u03b9 \u03ae \u03bd\u03b1 \u03b4\u03b9\u03b1\u03c6\u03ad\u03c1\u03bf\u03c5\u03bd \u03bf\u03c0\u03c4\u03b9\u03ba\u03ac \u03c7\u03c9\u03c1\u03af\u03c2 \u03bf\u03c5\u03c3\u03b9\u03b1\u03c3\u03c4\u03b9\u03ba\u03ae \u03b1\u03bb\u03bb\u03b1\u03b3\u03ae \u03c3\u03c4\u03b7 \u03b4\u03bf\u03c5\u03bb\u03b5\u03b9\u03ac. \u03a3\u03b5 production \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03bf \u03c3\u03c5\u03bd\u03b4\u03c5\u03b1\u03c3\u03bc\u03cc\u03c2 visual trace, structured events \u03ba\u03b1\u03b9 \u03b5\u03c0\u03b1\u03bb\u03ae\u03b8\u03b5\u03c5\u03c3\u03b7\u03c2 \u03c4\u03bf\u03c5 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03bf\u03cd state.<\/p>\n<h2 id=\"ypodomi-lacl\">\u0397 \u03c5\u03c0\u03bf\u03b4\u03bf\u03bc\u03ae \u03c0\u03af\u03c3\u03c9 \u03b1\u03c0\u03cc \u03c4\u03b7 LACL-GUI<\/h2>\n<p>\u0397 LACL-GUI \u03c3\u03c5\u03bd\u03bf\u03b4\u03b5\u03cd\u03b5\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03b1\u03c3\u03cd\u03b3\u03c7\u03c1\u03bf\u03bd\u03b7 \u03b1\u03c1\u03c7\u03b9\u03c4\u03b5\u03ba\u03c4\u03bf\u03bd\u03b9\u03ba\u03ae \u03c4\u03b5\u03c3\u03c3\u03ac\u03c1\u03c9\u03bd \u03bc\u03b5\u03c1\u03ce\u03bd. \u03a4\u03bf Rollout Service \u03c3\u03c5\u03bb\u03bb\u03ad\u03b3\u03b5\u03b9 trajectories \u03c0\u03b1\u03c1\u03ac\u03bb\u03bb\u03b7\u03bb\u03b1 \u03b1\u03c0\u03cc \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03ac GUI environments. \u03a4\u03bf Data Manager \u03ba\u03c1\u03b1\u03c4\u03ac \u03bf\u03bc\u03ac\u03b4\u03b5\u03c2 \u03c0\u03bf\u03c5 \u03c0\u03b5\u03c1\u03b9\u03ad\u03c7\u03bf\u03c5\u03bd \u03c4\u03cc\u03c3\u03bf \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b5\u03c2 \u03cc\u03c3\u03bf \u03ba\u03b1\u03b9 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b5\u03c2, \u03b5\u03c0\u03b5\u03b9\u03b4\u03ae \u03bf\u03bc\u03ac\u03b4\u03b5\u03c2 \u03bc\u03b5 \u03ad\u03bd\u03b1 \u03bc\u03cc\u03bd\u03bf outcome \u03b4\u03b5\u03bd \u03b4\u03af\u03bd\u03bf\u03c5\u03bd contrastive supervision. \u03a4\u03bf Success Buffer \u03b1\u03c0\u03bf\u03b8\u03b7\u03ba\u03b5\u03cd\u03b5\u03b9 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b5\u03c2 \u03b1\u03bd\u03b1\u03c6\u03bf\u03c1\u03ad\u03c2 \u03b1\u03bd\u03ac task. \u039f Trainer \u03b5\u03bd\u03b7\u03bc\u03b5\u03c1\u03ce\u03bd\u03b5\u03b9 \u03c4\u03b7\u03bd policy \u03ba\u03b1\u03b9 \u03c3\u03c5\u03b3\u03c7\u03c1\u03bf\u03bd\u03af\u03b6\u03b5\u03b9 \u03c0\u03b5\u03c1\u03b9\u03bf\u03b4\u03b9\u03ba\u03ac \u03c4\u03b1 \u03bd\u03ad\u03b1 \u03b2\u03ac\u03c1\u03b7 \u03c3\u03c4\u03bf\u03c5\u03c2 rollout workers.<\/p>\n<p>\u0397 \u03b1\u03c3\u03cd\u03b3\u03c7\u03c1\u03bf\u03bd\u03b7 \u03c3\u03c5\u03bb\u03bb\u03bf\u03b3\u03ae \u03b5\u03af\u03bd\u03b1\u03b9 \u03ba\u03c1\u03af\u03c3\u03b9\u03bc\u03b7 \u03c3\u03b5 long-horizon \u03b1\u03bb\u03bb\u03b7\u03bb\u03b5\u03c0\u03b9\u03b4\u03c1\u03ac\u03c3\u03b5\u03b9\u03c2, \u03cc\u03c0\u03bf\u03c5 \u03bf\u03b9 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ad\u03c2 UI \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b5\u03c2 \u03b5\u03af\u03bd\u03b1\u03b9 \u03b1\u03c1\u03b3\u03ad\u03c2 \u03ba\u03b1\u03b9 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03ac tasks \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 \u03ac\u03bd\u03b9\u03c3\u03bf \u03c7\u03c1\u03cc\u03bd\u03bf. \u03a4\u03bf buffer \u03b1\u03bd\u03c4\u03b9\u03bc\u03b5\u03c4\u03c9\u03c0\u03af\u03b6\u03b5\u03b9 \u03b5\u03c0\u03af\u03c3\u03b7\u03c2 \u03c4\u03b1 sparse rewards: \u03cc\u03c4\u03b1\u03bd \u03ad\u03bd\u03b1 \u03c4\u03c1\u03ad\u03c7\u03bf\u03bd batch \u03b4\u03b5\u03bd \u03ad\u03c7\u03b5\u03b9 \u03b1\u03c1\u03ba\u03b5\u03c4\u03ad\u03c2 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b5\u03c2, \u03c5\u03c0\u03ac\u03c1\u03c7\u03bf\u03c5\u03bd \u03c0\u03b1\u03bb\u03b9\u03cc\u03c4\u03b5\u03c1\u03b5\u03c2 \u03b1\u03bd\u03b1\u03c6\u03bf\u03c1\u03ad\u03c2 \u03b3\u03b9\u03b1 \u03c3\u03cd\u03b3\u03ba\u03c1\u03b9\u03c3\u03b7.<\/p>\n<p>\u0397 \u03bc\u03b5\u03bb\u03ad\u03c4\u03b7 \u03b4\u03b5\u03bd \u03b1\u03c0\u03bf\u03bc\u03bf\u03bd\u03ce\u03bd\u03b5\u03b9 \u03cc\u03bc\u03c9\u03c2 \u03c4\u03bf \u03bb\u03b5\u03b9\u03c4\u03bf\u03c5\u03c1\u03b3\u03b9\u03ba\u03cc \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2 \u03b1\u03c5\u03c4\u03ae\u03c2 \u03c4\u03b7\u03c2 \u03c5\u03c0\u03bf\u03b4\u03bf\u03bc\u03ae\u03c2 \u03c3\u03b5 \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03ae. \u0395\u03ba\u03c4\u03cc\u03c2 \u03b1\u03c0\u03cc inference, \u03bc\u03b9\u03b1 \u03b5\u03c0\u03b9\u03c7\u03b5\u03af\u03c1\u03b7\u03c3\u03b7 \u03b8\u03b1 \u03c7\u03c1\u03b5\u03b9\u03b1\u03b6\u03cc\u03c4\u03b1\u03bd \u03b1\u03c0\u03bf\u03b8\u03ae\u03ba\u03b5\u03c5\u03c3\u03b7 trajectories, screenshots, versioned policies, verifier logs \u03ba\u03b1\u03b9 \u03c0\u03bf\u03bb\u03b9\u03c4\u03b9\u03ba\u03ae \u03b4\u03b9\u03b1\u03c4\u03ae\u03c1\u03b7\u03c3\u03b7\u03c2 \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03c9\u03bd. \u0397 \u03bc\u03b5\u03c4\u03ac\u03b2\u03b1\u03c3\u03b7 \u03b1\u03c0\u03cc \u03ad\u03bd\u03b1 <a href=\"https:\/\/twodots.gr\/ai-agent-demo-paragogi-pente-teixi-epicheirisis\/\">AI agent demo \u03c3\u03c4\u03b7\u03bd \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03ae<\/a> \u03ba\u03c1\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03c3\u03c5\u03c7\u03bd\u03ac \u03b1\u03ba\u03c1\u03b9\u03b2\u03ce\u03c2 \u03c3\u03b5 \u03b1\u03c5\u03c4\u03ac \u03c4\u03b1 operational \u03c3\u03c4\u03c1\u03ce\u03bc\u03b1\u03c4\u03b1.<\/p>\n<h2 id=\"osworld-apotelesmata\">\u03a4\u03b9 \u03ad\u03b4\u03b5\u03b9\u03be\u03b1\u03bd \u03bf\u03b9 222 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b5\u03c2 \u03c4\u03bf\u03c5 OSWorld<\/h2>\n<p>\u0397 \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7 \u03ad\u03b3\u03b9\u03bd\u03b5 \u03c3\u03b5 222 \u03b1\u03c0\u03cc \u03c4\u03b9\u03c2 369 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b5\u03c2 \u03c4\u03bf\u03c5 \u03b1\u03c1\u03c7\u03b9\u03ba\u03bf\u03cd OSWorld. \u0391\u03c6\u03b1\u03b9\u03c1\u03ad\u03b8\u03b7\u03ba\u03b1\u03bd Chrome \u03ba\u03b1\u03b9 multi-app tasks \u03b5\u03c0\u03b5\u03b9\u03b4\u03ae \u03c4\u03bf \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03bf \u03c0\u03b5\u03c1\u03b9\u03b2\u03ac\u03bb\u03bb\u03bf\u03bd Baidu RLE \u03b5\u03af\u03c7\u03b5 \u03b1\u03c3\u03c4\u03b1\u03b8\u03ad\u03c2 \u03b4\u03af\u03ba\u03c4\u03c5\u03bf \u03ba\u03b1\u03b9 \u03b5\u03bb\u03bb\u03b9\u03c0\u03ae \u03b5\u03be\u03c9\u03c4\u03b5\u03c1\u03b9\u03ba\u03ac assets. \u039f\u03b9 \u03c5\u03c0\u03cc\u03bb\u03bf\u03b9\u03c0\u03b5\u03c2 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b5\u03c2 \u03ba\u03ac\u03bb\u03c5\u03c8\u03b1\u03bd Gimp, LibreOffice Calc, Impress \u03ba\u03b1\u03b9 Writer, \u03bb\u03b5\u03b9\u03c4\u03bf\u03c5\u03c1\u03b3\u03b9\u03ba\u03cc \u03c3\u03cd\u03c3\u03c4\u03b7\u03bc\u03b1, Thunderbird, VLC \u03ba\u03b1\u03b9 VS Code.