{"id":1078,"date":"2026-10-31T18:05:13","date_gmt":"2026-10-31T15:05:13","guid":{"rendered":"https:\/\/yarimada.gen.tr\/?p=1078"},"modified":"2026-10-25T18:13:07","modified_gmt":"2026-10-25T15:13:07","slug":"acik-agirliklar-yapay-zekada-aciklik-yanilsamasi","status":"publish","type":"post","link":"https:\/\/yarimada.gen.tr\/?p=1078","title":{"rendered":"A\u00e7\u0131k A\u011f\u0131rl\u0131klar: Yapay Zek\u00e2da A\u00e7\u0131kl\u0131k Yan\u0131lsamas\u0131"},"content":{"rendered":"\n<p class=\"wp-block-paragraph\">Richard Stallman\u2019\u0131n GNU (GNU\u2019s Not Unix \u2013 GNU, Unix De\u011fildir) projesini duyurmas\u0131n\u0131n \u00fczerinden 43 y\u0131l ge\u00e7ti. GNU projesi, \u00f6zg\u00fcr yaz\u0131l\u0131m (<em>free software<\/em>) hareketinin ba\u015flang\u0131\u00e7 noktas\u0131 oldu. Stallman\u2019\u0131n yakla\u015f\u0131m\u0131 tavizsizdi: Geli\u015ftiricilerin ve kullan\u0131c\u0131lar\u0131n \u00f6zg\u00fcrl\u00fcklerini merkeze al\u0131yordu. Bu kat\u0131 tutum, \u00f6zg\u00fcr yaz\u0131l\u0131m\u0131n i\u015f d\u00fcnyas\u0131nda daha kolay benimsenmesini isteyen kesimlerden \u00e7e\u015fitli ele\u015ftiriler al\u0131yordu. Bu ele\u015ftirilerle birlikte, 1998 y\u0131l\u0131ndan itibaren a\u00e7\u0131k kaynak (<em>open source<\/em>) terimi yayg\u0131nla\u015fmaya ba\u015flad\u0131.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u00d6zg\u00fcr yaz\u0131l\u0131mdan a\u00e7\u0131k kayna\u011fa do\u011fru ya\u015fanan kayma yaln\u0131zca s\u00f6ylemsel d\u00fczeyde kalmad\u0131; yaz\u0131l\u0131m lisanslar\u0131 d\u00fczeyinde de kendini g\u00f6sterdi. Geli\u015ftiricilerin ve kullan\u0131c\u0131lar\u0131n haklar\u0131n\u0131 temel alan, yaz\u0131l\u0131msal \u00f6zg\u00fcrl\u00fcklerin s\u00fcreklili\u011fini g\u00fcvence alt\u0131na almay\u0131 hedefleyen lisanslar\u0131n yan\u0131nda, \u015firketlerin yaz\u0131l\u0131m geli\u015ftirme ve da\u011f\u0131t\u0131m s\u00fcre\u00e7lerinde daha esnek hareket etmesine olanak sa\u011flayan lisanslar da yayg\u0131nla\u015fmaya ba\u015flad\u0131.<\/p>\n\n\n\n<!--more-->\n\n\n\n<p class=\"wp-block-paragraph\">Bu s\u00fcre\u00e7te \u00f6zg\u00fcr yaz\u0131l\u0131m terimine y\u00f6nelik ele\u015ftirilerin en temel arg\u00fcmanlar\u0131ndan biri, \u0130ngilizcedeki <em>free<\/em> kelimesinin hem \u00f6zg\u00fcr hem de \u00fccretsiz anlam\u0131na gelmesiydi. Bu \u00e7ift anlaml\u0131l\u0131\u011f\u0131n, \u00f6zg\u00fcr yaz\u0131l\u0131m\u0131n anla\u015f\u0131lmas\u0131nda kafa kar\u0131\u015f\u0131kl\u0131\u011f\u0131na yol a\u00e7abilece\u011fi d\u00fc\u015f\u00fcn\u00fcl\u00fcyordu. Ancak aradan ge\u00e7en y\u0131llar\u0131n ard\u0131ndan, a\u00e7\u0131k kaynak teriminin \u00e7ok daha b\u00fcy\u00fck bir belirsizli\u011fe yol a\u00e7t\u0131\u011f\u0131n\u0131 g\u00f6r\u00fcyoruz. Nitekim a\u00e7\u0131k kaynakla birlikte, \u00f6zg\u00fcr yaz\u0131l\u0131m\u0131n \u00fcretim \u00f6zg\u00fcrl\u00fc\u011f\u00fcne yapt\u0131\u011f\u0131 vurgu (yani kullan\u0131c\u0131n\u0131n yaz\u0131l\u0131m\u0131 yaln\u0131zca kullanmakla kalmay\u0131p inceleyip de\u011fi\u015ftirebilmesi ve yeniden da\u011f\u0131tabilmesi) da silikle\u015fti. \u00dcretim \u00f6zg\u00fcrl\u00fc\u011f\u00fcn\u00fcn yerini, kaynak kodunun yaln\u0131zca &#8220;g\u00f6r\u00fclebilir&#8221; olmas\u0131 almaya ba\u015flad\u0131. Bug\u00fcn yapay zek\u00e2 alan\u0131nda a\u00e7\u0131k kaynak olarak sunulan sistemler ise kimi zaman kaynak kodunun g\u00f6r\u00fclebilirli\u011fi iddias\u0131n\u0131n bile gerisinde bir yerde konumlan\u0131yor.