{"id":92980,"date":"2026-09-07T14:46:56","date_gmt":"2026-09-07T11:46:56","guid":{"rendered":"https:\/\/twodots.gr\/?p=92980"},"modified":"2026-09-07T14:46:57","modified_gmt":"2026-09-07T11:46:57","slug":"direag-ai-den-empistevetai-tyfla-vevaiotita","status":"publish","type":"post","link":"https:\/\/twodots.gr\/en\/direag-ai-den-empistevetai-tyfla-vevaiotita\/","title":{"rendered":"DirEAG: \u03c0\u03ce\u03c2 \u03b7 AI \u03bc\u03b1\u03b8\u03b1\u03af\u03bd\u03b5\u03b9 \u03bd\u03b1 \u03bc\u03b7\u03bd \u03b5\u03bc\u03c0\u03b9\u03c3\u03c4\u03b5\u03cd\u03b5\u03c4\u03b1\u03b9 \u03c4\u03c5\u03c6\u03bb\u03ac \u03c4\u03b7 \u03b2\u03b5\u03b2\u03b1\u03b9\u03cc\u03c4\u03b7\u03c4\u03ac \u03c4\u03b7\u03c2"},"content":{"rendered":"<div class=\"td-article-lede\">\n<p><strong>\u03a4\u03bf DirEAG \u03b4\u03b5\u03bd \u03b1\u03bd\u03c4\u03b9\u03bc\u03b5\u03c4\u03c9\u03c0\u03af\u03b6\u03b5\u03b9 \u03c4\u03bf \u00ab95% \u03b2\u03b5\u03b2\u03b1\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1\u00bb \u03b5\u03bd\u03cc\u03c2 LLM \u03c9\u03c2 \u03ad\u03c4\u03bf\u03b9\u03bc\u03b7 \u03c0\u03b9\u03b8\u03b1\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1.<\/strong> \u03a3\u03c5\u03bd\u03b4\u03c5\u03ac\u03b6\u03b5\u03b9 \u03c0\u03ad\u03bd\u03c4\u03b5 answer-confidence observations, \u03b4\u03b9\u03bf\u03c1\u03b8\u03ce\u03bd\u03b5\u03b9 \u03c4\u03b7 \u03bc\u03b5\u03c1\u03bf\u03bb\u03b7\u03c8\u03af\u03b1 \u03ba\u03ac\u03b8\u03b5 steering prompt, \u03ba\u03c1\u03b1\u03c4\u03ac \u03c1\u03b7\u03c4\u03cc null state \u03b3\u03b9\u03b1 \u03c4\u03b7\u03bd \u03c0\u03b5\u03c1\u03af\u03c0\u03c4\u03c9\u03c3\u03b7 \u03cc\u03c0\u03bf\u03c5 \u03cc\u03bb\u03b5\u03c2 \u03bf\u03b9 \u03c5\u03c0\u03bf\u03c8\u03ae\u03c6\u03b9\u03b5\u03c2 \u03b1\u03c0\u03b1\u03bd\u03c4\u03ae\u03c3\u03b5\u03b9\u03c2 \u03b5\u03af\u03bd\u03b1\u03b9 \u03bb\u03ac\u03b8\u03bf\u03c2 \u03ba\u03b1\u03b9 \u03b2\u03b1\u03b8\u03bc\u03bf\u03bd\u03bf\u03bc\u03b5\u03af \u03be\u03b1\u03bd\u03ac \u03c4\u03bf \u03c4\u03b5\u03bb\u03b9\u03ba\u03cc score. \u03a4\u03b1 \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03ad\u03c3\u03bc\u03b1\u03c4\u03b1 \u03c3\u03b5 \u03c4\u03c1\u03af\u03b1 \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03b1 \u03ba\u03b1\u03b9 \u03c4\u03c1\u03af\u03b1 \u03bc\u03b1\u03b8\u03b7\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac benchmarks \u03b5\u03af\u03bd\u03b1\u03b9 \u03b5\u03bd\u03b8\u03b1\u03c1\u03c1\u03c5\u03bd\u03c4\u03b9\u03ba\u03ac, \u03b1\u03bb\u03bb\u03ac \u03b4\u03b5\u03bd \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03bf\u03cd\u03bd \u03ad\u03c4\u03bf\u03b9\u03bc\u03b7 \u03b5\u03b3\u03b3\u03cd\u03b7\u03c3\u03b7 \u03b3\u03b9\u03b1 production workflows \u03c7\u03c9\u03c1\u03af\u03c2 \u03b4\u03b9\u03ba\u03ac \u03c4\u03bf\u03c5\u03c2 labeled outcomes \u03ba\u03b1\u03b9 \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03b1 gates.<\/p>\n<\/div>\n<div class=\"td-article-toc\">\n<div class=\"td-toc-title\">Contents<\/div>\n<ul>\n<li><a href=\"#vevaiotita-llm-den-einai-pithanotita\">\u0393\u03b9\u03b1\u03c4\u03af \u03b7 \u03b2\u03b5\u03b2\u03b1\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b5\u03bd\u03cc\u03c2 LLM \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03b1\u03c0\u03cc \u03bc\u03cc\u03bd\u03b7 \u03c4\u03b7\u03c2 \u03c0\u03b9\u03b8\u03b1\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1<\/a><\/li>\n<li><a href=\"#symfonia-apantiseon-den-arkei\">\u0393\u03b9\u03b1\u03c4\u03af \u03b7 \u03c3\u03c5\u03bc\u03c6\u03c9\u03bd\u03af\u03b1 \u03bc\u03b5\u03c4\u03b1\u03be\u03cd \u03b1\u03c0\u03b1\u03bd\u03c4\u03ae\u03c3\u03b5\u03c9\u03bd \u03b4\u03b5\u03bd \u03b1\u03c1\u03ba\u03b5\u03af<\/a><\/li>\n<li><a href=\"#pente-prompts-biased-klimakes\">\u03a0\u03ad\u03bd\u03c4\u03b5 prompts \u03ba\u03b1\u03b9 \u03c0\u03ad\u03bd\u03c4\u03b5 \u03bc\u03b5\u03c1\u03bf\u03bb\u03b7\u03c0\u03c4\u03b9\u03ba\u03ad\u03c2 \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03b5\u03c2<\/a><\/li>\n<li><a href=\"#dirichlet-logistiko-vivlio-evidence\">\u0397 Dirichlet \u03c9\u03c2 \u03bb\u03bf\u03b3\u03b9\u03c3\u03c4\u03b9\u03ba\u03cc \u03b2\u03b9\u03b2\u03bb\u03af\u03bf \u03c4\u03b5\u03ba\u03bc\u03b7\u03c1\u03af\u03c9\u03bd<\/a><\/li>\n<li><a href=\"#null-state-kamia-sosti-apantisi\">\u039f null state \u03b4\u03af\u03bd\u03b5\u03b9 \u03c7\u03ce\u03c1\u03bf \u03c3\u03c4\u03bf \u00ab\u03ba\u03b1\u03bc\u03af\u03b1 \u03c3\u03c9\u03c3\u03c4\u03ae \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7\u00bb<\/a><\/li>\n<li><a href=\"#deytero-stadio-calibration\">\u0393\u03b9\u03b1\u03c4\u03af \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03b4\u03b5\u03cd\u03c4\u03b5\u03c1\u03bf \u03c3\u03c4\u03ac\u03b4\u03b9\u03bf calibration<\/a><\/li>\n<li><a href=\"#peiramata-montela-datasets-metrics\">\u03a0\u03ce\u03c2 \u03c3\u03c7\u03b5\u03b4\u03b9\u03ac\u03c3\u03c4\u03b7\u03ba\u03b1\u03bd \u03c4\u03b1 \u03c0\u03b5\u03b9\u03c1\u03ac\u03bc\u03b1\u03c4\u03b1<\/a><\/li>\n<li><a href=\"#apotelesmata-kai-sosti-anagnosi\">\u03a4\u03b9 \u03b4\u03b5\u03af\u03c7\u03bd\u03bf\u03c5\u03bd \u03c4\u03b1 \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03ad\u03c3\u03bc\u03b1\u03c4\u03b1 \u03ba\u03b1\u03b9 \u03c0\u03ce\u03c2 \u03b4\u03b9\u03b1\u03b2\u03ac\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9<\/a><\/li>\n<li><a href=\"#production-ai-workflow\">\u03a4\u03b9 \u03b1\u03bb\u03bb\u03ac\u03b6\u03b5\u03b9 \u03c3\u03b5 \u03ad\u03bd\u03b1 production AI workflow<\/a><\/li>\n<li><a href=\"#pilot-vathmonomisis-vevaiotitas\">\u03a0\u03ce\u03c2 \u03c3\u03c4\u03ae\u03bd\u03b5\u03c4\u03b1\u03b9 \u03ad\u03bd\u03b1 pilot \u03b2\u03b1\u03b8\u03bc\u03bf\u03bd\u03cc\u03bc\u03b7\u03c3\u03b7\u03c2 \u03b2\u03b5\u03b2\u03b1\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1\u03c2<\/a><\/li>\n<li><a href=\"#oria-erevnas-kai-genikeysis\">\u03a4\u03b1 \u03cc\u03c1\u03b9\u03b1 \u03c4\u03b7\u03c2 \u03ad\u03c1\u03b5\u03c5\u03bd\u03b1\u03c2 \u03ba\u03b1\u03b9 \u03c4\u03b7\u03c2 \u03b3\u03b5\u03bd\u03af\u03ba\u03b5\u03c5\u03c3\u03b7\u03c2<\/a><\/li>\n<li><a href=\"#empistosyni-me-dikaioma-amfivolias\">\u0395\u03bc\u03c0\u03b9\u03c3\u03c4\u03bf\u03c3\u03cd\u03bd\u03b7 \u03bc\u03b5 \u03b4\u03b9\u03ba\u03b1\u03af\u03c9\u03bc\u03b1 \u03c3\u03c4\u03b7\u03bd \u03b1\u03bc\u03c6\u03b9\u03b2\u03bf\u03bb\u03af\u03b1<\/a><\/li>\n<\/ul>\n<\/div>\n<p>\u0391\u03bd \u03ad\u03bd\u03b1 \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf AI \u03b4\u03b7\u03bb\u03ce\u03bd\u03b5\u03b9 \u03cc\u03c4\u03b9 \u03b5\u03af\u03bd\u03b1\u03b9 95% \u03b2\u03ad\u03b2\u03b1\u03b9\u03bf, \u03c0\u03cc\u03c3\u03bf \u03c3\u03c5\u03c7\u03bd\u03ac \u03ad\u03c7\u03b5\u03b9 \u03c0\u03c1\u03ac\u03b3\u03bc\u03b1\u03c4\u03b9 \u03b4\u03af\u03ba\u03b9\u03bf; \u0397 \u03b5\u03c1\u03ce\u03c4\u03b7\u03c3\u03b7 \u03b2\u03c1\u03af\u03c3\u03ba\u03b5\u03c4\u03b1\u03b9 \u03c3\u03c4\u03bf\u03bd \u03c0\u03c5\u03c1\u03ae\u03bd\u03b1 \u03ba\u03ac\u03b8\u03b5 \u03c3\u03bf\u03b2\u03b1\u03c1\u03ae\u03c2 \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03b7\u03bc\u03b1\u03c4\u03b9\u03ba\u03ae\u03c2 \u03c7\u03c1\u03ae\u03c3\u03b7\u03c2 \u03c4\u03b7\u03c2 \u03c4\u03b5\u03c7\u03bd\u03b7\u03c4\u03ae\u03c2 \u03bd\u03bf\u03b7\u03bc\u03bf\u03c3\u03cd\u03bd\u03b7\u03c2. \u039c\u03b9\u03b1 \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7 \u03b4\u03b5\u03bd \u03b3\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03b1\u03be\u03b9\u03bf\u03c0\u03bf\u03b9\u03ae\u03c3\u03b9\u03bc\u03b7 \u03b5\u03c0\u03b5\u03b9\u03b4\u03ae \u03c3\u03c5\u03bd\u03bf\u03b4\u03b5\u03cd\u03b5\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03b5\u03bd\u03c4\u03c5\u03c0\u03c9\u03c3\u03b9\u03b1\u03ba\u03cc \u03b1\u03c1\u03b9\u03b8\u03bc\u03cc. \u039f \u03b1\u03c1\u03b9\u03b8\u03bc\u03cc\u03c2 \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b1\u03bd\u03c4\u03b9\u03c3\u03c4\u03bf\u03b9\u03c7\u03b5\u03af \u03bc\u03b5 \u03c3\u03c5\u03bd\u03ad\u03c0\u03b5\u03b9\u03b1 \u03c3\u03c4\u03b7\u03bd \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ae \u03c0\u03b9\u03b8\u03b1\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1 \u03bf\u03c1\u03b8\u03cc\u03c4\u03b7\u03c4\u03b1\u03c2 \u03b3\u03b9\u03b1 \u03c4\u03bf \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03bf task \u03ba\u03b1\u03b9 workload.<\/p>\n<p>\u0397 \u03c0\u03c1\u03bf\u03b4\u03b7\u03bc\u03bf\u03c3\u03af\u03b5\u03c5\u03c3\u03b7 <strong>DirEAG<\/strong> \u03c4\u03c9\u03bd Haorui Xu, Yuzhou Zhu \u03ba\u03b1\u03b9 Liyuan Gao \u03b5\u03be\u03b5\u03c4\u03ac\u03b6\u03b5\u03b9 \u03b1\u03ba\u03c1\u03b9\u03b2\u03ce\u03c2 \u03b1\u03c5\u03c4\u03cc \u03c4\u03bf \u03c0\u03c1\u03cc\u03b2\u03bb\u03b7\u03bc\u03b1 \u03c3\u03b5 mathematical reasoning. \u0391\u03bd\u03c4\u03af \u03bd\u03b1 \u03b5\u03bc\u03c0\u03b9\u03c3\u03c4\u03b5\u03cd\u03b5\u03c4\u03b1\u03b9 \u03bc\u03af\u03b1 \u03b4\u03ae\u03bb\u03c9\u03c3\u03b7 confidence, \u03c1\u03c9\u03c4\u03ac \u03c4\u03bf \u03af\u03b4\u03b9\u03bf black-box \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03bc\u03b5 \u03c0\u03ad\u03bd\u03c4\u03b5 confidence-steering prompts. \u0388\u03c0\u03b5\u03b9\u03c4\u03b1 \u03bc\u03b5\u03c4\u03b1\u03c4\u03c1\u03ad\u03c0\u03b5\u03b9 \u03c4\u03b9\u03c2 \u03b1\u03c0\u03b1\u03bd\u03c4\u03ae\u03c3\u03b5\u03b9\u03c2 \u03ba\u03b1\u03b9 \u03c4\u03b1 self-reported scores \u03c3\u03b5 \u03b2\u03b1\u03b8\u03bc\u03bf\u03bd\u03bf\u03bc\u03b7\u03bc\u03ad\u03bd\u03b1 \u03c4\u03b5\u03ba\u03bc\u03ae\u03c1\u03b9\u03b1, \u03c4\u03b1 \u03c3\u03c5\u03b3\u03ba\u03b5\u03bd\u03c4\u03c1\u03ce\u03bd\u03b5\u03b9 \u03bc\u03b5 \u03ba\u03b1\u03c4\u03b1\u03bd\u03bf\u03bc\u03ae Dirichlet \u03ba\u03b1\u03b9 \u03b5\u03c6\u03b1\u03c1\u03bc\u03cc\u03b6\u03b5\u03b9 \u03b4\u03b5\u03cd\u03c4\u03b5\u03c1\u03bf calibration \u03c3\u03c4\u03b7\u03bd \u03c4\u03b5\u03bb\u03b9\u03ba\u03ae \u03c0\u03b9\u03b8\u03b1\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1.<\/p>\n<p>\u03a4\u03bf \u03c0\u03b9\u03bf \u03c7\u03c1\u03ae\u03c3\u03b9\u03bc\u03bf \u03c3\u03c4\u03bf\u03b9\u03c7\u03b5\u03af\u03bf \u03b3\u03b9\u03b1 \u03bc\u03b9\u03b1 \u03bf\u03bc\u03ac\u03b4\u03b1 \u03c0\u03bf\u03c5 \u03c0\u03b5\u03c1\u03bd\u03ac <a href=\"https:\/\/twodots.gr\/ai-agents-paragogi-leitourgikes-astochies\/\">AI agents \u03b1\u03c0\u03cc demo \u03c3\u03b5 \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03ae<\/a> \u03b5\u03af\u03bd\u03b1\u03b9 \u03bf \u03c1\u03b7\u03c4\u03cc\u03c2 <em>null state<\/em>: \u03b7 \u03b4\u03c5\u03bd\u03b1\u03c4\u03cc\u03c4\u03b7\u03c4\u03b1 \u03bd\u03b1 \u03b1\u03bd\u03b1\u03b3\u03bd\u03c9\u03c1\u03b9\u03c3\u03c4\u03b5\u03af \u03cc\u03c4\u03b9 \u03af\u03c3\u03c9\u03c2 \u03ba\u03b1\u03bc\u03af\u03b1 \u03b1\u03c0\u03cc \u03c4\u03b9\u03c2 \u03b4\u03b9\u03b1\u03b8\u03ad\u03c3\u03b9\u03bc\u03b5\u03c2 \u03b1\u03c0\u03b1\u03bd\u03c4\u03ae\u03c3\u03b5\u03b9\u03c2 \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03c3\u03c9\u03c3\u03c4\u03ae. \u0391\u03c5\u03c4\u03ae \u03b7 \u03b1\u03c1\u03c7\u03ae \u03b4\u03b5\u03bd \u03b1\u03c1\u03ba\u03b5\u03af \u03b1\u03c0\u03cc \u03bc\u03cc\u03bd\u03b7 \u03c4\u03b7\u03c2 \u03b3\u03b9\u03b1 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ae \u03b1\u03c5\u03c4\u03bf\u03bc\u03b1\u03c4\u03bf\u03c0\u03bf\u03af\u03b7\u03c3\u03b7, \u03b1\u03bb\u03bb\u03ac \u03b4\u03b9\u03bf\u03c1\u03b8\u03ce\u03bd\u03b5\u03b9 \u03ad\u03bd\u03b1 \u03ba\u03c1\u03af\u03c3\u03b9\u03bc\u03bf \u03c3\u03c7\u03b5\u03b4\u03b9\u03b1\u03c3\u03c4\u03b9\u03ba\u03cc \u03c3\u03c6\u03ac\u03bb\u03bc\u03b1: \u03c4\u03b7\u03bd \u03c5\u03c0\u03bf\u03c7\u03c1\u03b5\u03c9\u03c4\u03b9\u03ba\u03ae \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae \u03b1\u03ba\u03cc\u03bc\u03b7 \u03ba\u03b1\u03b9 \u03cc\u03c4\u03b1\u03bd \u03c4\u03b1 \u03c4\u03b5\u03ba\u03bc\u03ae\u03c1\u03b9\u03b1 \u03b4\u03b5\u03bd \u03b5\u03c0\u03b1\u03c1\u03ba\u03bf\u03cd\u03bd.