<\/p>\n<p>\u03a7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03ae\u03b8\u03b7\u03ba\u03b1\u03bd Qwen3-VL-Thinking \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03b1 4B \u03ba\u03b1\u03b9 8B, \u03ad\u03c9\u03c2 50 \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1 \u03b1\u03bd\u03ac trajectory, \u03ad\u03be\u03b9 tasks \u03b1\u03bd\u03ac iteration \u03ba\u03b1\u03b9 \u03bf\u03ba\u03c4\u03ce trajectories \u03b1\u03bd\u03ac task. \u0397 \u03b5\u03ba\u03c0\u03b1\u03af\u03b4\u03b5\u03c5\u03c3\u03b7 \u03ad\u03c4\u03c1\u03b5\u03be\u03b5 \u03c3\u03b5 \u03bf\u03ba\u03c4\u03ce H100 \u03b3\u03b9\u03b1 \u03ad\u03c9\u03c2 50 iterations \u03ba\u03b1\u03b9 \u03b1\u03bd\u03b1\u03c6\u03ad\u03c1\u03b8\u03b7\u03ba\u03b5 \u03c4\u03bf \u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf checkpoint \u03c3\u03cd\u03bc\u03c6\u03c9\u03bd\u03b1 \u03bc\u03b5 \u03c4\u03b7\u03bd \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7 \u03ba\u03b1\u03c4\u03ac \u03c4\u03b7 \u03b4\u03b9\u03ac\u03c1\u03ba\u03b5\u03b9\u03b1 \u03c4\u03b7\u03c2 \u03b5\u03ba\u03c0\u03b1\u03af\u03b4\u03b5\u03c5\u03c3\u03b7\u03c2.<\/p>\n<div class=\"td-chart td-chart--metrics\">\n<div class=\"td-chart-head\">\n<p class=\"td-chart-title\">\u03a4\u03ad\u03c3\u03c3\u03b5\u03c1\u03b1 \u03c4\u03b5\u03ba\u03bc\u03b7\u03c1\u03b9\u03c9\u03bc\u03ad\u03bd\u03b1 \u03bc\u03b5\u03b3\u03ad\u03b8\u03b7 \u03c4\u03b7\u03c2 \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7\u03c2<\/p>\n<p class=\"td-chart-subtitle\">\u039f\u03b9 \u03c4\u03b9\u03bc\u03ad\u03c2 \u03b1\u03c6\u03bf\u03c1\u03bf\u03cd\u03bd \u03c4\u03bf \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03bf \u03c6\u03b9\u03bb\u03c4\u03c1\u03b1\u03c1\u03b9\u03c3\u03bc\u03ad\u03bd\u03bf OSWorld setup \u03ba\u03b1\u03b9 \u03c4\u03b1 \u03ba\u03bf\u03b9\u03bd\u03ac outcomes \u03c4\u03c9\u03bd \u03c3\u03c5\u03b3\u03ba\u03c1\u03af\u03c3\u03b5\u03c9\u03bd \u03bc\u03ae\u03ba\u03bf\u03c5\u03c2, \u03cc\u03c7\u03b9 \u03ba\u03ac\u03b8\u03b5 GUI agent \u03ae \u03ba\u03ac\u03b8\u03b5 \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03b9\u03ba\u03cc \u03c0\u03b5\u03c1\u03b9\u03b2\u03ac\u03bb\u03bb\u03bf\u03bd.<\/p>\n<\/div>\n<div class=\"td-chart-body\">\n<div class=\"td-metric-grid\">\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">222<\/span><span class=\"td-metric-label\">\u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b5\u03c2 \u03c3\u03b5 \u03bf\u03ba\u03c4\u03ce application domains \u03bc\u03b5\u03c4\u03ac \u03c4\u03bf \u03c6\u03b9\u03bb\u03c4\u03c1\u03ac\u03c1\u03b9\u03c3\u03bc\u03b1<\/span><\/div>\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">50,0%<\/span><span class=\"td-metric-label\">overall success \u03c4\u03bf\u03c5 LACL-GUI-8B \u03c3\u03c4\u03bf \u03b5\u03be\u03b5\u03c4\u03b1\u03b6\u03cc\u03bc\u03b5\u03bd\u03bf subset<\/span><\/div>\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">8,19<\/span><span class=\"td-metric-label\">\u03bc\u03ad\u03c3\u03b1 \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1 \u03c3\u03b5 70 tasks \u03c0\u03bf\u03c5 \u03c0\u03ad\u03c4\u03c5\u03c7\u03b1\u03bd \u03ba\u03b1\u03b9 \u03bf\u03b9 \u03c4\u03c1\u03b5\u03b9\u03c2 \u03bc\u03ad\u03b8\u03bf\u03b4\u03bf\u03b9<\/span><\/div>\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">26,72<\/span><span class=\"td-metric-label\">\u03bc\u03ad\u03c3\u03b1 \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1 \u03c3\u03b5 93 tasks \u03cc\u03c0\u03bf\u03c5 \u03ba\u03b1\u03b9 \u03bf\u03b9 \u03c4\u03c1\u03b5\u03b9\u03c2 \u03bc\u03ad\u03b8\u03bf\u03b4\u03bf\u03b9 \u03b1\u03c0\u03ad\u03c4\u03c5\u03c7\u03b1\u03bd<\/span><\/div>\n<\/div>\n<\/div>\n<\/div>\n<p>\u03a4\u03bf LACL-GUI-8B \u03ba\u03b1\u03c4\u03ad\u03b3\u03c1\u03b1\u03c8\u03b5 50,0% \u03c3\u03c5\u03bd\u03bf\u03bb\u03b9\u03ba\u03ae \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b1, \u03ad\u03bd\u03b1\u03bd\u03c4\u03b9 47,2% \u03b3\u03b9\u03b1 GRPO, 47,3% \u03b3\u03b9\u03b1 REAL \u03ba\u03b1\u03b9 42,7% \u03b3\u03b9\u03b1 \u03c4\u03bf Qwen3-VL-8B-Thinking backbone. \u03a3\u03c4\u03bf 4B scale \u03ad\u03c6\u03c4\u03b1\u03c3\u03b5 44,5%, \u03ad\u03bd\u03b1\u03bd\u03c4\u03b9 42,2% \u03b3\u03b9\u03b1 GRPO \u03ba\u03b1\u03b9 43,0% \u03b3\u03b9\u03b1 REAL. \u03a0\u03c1\u03cc\u03ba\u03b5\u03b9\u03c4\u03b1\u03b9 \u03b3\u03b9\u03b1 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03ad\u03c2 2,8 \u03ba\u03b1\u03b9 2,7 \u03c0\u03bf\u03c3\u03bf\u03c3\u03c4\u03b9\u03b1\u03af\u03c9\u03bd \u03bc\u03bf\u03bd\u03ac\u03b4\u03c9\u03bd \u03c3\u03c4\u03bf 8B \u03cc\u03c4\u03b1\u03bd \u03b7 \u03c3\u03cd\u03b3\u03ba\u03c1\u03b9\u03c3\u03b7 \u03bc\u03ad\u03bd\u03b5\u03b9 \u03c3\u03c4\u03bf \u03af\u03b4\u03b9\u03bf backbone.<\/p>\n<p>\u03a4\u03bf incremental ablation \u03b1\u03bd\u03ad\u03b2\u03b1\u03c3\u03b5 \u03c4\u03bf 8B \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1 \u03b1\u03c0\u03cc 42,7% \u03c3\u03b5 45,4% \u03bc\u03b5 decoupled architecture, 46,5% \u03bc\u03b5 Success Buffer, 47,2% \u03bc\u03b5 trajectory filter \u03ba\u03b1\u03b9 50,0% \u03bc\u03b5 \u03c4\u03b7\u03bd LACL-GUI objective. \u0386\u03c1\u03b1 \u03c4\u03bf \u03c4\u03b5\u03bb\u03b9\u03ba\u03cc \u03ba\u03ad\u03c1\u03b4\u03bf\u03c2 \u03b4\u03b5\u03bd \u03b1\u03c0\u03bf\u03b4\u03af\u03b4\u03b5\u03c4\u03b1\u03b9 \u03c3\u03b5 \u03ad\u03bd\u03b1 \u03bc\u03cc\u03bd\u03bf \u03c3\u03c4\u03bf\u03b9\u03c7\u03b5\u03af\u03bf\u00b7 \u03b7 objective \u03c0\u03c1\u03bf\u03c3\u03c4\u03ad\u03b8\u03b7\u03ba\u03b5 \u03c0\u03ac\u03bd\u03c9 \u03c3\u03b5 \u03ae\u03b4\u03b7 \u03b2\u03b5\u03bb\u03c4\u03b9\u03c9\u03bc\u03ad\u03bd\u03b7 \u03c5\u03c0\u03bf\u03b4\u03bf\u03bc\u03ae.<\/p>\n<h2 id=\"ligotera-vimata\">\u039b\u03b9\u03b3\u03cc\u03c4\u03b5\u03c1\u03b1 \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1, \u03b1\u03bb\u03bb\u03ac \u03cc\u03c7\u03b9 \u03c0\u03b1\u03bd\u03c4\u03bf\u03cd<\/h2>\n<p>\u0393\u03b9\u03b1 \u03bd\u03b1 \u03bc\u03b7 \u03c3\u03c5\u03b3\u03c7\u03ad\u03b5\u03c4\u03b1\u03b9 \u03b7 \u03b1\u03c0\u03bf\u03b4\u03bf\u03c4\u03b9\u03ba\u03cc\u03c4\u03b7\u03c4\u03b1 \u03bc\u03b5 \u03c4\u03b7 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03ae \u03c0\u03b9\u03b8\u03b1\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b1\u03c2, \u03bf\u03b9 \u03c3\u03c5\u03b3\u03b3\u03c1\u03b1\u03c6\u03b5\u03af\u03c2 \u03c3\u03c5\u03bd\u03ad\u03ba\u03c1\u03b9\u03bd\u03b1\u03bd \u03bc\u03ae\u03ba\u03bf\u03c2 \u03bc\u03cc\u03bd\u03bf \u03b5\u03ba\u03b5\u03af \u03cc\u03c0\u03bf\u03c5 base model, REAL \u03ba\u03b1\u03b9 LACL-GUI \u03b5\u03af\u03c7\u03b1\u03bd \u03c4\u03bf \u03af\u03b4\u03b9\u03bf outcome. \u03a3\u03b5 70 \u03ba\u03bf\u03b9\u03bd\u03ad\u03c2 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b5\u03c2, \u03b7 LACL-GUI \u03b5\u03af\u03c7\u03b5 8,19 \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1 \u03ba\u03b1\u03c4\u03ac \u03bc\u03ad\u03c3\u03bf \u03cc\u03c1\u03bf \u03ad\u03bd\u03b1\u03bd\u03c4\u03b9 9,13 \u03c4\u03b7\u03c2 REAL, \u03bc\u03b5 \u03bc\u03b5\u03af\u03c9\u03c3\u03b7 10,3%. \u03a3\u03b5 93 \u03ba\u03bf\u03b9\u03bd\u03ad\u03c2 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b5\u03c2, \u03bf\u03b9 \u03b1\u03bd\u03c4\u03af\u03c3\u03c4\u03bf\u03b9\u03c7\u03b5\u03c2 \u03c4\u03b9\u03bc\u03ad\u03c2 \u03ae\u03c4\u03b1\u03bd 26,72 \u03ba\u03b1\u03b9 29,87 \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1, \u03bc\u03b5 \u03bc\u03b5\u03af\u03c9\u03c3\u03b7 10,5%.