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Son zamanlarda teknoloji \u015firketleri a\u00e7\u0131k kaynak yapay zek\u00e2 terimine daha s\u0131k ba\u015fvurmaya ba\u015flad\u0131. \u0130tibar kazanmak veya reg\u00fclasyon s\u00fcre\u00e7lerinde avantaj elde etmek amac\u0131yla a\u00e7\u0131kl\u0131k s\u00f6ylemini stratejik bir halkla ili\u015fkiler ve pazarlama arac\u0131na d\u00f6n\u00fc\u015ft\u00fcren \u015firketler de var. \u00d6rne\u011fin Meta\u2019n\u0131n CEO\u2019su Mark Zuckerberg, Temmuz ay\u0131nda <em>The New York Times<\/em> ile yapt\u0131\u011f\u0131 s\u00f6yle\u015fide, isim vermeden Anthropic ve OpenAI\u2019yi, yapay zek\u00e2 teknolojisi \u00fczerindeki kontrol\u00fc s\u0131k\u0131la\u015ft\u0131rmakla su\u00e7lad\u0131. Zuckerberg, yapay zek\u00e2 g\u00fcc\u00fcn\u00fcn merkezile\u015ftirilmesi yerine m\u00fcmk\u00fcn oldu\u011funca geni\u015f bir kesime da\u011f\u0131t\u0131lmas\u0131 gerekti\u011fini savunuyor. &#8220;Ki\u015fisel s\u00fcper zek\u00e2&#8221; olarak adland\u0131rd\u0131\u011f\u0131 bu yakla\u015f\u0131mda, herkesin kendi ama\u00e7lar\u0131 ve de\u011ferleri do\u011frultusunda y\u00f6nlendirebilece\u011fi g\u00fc\u00e7l\u00fc yapay zek\u00e2 sistemlerine eri\u015febilmesini \u00f6ng\u00f6r\u00fcyor. B\u00fct\u00fcn insanl\u0131k ad\u0131na karar veren tek bir merkezi s\u00fcper zek\u00e2 yerine, insanlar\u0131n kendi ama\u00e7lar\u0131 do\u011frultusunda kullanabilece\u011fi \u00e7ok say\u0131da ki\u015fisel s\u00fcper zek\u00e2n\u0131n varl\u0131\u011f\u0131n\u0131 daha makul buluyor[1].<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Zuckerberg\u2019in bu s\u00f6yle\u015fisinden k\u0131sa bir s\u00fcre sonra Meta, Muse Glimmer adl\u0131 yapay zek\u00e2 modelini yay\u0131mlad\u0131. Muse Glimmer, Meta\u2019n\u0131n g\u00fc\u00e7l\u00fc yapay zek\u00e2 modellerinden biri olan Muse Spark\u2019a benzer \u00f6zelliklere sahip; kod yazma, metin \u00fcretme, g\u00f6r\u00fcnt\u00fcleri anlama ve \u00e7ok ad\u0131ml\u0131 g\u00f6revleri yerine getirme gibi yetenekler sunuyor. Ancak Muse Spark\u2019a eri\u015fim Meta\u2019n\u0131n hizmetleri \u00fczerinden sa\u011flan\u0131rken Muse Glimmer, a\u00e7\u0131k a\u011f\u0131rl\u0131kl\u0131 bir model olarak yay\u0131mland\u0131. Burada yaz\u0131l\u0131mdaki anlam\u0131yla tamamen a\u00e7\u0131k kaynakl\u0131 bir modelden s\u00f6z etmiyoruz. A\u00e7\u0131k olan, modelin e\u011fitilmesi sonucunda ortaya \u00e7\u0131kan a\u011f\u0131rl\u0131klar, yani modelin girdilerden \u00e7\u0131kt\u0131lar \u00fcretmesini sa\u011flayan milyarlarca parametrenin olu\u015fturdu\u011fu yap\u0131[2].<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">A\u00e7\u0131k a\u011f\u0131rl\u0131klar\u0131n sundu\u011fu olanaklar ile a\u00e7\u0131k kaynak yaz\u0131l\u0131m\u0131n sa\u011flad\u0131\u011f\u0131 \u00f6zg\u00fcrl\u00fckler ayn\u0131 olmamas\u0131na ra\u011fmen, haberlerde bu iki terim s\u0131kl\u0131kla i\u00e7 i\u00e7e ge\u00e7ebiliyor. A\u00e7\u0131k a\u011f\u0131rl\u0131klar, a\u00e7\u0131k kayna\u011f\u0131n (ya da \u00f6zg\u00fcr yaz\u0131l\u0131m\u0131n!) sahip oldu\u011fu olumlu imajdan yararlan\u0131yor ve geli\u015ftiricilere benzer \u00f6zg\u00fcrl\u00fckler sunuluyormu\u015f gibi bir alg\u0131n\u0131n olu\u015fmas\u0131na neden olabiliyor. \u00d6rne\u011fin, 24 Temmuz 2026 tarihinde ABD\u2019li teknoloji \u015firketlerinin yay\u0131mlad\u0131\u011f\u0131 ortak bildiri, a\u00e7\u0131k kaynak yaz\u0131l\u0131ma \u00f6vg\u00fcyle ba\u015fl\u0131yor[3]. Bildiride a\u00e7\u0131k kaynak yaz\u0131l\u0131m\u0131n geli\u015ftiricilere sundu\u011fu inceleme, de\u011fi\u015ftirme ve geli\u015ftirme \u00f6zg\u00fcrl\u00fckleri vurguland\u0131ktan sonra benzer bir yakla\u015f\u0131m\u0131n yapay zek\u00e2ya uygulanmas\u0131 savunuluyor. Bildirinin devam\u0131nda ise herkesin indirebilece\u011fi, inceleyebilece\u011fi, de\u011fi\u015ftirebilece\u011fi ve kendi altyap\u0131s\u0131nda \u00e7al\u0131\u015ft\u0131rabilece\u011fi a\u00e7\u0131k model yapay zek\u00e2 sistemlerinin \u00f6nemi anlat\u0131l\u0131yor.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Peki a\u00e7\u0131k a\u011f\u0131rl\u0131k nedir? A\u00e7\u0131k a\u011f\u0131rl\u0131k ile a\u00e7\u0131k kaynak yapay zek\u00e2 ayn\u0131 \u015fey midir?