<\/p>\n<h2 id=\"vevaiotita-llm-den-einai-pithanotita\">\u0393\u03b9\u03b1\u03c4\u03af \u03b7 \u03b2\u03b5\u03b2\u03b1\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b5\u03bd\u03cc\u03c2 LLM \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03b1\u03c0\u03cc \u03bc\u03cc\u03bd\u03b7 \u03c4\u03b7\u03c2 \u03c0\u03b9\u03b8\u03b1\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1<\/h2>\n<p>\u03a4\u03b1 language models \u03bc\u03c0\u03bf\u03c1\u03bf\u03cd\u03bd \u03bd\u03b1 \u03c0\u03b1\u03c1\u03ac\u03b3\u03bf\u03c5\u03bd \u03bc\u03b9\u03b1 \u03b1\u03c1\u03b9\u03b8\u03bc\u03b7\u03c4\u03b9\u03ba\u03ae \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7 \u03ba\u03b1\u03b9 \u03bd\u03b1 \u03b4\u03b7\u03bb\u03ce\u03c3\u03bf\u03c5\u03bd \u03c0\u03cc\u03c3\u03bf \u03c3\u03af\u03b3\u03bf\u03c5\u03c1\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03b3\u03b9\u2019 \u03b1\u03c5\u03c4\u03ae. \u0397 verbalized confidence \u03b5\u03af\u03bd\u03b1\u03b9 \u03b4\u03b9\u03b1\u03b8\u03ad\u03c3\u03b9\u03bc\u03b7 \u03b1\u03ba\u03cc\u03bc\u03b7 \u03ba\u03b1\u03b9 \u03cc\u03c4\u03b1\u03bd \u03b4\u03b5\u03bd \u03ad\u03c7\u03bf\u03c5\u03bc\u03b5 \u03c0\u03c1\u03cc\u03c3\u03b2\u03b1\u03c3\u03b7 \u03c3\u03c4\u03b1 token probabilities. \u0391\u03c5\u03c4\u03cc \u03c4\u03b7\u03bd \u03ba\u03ac\u03bd\u03b5\u03b9 \u03b5\u03bb\u03ba\u03c5\u03c3\u03c4\u03b9\u03ba\u03ae \u03b3\u03b9\u03b1 black-box \u03b5\u03c6\u03b1\u03c1\u03bc\u03bf\u03b3\u03ad\u03c2, \u03b1\u03bb\u03bb\u03ac \u03cc\u03c7\u03b9 \u03b1\u03c5\u03c4\u03cc\u03bc\u03b1\u03c4\u03b1 \u03b1\u03be\u03b9\u03cc\u03c0\u03b9\u03c3\u03c4\u03b7.<\/p>\n<p>\u0388\u03bd\u03b1 prompt \u03c0\u03bf\u03c5 \u03b6\u03b7\u03c4\u03ac \u03b1\u03c0\u03cc \u03c4\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03bd\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03b5\u03c0\u03b9\u03c6\u03c5\u03bb\u03b1\u03ba\u03c4\u03b9\u03ba\u03cc \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03bc\u03b5\u03c4\u03b1\u03ba\u03b9\u03bd\u03ae\u03c3\u03b5\u03b9 \u03cc\u03bb\u03b7 \u03c4\u03b7\u03bd \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03b1 \u03c0\u03c1\u03bf\u03c2 \u03c7\u03b1\u03bc\u03b7\u03bb\u03cc\u03c4\u03b5\u03c1\u03b5\u03c2 \u03c4\u03b9\u03bc\u03ad\u03c2. \u0388\u03bd\u03b1 prompt \u03c0\u03bf\u03c5 \u03c4\u03bf \u03b5\u03bd\u03b8\u03b1\u03c1\u03c1\u03cd\u03bd\u03b5\u03b9 \u03bd\u03b1 \u03b5\u03bc\u03c6\u03b1\u03bd\u03b9\u03c3\u03c4\u03b5\u03af \u03c0\u03bf\u03bb\u03cd \u03b2\u03ad\u03b2\u03b1\u03b9\u03bf \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03c0\u03b1\u03c1\u03ac\u03b3\u03b5\u03b9 \u03c5\u03c8\u03b7\u03bb\u03cc\u03c4\u03b5\u03c1\u03b1 scores \u03b3\u03b9\u03b1 \u03c0\u03b1\u03c1\u03cc\u03bc\u03bf\u03b9\u03b1 \u03b9\u03ba\u03b1\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1. \u0397 \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03b1 \u03b1\u03bb\u03bb\u03ac\u03b6\u03b5\u03b9 \u03b5\u03c0\u03af\u03c3\u03b7\u03c2 \u03b1\u03bd\u03ac \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03ba\u03b1\u03b9 dataset. \u0386\u03c1\u03b1 \u03c4\u03bf \u00ab80%\u00bb \u03b4\u03b5\u03bd \u03ad\u03c7\u03b5\u03b9 \u03b1\u03bd\u03b1\u03b3\u03ba\u03b1\u03c3\u03c4\u03b9\u03ba\u03ac \u03c4\u03b7\u03bd \u03af\u03b4\u03b9\u03b1 \u03b5\u03bc\u03c0\u03b5\u03b9\u03c1\u03b9\u03ba\u03ae \u03c3\u03b7\u03bc\u03b1\u03c3\u03af\u03b1 \u03c3\u03b5 \u03b4\u03cd\u03bf prompts \u03ae \u03b4\u03cd\u03bf \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b5\u03c2.<\/p>\n<p>\u0391\u03c5\u03c4\u03cc \u03b5\u03af\u03bd\u03b1\u03b9 \u03c4\u03bf \u03c0\u03c1\u03cc\u03b2\u03bb\u03b7\u03bc\u03b1 \u03c4\u03b7\u03c2 \u03b2\u03b1\u03b8\u03bc\u03bf\u03bd\u03cc\u03bc\u03b7\u03c3\u03b7\u03c2. \u0391\u03bd \u03b5\u03ba\u03b1\u03c4\u03cc \u03c0\u03c1\u03bf\u03b2\u03bb\u03ad\u03c8\u03b5\u03b9\u03c2 \u03b4\u03b7\u03bb\u03ce\u03bd\u03bf\u03bd\u03c4\u03b1\u03b9 \u03bc\u03b5 confidence 80%, \u03ad\u03bd\u03b1 \u03ba\u03b1\u03bb\u03ac \u03b2\u03b1\u03b8\u03bc\u03bf\u03bd\u03bf\u03bc\u03b7\u03bc\u03ad\u03bd\u03bf \u03c3\u03cd\u03c3\u03c4\u03b7\u03bc\u03b1 \u03b8\u03b1 \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03c3\u03c9\u03c3\u03c4\u03cc \u03c0\u03b5\u03c1\u03af\u03c0\u03bf\u03c5 \u03c3\u03c4\u03b9\u03c2 \u03bf\u03b3\u03b4\u03cc\u03bd\u03c4\u03b1. \u0397 calibration \u03b4\u03b5\u03bd \u03c4\u03b1\u03c5\u03c4\u03af\u03b6\u03b5\u03c4\u03b1\u03b9 \u03bc\u03b5 \u03c4\u03b7\u03bd accuracy: \u03ad\u03bd\u03b1 \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b5\u03c0\u03b9\u03bb\u03ad\u03b3\u03b5\u03b9 \u03c3\u03c5\u03c7\u03bd\u03ac \u03c3\u03c9\u03c3\u03c4\u03ae \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7 \u03b1\u03bb\u03bb\u03ac \u03bd\u03b1 \u03c5\u03c0\u03b5\u03c1\u03b5\u03ba\u03c4\u03b9\u03bc\u03ac \u03c4\u03b7 \u03b2\u03b5\u03b2\u03b1\u03b9\u03cc\u03c4\u03b7\u03c4\u03ac \u03c4\u03bf\u03c5.<\/p>\n<h2 id=\"symfonia-apantiseon-den-arkei\">\u0393\u03b9\u03b1\u03c4\u03af \u03b7 \u03c3\u03c5\u03bc\u03c6\u03c9\u03bd\u03af\u03b1 \u03bc\u03b5\u03c4\u03b1\u03be\u03cd \u03b1\u03c0\u03b1\u03bd\u03c4\u03ae\u03c3\u03b5\u03c9\u03bd \u03b4\u03b5\u03bd \u03b1\u03c1\u03ba\u03b5\u03af<\/h2>\n<p>\u03a0\u03bf\u03bb\u03bb\u03ad\u03c2 black-box \u03bc\u03ad\u03b8\u03bf\u03b4\u03bf\u03b9 \u03c0\u03b1\u03c1\u03ac\u03b3\u03bf\u03c5\u03bd \u03b1\u03c1\u03ba\u03b5\u03c4\u03ad\u03c2 \u03b1\u03c0\u03b1\u03bd\u03c4\u03ae\u03c3\u03b5\u03b9\u03c2 \u03ba\u03b1\u03b9 \u03bc\u03b5\u03c4\u03c1\u03bf\u03cd\u03bd \u03c0\u03cc\u03c3\u03b5\u03c2 \u03c6\u03bf\u03c1\u03ad\u03c2 \u03b5\u03c0\u03b1\u03bd\u03b1\u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03b5\u03c4\u03b1\u03b9 \u03ba\u03ac\u03b8\u03b5 \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1 \u03ae \u03c5\u03c0\u03bf\u03bb\u03bf\u03b3\u03af\u03b6\u03bf\u03c5\u03bd entropy \u03c3\u03c4\u03b7 \u03b4\u03b9\u03b1\u03bd\u03bf\u03bc\u03ae \u03c4\u03c9\u03bd outputs. \u0397 self-consistency \u03ba\u03b1\u03b9 \u03bf\u03b9 \u03c3\u03c5\u03b3\u03b3\u03b5\u03bd\u03b9\u03ba\u03ad\u03c2 \u03c0\u03c1\u03bf\u03c3\u03b5\u03b3\u03b3\u03af\u03c3\u03b5\u03b9\u03c2 \u03b4\u03af\u03bd\u03bf\u03c5\u03bd \u03c7\u03c1\u03ae\u03c3\u03b9\u03bc\u03bf \u03c3\u03ae\u03bc\u03b1 \u03b3\u03b9\u03b1 \u03c4\u03b7 \u03bc\u03b5\u03c4\u03b1\u03b2\u03bb\u03b7\u03c4\u03cc\u03c4\u03b7\u03c4\u03b1 \u03c4\u03bf\u03c5 \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf\u03c5, \u03cc\u03c0\u03c9\u03c2 \u03ba\u03b1\u03b9 \u03bf\u03b9 \u03c4\u03b5\u03c7\u03bd\u03b9\u03ba\u03ad\u03c2 \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03b7\u03c2 \u03b1\u03b2\u03b5\u03b2\u03b1\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1\u03c2 \u03c0\u03bf\u03c5 \u03c3\u03c5\u03b6\u03b7\u03c4\u03ac\u03bc\u03b5 \u03c3\u03c4\u03bf <a href=\"https:\/\/twodots.gr\/cas-ai-search-agents-conformal-prediction\/\">CAS \u03b3\u03b9\u03b1 AI search agents<\/a>.<\/p>\n<p>\u03a9\u03c3\u03c4\u03cc\u03c3\u03bf, \u03b7 \u03c3\u03c5\u03c7\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b4\u03b5\u03bd \u03b1\u03be\u03b9\u03bf\u03c0\u03bf\u03b9\u03b5\u03af \u03c4\u03bf \u03b1\u03c1\u03b9\u03b8\u03bc\u03b7\u03c4\u03b9\u03ba\u03cc \u03c0\u03b5\u03c1\u03b9\u03b5\u03c7\u03cc\u03bc\u03b5\u03bd\u03bf \u03c4\u03b7\u03c2 \u03b4\u03ae\u03bb\u03c9\u03c3\u03b7\u03c2 confidence. \u03a0\u03ad\u03bd\u03c4\u03b5 \u03cc\u03bc\u03bf\u03b9\u03b5\u03c2 \u03b1\u03c0\u03b1\u03bd\u03c4\u03ae\u03c3\u03b5\u03b9\u03c2 \u03bc\u03b5 \u03c7\u03b1\u03bc\u03b7\u03bb\u03ae \u03b1\u03c5\u03c4\u03bf\u03c0\u03b5\u03c0\u03bf\u03af\u03b8\u03b7\u03c3\u03b7 \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03b1\u03c0\u03b1\u03c1\u03b1\u03af\u03c4\u03b7\u03c4\u03b1 \u03b9\u03c3\u03bf\u03b4\u03cd\u03bd\u03b1\u03bc\u03b5\u03c2 \u03bc\u03b5 \u03c0\u03ad\u03bd\u03c4\u03b5 \u03cc\u03bc\u03bf\u03b9\u03b5\u03c2 \u03b1\u03c0\u03b1\u03bd\u03c4\u03ae\u03c3\u03b5\u03b9\u03c2 \u03c5\u03c8\u03b7\u03bb\u03ae\u03c2 \u03b1\u03c5\u03c4\u03bf\u03c0\u03b5\u03c0\u03bf\u03af\u03b8\u03b7\u03c3\u03b7\u03c2. \u03a3\u03c4\u03b7\u03bd \u03b1\u03bd\u03c4\u03af\u03b8\u03b5\u03c4\u03b7 \u03ba\u03b1\u03c4\u03b5\u03cd\u03b8\u03c5\u03bd\u03c3\u03b7, \u03bf \u03b1\u03c0\u03bb\u03cc\u03c2 \u03bc\u03ad\u03c3\u03bf\u03c2 \u03cc\u03c1\u03bf\u03c2 \u03c4\u03c9\u03bd scores \u03c5\u03c0\u03bf\u03b8\u03ad\u03c4\u03b5\u03b9 \u03cc\u03c4\u03b9 \u03bf\u03b9 \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03b5\u03c2 \u03b5\u03af\u03bd\u03b1\u03b9 \u03ae\u03b4\u03b7 \u03c3\u03c5\u03b3\u03ba\u03c1\u03af\u03c3\u03b9\u03bc\u03b5\u03c2 \u03ba\u03b1\u03b9 \u03cc\u03c4\u03b9 \u03b4\u03b5\u03bd \u03c5\u03c0\u03ac\u03c1\u03c7\u03b5\u03b9 prompt-dependent bias.<\/p>\n<p>\u03a4\u03bf DirEAG \u03b1\u03bd\u03c4\u03b9\u03bc\u03b5\u03c4\u03c9\u03c0\u03af\u03b6\u03b5\u03b9 \u03c4\u03b1 \u03c0\u03bf\u03bb\u03bb\u03b1\u03c0\u03bb\u03ac observations \u03c9\u03c2 \u03c3\u03c4\u03b1\u03c4\u03b9\u03c3\u03c4\u03b9\u03ba\u03cc calibration problem. \u0394\u03b5\u03bd \u03c0\u03b5\u03c4\u03ac \u03c4\u03b7 verbalized confidence, \u03b1\u03bb\u03bb\u03ac \u03bf\u03cd\u03c4\u03b5 \u03c4\u03b7\u03bd \u03c0\u03b1\u03af\u03c1\u03bd\u03b5\u03b9 \u03ba\u03c5\u03c1\u03b9\u03bf\u03bb\u03b5\u03ba\u03c4\u03b9\u03ba\u03ac. \u039c\u03b1\u03b8\u03b1\u03af\u03bd\u03b5\u03b9 \u03b1\u03c0\u03cc calibration examples \u03c0\u03ce\u03c2 \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b5\u03c0\u03b7\u03c1\u03b5\u03ac\u03b6\u03b5\u03b9 \u03ba\u03ac\u03b8\u03b5 score \u03c4\u03b7\u03bd \u03c5\u03c0\u03bf\u03c3\u03c4\u03ae\u03c1\u03b9\u03be\u03b7 \u03bc\u03b9\u03b1\u03c2 \u03c5\u03c0\u03bf\u03c8\u03ae\u03c6\u03b9\u03b1\u03c2 \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7\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\">\u0388\u03bd\u03b1 self-reported score<\/p>\n<p>\u0395\u03af\u03bd\u03b1\u03b9 \u03b5\u03cd\u03ba\u03bf\u03bb\u03bf \u03bd\u03b1 \u03bb\u03b7\u03c6\u03b8\u03b5\u03af \u03b1\u03c0\u03cc black-box LLM, \u03b1\u03bb\u03bb\u03ac \u03b7 \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03ac \u03c4\u03bf\u03c5 \u03bc\u03b5\u03c4\u03b1\u03ba\u03b9\u03bd\u03b5\u03af\u03c4\u03b1\u03b9 \u03bc\u03b5 \u03c4\u03bf prompt, \u03c4\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03ba\u03b1\u03b9 \u03c4\u03bf dataset. \u0394\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03ad\u03c4\u03bf\u03b9\u03bc\u03b7 \u03c0\u03b9\u03b8\u03b1\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1 \u03bf\u03c1\u03b8\u03cc\u03c4\u03b7\u03c4\u03b1\u03c2.