{"id":33141,"date":"2025-08-06T15:07:11","date_gmt":"2025-08-06T12:07:11","guid":{"rendered":"https:\/\/kibrisayna.com\/index.php\/2025\/08\/06\/openai-acik-agirlikli-modellerini-yayinladi-yapay-zeka-demokratiklesiyor-mu-son-dakika-teknoloji-haberleri\/"},"modified":"2025-08-06T15:07:11","modified_gmt":"2025-08-06T12:07:11","slug":"openai-acik-agirlikli-modellerini-yayinladi-yapay-zeka-demokratiklesiyor-mu-son-dakika-teknoloji-haberleri","status":"publish","type":"post","link":"https:\/\/kibrisayna.com\/index.php\/2025\/08\/06\/openai-acik-agirlikli-modellerini-yayinladi-yapay-zeka-demokratiklesiyor-mu-son-dakika-teknoloji-haberleri\/","title":{"rendered":"OpenAI a\u00e7\u0131k a\u011f\u0131rl\u0131kl\u0131 modellerini yay\u0131nlad\u0131: Yapay zeka demokratikle\u015fiyor mu? – Son Dakika Teknoloji Haberleri"},"content":{"rendered":"


\n<\/p>\n

\n

OpenAI, Apache 2.0 lisans\u0131 alt\u0131nda iki yeni a\u00e7\u0131k a\u011f\u0131rl\u0131kl\u0131 (open-weight) dil modelini yay\u0131nlad\u0131.<\/p>\n

“A\u00e7\u0131k a\u011f\u0131rl\u0131kl\u0131” modeller, yapay zeka<\/a> alan\u0131nda bir modelin a\u011f\u0131rl\u0131klar\u0131n\u0131n, yani e\u011fitim s\u0131ras\u0131nda \u00f6\u011frenilen parametrelerin kamuya a\u00e7\u0131k \u015fekilde payla\u015f\u0131lmas\u0131 anlam\u0131na geliyor. Bu sayede kullan\u0131c\u0131lar\u0131n bu modelleri kendi bilgisayar\u0131na kurup yerel olarak \u00e7al\u0131\u015ft\u0131rmas\u0131, pahal\u0131 aray\u00fcz hizmetlerine para \u00f6demeden kendi sistemlerine entegre edip \u00f6zelle\u015ftirmesi m\u00fcmk\u00fcn hale geliyor.<\/p>\n

\u015eirkete g\u00f6re yeni yay\u0131nlanan bu modeller, y\u00fcksek performans sunarken, t\u00fcketici d\u00fczeyindeki donan\u0131mlarda bile \u00e7al\u0131\u015fabilecek \u015fekilde tasarland\u0131. <\/p>\n

\u00d6zellikle 16 GB GPU\u2019ya sahip bir diz\u00fcst\u00fc bilgisayarda bile \u00e7al\u0131\u015fabilen bir model dikkat \u00e7ekiyor. Grafik i\u015flem birimi (GPU), matematiksel hesaplamalar\u0131 y\u00fcksek h\u0131zda ger\u00e7ekle\u015ftirebilen elektronik bir devre. Ayn\u0131 zamanda bug\u00fcn yapay zeka ara\u00e7lar\u0131n\u0131n en kritik bile\u015fenlerinden biri.<\/p>\n

YEN\u0130 MODELLER\u0130N TEKN\u0130K \u00d6ZELL\u0130KLER\u0130<\/strong><\/p>\n

\u015eirketin a\u00e7\u0131k a\u011f\u0131rl\u0131kl\u0131 olarak yay\u0131nlad\u0131\u011f\u0131 modeller ve \u00f6zellikleri ise \u015fu \u015fekilde:<\/p>\n

– gpt-oss-120b (117 milyar parametre):<\/strong><\/p>\n

OpenAI\u2019nin o4-mini modeliyle benzer ak\u0131l y\u00fcr\u00fctme performans\u0131 sunuyor ve yaln\u0131zca 80 GB GPU ile \u00e7al\u0131\u015fabiliyor.<\/p>\n

– gpt-oss-20b (21 milyar parametre):<\/strong><\/p>\n

o3-mini seviyesinde performans sunuyor ve sadece 16 GB GPU ile verimli \u015fekilde \u00e7al\u0131\u015fabiliyor.<\/p>\n

Bu parametreler, bir yapay zeka modelinin ne kadar b\u00fcy\u00fck ve karma\u015f\u0131k oldu\u011funu ifade eden teknik bir \u00f6l\u00e7\u00fc. Parametreler, bir modelin karar verirken neye odaklanaca\u011f\u0131n\u0131, neyi nas\u0131l yorumlayaca\u011f\u0131n\u0131 belirliyor.<\/p>\n

\u00d6rne\u011fin bir kelimeyi tan\u0131ma, bir c\u00fcmledeki anlam ili\u015fkisini \u00e7\u00f6zme gibi g\u00f6revlerde bu parametreler kullan\u0131l\u0131yor. Ne kadar \u00e7ok parametre varsa model de o kadar \u00e7ok bilgiyi temsil edebilir ve daha karma\u015f\u0131k ba\u011flant\u0131lar\u0131 anlayabilir.<\/p>\n