<\/p>\n<p>\u0397 \u03b5\u03b9\u03ba\u03cc\u03bd\u03b1 \u03b4\u03b5\u03bd \u03ae\u03c4\u03b1\u03bd \u03bf\u03bc\u03bf\u03b9\u03cc\u03bc\u03bf\u03c1\u03c6\u03b7. \u03a3\u03c4\u03b9\u03c2 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b5\u03c2, \u03c4\u03bf Writer \u03c0\u03b1\u03c1\u03bf\u03c5\u03c3\u03af\u03b1\u03c3\u03b5 \u03bc\u03b9\u03ba\u03c1\u03ae \u03b1\u03cd\u03be\u03b7\u03c3\u03b7 \u03ba\u03b1\u03b9 \u03c4\u03bf VLC \u03bc\u03b5\u03b3\u03ac\u03bb\u03b7 \u03b1\u03cd\u03be\u03b7\u03c3\u03b7 \u03bc\u03ae\u03ba\u03bf\u03c5\u03c2 \u03ad\u03bd\u03b1\u03bd\u03c4\u03b9 REAL. \u03a3\u03c4\u03b9\u03c2 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b5\u03c2, Writer \u03ba\u03b1\u03b9 VS Code \u03b5\u03af\u03c7\u03b1\u03bd \u03b5\u03c0\u03af\u03c3\u03b7\u03c2 \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b5\u03c2 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ad\u03c2. \u039f\u03b9 \u03b5\u03be\u03b1\u03b9\u03c1\u03ad\u03c3\u03b5\u03b9\u03c2 \u03b5\u03bc\u03c0\u03bf\u03b4\u03af\u03b6\u03bf\u03c5\u03bd \u03bc\u03b9\u03b1 \u03c5\u03c0\u03b5\u03c1\u03b2\u03bf\u03bb\u03b9\u03ba\u03ae \u03c5\u03c0\u03cc\u03c3\u03c7\u03b5\u03c3\u03b7 \u03cc\u03c4\u03b9 \u03ba\u03ac\u03b8\u03b5 \u03b5\u03c6\u03b1\u03c1\u03bc\u03bf\u03b3\u03ae \u03b3\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03c3\u03c5\u03bd\u03c4\u03bf\u03bc\u03cc\u03c4\u03b5\u03c1\u03b7.<\/p>\n<aside class=\"td-article-note\"><strong>\u03a3\u03c9\u03c3\u03c4\u03ae \u03b1\u03bd\u03ac\u03b3\u03bd\u03c9\u03c3\u03b7 \u03c4\u03c9\u03bd \u03bc\u03ad\u03c3\u03c9\u03bd \u03cc\u03c1\u03c9\u03bd:<\/strong> \u03c4\u03b1 8,19 \u03ba\u03b1\u03b9 26,72 \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1 \u03c0\u03c1\u03bf\u03ad\u03c1\u03c7\u03bf\u03bd\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03ac \u03c3\u03cd\u03bd\u03bf\u03bb\u03b1 tasks \u03ba\u03b1\u03b9 outcomes. \u0394\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03b4\u03cd\u03bf \u03c3\u03c4\u03ac\u03b4\u03b9\u03b1 \u03c4\u03b7\u03c2 \u03af\u03b4\u03b9\u03b1\u03c2 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ae\u03c2 \u03bf\u03cd\u03c4\u03b5 \u03b3\u03b5\u03bd\u03b9\u03ba\u03ac SLA \u03b3\u03b9\u03b1 computer-use agents.<\/aside>\n<p>\u0391\u03c5\u03c4\u03cc\u03c2 \u03b5\u03af\u03bd\u03b1\u03b9 \u03ba\u03b1\u03b9 \u03bf \u03bb\u03cc\u03b3\u03bf\u03c2 \u03c0\u03bf\u03c5 \u03ad\u03bd\u03b1 <a href=\"https:\/\/twodots.gr\/ai-benchmark-harness-allazei-nikiti\/\">AI benchmark \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b1\u03bb\u03bb\u03ac\u03be\u03b5\u03b9 \u03c4\u03bf\u03bd \u03bd\u03b9\u03ba\u03b7\u03c4\u03ae<\/a>. \u03a4\u03bf \u03c4\u03b5\u03bb\u03b9\u03ba\u03cc score \u03b5\u03be\u03b1\u03c1\u03c4\u03ac\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03c4\u03bf subset, \u03c4\u03bf\u03bd verifier, \u03c4\u03b1 environment failures, \u03c4\u03bf \u03b5\u03c0\u03b9\u03c4\u03c1\u03b5\u03c0\u03cc\u03bc\u03b5\u03bd\u03bf horizon \u03ba\u03b1\u03b9 \u03c4\u03bf\u03bd \u03c4\u03c1\u03cc\u03c0\u03bf \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae\u03c2 checkpoint.<\/p>\n<h2 id=\"gimp-paradeigma\">\u03a4\u03bf \u03c0\u03b1\u03c1\u03ac\u03b4\u03b5\u03b9\u03b3\u03bc\u03b1 \u03c4\u03bf\u03c5 Gimp \u03ba\u03ac\u03bd\u03b5\u03b9 \u03c4\u03b7 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03ac \u03bf\u03c1\u03b1\u03c4\u03ae<\/h2>\n<p>\u039f\u03b9 \u03c3\u03c5\u03b3\u03b3\u03c1\u03b1\u03c6\u03b5\u03af\u03c2 \u03b1\u03bd\u03b1\u03bb\u03cd\u03bf\u03c5\u03bd task \u03b1\u03bb\u03bb\u03b1\u03b3\u03ae\u03c2 \u03c4\u03bf\u03c5 theme \u03c4\u03bf\u03c5 Gimp \u03b1\u03c0\u03cc dark \u03c3\u03b5 light. \u0397 LACL-GUI \u03bf\u03bb\u03bf\u03ba\u03bb\u03b7\u03c1\u03ce\u03bd\u03b5\u03b9 \u03c4\u03b7\u03bd \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 \u03c3\u03b5 \u03ad\u03be\u03b9 \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1: \u03b1\u03bd\u03bf\u03af\u03b3\u03b5\u03b9 \u03c4\u03bf Edit, \u03bc\u03c0\u03b1\u03af\u03bd\u03b5\u03b9 \u03c3\u03c4\u03b1 Preferences, \u03b5\u03c0\u03b9\u03bb\u03ad\u03b3\u03b5\u03b9 Theme, \u03b1\u03bb\u03bb\u03ac\u03b6\u03b5\u03b9 \u03c3\u03b5 Light \u03ba\u03b1\u03b9 \u03b5\u03c0\u03b9\u03b2\u03b5\u03b2\u03b1\u03b9\u03ce\u03bd\u03b5\u03b9. \u039c\u03b9\u03b1 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b7 REAL trajectory \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 27 \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1.<\/p>\n<p>\u0397 \u03bc\u03b1\u03ba\u03c1\u03cd\u03c4\u03b5\u03c1\u03b7 \u03b5\u03ba\u03c4\u03ad\u03bb\u03b5\u03c3\u03b7 \u03c0\u03b5\u03c1\u03b9\u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03b5\u03b9 \u03b5\u03c0\u03b1\u03bd\u03b1\u03bb\u03b1\u03bc\u03b2\u03b1\u03bd\u03cc\u03bc\u03b5\u03bd\u03b5\u03c2 \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b5\u03c2 \u03c7\u03c9\u03c1\u03af\u03c2 \u03c0\u03c1\u03cc\u03bf\u03b4\u03bf \u03ba\u03b1\u03b9, \u03cc\u03c4\u03b1\u03bd \u03b1\u03bb\u03bb\u03ac\u03b6\u03b5\u03b9 \u03c4\u03bf UI state, \u03c3\u03c5\u03bd\u03b5\u03c7\u03af\u03b6\u03b5\u03b9 \u03bd\u03b1 \u03ba\u03ac\u03bd\u03b5\u03b9 \u03ba\u03bb\u03b9\u03ba \u03c3\u03b5 \u03c0\u03b1\u03bb\u03b9\u03ac \u03b8\u03ad\u03c3\u03b7. \u039a\u03b1\u03b9 \u03bf\u03b9 \u03b4\u03cd\u03bf trajectories \u03c0\u03b1\u03af\u03c1\u03bd\u03bf\u03c5\u03bd reward 1, \u03b1\u03bb\u03bb\u03ac \u03bc\u03cc\u03bd\u03bf \u03b7 \u03c0\u03c1\u03ce\u03c4\u03b7 \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03b5\u03af \u03ba\u03b1\u03b8\u03b1\u03c1\u03cc \u03c0\u03c1\u03cc\u03c4\u03c5\u03c0\u03bf efficiency.<\/p>\n<p>\u03a3\u03c4\u03b9\u03c2 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b5\u03c2, \u03b4\u03cd\u03bf trajectories \u03b1\u03c0\u03bf\u03ba\u03bb\u03af\u03bd\u03bf\u03c5\u03bd \u03c3\u03c4\u03bf \u03af\u03b4\u03b9\u03bf \u03ba\u03c1\u03af\u03c3\u03b9\u03bc\u03bf \u03c3\u03b7\u03bc\u03b5\u03af\u03bf. \u0397 \u03bc\u03af\u03b1 \u03c3\u03c4\u03b1\u03bc\u03b1\u03c4\u03ac \u03c0\u03c1\u03cc\u03c9\u03c1\u03b1 \u03b5\u03c0\u03b5\u03b9\u03b4\u03ae \u03b8\u03b5\u03c9\u03c1\u03b5\u03af \u03bb\u03b1\u03bd\u03b8\u03b1\u03c3\u03bc\u03ad\u03bd\u03b1 \u03cc\u03c4\u03b9 \u03bf\u03bb\u03bf\u03ba\u03bb\u03ae\u03c1\u03c9\u03c3\u03b5 \u03c4\u03b7 \u03b4\u03bf\u03c5\u03bb\u03b5\u03b9\u03ac\u00b7 \u03b7 \u03ac\u03bb\u03bb\u03b7 \u03c3\u03c5\u03bd\u03b5\u03c7\u03af\u03b6\u03b5\u03b9 \u03bc\u03b5 \u03ac\u03ba\u03b1\u03c1\u03c0\u03b7 \u03b5\u03be\u03b5\u03c1\u03b5\u03cd\u03bd\u03b7\u03c3\u03b7. \u03a4\u03bf \u03b4\u03c5\u03b1\u03b4\u03b9\u03ba\u03cc reward 0 \u03b4\u03b5\u03bd \u03b1\u03c0\u03bf\u03c4\u03c5\u03c0\u03ce\u03bd\u03b5\u03b9 \u03c0\u03cc\u03c3\u03bf \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03ae \u03b5\u03af\u03bd\u03b1\u03b9 \u03b7 \u03c0\u03b9\u03b8\u03b1\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b1\u03bd\u03ac\u03ba\u03b1\u03bc\u03c8\u03b7\u03c2, \u03c4\u03bf \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2 \u03ba\u03b1\u03b9 \u03b7 \u03b1\u03b9\u03c4\u03af\u03b1 \u03c4\u03bf\u03c5 \u03bb\u03ac\u03b8\u03bf\u03c5\u03c2.<\/p>\n<h2 id=\"production-metafora\">\u0391\u03c0\u03cc \u03c4\u03bf paper \u03c3\u03b5 production GUI agents<\/h2>\n<p>\u0397 \u03ac\u03bc\u03b5\u03c3\u03b7 \u03c3\u03c5\u03bd\u03b5\u03b9\u03c3\u03c6\u03bf\u03c1\u03ac \u03c4\u03b7\u03c2 LACL-GUI \u03b1\u03c6\u03bf\u03c1\u03ac training \u03c3\u03b5 benchmark \u03ba\u03b1\u03b9 \u03cc\u03c7\u03b9 \u03ad\u03c4\u03bf\u03b9\u03bc\u03b7 \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03b7\u03bc\u03b1\u03c4\u03b9\u03ba\u03ae \u03c0\u03bb\u03b1\u03c4\u03c6\u03cc\u03c1\u03bc\u03b1. \u03a9\u03c3\u03c4\u03cc\u03c3\u03bf, \u03b7 \u03b1\u03c1\u03c7\u03ae \u03bc\u03b5\u03c4\u03b1\u03c6\u03ad\u03c1\u03b5\u03c4\u03b1\u03b9 \u03c7\u03c1\u03ae\u03c3\u03b9\u03bc\u03b1 \u03c3\u03b5 agents \u03c0\u03bf\u03c5 \u03b5\u03bd\u03b7\u03bc\u03b5\u03c1\u03ce\u03bd\u03bf\u03c5\u03bd \u03c0\u03c1\u03bf\u03ca\u03cc\u03bd\u03c4\u03b1, \u03b5\u03be\u03ac\u03b3\u03bf\u03c5\u03bd reports, \u03bf\u03c1\u03b3\u03b1\u03bd\u03ce\u03bd\u03bf\u03c5\u03bd \u03ba\u03b1\u03bc\u03c0\u03ac\u03bd\u03b9\u03b5\u03c2, \u03c7\u03b5\u03b9\u03c1\u03af\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 tickets \u03ae \u03c3\u03c5\u03bc\u03c0\u03bb\u03b7\u03c1\u03ce\u03bd\u03bf\u03c5\u03bd back-office \u03c3\u03c5\u03c3\u03c4\u03ae\u03bc\u03b1\u03c4\u03b1.