{"id":11402,"date":"2026-10-04T00:45:44","date_gmt":"2026-10-04T00:45:44","guid":{"rendered":"https:\/\/bitep.net\/blog\/2026\/10\/04\/bloggoogle-2\/"},"modified":"2026-10-04T00:45:44","modified_gmt":"2026-10-04T00:45:44","slug":"bloggoogle-2","status":"publish","type":"post","link":"https:\/\/bitep.net\/blog\/2026\/10\/04\/bloggoogle-2\/","title":{"rendered":"blog.google"},"content":{"rendered":"<h2>S\u00fcni \u0130ntellektd\u0259 Yeni M\u0259rh\u0259l\u0259: Sad\u0259 \u00c7atbotlardan Avtonom &#8220;R\u0259q\u0259msal H\u0259mkarlar&#8221;a Ke\u00e7id<\/h2>\n<p>Generativ s\u00fcni intellekt d\u00fcnyas\u0131 sad\u0259 m\u0259tn generasiyas\u0131 v\u0259 prompt-\u0259sasl\u0131 dialoqlardan ibar\u0259t ilkin m\u0259rh\u0259l\u0259ni gerid\u0259 qoyur. Google-\u0131n son texnoloji hesabatlar\u0131 v\u0259 frontier model inki\u015faflar\u0131 g\u00f6st\u0259rir ki, s\u0259naye art\u0131q sad\u0259c\u0259 parametr say\u0131n\u0131 art\u0131rmaqdan (brute-force scaling) \u00e7\u0131xaraq, <strong>d\u0259rin m\u0259ntiqi m\u00fchakim\u0259 (deep logical reasoning)<\/strong> v\u0259 m\u00fcr\u0259kk\u0259b \u0259m\u0259liyyatlar\u0131 icra ed\u0259n <strong>avtonom agent arxitekturalar\u0131na<\/strong> y\u00f6n\u0259lib. Art\u0131q s\u00fcni intellekt yaln\u0131z e-m\u0259ktub qaralamas\u0131 haz\u0131rlayan k\u00f6m\u0259k\u00e7i deyil, t\u0259\u015fkilati prosesl\u0259ri u\u00e7dan-uca idar\u0259 ed\u0259n &#8220;R\u0259q\u0259msal H\u0259mkar&#8221; (Digital Colleague) rolunu icra edir.<\/p>\n<h2>D\u0259rin M\u0259ntiq v\u0259 Qabaqc\u0131l Arxitekturalar: Gemini 4 Argon v\u0259 MoE<\/h2>\n<p>Son n\u0259sil modell\u0259r, x\u00fcsusil\u0259 Google-\u0131n <em>Gemini 4 Argon<\/em> konsepti v\u0259 qabaqc\u0131l Mixture-of-Experts (MoE) arxitekturalar\u0131 sistemin daxili i\u015fl\u0259m\u0259 prinsipl\u0259rini k\u00f6k\u00fcnd\u0259n d\u0259yi\u015fib. Bu arxitekturalar cavab verm\u0259zd\u0259n \u0259vv\u0259l bir ne\u00e7\u0259 m\u0259rh\u0259l\u0259li daxili yoxlama, f\u0259rziyy\u0259l\u0259rin s\u0131naqdan ke\u00e7irilm\u0259si v\u0259 d\u00fc\u015f\u00fcnc\u0259 z\u0259nciri (chain-of-thought) add\u0131mlar\u0131n\u0131 i\u015f\u0259 sal\u0131r.<\/p>\n<ul>\n<li><strong>Test-Time Compute (\u0130cra Zaman\u0131 Hesablama):<\/strong> Model \u0259n\u0259n\u0259vi token proqnozla\u015fd\u0131rmas\u0131ndan f\u0259rqli olaraq, cavab generasiyas\u0131na ba\u015flamazdan \u0259vv\u0259l hesablama resurslar\u0131n\u0131 sual\u0131n m\u00fcr\u0259kk\u0259bliyin\u0259 uy\u011fun b\u00f6l\u00fc\u015fd\u00fcr\u00fcr. Bu, x\u00fcsusil\u0259 ali riyaziyyat, arxitektur proqramla\u015fd\u0131rma v\u0259 strateji planla\u015fd\u0131rmada x\u0259talar\u0131 minimuma endirir.