Bu nedenle OpenAI\u2019\u0131n yay\u0131nlad\u0131\u011f\u0131 son a\u00e7\u0131k a\u011f\u0131rl\u0131kl\u0131 modeller, \u00f6zellikle geli\u015ftiricilerin pahal\u0131 sunucu altyap\u0131lar\u0131na ihtiya\u00e7 duymadan g\u00fc\u00e7l\u00fc yapay zeka uygulamalar\u0131 geli\u015ftirmesini kolayla\u015ft\u0131rabilir.<\/p>\n

AKIL Y\u00dcR\u00dcTMEDE DAHA \u0130Y\u0130<\/strong><\/p>\n

OpenAI\u2019\u0131n a\u00e7\u0131klamas\u0131na g\u00f6re yeni modeller, ayn\u0131 boyuttaki di\u011fer a\u00e7\u0131k kaynak modellerden daha iyi ak\u0131l y\u00fcr\u00fctme ve ara\u00e7 kullan\u0131m\u0131 performans\u0131 g\u00f6steriyor. <\/p>\n

Modellerin internette arama yapma ve Python yaz\u0131l\u0131m<\/a> kodlar\u0131n\u0131 \u00e7al\u0131\u015ft\u0131rma gibi ara\u00e7 kullan\u0131m\u0131 gerektiren g\u00f6revlerde ba\u015far\u0131l\u0131 oldu\u011fu, ayr\u0131ca talimatlara uyma ve g\u00f6revin karma\u015f\u0131kl\u0131\u011f\u0131na g\u00f6re ak\u0131l y\u00fcr\u00fctme d\u00fczeyini ayarlama yetene\u011fine sahip oldu\u011fu belirtiliyor.<\/p>\n

OpenAI, modellerin \u00f6zellikle yaz\u0131l\u0131mc\u0131lar\u0131n kodlar\u0131 payla\u015fmak kulland\u0131\u011f\u0131 Hugging Face, GitHub, vLLM, Ollama ve llama.cpp gibi platformlarla kolayca entegre edilebilmesi i\u00e7in rehberler de yay\u0131nlad\u0131.<\/p>\n

G\u00dcVENL\u0130K VE \u015eEFFAFLIK \u00d6NLEMLER\u0130<\/strong><\/p>\n

OpenAI, bu a\u00e7\u0131k a\u011f\u0131rl\u0131kl\u0131 modelleri geli\u015ftirirken g\u00fcvenli\u011fi \u00f6n planda tuttu\u011funu s\u00f6yl\u00fcyor. Yap\u0131lan testlerde, k\u00f6t\u00fc niyetli ince ayarlarla bile modellerin biyolojik, kimyasal veya siber risk alanlar\u0131nda kullan\u0131m\u0131 a\u00e7\u0131s\u0131ndan tehlikeli seviyelere ula\u015fmad\u0131\u011f\u0131 belirlendi.<\/p>\n

\u00d6te yandan \u015firket, modellerin d\u00fc\u015f\u00fcnme s\u00fcrecini \u015feffaf \u015fekilde izleyebilmek i\u00e7in \u201cd\u00fc\u015f\u00fcnce zinciri\u201d (yapay zeka modellerinin bir soruya cevap vermeden \u00f6nce ad\u0131m ad\u0131m d\u00fc\u015f\u00fcnmesini sa\u011flayan bir teknik) \u00e7\u0131kt\u0131lar\u0131nda herhangi bir filtreleme uygulamad\u0131. Bu durum, modellerin bazen ger\u00e7ekle ba\u011fda\u015fmayan veya \u201chal\u00fcsinatif\u201d i\u00e7erikler \u00fcretmesine neden olabilir.<\/p>\n

\u015eirketin a\u00e7\u0131klamas\u0131nda, \u201cD\u00fc\u015f\u00fcnce zincirlerini bast\u0131rmak, modellerin zararl\u0131 d\u00fc\u015f\u00fcncelerini gizlemesine yol a\u00e7abilir. Bu nedenle bask\u0131 uygulamaktan ka\u00e7\u0131nd\u0131k. Geli\u015ftiricilerin bu \u00e7\u0131kt\u0131lar\u0131 izleyerek modellerin davran\u0131\u015flar\u0131n\u0131 denetleyebilmesini ama\u00e7l\u0131yoruz\u201d dendi.<\/p>\n

TESTLER NE G\u00d6STER\u0130YOR?<\/strong><\/p>\n

Yap\u0131lan k\u0131yaslamalarda, bu yeni modellerin hal\u00fcsinasyon konusunda o4-mini kadar iyi sonu\u00e7lar vermedi\u011fi g\u00f6r\u00fcld\u00fc. OpenAI, bunun model boyutlar\u0131n\u0131n daha k\u00fc\u00e7\u00fck olmas\u0131ndan kaynakland\u0131\u011f\u0131n\u0131 belirtiyor.<\/p>\n

