{"id":98245,"date":"2026-10-06T10:29:42","date_gmt":"2026-10-06T07:29:42","guid":{"rendered":"https:\/\/twodots.gr\/?p=98245"},"modified":"2026-10-06T10:29:43","modified_gmt":"2026-10-06T07:29:43","slug":"proof-gen-apotychimenes-diadromes-ai-agent-dedomena-ekpaidefsis","status":"publish","type":"post","link":"https:\/\/twodots.gr\/bg\/proof-gen-apotychimenes-diadromes-ai-agent-dedomena-ekpaidefsis\/","title":{"rendered":"PROOF-Gen: \u03c0\u03ce\u03c2 \u03bf\u03b9 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b5\u03c2 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ad\u03c2 \u03b5\u03bd\u03cc\u03c2 AI agent \u03b3\u03af\u03bd\u03bf\u03bd\u03c4\u03b1\u03b9 \u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b1 \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03b1 \u03b5\u03ba\u03c0\u03b1\u03af\u03b4\u03b5\u03c5\u03c3\u03b7\u03c2"},"content":{"rendered":"<div class=\"td-article-lede\">\n<p><strong>\u041e\u0442\u0433\u043e\u0432\u043e\u0440 \u043f\u044a\u0440\u0432\u043e:<\/strong> \u03a4\u03bf PROOF-Gen \u03b4\u03b5\u03bd \u03b5\u03ba\u03c0\u03b1\u03b9\u03b4\u03b5\u03cd\u03b5\u03b9 \u03ad\u03bd\u03b1\u03bd AI agent \u03c0\u03ac\u03bd\u03c9 \u03c3\u03c4\u03b9\u03c2 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b5\u03c2 \u03c4\u03bf\u03c5 \u03b1\u03c5\u03c4\u03bf\u03cd\u03c3\u03b9\u03b5\u03c2. \u03a7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03b5\u03af \u03c4\u03bf trace \u03ba\u03b1\u03b9 \u03c4\u03bf feedback \u03c4\u03bf\u03c5 verifier \u03b3\u03b9\u03b1 \u03bd\u03b1 \u03b1\u03bd\u03b1\u03ba\u03c4\u03ae\u03c3\u03b5\u03b9 \u03bc\u03b9\u03b1 \u03c0\u03bb\u03ae\u03c1\u03b7, \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b7 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ae, \u03b1\u03c6\u03b1\u03b9\u03c1\u03b5\u03af \u03c4\u03b7\u03bd \u03c0\u03c1\u03bf\u03c3\u03c9\u03c1\u03b9\u03bd\u03ae \u03ba\u03b1\u03b8\u03bf\u03b4\u03ae\u03b3\u03b7\u03c3\u03b7 \u03ba\u03b1\u03b9 \u03ba\u03c1\u03b1\u03c4\u03ac \u03bc\u03cc\u03bd\u03bf \u03c4\u03b7\u03bd \u03ba\u03b1\u03b8\u03b1\u03c1\u03ae \u03b5\u03c0\u03af\u03b4\u03b5\u03b9\u03be\u03b7 \u03b3\u03b9\u03b1 distillation. \u0397 \u03b1\u03be\u03af\u03b1 \u03b2\u03c1\u03af\u03c3\u03ba\u03b5\u03c4\u03b1\u03b9 \u03c3\u03c4\u03b7 \u03bc\u03b5\u03c4\u03b1\u03c4\u03c1\u03bf\u03c0\u03ae \u03b5\u03bd\u03cc\u03c2 near-miss \u03c3\u03b5 \u03b5\u03bb\u03ad\u03b3\u03be\u03b9\u03bc\u03bf training signal.<\/p>\n<\/div>\n<p>\u03a4\u03bf PROOF-Gen \u03b1\u03bd\u03b1\u03ba\u03c4\u03ac \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b5\u03c2 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ad\u03c2 \u03b1\u03c0\u03cc failures AI agents \u03ba\u03b1\u03b9 \u03c4\u03b9\u03c2 \u03bc\u03b5\u03c4\u03b1\u03c4\u03c1\u03ad\u03c0\u03b5\u03b9 \u03c3\u03b5 \u03ba\u03b1\u03b8\u03b1\u03c1\u03ac \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03b1 \u03b3\u03b9\u03b1 \u03c0\u03b9\u03bf \u03b1\u03c3\u03c6\u03b1\u03bb\u03ad\u03c2 model distillation \u03bc\u03b5 \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf.<\/p>\n<p>\u03a4\u03b9 \u03b1\u03be\u03af\u03b1 \u03ad\u03c7\u03b5\u03b9 \u03ad\u03bd\u03b1\u03c2 AI agent \u03c0\u03bf\u03c5 \u03b5\u03ba\u03c4\u03b5\u03bb\u03b5\u03af \u03c3\u03c9\u03c3\u03c4\u03ac \u03c3\u03c7\u03b5\u03b4\u03cc\u03bd \u03cc\u03bb\u03b1 \u03c4\u03b1 \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1 \u03bc\u03b9\u03b1\u03c2 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1\u03c2, \u03b1\u03bb\u03bb\u03ac \u03b1\u03c0\u03bf\u03c4\u03c5\u03b3\u03c7\u03ac\u03bd\u03b5\u03b9 \u03c3\u03c4\u03bf \u03ad\u03bd\u03b1 \u03b2\u03ae\u03bc\u03b1 \u03c0\u03bf\u03c5 \u03ba\u03c1\u03af\u03bd\u03b5\u03b9 \u03c4\u03bf \u03c4\u03b5\u03bb\u03b9\u03ba\u03cc \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1; \u03a3\u03c4\u03b1 \u03c0\u03b5\u03c1\u03b9\u03c3\u03c3\u03cc\u03c4\u03b5\u03c1\u03b1 generate-and-filter pipelines \u03b7 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ae \u03b1\u03c0\u03bf\u03c1\u03c1\u03af\u03c0\u03c4\u03b5\u03c4\u03b1\u03b9 \u03bf\u03bb\u03cc\u03ba\u03bb\u03b7\u03c1\u03b7. \u0397 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 <strong>PROOF-Gen: From Optimized Data to Better Distillation<\/strong> \u03c4\u03c9\u03bd Anh Ta, Junjie Zhu \u03ba\u03b1\u03b9 Shahin Shayandeh \u03c5\u03c0\u03bf\u03c3\u03c4\u03b7\u03c1\u03af\u03b6\u03b5\u03b9 \u03cc\u03c4\u03b9 \u03ad\u03c4\u03c3\u03b9 \u03c7\u03ac\u03bd\u03bf\u03bd\u03c4\u03b1\u03b9 \u03b1\u03ba\u03c1\u03b9\u03b2\u03ce\u03c2 \u03c4\u03b1 \u03b4\u03cd\u03c3\u03ba\u03bf\u03bb\u03b1 \u03c3\u03b5\u03bd\u03ac\u03c1\u03b9\u03b1 \u03c0\u03bf\u03c5 \u03ad\u03bd\u03b1 \u03bc\u03b9\u03ba\u03c1\u03cc\u03c4\u03b5\u03c1\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03bc\u03ac\u03b8\u03b5\u03b9 \u03bd\u03b1 \u03bf\u03bb\u03bf\u03ba\u03bb\u03b7\u03c1\u03ce\u03bd\u03b5\u03b9.<\/p>\n<p>\u03a4\u03bf PROOF-Gen, \u03b1\u03c0\u03cc \u03c4\u03bf <em>Per-scenario Reflective Optimization to Overcome Failed Generation<\/em>, \u03c0\u03b1\u03c1\u03b5\u03bc\u03b2\u03ac\u03bb\u03bb\u03b5\u03b9 \u03ad\u03bd\u03b1 \u03c3\u03c4\u03ac\u03b4\u03b9\u03bf \u03b1\u03bd\u03ac\u03ba\u03c4\u03b7\u03c3\u03b7\u03c2 \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03c4\u03bf supervised fine-tuning. \u0388\u03bd\u03b1\u03c2 reflector \u03b4\u03b9\u03b1\u03b2\u03ac\u03b6\u03b5\u03b9 \u03c4\u03bf \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03bf execution trace \u03ba\u03b1\u03b9 \u03c4\u03bf feedback \u03c4\u03c9\u03bd evaluators, \u03c0\u03b1\u03c1\u03ac\u03b3\u03b5\u03b9 \u03c0\u03c1\u03bf\u03c3\u03c9\u03c1\u03b9\u03bd\u03ae \u03b4\u03b9\u03bf\u03c1\u03b8\u03c9\u03c4\u03b9\u03ba\u03ae \u03ba\u03b1\u03b8\u03bf\u03b4\u03ae\u03b3\u03b7\u03c3\u03b7 \u03ba\u03b1\u03b9 \u03bf\u03b4\u03b7\u03b3\u03b5\u03af \u03c4\u03bf\u03bd \u03af\u03b4\u03b9\u03bf executor \u03c3\u03b5 \u03bd\u03ad\u03b1 \u03b5\u03ba\u03c4\u03ad\u03bb\u03b5\u03c3\u03b7 \u03b1\u03c0\u03cc \u03c4\u03b7\u03bd \u03b1\u03c1\u03c7\u03ae. \u0391\u03bd \u03cc\u03bb\u03bf\u03b9 \u03bf\u03b9 \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf\u03b9 \u03c0\u03b5\u03c1\u03ac\u03c3\u03bf\u03c5\u03bd, \u03b7 \u03ba\u03b1\u03b8\u03bf\u03b4\u03ae\u03b3\u03b7\u03c3\u03b7 \u03b1\u03c6\u03b1\u03b9\u03c1\u03b5\u03af\u03c4\u03b1\u03b9 \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03c4\u03b7\u03bd \u03b5\u03ba\u03c0\u03b1\u03af\u03b4\u03b5\u03c5\u03c3\u03b7 \u03c4\u03bf\u03c5 student.<\/p>\n<aside class=\"td-article-note\"><strong>\u0397 \u03c3\u03c9\u03c3\u03c4\u03ae \u03b1\u03bd\u03ac\u03b3\u03bd\u03c9\u03c3\u03b7:<\/strong> \u03c4\u03bf 93% recovery \u03b1\u03c6\u03bf\u03c1\u03ac 279 \u03b1\u03c0\u03cc 300 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b1 telecom tasks \u03c0\u03bf\u03c5 \u03b5\u03af\u03c7\u03b1\u03bd \u03b5\u03c0\u03b9\u03bb\u03b5\u03b3\u03b5\u03af \u03b1\u03c0\u03cc failures \u03c4\u03bf\u03c5 GPT-4o \u03c3\u03c4\u03bf \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03bf \u03c42-bench setup. \u0394\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03b5\u03b3\u03b3\u03c5\u03b7\u03bc\u03ad\u03bd\u03bf recovery rate \u03b3\u03b9\u03b1 \u03ba\u03ac\u03b8\u03b5 agent, verifier \u03ae \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03b7\u03c3\u03b9\u03b1\u03ba\u03cc workflow.<\/aside>\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=\"#generate-filter-gap\">\u03a4\u03bf \u03ba\u03b5\u03bd\u03cc \u03c4\u03bf\u03c5 generate-and-filter<\/a><\/li>\n<li><a href=\"#proof-gen-method\">\u03a0\u03ce\u03c2 \u03bb\u03b5\u03b9\u03c4\u03bf\u03c5\u03c1\u03b3\u03b5\u03af \u03c4\u03bf PROOF-Gen<\/a><\/li>\n<li><a href=\"#benchmarks-results\">\u03a4\u03b9 \u03ad\u03b4\u03b5\u03b9\u03be\u03b1\u03bd \u03c42-bench \u03ba\u03b1\u03b9 BFCL<\/a><\/li>\n<li><a href=\"#partial-credit\">\u0393\u03b9\u03b1\u03c4\u03af \u03c4\u03bf partial credit \u03b4\u03b5\u03bd \u03b1\u03c1\u03ba\u03b5\u03af<\/a><\/li>\n<li><a href=\"#reflector-quality\">\u0397 \u03c0\u03bf\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1 \u03c4\u03bf\u03c5 reflector \u03ba\u03b1\u03b8\u03bf\u03c1\u03af\u03b6\u03b5\u03b9 \u03c4\u03bf recovery<\/a><\/li>\n<li><a href=\"#capability-voice\">Capability \u03ba\u03b1\u03b9 brand voice \u03bc\u03c0\u03bf\u03c1\u03bf\u03cd\u03bd \u03bd\u03b1 \u03b4\u03b9\u03b1\u03c7\u03c9\u03c1\u03b9\u03c3\u03c4\u03bf\u03cd\u03bd<\/a><\/li>\n<li><a href=\"#shortcut-audit\">\u039f \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf\u03c2 \u03b3\u03b9\u03b1 shortcuts \u03ba\u03b1\u03b9 \u03b4\u03b9\u03b1\u03c1\u03c1\u03bf\u03ae<\/a><\/li>\n<li><a href=\"#production-agents\">\u03a4\u03b9 \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u03b3\u03b9\u03b1 production AI agents<\/a><\/li>\n<li><a href=\"#praktiko-protokollo\">\u0395\u03c0\u03c4\u03ac \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1 \u03b3\u03b9\u03b1 \u03b5\u03bb\u03ad\u03b3\u03be\u03b9\u03bc\u03bf recovery pipeline<\/a><\/li>\n<li><a href=\"#oria-kostos\">\u038c\u03c1\u03b9\u03b1, \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2 \u03ba\u03b1\u03b9 governance<\/a><\/li>\n<li><a href=\"#apotychia-signal\">\u0397 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b1 \u03c9\u03c2 training signal<\/a><\/li>\n<\/ul>\n<\/div>\n<h2 id=\"generate-filter-gap\">\u03a4\u03bf \u03ba\u03b5\u03bd\u03cc \u03c4\u03bf\u03c5 generate-and-filter<\/h2>\n<p>\u0397 \u03c3\u03c5\u03bd\u03b7\u03b8\u03b9\u03c3\u03bc\u03ad\u03bd\u03b7 \u03c0\u03c1\u03ce\u03c4\u03b7 \u03c6\u03ac\u03c3\u03b7 \u03c4\u03bf\u03c5 model distillation \u03b3\u03b9\u03b1 tool-calling agents \u03b5\u03af\u03bd\u03b1\u03b9 supervised fine-tuning \u03c0\u03ac\u03bd\u03c9 \u03c3\u03b5 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ad\u03c2 \u03c0\u03bf\u03c5 \u03c0\u03b1\u03c1\u03ac\u03b3\u03b5\u03b9 \u03ad\u03bd\u03b1 \u03b9\u03c3\u03c7\u03c5\u03c1\u03cc teacher model. \u03a4\u03bf pipeline \u03b5\u03ba\u03c4\u03b5\u03bb\u03b5\u03af \u03ba\u03ac\u03b8\u03b5 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1, \u03b5\u03bb\u03ad\u03b3\u03c7\u03b5\u03b9 \u03c4\u03bf \u03c4\u03b5\u03bb\u03b9\u03ba\u03cc \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1 \u03ba\u03b1\u03b9 \u03ba\u03c1\u03b1\u03c4\u03ac \u03bc\u03cc\u03bd\u03bf \u03c4\u03b1 traces \u03c0\u03bf\u03c5 \u03c0\u03ad\u03c1\u03b1\u03c3\u03b1\u03bd. \u0397 \u03bb\u03bf\u03b3\u03b9\u03ba\u03ae \u03c0\u03c1\u03bf\u03c3\u03c4\u03b1\u03c4\u03b5\u03cd\u03b5\u03b9 \u03c4\u03bf training set \u03b1\u03c0\u03cc \u03b5\u03bc\u03c6\u03b1\u03bd\u03ce\u03c2 \u03bb\u03b1\u03bd\u03b8\u03b1\u03c3\u03bc\u03ad\u03bd\u03b1 \u03c0\u03b1\u03c1\u03b1\u03b4\u03b5\u03af\u03b3\u03bc\u03b1\u03c4\u03b1, \u03b1\u03bb\u03bb\u03ac \u03ad\u03bd\u03b1 \u03b4\u03c5\u03b1\u03b4\u03b9\u03ba\u03cc \u03c6\u03af\u03bb\u03c4\u03c1\u03bf \u03b1\u03bd\u03c4\u03b9\u03bc\u03b5\u03c4\u03c9\u03c0\u03af\u03b6\u03b5\u03b9 \u03c4\u03bf \u03af\u03b4\u03b9\u03bf \u03bc\u03b9\u03b1 \u03b5\u03ba\u03c4\u03ad\u03bb\u03b5\u03c3\u03b7 \u03c0\u03bf\u03c5 \u03b1\u03c0\u03ad\u03c4\u03c5\u03c7\u03b5 \u03b1\u03bc\u03ad\u03c3\u03c9\u03c2 \u03ba\u03b1\u03b9 \u03bc\u03b9\u03b1 \u03b5\u03ba\u03c4\u03ad\u03bb\u03b5\u03c3\u03b7 \u03c0\u03bf\u03c5 \u03bf\u03bb\u03bf\u03ba\u03bb\u03ae\u03c1\u03c9\u03c3\u03b5 \u03c4\u03ad\u03c3\u03c3\u03b5\u03c1\u03b1 \u03b1\u03c0\u03cc \u03c4\u03b1 \u03c0\u03ad\u03bd\u03c4\u03b5 \u03b1\u03c0\u03b1\u03c1\u03b1\u03af\u03c4\u03b7\u03c4\u03b1 \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1.