<\/p>\n<p>\u0397 \u03bf\u03bc\u03ac\u03b4\u03b1 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03bd\u03b1 \u03ba\u03b1\u03c4\u03b1\u03b3\u03c1\u03ac\u03c6\u03b5\u03b9 \u03cc\u03c7\u03b9 \u03bc\u03cc\u03bd\u03bf completion \u03b1\u03bb\u03bb\u03ac \u03ba\u03b1\u03b9 \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1, retries, \u03b5\u03c0\u03b9\u03c3\u03c4\u03c1\u03bf\u03c6\u03ad\u03c2 \u03c3\u03b5 \u03c0\u03c1\u03bf\u03b7\u03b3\u03bf\u03cd\u03bc\u03b5\u03bd\u03b7 \u03bf\u03b8\u03cc\u03bd\u03b7, \u03c7\u03c1\u03cc\u03bd\u03bf \u03b1\u03bd\u03ac \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b1, \u03c0\u03c1\u03ce\u03c4\u03b7 \u03b1\u03c0\u03cc\u03ba\u03bb\u03b9\u03c3\u03b7, recovery \u03ba\u03b1\u03b9 \u03c4\u03b5\u03bb\u03b9\u03ba\u03ae verification. \u0397 \u03b1\u03bd\u03ac\u03bb\u03c5\u03c3\u03b7 \u03b3\u03b9\u03b1 \u03c4\u03bf\u03c5\u03c2 <a href=\"https:\/\/twodots.gr\/multimodal-ai-agents-antilipsi-drasi\/\">multimodal AI agents \u03ba\u03b1\u03b9 \u03c4\u03bf \u03c7\u03ac\u03c3\u03bc\u03b1 \u03b1\u03bd\u03ac\u03bc\u03b5\u03c3\u03b1 \u03c3\u03c4\u03b7\u03bd \u03b1\u03bd\u03c4\u03af\u03bb\u03b7\u03c8\u03b7 \u03ba\u03b1\u03b9 \u03c4\u03b7 \u03c3\u03c9\u03c3\u03c4\u03ae \u03b4\u03c1\u03ac\u03c3\u03b7<\/a> \u03b5\u03be\u03b7\u03b3\u03b5\u03af \u03b3\u03b9\u03b1\u03c4\u03af \u03ad\u03bd\u03b1 \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u00ab\u03b2\u03bb\u03ad\u03c0\u03b5\u03b9\u00bb \u03c4\u03bf interface \u03c7\u03c9\u03c1\u03af\u03c2 \u03bd\u03b1 \u03b5\u03bb\u03ad\u03b3\u03c7\u03b5\u03b9 \u03b1\u03be\u03b9\u03cc\u03c0\u03b9\u03c3\u03c4\u03b1 \u03c4\u03b7\u03bd \u03b1\u03ba\u03bf\u03bb\u03bf\u03c5\u03b8\u03af\u03b1.<\/p>\n<p>\u0394\u03b5\u03bd \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b1\u03bd\u03c4\u03b1\u03bc\u03b5\u03af\u03b2\u03b5\u03c4\u03b1\u03b9 \u03bc\u03b7\u03c7\u03b1\u03bd\u03b9\u03ba\u03ac \u03b7 \u03c3\u03c5\u03bd\u03c4\u03bf\u03bc\u03af\u03b1. \u0388\u03bd\u03b1\u03c2 agent \u03c0\u03bf\u03c5 \u03c0\u03b1\u03c1\u03b1\u03bb\u03b5\u03af\u03c0\u03b5\u03b9 preview, \u03b4\u03b5\u03cd\u03c4\u03b5\u03c1\u03bf \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf \u03c4\u03b9\u03bc\u03ae\u03c2 \u03ae confirmation \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03bc\u03b1\u03b6\u03b9\u03ba\u03ae \u03b1\u03bb\u03bb\u03b1\u03b3\u03ae \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03c6\u03b1\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03b3\u03c1\u03ae\u03b3\u03bf\u03c1\u03bf\u03c2 \u03ba\u03b1\u03b9 \u03bd\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03b5\u03c0\u03b9\u03ba\u03af\u03bd\u03b4\u03c5\u03bd\u03bf\u03c2. \u0397 \u03c3\u03c9\u03c3\u03c4\u03ae \u03c0\u03c1\u03bf\u03c4\u03af\u03bc\u03b7\u03c3\u03b7 \u03b5\u03af\u03bd\u03b1\u03b9 \u03b7 \u03c3\u03c5\u03bd\u03c4\u03bf\u03bc\u03cc\u03c4\u03b5\u03c1\u03b7 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ae \u03c0\u03bf\u03c5 \u03b9\u03ba\u03b1\u03bd\u03bf\u03c0\u03bf\u03b9\u03b5\u03af \u03cc\u03bb\u03b1 \u03c4\u03b1 business \u03ba\u03b1\u03b9 safety invariants.<\/p>\n<div class=\"td-decision-band\">\n<p class=\"td-decision-kicker\">\u039a\u03b1\u03bd\u03cc\u03bd\u03b1\u03c2 \u03b1\u03c0\u03cc\u03c6\u03b1\u03c3\u03b7\u03c2 \u03b3\u03b9\u03b1 GUI automation<\/p>\n<p class=\"td-decision-title\">\u039c\u03b7\u03bd \u03ba\u03ac\u03bd\u03b5\u03c4\u03b5 rollout \u03bc\u03cc\u03bd\u03bf \u03b5\u03c0\u03b5\u03b9\u03b4\u03ae \u03b1\u03bd\u03ad\u03b2\u03b7\u03ba\u03b5 \u03c4\u03bf completion rate<\/p>\n<p>Go \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u03b5\u03c0\u03b1\u03bb\u03b7\u03b8\u03b5\u03c5\u03bc\u03ad\u03bd\u03bf \u03c4\u03b5\u03bb\u03b9\u03ba\u03cc state, \u03b1\u03c0\u03bf\u03b4\u03b5\u03ba\u03c4\u03cc \u03bc\u03ae\u03ba\u03bf\u03c2, \u03c7\u03b1\u03bc\u03b7\u03bb\u03cc retry rate, traceable divergence, \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03bf recovery, \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2 \u03bc\u03ad\u03c3\u03b1 \u03c3\u03c4\u03bf budget \u03ba\u03b1\u03b9 \u03c3\u03b1\u03c6\u03ae human-handoff gates. No-go \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b1 \u03c7\u03c9\u03c1\u03af\u03c2 audit trail, \u03c3\u03c5\u03bd\u03c4\u03cc\u03bc\u03b5\u03c5\u03c3\u03b7 \u03c0\u03bf\u03c5 \u03c0\u03b1\u03c1\u03b1\u03bb\u03b5\u03af\u03c0\u03b5\u03b9 \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf \u03ae benchmark \u03c0\u03bf\u03c5 \u03b5\u03be\u03b1\u03b9\u03c1\u03b5\u03af \u03b1\u03ba\u03c1\u03b9\u03b2\u03ce\u03c2 \u03c4\u03b1 production workflows.<\/p>\n<\/div>\n<p>\u03a4\u03bf \u03af\u03b4\u03b9\u03bf \u03ba\u03c1\u03b9\u03c4\u03ae\u03c1\u03b9\u03bf \u03b5\u03bc\u03c6\u03b1\u03bd\u03af\u03b6\u03b5\u03c4\u03b1\u03b9 \u03c3\u03c4\u03bf <a href=\"https:\/\/twodots.gr\/k-bench-ai-agents-epistimoniki-akriveia-artifacts\/\">K-Bench \u03b3\u03b9\u03b1 AI agents \u03ba\u03b1\u03b9 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac artifacts<\/a>: \u03bc\u03b9\u03b1 \u03c0\u03b5\u03b9\u03c3\u03c4\u03b9\u03ba\u03ae \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7 \u03ae \u03ad\u03bd\u03b1 success flag \u03b4\u03b5\u03bd \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03b5\u03af \u03b1\u03c0\u03cc\u03b4\u03b5\u03b9\u03be\u03b7 \u03cc\u03c4\u03b9 \u03c0\u03b1\u03c1\u03b1\u03b4\u03cc\u03b8\u03b7\u03ba\u03b5 \u03c3\u03c9\u03c3\u03c4\u03ac \u03bf\u03bb\u03cc\u03ba\u03bb\u03b7\u03c1\u03b7 \u03b7 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1.<\/p>\n<h2 id=\"elegxoi-ylopoiisis\">\u0395\u03c0\u03c4\u03ac \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf\u03b9 \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03b9\u03ba\u03cc rollout<\/h2>\n<div class=\"td-step-list\">\n<p class=\"td-step-list-title\">\u0391\u03c0\u03cc \u03c4\u03bf binary outcome \u03c3\u03b5 \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03b7 trajectory quality<\/p>\n<ol>\n<li><span class=\"td-step-kicker\">\u0421\u0442\u044a\u043f\u043a\u0430 1<\/span><strong>\u039f\u03c1\u03af\u03c3\u03c4\u03b5 \u03c4\u03bf \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03cc \u03c4\u03b5\u03bb\u03b9\u03ba\u03cc state<\/strong>\n<p>\u03a7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03ae\u03c3\u03c4\u03b5 verifier \u03c0\u03bf\u03c5 \u03b5\u03bb\u03ad\u03b3\u03c7\u03b5\u03b9 \u03c4\u03bf \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1 \u03c3\u03c4\u03bf \u03c3\u03cd\u03c3\u03c4\u03b7\u03bc\u03b1 \u03ba\u03b1\u03b9 \u03cc\u03c7\u03b9 \u03bc\u03cc\u03bd\u03bf \u03c4\u03bf \u03c4\u03b5\u03bb\u03b5\u03c5\u03c4\u03b1\u03af\u03bf \u03bc\u03ae\u03bd\u03c5\u03bc\u03b1 \u03c4\u03bf\u03c5 GUI agent.