<\/p>\n\n\n\n<h1 class=\"wp-block-heading\">A\u00e7\u0131k A\u011f\u0131rl\u0131klar<\/h1>\n\n\n\n<p class=\"wp-block-paragraph\">Makine \u00f6\u011frenimi uygulamalar\u0131, sadece yaz\u0131l\u0131mdan ibaret de\u011fildir. Yaz\u0131l\u0131m\u0131n yan\u0131 s\u0131ra e\u011fitim sonucunda elde edilen verileri de i\u00e7erir. Bu t\u00fcr verilere genel olarak model parametreleri denilebilir. Sinir a\u011f\u0131 tabanl\u0131 uygulamalarda bunlar model a\u011f\u0131rl\u0131klar\u0131 olarak bilinir. Model a\u011f\u0131rl\u0131klar\u0131, bir makine \u00f6\u011frenimi sisteminin yaz\u0131l\u0131m\u0131n\u0131n hem \u00e7\u0131kt\u0131lar\u0131 hem de girdileridir. Bu a\u011f\u0131rl\u0131klar, sistemin tepkilerini etkiler veya kontrol eder.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Model a\u011f\u0131rl\u0131klar\u0131, makine \u00f6\u011frenimi e\u011fitim yaz\u0131l\u0131m\u0131n\u0131n e\u011fitim verileri \u00fczerinde \u00e7al\u0131\u015ft\u0131r\u0131lmas\u0131yla olu\u015fturulur. Ancak bu a\u011f\u0131rl\u0131klar, bir insan geli\u015ftirici taraf\u0131ndan yaz\u0131lan kodlar\u0131n do\u011frudan bir derlemesi ya da \u00e7evirisi de\u011fildir. Model parametreleri insanlar taraf\u0131ndan do\u011frudan anla\u015f\u0131lamaz. Bu nedenle bir makine \u00f6\u011frenimi uygulamas\u0131n\u0131 model parametrelerini do\u011frudan analiz ederek veya d\u00fczenleyerek incelemek veya uyarlamak pratik de\u011fildir. Ayr\u0131ca, model parametrelerinin hesaplanmas\u0131 genellikle e\u011fitim verilerinde toplanan \u00e7ok say\u0131da \u00f6rne\u011fin i\u015flenmesini gerektirir. Bu \u015fekilde olu\u015fturulan kontrol verileri \u00fczerindeki herhangi bir \u00f6rne\u011fin etkisi \u00f6rt\u00fck ve dolayl\u0131 olabilir. Pratikte bir makine \u00f6\u011frenimi uygulamas\u0131n\u0131 incelemek ve \u00f6zelle\u015ftirmek; modeli farkl\u0131 istem (<em>prompt<\/em>) k\u00fcmeleriyle test etmeyi, e\u011fitim verilerini analiz etmeyi ya da modeli s\u0131f\u0131rdan veya art\u0131ml\u0131 olarak yeniden e\u011fitmeyi gerektirir[4].<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">GPT (OpenAI) veya Claude (Anthropic) gibi yapay zek\u00e2 modellerine genellikle bir API (Application Programming Interface &#8211; Uygulama Programlama Aray\u00fcz\u00fc) \u00fczerinden eri\u015filir. Llama (Meta), Mistral, Qwen (Alibaba), DeepSeek ve Gemma (Google) gibi a\u00e7\u0131k a\u011f\u0131rl\u0131kl\u0131 modellerde ise e\u011fitim s\u00fcrecinde elde edilen a\u011f\u0131rl\u0131klar, kullan\u0131c\u0131lar\u0131n indirip kendi altyap\u0131lar\u0131nda \u00e7al\u0131\u015ft\u0131rabilece\u011fi \u015fekilde yay\u0131mlan\u0131r. B\u00f6ylece yeterli donan\u0131ma sahip kullan\u0131c\u0131lar; geli\u015ftiriciye belirte\u00e7 (<em>token<\/em>) ba\u015f\u0131na API \u00fccreti \u00f6demeden modeli kendi sistemlerinde \u00e7al\u0131\u015ft\u0131rabilir, ince ayar (<em>fine-tuning<\/em>) uygulayabilir ve kendi yaz\u0131l\u0131mlar\u0131na entegre edebilir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Bir makine \u00f6\u011frenimi modelinin a\u011f\u0131rl\u0131klar\u0131na eri\u015fimi olan kullan\u0131c\u0131lar,<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Belirli bir g\u00f6rev veya alan i\u00e7in kendi verilerini kullanarak modele ince ayar yapabilir.<\/li>\n\n\n\n<li>Modelin a\u011f\u0131rl\u0131klar\u0131n\u0131 inceleyebilir, de\u011fi\u015ftirebilir veya budayabilir.<\/li>\n\n\n\n<li>Modeli daha k\u00fc\u00e7\u00fck veya daha ucuz donan\u0131mlarda \u00e7al\u0131\u015ft\u0131rmak i\u00e7in modelin boyutunu azaltabilir.<\/li>\n\n\n\n<li>Modeli tamamen kendi altyap\u0131lar\u0131nda \u00e7al\u0131\u015ft\u0131rabilir. Verileri model sa\u011flay\u0131c\u0131s\u0131n\u0131n sunucular\u0131na g\u00f6nderme zorunlulu\u011funun ortadan kalkmas\u0131, \u00f6zellikle sa\u011fl\u0131k, hukuk ve finans gibi hassas ve s\u0131k\u0131 reg\u00fclasyona tabi alanlarda \u00f6nemli bir avantaj sa\u011flar.<\/li>\n\n\n\n<li>Belirte\u00e7 ba\u015f\u0131na API \u00fccreti \u00f6deme zorunlulu\u011fu ortadan kalkar<\/li>\n\n\n\n<li>\u015eirketin fiyatland\u0131rmas\u0131n\u0131 de\u011fi\u015ftirmesi, modeli kullan\u0131mdan kald\u0131rmas\u0131, sessiz bir g\u00fcncellemeyle davran\u0131\u015f\u0131n\u0131 de\u011fi\u015ftirmesi veya eri\u015fimi kapatmas\u0131 gibi d\u0131\u015f ba\u011f\u0131ml\u0131l\u0131klardan kaynaklanan riskleri azaltabilir.