<\/p>\n<div class=\"td-badge-row\"><span class=\"td-badge\">\u03a7\u03b1\u03bc\u03b7\u03bb\u03cc \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2<\/span><span class=\"td-badge\">Scale bias<\/span><\/div>\n<\/div>\n<div class=\"td-platform-card\">\n<p class=\"td-comparison-title\">\u03a3\u03c5\u03bc\u03c6\u03c9\u03bd\u03af\u03b1 \u03ae entropy<\/p>\n<p>\u0391\u03be\u03b9\u03bf\u03c0\u03bf\u03b9\u03b5\u03af \u03c4\u03b7\u03bd \u03b5\u03c0\u03b1\u03bd\u03ac\u03bb\u03b7\u03c8\u03b7 \u03ba\u03b1\u03b9 \u03c4\u03b7 \u03b4\u03b9\u03b1\u03c6\u03c9\u03bd\u03af\u03b1 \u03bc\u03b5\u03c4\u03b1\u03be\u03cd samples, \u03cc\u03bc\u03c9\u03c2 \u03c7\u03ac\u03bd\u03b5\u03b9 \u03c4\u03bf \u03b1\u03c1\u03b9\u03b8\u03bc\u03b7\u03c4\u03b9\u03ba\u03cc \u03c3\u03ae\u03bc\u03b1 \u03c4\u03bf\u03c5 confidence \u03ba\u03b1\u03b9 \u03b4\u03b5\u03bd \u03b5\u03b3\u03b3\u03c5\u03ac\u03c4\u03b1\u03b9 \u03b2\u03b1\u03b8\u03bc\u03bf\u03bd\u03bf\u03bc\u03b7\u03bc\u03ad\u03bd\u03b7 \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03b1.<\/p>\n<div class=\"td-badge-row\"><span class=\"td-badge\">\u03a7\u03c1\u03ae\u03c3\u03b9\u03bc\u03bf ranking<\/span><span class=\"td-badge\">\u03a7\u03c9\u03c1\u03af\u03c2 score semantics<\/span><\/div>\n<\/div>\n<div class=\"td-platform-card td-platform-card--navy\">\n<p class=\"td-comparison-title\">DirEAG aggregation<\/p>\n<p>\u0394\u03b9\u03bf\u03c1\u03b8\u03ce\u03bd\u03b5\u03b9 \u03ba\u03ac\u03b8\u03b5 steering scale, \u03b6\u03c5\u03b3\u03af\u03b6\u03b5\u03b9 \u03c4\u03bf evidence, \u03ba\u03c1\u03b1\u03c4\u03ac null state \u03ba\u03b1\u03b9 \u03b2\u03b1\u03b8\u03bc\u03bf\u03bd\u03bf\u03bc\u03b5\u03af \u03be\u03b1\u03bd\u03ac \u03c4\u03bf\u03bd \u03bd\u03b9\u03ba\u03b7\u03c4\u03ae. \u03a7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 labeled calibration data.<\/p>\n<div class=\"td-badge-row\"><span class=\"td-badge\">Learned evidence<\/span><span class=\"td-badge\">\u03a1\u03b7\u03c4\u03ae \u03b1\u03c0\u03bf\u03c7\u03ae<\/span><\/div>\n<\/div>\n<\/div>\n<\/div>\n<h2 id=\"pente-prompts-biased-klimakes\">\u03a0\u03ad\u03bd\u03c4\u03b5 prompts \u03ba\u03b1\u03b9 \u03c0\u03ad\u03bd\u03c4\u03b5 \u03bc\u03b5\u03c1\u03bf\u03bb\u03b7\u03c0\u03c4\u03b9\u03ba\u03ad\u03c2 \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03b5\u03c2<\/h2>\n<p>\u0393\u03b9\u03b1 \u03ba\u03ac\u03b8\u03b5 \u03bc\u03b1\u03b8\u03b7\u03bc\u03b1\u03c4\u03b9\u03ba\u03cc \u03c0\u03c1\u03cc\u03b2\u03bb\u03b7\u03bc\u03b1, \u03c4\u03bf \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03b5\u03c1\u03c9\u03c4\u03ac\u03c4\u03b1\u03b9 \u03bc\u03b5 \u03c0\u03ad\u03bd\u03c4\u03b5 confidence-steering prompts, \u03b1\u03c0\u03cc \u03c0\u03b9\u03bf \u03b5\u03c0\u03b9\u03c6\u03c5\u03bb\u03b1\u03ba\u03c4\u03b9\u03ba\u03ac \u03ad\u03c9\u03c2 \u03c0\u03b9\u03bf \u03b2\u03ad\u03b2\u03b1\u03b9\u03b1. \u039a\u03ac\u03b8\u03b5 \u03b5\u03ba\u03c4\u03ad\u03bb\u03b5\u03c3\u03b7 \u03b5\u03c0\u03b9\u03c3\u03c4\u03c1\u03ad\u03c6\u03b5\u03b9 \u03b6\u03b5\u03cd\u03b3\u03bf\u03c2 \u03b1\u03c1\u03b9\u03b8\u03bc\u03b7\u03c4\u03b9\u03ba\u03ae\u03c2 \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7\u03c2 \u03ba\u03b1\u03b9 \u03b4\u03b7\u03bb\u03c9\u03bc\u03ad\u03bd\u03bf\u03c5 confidence. \u039f\u03b9 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03ad\u03c2 \u03b1\u03c0\u03b1\u03bd\u03c4\u03ae\u03c3\u03b5\u03b9\u03c2 \u03c3\u03c7\u03b7\u03bc\u03b1\u03c4\u03af\u03b6\u03bf\u03c5\u03bd \u03ad\u03bd\u03b1 \u03c0\u03b5\u03c0\u03b5\u03c1\u03b1\u03c3\u03bc\u03ad\u03bd\u03bf candidate set.<\/p>\n<p>\u03a0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03c4\u03b7 \u03c3\u03c5\u03b3\u03ba\u03ad\u03bd\u03c4\u03c1\u03c9\u03c3\u03b7, \u03ba\u03ac\u03b8\u03b5 raw score \u03c0\u03b5\u03c1\u03bd\u03ac \u03b1\u03c0\u03cc monotonic calibration transform \u03c4\u03cd\u03c0\u03bf\u03c5 Platt. \u0397 \u03bc\u03b5\u03c4\u03b1\u03c4\u03c1\u03bf\u03c0\u03ae \u03ad\u03c7\u03b5\u03b9 bias \u03b5\u03b9\u03b4\u03b9\u03ba\u03cc \u03b3\u03b9\u03b1 \u03ba\u03ac\u03b8\u03b5 steering level \u03ba\u03b1\u03b9 \u03ba\u03bf\u03b9\u03bd\u03ae \u03b8\u03b5\u03c4\u03b9\u03ba\u03ae \u03ba\u03bb\u03af\u03c3\u03b7. \u0388\u03c4\u03c3\u03b9 \u03b4\u03b9\u03bf\u03c1\u03b8\u03ce\u03bd\u03b5\u03b9 \u03c4\u03b7\u03bd \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03b1 \u03c7\u03c9\u03c1\u03af\u03c2 \u03bd\u03b1 \u03b1\u03bd\u03c4\u03b9\u03c3\u03c4\u03c1\u03ad\u03c6\u03b5\u03b9 \u03c4\u03b7 \u03c3\u03b5\u03b9\u03c1\u03ac: \u03bc\u03ad\u03c3\u03b1 \u03c3\u03c4\u03bf \u03af\u03b4\u03b9\u03bf level, \u03ad\u03bd\u03b1 \u03c5\u03c8\u03b7\u03bb\u03cc\u03c4\u03b5\u03c1\u03bf raw confidence \u03c0\u03b1\u03c1\u03b1\u03bc\u03ad\u03bd\u03b5\u03b9 \u03c4\u03bf\u03c5\u03bb\u03ac\u03c7\u03b9\u03c3\u03c4\u03bf\u03bd \u03c4\u03cc\u03c3\u03bf \u03b9\u03c3\u03c7\u03c5\u03c1\u03cc \u03bc\u03b5\u03c4\u03ac \u03c4\u03bf calibration.<\/p>\n<p>\u039f\u03b9 \u03c0\u03b1\u03c1\u03ac\u03bc\u03b5\u03c4\u03c1\u03bf\u03b9 \u03b5\u03af\u03bd\u03b1\u03b9 \u03bb\u03af\u03b3\u03bf\u03b9 \u03ba\u03b1\u03b9 \u03ba\u03bf\u03b9\u03bd\u03ad\u03c2 \u03b3\u03b9\u03b1 \u03c4\u03b1 \u03c0\u03b1\u03c1\u03b1\u03b4\u03b5\u03af\u03b3\u03bc\u03b1\u03c4\u03b1 \u03c4\u03b7\u03c2 \u03af\u03b4\u03b9\u03b1\u03c2 calibration setting. \u0397 \u03bc\u03ad\u03b8\u03bf\u03b4\u03bf\u03c2 \u03b4\u03b5\u03bd \u03b1\u03bb\u03bb\u03ac\u03b6\u03b5\u03b9 \u03bf\u03cd\u03c4\u03b5 \u03ba\u03ac\u03bd\u03b5\u03b9 fine-tune \u03c4\u03bf \u03b2\u03b1\u03c3\u03b9\u03ba\u03cc LLM. \u0397 \u03b5\u03ba\u03bc\u03ac\u03b8\u03b7\u03c3\u03b7 \u03b3\u03af\u03bd\u03b5\u03c4\u03b1\u03b9 \u03c3\u03c4\u03bf \u03b5\u03c0\u03af\u03c0\u03b5\u03b4\u03bf \u03c4\u03b7\u03c2 \u03c3\u03c5\u03b3\u03ba\u03ad\u03bd\u03c4\u03c1\u03c9\u03c3\u03b7\u03c2 \u03c4\u03c9\u03bd observations, \u03bc\u03b9\u03b1 \u03c7\u03c1\u03ae\u03c3\u03b9\u03bc\u03b7 \u03b4\u03b9\u03ac\u03ba\u03c1\u03b9\u03c3\u03b7 \u03cc\u03c4\u03b1\u03bd \u03b7 \u03bf\u03bc\u03ac\u03b4\u03b1 \u03b4\u03b5\u03bd \u03b5\u03bb\u03ad\u03b3\u03c7\u03b5\u03b9 \u03c4\u03b1 \u03b2\u03ac\u03c1\u03b7 \u03c4\u03bf\u03c5 \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf\u03c5.<\/p>\n<h2 id=\"dirichlet-logistiko-vivlio-evidence\">\u0397 Dirichlet \u03c9\u03c2 \u03bb\u03bf\u03b3\u03b9\u03c3\u03c4\u03b9\u03ba\u03cc \u03b2\u03b9\u03b2\u03bb\u03af\u03bf \u03c4\u03b5\u03ba\u03bc\u03b7\u03c1\u03af\u03c9\u03bd<\/h2>\n<p>\u03a4\u03bf DirEAG \u03b1\u03bd\u03b1\u03b8\u03ad\u03c4\u03b5\u03b9 \u03bc\u03b7 \u03b1\u03c1\u03bd\u03b7\u03c4\u03b9\u03ba\u03ae \u03c3\u03c5\u03b3\u03ba\u03ad\u03bd\u03c4\u03c1\u03c9\u03c3\u03b7 \u03c3\u03b5 \u03ba\u03ac\u03b8\u03b5 \u03c5\u03c0\u03bf\u03c8\u03ae\u03c6\u03b9\u03b1 \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7. \u038c\u03c4\u03b1\u03bd \u03ad\u03bd\u03b1 steering level \u03b5\u03c0\u03b9\u03c3\u03c4\u03c1\u03ad\u03c6\u03b5\u03b9 \u03bc\u03b9\u03b1 \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03b7 \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7, \u03c0\u03c1\u03bf\u03c3\u03b8\u03ad\u03c4\u03b5\u03b9 \u03c4\u03b5\u03ba\u03bc\u03ae\u03c1\u03b9\u03bf \u03b1\u03bd\u03ac\u03bb\u03bf\u03b3\u03bf \u03bc\u03b5 \u03c4\u03bf calibrated confidence \u03ba\u03b1\u03b9 \u03ad\u03bd\u03b1 learned reliability weight \u03c4\u03bf\u03c5 level. \u0397 \u03b5\u03c0\u03b1\u03bd\u03ac\u03bb\u03b7\u03c8\u03b7 \u03ad\u03c7\u03b5\u03b9 \u03c3\u03b7\u03bc\u03b1\u03c3\u03af\u03b1, \u03b1\u03bb\u03bb\u03ac \u03cc\u03c7\u03b9 \u03c9\u03c2 \u03b1\u03c0\u03bb\u03ae \u03c8\u03ae\u03c6\u03bf\u03c2: \u03b7 \u03b9\u03c3\u03c7\u03cd\u03c2 \u03c4\u03b7\u03c2 \u03b5\u03be\u03b1\u03c1\u03c4\u03ac\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03c4\u03b7\u03bd \u03b1\u03be\u03b9\u03bf\u03c0\u03b9\u03c3\u03c4\u03af\u03b1 \u03c4\u03bf\u03c5 prompt \u03ba\u03b1\u03b9 \u03c4\u03b7 \u03b2\u03b1\u03b8\u03bc\u03bf\u03bd\u03bf\u03bc\u03b7\u03bc\u03ad\u03bd\u03b7 \u03b2\u03b5\u03b2\u03b1\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1.<\/p>\n<p>\u0397 \u03ba\u03b1\u03c4\u03b1\u03bd\u03bf\u03bc\u03ae Dirichlet \u03c0\u03b5\u03c1\u03b9\u03b3\u03c1\u03ac\u03c6\u03b5\u03b9 \u03c0\u03b9\u03b8\u03b1\u03bd\u03cc\u03c4\u03b7\u03c4\u03b5\u03c2 \u03c0\u03ac\u03bd\u03c9 \u03c3\u03b5 \u03bc\u03b5\u03c4\u03b1\u03b2\u03bb\u03b7\u03c4\u03cc \u03b1\u03c1\u03b9\u03b8\u03bc\u03cc \u03ba\u03b1\u03c4\u03b7\u03b3\u03bf\u03c1\u03b9\u03ba\u03ce\u03bd states. \u039c\u03b5\u03c4\u03ac \u03c4\u03b7 \u03c3\u03c5\u03c3\u03c3\u03ce\u03c1\u03b5\u03c5\u03c3\u03b7, \u03bf posterior mean \u03b4\u03af\u03bd\u03b5\u03b9 \u03c4\u03b7 \u03c3\u03c7\u03b5\u03c4\u03b9\u03ba\u03ae \u03c5\u03c0\u03bf\u03c3\u03c4\u03ae\u03c1\u03b9\u03be\u03b7 \u03b3\u03b9\u03b1 \u03ba\u03ac\u03b8\u03b5 candidate. \u0397 \u03c4\u03b5\u03bb\u03b9\u03ba\u03ae \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7 \u03b5\u03c0\u03b9\u03bb\u03ad\u03b3\u03b5\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03c4\u03b7\u03bd \u03c5\u03c0\u03bf\u03c8\u03ae\u03c6\u03b9\u03b1 \u03bc\u03b5 \u03c4\u03b7 \u03bc\u03b5\u03b3\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b7 posterior probability.<\/p>\n<p>\u0397 \u03b1\u03bd\u03b1\u03c0\u03b1\u03c1\u03ac\u03c3\u03c4\u03b1\u03c3\u03b7 \u03b4\u03b9\u03b1\u03c7\u03c9\u03c1\u03af\u03b6\u03b5\u03b9 \u03b4\u03cd\u03bf \u03ad\u03bd\u03bd\u03bf\u03b9\u03b5\u03c2 \u03c0\u03bf\u03c5 \u03c3\u03c5\u03c7\u03bd\u03ac \u03bc\u03c0\u03b5\u03c1\u03b4\u03b5\u03cd\u03bf\u03bd\u03c4\u03b1\u03b9: \u03c0\u03cc\u03c3\u03bf \u03b9\u03c3\u03c7\u03c5\u03c1\u03ac \u03c5\u03c0\u03bf\u03c3\u03c4\u03b7\u03c1\u03af\u03b6\u03b5\u03c4\u03b1\u03b9 \u03bc\u03b9\u03b1 \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae \u03c3\u03b5 \u03c3\u03c7\u03ad\u03c3\u03b7 \u03bc\u03b5 \u03c4\u03b9\u03c2 \u03ac\u03bb\u03bb\u03b5\u03c2 \u03b4\u03b9\u03b1\u03b8\u03ad\u03c3\u03b9\u03bc\u03b5\u03c2 \u03ba\u03b1\u03b9 \u03c0\u03cc\u03c3\u03bf \u03c0\u03b9\u03b8\u03b1\u03bd\u03cc \u03b5\u03af\u03bd\u03b1\u03b9 \u03bd\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac \u03c3\u03c9\u03c3\u03c4\u03ae. \u0397 \u03c0\u03c1\u03ce\u03c4\u03b7 \u03c0\u03c1\u03bf\u03ba\u03cd\u03c0\u03c4\u03b5\u03b9 \u03b1\u03c0\u03cc \u03c4\u03b7\u03bd evidence aggregation. \u0397 \u03b4\u03b5\u03cd\u03c4\u03b5\u03c1\u03b7 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03be\u03b5\u03c7\u03c9\u03c1\u03b9\u03c3\u03c4\u03ae \u03b5\u03bc\u03c0\u03b5\u03b9\u03c1\u03b9\u03ba\u03ae \u03b2\u03b1\u03b8\u03bc\u03bf\u03bd\u03cc\u03bc\u03b7\u03c3\u03b7 \u03bc\u03b5\u03c4\u03ac \u03c4\u03b7\u03bd \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae.<\/p>\n<h2 id=\"null-state-kamia-sosti-apantisi\">\u039f null state \u03b4\u03af\u03bd\u03b5\u03b9 \u03c7\u03ce\u03c1\u03bf \u03c3\u03c4\u03bf \u00ab\u03ba\u03b1\u03bc\u03af\u03b1 \u03c3\u03c9\u03c3\u03c4\u03ae \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7\u00bb<\/h2>\n<p>\u0391\u03bd \u03c4\u03bf \u03c3\u03cd\u03c3\u03c4\u03b7\u03bc\u03b1 \u03bc\u03bf\u03af\u03c1\u03b1\u03b6\u03b5 \u03cc\u03bb\u03b7 \u03c4\u03b7 \u03b4\u03b9\u03b1\u03b8\u03ad\u03c3\u03b9\u03bc\u03b7 \u03c0\u03b9\u03b8\u03b1\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1 \u03bc\u03cc\u03bd\u03bf \u03c3\u03c4\u03b9\u03c2 \u03b1\u03c0\u03b1\u03bd\u03c4\u03ae\u03c3\u03b5\u03b9\u03c2 \u03c0\u03bf\u03c5 \u03c0\u03b1\u03c1\u03ae\u03c7\u03b8\u03b7\u03c3\u03b1\u03bd, \u03ba\u03ac\u03c0\u03bf\u03b9\u03b1 \u03b8\u03b1 \u03ba\u03ad\u03c1\u03b4\u03b9\u03b6\u03b5 \u03b1\u03bd\u03b1\u03b3\u03ba\u03b1\u03c3\u03c4\u03b9\u03ba\u03ac \u03b1\u03ba\u03cc\u03bc\u03b7 \u03ba\u03b1\u03b9 \u03cc\u03c4\u03b1\u03bd \u03cc\u03bb\u03b5\u03c2 \u03ae\u03c4\u03b1\u03bd \u03bb\u03ac\u03b8\u03bf\u03c2. \u03a4\u03bf DirEAG \u03c0\u03c1\u03bf\u03c3\u03b8\u03ad\u03c4\u03b5\u03b9 null state, \u03bf \u03bf\u03c0\u03bf\u03af\u03bf\u03c2 \u03b1\u03bd\u03c4\u03b9\u03c0\u03c1\u03bf\u03c3\u03c9\u03c0\u03b5\u03cd\u03b5\u03b9 \u03c4\u03b7\u03bd \u03b1\u03c0\u03bf\u03c5\u03c3\u03af\u03b1 \u03c4\u03b7\u03c2 \u03c3\u03c9\u03c3\u03c4\u03ae\u03c2 \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7\u03c2 \u03b1\u03c0\u03cc \u03c4\u03bf candidate set.<\/p>\n<p>\u039f null state \u03c0\u03b1\u03af\u03c1\u03bd\u03b5\u03b9 prior mass, \u03b2\u03b1\u03c3\u03b9\u03ba\u03ae \u03c3\u03c5\u03b3\u03ba\u03ad\u03bd\u03c4\u03c1\u03c9\u03c3\u03b7 \u03ba\u03b1\u03b9 \u03c0\u03c1\u03cc\u03c3\u03b8\u03b5\u03c4\u03bf uncertainty evidence \u03cc\u03c4\u03b1\u03bd \u03c4\u03b1 calibrated confidence scores \u03b5\u03af\u03bd\u03b1\u03b9 \u03c7\u03b1\u03bc\u03b7\u03bb\u03ac. \u0388\u03c4\u03c3\u03b9 \u03b7 \u03b1\u03b4\u03cd\u03bd\u03b1\u03bc\u03b7 \u03c5\u03c0\u03bf\u03c3\u03c4\u03ae\u03c1\u03b9\u03be\u03b7 \u03b4\u03b5\u03bd \u03b1\u03bd\u03b1\u03b4\u03b9\u03b1\u03bd\u03ad\u03bc\u03b5\u03c4\u03b1\u03b9 \u03c4\u03b5\u03c7\u03bd\u03b7\u03c4\u03ac \u03c5\u03c0\u03ad\u03c1 \u03bc\u03b9\u03b1\u03c2 \u03b4\u03b9\u03b1\u03b8\u03ad\u03c3\u03b9\u03bc\u03b7\u03c2 \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7\u03c2. \u03a3\u03c4\u03b1 diagnostics \u03c4\u03bf\u03c5 paper, \u03b7 \u03bc\u03ad\u03c3\u03b7 \u03c0\u03b9\u03b8\u03b1\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1 \u03c4\u03bf\u03c5 null state \u03ae\u03c4\u03b1\u03bd \u03c3\u03c5\u03c3\u03c4\u03b7\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac \u03c5\u03c8\u03b7\u03bb\u03cc\u03c4\u03b5\u03c1\u03b7 \u03cc\u03c4\u03b1\u03bd \u03b7 \u03c3\u03c9\u03c3\u03c4\u03ae \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7 \u03ad\u03bb\u03b5\u03b9\u03c0\u03b5 \u03b1\u03c0\u03cc \u03c4\u03b1 generated candidates \u03b1\u03c0\u03cc \u03cc,\u03c4\u03b9 \u03c3\u03c4\u03b9\u03c2 \u03c3\u03c9\u03c3\u03c4\u03ad\u03c2 \u03c4\u03b5\u03bb\u03b9\u03ba\u03ad\u03c2 \u03c0\u03c1\u03bf\u03b2\u03bb\u03ad\u03c8\u03b5\u03b9\u03c2.<\/p>\n<p>\u0393\u03b9\u03b1 \u03ad\u03bd\u03b1 business workflow, \u03b7 \u03b1\u03bd\u03b1\u03bb\u03bf\u03b3\u03af\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03c3\u03c7\u03b5\u03b4\u03b9\u03b1\u03c3\u03c4\u03b9\u03ba\u03ae \u03b5\u03c1\u03bc\u03b7\u03bd\u03b5\u03af\u03b1 \u03ba\u03b1\u03b9 \u03cc\u03c7\u03b9 \u03c0\u03b5\u03b9\u03c1\u03b1\u03bc\u03b1\u03c4\u03b9\u03ba\u03cc \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1 \u03c4\u03b7\u03c2 \u03bc\u03b5\u03bb\u03ad\u03c4\u03b7\u03c2. \u0388\u03bd\u03b1\u03c2 product recommender, support agent \u03ae classifier \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae \u00ab\u03b1\u03bd\u03b5\u03c0\u03b1\u03c1\u03ba\u03ae \u03c3\u03c4\u03bf\u03b9\u03c7\u03b5\u03af\u03b1 \/ \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03bf\u03c2 \u03ad\u03bb\u03b5\u03b3\u03c7\u03bf\u03c2\u00bb. \u0391\u03bd \u03c4\u03bf interface \u03b5\u03c0\u03b9\u03b2\u03ac\u03bb\u03bb\u03b5\u03b9 \u03c0\u03ac\u03bd\u03c4\u03b1 \u03bc\u03af\u03b1 \u03ba\u03b1\u03c4\u03b7\u03b3\u03bf\u03c1\u03af\u03b1, \u03b4\u03b7\u03bc\u03b9\u03bf\u03c5\u03c1\u03b3\u03b5\u03af \u03c8\u03b5\u03c5\u03b4\u03ae \u03b2\u03b5\u03b2\u03b1\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b1\u03c0\u03cc \u03c4\u03bf\u03bd \u03af\u03b4\u03b9\u03bf \u03c4\u03bf\u03c5 \u03c4\u03bf\u03bd \u03c3\u03c7\u03b5\u03b4\u03b9\u03b1\u03c3\u03bc\u03cc.<\/p>\n<aside class=\"td-article-note\">\n<p><strong>\u039f null state \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03ac\u03b4\u03b5\u03b9\u03b1 \u03b3\u03b9\u03b1 \u03c4\u03c5\u03c6\u03bb\u03ae \u03b1\u03c5\u03c4\u03bf\u03bc\u03b1\u03c4\u03bf\u03c0\u03bf\u03af\u03b7\u03c3\u03b7.<\/strong> \u03a3\u03b7\u03bc\u03b1\u03c4\u03bf\u03b4\u03bf\u03c4\u03b5\u03af candidate-set failure \u03bc\u03ad\u03c3\u03b1 \u03c3\u03c4\u03bf \u03bc\u03b1\u03b8\u03b7\u03bc\u03b1\u03c4\u03b9\u03ba\u03cc \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf. \u03a3\u03b5 \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03ae, \u03b7 \u03b1\u03bd\u03c4\u03af\u03c3\u03c4\u03bf\u03b9\u03c7\u03b7 \u03b1\u03c0\u03bf\u03c7\u03ae \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03c3\u03c5\u03bd\u03b4\u03ad\u03b5\u03c4\u03b1\u03b9 \u03bc\u03b5 \u03c3\u03b1\u03c6\u03ad\u03c2 handoff, \u03ba\u03b1\u03c4\u03b1\u03b3\u03c1\u03b1\u03c6\u03ae \u03b1\u03b9\u03c4\u03af\u03b1\u03c2 \u03ba\u03b1\u03b9 \u03c0\u03bf\u03bb\u03b9\u03c4\u03b9\u03ba\u03ae \u03b3\u03b9\u03b1 \u03c4\u03bf \u03c0\u03bf\u03b9\u03bf\u03c2 \u03b5\u03c0\u03b9\u03b2\u03b5\u03b2\u03b1\u03b9\u03ce\u03bd\u03b5\u03b9 \u03c4\u03b7\u03bd \u03b5\u03c0\u03cc\u03bc\u03b5\u03bd\u03b7 \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b1.<\/p>\n<\/aside>\n<h2 id=\"deytero-stadio-calibration\">\u0393\u03b9\u03b1\u03c4\u03af \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03b4\u03b5\u03cd\u03c4\u03b5\u03c1\u03bf \u03c3\u03c4\u03ac\u03b4\u03b9\u03bf calibration<\/h2>\n<p>\u0397 posterior mass \u03c4\u03bf\u03c5 \u03b5\u03c0\u03b9\u03bb\u03b5\u03b3\u03bc\u03ad\u03bd\u03bf\u03c5 candidate \u03b5\u03af\u03bd\u03b1\u03b9 \u03b5\u03bd\u03b4\u03bf\u03c0\u03b1\u03c1\u03b1\u03b4\u03b5\u03b9\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ae \u03bc\u03ad\u03c4\u03c1\u03b7\u03c3\u03b7: \u03b4\u03b5\u03af\u03c7\u03bd\u03b5\u03b9 \u03c0\u03cc\u03c3\u03bf \u03b9\u03c3\u03c7\u03c5\u03c1\u03ae \u03b5\u03af\u03bd\u03b1\u03b9 \u03b7 \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae \u03b1\u03c0\u03ad\u03bd\u03b1\u03bd\u03c4\u03b9 \u03c3\u03c4\u03b9\u03c2 \u03ac\u03bb\u03bb\u03b5\u03c2 \u03b1\u03c0\u03b1\u03bd\u03c4\u03ae\u03c3\u03b5\u03b9\u03c2 \u03ba\u03b1\u03b9 \u03c3\u03c4\u03bf\u03bd null state. \u039f\u03b9 calibration metrics \u03b6\u03b7\u03c4\u03bf\u03cd\u03bd \u03ba\u03ac\u03c4\u03b9 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03cc: \u03b1\u03bd\u03ac\u03bc\u03b5\u03c3\u03b1 \u03c3\u03b5 \u03c0\u03bf\u03bb\u03bb\u03ad\u03c2 \u03c0\u03c1\u03bf\u03b2\u03bb\u03ad\u03c8\u03b5\u03b9\u03c2 \u03bc\u03b5 \u03c0\u03b1\u03c1\u03cc\u03bc\u03bf\u03b9\u03bf score, \u03c0\u03bf\u03b9\u03bf \u03c0\u03bf\u03c3\u03bf\u03c3\u03c4\u03cc \u03b5\u03af\u03bd\u03b1\u03b9 \u03c0\u03c1\u03ac\u03b3\u03bc\u03b1\u03c4\u03b9 \u03c3\u03c9\u03c3\u03c4\u03cc;<\/p>\n<p>\u0397 \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae \u03c4\u03bf\u03c5 \u03bc\u03ad\u03b3\u03b9\u03c3\u03c4\u03bf\u03c5 \u03ba\u03b1\u03b9 \u03b7 \u03c0\u03c1\u03bf\u03c3\u03b5\u03b3\u03b3\u03b9\u03c3\u03c4\u03b9\u03ba\u03ae \u03c6\u03cd\u03c3\u03b7 \u03c4\u03bf\u03c5 evidence model \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03bf\u03c5\u03bd \u03cc\u03c4\u03b9 \u03c4\u03b1 \u03b4\u03cd\u03bf \u03bc\u03b5\u03b3\u03ad\u03b8\u03b7 \u03b4\u03b5\u03bd \u03c3\u03c5\u03bc\u03c0\u03af\u03c0\u03c4\u03bf\u03c5\u03bd \u03b1\u03bd\u03b1\u03b3\u03ba\u03b1\u03c3\u03c4\u03b9\u03ba\u03ac. \u0393\u03b9\u2019 \u03b1\u03c5\u03c4\u03cc \u03bf\u03b9 \u03b5\u03c1\u03b5\u03c5\u03bd\u03b7\u03c4\u03ad\u03c2 \u03b5\u03c6\u03b1\u03c1\u03bc\u03cc\u03b6\u03bf\u03c5\u03bd \u03b4\u03b5\u03cd\u03c4\u03b5\u03c1\u03b7 Platt-style mapping \u03c0\u03ac\u03bd\u03c9 \u03c3\u03c4\u03b7\u03bd posterior mass \u03c4\u03bf\u03c5 \u03bd\u03b9\u03ba\u03b7\u03c4\u03ae. \u0397 \u03c7\u03b1\u03c1\u03c4\u03bf\u03b3\u03c1\u03ac\u03c6\u03b7\u03c3\u03b7 \u03c0\u03c1\u03bf\u03c3\u03b1\u03c1\u03bc\u03cc\u03b6\u03b5\u03c4\u03b1\u03b9 \u03c3\u03b5 held-out calibration split \u03ba\u03b1\u03b9 \u03bc\u03b5\u03c4\u03b1\u03c4\u03c1\u03ad\u03c0\u03b5\u03b9 \u03c4\u03bf aggregated score \u03c3\u03b5 binary correctness probability.<\/p>\n<p>\u039f\u03b9 ablations \u03b4\u03b5\u03af\u03c7\u03bd\u03bf\u03c5\u03bd \u03cc\u03c4\u03b9 \u03c4\u03b1 \u03c3\u03c4\u03ac\u03b4\u03b9\u03b1 \u03ba\u03ac\u03bd\u03bf\u03c5\u03bd \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03ae \u03b4\u03bf\u03c5\u03bb\u03b5\u03b9\u03ac. \u03a4\u03bf confidence evidence \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b1\u03bb\u03bb\u03ac\u03be\u03b5\u03b9 \u03c4\u03bf candidate scoring \u03ba\u03b1\u03b9 \u03bc\u03b5\u03c1\u03b9\u03ba\u03ad\u03c2 \u03c6\u03bf\u03c1\u03ad\u03c2 \u03c4\u03b7\u03bd \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae. \u03a7\u03c9\u03c1\u03af\u03c2 \u03c4\u03bf \u03c4\u03b5\u03bb\u03b5\u03c5\u03c4\u03b1\u03af\u03bf calibration, \u03cc\u03bc\u03c9\u03c2, \u03c4\u03bf raw posterior \u03b5\u03af\u03c7\u03b5 \u03c5\u03c8\u03b7\u03bb\u03cc ECE \u03ba\u03b1\u03b9 Brier score \u03c3\u03b5 \u03b1\u03c1\u03ba\u03b5\u03c4\u03ad\u03c2 \u03c1\u03c5\u03b8\u03bc\u03af\u03c3\u03b5\u03b9\u03c2. \u0397 \u03c4\u03b5\u03bb\u03b9\u03ba\u03ae monotonic mapping \u03b4\u03b9\u03bf\u03c1\u03b8\u03ce\u03bd\u03b5\u03b9 \u03c4\u03b7\u03bd \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03b1 \u03c7\u03c9\u03c1\u03af\u03c2 \u03bd\u03b1 \u03b1\u03bb\u03bb\u03ac\u03b6\u03b5\u03b9 \u03c4\u03bf\u03bd \u03b5\u03c0\u03b9\u03bb\u03b5\u03b3\u03bc\u03ad\u03bd\u03bf candidate.<\/p>\n<h2 id=\"peiramata-montela-datasets-metrics\">\u03a0\u03ce\u03c2 \u03c3\u03c7\u03b5\u03b4\u03b9\u03ac\u03c3\u03c4\u03b7\u03ba\u03b1\u03bd \u03c4\u03b1 \u03c0\u03b5\u03b9\u03c1\u03ac\u03bc\u03b1\u03c4\u03b1<\/h2>\n<p>\u0397 \u03bc\u03b5\u03bb\u03ad\u03c4\u03b7 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03b5\u03af \u03c4\u03c1\u03af\u03b1 instruction-tuned open-weight models \u03bc\u03ad\u03c4\u03c1\u03b9\u03b1\u03c2 \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03b1\u03c2: Qwen2.5-7B-Instruct, Mistral-7B-Instruct-v0.3 \u03ba\u03b1\u03b9 Gemma-2-9B-it. \u03a4\u03b1 benchmarks \u03b5\u03af\u03bd\u03b1\u03b9 GSM8K, SVAMP \u03ba\u03b1\u03b9 GSM-Hard. \u0397 \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae \u03ba\u03c1\u03b1\u03c4\u03ac \u03c4\u03bf task \u03c3\u03b5 \u03b1\u03c1\u03b9\u03b8\u03bc\u03b7\u03c4\u03b9\u03ba\u03cc reasoning, \u03cc\u03c0\u03bf\u03c5 \u03b7 \u03c4\u03b5\u03bb\u03b9\u03ba\u03ae \u03bf\u03c1\u03b8\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b5\u03bb\u03ad\u03b3\u03c7\u03b5\u03c4\u03b1\u03b9 \u03bc\u03b5 \u03b5\u03be\u03b1\u03b3\u03c9\u03b3\u03ae \u03ba\u03b1\u03b9 \u03b1\u03ba\u03c1\u03b9\u03b2\u03ae \u03c3\u03cd\u03b3\u03ba\u03c1\u03b9\u03c3\u03b7 \u03c4\u03b7\u03c2 \u03b1\u03c1\u03b9\u03b8\u03bc\u03b7\u03c4\u03b9\u03ba\u03ae\u03c2 \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7\u03c2.