<\/p>\n<p>\u03a3\u03c4\u03b7\u03bd \u03b1\u03bd\u03ac\u03bb\u03c5\u03c3\u03b7 \u03c4\u03b7\u03c2 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1\u03c2 \u03c0\u03ac\u03bd\u03c9 \u03c3\u03c4\u03bf \u03c42-bench, 473 \u03b1\u03c0\u03cc 834 trials \u03b1\u03c0\u03ad\u03c4\u03c5\u03c7\u03b1\u03bd. \u0391\u03c0\u03cc \u03b1\u03c5\u03c4\u03ad\u03c2 \u03c4\u03b9\u03c2 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b5\u03c2, \u03c4\u03bf 67% \u03b5\u03af\u03c7\u03b5 \u03b5\u03ba\u03c4\u03b5\u03bb\u03ad\u03c3\u03b5\u03b9 \u03c3\u03c9\u03c3\u03c4\u03ac \u03c0\u03ac\u03bd\u03c9 \u03b1\u03c0\u03cc \u03c4\u03b9\u03c2 \u03bc\u03b9\u03c3\u03ad\u03c2 \u03b1\u03c0\u03b1\u03b9\u03c4\u03bf\u03cd\u03bc\u03b5\u03bd\u03b5\u03c2 tool calls. \u039f\u03b9 \u03b5\u03c1\u03b5\u03c5\u03bd\u03b7\u03c4\u03ad\u03c2 \u03c0\u03b5\u03c1\u03b9\u03b3\u03c1\u03ac\u03c6\u03bf\u03c5\u03bd near-misses \u03cc\u03c0\u03bf\u03c5 \u03bf agent \u03b5\u03af\u03c7\u03b5 \u03c7\u03b5\u03b9\u03c1\u03b9\u03c3\u03c4\u03b5\u03af \u03c4\u03b9\u03c2 \u03c1\u03bf\u03c5\u03c4\u03af\u03bd\u03b5\u03c2 \u03b1\u03bb\u03bb\u03ac \u03c0\u03b1\u03c1\u03ad\u03bb\u03b5\u03b9\u03c8\u03b5 \u03bc\u03af\u03b1 \u03ba\u03c1\u03af\u03c3\u03b9\u03bc\u03b7 \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b1, \u03cc\u03c0\u03c9\u03c2 \u03ad\u03bd\u03b1 roaming toggle, \u03bc\u03b9\u03b1 \u03ac\u03b4\u03b5\u03b9\u03b1 SMS \u03ae \u03ad\u03bd\u03b1 data refuel. \u039c\u03af\u03b1 \u03c0\u03b1\u03c1\u03ac\u03bb\u03b5\u03b9\u03c8\u03b7 \u03b1\u03c1\u03ba\u03bf\u03cd\u03c3\u03b5 \u03b3\u03b9\u03b1 \u03bb\u03ac\u03b8\u03bf\u03c2 environment \u03ae database state.<\/p>\n<p>\u0391\u03c5\u03c4\u03cc \u03ad\u03c7\u03b5\u03b9 \u03ac\u03bc\u03b5\u03c3\u03b7 \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03b7\u03c3\u03b9\u03b1\u03ba\u03ae \u03c3\u03b7\u03bc\u03b1\u03c3\u03af\u03b1. \u03a3\u03b5 \u03c5\u03c0\u03bf\u03c3\u03c4\u03ae\u03c1\u03b9\u03be\u03b7 \u03c0\u03b5\u03bb\u03b1\u03c4\u03ce\u03bd, e-commerce operations \u03ae \u03b5\u03c3\u03c9\u03c4\u03b5\u03c1\u03b9\u03ba\u03ac workflows, \u03ad\u03bd\u03b1 training set \u03c0\u03bf\u03c5 \u03ba\u03c1\u03b1\u03c4\u03ac \u03bc\u03cc\u03bd\u03bf \u03c4\u03b9\u03c2 \u03b5\u03cd\u03ba\u03bf\u03bb\u03b5\u03c2 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03af\u03b5\u03c2 \u03b3\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03ba\u03b1\u03b8\u03c1\u03ad\u03c6\u03c4\u03b7\u03c2 \u03cc\u03c3\u03c9\u03bd \u03bf teacher \u03ae\u03b4\u03b7 \u03bb\u03cd\u03bd\u03b5\u03b9. \u03a4\u03b1 \u03b4\u03cd\u03c3\u03ba\u03bf\u03bb\u03b1 \u03c3\u03b7\u03bc\u03b5\u03af\u03b1 \u03b1\u03c0\u03cc\u03c6\u03b1\u03c3\u03b7\u03c2 \u03b1\u03c0\u03bf\u03c5\u03c3\u03b9\u03ac\u03b6\u03bf\u03c5\u03bd, \u03c0\u03b1\u03c1\u03cc\u03c4\u03b9 \u03b5\u03ba\u03b5\u03af \u03c3\u03c5\u03b3\u03ba\u03b5\u03bd\u03c4\u03c1\u03ce\u03bd\u03b5\u03c4\u03b1\u03b9 \u03bf \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03cc\u03c2 \u03ba\u03af\u03bd\u03b4\u03c5\u03bd\u03bf\u03c2. \u0397 \u03b4\u03b9\u03ac\u03ba\u03c1\u03b9\u03c3\u03b7 \u03b1\u03bd\u03ac\u03bc\u03b5\u03c3\u03b1 \u03c3\u03c4\u03b7 \u03c3\u03c9\u03c3\u03c4\u03ae \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7 \u03ba\u03b1\u03b9 \u03c3\u03c4\u03bf \u03c3\u03c9\u03c3\u03c4\u03cc artifact \u03c6\u03b1\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03ba\u03b1\u03b9 \u03c3\u03c4\u03bf <a href=\"https:\/\/twodots.gr\/k-bench-ai-agents-epistimoniki-akriveia-artifacts\/\">K-Bench \u03b3\u03b9\u03b1 \u03c4\u03b7\u03bd \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7 AI agents<\/a>.<\/p>\n<h2 id=\"proof-gen-method\">\u03a0\u03ce\u03c2 \u03bb\u03b5\u03b9\u03c4\u03bf\u03c5\u03c1\u03b3\u03b5\u03af \u03c4\u03bf PROOF-Gen<\/h2>\n<p>\u0393\u03b9\u03b1 \u03ba\u03ac\u03b8\u03b5 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03bf task, \u03bf reflector \u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03b5\u03b9 \u03bf\u03bb\u03cc\u03ba\u03bb\u03b7\u03c1\u03bf \u03c4\u03bf execution trace \u03ba\u03b1\u03b9 \u03c4\u03bf feedback \u03c4\u03bf\u03c5 verifier: \u03c0\u03bf\u03b9\u03bf assertion \u03b1\u03c0\u03ad\u03c4\u03c5\u03c7\u03b5, \u03c0\u03bf\u03b9\u03b1 \u03ba\u03b1\u03c4\u03ac\u03c3\u03c4\u03b1\u03c3\u03b7 \u03ad\u03bc\u03b5\u03b9\u03bd\u03b5 \u03bb\u03b1\u03bd\u03b8\u03b1\u03c3\u03bc\u03ad\u03bd\u03b7 \u03ba\u03b1\u03b9 \u03c0\u03bf\u03cd \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03ac\u03bb\u03bb\u03b7 \u03b1\u03c0\u03cc\u03c6\u03b1\u03c3\u03b7. \u03a3\u03c4\u03b7 \u03c3\u03c5\u03bd\u03ad\u03c7\u03b5\u03b9\u03b1 \u03b3\u03c1\u03ac\u03c6\u03b5\u03b9 \u03ad\u03bd\u03b1 \u03c0\u03c1\u03bf\u03c3\u03c9\u03c1\u03b9\u03bd\u03cc cheatsheet \u03bc\u03b5 \u03b4\u03b9\u03b1\u03b4\u03b9\u03ba\u03b1\u03c3\u03c4\u03b9\u03ba\u03ae \u03ba\u03b1\u03b8\u03bf\u03b4\u03ae\u03b3\u03b7\u03c3\u03b7, \u03cc\u03c0\u03c9\u03c2 \u03c4\u03b7 \u03c3\u03c9\u03c3\u03c4\u03ae \u03c3\u03b5\u03b9\u03c1\u03ac \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03c9\u03bd \u03ae \u03c4\u03b7\u03bd \u03c0\u03c1\u03bf\u03cb\u03c0\u03cc\u03b8\u03b5\u03c3\u03b7 \u03c0\u03bf\u03c5 \u03b1\u03b3\u03bd\u03bf\u03ae\u03b8\u03b7\u03ba\u03b5.<\/p>\n<p>\u039f executor \u03be\u03b5\u03ba\u03b9\u03bd\u03ac \u03c4\u03b7\u03bd \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 \u03b1\u03c0\u03cc \u03c4\u03b7\u03bd \u03b1\u03c1\u03c7\u03ae \u03bc\u03b5 \u03b1\u03c5\u03c4\u03ae \u03c4\u03b7\u03bd \u03ba\u03b1\u03b8\u03bf\u03b4\u03ae\u03b3\u03b7\u03c3\u03b7. \u0391\u03bd \u03b1\u03c0\u03bf\u03c4\u03cd\u03c7\u03b5\u03b9 \u03be\u03b1\u03bd\u03ac, \u03bf reflector \u03b1\u03bd\u03b1\u03b8\u03b5\u03c9\u03c1\u03b5\u03af \u03c4\u03bf cheatsheet \u03ba\u03b1\u03b9 \u03b1\u03ba\u03bf\u03bb\u03bf\u03c5\u03b8\u03b5\u03af \u03bd\u03ad\u03b1 \u03c0\u03c1\u03bf\u03c3\u03c0\u03ac\u03b8\u03b5\u03b9\u03b1, \u03ad\u03c9\u03c2 \u03b4\u03ad\u03ba\u03b1 iterations \u03c3\u03c4\u03bf \u03c0\u03b5\u03af\u03c1\u03b1\u03bc\u03b1. \u0397 \u03b2\u03b5\u03bb\u03c4\u03b9\u03c3\u03c4\u03bf\u03c0\u03bf\u03af\u03b7\u03c3\u03b7 \u03b3\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03b1\u03bd\u03ac \u03c3\u03b5\u03bd\u03ac\u03c1\u03b9\u03bf. \u03a3\u03b5 \u03b1\u03bd\u03c4\u03af\u03b8\u03b5\u03c3\u03b7 \u03bc\u03b5 \u03ad\u03bd\u03b1 \u03b3\u03b5\u03bd\u03b9\u03ba\u03cc prompt optimization \u03c0\u03bf\u03c5 \u03b1\u03bd\u03b1\u03b6\u03b7\u03c4\u03ac \u03bf\u03b4\u03b7\u03b3\u03af\u03b5\u03c2 \u03b3\u03b9\u03b1 \u03c0\u03bf\u03bb\u03bb\u03ac inputs, \u03b5\u03b4\u03ce \u03c4\u03bf prompt \u03b5\u03c0\u03b9\u03c4\u03c1\u03ad\u03c0\u03b5\u03c4\u03b1\u03b9 \u03bd\u03b1 \u03c5\u03c0\u03b5\u03c1\u03c0\u03c1\u03bf\u03c3\u03b1\u03c1\u03bc\u03bf\u03c3\u03c4\u03b5\u03af \u03c3\u03b5 \u03bc\u03af\u03b1 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1, \u03b5\u03c0\u03b5\u03b9\u03b4\u03ae \u03b5\u03af\u03bd\u03b1\u03b9 \u03c0\u03c1\u03bf\u03c3\u03c9\u03c1\u03b9\u03bd\u03cc \u03ba\u03b1\u03b9 \u03b8\u03b1 \u03b1\u03c0\u03bf\u03c1\u03c1\u03b9\u03c6\u03b8\u03b5\u03af.<\/p>\n<p>\u03a4\u03bf \u03c0\u03c1\u03bf\u03ca\u03cc\u03bd \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03c4\u03bf cheatsheet \u03b1\u03bb\u03bb\u03ac \u03b7 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b7 trajectory. \u03a0\u03c1\u03b9\u03bd \u03bc\u03c0\u03b5\u03b9 \u03c3\u03c4\u03bf training set, \u03b7 \u03c0\u03c1\u03bf\u03c3\u03c9\u03c1\u03b9\u03bd\u03ae \u03ba\u03b1\u03b8\u03bf\u03b4\u03ae\u03b3\u03b7\u03c3\u03b7 \u03b1\u03c6\u03b1\u03b9\u03c1\u03b5\u03af\u03c4\u03b1\u03b9. \u039f student \u03b2\u03bb\u03ad\u03c0\u03b5\u03b9 \u03c4\u03b9\u03c2 \u03b1\u03c0\u03bf\u03c6\u03ac\u03c3\u03b5\u03b9\u03c2, \u03c4\u03b9\u03c2 tool calls \u03ba\u03b1\u03b9 \u03c4\u03bf \u03c4\u03b5\u03bb\u03b9\u03ba\u03cc \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1, \u03cc\u03c7\u03b9 \u03c4\u03b7 \u03c3\u03ba\u03b1\u03bb\u03c9\u03c3\u03b9\u03ac \u03c0\u03bf\u03c5 \u03b2\u03bf\u03ae\u03b8\u03b7\u03c3\u03b5 \u03c4\u03bf\u03bd teacher \u03bd\u03b1 \u03c4\u03b1 \u03c0\u03b1\u03c1\u03ac\u03b3\u03b5\u03b9. \u0388\u03c4\u03c3\u03b9 \u03c4\u03bf pipeline \u03b5\u03c0\u03b9\u03b4\u03b9\u03ce\u03ba\u03b5\u03b9 \u03bd\u03b1 \u03bc\u03b5\u03c4\u03b1\u03c6\u03ad\u03c1\u03b5\u03b9 \u03c3\u03c5\u03bc\u03c0\u03b5\u03c1\u03b9\u03c6\u03bf\u03c1\u03ac \u03b5\u03c0\u03af\u03bb\u03c5\u03c3\u03b7\u03c2 \u03c7\u03c9\u03c1\u03af\u03c2 \u03bd\u03b1 \u03b1\u03bd\u03c4\u03b9\u03b3\u03c1\u03ac\u03c6\u03b5\u03b9 task-specific \u03c5\u03c0\u03bf\u03b4\u03b5\u03af\u03be\u03b5\u03b9\u03c2.<\/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\">Generate-and-filter<\/p>\n<p>\u039a\u03c1\u03b1\u03c4\u03ac \u03bc\u03cc\u03bd\u03bf \u03cc,\u03c4\u03b9 \u03c0\u03ad\u03c4\u03c5\u03c7\u03b5 \u03c3\u03c4\u03b7\u03bd \u03c0\u03c1\u03ce\u03c4\u03b7 \u03b5\u03ba\u03c4\u03ad\u03bb\u03b5\u03c3\u03b7. \u0395\u03af\u03bd\u03b1\u03b9 \u03b1\u03c0\u03bb\u03cc, \u03b1\u03bb\u03bb\u03ac \u03b1\u03c6\u03ae\u03bd\u03b5\u03b9 \u03c4\u03b1 \u03b4\u03cd\u03c3\u03ba\u03bf\u03bb\u03b1 failures \u03c7\u03c9\u03c1\u03af\u03c2 training signal.<\/p>\n<div class=\"td-badge-row\"><span class=\"td-badge\">Binary filter<\/span><span class=\"td-badge\">\u0395\u03cd\u03ba\u03bf\u03bb\u03b1 cases<\/span><\/div>\n<\/div>\n<div class=\"td-platform-card\">\n<p class=\"td-comparison-title\">Partial-credit training<\/p>\n<p>\u039c\u03b1\u03b8\u03b1\u03af\u03bd\u03b5\u03b9 \u03b1\u03c0\u03cc \u03c3\u03c9\u03c3\u03c4\u03ad\u03c2 \u03b5\u03bd\u03b4\u03b9\u03ac\u03bc\u03b5\u03c3\u03b5\u03c2 \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b5\u03c2, \u03c7\u03c9\u03c1\u03af\u03c2 \u03b1\u03c0\u03b1\u03c1\u03b1\u03af\u03c4\u03b7\u03c4\u03b1 \u03bd\u03b1 \u03b4\u03b5\u03af\u03c7\u03bd\u03b5\u03b9 \u03c4\u03b7\u03bd \u03c0\u03bb\u03ae\u03c1\u03b7 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ae \u03bc\u03ad\u03c3\u03b1 \u03b1\u03c0\u03cc \u03c4\u03bf \u03ba\u03c1\u03af\u03c3\u03b9\u03bc\u03bf decision point.<\/p>\n<div class=\"td-badge-row\"><span class=\"td-badge\">Tool accuracy<\/span><span class=\"td-badge\">\u0391\u03c4\u03b5\u03bb\u03ad\u03c2 \u03c4\u03ad\u03bb\u03bf\u03c2<\/span><\/div>\n<\/div>\n<div class=\"td-platform-card td-platform-card--navy\">\n<p class=\"td-comparison-title\">PROOF-Gen recovery<\/p>\n<p>\u03a7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03b5\u03af reflection \u03b3\u03b9\u03b1 \u03bd\u03b1 \u03c0\u03b1\u03c1\u03ac\u03b3\u03b5\u03b9 \u03bf\u03bb\u03bf\u03ba\u03bb\u03b7\u03c1\u03c9\u03bc\u03ad\u03bd\u03b7, \u03b5\u03c0\u03b1\u03bb\u03b7\u03b8\u03b5\u03c5\u03bc\u03ad\u03bd\u03b7 trajectory \u03ba\u03b1\u03b9 \u03b1\u03c6\u03b1\u03b9\u03c1\u03b5\u03af \u03c4\u03bf \u03c0\u03c1\u03bf\u03c3\u03c9\u03c1\u03b9\u03bd\u03cc scaffold \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03c4\u03bf SFT.