<\/li>\n<li><strong>Geni\u015fl\u0259ndirilmi\u015f \u00c7oxmodall\u0131q (Multimodality):<\/strong> Yeni n\u0259sil sisteml\u0259r saatlarla davam ed\u0259n video ax\u0131nlar\u0131n\u0131, real-vaxt audio siqnallar\u0131n\u0131 v\u0259 minl\u0259rl\u0259 s\u0259hif\u0259lik s\u0259n\u0259dl\u0259ri eyni anda emal edir. Giri\u015f m\u0259lumatlar\u0131n\u0131n format\u0131ndan as\u0131l\u0131 olmayaraq, daxili vektor f\u0259zas\u0131nda vahid kontekst formala\u015f\u0131r.<\/li>\n<li><strong>Ultra A\u015fa\u011f\u0131 Gecikm\u0259li T\u0259bii S\u0259s Modeli:<\/strong> \u0130nsan dan\u0131\u015f\u0131\u011f\u0131ndak\u0131 intonasiya, fasil\u0259l\u0259r v\u0259 emosional \u00e7alarlar\u0131 d\u0259rk ed\u0259n nitq modell\u0259ri API s\u0259viyy\u0259sind\u0259 inteqrasiya edil\u0259r\u0259k ma\u015f\u0131nla qar\u015f\u0131l\u0131ql\u0131 \u0259laq\u0259ni real-vaxt s\u00f6hb\u0259t dinamikas\u0131na g\u0259tirir.<\/li>\n<\/ul>\n<h2>&#8220;R\u0259q\u0259msal H\u0259mkarlar&#8221;: M\u00fc\u0259ssis\u0259 S\u0259viyy\u0259sind\u0259 \u0130\u015f Ax\u0131nlar\u0131n\u0131n Avtomatla\u015fd\u0131r\u0131lmas\u0131<\/h2>\n<p>\u018fvv\u0259lki n\u0259sil s\u00fcni intellekt al\u0259tl\u0259ri h\u0259r add\u0131mda insan m\u00fcdaxil\u0259si t\u0259l\u0259b edirdis\u0259, yeni agentl\u0259r <em>action-space<\/em> (f\u0259aliyy\u0259t m\u00fchiti) daxilind\u0259 tam m\u00fcst\u0259qil q\u0259rarlar q\u0259bul ed\u0259 bilir. Sistem m\u00fc\u0259ssis\u0259 daxilind\u0259 a\u015fa\u011f\u0131dak\u0131 kritik \u0259m\u0259liyyatlar\u0131 icra edir:<\/p>\n<ul>\n<li><strong>\u0130T D\u0259st\u0259k v\u0259 \u0130nsident \u0130dar\u0259etm\u0259si:<\/strong> Server loqlar\u0131n\u0131 analiz edir, x\u0259tan\u0131n k\u00f6k s\u0259b\u0259bini (root cause analysis) tap\u0131r, m\u00fcvafiq patch-i haz\u0131rlay\u0131r v\u0259 bileti (ticket) ba\u011flay\u0131r.<\/li>\n<li><strong>Maliyy\u0259 \u018fm\u0259liyyatlar\u0131 v\u0259 Uzla\u015fd\u0131rma:<\/strong> \u00c7oxsayl\u0131 ERP sisteml\u0259ri aras\u0131nda hesab-fakturalar\u0131 avtomatik yoxlay\u0131r, uy\u011funsuzluqlar\u0131 a\u015fkarlay\u0131r v\u0259 maliyy\u0259 hesabatlar\u0131n\u0131 t\u0259rtib edir.<\/li>\n<li><strong>Dinamik Koordinasiya:<\/strong> H\u0259m korporativ, h\u0259m d\u0259 f\u0259rdi qrafikl\u0259ri \u00e7arpaz t\u0259hlil ed\u0259r\u0259k logistika v\u0259 layih\u0259 idar\u0259etm\u0259sini optimalla\u015fd\u0131r\u0131r.