<\/p>\n<p>\u039f\u03b9 \u03c3\u03c5\u03b3\u03ba\u03c1\u03af\u03c3\u03b5\u03b9\u03c2 \u03c0\u03b5\u03c1\u03b9\u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03bf\u03c5\u03bd vanilla verbalized confidence, mean confidence, self-consistency, answer-entropy confidence, Top-K prompting \u03ba\u03b1\u03b9 SteerConf. \u038c\u03c0\u03bf\u03c5 \u03ae\u03c4\u03b1\u03bd \u03b5\u03c6\u03b9\u03ba\u03c4\u03cc, \u03bf\u03b9 \u03bc\u03ad\u03b8\u03bf\u03b4\u03bf\u03b9 \u03b5\u03c5\u03b8\u03c5\u03b3\u03c1\u03b1\u03bc\u03bc\u03af\u03c3\u03c4\u03b7\u03ba\u03b1\u03bd \u03c3\u03b5 \u03c0\u03ad\u03bd\u03c4\u03b5 observations: DirEAG, mean confidence \u03ba\u03b1\u03b9 SteerConf \u03bc\u03bf\u03b9\u03c1\u03ac\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 \u03c4\u03b1 \u03af\u03b4\u03b9\u03b1 \u03c0\u03ad\u03bd\u03c4\u03b5 answer-confidence pairs, \u03b5\u03bd\u03ce self-consistency \u03ba\u03b1\u03b9 entropy \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03bf\u03cd\u03bd \u03c0\u03ad\u03bd\u03c4\u03b5 vanilla samples.<\/p>\n<p>\u0397 \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7 \u03c0\u03b5\u03c1\u03b9\u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03b5\u03b9 accuracy, Expected Calibration Error, Brier score, AUROC, PR-P \u03ba\u03b1\u03b9 PR-N. \u03a4\u03bf ECE \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03b5\u03af \u03b4\u03ad\u03ba\u03b1 bins \u03af\u03c3\u03bf\u03c5 \u03c0\u03bb\u03ac\u03c4\u03bf\u03c5\u03c2. \u03a4\u03b1 evaluated examples \u03b1\u03c0\u03bf\u03ba\u03bb\u03b5\u03af\u03bf\u03bd\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03c4\u03bf fitting \u03c4\u03c9\u03bd \u03c0\u03b1\u03c1\u03b1\u03bc\u03ad\u03c4\u03c1\u03c9\u03bd \u03ba\u03b1\u03b9 \u03b1\u03c0\u03cc \u03c4\u03b7\u03bd \u03c4\u03b5\u03bb\u03b9\u03ba\u03ae Platt calibration\u00b7 \u03b3\u03b9\u03b1 SVAMP \u03ba\u03b1\u03b9 GSM-Hard \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03b5\u03af\u03c4\u03b1\u03b9 five-fold cross-fitting.<\/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 \u03c3\u03b7\u03bc\u03b5\u03af\u03b1 \u03b1\u03c0\u03cc \u03c4\u03b1 \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03ad\u03c3\u03bc\u03b1\u03c4\u03b1<\/p>\n<p class=\"td-chart-subtitle\">\u039f\u03b9 \u03b4\u03cd\u03bf \u03c0\u03c1\u03ce\u03c4\u03b5\u03c2 \u03c4\u03b9\u03bc\u03ad\u03c2 \u03b5\u03af\u03bd\u03b1\u03b9 \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03b1 model\u2013dataset \u03c0\u03b1\u03c1\u03b1\u03b4\u03b5\u03af\u03b3\u03bc\u03b1\u03c4\u03b1. \u03a4\u03b1 ranges \u03c0\u03c1\u03bf\u03ad\u03c1\u03c7\u03bf\u03bd\u03c4\u03b1\u03b9 \u03b1\u03c0\u03cc \u03c4\u03bf diagnostic \u03ba\u03b1\u03b9 \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03b3\u03b5\u03bd\u03b9\u03ba\u03ae \u03c0\u03c1\u03cc\u03b2\u03bb\u03b5\u03c8\u03b7 \u03b3\u03b9\u03b1 \u03ac\u03bb\u03bb\u03b1 workloads.<\/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\">0,0176<\/span><span class=\"td-metric-label\">ECE \u03c4\u03bf\u03c5 DirEAG \u03c3\u03c4\u03bf Qwen2.5-7B \u03bc\u03b5 SVAMP<\/span><\/div>\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">0,0387<\/span><span class=\"td-metric-label\">ECE \u03c4\u03bf\u03c5 DirEAG \u03c3\u03c4\u03bf Mistral-7B \u03bc\u03b5 GSM-Hard<\/span><\/div>\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">0,021\u20130,086<\/span><span class=\"td-metric-label\">\u03b1\u03c0\u03cc\u03bb\u03c5\u03c4\u03b7 \u03bc\u03b5\u03af\u03c9\u03c3\u03b7 Brier \u03ad\u03bd\u03b1\u03bd\u03c4\u03b9 constant base-rate \u03c3\u03c4\u03b1 9 settings<\/span><\/div>\n<div class=\"td-metric-card\"><span class=\"td-metric-value\">+0,229\u20130,362<\/span><span class=\"td-metric-label\">AUROC gain \u03ad\u03bd\u03b1\u03bd\u03c4\u03b9 \u03c4\u03bf\u03c5 constant predictor \u03c3\u03c4\u03b1 diagnostics<\/span><\/div>\n<\/div>\n<\/div>\n<\/div>\n<aside class=\"td-article-note\">\n<p><strong>\u0397 \u03b5\u03c1\u03b3\u03b1\u03c3\u03af\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 arXiv preprint v1 \u03c4\u03b7\u03c2 21\u03b7\u03c2 \u0391\u03c5\u03b3\u03bf\u03cd\u03c3\u03c4\u03bf\u03c5 2026.<\/strong> \u03a4\u03b1 \u03b5\u03bd\u03bd\u03ad\u03b1 model\u2013dataset settings \u03b1\u03c6\u03bf\u03c1\u03bf\u03cd\u03bd \u03b1\u03c1\u03b9\u03b8\u03bc\u03b7\u03c4\u03b9\u03ba\u03ac benchmarks \u03ba\u03b1\u03b9 open-weight \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03b1 \u03bc\u03ad\u03c4\u03c1\u03b9\u03b1\u03c2 \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03b1\u03c2. \u03a4\u03b1 scores \u03b4\u03b5\u03bd \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03bf\u03cd\u03bd \u03b5\u03b3\u03b3\u03cd\u03b7\u03c3\u03b7 \u03b3\u03b9\u03b1 frontier models, open-ended generation \u03ae \u03b5\u03c4\u03b1\u03b9\u03c1\u03b9\u03ba\u03ac \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03b1.<\/p>\n<\/aside>\n<h2 id=\"apotelesmata-kai-sosti-anagnosi\">\u03a4\u03b9 \u03b4\u03b5\u03af\u03c7\u03bd\u03bf\u03c5\u03bd \u03c4\u03b1 \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03ad\u03c3\u03bc\u03b1\u03c4\u03b1 \u03ba\u03b1\u03b9 \u03c0\u03ce\u03c2 \u03b4\u03b9\u03b1\u03b2\u03ac\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9<\/h2>\n<p>\u03a3\u03c4\u03bf SVAMP \u03bc\u03b5 Qwen2.5-7B, \u03c4\u03bf ECE \u03b5\u03af\u03bd\u03b1\u03b9 0,0176 \u03b3\u03b9\u03b1 \u03c4\u03bf DirEAG, \u03ad\u03bd\u03b1\u03bd\u03c4\u03b9 0,0477 \u03b3\u03b9\u03b1 mean confidence \u03ba\u03b1\u03b9 0,0315 \u03b3\u03b9\u03b1 SteerConf. \u03a3\u03c4\u03bf GSM-Hard \u03bc\u03b5 Mistral-7B, \u03c4\u03bf ECE \u03b5\u03af\u03bd\u03b1\u03b9 0,0387, \u03ad\u03bd\u03b1\u03bd\u03c4\u03b9 0,1738 \u03ba\u03b1\u03b9 0,1886 \u03b1\u03bd\u03c4\u03af\u03c3\u03c4\u03bf\u03b9\u03c7\u03b1. \u03a3\u03c5\u03bd\u03bf\u03bb\u03b9\u03ba\u03ac, \u03c4\u03bf paper \u03b1\u03bd\u03b1\u03c6\u03ad\u03c1\u03b5\u03b9 \u03c7\u03b1\u03bc\u03b7\u03bb\u03cc\u03c4\u03b5\u03c1\u03bf ECE \u03b1\u03c0\u03cc direct confidence averaging \u03ba\u03b1\u03b9 SteerConf-style aggregation \u03c3\u03c4\u03b9\u03c2 \u03c0\u03b5\u03c1\u03b9\u03c3\u03c3\u03cc\u03c4\u03b5\u03c1\u03b5\u03c2 \u03c1\u03c5\u03b8\u03bc\u03af\u03c3\u03b5\u03b9\u03c2.<\/p>\n<p>\u0397 \u03b5\u03b9\u03ba\u03cc\u03bd\u03b1 \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u00ab\u03ba\u03b5\u03c1\u03b4\u03af\u03b6\u03b5\u03b9 \u03c0\u03b1\u03bd\u03c4\u03bf\u03cd\u00bb. \u03a3\u03c4\u03bf GSM8K \u03bc\u03b5 Qwen2.5-7B, \u03b7 self-consistency \u03ad\u03c7\u03b5\u03b9 accuracy 0,9265 \u03ba\u03b1\u03b9 ECE 0,0170, \u03b5\u03bd\u03ce \u03c4\u03bf DirEAG 0,8908 \u03ba\u03b1\u03b9 0,0280. \u03a3\u03b5 \u03ac\u03bb\u03bb\u03b1 blocks, \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03ad\u03c2 \u03bc\u03ad\u03b8\u03bf\u03b4\u03bf\u03b9 \u03c0\u03b1\u03c1\u03b1\u03bc\u03ad\u03bd\u03bf\u03c5\u03bd \u03b1\u03bd\u03c4\u03b1\u03b3\u03c9\u03bd\u03b9\u03c3\u03c4\u03b9\u03ba\u03ad\u03c2 \u03c3\u03b5 AUROC \u03ae PR-N. \u0397 \u03c3\u03c9\u03c3\u03c4\u03ae \u03b1\u03bd\u03ac\u03b3\u03bd\u03c9\u03c3\u03b7 \u03b5\u03af\u03bd\u03b1\u03b9 \u03cc\u03c4\u03b9 \u03b7 \u03bc\u03ad\u03b8\u03bf\u03b4\u03bf\u03c2 \u03b2\u03b5\u03bb\u03c4\u03b9\u03ce\u03bd\u03b5\u03b9 \u03c3\u03c5\u03c7\u03bd\u03ac \u03c4\u03b7\u03bd \u03b1\u03be\u03b9\u03bf\u03c0\u03bf\u03af\u03b7\u03c3\u03b7 \u03ba\u03b1\u03b9 \u03c4\u03b7 \u03b2\u03b1\u03b8\u03bc\u03bf\u03bd\u03cc\u03bc\u03b7\u03c3\u03b7 verbalized confidence, \u03cc\u03c7\u03b9 \u03cc\u03c4\u03b9 \u03c5\u03c0\u03b5\u03c1\u03ad\u03c7\u03b5\u03b9 \u03ba\u03b1\u03b8\u03bf\u03bb\u03b9\u03ba\u03ac \u03c3\u03b5 \u03ba\u03ac\u03b8\u03b5 metric.<\/p>\n<p>\u03a4\u03bf diagnostic \u03b1\u03c0\u03ad\u03bd\u03b1\u03bd\u03c4\u03b9 \u03c3\u03b5 constant base-rate predictor \u03b5\u03af\u03bd\u03b1\u03b9 \u03b5\u03c0\u03af\u03c3\u03b7\u03c2 \u03c3\u03b7\u03bc\u03b1\u03bd\u03c4\u03b9\u03ba\u03cc. \u03a4\u03bf DirEAG \u03b5\u03af\u03c7\u03b5 \u03c7\u03b1\u03bc\u03b7\u03bb\u03cc\u03c4\u03b5\u03c1\u03bf Brier score \u03ba\u03b1\u03b9 \u03c3\u03c4\u03b1 \u03b5\u03bd\u03bd\u03ad\u03b1 settings, \u03bc\u03b5 \u03b1\u03c0\u03cc\u03bb\u03c5\u03c4\u03b7 \u03bc\u03b5\u03af\u03c9\u03c3\u03b7 0,021 \u03ad\u03c9\u03c2 0,086, \u03ba\u03b1\u03b9 AUROC gain 0,229 \u03ad\u03c9\u03c2 0,362. \u0391\u03c5\u03c4\u03cc \u03b4\u03b5\u03af\u03c7\u03bd\u03b5\u03b9 \u03cc\u03c4\u03b9 \u03c4\u03b1 observations \u03c0\u03b5\u03c1\u03b9\u03ad\u03c7\u03bf\u03c5\u03bd instance-level \u03c0\u03bb\u03b7\u03c1\u03bf\u03c6\u03bf\u03c1\u03af\u03b1 \u03ba\u03b1\u03b9 \u03b4\u03b5\u03bd \u03b1\u03bd\u03b1\u03c0\u03b1\u03c1\u03ac\u03b3\u03bf\u03c5\u03bd \u03b1\u03c0\u03bb\u03ce\u03c2 \u03c4\u03b7 \u03bc\u03ad\u03c3\u03b7 accuracy \u03c4\u03bf\u03c5 dataset.<\/p>\n<h2 id=\"production-ai-workflow\">\u03a4\u03b9 \u03b1\u03bb\u03bb\u03ac\u03b6\u03b5\u03b9 \u03c3\u03b5 \u03ad\u03bd\u03b1 production AI workflow<\/h2>\n<p>\u0397 \u03c0\u03c1\u03b1\u03ba\u03c4\u03b9\u03ba\u03ae \u03b1\u03be\u03af\u03b1 \u03b4\u03b5\u03bd \u03b5\u03af\u03bd\u03b1\u03b9 \u03bd\u03b1 \u03b1\u03bd\u03c4\u03b9\u03b3\u03c1\u03b1\u03c6\u03bf\u03cd\u03bd \u03bc\u03b7\u03c7\u03b1\u03bd\u03b9\u03ba\u03ac \u03bf\u03b9 \u03b5\u03be\u03b9\u03c3\u03ce\u03c3\u03b5\u03b9\u03c2. \u0395\u03af\u03bd\u03b1\u03b9 \u03bd\u03b1 \u03b4\u03b9\u03b1\u03c7\u03c9\u03c1\u03b9\u03c3\u03c4\u03bf\u03cd\u03bd \u03c4\u03ad\u03c3\u03c3\u03b5\u03c1\u03b1 \u03b5\u03c0\u03af\u03c0\u03b5\u03b4\u03b1: \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03ae \u03c0\u03bf\u03bb\u03bb\u03b1\u03c0\u03bb\u03ce\u03bd observations, \u03b4\u03b9\u03cc\u03c1\u03b8\u03c9\u03c3\u03b7 \u03c3\u03c5\u03c3\u03c4\u03b7\u03bc\u03b1\u03c4\u03b9\u03ba\u03ce\u03bd biases, \u03c3\u03c5\u03b3\u03ba\u03ad\u03bd\u03c4\u03c1\u03c9\u03c3\u03b7 \u03c4\u03b5\u03ba\u03bc\u03b7\u03c1\u03af\u03c9\u03bd \u03bc\u03b5 \u03c1\u03b7\u03c4\u03ae \u03b4\u03c5\u03bd\u03b1\u03c4\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b1\u03c0\u03bf\u03c7\u03ae\u03c2 \u03ba\u03b1\u03b9 \u03c4\u03b5\u03bb\u03b9\u03ba\u03ae \u03b2\u03b1\u03b8\u03bc\u03bf\u03bd\u03cc\u03bc\u03b7\u03c3\u03b7 \u03c0\u03ac\u03bd\u03c9 \u03c3\u03b5 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac labeled outcomes. \u0397 \u03b4\u03b9\u03ac\u03ba\u03c1\u03b9\u03c3\u03b7 evidence \u03b1\u03c0\u03cc \u03c4\u03b5\u03bb\u03b9\u03ba\u03ae \u03c0\u03b9\u03b8\u03b1\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1 \u03c3\u03c5\u03bc\u03c0\u03bb\u03b7\u03c1\u03ce\u03bd\u03b5\u03b9 \u03c4\u03b7\u03bd \u03b1\u03bd\u03ac\u03b3\u03ba\u03b7 \u03b3\u03b9\u03b1 <a href=\"https:\/\/twodots.gr\/consilience-ai-agent-communication-control\/\">\u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03b7 \u03c3\u03cd\u03bd\u03b8\u03b5\u03c3\u03b7 \u03c0\u03bf\u03bb\u03bb\u03ce\u03bd AI \u03c3\u03b7\u03bc\u03ac\u03c4\u03c9\u03bd<\/a>.