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u0421\u0442\u044a\u043f\u043a\u0430 2<\/span><strong>\u039a\u03b1\u03c4\u03b1\u03b3\u03c1\u03ac\u03c8\u03c4\u03b5 \u03bf\u03bb\u03cc\u03ba\u03bb\u03b7\u03c1\u03b7 \u03c4\u03b7 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ae<\/strong>\n<p>\u039a\u03c1\u03b1\u03c4\u03ae\u03c3\u03c4\u03b5 versioned actions, observations, timestamps, policy \u03ad\u03ba\u03b4\u03bf\u03c3\u03b7 \u03ba\u03b1\u03b9 tool responses \u03bc\u03b5 \u03ba\u03b1\u03bd\u03cc\u03bd\u03b5\u03c2 \u03b5\u03bb\u03b1\u03c7\u03b9\u03c3\u03c4\u03bf\u03c0\u03bf\u03af\u03b7\u03c3\u03b7\u03c2 \u03ba\u03b1\u03b9 \u03b4\u03b9\u03b1\u03c4\u03ae\u03c1\u03b7\u03c3\u03b7\u03c2 \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03c9\u03bd.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u0421\u0442\u044a\u043f\u043a\u0430 3<\/span><strong>\u039c\u03b5\u03c4\u03c1\u03ae\u03c3\u03c4\u03b5 \u03b1\u03bd\u03b1\u03b3\u03ba\u03b1\u03af\u03b1 \u03ba\u03b1\u03b9 \u03c0\u03b5\u03c1\u03b9\u03c4\u03c4\u03ac \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1<\/strong>\n<p>\u039e\u03b5\u03c7\u03c9\u03c1\u03af\u03c3\u03c4\u03b5 \u03c4\u03b1 business-required checks \u03b1\u03c0\u03cc retries, loops, \u03b5\u03c0\u03b9\u03c3\u03c4\u03c1\u03bf\u03c6\u03ad\u03c2 \u03ba\u03b1\u03b9 \u03ba\u03bb\u03b9\u03ba \u03c0\u03bf\u03c5 \u03b4\u03b5\u03bd \u03b1\u03bb\u03bb\u03ac\u03b6\u03bf\u03c5\u03bd state.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u0421\u0442\u044a\u043f\u043a\u0430 4<\/span><strong>\u0395\u03bd\u03c4\u03bf\u03c0\u03af\u03c3\u03c4\u03b5 \u03c4\u03b7\u03bd \u03c0\u03c1\u03ce\u03c4\u03b7 \u03bf\u03c5\u03c3\u03b9\u03b1\u03c3\u03c4\u03b9\u03ba\u03ae \u03b1\u03c0\u03cc\u03ba\u03bb\u03b9\u03c3\u03b7<\/strong>\n<p>\u03a3\u03c5\u03bd\u03b4\u03c5\u03ac\u03c3\u03c4\u03b5 visual similarity \u03bc\u03b5 structured state \u03ba\u03b1\u03b9 event logs, \u03ce\u03c3\u03c4\u03b5 \u03ad\u03bd\u03b1 cosmetic UI change \u03bd\u03b1 \u03bc\u03b7 \u03b8\u03b5\u03c9\u03c1\u03b5\u03af\u03c4\u03b1\u03b9 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b1 \u03ba\u03b1\u03b9 \u03ad\u03bd\u03b1 \u03ba\u03c1\u03c5\u03c6\u03cc \u03bb\u03ac\u03b8\u03bf\u03c2 \u03bd\u03b1 \u03bc\u03b7 \u03b4\u03b9\u03b1\u03c6\u03b5\u03cd\u03b3\u03b5\u03b9.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u0421\u0442\u044a\u043f\u043a\u0430 5<\/span><strong>\u03a4\u03b1\u03be\u03b9\u03bd\u03bf\u03bc\u03ae\u03c3\u03c4\u03b5 recovery \u03ba\u03b1\u03b9 stop behavior<\/strong>\n<p>\u039c\u03b5\u03c4\u03c1\u03ae\u03c3\u03c4\u03b5 \u03b1\u03bd \u03bf agent \u03b1\u03bd\u03b1\u03b3\u03bd\u03c9\u03c1\u03af\u03b6\u03b5\u03b9 \u03c4\u03bf \u03bb\u03ac\u03b8\u03bf\u03c2, \u03b1\u03bd \u03b5\u03c0\u03b9\u03c3\u03c4\u03c1\u03ad\u03c6\u03b5\u03b9 \u03c3\u03b5 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ad\u03c2 checkpoint \u03ae \u03b1\u03bd \u03c3\u03c5\u03bd\u03b5\u03c7\u03af\u03b6\u03b5\u03b9 \u03ac\u03c3\u03ba\u03bf\u03c0\u03b1 \u03bc\u03ad\u03c7\u03c1\u03b9 \u03c4\u03bf step limit.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u0421\u0442\u044a\u043f\u043a\u0430 6<\/span><strong>\u039a\u03ac\u03bd\u03c4\u03b5 slice \u03b1\u03bd\u03ac \u03b5\u03c6\u03b1\u03c1\u03bc\u03bf\u03b3\u03ae \u03ba\u03b1\u03b9 workflow<\/strong>\n<p>\u039c\u03b7\u03bd \u03ba\u03c1\u03cd\u03b2\u03b5\u03c4\u03b5 Writer, VLC, checkout \u03ae CRM failures \u03bc\u03ad\u03c3\u03b1 \u03c3\u03b5 \u03c3\u03c5\u03bd\u03bf\u03bb\u03b9\u03ba\u03cc \u03bc\u03ad\u03c3\u03bf \u03cc\u03c1\u03bf\u00b7 \u03ba\u03ac\u03b8\u03b5 domain \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03b4\u03b9\u03ba\u03cc \u03c4\u03bf\u03c5 error budget.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u0421\u0442\u044a\u043f\u043a\u0430 7<\/span><strong>\u03a3\u03c5\u03bd\u03b4\u03ad\u03c3\u03c4\u03b5 rollout gates \u03bc\u03b5 \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03b7 \u03b5\u03c0\u03bf\u03c0\u03c4\u03b5\u03af\u03b1<\/strong>\n<p>\u039f\u03c1\u03af\u03c3\u03c4\u03b5 \u03c0\u03cc\u03c4\u03b5 \u03b5\u03c0\u03b9\u03c4\u03c1\u03ad\u03c0\u03b5\u03c4\u03b1\u03b9 \u03b1\u03c5\u03c4\u03cc\u03bc\u03b1\u03c4\u03b7 \u03bf\u03bb\u03bf\u03ba\u03bb\u03ae\u03c1\u03c9\u03c3\u03b7, \u03c0\u03cc\u03c4\u03b5 \u03b1\u03c0\u03b1\u03b9\u03c4\u03b5\u03af\u03c4\u03b1\u03b9 confirmation \u03ba\u03b1\u03b9 \u03c0\u03cc\u03c4\u03b5 \u03b7 trajectory \u03c3\u03c4\u03b1\u03bc\u03b1\u03c4\u03ac \u03ba\u03b1\u03b9 \u03c0\u03b1\u03c1\u03b1\u03b4\u03af\u03b4\u03b5\u03c4\u03b1\u03b9 \u03c3\u03b5 \u03ac\u03bd\u03b8\u03c1\u03c9\u03c0\u03bf.<\/p>\n<\/li>\n<\/ol>\n<\/div>\n<p>\u0397 \u03b4\u03b9\u03b1\u03b4\u03b9\u03ba\u03b1\u03c3\u03af\u03b1 \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b5\u03c0\u03b1\u03bd\u03b1\u03bb\u03b7\u03c6\u03b8\u03b5\u03af \u03c3\u03b5 production-like \u03c0\u03b5\u03c1\u03b9\u03b2\u03ac\u03bb\u03bb\u03bf\u03bd \u03ba\u03b1\u03b9 \u03cc\u03c7\u03b9 \u03bc\u03cc\u03bd\u03bf \u03c3\u03c4\u03bf lab harness. \u0393\u03b9\u03b1 high-impact flows, \u03ba\u03ac\u03b8\u03b5 policy \u03ae UI \u03b1\u03bb\u03bb\u03b1\u03b3\u03ae \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 regression set \u03c0\u03bf\u03c5 \u03c0\u03b5\u03c1\u03b9\u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03b5\u03b9 \u03ba\u03bf\u03bd\u03c4\u03b9\u03bd\u03ad\u03c2 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b5\u03c2, \u03c0\u03c1\u03ce\u03b9\u03bc\u03b5\u03c2 \u03b1\u03c0\u03bf\u03ba\u03bb\u03af\u03c3\u03b5\u03b9\u03c2, loops \u03ba\u03b1\u03b9 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ae handoffs. \u0397 NIST AI RMF \u03bf\u03c1\u03b3\u03ac\u03bd\u03c9\u03c3\u03b7 \u03c3\u03b5 Govern, Map, Measure \u03ba\u03b1\u03b9 Manage \u03b5\u03af\u03bd\u03b1\u03b9 \u03c7\u03c1\u03ae\u03c3\u03b9\u03bc\u03bf \u03c0\u03bb\u03b1\u03af\u03c3\u03b9\u03bf \u03b3\u03b9\u03b1 \u03bd\u03b1 \u03bc\u03b7 \u03bc\u03b5\u03af\u03bd\u03b5\u03b9 \u03b7 \u03c4\u03b5\u03c7\u03bd\u03b9\u03ba\u03ae \u03bc\u03ad\u03c4\u03c1\u03b7\u03c3\u03b7 \u03b1\u03c0\u03bf\u03c3\u03c5\u03bd\u03b4\u03b5\u03b4\u03b5\u03bc\u03ad\u03bd\u03b7 \u03b1\u03c0\u03cc ownership \u03ba\u03b1\u03b9 risk response.<\/p>\n<h2 id=\"oria-meletis\">\u038c\u03c1\u03b9\u03b1 \u03c0\u03bf\u03c5 \u03b4\u03b5\u03bd \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03ba\u03c1\u03cd\u03b2\u03bf\u03bd\u03c4\u03b1\u03b9<\/h2>\n<p>\u03a4\u03b1 \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03ad\u03c3\u03bc\u03b1\u03c4\u03b1 \u03c0\u03c1\u03bf\u03ad\u03c1\u03c7\u03bf\u03bd\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03c6\u03b9\u03bb\u03c4\u03c1\u03b1\u03c1\u03b9\u03c3\u03bc\u03ad\u03bd\u03bf \u03c5\u03c0\u03bf\u03c3\u03cd\u03bd\u03bf\u03bb\u03bf \u03c4\u03bf\u03c5 OSWorld \u03ba\u03b1\u03b9 \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03bf Baidu RLE environment. \u0397 \u03b1\u03c6\u03b1\u03af\u03c1\u03b5\u03c3\u03b7 Chrome \u03ba\u03b1\u03b9 multi-app \u03b5\u03c1\u03b3\u03b1\u03c3\u03b9\u03ce\u03bd \u03bc\u03b5\u03af\u03c9\u03c3\u03b5 \u03c4\u03bf evaluation noise, \u03b1\u03bb\u03bb\u03ac \u03c0\u03b5\u03c1\u03b9\u03cc\u03c1\u03b9\u03c3\u03b5 \u03b1\u03ba\u03c1\u03b9\u03b2\u03ce\u03c2 \u03c4\u03b1 web \u03ba\u03b1\u03b9 cross-application \u03c3\u03b5\u03bd\u03ac\u03c1\u03b9\u03b1 \u03c0\u03bf\u03c5 \u03b5\u03af\u03bd\u03b1\u03b9 \u03c3\u03c5\u03c7\u03bd\u03ac \u03ba\u03c1\u03af\u03c3\u03b9\u03bc\u03b1 \u03c3\u03c4\u03b7\u03bd \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03ae.