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">Fakat kullan\u0131c\u0131n\u0131n a\u00e7\u0131k a\u011f\u0131rl\u0131klar\u0131 kullanarak kendi modelini \u00e7al\u0131\u015ft\u0131rmas\u0131n\u0131n baz\u0131 dezavantajlar\u0131 da vard\u0131r:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Modelin verimli bir \u015fekilde \u00e7al\u0131\u015fmas\u0131 i\u00e7in yeterli i\u015flem g\u00fcc\u00fcne, \u00f6zellikle GPU\u2019lara ve teknik uzmanl\u0131\u011fa ihtiya\u00e7 vard\u0131r.<\/li>\n\n\n\n<li>A\u00e7\u0131k a\u011f\u0131rl\u0131kl\u0131 sunulan sistemler, genel yetenekler a\u00e7\u0131s\u0131ndan bazen en geli\u015fmi\u015f kapal\u0131 modellerin gerisinde kalabilmektedir.<\/li>\n\n\n\n<li>G\u00fcvenlik, izleme ve g\u00fcncelleme sorumlulu\u011fu kullan\u0131c\u0131ya b\u0131rak\u0131l\u0131r.<\/li>\n\n\n\n<li>Lisans ko\u015fullar\u0131 her a\u00e7\u0131k a\u011f\u0131rl\u0131kl\u0131 modelde ayn\u0131 olmayabilir. \u00d6rne\u011fin baz\u0131 modeller, belirli bir \u00f6l\u00e7e\u011fin \u00fczerindeki ticari kullan\u0131mlara k\u0131s\u0131tlamalar getirebilir.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><a><\/a>T\u00fcm bu dezavantajlar\u0131na ra\u011fmen, bir\u00e7ok durumda a\u00e7\u0131k a\u011f\u0131rl\u0131klar\u0131n \u00f6zel m\u00fclkiyetli yapay zek\u00e2 modellerine API \u00fczerinden eri\u015fmekten daha avantajl\u0131 bir se\u00e7enek oldu\u011funu s\u00f6yleyebiliriz. Ancak a\u00e7\u0131k a\u011f\u0131rl\u0131klar\u0131n bu a\u00e7\u0131dan daha avantajl\u0131 olmas\u0131, onlar\u0131n a\u00e7\u0131k kaynak yapay zek\u00e2 olarak adland\u0131r\u0131labilece\u011fi anlam\u0131na gelmez.<\/p>\n\n\n\n<h1 class=\"wp-block-heading\"><a><\/a>A\u00e7\u0131k Kaynak Yapay Zek\u00e2<\/h1>\n\n\n\n<p class=\"wp-block-paragraph\">A\u00e7\u0131k a\u011f\u0131rl\u0131kl\u0131 yapay zek\u00e2 modelleri, kullan\u0131c\u0131lar\u0131n teknoloji \u00fczerindeki kontrol\u00fcn\u00fc art\u0131r\u0131r. Fakat a\u00e7\u0131k a\u011f\u0131rl\u0131klar\u0131n payla\u015f\u0131lmas\u0131, bir yapay zek\u00e2 sistemini a\u00e7\u0131k kaynak yapay zek\u00e2 olarak tan\u0131mlamak i\u00e7in yeterli de\u011fildir. OSI (Open Source Initiative &#8211; A\u00e7\u0131k Kaynak \u0130nisiyatifi) taraf\u0131ndan da vurguland\u0131\u011f\u0131 \u00fczere, a\u00e7\u0131k kaynak yapay zek\u00e2 a\u015fa\u011f\u0131daki \u00f6zg\u00fcrl\u00fckleri sa\u011flamal\u0131d\u0131r[5]:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Sistemi herhangi bir ama\u00e7la ve izin istemek zorunda kalmadan kullanmak.<\/li>\n\n\n\n<li>Sistemin nas\u0131l \u00e7al\u0131\u015ft\u0131\u011f\u0131n\u0131 incelemek ve bile\u015fenlerini denetlemek.<\/li>\n\n\n\n<li>\u00c7\u0131kt\u0131lar\u0131n\u0131 \u00f6zelle\u015ftirmek de d\u00e2hil olmak \u00fczere, sistemi ihtiya\u00e7lar do\u011frultusunda de\u011fi\u015ftirmek.<\/li>\n\n\n\n<li>Sistemi, \u00fczerinde de\u011fi\u015fiklik yap\u0131lm\u0131\u015f ya da yap\u0131lmam\u0131\u015f haliyle, ba\u015fkalar\u0131n\u0131n kullan\u0131m\u0131na sunmak veya payla\u015fmak.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">Bu \u00f6zg\u00fcrl\u00fcklerin kullan\u0131labilmesi i\u00e7in temel \u00f6n ko\u015ful, sistemde de\u011fi\u015fiklik yapmaya uygun formatlara ve bu de\u011fi\u015fiklikleri ger\u00e7ekle\u015ftirecek pratik ara\u00e7lara eri\u015fimdir. \u00d6zg\u00fcr ve a\u00e7\u0131k kaynak yaz\u0131l\u0131mdaki kullan\u0131m, inceleme, de\u011fi\u015ftirme ve payla\u015fma olanaklar\u0131n\u0131 yapay zek\u00e2da da sa\u011flayabilmek i\u00e7in kullan\u0131c\u0131lar\u0131n model a\u011f\u0131rl\u0131klar\u0131n yan\u0131nda a\u015fa\u011f\u0131daki bilgilere ve bile\u015fenlere eri\u015febilmesi gerekir:<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Veri Bilgisi:<\/strong> Yetkin bir ki\u015finin b\u00fcy\u00fck \u00f6l\u00e7\u00fcde e\u015fde\u011fer bir sistem olu\u015fturabilmesini sa\u011flayacak d\u00fczeyde, sistemin e\u011fitilmesinde kullan\u0131lan verilere ili\u015fkin yeterince ayr\u0131nt\u0131l\u0131 bilgi. Veri Bilgisi, OSI onayl\u0131 ko\u015fullar alt\u0131nda kullan\u0131ma sunulmal\u0131d\u0131r.