{"id":105947,"date":"2025-10-14T08:48:07","date_gmt":"2025-10-14T08:48:07","guid":{"rendered":"https:\/\/ekamu.net\/?p=105947"},"modified":"2025-10-14T08:48:07","modified_gmt":"2025-10-14T08:48:07","slug":"openai-o3-tanitildi-iste-tarihin-en-zeki-yapay-zekasi","status":"publish","type":"post","link":"https:\/\/ekamu.net\/index.php\/2025\/10\/14\/openai-o3-tanitildi-iste-tarihin-en-zeki-yapay-zekasi\/","title":{"rendered":"OpenAI o3 tan\u0131t\u0131ld\u0131: \u0130\u015fte tarihin \u201cen zeki\u201d yapay zekas\u0131"},"content":{"rendered":"

Yapay zekan\u0131n lider \u015firketi OpenAI<\/strong>, \u201c12 Days of OpenA\u201d kapsam\u0131nda on iki g\u00fcn boyunca \u00e7e\u015fitli tan\u0131t\u0131mlar ve duyurular ger\u00e7ekle\u015ftirdi. Bu uzun etkinli\u011fin son g\u00fcn\u00fcnde ise firman\u0131n \u201cd\u00fc\u015f\u00fcnebilen\u201d yapay zeka modeli olan o1\u2019in yeni s\u00fcr\u00fcm\u00fc olan OpenAI o3 <\/strong>ve OpenAI o3-mini tan\u0131t\u0131ld\u0131<\/strong>. Firma yeni yapay zeka modelinin mevcut modellerden daha insan benzeri muhakeme yetene\u011fine sahip oldu\u011funun alt\u0131n\u0131 \u00e7izdi.<\/p>\n

OpenAI o3 ile yeni bir d\u00f6nem ba\u015fl\u0131yor<\/b><\/p>\n

Google’\u0131n bu t\u00fcrdeki ilk modelini duyurmas\u0131ndan sadece bir g\u00fcn sonra OpenAI, bug\u00fcne kadarki en yetenekli yapay zeka modelinin geli\u015ftirilmi\u015f bir versiyonunu – sorular \u00fczerinde d\u00fc\u015f\u00fcnmek i\u00e7in daha da fazla zaman alan \u2013 ile kar\u015f\u0131m\u0131za \u00e7\u0131k\u0131yor. OpenAI o3, t\u0131pk\u0131 o1 gibi sorulara daha iyi yan\u0131tlar verebilmek i\u00e7in \u201ckafa yorarak zaman ge\u00e7iriyor\u201d ve bu s\u00fcre\u00e7te mant\u0131ksal ak\u0131l y\u00fcr\u00fctme ger\u00e7ekle\u015ftiriyor.<\/p>\n

OpenAI o3, \u00f6zellikle karma\u015f\u0131k matematik<\/strong>, ileri d\u00fczey bilim sorular\u0131 <\/strong>ve kodlama <\/strong>gibi alanlarda \u00f6nceki modelden \u00fc\u00e7 kat daha iyi performans<\/strong> g\u00f6steriyor. Model, ARC-AGI <\/strong>gibi yapay zekalar\u0131n mant\u0131ksal \u00e7\u0131kar\u0131m yeteneklerini test eden bir benchmarkta da \u00fcst\u00fcn ba\u015far\u0131<\/strong> sa\u011fl\u0131yor.<\/p>\n

<\/span> o3, mant\u0131ksal d\u00fc\u015f\u00fcnme gerektiren problemleri ad\u0131m ad\u0131m \u00e7\u00f6zme konusunda e\u011fitim alm\u0131\u015f bir model. OpenAI, bu y\u00f6ntemin yapay zekan\u0131n karma\u015f\u0131k sorunlar\u0131 \u00e7\u00f6zme kabiliyetini \u00f6nemli \u00f6l\u00e7\u00fcde art\u0131rd\u0131\u011f\u0131n\u0131 belirtiyor. \u015eirket, O3’\u00fcn, SWE-Bench<\/strong> gibi bir modelin problem \u00e7\u00f6zme kapasitesini \u00f6l\u00e7en testlerde, o1’den y\u00fczde 20 daha iyi performans<\/strong> sergiledi\u011fini a\u00e7\u0131klad\u0131. <\/figure>\n