<\/li>\n<\/ul>\n<h2>Fundamental Elmd\u0259 AI: Biologiyadan \u0130qlim D\u0259yi\u015fikliyin\u0259<\/h2>\n<p>M\u0259ntiqi t\u0259f\u0259kk\u00fcr qabiliyy\u0259tinin y\u00fcks\u0259lm\u0259si frontier modell\u0259rin n\u0259z\u0259ri elmi probleml\u0259ri h\u0259ll etm\u0259sin\u0259 yol a\u00e7\u0131b. Google-\u0131n bu sah\u0259d\u0259ki nailiyy\u0259tl\u0259ri yaln\u0131z kod yazmaqla m\u0259hdudla\u015fm\u0131r:<\/p>\n<ul>\n<li><strong>Konnektomika v\u0259 Biologiya:<\/strong> Meyv\u0259 mil\u00e7\u0259yinin (Drosophila) beyin x\u0259rit\u0259sinin tam modell\u0259\u015fdirilm\u0259si v\u0259 insan DNT-sinin variant analizl\u0259ri s\u00fcr\u0259tl\u0259ndirilib. Bu, n\u0259z\u0259ri molekulyar biologiyan\u0131 birba\u015fa h\u0259d\u0259fli terapevtik m\u00fcalic\u0259 metodlar\u0131na \u00e7evirir.<\/li>\n<li><strong>Peyk M\u0259lumatlar\u0131 v\u0259 Qlobal \u0130qlim:<\/strong> Orbitd\u0259n daxil olan petabaytlarla vizual v\u0259 spektral m\u0259lumat real-vaxt rejimind\u0259 i\u015fl\u0259n\u0259r\u0259k metan s\u0131zmalar\u0131, me\u015f\u0259 q\u0131r\u0131lmalar\u0131 v\u0259 qlobal istil\u0259\u015fm\u0259 modell\u0259ri proqnozla\u015fd\u0131r\u0131l\u0131r.<\/li>\n<\/ul>\n<h2>Praktiki T\u0259tbiq: AI Agent \u0130\u015f Ax\u0131n\u0131n\u0131n Qurulmas\u0131<\/h2>\n<p>A\u015fa\u011f\u0131dak\u0131 b\u0259l\u0259d\u00e7id\u0259 daxil olan texniki tap\u015f\u0131r\u0131qlar\u0131 t\u0259hlil ed\u0259n, q\u0259rar ver\u0259n v\u0259 m\u00fcvafiq skripti icra ed\u0259n sad\u0259l\u0259\u015fdirilmi\u015f avtonom agent arxitekturas\u0131n\u0131n Python v\u0259 <code>google-genai<\/code> SDK vasit\u0259sil\u0259 qurulmas\u0131 n\u00fcmayi\u015f etdirilir.<\/p>\n<h3>Add\u0131m 1: M\u00fchitin Haz\u0131rlanmas\u0131<\/h3>\n<p>T\u0259l\u0259b olunan as\u0131l\u0131l\u0131qlar\u0131 virtual m\u00fchitiniz\u0259 qura\u015fd\u0131r\u0131n:<\/p>\n<pre><code class=\"language-bash\">python3 -m venv ai-agent-env\nsource ai-agent-env\/bin\/activate\npip install google-genai pydantic requests<\/code><\/pre>\n<h3>Add\u0131m 2: Al\u0259tl\u0259rin (Tools) v\u0259 Agent Strukturunun T\u0259yini<\/h3>\n<p>Agentin icra ed\u0259 bil\u0259c\u0259yi funksiyalar\u0131 (tools) m\u00fc\u0259yy\u0259nl\u0259\u015fdirin v\u0259 model\u0259 m\u00fchakim\u0259 y\u00fcr\u00fcd\u0259r\u0259k bu funksiyalar\u0131 \u00e7a\u011f\u0131rmaq icaz\u0259si verin:<\/p>\n<pre><code class=\"language-python\">import os\nfrom google import genai\nfrom google.genai import types\n\n# M\u00fc\u015ft\u0259rinin inisializasiyas\u0131\nclient = genai.Client(api_key=os.environ.get(\"GEMINI_API_KEY\"))\n\n# Agent \u00fc\u00e7\u00fcn t\u0259yin edilmi\u015f funksiya (Tool)\ndef resolve_it_ticket(ticket_id: str, action: str) -> dict:\n    \"\"\"Sistemd\u0259ki IT biletini h\u0259ll edir v\u0259 statusu yenil\u0259yir.