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Bu bilgiler \u00f6zellikle \u015funlar\u0131 i\u00e7ermelidir:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Payla\u015f\u0131lamayan veriler de d\u00e2hil olmak \u00fczere, e\u011fitimde kullan\u0131lan t\u00fcm verilerin eksiksiz bir a\u00e7\u0131klamas\u0131. Bu kapsamda verilerin kayna\u011f\u0131, kapsam\u0131, nitelikleri, nas\u0131l elde edildi\u011fi ve se\u00e7ildi\u011fi, etiketleme prosed\u00fcrleri ile veri i\u015fleme ve filtreleme y\u00f6ntemleri ayr\u0131nt\u0131lar\u0131yla belirtilmelidir.<\/li>\n\n\n\n<li>Herkese a\u00e7\u0131k t\u00fcm e\u011fitim verilerinin ve bunlara eri\u015fim sa\u011flanabilecek kaynaklar\u0131n d\u00f6k\u00fcm\u00fc.<\/li>\n\n\n\n<li>\u00dccretli olanlar da d\u00e2hil olmak \u00fczere, \u00fc\u00e7\u00fcnc\u00fc taraflardan temin edilen e\u011fitim verilerinin ve bunlar\u0131n sa\u011fland\u0131\u011f\u0131 platformlar\u0131n bir listesi.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Kod:<\/strong> Sistemin e\u011fitilmesi ve \u00e7al\u0131\u015ft\u0131r\u0131lmas\u0131 i\u00e7in kullan\u0131lan kaynak kodun tamam\u0131. Kod, verilerin nas\u0131l i\u015flenip filtrelendi\u011fine ve e\u011fitimin nas\u0131l ger\u00e7ekle\u015ftirildi\u011fine ili\u015fkin t\u00fcm teknik ayr\u0131nt\u0131lar\u0131 yans\u0131tmal\u0131d\u0131r. Kod, OSI onayl\u0131 lisanslar alt\u0131nda kullan\u0131ma sunulmal\u0131d\u0131r.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u00d6rne\u011fin; veri i\u015fleme ve filtreleme kodlar\u0131, kullan\u0131lan arg\u00fcmanlar ve ayarlar d\u00e2hil olmak \u00fczere e\u011fitim kodlar\u0131, do\u011frulama ve test s\u00fcre\u00e7leri, belirte\u00e7le\u015ftiriciler (tokenizer\u2019lar) ile hiperparametre arama kodlar\u0131 gibi destekleyici k\u00fct\u00fcphaneler, \u00e7\u0131kar\u0131m (inference) kodu ve model mimarisi bu kapsama d\u00e2hil edilmelidir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>Parametreler:<\/strong> A\u011f\u0131rl\u0131klar veya di\u011fer yap\u0131land\u0131rma ayarlar\u0131 gibi model parametreleri. Parametreler, OSI onayl\u0131 ko\u015fullar alt\u0131nda sunulmal\u0131d\u0131r.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Bu kapsamda e\u011fitimin kritik ara a\u015famalar\u0131ndan al\u0131nan kontrol noktalar\u0131 (<em>checkpoints<\/em>) ve son optimizasyon durumu da yer alabilir.<\/p>\n\n\n\n<h1 class=\"wp-block-heading\">A\u00e7\u0131k A\u011f\u0131rl\u0131klar\u0131n Dezavantajlar\u0131<\/h1>\n\n\n\n<p class=\"wp-block-paragraph\">A\u00e7\u0131k a\u011f\u0131rl\u0131klar\u0131n, tam anlam\u0131yla a\u00e7\u0131k kaynak yapay zek\u00e2 sistemleriyle kar\u015f\u0131la\u015ft\u0131r\u0131ld\u0131\u011f\u0131nda d\u00f6rt temel dezavantaj\u0131 bulunmaktad\u0131r[6]:<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>1. Tekrarlanabilirlik<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Tekrarlanabilirlik, bilimsel ve teknolojik ilerlemede kritik bir \u00f6neme sahiptir. E\u011fitim koduna veya ara kontrol noktalar\u0131na eri\u015fim olmad\u0131\u011f\u0131nda, ara\u015ft\u0131rmac\u0131lar ve denet\u00e7iler modelin geli\u015ftirme s\u00fcrecini s\u0131f\u0131rdan tekrarlayamaz veya do\u011frulayamaz. Bu eksiklik, modellerdeki \u00f6nyarg\u0131lar\u0131n (<em>bias<\/em>) tam olarak nerede ve ne zaman ortaya \u00e7\u0131kt\u0131\u011f\u0131n\u0131 tespit etmeyi zorla\u015ft\u0131rd\u0131\u011f\u0131 gibi hatalar\u0131n veya g\u00fcvenlik a\u00e7\u0131klar\u0131n\u0131n temel nedenlerini ara\u015ft\u0131rma ve giderme olanaklar\u0131n\u0131 da ciddi \u015fekilde s\u0131n\u0131rlar.