Neden o2 de\u011fil de o3?<\/b><\/p>\n

Detaylara girmeden \u00f6nce hemen ak\u0131llara gelen ilk soruyu cevapland\u0131ral\u0131m: Neden o2 de\u011fil de o3? o2\u2019nin kullan\u0131lmamas\u0131n\u0131n ard\u0131nda, \u0130ngiliz telekom\u00fcnikasyon \u015firketi O2 ile ya\u015fanabilecek olas\u0131 bir ticari marka ihtilaf\u0131 oldu\u011fu belirtiliyor.<\/p>\n

Bununla birlikte o3 ve o3-mini hen\u00fcz genel kullan\u0131ma sunulmu\u015f de\u011fil. Ancak, g\u00fcvenlik ara\u015ft\u0131rmac\u0131lar\u0131 bug\u00fcnden itibaren bir \u00f6nizleme s\u00fcr\u00fcm\u00fcne eri\u015fim i\u00e7in ba\u015fvurabilecekler. OpenAI CEO\u2019su Sam Altman\u2019\u0131n a\u00e7\u0131klamalar\u0131na g\u00f6re, bu yeni modellerin geni\u015f \u00e7apl\u0131 kullan\u0131labilir hale gelmesi biraz zaman alabilir.<\/p>\n

AGI\u2019ye bir ad\u0131m daha<\/b><\/p>\n

o3, o1 gibi, yan\u0131t \u00fcretmeden \u00f6nce \u201c\u00f6zel bir d\u00fc\u015f\u00fcnce zinciri\u201d<\/strong> \u00fczerinden mant\u0131k y\u00fcr\u00fctebiliyor. Bu s\u00fcre\u00e7, modelin kendini adeta \u201ckontrol etmesini\u201d sa\u011fl\u0131yor ve yanl\u0131\u015f bilgilendirme riskini azalt\u0131yor. Fizik, bilim ve matematik gibi karma\u015f\u0131k alanlarda g\u00fcvenilirli\u011fi art\u0131ran bu y\u00f6ntem, ayn\u0131 zamanda bir dezavantaj\u0131 da beraberinde getiriyor: Yan\u0131t s\u00fcresi. Geleneksel modellerden farkl\u0131 olarak o3, \u00e7\u00f6z\u00fcm \u00fcretmek i\u00e7in daha fazla zaman harc\u0131yor.<\/p>\n

O3\u2019\u00fcn as\u0131l fark\u0131, bir g\u00f6revi yerine getirirken birka\u00e7 ad\u0131m\u0131 \u00f6nceden planlayarak <\/strong>\u00e7\u00f6z\u00fcm odakl\u0131 bir s\u00fcre\u00e7 olu\u015fturmas\u0131. Verilen bir komutu yan\u0131tlamadan \u00f6nce, ilgili konular\u0131 de\u011ferlendirip mant\u0131\u011f\u0131n\u0131 a\u00e7\u0131kl\u0131yor ve ard\u0131ndan en do\u011fru cevab\u0131 \u00f6zetliyor.\u00a0Ek olarak modeller d\u00fc\u015f\u00fck, orta veya y\u00fcksek d\u00fc\u015f\u00fcnme s\u00fcresine ayarlanabiliyor.<\/p>\n