<\/p>\n<p>\u03a3\u03b5 e-commerce \u03ae customer support, \u03ad\u03bd\u03b1 \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b5\u03ba\u03c4\u03b9\u03bc\u03ac \u03c0\u03c1\u03cc\u03b8\u03b5\u03c3\u03b7 \u03bc\u03b7\u03bd\u03cd\u03bc\u03b1\u03c4\u03bf\u03c2, \u03ba\u03b1\u03c4\u03b7\u03b3\u03bf\u03c1\u03af\u03b1 \u03b5\u03c0\u03b9\u03c3\u03c4\u03c1\u03bf\u03c6\u03ae\u03c2 \u03ae \u03ba\u03b1\u03c4\u03b1\u03bb\u03bb\u03b7\u03bb\u03cc\u03c4\u03b7\u03c4\u03b1 \u03c0\u03c1\u03bf\u03ca\u03cc\u03bd\u03c4\u03bf\u03c2. \u0394\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03ac prompts \u03ae samples \u03bc\u03c0\u03bf\u03c1\u03bf\u03cd\u03bd \u03bd\u03b1 \u03bb\u03b5\u03b9\u03c4\u03bf\u03c5\u03c1\u03b3\u03ae\u03c3\u03bf\u03c5\u03bd \u03c9\u03c2 \u03c3\u03ae\u03bc\u03b1\u03c4\u03b1, \u03b1\u03bb\u03bb\u03ac \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7 \u03b1\u03bd\u03ac prompt \u03ba\u03b1\u03b9 workload. \u0388\u03bd\u03b1 self-reported \u00ab90%\u00bb \u03b4\u03b5\u03bd \u03b9\u03c3\u03bf\u03b4\u03c5\u03bd\u03b1\u03bc\u03b5\u03af \u03bc\u03b5 SLA \u03c0\u03bf\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1\u03c2 \u03bf\u03cd\u03c4\u03b5 \u03c5\u03c0\u03bf\u03ba\u03b1\u03b8\u03b9\u03c3\u03c4\u03ac \u03c4\u03bf <a href=\"https:\/\/twodots.gr\/ai-oikonomika-audit-trail-kdaf\/\">audit trail \u03b3\u03b9\u03b1 \u03b1\u03c0\u03bf\u03c6\u03ac\u03c3\u03b5\u03b9\u03c2 AI<\/a>.<\/p>\n<p>\u0397 accuracy \u03b1\u03c0\u03b1\u03bd\u03c4\u03ac \u00ab\u03c0\u03cc\u03c3\u03bf \u03c3\u03c5\u03c7\u03bd\u03ac \u03c0\u03ad\u03c4\u03c5\u03c7\u03b5;\u00bb. \u0397 calibration \u03b1\u03c0\u03b1\u03bd\u03c4\u03ac \u00ab\u03cc\u03c4\u03b1\u03bd \u03b4\u03ae\u03bb\u03c9\u03c3\u03b5 \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03b7 \u03b2\u03b5\u03b2\u03b1\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1, \u03c0\u03cc\u03c3\u03bf \u03b1\u03be\u03b9\u03cc\u03c0\u03b9\u03c3\u03c4\u03b7 \u03ae\u03c4\u03b1\u03bd \u03b1\u03c5\u03c4\u03ae \u03b7 \u03b4\u03ae\u03bb\u03c9\u03c3\u03b7;\u00bb. \u03a4\u03bf <a href=\"https:\/\/twodots.gr\/ai-agents-behavioral-testing\/\">behavioral testing \u03b3\u03b9\u03b1 AI agents<\/a> \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03bc\u03b5\u03c4\u03c1\u03ac \u03ba\u03b1\u03b9 \u03c4\u03b1 \u03b4\u03cd\u03bf, \u03b5\u03c0\u03b5\u03b9\u03b4\u03ae \u03ad\u03bd\u03b1 threshold \u03b1\u03c5\u03c4\u03bf\u03bc\u03b1\u03c4\u03b9\u03c3\u03bc\u03bf\u03cd \u03b5\u03af\u03bd\u03b1\u03b9 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ad\u03c2 \u03bc\u03cc\u03bd\u03bf \u03cc\u03c4\u03b1\u03bd \u03c4\u03bf score \u03ad\u03c7\u03b5\u03b9 \u03b5\u03bb\u03b5\u03b3\u03c7\u03b8\u03b5\u03af \u03c3\u03c4\u03bf \u03af\u03b4\u03b9\u03bf \u03b5\u03af\u03b4\u03bf\u03c2 \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03c9\u03bd \u03ba\u03b1\u03b9 \u03ba\u03cc\u03c3\u03c4\u03bf\u03c5\u03c2 \u03c3\u03c6\u03ac\u03bb\u03bc\u03b1\u03c4\u03bf\u03c2.<\/p>\n<div class=\"td-decision-band\">\n<p class=\"td-decision-band-kicker\">\u039a\u03b1\u03bd\u03cc\u03bd\u03b1\u03c2 \u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03b9\u03ba\u03ae\u03c2 \u03b1\u03c0\u03cc\u03c6\u03b1\u03c3\u03b7\u03c2<\/p>\n<h3>\u03a0\u03cc\u03c4\u03b5 \u03ad\u03bd\u03b1 calibrated confidence \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b5\u03bd\u03b5\u03c1\u03b3\u03bf\u03c0\u03bf\u03b9\u03ae\u03c3\u03b5\u03b9 \u03b1\u03c5\u03c4\u03bf\u03bc\u03b1\u03c4\u03b9\u03c3\u03bc\u03cc<\/h3>\n<p>\u039c\u03cc\u03bd\u03bf \u03cc\u03c4\u03b1\u03bd \u03c4\u03bf confidence bucket \u03ad\u03c7\u03b5\u03b9 \u03b5\u03c0\u03b1\u03bb\u03b7\u03b8\u03b5\u03c5\u03c4\u03b5\u03af \u03c3\u03b5 labeled outcomes \u03c4\u03bf\u03c5 \u03af\u03b4\u03b9\u03bf\u03c5 workload, \u03c4\u03bf null \u03ae abstain path \u03bf\u03b4\u03b7\u03b3\u03b5\u03af \u03c3\u03b5 \u03c3\u03b1\u03c6\u03ad\u03c2 human handoff, ECE \u03ba\u03b1\u03b9 Brier \u03c0\u03b1\u03c1\u03b1\u03bc\u03ad\u03bd\u03bf\u03c5\u03bd \u03b5\u03bd\u03c4\u03cc\u03c2 \u03c3\u03c5\u03bc\u03c6\u03c9\u03bd\u03b7\u03bc\u03ad\u03bd\u03c9\u03bd \u03bf\u03c1\u03af\u03c9\u03bd \u03bc\u03b5\u03c4\u03ac \u03b1\u03c0\u03cc drift checks \u03ba\u03b1\u03b9 \u03bf\u03b9 \u03b5\u03bd\u03ad\u03c1\u03b3\u03b5\u03b9\u03b5\u03c2 \u03c5\u03c8\u03b7\u03bb\u03bf\u03cd \u03b1\u03bd\u03c4\u03af\u03ba\u03c4\u03c5\u03c0\u03bf\u03c5 \u03b4\u03b9\u03b1\u03c4\u03b7\u03c1\u03bf\u03cd\u03bd \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03b7 \u03ad\u03b3\u03ba\u03c1\u03b9\u03c3\u03b7. \u0388\u03bb\u03bb\u03b5\u03b9\u03c8\u03b7 \u03c4\u03b5\u03ba\u03bc\u03b7\u03c1\u03af\u03bf\u03c5 \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u03b1\u03c0\u03bf\u03c7\u03ae, \u03cc\u03c7\u03b9 \u03b1\u03bd\u03b1\u03b3\u03ba\u03b1\u03c3\u03c4\u03b9\u03ba\u03ae \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae.<\/p>\n<\/div>\n<h2 id=\"pilot-vathmonomisis-vevaiotitas\">\u03a0\u03ce\u03c2 \u03c3\u03c4\u03ae\u03bd\u03b5\u03c4\u03b1\u03b9 \u03ad\u03bd\u03b1 pilot \u03b2\u03b1\u03b8\u03bc\u03bf\u03bd\u03cc\u03bc\u03b7\u03c3\u03b7\u03c2 \u03b2\u03b5\u03b2\u03b1\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1\u03c2<\/h2>\n<p>\u03a4\u03bf DirEAG \u03b5\u03af\u03bd\u03b1\u03b9 research method, \u03cc\u03c7\u03b9 plug-and-play policy engine. \u0388\u03bd\u03b1 \u03c7\u03c1\u03ae\u03c3\u03b9\u03bc\u03bf pilot \u03be\u03b5\u03ba\u03b9\u03bd\u03ac \u03b1\u03c0\u03cc \u03bc\u03af\u03b1 \u03b1\u03c0\u03cc\u03c6\u03b1\u03c3\u03b7 \u03bc\u03b5 \u03b3\u03bd\u03c9\u03c3\u03c4\u03cc ground truth \u03ba\u03b1\u03b9 \u03bc\u03b5\u03c4\u03c1\u03ae\u03c3\u03b9\u03bc\u03bf \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2 \u03bb\u03ac\u03b8\u03bf\u03c5\u03c2. \u0393\u03b9\u03b1 \u03c0\u03b1\u03c1\u03ac\u03b4\u03b5\u03b9\u03b3\u03bc\u03b1, \u03c3\u03b5 <a href=\"https:\/\/twodots.gr\/6-ai-agents-customer-support-gia-epicheiriseis\/\">AI agents \u03b3\u03b9\u03b1 customer support<\/a> \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b2\u03b1\u03b8\u03bc\u03bf\u03bd\u03bf\u03bc\u03b5\u03af\u03c4\u03b1\u03b9 \u03b7 \u03c4\u03b1\u03be\u03b9\u03bd\u03cc\u03bc\u03b7\u03c3\u03b7 \u03b1\u03b9\u03c4\u03ae\u03bc\u03b1\u03c4\u03bf\u03c2, \u03b5\u03bd\u03ce \u03b7 \u03b1\u03c0\u03bf\u03c3\u03c4\u03bf\u03bb\u03ae \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7\u03c2 \u03ba\u03b1\u03b9 \u03ba\u03ac\u03b8\u03b5 \u03b1\u03bb\u03bb\u03b1\u03b3\u03ae \u03bb\u03bf\u03b3\u03b1\u03c1\u03b9\u03b1\u03c3\u03bc\u03bf\u03cd \u03bc\u03ad\u03bd\u03bf\u03c5\u03bd \u03c0\u03af\u03c3\u03c9 \u03b1\u03c0\u03cc human gate.<\/p>\n<div class=\"td-step-list\">\n<p class=\"td-step-list-title\">\u0391\u03c0\u03cc \u03c4\u03bf confidence score \u03c3\u03b5 \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03bf production threshold<\/p>\n<ol>\n<li><span class=\"td-step-kicker\">Step 1<\/span><strong>\u039f\u03c1\u03af\u03c3\u03c4\u03b5 \u03bc\u03af\u03b1 \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03b7 \u03b1\u03c0\u03cc\u03c6\u03b1\u03c3\u03b7 \u03ba\u03b1\u03b9 \u03c4\u03bf \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2 \u03c3\u03c6\u03ac\u03bb\u03bc\u03b1\u03c4\u03bf\u03c2<\/strong>\n<p>\u039e\u03b5\u03ba\u03b9\u03bd\u03ae\u03c3\u03c4\u03b5 \u03bc\u03b5 bounded task, \u03cc\u03c0\u03c9\u03c2 \u03c4\u03b1\u03be\u03b9\u03bd\u03cc\u03bc\u03b7\u03c3\u03b7 intent \u03ae product match. \u039a\u03b1\u03c4\u03b1\u03b3\u03c1\u03ac\u03c8\u03c4\u03b5 \u03c7\u03c9\u03c1\u03b9\u03c3\u03c4\u03ac \u03c4\u03bf \u03ba\u03cc\u03c3\u03c4\u03bf\u03c2 false positive, false negative \u03ba\u03b1\u03b9 \u03b1\u03c0\u03bf\u03c7\u03ae\u03c2 \u03c0\u03c1\u03b9\u03bd \u03c3\u03c5\u03b6\u03b7\u03c4\u03ae\u03c3\u03b5\u03c4\u03b5 threshold.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">Step 2<\/span><strong>\u03a7\u03c4\u03af\u03c3\u03c4\u03b5 labeled calibration set \u03b1\u03c0\u03cc \u03c4\u03bf \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03cc workload<\/strong>\n<p>\u03a3\u03c5\u03bb\u03bb\u03ad\u03be\u03c4\u03b5 \u03b1\u03bd\u03c4\u03b9\u03c0\u03c1\u03bf\u03c3\u03c9\u03c0\u03b5\u03c5\u03c4\u03b9\u03ba\u03ac cases, \u03bf\u03c1\u03af\u03c3\u03c4\u03b5 \u03c3\u03b1\u03c6\u03ae verifier \u03ba\u03b1\u03b9 \u03b5\u03c0\u03b9\u03bb\u03cd\u03c3\u03c4\u03b5 \u03c4\u03b9\u03c2 \u03b4\u03b9\u03b1\u03c6\u03c9\u03bd\u03af\u03b5\u03c2 reviewers. \u03a4\u03b1 training examples \u03b4\u03b5\u03bd \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03b5\u03c0\u03b1\u03bd\u03b5\u03bc\u03c6\u03b1\u03bd\u03af\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 \u03c3\u03c4\u03bf held-out calibration \u03ba\u03b1\u03b9 test split.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">Step 3<\/span><strong>\u03a0\u03b1\u03c1\u03ac\u03b3\u03b5\u03c4\u03b5 \u03c0\u03bf\u03bb\u03bb\u03b1\u03c0\u03bb\u03ac, \u03b5\u03bb\u03b5\u03b3\u03c7\u03cc\u03bc\u03b5\u03bd\u03b1 observations<\/strong>\n<p>\u03a7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03ae\u03c3\u03c4\u03b5 \u03c3\u03c4\u03b1\u03b8\u03b5\u03c1\u03ad\u03c2 prompt \u03c0\u03b1\u03c1\u03b1\u03bb\u03bb\u03b1\u03b3\u03ad\u03c2 \u03ae samples \u03ba\u03b1\u03b9 \u03b1\u03c0\u03bf\u03b8\u03b7\u03ba\u03b5\u03cd\u03c3\u03c4\u03b5 \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7, raw confidence, prompt version \u03ba\u03b1\u03b9 outcome. \u039c\u03b7\u03bd \u03b1\u03bd\u03c4\u03b9\u03bc\u03b5\u03c4\u03c9\u03c0\u03af\u03b6\u03b5\u03c4\u03b5 \u03c3\u03c5\u03c3\u03c7\u03b5\u03c4\u03b9\u03c3\u03bc\u03ad\u03bd\u03b5\u03c2 \u03b4\u03b9\u03b1\u03c4\u03c5\u03c0\u03ce\u03c3\u03b5\u03b9\u03c2 \u03c9\u03c2 \u03c0\u03bb\u03ae\u03c1\u03c9\u03c2 \u03b1\u03bd\u03b5\u03be\u03ac\u03c1\u03c4\u03b7\u03c4\u03b1 \u03c4\u03b5\u03ba\u03bc\u03ae\u03c1\u03b9\u03b1.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">Step 4<\/span><strong>\u0392\u03b1\u03b8\u03bc\u03bf\u03bd\u03bf\u03bc\u03ae\u03c3\u03c4\u03b5 \u03ba\u03ac\u03b8\u03b5 \u03ba\u03b1\u03bd\u03ac\u03bb\u03b9 \u03c0\u03c1\u03b9\u03bd \u03b1\u03c0\u03cc \u03c4\u03b7 \u03c3\u03c5\u03b3\u03ba\u03ad\u03bd\u03c4\u03c1\u03c9\u03c3\u03b7<\/strong>\n<p>\u039c\u03b5\u03c4\u03c1\u03ae\u03c3\u03c4\u03b5 \u03c0\u03ce\u03c2 \u03bc\u03b5\u03c4\u03b1\u03ba\u03b9\u03bd\u03b5\u03af\u03c4\u03b1\u03b9 \u03b7 \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03b1 \u03b1\u03bd\u03ac prompt, \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03bf \u03ba\u03b1\u03b9 \u03ba\u03b1\u03c4\u03b7\u03b3\u03bf\u03c1\u03af\u03b1 task. \u0394\u03b9\u03b1\u03c4\u03b7\u03c1\u03ae\u03c3\u03c4\u03b5 \u03c4\u03b7 monotonic \u03c3\u03b5\u03b9\u03c1\u03ac, \u03b1\u03bb\u03bb\u03ac \u03b4\u03b9\u03bf\u03c1\u03b8\u03ce\u03c3\u03c4\u03b5 systematic overconfidence \u03ae underconfidence.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">Step 5<\/span><strong>\u03a0\u03c1\u03bf\u03c3\u03b8\u03ad\u03c3\u03c4\u03b5 \u03c1\u03b7\u03c4\u03ae abstain \u03ae null \u03b4\u03b9\u03b1\u03b4\u03c1\u03bf\u03bc\u03ae<\/strong>\n<p>\u039f\u03c1\u03af\u03c3\u03c4\u03b5 \u03c4\u03b9 \u03c3\u03c5\u03bc\u03b2\u03b1\u03af\u03bd\u03b5\u03b9 \u03cc\u03c4\u03b1\u03bd \u03bb\u03b5\u03af\u03c0\u03b5\u03b9 \u03ba\u03b1\u03c4\u03ac\u03bb\u03bb\u03b7\u03bb\u03bf\u03c2 candidate, \u03c4\u03b1 observations \u03b4\u03b9\u03b1\u03c6\u03c9\u03bd\u03bf\u03cd\u03bd \u03ae \u03c4\u03bf calibrated evidence \u03b5\u03af\u03bd\u03b1\u03b9 \u03b1\u03c3\u03b8\u03b5\u03bd\u03ad\u03c2. \u03a4\u03bf \u03b1\u03c0\u03bf\u03c4\u03ad\u03bb\u03b5\u03c3\u03bc\u03b1 \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03bf\u03b4\u03b7\u03b3\u03b5\u03af \u03c3\u03b5 human review \u03ba\u03b1\u03b9 \u03cc\u03c7\u03b9 \u03c3\u03b5 \u03b1\u03c5\u03b8\u03b1\u03af\u03c1\u03b5\u03c4\u03b7 \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">Step 6<\/span><strong>\u03a4\u03c1\u03ad\u03be\u03c4\u03b5 shadow mode \u03ba\u03b1\u03b9 \u03b5\u03c0\u03b9\u03bb\u03ad\u03be\u03c4\u03b5 thresholds<\/strong>\n<p>\u039c\u03b5\u03c4\u03c1\u03ae\u03c3\u03c4\u03b5 accuracy, ECE, Brier, AUROC \u03ba\u03b1\u03b9 \u03c0\u03bf\u03c3\u03bf\u03c3\u03c4\u03cc handoff \u03c7\u03c9\u03c1\u03af\u03c2 \u03b1\u03c5\u03c4\u03cc\u03bc\u03b1\u03c4\u03b7 \u03b5\u03c0\u03af\u03b4\u03c1\u03b1\u03c3\u03b7 \u03c3\u03b5 \u03c0\u03b5\u03bb\u03ac\u03c4\u03b5\u03c2. \u0395\u03bb\u03ad\u03b3\u03be\u03c4\u03b5 \u03c4\u03b1 thresholds \u03b1\u03bd\u03ac business cost \u03ba\u03b1\u03b9 \u03b5\u03bd\u03c3\u03c9\u03bc\u03b1\u03c4\u03ce\u03c3\u03c4\u03b5 \u03c4\u03b1 \u03c3\u03c4\u03bf <a href=\"https:\/\/twodots.gr\/agentic-ai-security-stack-controls-technologikes-ependyseis\/\">agentic AI security stack<\/a>.