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>2. Veri \u015eeffafl\u0131\u011f\u0131<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Yapay zek\u00e2 sistemlerinin e\u011fitimi i\u00e7in kullan\u0131lan verilerin temsil yetene\u011fi zay\u0131fsa veya veriler etik olmayan y\u00f6ntemlerle elde edilmi\u015fse, bu kusurlar modelin \u00e7\u0131kt\u0131lar\u0131na da yans\u0131yabilir. Yaln\u0131zca a\u011f\u0131rl\u0131klar\u0131 payla\u015f\u0131lan modellerde veri setinin nas\u0131l derlendi\u011finin, se\u00e7ildi\u011finin veya temizlendi\u011finin \u015feffafl\u0131kla a\u00e7\u0131klanmamas\u0131 ara\u015ft\u0131rmac\u0131lar\u0131n kullan\u0131lan verinin kalitesini, \u00e7e\u015fitlili\u011fini ve g\u00fcvenilirli\u011fini tam olarak de\u011ferlendirmesini engeller.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Bir modelin adilli\u011fi b\u00fcy\u00fck \u00f6l\u00e7\u00fcde veri kalitesine ve e\u011fitim s\u00fcre\u00e7lerinin nas\u0131l tasarland\u0131\u011f\u0131na ba\u011fl\u0131d\u0131r. A\u00e7\u0131k kaynak yapay zek\u00e2da ara\u015ft\u0131rmac\u0131lar ve denet\u00e7iler, kullan\u0131lan verileri ve e\u011fitim s\u00fcre\u00e7lerini analiz ederek potansiyel yanl\u0131l\u0131klar\u0131 ve \u00f6nyarg\u0131lar\u0131 daha erken a\u015famada tespit edebilir, bunlar\u0131n giderilmesine katk\u0131da bulunabilir. Yaln\u0131zca model a\u011f\u0131rl\u0131klar\u0131n\u0131n payla\u015f\u0131lmas\u0131 ise bu nitelikte bir denetim i\u00e7in gerekli verileri yeterli d\u00fczeyde sunamaz.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Veri ihlalleri ve algoritmik tart\u0131\u015fmalar\u0131n ya\u015fand\u0131\u011f\u0131 bir d\u00f6nemde, yapay zek\u00e2 sistemlerine duyulan g\u00fcven de \u015feffafl\u0131ktan etkilenmektedir. \u00d6nyarg\u0131, mahremiyet ve hesap verebilirlik konular\u0131ndaki duyarl\u0131l\u0131\u011f\u0131n artmas\u0131, yaln\u0131zca a\u011f\u0131rl\u0131klar\u0131 eri\u015fime a\u00e7\u0131lan modellerin kamuoyu taraf\u0131ndan nas\u0131l de\u011ferlendirilece\u011fi sorusunu daha \u00f6nemli h\u00e2le getiriyor. Modelin nas\u0131l geli\u015ftirildi\u011fine ili\u015fkin daha kapsaml\u0131 bilgilerin sunulmas\u0131, g\u00fcvenin olu\u015fmas\u0131na ve sistemlerin kabul g\u00f6rmesine katk\u0131da bulunabilir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>3. D\u00fczenleyici Engeller<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\">D\u00fcnya genelindeki h\u00fck\u00fcmetler ve d\u00fczenleyici kurumlar, \u00f6zellikle finans, sa\u011fl\u0131k ve kamu y\u00f6netimi gibi hassas alanlarda konumland\u0131r\u0131lan yapay zek\u00e2 sistemleri i\u00e7in \u00e7ok daha kat\u0131 \u015feffafl\u0131k standartlar\u0131 \u00f6ng\u00f6ren politikalar geli\u015ftiriyor. Yaln\u0131zca nihai a\u011f\u0131rl\u0131klar\u0131n a\u00e7\u0131klanmas\u0131, bu t\u00fcr d\u00fczenlemelerin gerektirdi\u011fi \u015feffafl\u0131k d\u00fczeyini kar\u015f\u0131lamayabilir. E\u011fitim kodunun ya da veri setlerine dair ayr\u0131nt\u0131lar\u0131n eksikli\u011fi sistemin adillik, gizlilik ve a\u00e7\u0131klanabilirlik gibi yasal gereklilikler a\u00e7\u0131s\u0131ndan denetlenmesini ve onaylanmas\u0131n\u0131 olduk\u00e7a zorla\u015ft\u0131rmaktad\u0131r.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><strong>4. S\u0131n\u0131rl\u0131 Topluluk \u0130\u015f Birli\u011fi<\/strong><\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a><\/a> A\u00e7\u0131k kaynak yapay zek\u00e2n\u0131n temel g\u00fc\u00e7l\u00fc y\u00f6nlerinden biri, ortaya \u00e7\u0131kard\u0131\u011f\u0131 i\u015f birli\u011fi potansiyelidir. E\u011fitim kodlar\u0131, veri k\u00fcmesi olu\u015fturma s\u00fcre\u00e7leri ve ara kontrol noktalar\u0131 d\u00e2hil t\u00fcm i\u015f ak\u0131\u015f\u0131 \u015feffaf\u00e7a payla\u015f\u0131ld\u0131\u011f\u0131nda, k\u00fcresel ara\u015ft\u0131rmac\u0131 toplulu\u011fu modelin iyile\u015ftirilmesi, hatalar\u0131n ay\u0131klanmas\u0131 ve etik kayg\u0131lar\u0131n giderilmesi i\u00e7in kolektif bir \u015fekilde \u00e7al\u0131\u015fabilir. Buna kar\u015f\u0131l\u0131k, yaln\u0131zca a\u00e7\u0131k a\u011f\u0131rl\u0131klar\u0131n sunuldu\u011fu modellerde bu olanaklar \u00f6nemli \u00f6l\u00e7\u00fcde k\u0131s\u0131tlan\u0131r ve toplulu\u011fun katk\u0131s\u0131 daha \u00e7ok mevcut model \u00fczerinde yap\u0131lan ince ayarlarla s\u0131n\u0131rl\u0131 kal\u0131r.