<\/p>\n<p>\u0397 \u03bc\u03b5\u03bb\u03ad\u03c4\u03b7 \u03b1\u03bd\u03b1\u03c6\u03ad\u03c1\u03b5\u03b9 \u03c4\u03bf \u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf checkpoint \u03c3\u03b5 \u03ad\u03c9\u03c2 50 iterations. \u0391\u03c5\u03c4\u03ae \u03b7 \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b4\u03ce\u03c3\u03b5\u03b9 \u03c0\u03b9\u03bf \u03b1\u03b9\u03c3\u03b9\u03cc\u03b4\u03bf\u03be\u03b7 \u03b5\u03b9\u03ba\u03cc\u03bd\u03b1 \u03b1\u03c0\u03cc \u03ad\u03bd\u03b1\u03bd fixed-checkpoint \u03ba\u03b1\u03bd\u03cc\u03bd\u03b1 \u03ae \u03b1\u03bd\u03b5\u03be\u03ac\u03c1\u03c4\u03b7\u03c4\u03bf validation protocol. \u0395\u03c0\u03af\u03c3\u03b7\u03c2, \u03b7 divergence estimation \u03b5\u03be\u03b1\u03c1\u03c4\u03ac\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc SSIM \u03ba\u03b1\u03b9 \u03ba\u03b1\u03c4\u03ce\u03c6\u03bb\u03b9. \u03a4\u03b1 \u03c3\u03c5\u03bd\u03bf\u03bb\u03b9\u03ba\u03ac success rates \u03ae\u03c4\u03b1\u03bd 49,5%, 50,0% \u03ba\u03b1\u03b9 49,9% \u03b3\u03b9\u03b1 thresholds 0,80, 0,85 \u03ba\u03b1\u03b9 0,90, \u03b1\u03bb\u03bb\u03ac \u03bf\u03b9 domain-level \u03bc\u03b5\u03c4\u03b1\u03b2\u03bf\u03bb\u03ad\u03c2 \u03ae\u03c4\u03b1\u03bd \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b5\u03c2.<\/p>\n<p>\u0397 \u03c3\u03c5\u03bd\u03c4\u03bf\u03bc\u03cc\u03c4\u03b5\u03c1\u03b7 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b7 reference \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03b1\u03bd\u03b1\u03b3\u03ba\u03b1\u03c3\u03c4\u03b9\u03ba\u03ac \u03b7 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ad\u03c3\u03c4\u03b5\u03c1\u03b7, \u03b7 \u03c0\u03b9\u03bf \u03b5\u03be\u03b7\u03b3\u03ae\u03c3\u03b9\u03bc\u03b7 \u03ae \u03b7 \u03c0\u03b9\u03bf \u03c3\u03c5\u03bc\u03bc\u03bf\u03c1\u03c6\u03c9\u03bc\u03ad\u03bd\u03b7 \u03c3\u03b5 \u03ba\u03ac\u03b8\u03b5 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03cc workflow. \u03a4\u03bf Success Buffer \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03b5\u03c0\u03af\u03c3\u03b7\u03c2 \u03bd\u03b1 \u03c0\u03b1\u03bb\u03b9\u03ce\u03c3\u03b5\u03b9 \u03cc\u03c4\u03b1\u03bd \u03b1\u03bb\u03bb\u03ac\u03be\u03b5\u03b9 \u03c4\u03bf UI. \u03a7\u03c1\u03b5\u03b9\u03ac\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 reference freshness, invalidation rules \u03ba\u03b1\u03b9 tests \u03bc\u03b5\u03c4\u03ac \u03b1\u03c0\u03cc \u03b1\u03bb\u03bb\u03b1\u03b3\u03ad\u03c2 \u03c3\u03b5 layout, \u03b4\u03b9\u03ba\u03b1\u03b9\u03ce\u03bc\u03b1\u03c4\u03b1 \u03ae business policy.<\/p>\n<p>\u03a4\u03ad\u03bb\u03bf\u03c2, \u03c4\u03b1 \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03ad\u03c3\u03bc\u03b1\u03c4\u03b1 \u03b4\u03b5\u03bd \u03b1\u03c0\u03bf\u03b4\u03b5\u03b9\u03ba\u03bd\u03cd\u03bf\u03c5\u03bd \u03cc\u03c4\u03b9 \u03b7 LACL-GUI \u03b8\u03b1 \u03ad\u03c7\u03b5\u03b9 \u03c4\u03bf \u03af\u03b4\u03b9\u03bf uplift \u03c3\u03b5 e-commerce, marketing \u03ae customer support. \u0394\u03b5\u03af\u03c7\u03bd\u03bf\u03c5\u03bd \u03cc\u03c4\u03b9 \u03c4\u03bf trajectory-level supervision \u03b1\u03be\u03af\u03b6\u03b5\u03b9 \u03bd\u03b1 \u03b5\u03be\u03b5\u03c4\u03b1\u03c3\u03c4\u03b5\u03af \u03bc\u03b5 \u03b1\u03c5\u03c3\u03c4\u03b7\u03c1\u03ac ablations \u03ba\u03b1\u03b9 production-specific verifiers.<\/p>\n<h2 id=\"symperasma\">\u03a4\u03bf \u03c3\u03c9\u03c3\u03c4\u03cc \u03c3\u03c5\u03bc\u03c0\u03ad\u03c1\u03b1\u03c3\u03bc\u03b1<\/h2>\n<p>\u0397 LACL-GUI \u03b1\u03bb\u03bb\u03ac\u03b6\u03b5\u03b9 \u03c4\u03bf \u03b5\u03c1\u03ce\u03c4\u03b7\u03bc\u03b1 \u03b1\u03c0\u03cc \u00ab\u03c0\u03ad\u03c4\u03c5\u03c7\u03b5 \u03ae \u03b1\u03c0\u03ad\u03c4\u03c5\u03c7\u03b5;\u00bb \u03c3\u03b5 \u00ab\u03c0\u03ce\u03c2 \u03c0\u03ad\u03c4\u03c5\u03c7\u03b5 \u03ba\u03b1\u03b9 \u03c0\u03cc\u03c3\u03bf \u03ba\u03bf\u03bd\u03c4\u03ac \u03ad\u03c6\u03c4\u03b1\u03c3\u03b5 \u03cc\u03c4\u03b1\u03bd \u03b1\u03c0\u03ad\u03c4\u03c5\u03c7\u03b5;\u00bb. \u039c\u03b5 length-aware \u03c0\u03c1\u03bf\u03c4\u03af\u03bc\u03b7\u03c3\u03b7 \u03c3\u03c4\u03b9\u03c2 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b5\u03c2, divergence-aware \u03b1\u03bd\u03c4\u03b9\u03bc\u03b5\u03c4\u03ce\u03c0\u03b9\u03c3\u03b7 \u03c4\u03c9\u03bd \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03b9\u03ce\u03bd, Success Buffer \u03ba\u03b1\u03b9 \u03b1\u03c3\u03cd\u03b3\u03c7\u03c1\u03bf\u03bd\u03b7 \u03c3\u03c5\u03bb\u03bb\u03bf\u03b3\u03ae trajectories, \u03c0\u03c1\u03bf\u03c3\u03b8\u03ad\u03c4\u03b5\u03b9 \u03c0\u03bf\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b4\u03b9\u03b1\u03b4\u03b9\u03ba\u03b1\u03c3\u03af\u03b1\u03c2 \u03c0\u03ac\u03bd\u03c9 \u03b1\u03c0\u03cc \u03c4\u03bf \u03b4\u03c5\u03b1\u03b4\u03b9\u03ba\u03cc \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1.<\/p>\n<p>\u03a3\u03c4\u03bf \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03bf OSWorld subset, \u03b7 \u03bc\u03ad\u03b8\u03bf\u03b4\u03bf\u03c2 \u03b2\u03b5\u03bb\u03c4\u03af\u03c9\u03c3\u03b5 \u03c4\u03bf overall success \u03c4\u03bf\u03c5 8B backbone \u03ba\u03b1\u03b9 \u03bc\u03b5\u03af\u03c9\u03c3\u03b5 \u03c4\u03bf \u03bc\u03ad\u03c3\u03bf \u03bc\u03ae\u03ba\u03bf\u03c2 \u03c3\u03b5 \u03ba\u03bf\u03b9\u03bd\u03ac outcomes. \u03a4\u03b1 domain exceptions, \u03c4\u03bf \u03c6\u03b9\u03bb\u03c4\u03c1\u03ac\u03c1\u03b9\u03c3\u03bc\u03b1 \u03c4\u03bf\u03c5 benchmark \u03ba\u03b1\u03b9 \u03b7 \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae \u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf\u03c5 checkpoint \u03b1\u03c0\u03b1\u03b9\u03c4\u03bf\u03cd\u03bd \u03cc\u03bc\u03c9\u03c2 \u03c0\u03c1\u03bf\u03c3\u03b5\u03ba\u03c4\u03b9\u03ba\u03ae \u03b5\u03c1\u03bc\u03b7\u03bd\u03b5\u03af\u03b1.<\/p>\n<p>\u0393\u03b9\u03b1 \u03bc\u03b9\u03b1 \u03b5\u03c0\u03b9\u03c7\u03b5\u03af\u03c1\u03b7\u03c3\u03b7, \u03b7 \u03c0\u03b9\u03bf \u03c7\u03c1\u03ae\u03c3\u03b9\u03bc\u03b7 \u03c0\u03c1\u03b1\u03ba\u03c4\u03b9\u03ba\u03ae \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03bd\u03b1 \u03b1\u03bd\u03c4\u03b9\u03b3\u03c1\u03ac\u03c8\u03b5\u03b9 \u03ad\u03bd\u03b1\u03bd \u03b1\u03ba\u03b1\u03b4\u03b7\u03bc\u03b1\u03ca\u03ba\u03cc loss. \u0395\u03af\u03bd\u03b1\u03b9 \u03bd\u03b1 \u03ba\u03b1\u03c4\u03b1\u03c3\u03ba\u03b5\u03c5\u03ac\u03c3\u03b5\u03b9 \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7 \u03c0\u03bf\u03c5 \u03be\u03b5\u03c7\u03c9\u03c1\u03af\u03b6\u03b5\u03b9 outcome, efficiency, divergence, recovery \u03ba\u03b1\u03b9 safety checks. \u039c\u03cc\u03bd\u03bf \u03c4\u03cc\u03c4\u03b5 \u03b7 \u03b2\u03b5\u03bb\u03c4\u03af\u03c9\u03c3\u03b7 \u03b5\u03bd\u03cc\u03c2 GUI agent \u03b1\u03bd\u03c4\u03b9\u03c3\u03c4\u03bf\u03b9\u03c7\u03b5\u03af \u03c3\u03b5 \u03c0\u03b9\u03bf \u03b1\u03be\u03b9\u03cc\u03c0\u03b9\u03c3\u03c4\u03b7 \u03b4\u03bf\u03c5\u03bb\u03b5\u03b9\u03ac \u03ba\u03b1\u03b9 \u03cc\u03c7\u03b9 \u03b1\u03c0\u03bb\u03ce\u03c2 \u03c3\u03b5 \u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf score.