<\/p>\n<\/li>\n<li><span class=\"td-step-kicker\">Step 7<\/span><strong>\u03a0\u03b1\u03c1\u03b1\u03ba\u03bf\u03bb\u03bf\u03c5\u03b8\u03ae\u03c3\u03c4\u03b5 drift \u03ba\u03b1\u03b9 \u03b5\u03c0\u03b1\u03bd\u03b1\u03b2\u03b1\u03b8\u03bc\u03bf\u03bd\u03bf\u03bc\u03ae\u03c3\u03c4\u03b5<\/strong>\n<p>\u039a\u03c1\u03b1\u03c4\u03ae\u03c3\u03c4\u03b5 model version, prompt version, confidence bucket, \u03c4\u03b5\u03bb\u03b9\u03ba\u03cc outcome \u03ba\u03b1\u03b9 reviewer override. \u038c\u03c4\u03b1\u03bd \u03b1\u03bb\u03bb\u03ac\u03b6\u03b5\u03b9 \u03c4\u03bf workload \u03ae \u03c5\u03c0\u03bf\u03b2\u03b1\u03b8\u03bc\u03af\u03b6\u03b5\u03c4\u03b1\u03b9 \u03b7 calibration, \u03c0\u03b5\u03c1\u03b9\u03bf\u03c1\u03af\u03c3\u03c4\u03b5 \u03c4\u03b7\u03bd \u03b1\u03c5\u03c4\u03bf\u03bd\u03bf\u03bc\u03af\u03b1 \u03ba\u03b1\u03b9 \u03b5\u03c0\u03b1\u03bd\u03b1\u03bb\u03ac\u03b2\u03b5\u03c4\u03b5 \u03c4\u03bf fitting.<\/p>\n<\/li>\n<\/ol>\n<\/div>\n<h2 id=\"oria-erevnas-kai-genikeysis\">\u03a4\u03b1 \u03cc\u03c1\u03b9\u03b1 \u03c4\u03b7\u03c2 \u03ad\u03c1\u03b5\u03c5\u03bd\u03b1\u03c2 \u03ba\u03b1\u03b9 \u03c4\u03b7\u03c2 \u03b3\u03b5\u03bd\u03af\u03ba\u03b5\u03c5\u03c3\u03b7\u03c2<\/h2>\n<p>\u039f\u03b9 \u03c3\u03c5\u03b3\u03b3\u03c1\u03b1\u03c6\u03b5\u03af\u03c2 \u03c0\u03b5\u03c1\u03b9\u03bf\u03c1\u03af\u03b6\u03bf\u03c5\u03bd \u03c1\u03b7\u03c4\u03ac \u03c4\u03b7 \u03bc\u03b5\u03bb\u03ad\u03c4\u03b7 \u03c3\u03b5 numerical mathematical reasoning. \u0394\u03b5\u03bd \u03ba\u03b1\u03bb\u03cd\u03c0\u03c4\u03b5\u03b9 open-ended generation, dialogue, proof-oriented reasoning, multimodal tasks \u03ae domains \u03cc\u03c0\u03bf\u03c5 \u03b7 \u03b9\u03c3\u03bf\u03b4\u03c5\u03bd\u03b1\u03bc\u03af\u03b1 \u03c4\u03c9\u03bd \u03b1\u03c0\u03b1\u03bd\u03c4\u03ae\u03c3\u03b5\u03c9\u03bd \u03b1\u03c0\u03b1\u03b9\u03c4\u03b5\u03af semantic judgment. \u03a3\u03b5 \u03c4\u03ad\u03c4\u03bf\u03b9\u03b5\u03c2 \u03c0\u03b5\u03c1\u03b9\u03c0\u03c4\u03ce\u03c3\u03b5\u03b9\u03c2 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 task-specific verification \u03ba\u03b1\u03b9 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03bf\u03af \u03bf\u03c1\u03b9\u03c3\u03bc\u03bf\u03af candidate states.<\/p>\n<p>\u03a4\u03b1 \u03c0\u03b5\u03b9\u03c1\u03ac\u03bc\u03b1\u03c4\u03b1 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03bf\u03cd\u03bd moderate-scale open-weight models \u03ba\u03b1\u03b9 \u03cc\u03c7\u03b9 \u03c4\u03b1 \u03bd\u03b5\u03cc\u03c4\u03b5\u03c1\u03b1 frontier models. \u039f\u03b9 \u03b5\u03c1\u03b5\u03c5\u03bd\u03b7\u03c4\u03ad\u03c2 \u03b5\u03be\u03b7\u03b3\u03bf\u03cd\u03bd \u03cc\u03c4\u03b9 \u03c0\u03bf\u03bb\u03cd \u03b9\u03c3\u03c7\u03c5\u03c1\u03ac \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03b1 \u03bc\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03ba\u03ac\u03bd\u03bf\u03c5\u03bd \u03b5\u03bb\u03ac\u03c7\u03b9\u03c3\u03c4\u03b1 \u03bb\u03ac\u03b8\u03b7 \u03c3\u03c4\u03b1 \u03c3\u03c5\u03bd\u03b7\u03b8\u03b9\u03c3\u03bc\u03ad\u03bd\u03b1 arithmetic benchmarks, \u03b3\u03b5\u03b3\u03bf\u03bd\u03cc\u03c2 \u03c0\u03bf\u03c5 \u03b4\u03c5\u03c3\u03ba\u03bf\u03bb\u03b5\u03cd\u03b5\u03b9 failure-aware calibration analysis. \u03a0\u03c1\u03bf\u03c4\u03b5\u03af\u03bd\u03bf\u03c5\u03bd \u03b1\u03be\u03b9\u03bf\u03bb\u03cc\u03b3\u03b7\u03c3\u03b7 \u03c3\u03b5 \u03b4\u03c5\u03c3\u03ba\u03bf\u03bb\u03cc\u03c4\u03b5\u03c1\u03b1 benchmarks \u03bc\u03b5 \u03b1\u03c1\u03ba\u03b5\u03c4\u03ad\u03c2 \u03c3\u03c9\u03c3\u03c4\u03ad\u03c2 \u03ba\u03b1\u03b9 \u03bb\u03ac\u03b8\u03bf\u03c2 \u03c0\u03c1\u03bf\u03b2\u03bb\u03ad\u03c8\u03b5\u03b9\u03c2.<\/p>\n<p>\u0397 \u03bc\u03ad\u03b8\u03bf\u03b4\u03bf\u03c2 \u03b1\u03c0\u03b1\u03b9\u03c4\u03b5\u03af calibration examples \u03bc\u03b5 \u03b3\u03bd\u03c9\u03c3\u03c4\u03ae \u03bf\u03c1\u03b8\u03cc\u03c4\u03b7\u03c4\u03b1. \u0395\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03b7\u03c3\u03b9\u03b1\u03ba\u03ac, \u03b1\u03c5\u03c4\u03cc \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 curated validation set, \u03b1\u03be\u03b9\u03cc\u03c0\u03b9\u03c3\u03c4\u03bf verifier \u03ba\u03b1\u03b9 \u03c0\u03b1\u03c1\u03b1\u03ba\u03bf\u03bb\u03bf\u03cd\u03b8\u03b7\u03c3\u03b7 \u03bc\u03b5\u03c4\u03b1\u03c4\u03cc\u03c0\u03b9\u03c3\u03b7\u03c2 \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03c9\u03bd. \u03a4\u03bf NIST AI RMF \u03c4\u03bf\u03bd\u03af\u03b6\u03b5\u03b9 \u03c4\u03b7 \u03c3\u03c5\u03bd\u03b5\u03c7\u03ae \u03bc\u03ad\u03c4\u03c1\u03b7\u03c3\u03b7, \u03c4\u03b7\u03bd \u03c4\u03b5\u03ba\u03bc\u03b7\u03c1\u03af\u03c9\u03c3\u03b7 \u03ba\u03b1\u03b9 \u03c4\u03bf\u03c5\u03c2 \u03c3\u03b1\u03c6\u03b5\u03af\u03c2 \u03c1\u03cc\u03bb\u03bf\u03c5\u03c2 \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03b7\u03c2 \u03b5\u03c0\u03bf\u03c0\u03c4\u03b5\u03af\u03b1\u03c2\u00b7 \u03c4\u03bf DirEAG \u03b4\u03af\u03bd\u03b5\u03b9 \u03ad\u03bd\u03b1 \u03c4\u03b5\u03c7\u03bd\u03b9\u03ba\u03cc \u03bc\u03bf\u03c4\u03af\u03b2\u03bf \u03b3\u03b9\u03b1 confidence aggregation, \u03cc\u03c7\u03b9 \u03bf\u03bb\u03cc\u03ba\u03bb\u03b7\u03c1\u03bf \u03c0\u03bb\u03b1\u03af\u03c3\u03b9\u03bf governance.<\/p>\n<h2 id=\"empistosyni-me-dikaioma-amfivolias\">\u0395\u03bc\u03c0\u03b9\u03c3\u03c4\u03bf\u03c3\u03cd\u03bd\u03b7 \u03bc\u03b5 \u03b4\u03b9\u03ba\u03b1\u03af\u03c9\u03bc\u03b1 \u03c3\u03c4\u03b7\u03bd \u03b1\u03bc\u03c6\u03b9\u03b2\u03bf\u03bb\u03af\u03b1<\/h2>\n<p>\u03a4\u03bf DirEAG \u03b1\u03bd\u03c4\u03b9\u03bc\u03b5\u03c4\u03c9\u03c0\u03af\u03b6\u03b5\u03b9 \u03c4\u03b7 confidence \u03c9\u03c2 \u03bc\u03b5\u03c4\u03c1\u03ae\u03c3\u03b9\u03bc\u03bf \u03b1\u03bb\u03bb\u03ac scale-dependent \u03c3\u03ae\u03bc\u03b1. \u039a\u03c1\u03b1\u03c4\u03ac \u03c4\u03b7 \u03c3\u03b5\u03b9\u03c1\u03ac \u03c4\u03c9\u03bd scores \u03bc\u03ad\u03c3\u03b1 \u03c3\u03b5 \u03ba\u03ac\u03b8\u03b5 steering level, \u03bc\u03b1\u03b8\u03b1\u03af\u03bd\u03b5\u03b9 \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03ad\u03c2 \u03b1\u03be\u03b9\u03bf\u03c0\u03b9\u03c3\u03c4\u03af\u03b1\u03c2, \u03c3\u03c5\u03c3\u03c3\u03c9\u03c1\u03b5\u03cd\u03b5\u03b9 evidence \u03c0\u03ac\u03bd\u03c9 \u03c3\u03c4\u03b9\u03c2 \u03b1\u03c0\u03b1\u03bd\u03c4\u03ae\u03c3\u03b5\u03b9\u03c2 \u03ba\u03b1\u03b9 \u03b1\u03c6\u03ae\u03bd\u03b5\u03b9 \u03c7\u03ce\u03c1\u03bf \u03b3\u03b9\u03b1 \u03c4\u03bf \u03b5\u03bd\u03b4\u03b5\u03c7\u03cc\u03bc\u03b5\u03bd\u03bf \u03bd\u03b1 \u03bc\u03b7\u03bd \u03c5\u03c0\u03ac\u03c1\u03c7\u03b5\u03b9 \u03c3\u03c9\u03c3\u03c4\u03ae \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ae.<\/p>\n<p>\u0393\u03b9\u03b1 \u03bc\u03b9\u03b1 \u03b5\u03c0\u03b9\u03c7\u03b5\u03af\u03c1\u03b7\u03c3\u03b7, \u03c4\u03bf \u03bc\u03ae\u03bd\u03c5\u03bc\u03b1 \u03b5\u03af\u03bd\u03b1\u03b9 \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03bf: \u03bc\u03b7\u03bd \u03b1\u03c6\u03ae\u03bd\u03b5\u03c4\u03b5 \u03ad\u03bd\u03b1 LLM \u03bd\u03b1 \u03bc\u03b5\u03c4\u03b1\u03c4\u03c1\u03ad\u03c0\u03b5\u03b9 \u03c4\u03b7\u03bd \u03b1\u03c5\u03c4\u03bf\u03c0\u03b5\u03c0\u03bf\u03af\u03b8\u03b7\u03c3\u03ae \u03c4\u03bf\u03c5 \u03b1\u03c0\u03b5\u03c5\u03b8\u03b5\u03af\u03b1\u03c2 \u03c3\u03b5 \u03b5\u03be\u03bf\u03c5\u03c3\u03af\u03b1 \u03b4\u03c1\u03ac\u03c3\u03b7\u03c2. \u039c\u03b5\u03c4\u03c1\u03ae\u03c3\u03c4\u03b5 \u03c4\u03b7 \u03c3\u03c7\u03ad\u03c3\u03b7 \u03ba\u03ac\u03b8\u03b5 score \u03bc\u03b5 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac outcomes, \u03c3\u03c4\u03bf \u03af\u03b4\u03b9\u03bf task \u03ba\u03b1\u03b9 \u03c3\u03c4\u03b1 \u03af\u03b4\u03b9\u03b1 \u03b4\u03b5\u03b4\u03bf\u03bc\u03ad\u03bd\u03b1, \u03ba\u03b1\u03b9 \u03c3\u03c7\u03b5\u03b4\u03b9\u03ac\u03c3\u03c4\u03b5 \u03b1\u03c0\u03cc \u03c0\u03c1\u03b9\u03bd \u03c4\u03b9 \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u03b1\u03c0\u03bf\u03c7\u03ae. \u0388\u03bd\u03b1 \u03b1\u03be\u03b9\u03cc\u03c0\u03b9\u03c3\u03c4\u03bf workflow \u03b4\u03b5\u03bd \u03b5\u03c0\u03b9\u03b4\u03b9\u03ce\u03ba\u03b5\u03b9 \u03bc\u03cc\u03bd\u03bf \u03b1\u03c0\u03bf\u03c6\u03b1\u03c3\u03b9\u03c3\u03c4\u03b9\u03ba\u03ad\u03c2 \u03b1\u03c0\u03b1\u03bd\u03c4\u03ae\u03c3\u03b5\u03b9\u03c2\u00b7 \u03b3\u03bd\u03c9\u03c1\u03af\u03b6\u03b5\u03b9 \u03c0\u03cc\u03c4\u03b5 \u03c4\u03bf evidence \u03b1\u03c1\u03ba\u03b5\u03af \u03ba\u03b1\u03b9 \u03c0\u03cc\u03c4\u03b5 \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03ac\u03bd\u03b8\u03c1\u03c9\u03c0\u03bf\u03c2.<\/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 confidence score \u03c3\u03b5 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ad\u03c2 action<\/p>\n<h3>\u03a3\u03c7\u03b5\u03b4\u03b9\u03ac\u03c3\u03c4\u03b5 AI \u03b1\u03c5\u03c4\u03bf\u03bc\u03b1\u03c4\u03b9\u03c3\u03bc\u03bf\u03cd\u03c2 \u03bc\u03b5 \u03b2\u03b1\u03b8\u03bc\u03bf\u03bd\u03cc\u03bc\u03b7\u03c3\u03b7 \u03ba\u03b1\u03b9 \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03b1 gates<\/h3>\n<p>\u0397 TWO DOTS \u03c7\u03b1\u03c1\u03c4\u03bf\u03b3\u03c1\u03b1\u03c6\u03b5\u03af labeled outcomes, confidence thresholds, abstain paths, audit trail, drift monitoring \u03ba\u03b1\u03b9 human approval, \u03ce\u03c3\u03c4\u03b5 \u03ad\u03bd\u03b1 AI workflow \u03bd\u03b1 \u03b5\u03bd\u03b5\u03c1\u03b3\u03b5\u03af \u03bc\u03cc\u03bd\u03bf \u03cc\u03c4\u03b1\u03bd \u03b7 \u03b2\u03b5\u03b2\u03b1\u03b9\u03cc\u03c4\u03b7\u03c4\u03ac \u03c4\u03bf\u03c5 \u03ad\u03c7\u03b5\u03b9 \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03b7\u03c3\u03b9\u03b1\u03ba\u03cc \u03bd\u03cc\u03b7\u03bc\u03b1.<\/p>\n<div class=\"td-service-cta-actions\"><a class=\"td-service-cta-button\" href=\"https:\/\/twodots.gr\/aftomatismoi-epicheiriseon-ai\/\">Check out Business Automation &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\">Frequently Asked Questions (FAQs)<\/p>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a4\u03b9 \u03b5\u03af\u03bd\u03b1\u03b9 \u03c4\u03bf DirEAG;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u03a4\u03bf DirEAG \u03b5\u03af\u03bd\u03b1\u03b9 \u03bc\u03ad\u03b8\u03bf\u03b4\u03bf\u03c2 Dirichlet Evidence Aggregation \u03c0\u03bf\u03c5 \u03c3\u03c5\u03bd\u03b4\u03c5\u03ac\u03b6\u03b5\u03b9 \u03c0\u03bf\u03bb\u03bb\u03b1\u03c0\u03bb\u03ad\u03c2 \u03b1\u03c0\u03b1\u03bd\u03c4\u03ae\u03c3\u03b5\u03b9\u03c2 \u03ba\u03b1\u03b9 \u03b4\u03b7\u03bb\u03ce\u03c3\u03b5\u03b9\u03c2 confidence \u03b5\u03bd\u03cc\u03c2 fixed black-box LLM, \u03ce\u03c3\u03c4\u03b5 \u03bd\u03b1 \u03b5\u03c0\u03b9\u03bb\u03ad\u03b3\u03b5\u03b9 candidate \u03ba\u03b1\u03b9 \u03bd\u03b1 \u03c0\u03b1\u03c1\u03ac\u03b3\u03b5\u03b9 \u03b2\u03b1\u03b8\u03bc\u03bf\u03bd\u03bf\u03bc\u03b7\u03bc\u03ad\u03bd\u03b7 \u03c0\u03b9\u03b8\u03b1\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1 \u03bf\u03c1\u03b8\u03cc\u03c4\u03b7\u03c4\u03b1\u03c2.