Ancak ger\u00e7ek d\u00fcnya uygulamalar\u0131nda, \u00f6rne\u011fin web’den bilgi \u00e7ekme (RAG) veya veritaban\u0131 sorgulama gibi senaryolarda hal\u00fcsinasyonlar\u0131n daha az g\u00f6r\u00fclmesi bekleniyor.<\/p>\n

A\u00c7IK KAYNAK VE A\u00c7IK A\u011eIRLIK<\/strong><\/p>\n

OpenAI\u2019\u0131n bu hamlesi, g\u00fc\u00e7l\u00fc yapay zeka modellerini daha eri\u015filebilir h\u00e2le getirirken, geli\u015ftiricilere yeni uygulama alanlar\u0131 a\u00e7may\u0131 hedefliyor.<\/p>\n

\u00d6zellikle d\u00fc\u015f\u00fck donan\u0131mda \u00e7al\u0131\u015fan bu modeller, yapay zekan\u0131n \u201cdemokratikle\u015fmesi\u201d yolunda \u00f6nemli bir ad\u0131m olarak de\u011ferlendiriliyor.
Son d\u00f6nemde uygulamalar\u0131n \u015feffafl\u0131\u011f\u0131 konusundaki artan \u00e7abalar kapsam\u0131nda \u015firketler, geli\u015ftirdikleri uygulamalar\u0131n kodlar\u0131n\u0131 herkese a\u00e7\u0131k hale getiriyor. Buna \u201ca\u00e7\u0131k kaynakl\u0131 kod\u201d ad\u0131 veriliyor.<\/p>\n

\u00d6te yandan OpenAI\u2019\u0131n yay\u0131nlad\u0131\u011f\u0131 modellerin a\u00e7\u0131k a\u011f\u0131rl\u0131kl\u0131 olmas\u0131, ayn\u0131 zamanda a\u00e7\u0131k kaynak kodlu oldu\u011fu anlam\u0131na gelmiyor. Zira a\u00e7\u0131k a\u011f\u0131rl\u0131k sadece modelin parametrelerinin ve mimarisinin a\u00e7\u0131k olmas\u0131yken, a\u00e7\u0131k kaynak ise modelin e\u011fitim verisinin, kodlar\u0131n\u0131n ve her \u015feyinin a\u00e7\u0131k olmas\u0131 anlam\u0131na geliyor.<\/p>\n<\/div>\n


\n
<\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

OpenAI, Apache 2.0 lisans\u0131 alt\u0131nda iki yeni a\u00e7\u0131k a\u011f\u0131rl\u0131kl\u0131 (open-weight) dil modelini yay\u0131nlad\u0131. “A\u00e7\u0131k a\u011f\u0131rl\u0131kl\u0131” modeller, yapay zeka alan\u0131nda bir modelin a\u011f\u0131rl\u0131klar\u0131n\u0131n, yani e\u011fitim s\u0131ras\u0131nda \u00f6\u011frenilen parametrelerin kamuya a\u00e7\u0131k \u015fekilde payla\u015f\u0131lmas\u0131 anlam\u0131na geliyor. Bu sayede kullan\u0131c\u0131lar\u0131n bu modelleri kendi bilgisayar\u0131na kurup yerel olarak \u00e7al\u0131\u015ft\u0131rmas\u0131, pahal\u0131 aray\u00fcz hizmetlerine para \u00f6demeden kendi sistemlerine entegre edip \u00f6zelle\u015ftirmesi m\u00fcmk\u00fcn […]<\/p>\n","protected":false},"author":908,"featured_media":19582,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_publicize_message":"","jetpack_publicize_feature_enabled":true,"jetpack_social_post_already_shared":true,"jetpack_social_options":{"image_generator_settings":{"template":"highway","default_image_id":0,"font":"","enabled":false},"version":2}},"categories":[12],"tags":[],"class_list":["post-33141","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-teknoloji"],"jetpack_publicize_connections":[],"jetpack_featured_media_url":"https:\/\/kibrisayna.com\/wp-content\/uploads\/2025\/03\/Q1AzNsuVHUm4AAb2rk0tMQ.jpg","jetpack_sharing_enabled":true,"_links":{"self":[{"href":"https:\/\/kibrisayna.com\/index.php\/wp-json\/wp\/v2\/posts\/33141","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/kibrisayna.com\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/kibrisayna.com\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/kibrisayna.com\/index.php\/wp-json\/wp\/v2\/users\/908"}],"replies":[{"embeddable":true,"href":"https:\/\/kibrisayna.com\/index.php\/wp-json\/wp\/v2\/comments?post=33141"}],"version-history":[{"count":0,"href":"https:\/\/kibrisayna.com\/index.php\/wp-json\/wp\/v2\/posts\/33141\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/kibrisayna.com\/index.php\/wp-json\/wp\/v2\/media\/19582"}],"wp:attachment":[{"href":"https:\/\/kibrisayna.com\/index.php\/wp-json\/wp\/v2\/media?parent=33141"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/kibrisayna.com\/index.php\/wp-json\/wp\/v2\/categories?post=33141"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/kibrisayna.com\/index.php\/wp-json\/wp\/v2\/tags?post=33141"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}