<\/p>\n<div class=\"td-badge-row\"><span class=\"td-badge\">Verifier<\/span><span class=\"td-badge\">End-to-end<\/span><\/div>\n<\/div>\n<\/div>\n<\/div>\n<h2 id=\"benchmarks-results\">\u03a4\u03b9 \u03ad\u03b4\u03b5\u03b9\u03be\u03b1\u03bd \u03c42-bench \u03ba\u03b1\u03b9 BFCL<\/h2>\n<p>\u03a3\u03c4\u03bf telecom training pool \u03c4\u03bf\u03c5 \u03c42-bench \u03c5\u03c0\u03ae\u03c1\u03c7\u03b1\u03bd 2.171 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b5\u03c2. \u039f GPT-4o teacher \u03c0\u03ad\u03c1\u03b1\u03c3\u03b5 \u03b1\u03c1\u03c7\u03b9\u03ba\u03ac 158 \u03ba\u03b1\u03b9 \u03b1\u03c0\u03ad\u03c4\u03c5\u03c7\u03b5 \u03c3\u03b5 2.013. \u039f\u03b9 \u03b5\u03c1\u03b5\u03c5\u03bd\u03b7\u03c4\u03ad\u03c2 \u03ad\u03c3\u03c4\u03b5\u03b9\u03bb\u03b1\u03bd 300 failures \u03c3\u03c4\u03bf recovery loop \u03ba\u03b1\u03b9 \u03b1\u03bd\u03ad\u03ba\u03c4\u03b7\u03c3\u03b1\u03bd 279, \u03b4\u03b7\u03bb\u03b1\u03b4\u03ae 93%. \u039f\u03b9 264 \u03b1\u03c0\u03bf\u03b4\u03cc\u03b8\u03b7\u03ba\u03b1\u03bd \u03c3\u03c4\u03b7\u03bd \u03ba\u03b1\u03b8\u03bf\u03b4\u03ae\u03b3\u03b7\u03c3\u03b7 \u03c4\u03bf\u03c5 cheatsheet \u03ba\u03b1\u03b9 \u03bf\u03b9 15 \u03c3\u03b5 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03bf stochastic replay \u03c4\u03bf\u03c5 user simulator. \u03a4\u03bf \u03c4\u03b5\u03bb\u03b9\u03ba\u03cc telecom training set \u03c0\u03b5\u03c1\u03b9\u03b5\u03af\u03c7\u03b5 437 trajectories, \u03b1\u03c0\u03cc \u03c4\u03b9\u03c2 \u03bf\u03c0\u03bf\u03af\u03b5\u03c2 \u03c4\u03bf 64% \u03c0\u03c1\u03bf\u03ae\u03bb\u03b8\u03b5 \u03b1\u03c0\u03cc recovery.<\/p>\n<p>\u03a3\u03c4\u03bf BFCL v4 multi-turn, \u03bf teacher \u03c0\u03ad\u03c1\u03b1\u03c3\u03b5 296 \u03b1\u03c0\u03cc 560 training tasks. \u0397 \u03bc\u03ad\u03b8\u03bf\u03b4\u03bf\u03c2 \u03b5\u03c6\u03b1\u03c1\u03bc\u03cc\u03c3\u03c4\u03b7\u03ba\u03b5 \u03ba\u03b1\u03b9 \u03c3\u03c4\u03b9\u03c2 264 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b5\u03c2 \u03ba\u03b1\u03b9 \u03b1\u03bd\u03ad\u03ba\u03c4\u03b7\u03c3\u03b5 89, \u03c0\u03bf\u03c3\u03bf\u03c3\u03c4\u03cc 33,7%. \u0397 \u03c7\u03b1\u03bc\u03b7\u03bb\u03cc\u03c4\u03b5\u03c1\u03b7 \u03b1\u03bd\u03ac\u03ba\u03c4\u03b7\u03c3\u03b7 \u03b4\u03b5\u03bd \u03b1\u03bd\u03b1\u03b9\u03c1\u03b5\u03af \u03c4\u03b7\u03bd \u03b1\u03c1\u03c7\u03ae\u00b7 \u03b4\u03b5\u03af\u03c7\u03bd\u03b5\u03b9 \u03cc\u03c4\u03b9 \u03c4\u03bf difficulty distribution \u03ba\u03b1\u03b9 \u03b7 \u03b4\u03c5\u03bd\u03b1\u03c4\u03cc\u03c4\u03b7\u03c4\u03b1 \u03c4\u03bf\u03c5 verifier \u03b1\u03bb\u03bb\u03ac\u03b6\u03bf\u03c5\u03bd \u03c4\u03bf \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1. \u03a4\u03bf recovery rate \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b4\u03b9\u03b1\u03b2\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03c0\u03ac\u03bd\u03c4\u03b1 \u03bc\u03b1\u03b6\u03af \u03bc\u03b5 \u03c4\u03bf \u03b1\u03c1\u03c7\u03b9\u03ba\u03cc failure pool.<\/p>\n<p>\u039c\u03b5 combined training, \u03c4\u03bf Qwen3-4B-Instruct-2507 \u03b2\u03b5\u03bb\u03c4\u03b9\u03ce\u03b8\u03b7\u03ba\u03b5 \u03c3\u03c4\u03bf \u03c42-bench \u03b1\u03c0\u03cc Pass@1 0,132 \u03c3\u03b5 0,529. \u03a4\u03bf paper \u03b1\u03bd\u03b1\u03c6\u03ad\u03c1\u03b5\u03b9 \u03b5\u03c0\u03af\u03c3\u03b7\u03c2 \u03b2\u03b5\u03bb\u03c4\u03af\u03c9\u03c3\u03b7 7,2 \u03c0\u03bf\u03c3\u03bf\u03c3\u03c4\u03b9\u03b1\u03af\u03c9\u03bd \u03bc\u03bf\u03bd\u03ac\u03b4\u03c9\u03bd \u03b3\u03b9\u03b1 \u03c4\u03bf Gemma 4 E4B-it \u03c3\u03c4\u03bf BFCL v4 multi-turn. \u0391\u03c5\u03c4\u03ac \u03b5\u03af\u03bd\u03b1\u03b9 \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03ad\u03c3\u03bc\u03b1\u03c4\u03b1 \u03b4\u03cd\u03bf benchmark setups, \u03cc\u03c7\u03b9 \u03b3\u03b5\u03bd\u03b9\u03ba\u03cc claim \u03cc\u03c4\u03b9 \u03ba\u03ac\u03b8\u03b5 \u03bc\u03b9\u03ba\u03c1\u03cc \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03b8\u03b1 \u03b1\u03c0\u03bf\u03ba\u03c4\u03ae\u03c3\u03b5\u03b9 \u03c4\u03b7\u03bd \u03af\u03b4\u03b9\u03b1 \u03b2\u03b5\u03bb\u03c4\u03af\u03c9\u03c3\u03b7.<\/p>\n<div class=\"td-chart td-chart--metrics\">\n<div class=\"td-chart-head\">\n<p class=\"td-chart-kicker\">\u03a4\u03b5\u03ba\u03bc\u03b7\u03c1\u03b9\u03c9\u03bc\u03ad\u03bd\u03bf scope<\/p>\n<p class=\"td-chart-title\">\u03a4\u03ad\u03c3\u03c3\u03b5\u03c1\u03b1 \u03bc\u03b5\u03b3\u03ad\u03b8\u03b7 \u03c4\u03bf\u03c5 PROOF-Gen<\/p>\n<p class=\"td-chart-intro\">\u039f\u03b9 \u03c4\u03b9\u03bc\u03ad\u03c2 \u03c0\u03c1\u03bf\u03ad\u03c1\u03c7\u03bf\u03bd\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03c4\u03b7 \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03b7 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 \u03ba\u03b1\u03b9 \u03c0\u03b5\u03c1\u03b9\u03b3\u03c1\u03ac\u03c6\u03bf\u03c5\u03bd datasets, recovery \u03ba\u03b1\u03b9 downstream evaluation. \u0394\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 benchmark \u03b1\u03b3\u03bf\u03c1\u03ac\u03c2 \u03ae \u03c5\u03c0\u03cc\u03c3\u03c7\u03b5\u03c3\u03b7 ROI.<\/p>\n<\/div>\n<div class=\"td-metric-grid\">\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">57%<\/span><span class=\"td-metric-label\">Teacher trials \u03c0\u03bf\u03c5 \u03b1\u03c0\u03ad\u03c4\u03c5\u03c7\u03b1\u03bd \u03c3\u03c4\u03b7\u03bd \u03b1\u03bd\u03ac\u03bb\u03c5\u03c3\u03b7 \u03c42-bench<\/span><\/div>\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">93%<\/span><span class=\"td-metric-label\">Recovery \u03c3\u03b5 300 \u03b5\u03c0\u03b9\u03bb\u03b5\u03b3\u03bc\u03ad\u03bd\u03b1 telecom failures<\/span><\/div>\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">0,132 \u2192 0,529<\/span><span class=\"td-metric-label\">Pass@1 \u03c4\u03bf\u03c5 Qwen \u03c3\u03c4\u03bf \u03c42-bench \u03bc\u03b5 combined training<\/span><\/div>\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">+7,2 \u03bc\u03bf\u03bd\u03ac\u03b4\u03b5\u03c2<\/span><span class=\"td-metric-label\">Task accuracy gain \u03c4\u03bf\u03c5 Gemma \u03c3\u03c4\u03bf BFCL v4 multi-turn<\/span><\/div>\n<\/div>\n<\/div>\n<h2 id=\"partial-credit\">\u0393\u03b9\u03b1\u03c4\u03af \u03c4\u03bf partial credit \u03b4\u03b5\u03bd \u03b1\u03c1\u03ba\u03b5\u03af<\/h2>\n<p>\u039c\u03b9\u03b1 \u03c6\u03c5\u03c3\u03b9\u03ba\u03ae \u03b5\u03bd\u03b1\u03bb\u03bb\u03b1\u03ba\u03c4\u03b9\u03ba\u03ae \u03b5\u03af\u03bd\u03b1\u03b9 \u03bd\u03b1 \u03ba\u03c1\u03b1\u03c4\u03b7\u03b8\u03bf\u03cd\u03bd \u03bf\u03b9 \u03c3\u03c9\u03c3\u03c4\u03ad\u03c2 \u03b5\u03c0\u03b9\u03bc\u03ad\u03c1\u03bf\u03c5\u03c2 \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b5\u03c2 \u03ba\u03b1\u03b9 \u03bd\u03b1 \u03b4\u03bf\u03b8\u03b5\u03af partial credit. \u0397 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 \u03b1\u03bd\u03b1\u03c6\u03ad\u03c1\u03b5\u03b9 \u03cc\u03c4\u03b9 \u03c4\u03bf filtered-only training \u03b1\u03cd\u03be\u03b7\u03c3\u03b5 \u03c4\u03b7\u03bd action accuracy \u03c0\u03b5\u03c1\u03af\u03c0\u03bf\u03c5 \u03ba\u03b1\u03c4\u03ac 25 \u03c0\u03bf\u03c3\u03bf\u03c3\u03c4\u03b9\u03b1\u03af\u03b5\u03c2 \u03bc\u03bf\u03bd\u03ac\u03b4\u03b5\u03c2, \u03b1\u03bb\u03bb\u03ac \u03b4\u03b5\u03bd \u03b2\u03b5\u03bb\u03c4\u03af\u03c9\u03c3\u03b5 \u03c4\u03bf end-to-end task completion \u03c3\u03b5 \u03ba\u03b1\u03bd\u03ad\u03bd\u03b1 \u03b1\u03c0\u03cc \u03c4\u03b1 \u03c4\u03ad\u03c3\u03c3\u03b5\u03c1\u03b1 \u03b6\u03b5\u03cd\u03b3\u03b7 \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf\u03c5 \u03ba\u03b1\u03b9 benchmark.<\/p>\n<p>\u0393\u03b9\u03b1 \u03c4\u03bf Qwen \u03c3\u03c4\u03bf telecom benchmark, \u03b7 action accuracy \u03b1\u03bd\u03ad\u03b2\u03b7\u03ba\u03b5 \u03b1\u03c0\u03cc 41,4% \u03c3\u03b5 66,3% \u03bc\u03b5 filtered-only QLoRA, \u03b5\u03bd\u03ce \u03b7 environment accuracy \u03ad\u03bc\u03b5\u03b9\u03bd\u03b5 \u03bf\u03c5\u03c3\u03b9\u03b1\u03c3\u03c4\u03b9\u03ba\u03ac \u03c3\u03c4\u03ac\u03c3\u03b9\u03bc\u03b7, \u03b1\u03c0\u03cc 23,5% \u03c3\u03b5 22,6%. \u039c\u03b5 \u03c4\u03bf combined dataset, \u03c0\u03bf\u03c5 \u03c0\u03b5\u03c1\u03b9\u03bb\u03ac\u03bc\u03b2\u03b1\u03bd\u03b5 recovered trajectories, \u03bf\u03b9 \u03b1\u03bd\u03c4\u03af\u03c3\u03c4\u03bf\u03b9\u03c7\u03b5\u03c2 \u03c4\u03b9\u03bc\u03ad\u03c2 \u03ad\u03c6\u03c4\u03b1\u03c3\u03b1\u03bd 83,3% \u03ba\u03b1\u03b9 54,5%. \u0397 database accuracy \u03b1\u03bd\u03ad\u03b2\u03b7\u03ba\u03b5 \u03b1\u03c0\u03cc 9,3% \u03c3\u03b5 15,6%.<\/p>\n<p>\u0397 \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03b7\u03c3\u03b9\u03b1\u03ba\u03ae \u03bc\u03b5\u03c4\u03ac\u03c6\u03c1\u03b1\u03c3\u03b7 \u03b5\u03af\u03bd\u03b1\u03b9 \u03b1\u03c0\u03bb\u03ae: \u03ad\u03bd\u03b1\u03c2 agent \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03c6\u03b1\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03b4\u03c1\u03b1\u03c3\u03c4\u03ae\u03c1\u03b9\u03bf\u03c2 \u03ba\u03b1\u03b9 \u03bd\u03b1 \u03ba\u03b1\u03bb\u03b5\u03af \u03c0\u03bf\u03bb\u03bb\u03ac \u03c3\u03c9\u03c3\u03c4\u03ac APIs \u03c7\u03c9\u03c1\u03af\u03c2 \u03bd\u03b1 \u03bf\u03bb\u03bf\u03ba\u03bb\u03b7\u03c1\u03ce\u03bd\u03b5\u03b9 \u03c4\u03b7\u03bd \u03c5\u03c0\u03cc\u03b8\u03b5\u03c3\u03b7. \u0391\u03bd \u03bc\u03b9\u03b1 \u03b1\u03bb\u03bb\u03b1\u03b3\u03ae \u03b4\u03b9\u03b5\u03cd\u03b8\u03c5\u03bd\u03c3\u03b7\u03c2, \u03b5\u03c0\u03b9\u03c3\u03c4\u03c1\u03bf\u03c6\u03ae \u03ae \u03b1\u03ba\u03cd\u03c1\u03c9\u03c3\u03b7 \u03ba\u03b1\u03c4\u03b1\u03bb\u03ae\u03b3\u03b5\u03b9 \u03c3\u03b5 \u03bb\u03ac\u03b8\u03bf\u03c2 \u03c4\u03b5\u03bb\u03b9\u03ba\u03ae \u03ba\u03b1\u03c4\u03ac\u03c3\u03c4\u03b1\u03c3\u03b7, \u03c4\u03bf \u03c0\u03bb\u03ae\u03b8\u03bf\u03c2 \u03c4\u03c9\u03bd \u03c3\u03c9\u03c3\u03c4\u03ce\u03bd tool calls \u03b4\u03b5\u03bd \u03c3\u03ce\u03b6\u03b5\u03b9 \u03c4\u03bf workflow. \u0393\u03b9\u2019 \u03b1\u03c5\u03c4\u03cc \u03b7 <a href=\"https:\/\/twodots.gr\/ai-agents-behavioral-testing\/\">\u03c3\u03c5\u03bc\u03c0\u03b5\u03c1\u03b9\u03c6\u03bf\u03c1\u03b9\u03ba\u03ae \u03b4\u03bf\u03ba\u03b9\u03bc\u03ae \u03c4\u03c9\u03bd AI agents<\/a> \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 end-to-end assertions \u03ba\u03b1\u03b9 \u03cc\u03c7\u03b9 \u03bc\u03cc\u03bd\u03bf partial metrics.