<\/span> OpenAI\u2019nin bu yeni modelleri yapay genel zeka<\/strong> (AGI) konusundaki tart\u0131\u015fmalar\u0131 da yeniden g\u00fcndeme ta\u015f\u0131yor. AGI, insan\u0131n yapabilece\u011fi hemen her g\u00f6revi ger\u00e7ekle\u015ftirebilen yapay zek\u00e2 sistemlerini ifade ediyor. OpenAI\u2019nin AGI\u2019ye dair kendi tan\u0131m\u0131 ise \u201cekonomik olarak de\u011ferli i\u015flerin \u00e7o\u011funda insanlardan daha iyi performans g\u00f6steren son derece otonom sistemler.\u201d <\/figure>\n
<\/span> o3\u2019\u00fcn AGI\u2019ye ne kadar yakla\u015ft\u0131\u011f\u0131n\u0131 anlamak i\u00e7in ARC-AGI <\/strong>test sonu\u00e7lar\u0131 referans al\u0131nabilir. Bu testte o1, y\u00fczde 25 ile y\u00fczde 32<\/strong> aras\u0131nda bir ba\u015far\u0131 oran\u0131 yakalayabiliyor. \u0130nsan d\u00fczeyine ula\u015fmak i\u00e7in gereken oran y\u00fczde 85 <\/strong>olsa da bu sonu\u00e7lar, \u00f6nemli bir ilerlemeye i\u015faret ediyor. Ancak s\u0131k\u0131 durun. o3\u2019\u00fcn bu testten ald\u0131\u011f\u0131 skor ise y\u00fczde 87.5!<\/strong> o1’in performans\u0131n\u0131n \u00fc\u00e7 kat\u0131na \u00e7\u0131k\u0131yor ve bu oranlara daha \u00f6nce yakla\u015f\u0131lamam\u0131\u015ft\u0131 bile. <\/figure>\n
<\/span> Di\u011fer k\u0131yaslamalarda da o3 yine rakiplerini geride b\u0131rakmay\u0131 ba\u015far\u0131yor. AIME 2024’te sadece bir soru ka\u00e7\u0131rarak y\u00fczde 96,7 puan al\u0131yor ve GPQA Diamond’da y\u00fczde 87,7’ye ula\u015f\u0131yor. Model ayr\u0131ca bilinen en zorlu de\u011ferlendirmelerde yeni rekorlar k\u0131rarak, ba\u015fka modellerin y\u00fczde 2\u2019yi bile a\u015famad\u0131\u011f\u0131 EpochAI’s Frontier Math’da problemlerin y\u00fczde 25,2’sini \u00e7\u00f6z\u00fcyor. Buradaki matematik problemlerini bu konuda uzman ki\u015filerin \u00e7\u00f6zmesi bile bazen g\u00fcnler s\u00fcrebiliyor. <\/figure>\n

Bu bir son dakika haberidir. \u0130\u00e7erik g\u00fcncelleniyor, sayfay\u0131 yenileyin.<\/em><\/p>\n\n

https:\/\/www.youtube.com\/embed\/SKBG1sqdyIU<\/span><\/p>\n

Kaynak :\u00a0https:\/\/www.donanimhaber.com\/openai-o3-tanitildi-iste-tarihin-en-zeki-yapay-zekasi–185597<\/span><\/p>\n","protected":false},"excerpt":{"rendered":"

Yapay zekan\u0131n lider \u015firketi OpenAI, \u201c12 Days of OpenA\u201d kapsam\u0131nda on iki g\u00fcn boyunca \u00e7e\u015fitli tan\u0131t\u0131mlar ve duyurular ger\u00e7ekle\u015ftirdi. Bu uzun etkinli\u011fin son g\u00fcn\u00fcnde ise firman\u0131n \u201cd\u00fc\u015f\u00fcnebilen\u201d yapay zeka modeli olan o1\u2019in yeni s\u00fcr\u00fcm\u00fc olan OpenAI o3 ve …<\/p>\n","protected":false},"author":1,"featured_media":105948,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[8],"tags":[301,566,2529,277,341],"class_list":["post-105947","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-teknoloji","tag-gun","tag-modeli","tag-modeller","tag-openai","tag-soru"],"_links":{"self":[{"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/posts\/105947","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=105947"}],"version-history":[{"count":1,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/posts\/105947\/revisions"}],"predecessor-version":[{"id":105953,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/posts\/105947\/revisions\/105953"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/media\/105948"}],"wp:attachment":[{"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/media?parent=105947"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/categories?post=105947"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/ekamu.net\/index.php\/wp-json\/wp\/v2\/tags?post=105947"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}