{"id":110707,"date":"2025-11-14T20:00:17","date_gmt":"2025-11-14T20:00:17","guid":{"rendered":"https:\/\/ekamu.net\/?p=110707"},"modified":"2025-11-14T20:00:17","modified_gmt":"2025-11-14T20:00:17","slug":"google-sima-2yi-tanitti-sanal-dunyada-kendi-kendine-ogrenen-ai-sistemi","status":"publish","type":"post","link":"https:\/\/ekamu.net\/index.php\/2025\/11\/14\/google-sima-2yi-tanitti-sanal-dunyada-kendi-kendine-ogrenen-ai-sistemi\/","title":{"rendered":"Google, SIMA 2\u2019yi tan\u0131tt\u0131: Sanal d\u00fcnyada kendi kendine \u00f6\u011frenen AI sistemi"},"content":{"rendered":"

\n

<\/span> <\/figure>\n<\/p>\n

Google DeepMind<\/strong>, yapay zeka ara\u015ft\u0131rmalar\u0131ndaki bir sonraki b\u00fcy\u00fck ad\u0131m\u0131 temsil eden SIMA 2<\/strong> i\u00e7in per\u015fembe g\u00fcn\u00fc kapsaml\u0131 bir ara\u015ft\u0131rma \u00f6n izlemesi payla\u015ft\u0131. Yeni nesil genel ama\u00e7l\u0131 ajan, Gemini\u2019nin geli\u015fmi\u015f dil ve ak\u0131l y\u00fcr\u00fctme yetenekleri ile birle\u015ferek art\u0131k yaln\u0131zca komutlar\u0131 takip eden bir sistem olmaktan \u00e7\u0131k\u0131yor. Bunun yerine bulundu\u011fu sanal d\u00fcnyay\u0131 anlamland\u0131r\u0131p etkile\u015fime girebilen<\/strong> bir yap\u0131ya kavu\u015fuyor.<\/p>\n

SIMA 2 ile \u00e7\u0131ta bir \u00fcste \u00e7\u0131kar\u0131l\u0131yor<\/b><\/p>\n

DeepMind, ilk versiyon olan SIMA 1\u2019i y\u00fczlerce saatlik oyun g\u00f6r\u00fcnt\u00fcs\u00fcyle e\u011fitmi\u015f ve ajan\u0131n \u00e7ok say\u0131da 3D oyunu insanlar gibi oynayabildi\u011fini g\u00f6stermi\u015fti. Ancak SIMA 1\u2019in karma\u015f\u0131k g\u00f6revleri tamamlama oran\u0131 yaln\u0131zca y\u00fczde 31 seviyesindeydi. Ayn\u0131 g\u00f6revlerde insanlar y\u00fczde 71 ba\u015far\u0131 sa\u011fl\u0131yordu. Bu s\u0131n\u0131rl\u0131l\u0131klar\u0131 a\u015fmak i\u00e7in geli\u015ftirilen SIMA 2 i\u00e7in yeni ajan\u0131n hem daha genel bir zeka d\u00fczeyine ula\u015ft\u0131\u011f\u0131 hem de kendi deneyimlerinden \u00f6\u011frenerek kendisini geli\u015ftirebildi\u011fi belirtiliyor.<\/p>\n

Bu \u00f6zellik, ara\u015ft\u0131rmac\u0131lara g\u00f6re daha kapsaml\u0131 robotik sistemlere ve AGI olarak tan\u0131mlanan genel ama\u00e7l\u0131 yapay zekaya do\u011fru at\u0131lan kritik bir ad\u0131m\u0131 temsil ediyor.<\/p>\n

<\/span> <\/figure>\n<\/p>\n

\u00d6te yandan SIMA 2, Gemini 2.5 Flash-Lite modeliyle g\u00fc\u00e7lendiriliyor.\u00a0 Bedenselle\u015ftirilmi\u015f ajanlar (Embodied agent) olarak tan\u0131mlanan bu yap\u0131, fiziksel veya sanal bir d\u00fcnyayla bir \u201cbeden\u201d \u00fczerinden etkile\u015fim kurarak \u00e7evresini g\u00f6zlemliyor ve buna uygun eylemler \u00fcretiyor. Bu yakla\u015f\u0131m, yaln\u0131zca takvim y\u00f6netimi ya da kod y\u00fcr\u00fctme gibi soyut i\u015flemler yapan geleneksel yapay zekalardan ayr\u0131l\u0131yor.<\/p>\n

<\/span> <\/figure>\n<\/p>\n

DeepMind\u2019da k\u0131demli ara\u015ft\u0131rmac\u0131 olan Jane Wang, SIMA 2\u2019nin art\u0131k yaln\u0131zca oyun oynamad\u0131\u011f\u0131n\u0131 kullan\u0131c\u0131 talimatlar\u0131n\u0131 ba\u011flam\u0131yla birlikte kavrayarak mant\u0131kl\u0131, tutarl\u0131 ve sa\u011fduyulu tepkiler verebildi\u011fini vurguluyor. Gemini entegrasyonu sayesinde SIMA 2\u2019nin performans\u0131 \u00f6nceki versiyonun iki kat\u0131na \u00e7\u0131km\u0131\u015f durumda.<\/p>\n