<\/p>\n<div class=\"td-decision-band\">\n<p class=\"td-decision-kicker\">\u039a\u03b1\u03bd\u03cc\u03bd\u03b1\u03c2 \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7\u03c2<\/p>\n<p class=\"td-decision-title\">\u039c\u03b7\u03bd \u03c0\u03c1\u03bf\u03c9\u03b8\u03b5\u03af\u03c4\u03b5 agent \u03b5\u03c0\u03b5\u03b9\u03b4\u03ae \u03b2\u03b5\u03bb\u03c4\u03b9\u03ce\u03b8\u03b7\u03ba\u03b5 \u03bc\u03cc\u03bd\u03bf \u03b7 tool-call accuracy<\/p>\n<p>\u0391\u03c0\u03b1\u03b9\u03c4\u03ae\u03c3\u03c4\u03b5 \u03c3\u03c9\u03c3\u03c4\u03ae \u03c4\u03b5\u03bb\u03b9\u03ba\u03ae \u03ba\u03b1\u03c4\u03ac\u03c3\u03c4\u03b1\u03c3\u03b7, policy compliance, provenance \u03c4\u03c9\u03bd tool arguments \u03ba\u03b1\u03b9 \u03b5\u03c0\u03b1\u03bd\u03ac\u03bb\u03b7\u03c8\u03b7 \u03c3\u03b5 holdout tasks. \u0397 \u03b5\u03bd\u03b4\u03b9\u03ac\u03bc\u03b5\u03c3\u03b7 \u03b1\u03ba\u03c1\u03af\u03b2\u03b5\u03b9\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03b4\u03b9\u03b1\u03b3\u03bd\u03c9\u03c3\u03c4\u03b9\u03ba\u03ae\u00b7 \u03c4\u03bf release gate \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03bc\u03b5\u03c4\u03c1\u03ac \u03b1\u03bd \u03bf\u03bb\u03bf\u03ba\u03bb\u03b7\u03c1\u03ce\u03b8\u03b7\u03ba\u03b5 \u03b7 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ae \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 \u03c7\u03c9\u03c1\u03af\u03c2 \u03c0\u03b1\u03c1\u03b1\u03b2\u03af\u03b1\u03c3\u03b7.<\/p>\n<\/div>\n<h2 id=\"reflector-quality\">\u0397 \u03c0\u03bf\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1 \u03c4\u03bf\u03c5 reflector \u03ba\u03b1\u03b8\u03bf\u03c1\u03af\u03b6\u03b5\u03b9 \u03c4\u03bf recovery<\/h2>\n<p>\u03a4\u03bf recovery \u03b4\u03b5\u03bd \u03c0\u03c1\u03bf\u03ad\u03ba\u03c5\u03c8\u03b5 \u03b1\u03c0\u03cc \u03c4\u03c5\u03c6\u03bb\u03ae \u03b5\u03c0\u03b1\u03bd\u03ac\u03bb\u03b7\u03c8\u03b7. \u03a3\u03b5 ablation 82 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03c9\u03bd \u03c3\u03b5\u03bd\u03b1\u03c1\u03af\u03c9\u03bd \u03b1\u03c0\u03cc airline, retail \u03ba\u03b1\u03b9 telecom, \u03b7 \u03b1\u03c5\u03c4\u03bf\u03ba\u03c1\u03b9\u03c4\u03b9\u03ba\u03ae \u03c4\u03bf\u03c5 GPT-4o \u03b1\u03bd\u03ad\u03ba\u03c4\u03b7\u03c3\u03b5 20 \u03b1\u03c0\u03cc \u03c4\u03b1 82, \u03ae 24,4%. \u039c\u03b5 GPT-5.1 \u03c9\u03c2 reflector, \u03b1\u03bd\u03b1\u03ba\u03c4\u03ae\u03b8\u03b7\u03ba\u03b1\u03bd 80 \u03b1\u03c0\u03cc \u03c4\u03b1 82, \u03ae 97,6%, \u03b5\u03bd\u03ce \u03bf executor \u03ba\u03b1\u03b9 \u03bf\u03b9 \u03c5\u03c0\u03cc\u03bb\u03bf\u03b9\u03c0\u03b5\u03c2 \u03c3\u03c5\u03bd\u03b8\u03ae\u03ba\u03b5\u03c2 \u03ad\u03bc\u03b5\u03b9\u03bd\u03b1\u03bd \u03c3\u03c4\u03b1\u03b8\u03b5\u03c1\u03ad\u03c2.<\/p>\n<p>\u0397 \u03c3\u03cd\u03b3\u03ba\u03bb\u03b9\u03c3\u03b7 \u03ae\u03c4\u03b1\u03bd \u03b3\u03c1\u03ae\u03b3\u03bf\u03c1\u03b7. \u0391\u03c0\u03cc \u03c4\u03b9\u03c2 264 cheatsheet-driven \u03b1\u03bd\u03b1\u03ba\u03c4\u03ae\u03c3\u03b5\u03b9\u03c2 \u03c3\u03c4\u03bf telecom run, \u03c4\u03bf 58,7% \u03c0\u03ad\u03c4\u03c5\u03c7\u03b5 \u03c3\u03c4\u03bf \u03c0\u03c1\u03ce\u03c4\u03bf iteration \u03ba\u03b1\u03b9 \u03c4\u03bf 84,1% \u03ad\u03c9\u03c2 \u03c4\u03bf \u03c4\u03c1\u03af\u03c4\u03bf. \u038c\u03c3\u03b1 \u03c3\u03b5\u03bd\u03ac\u03c1\u03b9\u03b1 \u03ad\u03bc\u03b5\u03bd\u03b1\u03bd \u03ac\u03bb\u03c5\u03c4\u03b1 \u03bc\u03b5\u03c4\u03ac \u03c4\u03b1 \u03c0\u03c1\u03ce\u03c4\u03b1 iterations \u03c3\u03c0\u03ac\u03bd\u03b9\u03b1 \u03c3\u03ce\u03b6\u03bf\u03bd\u03c4\u03b1\u03bd \u03b1\u03c1\u03b3\u03cc\u03c4\u03b5\u03c1\u03b1. \u0397 \u03c3\u03c9\u03c3\u03c4\u03ae \u03b4\u03b9\u03ac\u03b3\u03bd\u03c9\u03c3\u03b7 \u03c4\u03b7\u03c2 \u03b1\u03b9\u03c4\u03af\u03b1\u03c2 \u03b1\u03be\u03af\u03b6\u03b5\u03b9 \u03c0\u03b5\u03c1\u03b9\u03c3\u03c3\u03cc\u03c4\u03b5\u03c1\u03bf \u03b1\u03c0\u03cc \u03c0\u03bf\u03bb\u03bb\u03ad\u03c2 \u03b1\u03bd\u03b5\u03be\u03ad\u03bb\u03b5\u03b3\u03ba\u03c4\u03b5\u03c2 \u03c0\u03b1\u03c1\u03b1\u03bb\u03bb\u03b1\u03b3\u03ad\u03c2 \u03c4\u03bf\u03c5 \u03af\u03b4\u03b9\u03bf\u03c5 prompt.<\/p>\n<p>\u0397 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 \u03c3\u03c5\u03b3\u03ba\u03c1\u03af\u03bd\u03b5\u03b9 \u03b5\u03c0\u03af\u03c3\u03b7\u03c2 direct execution \u03b1\u03c0\u03cc \u03b9\u03c3\u03c7\u03c5\u03c1\u03cc\u03c4\u03b5\u03c1\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03bc\u03b5 \u03c4\u03bf \u03c3\u03c7\u03ae\u03bc\u03b1 \u03b9\u03c3\u03c7\u03c5\u03c1\u03cc\u03c2 reflector \u03ba\u03b1\u03b9 \u03c6\u03b8\u03b7\u03bd\u03cc\u03c4\u03b5\u03c1\u03bf\u03c2 executor. \u03a3\u03c4\u03b1 300 \u03b4\u03cd\u03c3\u03ba\u03bf\u03bb\u03b1 telecom tasks, \u03c4\u03bf GPT-5.1 \u03bc\u03b5 high reasoning \u03ad\u03bb\u03c5\u03c3\u03b5 250, \u03b5\u03bd\u03ce \u03bf \u03c3\u03c5\u03bd\u03b4\u03c5\u03b1\u03c3\u03bc\u03cc\u03c2 GPT-4o executor \u03ba\u03b1\u03b9 GPT-5.1 reflector \u03b1\u03bd\u03ad\u03ba\u03c4\u03b7\u03c3\u03b5 279 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03ce\u03bd\u03c4\u03b1\u03c2 \u03c4\u03bf \u03b1\u03ba\u03c1\u03b9\u03b2\u03cc\u03c4\u03b5\u03c1\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03bc\u03cc\u03bd\u03bf \u03b3\u03b9\u03b1 reflection, \u03ba\u03b1\u03c4\u03ac \u03bc\u03ad\u03c3\u03bf \u03cc\u03c1\u03bf 2,2 calls \u03b1\u03bd\u03ac \u03c3\u03b5\u03bd\u03ac\u03c1\u03b9\u03bf. \u03a4\u03bf \u03bf\u03b9\u03ba\u03bf\u03bd\u03bf\u03bc\u03b9\u03ba\u03cc \u03cc\u03c6\u03b5\u03bb\u03bf\u03c2 \u03b5\u03be\u03b1\u03c1\u03c4\u03ac\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03c4\u03b9\u03bc\u03ad\u03c2, \u03bc\u03ae\u03ba\u03bf\u03c2 trajectories \u03ba\u03b1\u03b9 \u03c0\u03bf\u03c3\u03bf\u03c3\u03c4\u03cc \u03b1\u03bd\u03ac\u03ba\u03c4\u03b7\u03c3\u03b7\u03c2\u00b7 \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03bc\u03b5\u03c4\u03c1\u03b7\u03b8\u03b5\u03af, \u03cc\u03c7\u03b9 \u03bd\u03b1 \u03c5\u03c0\u03bf\u03c4\u03b5\u03b8\u03b5\u03af.<\/p>\n<p>\u0391\u03c5\u03c4\u03ae \u03b7 \u03b1\u03c1\u03c7\u03b9\u03c4\u03b5\u03ba\u03c4\u03bf\u03bd\u03b9\u03ba\u03ae \u03bc\u03bf\u03b9\u03ac\u03b6\u03b5\u03b9 \u03bc\u03b5 \u03c4\u03b7 \u03bb\u03bf\u03b3\u03b9\u03ba\u03ae actor, judge \u03ba\u03b1\u03b9 editor: \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03cc\u03c2 \u03c1\u03cc\u03bb\u03bf\u03c2 \u03b5\u03ba\u03c4\u03b5\u03bb\u03b5\u03af, \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03cc\u03c2 \u03b5\u03bb\u03ad\u03b3\u03c7\u03b5\u03b9 \u03ba\u03b1\u03b9 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03cc\u03c2 \u03c0\u03c1\u03bf\u03c4\u03b5\u03af\u03bd\u03b5\u03b9 \u03b4\u03b9\u03bf\u03c1\u03b8\u03ce\u03c3\u03b5\u03b9\u03c2. \u0397 \u03c3\u03c7\u03b5\u03c4\u03b9\u03ba\u03ae \u03c0\u03c1\u03bf\u03c3\u03ad\u03b3\u03b3\u03b9\u03c3\u03b7 \u03b1\u03bd\u03b1\u03bb\u03cd\u03b5\u03c4\u03b1\u03b9 \u03c3\u03c4\u03bf <a href=\"https:\/\/twodots.gr\/actor-judge-editor-ai-schediasmos-choris-allagi-stochou\/\">Actor, Judge, Editor \u03b3\u03b9\u03b1 \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03bf AI \u03c3\u03c7\u03b5\u03b4\u03b9\u03b1\u03c3\u03bc\u03cc<\/a>.<\/p>\n<h2 id=\"capability-voice\">Capability \u03ba\u03b1\u03b9 brand voice \u03bc\u03c0\u03bf\u03c1\u03bf\u03cd\u03bd \u03bd\u03b1 \u03b4\u03b9\u03b1\u03c7\u03c9\u03c1\u03b9\u03c3\u03c4\u03bf\u03cd\u03bd<\/h2>\n<p>\u0388\u03bd\u03b1 \u03b5\u03bd\u03b4\u03b9\u03b1\u03c6\u03ad\u03c1\u03bf\u03bd \u03b5\u03cd\u03c1\u03b7\u03bc\u03b1 \u03b3\u03b9\u03b1 product teams \u03b5\u03af\u03bd\u03b1\u03b9 \u03cc\u03c4\u03b9 \u03bf reflector \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b2\u03b5\u03bb\u03c4\u03b9\u03ce\u03c3\u03b5\u03b9 \u03c4\u03bf \u03c4\u03b9 \u03ba\u03ac\u03bd\u03b5\u03b9 \u03bf agent \u03c7\u03c9\u03c1\u03af\u03c2 \u03bd\u03b1 \u03b5\u03c0\u03b9\u03b2\u03ac\u03bb\u03bb\u03b5\u03b9 \u03c4\u03bf\u03bd \u03b4\u03b9\u03ba\u03cc \u03c4\u03bf\u03c5 \u03c4\u03c1\u03cc\u03c0\u03bf \u03b5\u03c0\u03b9\u03ba\u03bf\u03b9\u03bd\u03c9\u03bd\u03af\u03b1\u03c2. \u03a3\u03c4\u03b1 matched telecom tasks \u03c4\u03b7\u03c2 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1\u03c2, \u03bf GPT-4o \u03bc\u03b5 cheatsheet \u03b4\u03b9\u03b1\u03c4\u03ae\u03c1\u03b7\u03c3\u03b5 \u03c4\u03bf \u03c3\u03cd\u03bd\u03c4\u03bf\u03bc\u03bf, \u03c3\u03c4\u03b1\u03b4\u03b9\u03b1\u03ba\u03cc \u03c3\u03c4\u03b9\u03bb \u03c4\u03bf\u03c5, \u03b5\u03bd\u03ce \u03b7 direct execution \u03b1\u03c0\u03cc \u03c4\u03bf \u03b9\u03c3\u03c7\u03c5\u03c1\u03cc\u03c4\u03b5\u03c1\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03c0\u03b1\u03c1\u03ae\u03b3\u03b1\u03b3\u03b5 \u03c0\u03b9\u03bf \u03b5\u03ba\u03c4\u03b5\u03bd\u03ae \u03c0\u03c1\u03c9\u03c4\u03cc\u03ba\u03bf\u03bb\u03bb\u03b1.<\/p>\n<p>\u03a4\u03bf cheatsheet \u03b1\u03bb\u03bb\u03ac\u03b6\u03b5\u03b9 \u03c4\u03b7 \u03c3\u03b5\u03b9\u03c1\u03ac \u03ba\u03b1\u03b9 \u03c4\u03b7\u03bd \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae \u03b5\u03bd\u03b5\u03c1\u03b3\u03b5\u03b9\u03ce\u03bd, \u03cc\u03c7\u03b9 \u03b1\u03bd\u03b1\u03b3\u03ba\u03b1\u03c3\u03c4\u03b9\u03ba\u03ac \u03c4\u03b7 \u03c6\u03c9\u03bd\u03ae \u03c4\u03bf\u03c5 executor. \u0393\u03b9\u03b1 \u03ad\u03bd\u03b1 brand \u03b1\u03c5\u03c4\u03cc \u03b5\u03af\u03bd\u03b1\u03b9 \u03bf\u03c5\u03c3\u03b9\u03b1\u03c3\u03c4\u03b9\u03ba\u03cc. \u0397 \u03b2\u03b5\u03bb\u03c4\u03af\u03c9\u03c3\u03b7 \u03c4\u03b7\u03c2 \u03ba\u03ac\u03bb\u03c5\u03c8\u03b7\u03c2 \u03b5\u03bd\u03cc\u03c2 customer-service agent \u03b4\u03b5\u03bd \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03bd\u03b1 \u03bc\u03b5\u03c4\u03b1\u03c4\u03c1\u03ad\u03c8\u03b5\u03b9 \u03cc\u03bb\u03b5\u03c2 \u03c4\u03b9\u03c2 \u03b1\u03c0\u03b1\u03bd\u03c4\u03ae\u03c3\u03b5\u03b9\u03c2 \u03c3\u03b5 \u03bc\u03b1\u03ba\u03c1\u03bf\u03c3\u03ba\u03b5\u03bb\u03ae \u03c4\u03b5\u03c7\u03bd\u03b9\u03ba\u03ac manuals. \u0397 \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae \u03c4\u03bf\u03c5 executor \u03c0\u03b1\u03c1\u03b1\u03bc\u03ad\u03bd\u03b5\u03b9 \u03bc\u03bf\u03c7\u03bb\u03cc\u03c2 \u03b3\u03b9\u03b1 tone, latency \u03ba\u03b1\u03b9 interaction design, \u03b5\u03bd\u03ce \u03bf reflector \u03bb\u03b5\u03b9\u03c4\u03bf\u03c5\u03c1\u03b3\u03b5\u03af \u03c9\u03c2 \u03bc\u03b7\u03c7\u03b1\u03bd\u03b9\u03c3\u03bc\u03cc\u03c2 \u03b5\u03c0\u03af\u03bb\u03c5\u03c3\u03b7\u03c2 \u03b4\u03cd\u03c3\u03ba\u03bf\u03bb\u03c9\u03bd \u03c0\u03b5\u03c1\u03b9\u03c0\u03c4\u03ce\u03c3\u03b5\u03c9\u03bd.<\/p>\n<p>\u039f \u03b4\u03b9\u03b1\u03c7\u03c9\u03c1\u03b9\u03c3\u03bc\u03cc\u03c2 \u03b1\u03c5\u03c4\u03cc\u03c2 \u03b4\u03b5\u03bd \u03b5\u03be\u03b1\u03bb\u03b5\u03af\u03c6\u03b5\u03b9 \u03c4\u03b7\u03bd \u03b1\u03bd\u03ac\u03b3\u03ba\u03b7 \u03b3\u03b9\u03b1 UX testing. \u0388\u03bd\u03b1 agent \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03bf\u03bb\u03bf\u03ba\u03bb\u03b7\u03c1\u03ce\u03bd\u03b5\u03b9 \u03c3\u03c9\u03c3\u03c4\u03ac \u03c4\u03b7\u03bd \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 \u03b1\u03bb\u03bb\u03ac \u03bd\u03b1 \u03b6\u03b7\u03c4\u03ac \u03b1\u03c3\u03b1\u03c6\u03ae confirmations, \u03bd\u03b1 \u03b1\u03c0\u03bf\u03ba\u03b1\u03bb\u03cd\u03c0\u03c4\u03b5\u03b9 \u03b5\u03c3\u03c9\u03c4\u03b5\u03c1\u03b9\u03ba\u03ae \u03bf\u03c1\u03bf\u03bb\u03bf\u03b3\u03af\u03b1 \u03ae \u03bd\u03b1 \u03b4\u03b7\u03bc\u03b9\u03bf\u03c5\u03c1\u03b3\u03b5\u03af \u03c0\u03b5\u03c1\u03b9\u03c4\u03c4\u03ae \u03c4\u03c1\u03b9\u03b2\u03ae. \u0393\u03b9\u03b1 \u03b1\u03c5\u03c4\u03cc \u03b7 \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7 \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03ba\u03b1\u03bb\u03cd\u03c0\u03c4\u03b5\u03b9 outcome, \u03b1\u03c3\u03c6\u03ac\u03bb\u03b5\u03b9\u03b1 \u03ba\u03b1\u03b9 \u03c3\u03c5\u03bd\u03bf\u03bc\u03b9\u03bb\u03b9\u03b1\u03ba\u03ae \u03b5\u03bc\u03c0\u03b5\u03b9\u03c1\u03af\u03b1, \u03cc\u03c0\u03c9\u03c2 \u03ba\u03b1\u03b9 \u03c3\u03c4\u03b7\u03bd \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae <a href=\"https:\/\/twodots.gr\/6-ai-agents-customer-support-gia-epicheiriseis\/\">AI \u0430\u0433\u0435\u043d\u0442 \u0437\u0430 \u043e\u0431\u0441\u043b\u0443\u0436\u0432\u0430\u043d\u0435 \u043d\u0430 \u043a\u043b\u0438\u0435\u043d\u0442\u0438<\/a>.