\"\"\"\n    # Real ssenarid\u0259 burada JIRA v\u0259 ya daxili API \u00e7a\u011f\u0131r\u0131\u015f\u0131 olur\n    return {\n        \"ticket_id\": ticket_id,\n        \"status\": \"RESOLVED\",\n        \"action_taken\": action\n    }\n\n# Agentin sistem t\u0259limat\u0131 (System Instruction)\nsystem_prompt = \"\"\"\nS\u0259n \u0130T infrastrukturunu idar\u0259 ed\u0259n R\u0259q\u0259msal H\u0259mkar Agentis\u0259n.\nDaxil olan insidentl\u0259ri analiz et, add\u0131m-add\u0131m m\u0259ntiqi n\u0259tic\u0259 \u00e7\u0131xar \nv\u0259 z\u0259rur\u0259t olduqda t\u0259qdim olunan al\u0259tl\u0259rd\u0259n istifad\u0259 ed\u0259r\u0259k problemi aradan qald\u0131r.\n\"\"\"\n\n# Sor\u011funun g\u00f6nd\u0259rilm\u0259si v\u0259 funksiya \u00e7a\u011f\u0131r\u0131\u015f\u0131\nresponse = client.models.generate_content(\n    model='gemini-2.5-pro',\n    contents='Ticket #IT-8842: NGINX serverind\u0259 502 Bad Gateway x\u0259tas\u0131 qeyd\u0259 al\u0131n\u0131b. Loqlara \u0259sas\u0259n backend servisi s\u00f6n\u00fcb.',\n    config=types.GenerateContentConfig(\n        system_instruction=system_prompt,\n        tools=[resolve_it_ticket],\n        temperature=0.1 # D\u0259qiq q\u0259rarlar \u00fc\u00e7\u00fcn a\u015fa\u011f\u0131 temperatur\n    )\n)\n\n# Modelin \u00e7a\u011f\u0131r\u0131\u015f q\u0259rar\u0131n\u0131 yoxlamaq\nfor call in response.function_calls:\n    print(f\"Agent Funksiyan\u0131 \u00c7a\u011f\u0131r\u0131r: {call.name}\")\n    print(f\"Arqumentl\u0259r: {call.args}\")\n<\/code><\/pre>\n<h3>Add\u0131m 3: Agent \u018fm\u0259liyyat\u0131n\u0131n \u0130\u015f\u0259 Sal\u0131nmas\u0131<\/h3>\n<p>Kodu icra ed\u0259r\u0259k agentin daxil olan x\u0259tan\u0131 analiz edib funksiyan\u0131 \u00e7a\u011f\u0131rma q\u0259rar\u0131n\u0131 m\u00fc\u015fahid\u0259 edin:<\/p>\n<pre><code class=\"language-bash\">export GEMINI_API_KEY=\"your-api-key-here\"\npython agent_runner.py<\/code><\/pre>\n<h2>\u0130T Sektoruna v\u0259 M\u00fch\u0259ndisl\u0259r\u0259 T\u0259siri<\/h2>\n<p>M\u0259ntiqi m\u00fchakim\u0259 aparan frontier modell\u0259rin meydana \u00e7\u0131xmas\u0131 proqram t\u0259minat\u0131 m\u00fch\u0259ndisliyi v\u0259 \u0130T idar\u0259etm\u0259sind\u0259 i\u015f b\u00f6lg\u00fcs\u00fcn\u00fc d\u0259yi\u015fir. \u018fn\u0259n\u0259vi &#8220;insan kodu yaz\u0131r, komp\u00fcter icra edir&#8221; sxemi, &#8220;insan arxitekturan\u0131 v\u0259 biznes m\u0259ntiqini dizayn edir, s\u00fcni intellekt modullar\u0131 qurur v\u0259 sazlay\u0131r&#8221; format\u0131na transformasiya olunur.