<\/p>\n\n\n\n<h1 class=\"wp-block-heading\">FSF: Yaz\u0131l\u0131m ya \u00f6zg\u00fcrd\u00fcr ya da de\u011fildir<\/h1>\n\n\n\n<p class=\"wp-block-paragraph\">FSF\u2019nin (Free Software Foundation &#8211; \u00d6zg\u00fcr Yaz\u0131l\u0131m Vakf\u0131) 2024\u2019te ba\u015flatt\u0131\u011f\u0131 \u00f6zg\u00fcr makine \u00f6\u011frenimi uygulamalar\u0131na ili\u015fkin kriterler, Eyl\u00fcl 2026 itibar\u0131yla hen\u00fcz nihai bi\u00e7imiyle yay\u0131mlanm\u0131\u015f g\u00f6r\u00fcnm\u00fcyor. Ancak 2024\u2019ten bu yana FSF temsilcilerinin kat\u0131ld\u0131\u011f\u0131 panellerde yapt\u0131klar\u0131 a\u00e7\u0131klamalar[7] ve GNU e-posta listesindeki de\u011ferlendirmeler, FSF\u2019nin makine \u00f6\u011frenimi uygulamalar\u0131nda \u00f6zg\u00fcrl\u00fc\u011f\u00fc yaln\u0131zca uygulama yaz\u0131l\u0131m\u0131yla s\u0131n\u0131rl\u0131 g\u00f6rmedi\u011fine i\u015faret ediyor. FSF\u2019nin yakla\u015f\u0131m\u0131; e\u011fitim verileri, model ve yaz\u0131l\u0131m kaynak kodunun birlikte kullan\u0131c\u0131ya sunulmas\u0131n\u0131 gerektiren daha kapsaml\u0131 bir \u00f6zg\u00fcrl\u00fck anlay\u0131\u015f\u0131na dayan\u0131yor. Ayr\u0131ca FSF, genel bir &#8220;yapay zek\u00e2&#8221; kavram\u0131 yerine tart\u0131\u015fmay\u0131 makine \u00f6\u011frenimine odakl\u0131yor. B\u00fcy\u00fck dil modellerini de bu \u00e7er\u00e7evede de\u011ferlendiriyor[4].<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Bu ba\u011flamda, \u00f6zg\u00fcr bir makine \u00f6\u011frenimi uygulamas\u0131nda yer alan t\u00fcm yaz\u0131l\u0131mlar\u0131n her kullan\u0131c\u0131ya \u00f6zg\u00fcr yaz\u0131l\u0131m\u0131 tan\u0131mlayan d\u00f6rt temel \u00f6zg\u00fcrl\u00fc\u011f\u00fc sunmas\u0131 gerekiyor:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>Herhangi bir ama\u00e7 i\u00e7in program\u0131 istedi\u011finiz gibi \u00e7al\u0131\u015ft\u0131rma \u00f6zg\u00fcrl\u00fc\u011f\u00fc <strong>(\u00f6zg\u00fcrl\u00fck 0)<\/strong>.<\/li>\n\n\n\n<li>Program\u0131n nas\u0131l \u00e7al\u0131\u015ft\u0131\u011f\u0131n\u0131 inceleme ve kendi ihtiya\u00e7lar\u0131n\u0131za g\u00f6re de\u011fi\u015ftirme \u00f6zg\u00fcrl\u00fc\u011f\u00fc <strong>(\u00f6zg\u00fcrl\u00fck 1)<\/strong>. Kaynak koduna eri\u015fim bunun i\u00e7in bir \u00f6n ko\u015fuldur.<\/li>\n\n\n\n<li>Ba\u015fkalar\u0131na yard\u0131m edebilmeniz i\u00e7in kopyalar\u0131 yeniden da\u011f\u0131tma \u00f6zg\u00fcrl\u00fc\u011f\u00fc <strong>(\u00f6zg\u00fcrl\u00fck 2)<\/strong>.<\/li>\n\n\n\n<li>De\u011fi\u015ftirilmi\u015f s\u00fcr\u00fcmlerin kopyalar\u0131n\u0131 ba\u015fkalar\u0131na da\u011f\u0131tma \u00f6zg\u00fcrl\u00fc\u011f\u00fc <strong>(\u00f6zg\u00fcrl\u00fck 3)<\/strong>. B\u00f6ylece yapt\u0131\u011f\u0131n\u0131z de\u011fi\u015fikliklerden t\u00fcm toplulu\u011fun yararlanmas\u0131n\u0131 sa\u011flayabilirsiniz. Kaynak koduna eri\u015fim bunun i\u00e7in bir \u00f6n ko\u015fuldur.<\/li>\n<\/ul>\n\n\n\n<p class=\"wp-block-paragraph\">D\u00f6rt temel \u00f6zg\u00fcrl\u00fc\u011f\u00fcn kullan\u0131c\u0131lara tan\u0131nmas\u0131, makine \u00f6\u011frenimi uygulamas\u0131n\u0131n e\u011fitim s\u00fcrecini temsil eden model parametrelerinin de sunulmas\u0131n\u0131 gerektiriyor. Kullan\u0131c\u0131lar, bu parametreleri ve bunlar\u0131n de\u011fi\u015ftirilmi\u015f versiyonlar\u0131n\u0131 kullanma ve yeniden da\u011f\u0131tma \u00f6zg\u00fcrl\u00fc\u011f\u00fcne sahip olmal\u0131lar[8]. Ancak \u00f6zg\u00fcrl\u00fck ister e\u011fitim verisini i\u015fleyen arka plan kodlar\u0131 olsun ister model parametrelerini anlaml\u0131 yan\u0131tlara d\u00f6n\u00fc\u015ft\u00fcren \u00e7\u0131kar\u0131m (<em>inference<\/em>) yaz\u0131l\u0131mlar\u0131 olsun, sistemin t\u00fcm bile\u015fenleri i\u00e7in ge\u00e7erli. Dolay\u0131s\u0131yla makine \u00f6\u011frenimi uygulamalar\u0131nda verinin ta\u015f\u0131d\u0131\u011f\u0131 \u00f6nem d\u00fc\u015f\u00fcn\u00fcld\u00fc\u011f\u00fcnde, yaz\u0131l\u0131m\u0131n \u00f6zg\u00fcr olmas\u0131 tek ba\u015f\u0131na yeterli olam\u0131yor. Bu nedenle FSF, t\u00fcm e\u011fitim verileri ile bu verileri i\u015fleyen kodlar d\u00f6rt temel \u00f6zg\u00fcrl\u00fc\u011fe uygun bi\u00e7imde sunulmad\u0131k\u00e7a, bir makine \u00f6\u011frenimi uygulamas\u0131n\u0131n \u00f6zg\u00fcr olarak nitelendirilemeyece\u011fini vurguluyor[9].