<\/p>\n<h2 id=\"shortcut-audit\">\u039f \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf\u03c2 \u03b3\u03b9\u03b1 shortcuts \u03ba\u03b1\u03b9 \u03b4\u03b9\u03b1\u03c1\u03c1\u03bf\u03ae<\/h2>\n<p>\u03a4\u03bf \u03ba\u03c1\u03af\u03c3\u03b9\u03bc\u03bf \u03c1\u03af\u03c3\u03ba\u03bf \u03b5\u03af\u03bd\u03b1\u03b9 \u03ad\u03bd\u03b1 cheatsheet \u03bd\u03b1 \u03b4\u03ce\u03c3\u03b5\u03b9 \u03c3\u03c4\u03bf\u03bd teacher \u03c3\u03c4\u03bf\u03b9\u03c7\u03b5\u03af\u03b1 \u03c0\u03bf\u03c5 \u03ba\u03b1\u03bd\u03bf\u03bd\u03b9\u03ba\u03ac \u03ad\u03c0\u03c1\u03b5\u03c0\u03b5 \u03bd\u03b1 \u03b1\u03bd\u03b1\u03ba\u03b1\u03bb\u03cd\u03c8\u03b5\u03b9 \u03bc\u03ad\u03c3\u03c9 \u03b5\u03c1\u03b3\u03b1\u03bb\u03b5\u03af\u03c9\u03bd. \u03a4\u03cc\u03c4\u03b5 \u03b7 recovered trajectory \u03b8\u03b1 \u03ad\u03bc\u03bf\u03b9\u03b1\u03b6\u03b5 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b7 \u03b1\u03bb\u03bb\u03ac \u03b8\u03b1 \u03b4\u03af\u03b4\u03b1\u03c3\u03ba\u03b5 \u03c3\u03c4\u03bf\u03bd student \u03bd\u03b1 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03b5\u03af magic values, \u03bc\u03b5\u03bb\u03bb\u03bf\u03bd\u03c4\u03b9\u03ba\u03ae \u03c0\u03bb\u03b7\u03c1\u03bf\u03c6\u03bf\u03c1\u03af\u03b1 \u03ae \u03bc\u03b7 \u03bd\u03cc\u03bc\u03b9\u03bc\u03b5\u03c2 \u03c0\u03b1\u03c1\u03b1\u03ba\u03ac\u03bc\u03c8\u03b5\u03b9\u03c2.<\/p>\n<p>\u039f\u03b9 \u03c3\u03c5\u03b3\u03b3\u03c1\u03b1\u03c6\u03b5\u03af\u03c2 \u03ad\u03bb\u03b5\u03b3\u03be\u03b1\u03bd recovered trajectories \u03bc\u03b5 \u03b4\u03cd\u03bf \u03b1\u03bd\u03b5\u03be\u03ac\u03c1\u03c4\u03b7\u03c4\u03bf\u03c5\u03c2 judges \u03b3\u03b9\u03b1 hardcoded values, shortcuts \u03ba\u03b1\u03b9 \u03c7\u03c1\u03ae\u03c3\u03b7 \u03bc\u03b5\u03bb\u03bb\u03bf\u03bd\u03c4\u03b9\u03ba\u03ae\u03c2 \u03c0\u03bb\u03b7\u03c1\u03bf\u03c6\u03bf\u03c1\u03af\u03b1\u03c2 \u03ba\u03b1\u03b9 \u03b5\u03be\u03ad\u03c4\u03b1\u03c3\u03b1\u03bd \u03c7\u03b5\u03b9\u03c1\u03bf\u03ba\u03af\u03bd\u03b7\u03c4\u03b1 \u03c4\u03b9\u03c2 \u03b1\u03c0\u03bf\u03ba\u03bb\u03af\u03c3\u03b5\u03b9\u03c2. \u03a4\u03bf audit \u03b4\u03b5\u03bd \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03b5\u03af \u03b1\u03c0\u03cc\u03bb\u03c5\u03c4\u03b7 \u03b5\u03b3\u03b3\u03cd\u03b7\u03c3\u03b7 \u03b3\u03b9\u03b1 \u03ba\u03ac\u03b8\u03b5 \u03bd\u03ad\u03bf dataset. \u03a5\u03c0\u03bf\u03b4\u03b5\u03b9\u03ba\u03bd\u03cd\u03b5\u03b9 \u03cc\u03bc\u03c9\u03c2 \u03c4\u03b7 \u03c3\u03c9\u03c3\u03c4\u03ae \u03b1\u03c1\u03c7\u03b9\u03c4\u03b5\u03ba\u03c4\u03bf\u03bd\u03b9\u03ba\u03ae: \u03ba\u03ac\u03b8\u03b5 tool argument \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b1\u03c0\u03bf\u03b4\u03bf\u03b8\u03b5\u03af \u03c3\u03c4\u03bf \u03b1\u03af\u03c4\u03b7\u03bc\u03b1, \u03c3\u03b5 \u03c0\u03c1\u03bf\u03b7\u03b3\u03bf\u03cd\u03bc\u03b5\u03bd\u03bf tool return \u03ae \u03c3\u03b5 \u03bd\u03cc\u03bc\u03b9\u03bc\u03bf schema default.<\/p>\n<p>\u0397 \u03af\u03b4\u03b9\u03b1 \u03b1\u03c0\u03b1\u03af\u03c4\u03b7\u03c3\u03b7 \u03b9\u03c3\u03c7\u03cd\u03b5\u03b9 \u03ba\u03b1\u03b9 \u03bc\u03b5\u03c4\u03ac \u03c4\u03b7\u03bd \u03b5\u03ba\u03c0\u03b1\u03af\u03b4\u03b5\u03c5\u03c3\u03b7. Logs, evaluator results \u03ba\u03b1\u03b9 policy decisions \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 lineage \u03ce\u03c3\u03c4\u03b5 \u03b7 \u03bf\u03bc\u03ac\u03b4\u03b1 \u03bd\u03b1 \u03be\u03ad\u03c1\u03b5\u03b9 \u03b3\u03b9\u03b1\u03c4\u03af \u03ad\u03bd\u03b1 trace \u03c7\u03b1\u03c1\u03b1\u03ba\u03c4\u03b7\u03c1\u03af\u03c3\u03c4\u03b7\u03ba\u03b5 \u03b1\u03c0\u03bf\u03b4\u03b5\u03ba\u03c4\u03cc. \u0397 \u03b1\u03bd\u03ac\u03b3\u03ba\u03b7 \u03b1\u03c5\u03c4\u03ae \u03c3\u03c5\u03bd\u03b4\u03ad\u03b5\u03c4\u03b1\u03b9 \u03bc\u03b5 \u03c4\u03bf <a href=\"https:\/\/twodots.gr\/airep-apodeiktiko-apofaseon-ai-runtime\/\">\u03b1\u03c0\u03bf\u03b4\u03b5\u03b9\u03ba\u03c4\u03b9\u03ba\u03cc \u03ba\u03ac\u03b8\u03b5 \u03b1\u03c0\u03cc\u03c6\u03b1\u03c3\u03b7\u03c2 \u03b5\u03bd\u03cc\u03c2 AI runtime<\/a> \u03ba\u03b1\u03b9 \u03bc\u03b5 \u03c4\u03b7 \u03b4\u03b9\u03b1\u03ba\u03c5\u03b2\u03ad\u03c1\u03bd\u03b7\u03c3\u03b7 \u03c4\u03c9\u03bd \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03c9\u03bd \u03c0\u03bf\u03c5 \u03bc\u03c0\u03b1\u03af\u03bd\u03bf\u03c5\u03bd \u03c3\u03c4\u03bf training set.<\/p>\n<aside class=\"td-article-note\"><strong>\u03a0\u03c1\u03b1\u03ba\u03c4\u03b9\u03ba\u03cc red flag:<\/strong> \u03b1\u03bd \u03bc\u03b9\u03b1 recovered trajectory \u03c0\u03b5\u03c1\u03b9\u03ad\u03c7\u03b5\u03b9 \u03c4\u03b9\u03bc\u03ae \u03c0\u03bf\u03c5 \u03b4\u03b5\u03bd \u03b5\u03bc\u03c6\u03b1\u03bd\u03af\u03b6\u03b5\u03c4\u03b1\u03b9 \u03c3\u03c4\u03bf user input, \u03c3\u03b5 tool return \u03ae \u03c3\u03b5 \u03b5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03bf default, \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 training example\u00b7 \u03b5\u03af\u03bd\u03b1\u03b9 \u03c0\u03b9\u03b8\u03b1\u03bd\u03ae \u03b4\u03b9\u03b1\u03c1\u03c1\u03bf\u03ae \u03c0\u03bf\u03c5 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc\u03c1\u03c1\u03b9\u03c8\u03b7 \u03ae \u03bd\u03ad\u03b1 \u03b5\u03ba\u03c4\u03ad\u03bb\u03b5\u03c3\u03b7.<\/aside>\n<h2 id=\"production-agents\">\u03a4\u03b9 \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u03b3\u03b9\u03b1 production AI agents<\/h2>\n<p>\u03a3\u03c4\u03bf production \u03c3\u03cd\u03c3\u03c4\u03b7\u03bc\u03b1 \u03c0\u03bf\u03c5 \u03c0\u03b5\u03c1\u03b9\u03b3\u03c1\u03ac\u03c6\u03b5\u03b9 \u03b7 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1, \u03c4\u03bf PROOF-Gen \u03b5\u03c6\u03b1\u03c1\u03bc\u03cc\u03c3\u03c4\u03b7\u03ba\u03b5 \u03c3\u03b5 \u03c7\u03b9\u03bb\u03b9\u03ac\u03b4\u03b5\u03c2 \u03c3\u03b5\u03bd\u03ac\u03c1\u03b9\u03b1, \u03bc\u03b5 \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b5\u03c2 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ad\u03c2 \u03ba\u03b1\u03b9 \u03c0\u03bf\u03bb\u03bb\u03b1\u03c0\u03bb\u03bf\u03cd\u03c2 evaluators. \u039f\u03b9 \u03c3\u03c5\u03b3\u03b3\u03c1\u03b1\u03c6\u03b5\u03af\u03c2 \u03b1\u03bd\u03b1\u03c6\u03ad\u03c1\u03bf\u03c5\u03bd \u03ac\u03bd\u03bf\u03b4\u03bf 6,3 \u03c0\u03bf\u03c3\u03bf\u03c3\u03c4\u03b9\u03b1\u03af\u03c9\u03bd \u03bc\u03bf\u03bd\u03ac\u03b4\u03c9\u03bd \u03c3\u03c4\u03bf goal completion \u03c4\u03b7\u03c2 \u03c0\u03bf\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1\u03c2 trajectories. \u03a3\u03b5 deployed on-device \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03b1\u03bd\u03b1\u03c6\u03ad\u03c1\u03bf\u03c5\u03bd +1,5 \u03bc\u03bf\u03bd\u03ac\u03b4\u03b1 \u03c3\u03c4\u03bf goal completion, \u03b2\u03b5\u03bb\u03c4\u03b9\u03ce\u03c3\u03b5\u03b9\u03c2 +1,7 \u03ad\u03c9\u03c2 +5,0 \u03bc\u03bf\u03bd\u03ac\u03b4\u03b5\u03c2 \u03c3\u03b5 response-quality metrics \u03ba\u03b1\u03b9 \u03bc\u03ad\u03c3\u03b7 \u03b2\u03b5\u03bb\u03c4\u03af\u03c9\u03c3\u03b7 +1,48 \u03bc\u03bf\u03bd\u03ac\u03b4\u03b1 \u03c3\u03c4\u03b9\u03c2 \u03bc\u03b7 \u03b1\u03b3\u03b3\u03bb\u03b9\u03ba\u03ad\u03c2 \u03b3\u03bb\u03ce\u03c3\u03c3\u03b5\u03c2.<\/p>\n<p>\u039f\u03b9 \u03b1\u03c1\u03b9\u03b8\u03bc\u03bf\u03af \u03c0\u03b5\u03c1\u03b9\u03b3\u03c1\u03ac\u03c6\u03bf\u03c5\u03bd \u03c4\u03b7 \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03b7 \u03b5\u03c1\u03b5\u03c5\u03bd\u03b7\u03c4\u03b9\u03ba\u03ae \u03ba\u03b1\u03b9 \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03b9\u03ba\u03ae \u03b5\u03b3\u03ba\u03b1\u03c4\u03ac\u03c3\u03c4\u03b1\u03c3\u03b7. \u0394\u03b5\u03bd \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03bf\u03cd\u03bd \u03c0\u03c1\u03cc\u03b2\u03bb\u03b5\u03c8\u03b7 \u03b3\u03b9\u03b1 \u03ac\u03bb\u03bb\u03b7 \u03b5\u03c4\u03b1\u03b9\u03c1\u03b5\u03af\u03b1. \u0397 \u03bc\u03b5\u03c4\u03b1\u03c6\u03ad\u03c1\u03c3\u03b9\u03bc\u03b7 \u03b9\u03b4\u03ad\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03cc\u03c4\u03b9 \u03c4\u03b1 failure logs \u03bc\u03c0\u03bf\u03c1\u03bf\u03cd\u03bd \u03bd\u03b1 \u03b3\u03af\u03bd\u03bf\u03c5\u03bd \u03bf\u03c5\u03c1\u03ac \u03c5\u03c0\u03bf\u03c8\u03b7\u03c6\u03af\u03c9\u03bd training examples, \u03b5\u03c6\u03cc\u03c3\u03bf\u03bd \u03c5\u03c0\u03ac\u03c1\u03c7\u03bf\u03c5\u03bd \u03c0\u03bb\u03ae\u03c1\u03b7 traces, \u03b1\u03be\u03b9\u03cc\u03c0\u03b9\u03c3\u03c4\u03bf\u03b9 verifiers, \u03c3\u03b1\u03c6\u03ad\u03c2 provenance \u03ba\u03b1\u03b9 \u03b1\u03c5\u03c3\u03c4\u03b7\u03c1\u03ae \u03b1\u03c6\u03b1\u03af\u03c1\u03b5\u03c3\u03b7 \u03c4\u03bf\u03c5 scaffold.<\/p>\n<p>\u03a3\u03b5 e-commerce, \u03c5\u03c0\u03bf\u03c8\u03ae\u03c6\u03b9\u03b1 cases \u03b5\u03af\u03bd\u03b1\u03b9 \u03b7 \u03b1\u03bb\u03bb\u03b1\u03b3\u03ae \u03b4\u03b9\u03b5\u03cd\u03b8\u03c5\u03bd\u03c3\u03b7\u03c2 \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03c4\u03b7\u03bd \u03b1\u03c0\u03bf\u03c3\u03c4\u03bf\u03bb\u03ae, \u03b7 \u03bc\u03b5\u03c1\u03b9\u03ba\u03ae \u03b5\u03c0\u03b9\u03c3\u03c4\u03c1\u03bf\u03c6\u03ae, \u03b7 \u03b1\u03ba\u03cd\u03c1\u03c9\u03c3\u03b7 \u03bc\u03b5 \u03c0\u03bf\u03bb\u03b9\u03c4\u03b9\u03ba\u03ad\u03c2 \u03b5\u03be\u03b1\u03b9\u03c1\u03ad\u03c3\u03b5\u03c9\u03bd, \u03bf \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf\u03c2 \u03b4\u03b9\u03b1\u03b8\u03b5\u03c3\u03b9\u03bc\u03cc\u03c4\u03b7\u03c4\u03b1\u03c2 \u03ae \u03ad\u03bd\u03b1\u03c2 \u03c3\u03c5\u03bd\u03b4\u03c5\u03b1\u03c3\u03bc\u03cc\u03c2 CRM \u03ba\u03b1\u03b9 logistics. \u03a4\u03bf \u03c3\u03c9\u03c3\u03c4\u03cc unit \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7\u03c2 \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03c0\u03cc\u03c3\u03b5\u03c2 API calls \u03ae\u03c4\u03b1\u03bd \u03bf\u03c1\u03b8\u03ad\u03c2, \u03b1\u03bb\u03bb\u03ac \u03b1\u03bd \u03c4\u03bf \u03c3\u03cd\u03c3\u03c4\u03b7\u03bc\u03b1 \u03c0\u03ad\u03c4\u03c5\u03c7\u03b5 \u03c4\u03b7\u03bd \u03b5\u03c0\u03b9\u03b8\u03c5\u03bc\u03b7\u03c4\u03ae \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03b7\u03c3\u03b9\u03b1\u03ba\u03ae \u03ba\u03b1\u03c4\u03ac\u03c3\u03c4\u03b1\u03c3\u03b7 \u03c7\u03c9\u03c1\u03af\u03c2 \u03c0\u03b1\u03c1\u03b1\u03b2\u03af\u03b1\u03c3\u03b7 \u03c0\u03bf\u03bb\u03b9\u03c4\u03b9\u03ba\u03ae\u03c2.