<\/p>\n<p>Kritik infrastrukturda i\u015fl\u0259y\u0259n m\u00fch\u0259ndisl\u0259r \u00fc\u00e7\u00fcn \u0259sas \u00e7a\u011f\u0131r\u0131\u015f bu sisteml\u0259rin izl\u0259nilm\u0259si (observability) v\u0259 deterministik olmayan (non-deterministic) davran\u0131\u015flar\u0131n s\u0259rh\u0259dl\u0259rini d\u00fczg\u00fcn t\u0259yin etm\u0259k olacaq. D\u0259rin m\u0259ntiq modell\u0259ri x\u0259talar\u0131 azaltsa da, agentl\u0259rin icra s\u0259lahiyy\u0259tl\u0259ri ciddi RBAC (Role-Based Access Control) qaydalar\u0131 il\u0259 m\u0259hdudla\u015fd\u0131r\u0131lmal\u0131d\u0131r.<\/p>\n","protected":false},"excerpt":{"rendered":"<p>S\u00fcni \u0130ntellektd\u0259 Yeni M\u0259rh\u0259l\u0259: Sad\u0259 \u00c7atbotlardan Avtonom &#8220;R\u0259q\u0259msal H\u0259mkarlar&#8221;a Ke\u00e7id Generativ s\u00fcni intellekt d\u00fcnyas\u0131 sad\u0259 m\u0259tn generasiyas\u0131 v\u0259 prompt-\u0259sasl\u0131 dialoqlardan ibar\u0259t ilkin m\u0259rh\u0259l\u0259ni gerid\u0259 qoyur. Google-\u0131n son texnoloji hesabatlar\u0131 v\u0259 frontier model inki\u015faflar\u0131 g\u00f6st\u0259rir ki, s\u0259naye art\u0131q sad\u0259c\u0259 parametr say\u0131n\u0131 art\u0131rmaqdan (brute-force scaling) \u00e7\u0131xaraq, d\u0259rin m\u0259ntiqi m\u00fchakim\u0259 (deep logical reasoning) v\u0259 m\u00fcr\u0259kk\u0259b \u0259m\u0259liyyatlar\u0131 icra ed\u0259n [&hellip;]<\/p>\n","protected":false},"author":1,"featured_media":11401,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"om_disable_all_campaigns":false,"_monsterinsights_skip_tracking":false,"_jetpack_memberships_contains_paid_content":false,"footnotes":""},"categories":[1],"tags":[],"class_list":["post-11402","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-uncategorized"],"aioseo_notices":[],"jetpack_featured_media_url":"https:\/\/bitep.net\/blog\/wp-content\/uploads\/2026\/10\/bitep-ai-1791074738606-21.jpg","jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/bitep.net\/blog\/wp-json\/wp\/v2\/posts\/11402","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/bitep.net\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/bitep.net\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/bitep.net\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/bitep.net\/blog\/wp-json\/wp\/v2\/comments?post=11402"}],"version-history":[{"count":0,"href":"https:\/\/bitep.net\/blog\/wp-json\/wp\/v2\/posts\/11402\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/bitep.net\/blog\/wp-json\/wp\/v2\/media\/11401"}],"wp:attachment":[{"href":"https:\/\/bitep.net\/blog\/wp-json\/wp\/v2\/media?parent=11402"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/bitep.net\/blog\/wp-json\/wp\/v2\/categories?post=11402"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/bitep.net\/blog\/wp-json\/wp\/v2\/tags?post=11402"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}