Kendi kendini e\u011fitip \u00f6\u011freniyor<\/b><\/p>\n

No Man\u2019s Sky\u201d \u00fczerindeki canl\u0131 demoda SIMA 2, gezegenin kayal\u0131k y\u00fczeyini tarif etti, \u00e7evredeki acil durum i\u015faretini tan\u0131y\u0131p bir sonraki ad\u0131m\u0131n\u0131 belirledi. Ba\u015fka bir \u00f6rnekte, \u201colgun bir domatesin rengindeki eve git<\/em>\u201d komutu verildi\u011finde \u201cDomates k\u0131rm\u0131z\u0131d\u0131r, o halde k\u0131rm\u0131z\u0131 eve gitmeliyim<\/em>\u201d diye d\u00fc\u015f\u00fcnd\u00fc ve ard\u0131ndan k\u0131rm\u0131z\u0131 eve gitti. SIMA 2 ayr\u0131ca emoji tabanl\u0131 komutlar\u0131 da anlayabiliyor<\/strong>. \u00d6rne\u011fin balta ve a\u011fa\u00e7 emojisi g\u00f6nderildi\u011finde ajan, bunu anlamland\u0131rarak gidip a\u011fa\u00e7 kesiyor. Ajan, DeepMind\u2019\u0131n Genie modeliyle olu\u015fturulan fotoger\u00e7ek\u00e7i yeni d\u00fcnyalarda da do\u011fru nesneleri tan\u0131y\u0131p banklar, a\u011fa\u00e7lar ve kelebekler gibi detaylarla etkile\u015fime girebiliyor.<\/p>\n

En dikkat \u00e7ekici yeniliklerden biri de kendi kendine \u00f6\u011frenme kapasitesi. SIMA 1 tamamen insan oynan\u0131\u015f verisiyle e\u011fitilirken SIMA 2, yaln\u0131zca ilk temelini bu veriden al\u0131yor. Sonras\u0131nda sistem, yeni ortamlara b\u0131rak\u0131ld\u0131\u011f\u0131nda ba\u015fka bir Gemini modelinden g\u00f6revler \u00fcrettiriyor<\/strong>, ba\u011f\u0131ms\u0131z bir \u00f6d\u00fcl modeli de ajan\u0131n performans\u0131n\u0131 puanl\u0131yor. Bu d\u00f6ng\u00fcde SIMA 2, kendi hatalar\u0131ndan ders \u00e7\u0131kararak tamamen AI taraf\u0131ndan \u00fcretilen geri bildirimle yeni davran\u0131\u015flar geli\u015ftiriyor<\/strong>.<\/p>\n

DeepMind, SIMA 2\u2019yi gelecekte daha kapsaml\u0131 robotik platformlara a\u00e7\u0131lacak bir kap\u0131 olarak g\u00f6r\u00fcyor. Ara\u015ft\u0131rmac\u0131lar ger\u00e7ek d\u00fcnyada g\u00f6rev yapan bir robotun y\u00fcksek seviyeli kavrama ve mant\u0131k y\u00fcr\u00fctme becerilerine ihtiya\u00e7 duydu\u011funu, SIMA 2\u2019nin de tam olarak bu \u00fcst katmanda \u00e7al\u0131\u015ft\u0131\u011f\u0131n\u0131 ifade ediyor. Buna kar\u015f\u0131n fiziksel eklemler veya tekerlekler gibi d\u00fc\u015f\u00fck seviyeli kontrol mekanizmalar\u0131 farkl\u0131 modeller taraf\u0131ndan y\u00f6netiliyor. \u015eimdilik ise SIMA 2\u2019nin fiziksel robotlara entegre edilmesi veya kamuya a\u00e7\u0131k bir s\u00fcr\u00fcm\u00fcn\u00fcn yay\u0131nlanmas\u0131 i\u00e7in belirlenmi\u015f bir takvim bulunmuyor.<\/p>\n\n

https:\/\/www.youtube.com\/embed\/Zphax4f6Rls<\/span><\/p>\n

Kaynak :\u00a0https:\/\/www.donanimhaber.com\/google-sima-2-sanal-dunyada-kendi-kendine-ogrenen-ai-sistemi–198588<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"

Google DeepMind, yapay zeka ara\u015ft\u0131rmalar\u0131ndaki bir sonraki b\u00fcy\u00fck ad\u0131m\u0131 temsil eden SIMA 2 i\u00e7in per\u015fembe g\u00fcn\u00fc kapsaml\u0131 bir ara\u015ft\u0131rma \u00f6n izlemesi payla\u015ft\u0131. Yeni nesil genel ama\u00e7l\u0131 ajan, Gemini\u2019nin geli\u015fmi\u015f dil ve ak\u0131l y\u00fcr\u00fctme yetenekleri ile …<\/p>\n","protected":false},"author":1,"featured_media":110708,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[8],"tags":[314,6565],"class_list":["post-110707","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-teknoloji","tag-arastirma","tag-yalnizca"],"_links":{"self":[{"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/posts\/110707","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/comments?post=110707"}],"version-history":[{"count":1,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/posts\/110707\/revisions"}],"predecessor-version":[{"id":110712,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/posts\/110707\/revisions\/110712"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/media\/110708"}],"wp:attachment":[{"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/media?parent=110707"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/categories?post=110707"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/tags?post=110707"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}