<\/p>\n<p>\u0397 \u03c0\u03c1\u03bf\u03c4\u03b5\u03c1\u03b1\u03b9\u03bf\u03c0\u03bf\u03af\u03b7\u03c3\u03b7 \u03c4\u03c9\u03bd failures \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03ba\u03b1\u03b9 \u03bf\u03b9\u03ba\u03bf\u03bd\u03bf\u03bc\u03b9\u03ba\u03cc \u03c6\u03af\u03bb\u03c4\u03c1\u03bf. \u0394\u03b5\u03bd \u03b1\u03be\u03af\u03b6\u03b5\u03b9 \u03ba\u03ac\u03b8\u03b5 \u03c3\u03c0\u03ac\u03bd\u03b9\u03bf edge case \u03c4\u03bf \u03af\u03b4\u03b9\u03bf recovery budget. \u0397 \u03bf\u03bc\u03ac\u03b4\u03b1 \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03c3\u03c5\u03bd\u03b4\u03ad\u03b5\u03b9 \u03c3\u03c5\u03c7\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1, \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2 \u03bb\u03ac\u03b8\u03bf\u03c5\u03c2, \u03c1\u03c5\u03b8\u03bc\u03b9\u03c3\u03c4\u03b9\u03ba\u03cc \u03ba\u03af\u03bd\u03b4\u03c5\u03bd\u03bf \u03ba\u03b1\u03b9 \u03c0\u03b9\u03b8\u03b1\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b1\u03bd\u03ac\u03ba\u03c4\u03b7\u03c3\u03b7\u03c2, \u03cc\u03c0\u03c9\u03c2 \u03c3\u03b5 \u03bc\u03b9\u03b1 \u03c3\u03bf\u03b2\u03b1\u03c1\u03ae \u03b1\u03bd\u03ac\u03bb\u03c5\u03c3\u03b7 <a href=\"https:\/\/twodots.gr\/ai-roi-axia-ergasias\/\">AI ROI \u03b1\u03c0\u03cc \u03c4\u03b1 tokens \u03c3\u03c4\u03b7\u03bd \u03b1\u03be\u03af\u03b1 \u03c4\u03b7\u03c2 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1\u03c2<\/a>.<\/p>\n<h2 id=\"praktiko-protokollo\">\u0395\u03c0\u03c4\u03ac \u03b2\u03ae\u03bc\u03b1\u03c4\u03b1 \u03b3\u03b9\u03b1 \u03b5\u03bb\u03ad\u03b3\u03be\u03b9\u03bc\u03bf recovery pipeline<\/h2>\n<p>\u0397 \u03b5\u03c6\u03b1\u03c1\u03bc\u03bf\u03b3\u03ae \u03b4\u03b5\u03bd \u03be\u03b5\u03ba\u03b9\u03bd\u03ac \u03b1\u03c0\u03cc \u03c4\u03bf fine-tuning \u03b1\u03bb\u03bb\u03ac \u03b1\u03c0\u03cc \u03c4\u03b7\u03bd \u03c0\u03b1\u03c1\u03b1\u03c4\u03b7\u03c1\u03b7\u03c3\u03b9\u03bc\u03cc\u03c4\u03b7\u03c4\u03b1 \u03ba\u03b1\u03b9 \u03c4\u03bf\u03c5\u03c2 verifiers. \u03a4\u03bf \u03b1\u03ba\u03cc\u03bb\u03bf\u03c5\u03b8\u03bf \u03c0\u03c1\u03c9\u03c4\u03cc\u03ba\u03bf\u03bb\u03bb\u03bf \u03bc\u03b5\u03c4\u03b1\u03c6\u03ad\u03c1\u03b5\u03b9 \u03c4\u03b7\u03bd \u03b9\u03b4\u03ad\u03b1 \u03c3\u03b5 \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03b7\u03c3\u03b9\u03b1\u03ba\u03bf\u03cd\u03c2 agents \u03c7\u03c9\u03c1\u03af\u03c2 \u03bd\u03b1 \u03b8\u03b5\u03c9\u03c1\u03b5\u03af \u03ba\u03ac\u03b8\u03b5 failure \u03ba\u03b1\u03c4\u03ac\u03bb\u03bb\u03b7\u03bb\u03bf \u03b3\u03b9\u03b1 training.<\/p>\n<div class=\"td-step-list\">\n<p class=\"td-step-list-title\">\u0391\u03c0\u03cc \u03c4\u03bf failure log \u03c3\u03b5 \u03b5\u03c0\u03b1\u03bb\u03b7\u03b8\u03b5\u03c5\u03bc\u03ad\u03bd\u03bf training example<\/p>\n<ol>\n<li><span class=\"td-step-kicker\">\u0421\u0442\u044a\u043f\u043a\u0430 1<\/span><strong>\u039a\u03b1\u03c4\u03b1\u03b3\u03c1\u03ac\u03c8\u03c4\u03b5 \u03c0\u03bb\u03ae\u03c1\u03b5\u03c2 execution trace<\/strong>\n<p>\u0391\u03c0\u03bf\u03b8\u03b7\u03ba\u03b5\u03cd\u03c3\u03c4\u03b5 \u03b1\u03af\u03c4\u03b7\u03bc\u03b1, model \u03ba\u03b1\u03b9 version, system prompt, tool inputs \u03ba\u03b1\u03b9 outputs, timestamps, evaluator results \u03ba\u03b1\u03b9 \u03c4\u03b5\u03bb\u03b9\u03ba\u03ae \u03ba\u03b1\u03c4\u03ac\u03c3\u03c4\u03b1\u03c3\u03b7. \u03a7\u03c9\u03c1\u03af\u03c2 \u03c0\u03bb\u03ae\u03c1\u03b5\u03c2 trace \u03bf reflector \u03b4\u03b5\u03bd \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b5\u03bd\u03c4\u03bf\u03c0\u03af\u03c3\u03b5\u03b9 \u03c4\u03b7 \u03c1\u03af\u03b6\u03b1 \u03c4\u03bf\u03c5 \u03bb\u03ac\u03b8\u03bf\u03c5\u03c2.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u0421\u0442\u044a\u043f\u043a\u0430 2<\/span><strong>\u039f\u03c1\u03af\u03c3\u03c4\u03b5 programmatic end-to-end verifiers<\/strong>\n<p>\u0395\u03bb\u03ad\u03b3\u03be\u03c4\u03b5 \u03b1\u03bd \u03b5\u03bd\u03b7\u03bc\u03b5\u03c1\u03ce\u03b8\u03b7\u03ba\u03b5 \u03b7 \u03c3\u03c9\u03c3\u03c4\u03ae \u03b5\u03b3\u03b3\u03c1\u03b1\u03c6\u03ae, \u03b5\u03c6\u03b1\u03c1\u03bc\u03cc\u03c3\u03c4\u03b7\u03ba\u03b5 \u03b7 \u03c0\u03bf\u03bb\u03b9\u03c4\u03b9\u03ba\u03ae, \u03b6\u03b7\u03c4\u03ae\u03b8\u03b7\u03ba\u03b5 \u03c4\u03bf \u03b1\u03c0\u03b1\u03b9\u03c4\u03bf\u03cd\u03bc\u03b5\u03bd\u03bf confirmation \u03ba\u03b1\u03b9 \u03c0\u03c1\u03bf\u03ad\u03ba\u03c5\u03c8\u03b5 \u03b7 \u03b5\u03c0\u03b9\u03b8\u03c5\u03bc\u03b7\u03c4\u03ae business state. \u039c\u03b7\u03bd \u03b1\u03c1\u03ba\u03b5\u03c3\u03c4\u03b5\u03af\u03c4\u03b5 \u03c3\u03b5 judge \u03c4\u03b7\u03c2 \u03c4\u03b5\u03bb\u03b9\u03ba\u03ae\u03c2 \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7\u03c2.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u0421\u0442\u044a\u043f\u043a\u0430 3<\/span><strong>\u03a4\u03b1\u03be\u03b9\u03bd\u03bf\u03bc\u03ae\u03c3\u03c4\u03b5 failure cause \u03ba\u03b1\u03b9 business risk<\/strong>\n<p>\u0394\u03b9\u03b1\u03c7\u03c9\u03c1\u03af\u03c3\u03c4\u03b5 missing tool call, \u03bb\u03ac\u03b8\u03bf\u03c2 \u03c3\u03b5\u03b9\u03c1\u03ac, \u03b5\u03c3\u03c6\u03b1\u03bb\u03bc\u03ad\u03bd\u03bf argument, policy violation, verifier ambiguity \u03ba\u03b1\u03b9 tool outage. \u0394\u03ce\u03c3\u03c4\u03b5 \u03c0\u03c1\u03bf\u03c4\u03b5\u03c1\u03b1\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1 \u03c3\u03b5 \u03c3\u03c5\u03c7\u03bd\u03ac \u03ae \u03c5\u03c8\u03b7\u03bb\u03bf\u03cd \u03ba\u03cc\u03c3\u03c4\u03bf\u03c5\u03c2 cases \u03c0\u03bf\u03c5 \u03bc\u03c0\u03bf\u03c1\u03bf\u03cd\u03bd \u03bd\u03b1 \u03b1\u03bd\u03b1\u03c0\u03b1\u03c1\u03b1\u03c7\u03b8\u03bf\u03cd\u03bd.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u0421\u0442\u044a\u043f\u043a\u0430 4<\/span><strong>\u0391\u03c0\u03bf\u03bc\u03bf\u03bd\u03ce\u03c3\u03c4\u03b5 \u03c4\u03bf \u03c0\u03c1\u03bf\u03c3\u03c9\u03c1\u03b9\u03bd\u03cc scaffold<\/strong>\n<p>\u0392\u03ac\u03bb\u03c4\u03b5 \u03c3\u03b1\u03c6\u03ae delimiters \u03c3\u03c4\u03bf per-scenario cheatsheet, \u03bc\u03b7\u03bd \u03c4\u03bf \u03b5\u03bc\u03c6\u03b1\u03bd\u03af\u03b6\u03b5\u03c4\u03b5 \u03c3\u03c4\u03bf\u03bd \u03c7\u03c1\u03ae\u03c3\u03c4\u03b7 \u03ba\u03b1\u03b9 \u03bc\u03b7\u03bd \u03c4\u03bf \u03ba\u03b1\u03c4\u03b1\u03b3\u03c1\u03ac\u03c6\u03b5\u03c4\u03b5 \u03c9\u03c2 \u03bc\u03ad\u03c1\u03bf\u03c2 \u03c4\u03b7\u03c2 \u03c4\u03b5\u03bb\u03b9\u03ba\u03ae\u03c2 demonstration. \u0397 guidance \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03b4\u03b9\u03b1\u03b4\u03b9\u03ba\u03b1\u03c3\u03c4\u03b9\u03ba\u03ae, \u03cc\u03c7\u03b9 \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7 \u03bc\u03b5 \u03ba\u03c1\u03c5\u03c6\u03ad\u03c2 \u03c4\u03b9\u03bc\u03ad\u03c2.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u0421\u0442\u044a\u043f\u043a\u0430 5<\/span><strong>\u0395\u03c0\u03b1\u03bd\u03b5\u03ba\u03c4\u03b5\u03bb\u03ad\u03c3\u03c4\u03b5 \u03b1\u03c0\u03cc \u03ba\u03b1\u03b8\u03b1\u03c1\u03ae \u03b1\u03c1\u03c7\u03b9\u03ba\u03ae \u03ba\u03b1\u03c4\u03ac\u03c3\u03c4\u03b1\u03c3\u03b7<\/strong>\n<p>\u039c\u03b7\u03bd \u03c3\u03c5\u03bd\u03b5\u03c7\u03af\u03b6\u03b5\u03c4\u03b5 \u03b1\u03c0\u03cc \u03bc\u03b9\u03c3\u03bf\u03b1\u03bb\u03bb\u03b1\u03b3\u03bc\u03ad\u03bd\u03b7 \u03b2\u03ac\u03c3\u03b7. \u0395\u03c0\u03b1\u03bd\u03b1\u03c6\u03ad\u03c1\u03b5\u03c4\u03b5 sandbox, fixtures \u03ba\u03b1\u03b9 conversation state \u03ce\u03c3\u03c4\u03b5 \u03b7 \u03bd\u03ad\u03b1 trajectory \u03bd\u03b1 \u03b1\u03c0\u03bf\u03b4\u03b5\u03b9\u03ba\u03bd\u03cd\u03b5\u03b9 \u03cc\u03c4\u03b9 \u03bf\u03bb\u03bf\u03ba\u03bb\u03b7\u03c1\u03ce\u03bd\u03b5\u03b9 \u03b1\u03c5\u03c4\u03cc\u03bd\u03bf\u03bc\u03b1 \u03c4\u03b7\u03bd \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u0421\u0442\u044a\u043f\u043a\u0430 6<\/span><strong>\u0395\u03bb\u03ad\u03b3\u03be\u03c4\u03b5 provenance \u03ba\u03b1\u03b9 \u03b1\u03c6\u03b1\u03b9\u03c1\u03ad\u03c3\u03c4\u03b5 \u03c4\u03bf cheatsheet<\/strong>\n<p>\u0391\u03c0\u03bf\u03b4\u03b5\u03af\u03be\u03c4\u03b5 \u03c4\u03b7\u03bd \u03c0\u03b7\u03b3\u03ae \u03ba\u03ac\u03b8\u03b5 tool argument, \u03c4\u03c1\u03ad\u03be\u03c4\u03b5 \u03cc\u03bb\u03bf\u03c5\u03c2 \u03c4\u03bf\u03c5\u03c2 verifiers \u03ba\u03b1\u03b9 \u03b1\u03c6\u03b1\u03b9\u03c1\u03ad\u03c3\u03c4\u03b5 \u03c0\u03bb\u03ae\u03c1\u03c9\u03c2 \u03c4\u03b7\u03bd \u03c0\u03c1\u03bf\u03c3\u03c9\u03c1\u03b9\u03bd\u03ae \u03ba\u03b1\u03b8\u03bf\u03b4\u03ae\u03b3\u03b7\u03c3\u03b7 \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03c4\u03bf dataset export. \u0391\u03c0\u03bf\u03c1\u03c1\u03af\u03c8\u03c4\u03b5 shortcuts \u03ba\u03b1\u03b9 magic values.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">\u0421\u0442\u044a\u043f\u043a\u0430 7<\/span><strong>\u039a\u03c1\u03b1\u03c4\u03ae\u03c3\u03c4\u03b5 holdout release gate<\/strong>\n<p>\u039c\u03b5\u03c4\u03c1\u03ae\u03c3\u03c4\u03b5 goal completion, policy compliance, latency, \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2 \u03ba\u03b1\u03b9 \u03b5\u03bc\u03c0\u03b5\u03b9\u03c1\u03af\u03b1 \u03c7\u03c1\u03ae\u03c3\u03c4\u03b7 \u03c3\u03b5 unseen tasks. \u0391\u03bd \u03b2\u03b5\u03bb\u03c4\u03b9\u03ce\u03bd\u03b5\u03c4\u03b1\u03b9 \u03bc\u03cc\u03bd\u03bf \u03b7 tool-call accuracy, \u03c4\u03bf \u03bd\u03ad\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03b4\u03b5\u03bd \u03c0\u03b5\u03c1\u03bd\u03ac \u03c3\u03b5 production.<\/p>\n<\/li>\n<\/ol>\n<\/div>\n<p>\u03a4\u03bf \u03c0\u03c1\u03c9\u03c4\u03cc\u03ba\u03bf\u03bb\u03bb\u03bf \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03c0\u03b5\u03c1\u03b9\u03bf\u03c1\u03b9\u03c3\u03bc\u03bf\u03cd\u03c2 \u03c0\u03c1\u03cc\u03c3\u03b2\u03b1\u03c3\u03b7\u03c2 \u03ba\u03b1\u03b9 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ae sandboxes. \u0388\u03bd\u03b1 recovery loop \u03c0\u03bf\u03c5 \u03b5\u03c0\u03b1\u03bd\u03b1\u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03b5\u03b9 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ad\u03c2 \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b5\u03c2 \u03c7\u03c9\u03c1\u03af\u03c2 isolation \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b4\u03b7\u03bc\u03b9\u03bf\u03c5\u03c1\u03b3\u03ae\u03c3\u03b5\u03b9 \u03b4\u03b9\u03c0\u03bb\u03ad\u03c2 \u03c0\u03b1\u03c1\u03b1\u03b3\u03b3\u03b5\u03bb\u03af\u03b5\u03c2, \u03bb\u03b1\u03bd\u03b8\u03b1\u03c3\u03bc\u03ad\u03bd\u03b5\u03c2 \u03b5\u03c0\u03b9\u03c3\u03c4\u03c1\u03bf\u03c6\u03ad\u03c2 \u03ae \u03bc\u03b7 \u03b1\u03bd\u03b1\u03c3\u03c4\u03c1\u03ad\u03c8\u03b9\u03bc\u03b5\u03c2 \u03b1\u03bb\u03bb\u03b1\u03b3\u03ad\u03c2. \u0397 \u03b1\u03c1\u03c7\u03b9\u03c4\u03b5\u03ba\u03c4\u03bf\u03bd\u03b9\u03ba\u03ae \u03b5\u03bb\u03ac\u03c7\u03b9\u03c3\u03c4\u03b7\u03c2 \u03c0\u03c1\u03cc\u03c3\u03b2\u03b1\u03c3\u03b7\u03c2 \u03b3\u03b9\u03b1 \u03c4\u03ad\u03c4\u03bf\u03b9\u03b1 \u03c3\u03c5\u03c3\u03c4\u03ae\u03bc\u03b1\u03c4\u03b1 \u03b1\u03bd\u03b1\u03bb\u03cd\u03b5\u03c4\u03b1\u03b9 \u03c3\u03c4\u03bf \u03ac\u03c1\u03b8\u03c1\u03bf \u03b3\u03b9\u03b1 <a href=\"https:\/\/twodots.gr\/dynamika-dikaiomata-ai-agents-elachisti-prosvasi\/\">\u03b4\u03c5\u03bd\u03b1\u03bc\u03b9\u03ba\u03ac \u03b4\u03b9\u03ba\u03b1\u03b9\u03ce\u03bc\u03b1\u03c4\u03b1 AI agents<\/a>.