<\/p>\n<section class=\"td-service-cta\">\n<div class=\"td-service-cta-content\">\n<p class=\"td-service-cta-kicker\">\u0391\u03c0\u03cc \u03c4\u03bf demo \u03c3\u03b5 \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03bf computer-use workflow<\/p>\n<p class=\"td-service-cta-title\">\u03a3\u03c7\u03b5\u03b4\u03b9\u03ac\u03c3\u03c4\u03b5 GUI agents \u03bc\u03b5 verifiers, trajectory observability \u03ba\u03b1\u03b9 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ae handoffs<\/p>\n<p>\u0397 TWO DOTS \u03c7\u03b1\u03c1\u03c4\u03bf\u03b3\u03c1\u03b1\u03c6\u03b5\u03af business invariants, state checks, retry budgets, failure slices, audit trails \u03ba\u03b1\u03b9 rollout gates \u03ce\u03c3\u03c4\u03b5 \u03bf agent \u03bd\u03b1 \u03bf\u03bb\u03bf\u03ba\u03bb\u03b7\u03c1\u03ce\u03bd\u03b5\u03b9 \u03c4\u03b7 \u03c3\u03c9\u03c3\u03c4\u03ae \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 \u03bc\u03b5 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ae \u03c0\u03bf\u03c5 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af\u03c4\u03b5 \u03bd\u03b1 \u03b5\u03bb\u03ad\u03b3\u03be\u03b5\u03c4\u03b5.<\/p>\n<div class=\"td-service-cta-actions\"><a class=\"td-service-cta-button\" href=\"https:\/\/twodots.gr\/aftomatismoi-epicheiriseon-ai\/\">\u03a3\u03c5\u03b6\u03b7\u03c4\u03ae\u03c3\u03c4\u03b5 \u03ad\u03bd\u03b1\u03bd \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03bf AI \u03b1\u03c5\u03c4\u03bf\u03bc\u03b1\u03c4\u03b9\u03c3\u03bc\u03cc<\/a><\/div>\n<\/div>\n<\/section>\n<section id=\"sychnes-erotiseis\" class=\"td-faq-section\">\n<div class=\"td-faq\">\n<p class=\"td-faq-heading\">\u0427\u0435\u0441\u0442\u043e \u0437\u0430\u0434\u0430\u0432\u0430\u043d\u0438 \u0432\u044a\u043f\u0440\u043e\u0441\u0438<\/p>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a4\u03b9 \u03b5\u03af\u03bd\u03b1\u03b9 \u03ad\u03bd\u03b1\u03c2 GUI agent;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u0395\u03af\u03bd\u03b1\u03b9 \u03c3\u03cd\u03c3\u03c4\u03b7\u03bc\u03b1 AI \u03c0\u03bf\u03c5 \u03bc\u03b5\u03c4\u03b1\u03c4\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bc\u03b9\u03b1 \u03bf\u03b4\u03b7\u03b3\u03af\u03b1 \u03c3\u03b5 \u03b1\u03ba\u03bf\u03bb\u03bf\u03c5\u03b8\u03af\u03b1 \u03b5\u03bd\u03b5\u03c1\u03b3\u03b5\u03b9\u03ce\u03bd \u03c0\u03ac\u03bd\u03c9 \u03c3\u03b5 \u03b3\u03c1\u03b1\u03c6\u03b9\u03ba\u03cc \u03c0\u03b5\u03c1\u03b9\u03b2\u03ac\u03bb\u03bb\u03bf\u03bd, \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03ce\u03bd\u03c4\u03b1\u03c2 \u03bf\u03c0\u03c4\u03b9\u03ba\u03ad\u03c2 \u03c0\u03b1\u03c1\u03b1\u03c4\u03b7\u03c1\u03ae\u03c3\u03b5\u03b9\u03c2, \u03b9\u03c3\u03c4\u03bf\u03c1\u03b9\u03ba\u03cc \u03ba\u03b1\u03b9 \u03b4\u03bf\u03bc\u03b7\u03bc\u03ad\u03bd\u03b5\u03c2 \u03b5\u03bd\u03c4\u03bf\u03bb\u03ad\u03c2 \u03cc\u03c0\u03c9\u03c2 \u03ba\u03bb\u03b9\u03ba \u03ae \u03c0\u03bb\u03b7\u03ba\u03c4\u03c1\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a4\u03b9 \u03c0\u03c1\u03bf\u03c3\u03b8\u03ad\u03c4\u03b5\u03b9 \u03b7 LACL-GUI \u03c3\u03c4\u03bf \u03b4\u03c5\u03b1\u03b4\u03b9\u03ba\u03cc reward;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u03a0\u03c1\u03bf\u03c3\u03b8\u03ad\u03c4\u03b5\u03b9 \u03c0\u03c1\u03bf\u03c4\u03af\u03bc\u03b7\u03c3\u03b7 \u03b3\u03b9\u03b1 \u03c0\u03b9\u03bf \u03c3\u03cd\u03bd\u03c4\u03bf\u03bc\u03b5\u03c2 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b5\u03c2 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ad\u03c2 \u03ba\u03b1\u03b9 \u03b7\u03c0\u03b9\u03cc\u03c4\u03b5\u03c1\u03b7 \u03b1\u03bd\u03c4\u03b9\u03bc\u03b5\u03c4\u03ce\u03c0\u03b9\u03c3\u03b7 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03b9\u03ce\u03bd \u03c0\u03bf\u03c5 \u03b1\u03c0\u03bf\u03ba\u03bb\u03af\u03bd\u03bf\u03c5\u03bd \u03b1\u03c1\u03b3\u03ac \u03b1\u03c0\u03cc \u03bc\u03b9\u03b1 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b7 \u03b1\u03bd\u03b1\u03c6\u03bf\u03c1\u03ac, \u03c7\u03c9\u03c1\u03af\u03c2 \u03bd\u03b1 \u03b1\u03bb\u03bb\u03ac\u03b6\u03b5\u03b9 \u03c4\u03bf \u03b2\u03b1\u03c3\u03b9\u03ba\u03cc \u03cc\u03c1\u03b9\u03bf \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b1\u03c2 \u03ba\u03b1\u03b9 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b1\u03c2.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u0393\u03b9\u03b1\u03c4\u03af \u03c0\u03c1\u03bf\u03c4\u03b9\u03bc\u03ce\u03bd\u03c4\u03b1\u03b9 \u03bf\u03b9 \u03c3\u03cd\u03bd\u03c4\u03bf\u03bc\u03b5\u03c2 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b5\u03c2 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ad\u03c2;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u039c\u03ad\u03c3\u03b1 \u03c3\u03c4\u03bf \u03af\u03b4\u03b9\u03bf task, \u03bc\u03b5 \u03af\u03b4\u03b9\u03bf \u03b1\u03c1\u03c7\u03b9\u03ba\u03cc state \u03ba\u03b1\u03b9 \u03af\u03b4\u03b9\u03bf \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03bf outcome, \u03c4\u03b1 \u03b5\u03c0\u03b9\u03c0\u03bb\u03ad\u03bf\u03bd \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1 \u03c3\u03c5\u03c7\u03bd\u03ac \u03b4\u03b5\u03af\u03c7\u03bd\u03bf\u03c5\u03bd \u03ac\u03c3\u03ba\u03bf\u03c0\u03b7 \u03b5\u03be\u03b5\u03c1\u03b5\u03cd\u03bd\u03b7\u03c3\u03b7, \u03b5\u03c0\u03b1\u03bd\u03b1\u03bb\u03ae\u03c8\u03b5\u03b9\u03c2 \u03ae \u03bb\u03ac\u03b8\u03b7 \u03bc\u03b5 \u03bc\u03b5\u03c4\u03b1\u03b3\u03b5\u03bd\u03ad\u03c3\u03c4\u03b5\u03c1\u03b7 \u03b1\u03bd\u03ac\u03ba\u03b1\u03bc\u03c8\u03b7. \u0394\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03ba\u03b1\u03b8\u03bf\u03bb\u03b9\u03ba\u03cc\u03c2 \u03ba\u03b1\u03bd\u03cc\u03bd\u03b1\u03c2 \u03b3\u03b9\u03b1 \u03ba\u03ac\u03b8\u03b5 workflow.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a0\u03ce\u03c2 \u03bc\u03b5\u03c4\u03c1\u03b9\u03ad\u03c4\u03b1\u03b9 \u03b7 \u03b1\u03c0\u03cc\u03ba\u03bb\u03b9\u03c3\u03b7 \u03bc\u03b9\u03b1\u03c2 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b7\u03c2 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ae\u03c2;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u039f\u03b9 \u03b4\u03b9\u03b1\u03b4\u03bf\u03c7\u03b9\u03ba\u03ad\u03c2 \u03bf\u03c0\u03c4\u03b9\u03ba\u03ad\u03c2 \u03c0\u03b1\u03c1\u03b1\u03c4\u03b7\u03c1\u03ae\u03c3\u03b5\u03b9\u03c2 \u03c3\u03c5\u03b3\u03ba\u03c1\u03af\u03bd\u03bf\u03bd\u03c4\u03b1\u03b9 \u03bc\u03b5 \u03c4\u03b7 \u03c3\u03c5\u03bd\u03c4\u03bf\u03bc\u03cc\u03c4\u03b5\u03c1\u03b7 \u03b4\u03b9\u03b1\u03b8\u03ad\u03c3\u03b9\u03bc\u03b7 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b7 \u03b1\u03bd\u03b1\u03c6\u03bf\u03c1\u03ac \u03c4\u03b7\u03c2 \u03af\u03b4\u03b9\u03b1\u03c2 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1\u03c2 \u03bc\u03ad\u03c3\u03c9 SSIM \u03ba\u03b1\u03b9 \u03ba\u03b1\u03c4\u03c9\u03c6\u03bb\u03af\u03bf\u03c5 \u03bf\u03bc\u03bf\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1\u03c2. \u03a4\u03bf \u03c0\u03c1\u03ce\u03c4\u03bf \u03c3\u03b7\u03bc\u03b5\u03af\u03bf \u03bf\u03c5\u03c3\u03b9\u03b1\u03c3\u03c4\u03b9\u03ba\u03ae\u03c2 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03ac\u03c2 \u03bf\u03c1\u03af\u03b6\u03b5\u03b9 \u03c4\u03bf divergence length.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a0\u03bf\u03b9\u03bf \u03ae\u03c4\u03b1\u03bd \u03c4\u03bf \u03b2\u03b1\u03c3\u03b9\u03ba\u03cc \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1 \u03c3\u03c4\u03bf OSWorld;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u03a3\u03c4\u03bf \u03c6\u03b9\u03bb\u03c4\u03c1\u03b1\u03c1\u03b9\u03c3\u03bc\u03ad\u03bd\u03bf \u03c3\u03cd\u03bd\u03bf\u03bb\u03bf 222 \u03b5\u03c1\u03b3\u03b1\u03c3\u03b9\u03ce\u03bd, \u03c4\u03bf LACL-GUI-8B \u03ad\u03c6\u03c4\u03b1\u03c3\u03b5 50,0% overall success, \u03ad\u03bd\u03b1\u03bd\u03c4\u03b9 47,2% \u03b3\u03b9\u03b1 GRPO, 47,3% \u03b3\u03b9\u03b1 REAL \u03ba\u03b1\u03b9 42,7% \u03b3\u03b9\u03b1 \u03c4\u03bf \u03b1\u03bd\u03c4\u03af\u03c3\u03c4\u03bf\u03b9\u03c7\u03bf backbone.