<\/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 \u03bf \u03bc\u03ad\u03c3\u03bf\u03c2 \u03cc\u03c1\u03bf\u03c2 \u03c4\u03c9\u03bd confidence scores;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u0394\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03ac prompts, \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03b1 \u03ba\u03b1\u03b9 datasets \u03bc\u03c0\u03bf\u03c1\u03bf\u03cd\u03bd \u03bd\u03b1 \u03bc\u03b5\u03c4\u03b1\u03ba\u03b9\u03bd\u03bf\u03cd\u03bd \u03c3\u03c5\u03c3\u03c4\u03b7\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac \u03c4\u03b7\u03bd \u03ba\u03bb\u03af\u03bc\u03b1\u03ba\u03b1 \u03c4\u03b7\u03c2 \u03b4\u03b7\u03bb\u03c9\u03bc\u03ad\u03bd\u03b7\u03c2 \u03b2\u03b5\u03b2\u03b1\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1\u03c2. \u039f \u03b1\u03c0\u03bb\u03cc\u03c2 \u03bc\u03ad\u03c3\u03bf\u03c2 \u03cc\u03c1\u03bf\u03c2 \u03b4\u03b5\u03bd \u03bc\u03b1\u03b8\u03b1\u03af\u03bd\u03b5\u03b9 \u03b1\u03c5\u03c4\u03ad\u03c2 \u03c4\u03b9\u03c2 \u03b1\u03c0\u03bf\u03ba\u03bb\u03af\u03c3\u03b5\u03b9\u03c2 \u03bf\u03cd\u03c4\u03b5 \u03c4\u03b7\u03bd \u03b1\u03be\u03b9\u03bf\u03c0\u03b9\u03c3\u03c4\u03af\u03b1 \u03ba\u03ac\u03b8\u03b5 steering level.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a4\u03b9 \u03c1\u03cc\u03bb\u03bf \u03c0\u03b1\u03af\u03b6\u03b5\u03b9 \u03bf null state;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u0395\u03ba\u03c6\u03c1\u03ac\u03b6\u03b5\u03b9 \u03c4\u03b7\u03bd \u03c0\u03b9\u03b8\u03b1\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1 \u03cc\u03c4\u03b9 \u03b7 \u03c3\u03c9\u03c3\u03c4\u03ae \u03b1\u03c0\u03ac\u03bd\u03c4\u03b7\u03c3\u03b7 \u03bb\u03b5\u03af\u03c0\u03b5\u03b9 \u03b1\u03c0\u03cc \u03c4\u03bf \u03c3\u03cd\u03bd\u03bf\u03bb\u03bf \u03c4\u03c9\u03bd generated candidates. \u03a7\u03c9\u03c1\u03af\u03c2 \u03b1\u03c5\u03c4\u03cc\u03bd, \u03c4\u03bf \u03c3\u03cd\u03c3\u03c4\u03b7\u03bc\u03b1 \u03b8\u03b1 \u03bc\u03bf\u03af\u03c1\u03b1\u03b6\u03b5 \u03cc\u03bb\u03b7 \u03c4\u03b7\u03bd \u03c0\u03b9\u03b8\u03b1\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b1\u03bd\u03ac\u03bc\u03b5\u03c3\u03b1 \u03c3\u03b5 \u03b4\u03b9\u03b1\u03b8\u03ad\u03c3\u03b9\u03bc\u03b5\u03c2 \u03b1\u03bb\u03bb\u03ac \u03c0\u03b9\u03b8\u03b1\u03bd\u03ce\u03c2 \u03bb\u03b1\u03bd\u03b8\u03b1\u03c3\u03bc\u03ad\u03bd\u03b5\u03c2 \u03b5\u03c0\u03b9\u03bb\u03bf\u03b3\u03ad\u03c2.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u0391\u03bb\u03bb\u03ac\u03b6\u03b5\u03b9 \u03ae \u03ba\u03ac\u03bd\u03b5\u03b9 fine-tune \u03c4\u03bf \u03b2\u03b1\u03c3\u03b9\u03ba\u03cc LLM;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u038c\u03c7\u03b9. \u0397 \u03bc\u03ad\u03b8\u03bf\u03b4\u03bf\u03c2 \u03bb\u03b5\u03b9\u03c4\u03bf\u03c5\u03c1\u03b3\u03b5\u03af post hoc \u03c0\u03ac\u03bd\u03c9 \u03c3\u03b5 answer-confidence observations \u03b5\u03bd\u03cc\u03c2 fixed black-box model \u03ba\u03b1\u03b9 \u03bc\u03b1\u03b8\u03b1\u03af\u03bd\u03b5\u03b9 \u03bc\u03b9\u03ba\u03c1\u03cc \u03b1\u03c1\u03b9\u03b8\u03bc\u03cc \u03c0\u03b1\u03c1\u03b1\u03bc\u03ad\u03c4\u03c1\u03c9\u03bd \u03b3\u03b9\u03b1 calibration, reliability weighting \u03ba\u03b1\u03b9 aggregation.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u03a3\u03b5 \u03c0\u03bf\u03b9\u03b1 \u03bc\u03bf\u03bd\u03c4\u03ad\u03bb\u03b1 \u03ba\u03b1\u03b9 datasets \u03b4\u03bf\u03ba\u03b9\u03bc\u03ac\u03c3\u03c4\u03b7\u03ba\u03b5;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u03a3\u03c4\u03b1 Qwen2.5-7B-Instruct, Mistral-7B-Instruct-v0.3 \u03ba\u03b1\u03b9 Gemma-2-9B-it, \u03c0\u03ac\u03bd\u03c9 \u03c3\u03c4\u03b1 \u03bc\u03b1\u03b8\u03b7\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac benchmarks GSM8K, SVAMP \u03ba\u03b1\u03b9 GSM-Hard.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u0397 \u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b7 calibration \u03c3\u03b7\u03bc\u03b1\u03af\u03bd\u03b5\u03b9 \u03c0\u03ac\u03bd\u03c4\u03b1 \u03ba\u03b1\u03bb\u03cd\u03c4\u03b5\u03c1\u03b7 accuracy;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u038c\u03c7\u03b9. Accuracy \u03ba\u03b1\u03b9 calibration \u03bc\u03b5\u03c4\u03c1\u03bf\u03cd\u03bd \u03b4\u03b9\u03b1\u03c6\u03bf\u03c1\u03b5\u03c4\u03b9\u03ba\u03ad\u03c2 \u03b9\u03b4\u03b9\u03cc\u03c4\u03b7\u03c4\u03b5\u03c2. \u03a4\u03b1 \u03b1\u03c0\u03bf\u03c4\u03b5\u03bb\u03ad\u03c3\u03bc\u03b1\u03c4\u03b1 \u03c0\u03b5\u03c1\u03b9\u03bb\u03b1\u03bc\u03b2\u03ac\u03bd\u03bf\u03c5\u03bd \u03c0\u03b5\u03c1\u03b9\u03c0\u03c4\u03ce\u03c3\u03b5\u03b9\u03c2 \u03cc\u03c0\u03bf\u03c5 \u03ac\u03bb\u03bb\u03b7 \u03bc\u03ad\u03b8\u03bf\u03b4\u03bf\u03c2 \u03ad\u03c7\u03b5\u03b9 \u03c5\u03c8\u03b7\u03bb\u03cc\u03c4\u03b5\u03c1\u03b7 accuracy \u03ae \u03c7\u03b1\u03bc\u03b7\u03bb\u03cc\u03c4\u03b5\u03c1\u03bf ECE, \u03b5\u03bd\u03ce \u03c4\u03bf DirEAG \u03b2\u03b5\u03bb\u03c4\u03b9\u03ce\u03bd\u03b5\u03b9 \u03c3\u03c5\u03c7\u03bd\u03ac \u03c3\u03c5\u03b3\u03ba\u03b5\u03ba\u03c1\u03b9\u03bc\u03ad\u03bd\u03b5\u03c2 calibration metrics.<\/p>\n<\/div>\n<\/details>\n<details class=\"td-faq-item\">\n<summary class=\"td-faq-title\">\u039c\u03c0\u03bf\u03c1\u03b5\u03af \u03bd\u03b1 \u03b5\u03c6\u03b1\u03c1\u03bc\u03bf\u03c3\u03c4\u03b5\u03af \u03b1\u03c5\u03c4\u03bf\u03cd\u03c3\u03b9\u03bf \u03c3\u03b5 customer support \u03ae e-commerce;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u038c\u03c7\u03b9 \u03c7\u03c9\u03c1\u03af\u03c2 \u03bd\u03ad\u03b1 \u03b5\u03c0\u03b9\u03ba\u03cd\u03c1\u03c9\u03c3\u03b7. \u0397 \u03ad\u03c1\u03b5\u03c5\u03bd\u03b1 \u03b1\u03c6\u03bf\u03c1\u03ac \u03b1\u03c1\u03b9\u03b8\u03bc\u03b7\u03c4\u03b9\u03ba\u03cc reasoning. \u039a\u03ac\u03b8\u03b5 \u03b5\u03c0\u03b9\u03c7\u03b5\u03b9\u03c1\u03b7\u03c3\u03b9\u03b1\u03ba\u03cc domain \u03c7\u03c1\u03b5\u03b9\u03ac\u03b6\u03b5\u03c4\u03b1\u03b9 \u03b4\u03b9\u03ba\u03bf\u03cd\u03c2 \u03c4\u03bf\u03c5 candidates, verifier, labeled calibration set, drift monitoring \u03ba\u03b1\u03b9 \u03c0\u03bf\u03bb\u03b9\u03c4\u03b9\u03ba\u03ae \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03b7\u03c2 \u03c0\u03b1\u03c1\u03ad\u03bc\u03b2\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 operational takeaway;<\/summary>\n<div class=\"td-faq-content\">\n<p>\u0397 \u03b1\u03c5\u03c4\u03bf\u03b4\u03b7\u03bb\u03c9\u03bc\u03ad\u03bd\u03b7 \u03b2\u03b5\u03b2\u03b1\u03b9\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b4\u03b5\u03bd \u03c0\u03c1\u03ad\u03c0\u03b5\u03b9 \u03bd\u03b1 \u03c7\u03c1\u03b7\u03c3\u03b9\u03bc\u03bf\u03c0\u03bf\u03b9\u03b5\u03af\u03c4\u03b1\u03b9 \u03c9\u03c2 \u03ad\u03c4\u03bf\u03b9\u03bc\u03b7 \u03c0\u03b9\u03b8\u03b1\u03bd\u03cc\u03c4\u03b7\u03c4\u03b1 \u03ae \u03ac\u03b4\u03b5\u03b9\u03b1 \u03b4\u03c1\u03ac\u03c3\u03b7\u03c2. \u03a7\u03c1\u03b5\u03b9\u03ac\u03b6\u03bf\u03bd\u03c4\u03b1\u03b9 \u03c0\u03c1\u03b1\u03b3\u03bc\u03b1\u03c4\u03b9\u03ba\u03ac outcomes, \u03c1\u03b7\u03c4\u03ae \u03b4\u03c5\u03bd\u03b1\u03c4\u03cc\u03c4\u03b7\u03c4\u03b1 \u03b1\u03c0\u03bf\u03c7\u03ae\u03c2 \u03ba\u03b1\u03b9 thresholds \u03c0\u03bf\u03c5 \u03ad\u03c7\u03bf\u03c5\u03bd \u03b5\u03c0\u03b9\u03ba\u03c5\u03c1\u03c9\u03b8\u03b5\u03af \u03c3\u03c4\u03bf \u03af\u03b4\u03b9\u03bf workload.<\/p>\n<\/div>\n<\/details>\n<\/div>\n<\/section>\n<div class=\"td-source-list\">\n<p id=\"piges\" class=\"td-source-list-title\">Sources<\/p>\n<ul>\n<li><a href=\"https:\/\/arxiv.org\/abs\/2608.20717\" target=\"_blank\" rel=\"noopener\">arXiv \u2014 DirEAG: Dirichlet Evidence Aggregation for Calibrating Verbalized Confidence in Mathematical Reasoning<\/a><\/li>\n<li><a href=\"https:\/\/github.com\/horacehsugithub\/DirEAG\" target=\"_blank\" rel=\"noopener\">GitHub \u2014 \u03b5\u03c0\u03af\u03c3\u03b7\u03bc\u03bf\u03c2 \u03ba\u03ce\u03b4\u03b9\u03ba\u03b1\u03c2 \u03ba\u03b1\u03b9 \u03c5\u03bb\u03b9\u03ba\u03cc \u03b1\u03bd\u03b1\u03c0\u03b1\u03c1\u03b1\u03b3\u03c9\u03b3\u03ae\u03c2 \u03c4\u03bf\u03c5 DirEAG<\/a><\/li>\n<li><a href=\"https:\/\/www.nist.gov\/itl\/ai-risk-management-framework\" target=\"_blank\" rel=\"noopener\">NIST \u2014 AI Risk Management Framework<\/a><\/li>\n<li><a href=\"https:\/\/airc.nist.gov\/airmf-resources\/airmf\/5-sec-core\/\" target=\"_blank\" rel=\"noopener\">NIST AIRC \u2014 AI RMF Core \u03b3\u03b9\u03b1 \u03bc\u03ad\u03c4\u03c1\u03b7\u03c3\u03b7, \u03c0\u03b1\u03c1\u03b1\u03ba\u03bf\u03bb\u03bf\u03cd\u03b8\u03b7\u03c3\u03b7 \u03ba\u03b1\u03b9 \u03b1\u03bd\u03b8\u03c1\u03ce\u03c0\u03b9\u03bd\u03b7 \u03b5\u03c0\u03bf\u03c0\u03c4\u03b5\u03af\u03b1<\/a><\/li>\n<\/ul>\n<\/div>","protected":false},"excerpt":{"rendered":"<p>\u03a4\u03bf DirEAG \u03b2\u03b1\u03b8\u03bc\u03bf\u03bd\u03bf\u03bc\u03b5\u03af \u03b4\u03b7\u03bb\u03ce\u03c3\u03b5\u03b9\u03c2 confidence \u03b5\u03bd\u03cc\u03c2 LLM, \u03c3\u03c5\u03bd\u03b4\u03c5\u03ac\u03b6\u03b5\u03b9 \u03c0\u03bf\u03bb\u03bb\u03b1\u03c0\u03bb\u03ac \u03c4\u03b5\u03ba\u03bc\u03ae\u03c1\u03b9\u03b1 \u03ba\u03b1\u03b9 \u03ba\u03c1\u03b1\u03c4\u03ac null state \u03b3\u03b9\u03b1 \u03b1\u03c3\u03c6\u03b1\u03bb\u03ae \u03b1\u03c0\u03bf\u03c7\u03ae.<\/p>","protected":false},"author":1,"featured_media":96186,"comment_status":"open","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"_gspb_post_css":"","content-type":"","footnotes":""},"categories":[199],"tags":[20050,20053,20054,20051,20052],"class_list":["post-92980","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-techniti-noimosyni","tag-ai-confidence","tag-ai-uncertainty","tag-direag","tag-llm-calibration","tag-mathematical-reasoning"],"blocksy_meta":{"styles_descriptor":{"styles":{"desktop":"","tablet":"","mobile":""},"google_fonts":[],"version":8}},"_links":{"self":[{"href":"https:\/\/twodots.gr\/en\/wp-json\/wp\/v2\/posts\/92980","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/twodots.gr\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/twodots.gr\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/twodots.gr\/en\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/twodots.gr\/en\/wp-json\/wp\/v2\/comments?post=92980"}],"version-history":[{"count":0,"href":"https:\/\/twodots.gr\/en\/wp-json\/wp\/v2\/posts\/92980\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/twodots.gr\/en\/wp-json\/wp\/v2\/media\/96186"}],"wp:attachment":[{"href":"https:\/\/twodots.gr\/en\/wp-json\/wp\/v2\/media?parent=92980"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/twodots.gr\/en\/wp-json\/wp\/v2\/categories?post=92980"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/twodots.gr\/en\/wp-json\/wp\/v2\/tags?post=92980"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}