<\/p>\n<h2 id=\"oria-kostos\">\u038c\u03c1\u03b9\u03b1, \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2 \u03ba\u03b1\u03b9 governance<\/h2>\n<p>\u0397 \u03bc\u03b5\u03bb\u03ad\u03c4\u03b7 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03af\u03b7\u03c3\u03b5 \u03ad\u03bd\u03b1\u03bd teacher, \u03c4\u03bf\u03bd GPT-4o, \u03ba\u03b1\u03b9 reflectors \u03b1\u03c0\u03cc \u03c4\u03bf\u03bd \u03af\u03b4\u03b9\u03bf provider. \u0394\u03b5\u03bd \u03ad\u03c7\u03b5\u03b9 \u03b1\u03c0\u03bf\u03b4\u03b5\u03b9\u03c7\u03b8\u03b5\u03af \u03cc\u03c4\u03b9 \u03b7 \u03af\u03b4\u03b9\u03b1 \u03c3\u03c7\u03ad\u03c3\u03b7 reflector quality \u03ba\u03b1\u03b9 recovery \u03b9\u03c3\u03c7\u03cd\u03b5\u03b9 \u03bc\u03b5 \u03ba\u03ac\u03b8\u03b5 \u03bf\u03b9\u03ba\u03bf\u03b3\u03ad\u03bd\u03b5\u03b9\u03b1 \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03c9\u03bd \u03ae open-weight \u03c3\u03cd\u03c3\u03c4\u03b7\u03bc\u03b1. \u03a4\u03b1 cheatsheets \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03bc\u03b5\u03b3\u03ac\u03bb\u03b1, \u03ac\u03c1\u03b1 \u03bf executor \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03b1\u03c1\u03ba\u03b5\u03c4\u03cc context \u03ba\u03b1\u03b9 \u03b1\u03be\u03b9\u03cc\u03c0\u03b9\u03c3\u03c4\u03b7 instruction following.<\/p>\n<p>\u03a4\u03b1 benchmarks \u03ad\u03c7\u03bf\u03c5\u03bd \u03b5\u03c0\u03af\u03c3\u03b7\u03c2 \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03b7 \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03b1 \u03ba\u03b1\u03b9 \u03c3\u03c4\u03bf\u03c7\u03b1\u03c3\u03c4\u03b9\u03ba\u03cc\u03c4\u03b7\u03c4\u03b1. \u03a3\u03c4\u03bf BFCL \u03bf\u03b9 \u03b5\u03c0\u03b9\u03bc\u03ad\u03c1\u03bf\u03c5\u03c2 \u03ba\u03b1\u03c4\u03b7\u03b3\u03bf\u03c1\u03af\u03b5\u03c2 \u03c0\u03b5\u03c1\u03b9\u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03bf\u03c5\u03bd \u03c0\u03b5\u03c1\u03b9\u03bf\u03c1\u03b9\u03c3\u03bc\u03ad\u03bd\u03bf \u03b1\u03c1\u03b9\u03b8\u03bc\u03cc test tasks, \u03b5\u03bd\u03ce \u03c4\u03bf \u03c42-bench \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03b5\u03af stochastic user simulator. \u0393\u03b9\u03b1 \u03b1\u03c5\u03c4\u03cc \u03c4\u03bf paper \u03b4\u03b9\u03b1\u03c7\u03c9\u03c1\u03af\u03b6\u03b5\u03b9 \u03c4\u03bf headline 93% \u03b1\u03c0\u03cc \u03c4\u03bf 88% \u03c0\u03bf\u03c5 \u03b1\u03c0\u03bf\u03b4\u03af\u03b4\u03b5\u03c4\u03b1\u03b9 \u03ac\u03bc\u03b5\u03c3\u03b1 \u03c3\u03c4\u03bf cheatsheet. \u039a\u03ac\u03b8\u03b5 \u03bf\u03bc\u03ac\u03b4\u03b1 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03b5\u03c0\u03b1\u03bd\u03b1\u03bb\u03ae\u03c8\u03b5\u03b9\u03c2 \u03ba\u03b1\u03b9 confidence intervals, \u03cc\u03c7\u03b9 \u03bc\u03cc\u03bd\u03bf \u03ad\u03bd\u03b1 recovery percentage.<\/p>\n<p>\u03a4\u03bf PROOF-Gen \u03b4\u03b5\u03bd \u03b5\u03be\u03b1\u03c6\u03b1\u03bd\u03af\u03b6\u03b5\u03b9 \u03c4\u03bf \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2. \u03a0\u03c1\u03bf\u03c3\u03b8\u03ad\u03c4\u03b5\u03b9 reflection, \u03b5\u03c0\u03b1\u03bd\u03b5\u03ba\u03c4\u03b5\u03bb\u03ad\u03c3\u03b5\u03b9\u03c2, verifier compute, human audit \u03ba\u03b1\u03b9 dataset governance. \u0397 \u03b1\u03be\u03af\u03b1 \u03c4\u03bf\u03c5 \u03b5\u03be\u03b1\u03c1\u03c4\u03ac\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03c4\u03bf \u03b1\u03bd \u03b7 \u03b1\u03bd\u03ac\u03ba\u03c4\u03b7\u03c3\u03b7 \u03b4\u03cd\u03c3\u03ba\u03bf\u03bb\u03c9\u03bd cases \u03b5\u03af\u03bd\u03b1\u03b9 \u03c6\u03b8\u03b7\u03bd\u03cc\u03c4\u03b5\u03c1\u03b7 \u03ba\u03b1\u03b9 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03cc\u03c4\u03b5\u03c1\u03b7 \u03b1\u03c0\u03cc \u03c0\u03b5\u03c1\u03b9\u03c3\u03c3\u03cc\u03c4\u03b5\u03c1\u03b1 \u03b5\u03cd\u03ba\u03bf\u03bb\u03b1 traces \u03ae \u03b1\u03c0\u03cc direct execution \u03bc\u03b5 \u03b9\u03c3\u03c7\u03c5\u03c1\u03cc\u03c4\u03b5\u03c1\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03c3\u03b5 \u03ba\u03ac\u03b8\u03b5 turn.<\/p>\n<p>\u03a4\u03ad\u03bb\u03bf\u03c2, \u03ad\u03bd\u03b1\u03c2 \u03ba\u03b1\u03ba\u03cc\u03c2 verifier \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b2\u03b5\u03bb\u03c4\u03b9\u03c3\u03c4\u03bf\u03c0\u03bf\u03b9\u03ae\u03c3\u03b5\u03b9 \u03c4\u03bf \u03bb\u03ac\u03b8\u03bf\u03c2 \u03c0\u03c1\u03ac\u03b3\u03bc\u03b1. \u0391\u03bd \u03bf \u03c3\u03c4\u03cc\u03c7\u03bf\u03c2 \u03b5\u03c0\u03b9\u03b2\u03c1\u03b1\u03b2\u03b5\u03cd\u03b5\u03b9 \u03bc\u03cc\u03bd\u03bf \u03c4\u03b7 \u03bc\u03b5\u03c4\u03b1\u03b2\u03bf\u03bb\u03ae \u03c4\u03b7\u03c2 \u03b2\u03ac\u03c3\u03b7\u03c2 \u03ba\u03b1\u03b9 \u03b1\u03b3\u03bd\u03bf\u03b5\u03af \u03c3\u03c5\u03bd\u03b1\u03af\u03bd\u03b5\u03c3\u03b7, \u03c0\u03bf\u03bb\u03b9\u03c4\u03b9\u03ba\u03ae \u03ae \u03c0\u03bf\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7\u03c2, \u03c4\u03bf recovery loop \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03c0\u03b1\u03c1\u03ac\u03b3\u03b5\u03b9 \u03c3\u03c5\u03bd\u03b5\u03c0\u03b5\u03af\u03c2 \u03b1\u03bb\u03bb\u03ac \u03bc\u03b7 \u03b1\u03c0\u03bf\u03b4\u03b5\u03ba\u03c4\u03ad\u03c2 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ad\u03c2. \u0397 governance \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03c3\u03c5\u03bd\u03b4\u03ad\u03b5\u03b9 context, measurement \u03ba\u03b1\u03b9 risk management, \u03cc\u03c7\u03b9 \u03bd\u03b1 \u03b5\u03bc\u03c6\u03b1\u03bd\u03af\u03b6\u03b5\u03c4\u03b1\u03b9 \u03c3\u03c4\u03bf \u03c4\u03ad\u03bb\u03bf\u03c2 \u03c9\u03c2 compliance checklist. \u0397 \u03af\u03b4\u03b9\u03b1 \u03b1\u03c1\u03c7\u03ae \u03b2\u03c1\u03af\u03c3\u03ba\u03b5\u03c4\u03b1\u03b9 \u03c3\u03c4\u03b9\u03c2 <a href=\"https:\/\/twodots.gr\/granite-trust-policy-tools-politikes-ai-safety\/\">\u03b5\u03ba\u03c4\u03b5\u03bb\u03ad\u03c3\u03b9\u03bc\u03b5\u03c2 \u03c0\u03bf\u03bb\u03b9\u03c4\u03b9\u03ba\u03ad\u03c2 AI safety<\/a>.<\/p>\n<h2 id=\"apotychia-signal\">\u0397 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b1 \u03c9\u03c2 training signal<\/h2>\n<p>\u03a4\u03bf PROOF-Gen \u03bc\u03b5\u03c4\u03b1\u03ba\u03b9\u03bd\u03b5\u03af \u03c4\u03bf bottleneck \u03c4\u03bf\u03c5 distillation \u03c0\u03c1\u03bf\u03c2 \u03c4\u03b1 \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03b1. \u0388\u03bd\u03b1\u03c2 \u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03bf\u03c2 optimizer \u03b4\u03b5\u03bd \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03bc\u03ac\u03b8\u03b5\u03b9 \u03c4\u03b7 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ae \u03bc\u03ad\u03c3\u03b1 \u03b1\u03c0\u03cc \u03c4\u03bf \u03ba\u03c1\u03af\u03c3\u03b9\u03bc\u03bf decision point \u03b1\u03bd \u03c4\u03bf training set \u03c4\u03b7\u03bd \u03ad\u03c7\u03b5\u03b9 \u03ae\u03b4\u03b7 \u03c0\u03b5\u03c4\u03ac\u03be\u03b5\u03b9. \u03a0\u03c1\u03ce\u03c4\u03b1 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 \u03c0\u03bb\u03ae\u03c1\u03b5\u03b9\u03c2, \u03b5\u03c0\u03b1\u03bb\u03b7\u03b8\u03b5\u03c5\u03bc\u03ad\u03bd\u03b5\u03c2 demonstrations \u03b3\u03b9\u03b1 \u03c4\u03b9\u03c2 \u03c0\u03b5\u03c1\u03b9\u03c0\u03c4\u03ce\u03c3\u03b5\u03b9\u03c2 \u03c0\u03bf\u03c5 \u03bf teacher \u03b4\u03b5\u03bd \u03ad\u03bb\u03c5\u03c3\u03b5 \u03c3\u03c4\u03b7\u03bd \u03c0\u03c1\u03ce\u03c4\u03b7 \u03c0\u03c1\u03bf\u03c3\u03c0\u03ac\u03b8\u03b5\u03b9\u03b1.<\/p>\n<p>\u0393\u03b9\u03b1 \u03c4\u03b9\u03c2 \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03ae\u03c3\u03b5\u03b9\u03c2, \u03c4\u03bf \u03bc\u03ae\u03bd\u03c5\u03bc\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03c0\u03c1\u03b1\u03ba\u03c4\u03b9\u03ba\u03cc: \u03ad\u03bd\u03b1 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03bf workflow \u03c0\u03b5\u03c1\u03b9\u03ad\u03c7\u03b5\u03b9 \u03c3\u03c5\u03c7\u03bd\u03ac \u03bc\u03b5\u03b3\u03ac\u03bb\u03bf \u03bc\u03ad\u03c1\u03bf\u03c2 \u03c4\u03b7\u03c2 \u03c3\u03c9\u03c3\u03c4\u03ae\u03c2 \u03b4\u03b9\u03b1\u03b4\u03b9\u03ba\u03b1\u03c3\u03af\u03b1\u03c2 \u03ba\u03b1\u03b9 \u03ad\u03bd\u03b1 \u03c0\u03bf\u03bb\u03cd\u03c4\u03b9\u03bc\u03bf \u03c3\u03b7\u03bc\u03ac\u03b4\u03b9 \u03b3\u03b9\u03b1 \u03c4\u03bf \u03ba\u03b5\u03bd\u03cc \u03c0\u03bf\u03c5 \u03bc\u03ad\u03bd\u03b5\u03b9. \u039c\u03b5 \u03b1\u03be\u03b9\u03cc\u03c0\u03b9\u03c3\u03c4\u03bf\u03c5\u03c2 verifiers, \u03b9\u03c3\u03c7\u03c5\u03c1\u03cc reflection, \u03ba\u03b1\u03b8\u03b1\u03c1\u03cc provenance, \u03b1\u03c6\u03b1\u03af\u03c1\u03b5\u03c3\u03b7 \u03c4\u03bf\u03c5 scaffold \u03ba\u03b1\u03b9 holdout \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7, \u03b1\u03c5\u03c4\u03cc \u03c4\u03bf \u03ba\u03b5\u03bd\u03cc \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b3\u03af\u03bd\u03b5\u03b9 \u03bd\u03ad\u03bf training signal.<\/p>\n<p>\u0397 \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03af\u03b1 \u03b4\u03b5\u03bd \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03bc\u03c0\u03b5\u03b9 \u03b1\u03c5\u03c4\u03bf\u03cd\u03c3\u03b9\u03b1 \u03c3\u03c4\u03b7\u03bd \u03b5\u03ba\u03c0\u03b1\u03af\u03b4\u03b5\u03c5\u03c3\u03b7 \u03ba\u03b1\u03b9 \u03b4\u03b5\u03bd \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03c1\u03bf\u03bc\u03b1\u03bd\u03c4\u03b9\u03ba\u03bf\u03c0\u03bf\u03b9\u03b7\u03b8\u03b5\u03af. \u03a0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03bc\u03b5\u03c4\u03b1\u03c4\u03c1\u03b1\u03c0\u03b5\u03af \u03c3\u03b5 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b7, \u03b1\u03bd\u03b1\u03c0\u03b1\u03c1\u03b1\u03b3\u03ce\u03b3\u03b9\u03bc\u03b7 \u03ba\u03b1\u03b9 \u03b5\u03bb\u03ad\u03b3\u03be\u03b9\u03bc\u03b7 \u03b5\u03c0\u03af\u03b4\u03b5\u03b9\u03be\u03b7. \u0391\u03c5\u03c4\u03ae \u03b5\u03af\u03bd\u03b1\u03b9 \u03b7 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03ac \u03b1\u03bd\u03ac\u03bc\u03b5\u03c3\u03b1 \u03c3\u03b5 \u03ad\u03bd\u03b1 log \u03c0\u03bf\u03c5 \u03b1\u03c0\u03bb\u03ce\u03c2 \u03b5\u03be\u03b7\u03b3\u03b5\u03af \u03c4\u03b9 \u03c0\u03ae\u03b3\u03b5 \u03c3\u03c4\u03c1\u03b1\u03b2\u03ac \u03ba\u03b1\u03b9 \u03c3\u03b5 \u03ad\u03bd\u03b1 dataset \u03c0\u03bf\u03c5 \u03ba\u03ac\u03bd\u03b5\u03b9 \u03c4\u03bf\u03bd \u03b5\u03c0\u03cc\u03bc\u03b5\u03bd\u03bf agent \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac \u03c0\u03b9\u03bf \u03b1\u03be\u03b9\u03cc\u03c0\u03b9\u03c3\u03c4\u03bf.