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03ae\u03b8\u03b7\u03ba\u03b1\u03bd \u03ba\u03b1\u03b9 \u03bf\u03b9 369 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b5\u03c2 \u03c4\u03bf\u03c5 OSWorld;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u038c\u03c7\u03b9. \u039f\u03b9 \u03b5\u03c1\u03b5\u03c5\u03bd\u03b7\u03c4\u03ad\u03c2 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03af\u03b7\u03c3\u03b1\u03bd 222 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b5\u03c2 \u03ba\u03b1\u03b9 \u03b1\u03c6\u03b1\u03af\u03c1\u03b5\u03c3\u03b1\u03bd Chrome \u03ba\u03b1\u03b9 multi-app tasks \u03bb\u03cc\u03b3\u03c9 \u03b1\u03c3\u03c4\u03b1\u03b8\u03bf\u03cd\u03c2 \u03b4\u03b9\u03ba\u03c4\u03cd\u03bf\u03c5 \u03ba\u03b1\u03b9 \u03b5\u03bb\u03bb\u03b9\u03c0\u03ce\u03bd \u03b5\u03be\u03c9\u03c4\u03b5\u03c1\u03b9\u03ba\u03ce\u03bd assets \u03c3\u03c4\u03bf \u03b4\u03b9\u03ba\u03cc \u03c4\u03bf\u03c5\u03c2 \u03c0\u03b5\u03c1\u03b9\u03b2\u03ac\u03bb\u03bb\u03bf\u03bd.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u0397 LACL-GUI \u03bc\u03b5\u03b9\u03ce\u03bd\u03b5\u03b9 \u03c0\u03ac\u03bd\u03c4\u03b1 \u03c4\u03bf\u03bd \u03b1\u03c1\u03b9\u03b8\u03bc\u03cc \u03b2\u03b7\u03bc\u03ac\u03c4\u03c9\u03bd;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u038c\u03c7\u03b9. \u039f\u03b9 \u03c3\u03c5\u03bd\u03bf\u03bb\u03b9\u03ba\u03bf\u03af \u03bc\u03ad\u03c3\u03bf\u03b9 \u03cc\u03c1\u03bf\u03b9 \u03bc\u03b5\u03b9\u03ce\u03b8\u03b7\u03ba\u03b1\u03bd \u03c3\u03c4\u03b1 \u03ba\u03bf\u03b9\u03bd\u03ac outcomes, \u03b1\u03bb\u03bb\u03ac \u03c5\u03c0\u03ae\u03c1\u03c7\u03b1\u03bd \u03b5\u03be\u03b1\u03b9\u03c1\u03ad\u03c3\u03b5\u03b9\u03c2 \u03b1\u03bd\u03ac \u03b5\u03c6\u03b1\u03c1\u03bc\u03bf\u03b3\u03ae: Writer \u03ba\u03b1\u03b9 VLC \u03c3\u03c4\u03b9\u03c2 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b5\u03c2, Writer \u03ba\u03b1\u03b9 VS Code \u03c3\u03c4\u03b9\u03c2 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b5\u03c2.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a0\u03bf\u03b9\u03bf \u03b5\u03af\u03bd\u03b1\u03b9 \u03c4\u03bf \u03c0\u03c1\u03b1\u03ba\u03c4\u03b9\u03ba\u03cc \u03bc\u03ac\u03b8\u03b7\u03bc\u03b1 \u03b3\u03b9\u03b1 \u03bc\u03b9\u03b1 \u03b5\u03c0\u03b9\u03c7\u03b5\u03af\u03c1\u03b7\u03c3\u03b7;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u039d\u03b1 \u03bc\u03b5\u03c4\u03c1\u03ac \u03c7\u03c9\u03c1\u03b9\u03c3\u03c4\u03ac \u03c4\u03b5\u03bb\u03b9\u03ba\u03cc outcome, \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1, \u03b5\u03c0\u03b1\u03bd\u03b1\u03bb\u03ae\u03c8\u03b5\u03b9\u03c2, \u03c3\u03b7\u03bc\u03b5\u03af\u03bf \u03c0\u03c1\u03ce\u03c4\u03b7\u03c2 \u03b1\u03c0\u03cc\u03ba\u03bb\u03b9\u03c3\u03b7\u03c2, \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2 \u03ba\u03b1\u03b9 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b1 \u03c4\u03b7\u03c2 \u03c4\u03b5\u03bb\u03b9\u03ba\u03ae\u03c2 \u03b5\u03c0\u03b1\u03bb\u03ae\u03b8\u03b5\u03c5\u03c3\u03b7\u03c2. \u0397 \u03c3\u03c5\u03bd\u03c4\u03bf\u03bc\u03cc\u03c4\u03b5\u03c1\u03b7 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ae \u03b4\u03b5\u03bd \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b1\u03bd\u03c4\u03b1\u03bc\u03b5\u03af\u03b2\u03b5\u03c4\u03b1\u03b9 \u03cc\u03c4\u03b1\u03bd \u03c0\u03b1\u03c1\u03b1\u03bb\u03b5\u03af\u03c0\u03b5\u03b9 \u03b1\u03bd\u03b1\u03b3\u03ba\u03b1\u03af\u03bf \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf \u03ae \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03bf handoff.<\/p>\n<\/div>\n<\/details>\n<\/div>\n<\/section>\n<div class=\"td-source-list\">\n<p id=\"piges\" class=\"td-source-list-title\">\u0418\u0437\u0442\u043e\u0447\u043d\u0438\u0446\u0438<\/p>\n<ul>\n<li><a href=\"https:\/\/arxiv.org\/abs\/2608.21830\" target=\"_blank\" rel=\"noopener\">arXiv \u2014 Beyond Success and Failure: Length-Aware Contrastive Learning for GUI Agents<\/a><\/li>\n<li><a href=\"https:\/\/arxiv.org\/html\/2608.21830\" target=\"_blank\" rel=\"noopener\">arXiv HTML \u2014 \u03c0\u03bb\u03ae\u03c1\u03b7\u03c2 \u03bc\u03b5\u03b8\u03bf\u03b4\u03bf\u03bb\u03bf\u03b3\u03af\u03b1, \u03c0\u03af\u03bd\u03b1\u03ba\u03b5\u03c2, ablations \u03ba\u03b1\u03b9 trajectories \u03c4\u03b7\u03c2 LACL-GUI<\/a><\/li>\n<li><a href=\"https:\/\/arxiv.org\/abs\/2602.05630\" target=\"_blank\" rel=\"noopener\">arXiv \u2014 Rewards as Labels: Revisiting RLVR from a Classification Perspective<\/a><\/li>\n<li><a href=\"https:\/\/osworld-v1.xlang.ai\/\" target=\"_blank\" rel=\"noopener\">OSWorld \u2014 \u03b5\u03c0\u03af\u03c3\u03b7\u03bc\u03bf benchmark \u03ba\u03b1\u03b9 \u03c4\u03b5\u03ba\u03bc\u03b7\u03c1\u03af\u03c9\u03c3\u03b7 \u03c4\u03bf\u03c5 real-computer environment<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/QwenLM\/Qwen3-VL\" target=\"_blank\" rel=\"noopener\">QwenLM \u2014 \u03b5\u03c0\u03af\u03c3\u03b7\u03bc\u03bf repository \u03c4\u03bf\u03c5 Qwen3-VL backbone<\/a><\/li>\n<li><a href=\"https:\/\/www.nist.gov\/itl\/ai-risk-management-framework\/nist-ai-rmf-playbook\" target=\"_blank\" rel=\"noopener\">NIST \u2014 AI Risk Management Framework Playbook<\/a><\/li>\n<\/ul>\n<\/div>","protected":false},"excerpt":{"rendered":"<p>\u0397 LACL-GUI \u03b4\u03b5\u03af\u03c7\u03bd\u03b5\u03b9 \u03b3\u03b9\u03b1\u03c4\u03af \u03c4\u03bf \u03c4\u03b5\u03bb\u03b9\u03ba\u03cc reward \u03b4\u03b5\u03bd \u03b1\u03c1\u03ba\u03b5\u03af \u03b3\u03b9\u03b1 GUI agents \u03ba\u03b1\u03b9 \u03c0\u03ce\u03c2 \u03bc\u03ae\u03ba\u03bf\u03c2, \u03b1\u03c0\u03cc\u03ba\u03bb\u03b9\u03c3\u03b7 \u03ba\u03b1\u03b9 trajectory quality \u03b2\u03b5\u03bb\u03c4\u03b9\u03ce\u03bd\u03bf\u03c5\u03bd \u03c4\u03b7\u03bd \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7.<\/p>","protected":false},"author":1,"featured_media":98881,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"_gspb_post_css":"","content-type":"","footnotes":""},"categories":[199],"tags":[7340,20639,20637,20638,18034],"class_list":["post-97816","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-techniti-noimosyni","tag-ai-automation","tag-contrastive-learning","tag-gui-agents","tag-osworld","tag-reinforcement-learning"],"blocksy_meta":{"styles_descriptor":{"styles":{"desktop":"","tablet":"","mobile":""},"google_fonts":[],"version":8}},"_links":{"self":[{"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/posts\/97816","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/comments?post=97816"}],"version-history":[{"count":1,"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/posts\/97816\/revisions"}],"predecessor-version":[{"id":98882,"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/posts\/97816\/revisions\/98882"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/media\/98881"}],"wp:attachment":[{"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/media?parent=97816"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/categories?post=97816"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/tags?post=97816"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}