<\/p>\n<section class=\"td-service-cta\">\n<div class=\"td-service-cta-content\">\n<p class=\"td-service-cta-kicker\">AI agents \u03bc\u03b5 \u03b5\u03bb\u03ad\u03b3\u03be\u03b9\u03bc\u03b1 outcomes<\/p>\n<p class=\"td-service-cta-title\">\u039c\u03b5\u03c4\u03b1\u03c4\u03c1\u03ad\u03c8\u03c4\u03b5 \u03c4\u03b1 failure logs \u03c3\u03b5 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ad\u03c2 \u03c0\u03bb\u03ac\u03bd\u03bf \u03b2\u03b5\u03bb\u03c4\u03af\u03c9\u03c3\u03b7\u03c2<\/p>\n<p>\u0397 TWO DOTS \u03c3\u03c7\u03b5\u03b4\u03b9\u03ac\u03b6\u03b5\u03b9 \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03b7\u03c3\u03b9\u03b1\u03ba\u03bf\u03cd\u03c2 AI \u03b1\u03c5\u03c4\u03bf\u03bc\u03b1\u03c4\u03b9\u03c3\u03bc\u03bf\u03cd\u03c2 \u03bc\u03b5 \u03c0\u03bb\u03ae\u03c1\u03b7 traces, programmatic verifiers, \u03c0\u03b5\u03c1\u03b9\u03bf\u03c1\u03b9\u03c3\u03bc\u03ad\u03bd\u03b1 tool permissions, evaluation gates \u03ba\u03b1\u03b9 human review. \u0388\u03c4\u03c3\u03b9 \u03ba\u03ac\u03b8\u03b5 recovery experiment \u03c3\u03c5\u03bd\u03b4\u03ad\u03b5\u03c4\u03b1\u03b9 \u03bc\u03b5 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03cc business outcome \u03ba\u03b1\u03b9 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b5\u03bb\u03b5\u03b3\u03c7\u03b8\u03b5\u03af \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03c4\u03bf production.<\/p>\n<div class=\"td-service-cta-actions\"><a class=\"td-service-cta-button\" href=\"https:\/\/twodots.gr\/aftomatismoi-epicheiriseon-ai\/\">\u0394\u03b5\u03af\u03c4\u03b5 \u0391\u03c5\u03c4\u03bf\u03bc\u03b1\u03c4\u03b9\u03c3\u03bc\u03bf\u03cd\u03c2 \u0395\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03ae\u03c3\u03b5\u03c9\u03bd &amp; AI<\/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 \u03c4\u03bf PROOF-Gen;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u0395\u03af\u03bd\u03b1\u03b9 \u03bc\u03ad\u03b8\u03bf\u03b4\u03bf\u03c2 \u03b1\u03bd\u03ac\u03ba\u03c4\u03b7\u03c3\u03b7\u03c2 \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03c9\u03bd tool-calling trajectories \u03b1\u03c0\u03cc \u03b1\u03c0\u03bf\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b5\u03c2 \u03b5\u03ba\u03c4\u03b5\u03bb\u03ad\u03c3\u03b5\u03b9\u03c2, \u03bc\u03b5 reflective prompt optimization \u03be\u03b5\u03c7\u03c9\u03c1\u03b9\u03c3\u03c4\u03ac \u03b3\u03b9\u03b1 \u03ba\u03ac\u03b8\u03b5 \u03c3\u03b5\u03bd\u03ac\u03c1\u03b9\u03bf.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u0395\u03ba\u03c0\u03b1\u03b9\u03b4\u03b5\u03cd\u03b5\u03c4\u03b1\u03b9 \u03bf student \u03c0\u03ac\u03bd\u03c9 \u03c3\u03c4\u03bf cheatsheet;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u038c\u03c7\u03b9. \u0397 \u03c0\u03c1\u03bf\u03c3\u03c9\u03c1\u03b9\u03bd\u03ae \u03ba\u03b1\u03b8\u03bf\u03b4\u03ae\u03b3\u03b7\u03c3\u03b7 \u03b1\u03c6\u03b1\u03b9\u03c1\u03b5\u03af\u03c4\u03b1\u03b9 \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03c4\u03bf supervised fine-tuning \u03ba\u03b1\u03b9 \u03bf student \u03b2\u03bb\u03ad\u03c0\u03b5\u03b9 \u03bc\u03cc\u03bd\u03bf \u03c4\u03b7\u03bd \u03ba\u03b1\u03b8\u03b1\u03c1\u03ae, \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b7 trajectory.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u0393\u03b9\u03b1\u03c4\u03af \u03b4\u03b5\u03bd \u03b1\u03c1\u03ba\u03b5\u03af \u03b7 tool-call accuracy;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u0395\u03c0\u03b5\u03b9\u03b4\u03ae \u03c0\u03bf\u03bb\u03bb\u03ad\u03c2 \u03c3\u03c9\u03c3\u03c4\u03ad\u03c2 \u03b5\u03bd\u03b4\u03b9\u03ac\u03bc\u03b5\u03c3\u03b5\u03c2 \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b5\u03c2 \u03bc\u03c0\u03bf\u03c1\u03bf\u03cd\u03bd \u03bd\u03b1 \u03c3\u03c5\u03bd\u03c5\u03c0\u03ac\u03c1\u03c7\u03bf\u03c5\u03bd \u03bc\u03b5 \u03bb\u03ac\u03b8\u03bf\u03c2 \u03c4\u03b5\u03bb\u03b9\u03ba\u03ae \u03ba\u03b1\u03c4\u03ac\u03c3\u03c4\u03b1\u03c3\u03b7. \u03a4\u03bf release gate \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03bc\u03b5\u03c4\u03c1\u03ac end-to-end goal completion \u03ba\u03b1\u03b9 policy compliance.<\/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 recovery rate;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u03a3\u03c4\u03bf \u03b4\u03b5\u03af\u03b3\u03bc\u03b1 300 telecom failures \u03b1\u03bd\u03b1\u03ba\u03c4\u03ae\u03b8\u03b7\u03ba\u03b1\u03bd 279, \u03b4\u03b7\u03bb\u03b1\u03b4\u03ae 93%. \u039f\u03b9 264 \u03b1\u03c0\u03bf\u03b4\u03cc\u03b8\u03b7\u03ba\u03b1\u03bd \u03c3\u03c4\u03bf cheatsheet \u03ba\u03b1\u03b9 \u03bf\u03b9 15 \u03c3\u03b5 stochastic replay \u03c4\u03bf\u03c5 user simulator.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u039c\u03c0\u03bf\u03c1\u03b5\u03af \u03c4\u03bf PROOF-Gen \u03bd\u03b1 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03b7\u03b8\u03b5\u03af \u03c3\u03b5 \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03b7\u03c3\u03b9\u03b1\u03ba\u03bf\u03cd\u03c2 agents;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u0397 \u03bb\u03bf\u03b3\u03b9\u03ba\u03ae \u03b5\u03c6\u03b1\u03c1\u03bc\u03cc\u03b6\u03b5\u03c4\u03b1\u03b9 \u03cc\u03c0\u03bf\u03c5 \u03c5\u03c0\u03ac\u03c1\u03c7\u03bf\u03c5\u03bd \u03c0\u03bb\u03ae\u03c1\u03b7 traces, \u03b1\u03be\u03b9\u03cc\u03c0\u03b9\u03c3\u03c4\u03bf\u03b9 verifiers, \u03b1\u03c3\u03c6\u03b1\u03bb\u03ad\u03c2 \u03c0\u03b5\u03c1\u03b9\u03b2\u03ac\u03bb\u03bb\u03bf\u03bd \u03b5\u03c0\u03b1\u03bd\u03b5\u03ba\u03c4\u03ad\u03bb\u03b5\u03c3\u03b7\u03c2 \u03ba\u03b1\u03b9 \u03b4\u03c5\u03bd\u03b1\u03c4\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b5\u03bb\u03ad\u03b3\u03c7\u03bf\u03c5 \u03c4\u03b7\u03c2 \u03c4\u03b5\u03bb\u03b9\u03ba\u03ae\u03c2 \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03b7\u03c3\u03b9\u03b1\u03ba\u03ae\u03c2 \u03ba\u03b1\u03c4\u03ac\u03c3\u03c4\u03b1\u03c3\u03b7\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 \u03b2\u03b1\u03c3\u03b9\u03ba\u03cc \u03c1\u03af\u03c3\u03ba\u03bf \u03c3\u03c4\u03b1 recovered traces;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u039d\u03b1 \u03b5\u03b9\u03c3\u03b1\u03c7\u03b8\u03bf\u03cd\u03bd shortcuts, magic values \u03ae \u03bc\u03b5\u03bb\u03bb\u03bf\u03bd\u03c4\u03b9\u03ba\u03ae \u03c0\u03bb\u03b7\u03c1\u03bf\u03c6\u03bf\u03c1\u03af\u03b1. \u039a\u03ac\u03b8\u03b5 tool argument \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03b1\u03c0\u03bf\u03b4\u03b5\u03b4\u03b5\u03b9\u03b3\u03bc\u03ad\u03bd\u03b7 \u03c0\u03c1\u03bf\u03ad\u03bb\u03b5\u03c5\u03c3\u03b7 \u03b1\u03c0\u03cc \u03c4\u03bf \u03b1\u03af\u03c4\u03b7\u03bc\u03b1, tool return \u03ae \u03b5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03bf default.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u0391\u03bd\u03c4\u03b9\u03ba\u03b1\u03b8\u03b9\u03c3\u03c4\u03ac \u03c4\u03bf PROOF-Gen \u03c4\u03bf fine-tuning;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u038c\u03c7\u03b9. \u0395\u03af\u03bd\u03b1\u03b9 \u03c3\u03c4\u03ac\u03b4\u03b9\u03bf \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03ae\u03c2 \u03ba\u03b1\u03b9 \u03b1\u03bd\u03ac\u03ba\u03c4\u03b7\u03c3\u03b7\u03c2 \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03c9\u03bd \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03c4\u03bf fine-tuning. \u03a3\u03c4\u03b7 \u03bc\u03b5\u03bb\u03ad\u03c4\u03b7, \u03c4\u03bf downstream training \u03ad\u03b3\u03b9\u03bd\u03b5 \u03bc\u03b5 QLoRA \u03b3\u03b9\u03b1 \u03c4\u03c1\u03af\u03b1 epochs.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a0\u03bf\u03b9\u03b5\u03c2 \u03bc\u03b5\u03c4\u03c1\u03ae\u03c3\u03b5\u03b9\u03c2 \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03c0\u03b1\u03c1\u03b1\u03ba\u03bf\u03bb\u03bf\u03c5\u03b8\u03b5\u03af \u03bc\u03b9\u03b1 \u03b5\u03c4\u03b1\u03b9\u03c1\u03b5\u03af\u03b1;<\/summary>\n<div class=\"td-faq-content\">\n<p>Goal completion, \u03c3\u03c9\u03c3\u03c4\u03ae \u03c4\u03b5\u03bb\u03b9\u03ba\u03ae \u03ba\u03b1\u03c4\u03ac\u03c3\u03c4\u03b1\u03c3\u03b7, policy compliance, provenance, \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2, latency \u03ba\u03b1\u03b9 \u03b5\u03bc\u03c0\u03b5\u03b9\u03c1\u03af\u03b1 \u03c7\u03c1\u03ae\u03c3\u03c4\u03b7. \u0397 tool-call accuracy \u03c0\u03b1\u03c1\u03b1\u03bc\u03ad\u03bd\u03b5\u03b9 \u03b2\u03bf\u03b7\u03b8\u03b7\u03c4\u03b9\u03ba\u03ae \u03b4\u03b9\u03b1\u03b3\u03bd\u03c9\u03c3\u03c4\u03b9\u03ba\u03ae \u03bc\u03ad\u03c4\u03c1\u03b7\u03c3\u03b7.<\/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\/2608.23911\" target=\"_blank\" rel=\"noopener\">Ta, Zhu &amp; Shayandeh \u2014 PROOF-Gen: From Optimized Data to Better Distillation<\/a><\/li>\n<li><a href=\"https:\/\/arxiv.org\/abs\/2506.07982\" target=\"_blank\" rel=\"noopener\">Barres et al. \u2014 \u03c42-bench: Evaluating Conversational Agents in a Dual-Control Environment<\/a><\/li>\n<li><a href=\"https:\/\/proceedings.mlr.press\/v267\/patil25a.html\" target=\"_blank\" rel=\"noopener\">Patil et al. \u2014 The Berkeley Function Calling Leaderboard (BFCL)<\/a><\/li>\n<li><a href=\"https:\/\/arxiv.org\/abs\/2507.19457\" target=\"_blank\" rel=\"noopener\">Agrawal et al. \u2014 GEPA: Reflective Prompt Evolution Can Outperform Reinforcement Learning<\/a><\/li>\n<li><a href=\"https:\/\/airc.nist.gov\/airmf-resources\/airmf\/5-sec-core\/\" target=\"_blank\" rel=\"noopener\">NIST AI RMF \u2014 Core: Govern, Map, Measure and Manage<\/a><\/li>\n<li><a href=\"https:\/\/www.nist.gov\/publications\/artificial-intelligence-risk-management-framework-generative-artificial-intelligence\" target=\"_blank\" rel=\"noopener\">NIST \u2014 Generative Artificial Intelligence Profile<\/a><\/li>\n<\/ul>\n<\/div>","protected":false},"excerpt":{"rendered":"<p>\u03a4\u03bf PROOF-Gen \u03b1\u03bd\u03b1\u03ba\u03c4\u03ac \u03b5\u03c0\u03b9\u03c4\u03c5\u03c7\u03b7\u03bc\u03ad\u03bd\u03b5\u03c2 \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ad\u03c2 \u03b1\u03c0\u03cc failures AI agents \u03ba\u03b1\u03b9 \u03c4\u03b9\u03c2 \u03bc\u03b5\u03c4\u03b1\u03c4\u03c1\u03ad\u03c0\u03b5\u03b9 \u03c3\u03b5 \u03ba\u03b1\u03b8\u03b1\u03c1\u03ac \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03b1 \u03b3\u03b9\u03b1 \u03c0\u03b9\u03bf \u03b1\u03c3\u03c6\u03b1\u03bb\u03ad\u03c2 model distillation \u03bc\u03b5 \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf.<\/p>","protected":false},"author":1,"featured_media":105874,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"_gspb_post_css":"","content-type":"","footnotes":""},"categories":[199],"tags":[6506,20890,20889,20572,20891],"class_list":["post-98245","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-techniti-noimosyni","tag-ai-agents","tag-model-distillation","tag-proof-gen","tag-synthetic-data","tag-tool-calling"],"blocksy_meta":{"styles_descriptor":{"styles":{"desktop":"","tablet":"","mobile":""},"google_fonts":[],"version":8}},"_links":{"self":[{"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/posts\/98245","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=98245"}],"version-history":[{"count":1,"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/posts\/98245\/revisions"}],"predecessor-version":[{"id":105875,"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/posts\/98245\/revisions\/105875"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/media\/105874"}],"wp:attachment":[{"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/media?parent=98245"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/categories?post=98245"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/twodots.gr\/bg\/wp-json\/wp\/v2\/tags?post=98245"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}