<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">FSF, teknik terimleri kullan\u0131rken g\u00f6sterdi\u011fi hassasiyeti \u00f6zg\u00fcr yaz\u0131l\u0131m\u0131n kapsam\u0131n\u0131 \u00e7izerken de g\u00f6steriyor. FSF i\u00e7in birinci derece \u00f6nemli olan konu, kullan\u0131c\u0131lar\u0131n kendi bilgisayar kullan\u0131mlar\u0131n\u0131 kontrol etme \u00f6zg\u00fcrl\u00fckleri.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Bu ba\u011flamda FSF, \u00f6zg\u00fcrl\u00fck ile adalet aras\u0131nda net bir ayr\u0131m yap\u0131yor. \u00d6rne\u011fin, baz\u0131 \u00f6zg\u00fcr olmayan makine \u00f6\u011frenimi uygulamalar\u0131nda ki\u015fisel t\u0131bbi veriler gibi e\u011fitim verilerinin yay\u0131mlanmamas\u0131 i\u00e7in ge\u00e7erli ahlaki nedenler bulunabilir. FSF bu t\u00fcr durumlar i\u00e7in \u00f6zg\u00fcr yaz\u0131l\u0131m tan\u0131m\u0131na bir istisna getirmiyor: Yaz\u0131l\u0131m ya \u00f6zg\u00fcrd\u00fcr ya da de\u011fildir; ortas\u0131 yok! Bununla birlikte, \u00f6zg\u00fcr olmayan bir yaz\u0131l\u0131m tedavi veya hastal\u0131k te\u015fhisi gibi toplum i\u00e7in hayati \u00f6nem ta\u015f\u0131yan \u00f6zel bir i\u015fin yap\u0131lmas\u0131na yard\u0131mc\u0131 oluyorsa, bu yaz\u0131l\u0131m\u0131n kullan\u0131m\u0131n\u0131n etik a\u00e7\u0131dan mazur g\u00f6r\u00fclebilece\u011fini belirtiyor.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Ancak \u00f6zg\u00fcr olmayan b\u00f6yle bir makine \u00f6\u011frenimi uygulamas\u0131n\u0131n kullan\u0131m\u0131n\u0131 etik a\u00e7\u0131dan kabul edilebilir de\u011ferlendirebilmek i\u00e7in baz\u0131 temel ko\u015fullar\u0131n sa\u011flanmas\u0131 beklenir. \u00d6ncelikle sistemi olu\u015fturan bile\u015fen yaz\u0131l\u0131mlar\u0131n \u00f6zg\u00fcr olmas\u0131 \u015fartt\u0131r. Ayr\u0131ca makine \u00f6\u011frenimi uygulamas\u0131n\u0131n bir b\u00fct\u00fcn olarak; kullan\u0131c\u0131lar\u0131n art\u0131ml\u0131 e\u011fitim (<em>incremental training<\/em>) yapmas\u0131n\u0131, modeli s\u0131f\u0131rdan yeniden e\u011fitmesini veya her ikisini de makul ve esnek bir bi\u00e7imde ger\u00e7ekle\u015ftirmesini destekleyecek \u015fekilde da\u011f\u0131t\u0131lmas\u0131 gerekmektedir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a><\/a> FSF\u2019nin \u00f6zg\u00fcr yaz\u0131l\u0131m\u0131n\u0131n kapsam\u0131 hakk\u0131ndaki ilkeli duru\u015funun bir ba\u015fka \u00f6rne\u011fini RAIL (Responsible AI Licenses &#8211; Sorumlu Yapay Zek\u00e2 Lisanslar\u0131) konusunda g\u00f6r\u00fcyoruz. RAIL\u2019de oldu\u011fu gibi baz\u0131 \u00e7evreler lisanslar\u0131 genel olarak sosyal adaletsizlikleri ele almak i\u00e7in bir ara\u00e7 olarak ele al\u0131yorlar. Ancak etik lisanslar olarak sunulan bu tip lisanslar, belirli ama\u00e7lar i\u00e7in kullan\u0131c\u0131lar\u0131n yaz\u0131l\u0131m\u0131 kullanma \u00f6zg\u00fcrl\u00fc\u011f\u00fc k\u0131s\u0131tlama yoluna gidebiliyor. FSF, \u00e7o\u011fu zaman yasaklanan faaliyetlerin telif haklar\u0131n\u0131 korumak amac\u0131yla geli\u015ftirilen lisanslarla korunmaya \u00e7al\u0131\u015f\u0131lmas\u0131n\u0131 hem anlams\u0131z buluyor hem de kullan\u0131m k\u0131s\u0131tlay\u0131c\u0131 lisanslar\u0131n hukuku bireylere, kurulu\u015flara veya i\u015fletmelere devretmesinin demokrasinin tam tersi oldu\u011funu savunuyor. RAIL \u00f6rne\u011finde oldu\u011fu gibi ahlaki a\u00e7\u0131dan y\u00fckl\u00fc terimlerle (&#8220;zarar verme,&#8221; &#8220;ayr\u0131mc\u0131l\u0131k,&#8221; &#8220;aldatma ama\u00e7lar\u0131&#8221;) kullan\u0131c\u0131lara lisans k\u0131s\u0131tlamalar\u0131 getirmek, lisans verenlerin yaln\u0131zca demokratik olarak se\u00e7ilmi\u015f yasama organlar\u0131 arac\u0131l\u0131\u011f\u0131yla adil bir \u015fekilde kullan\u0131labilecek yasal yetkiyi kendilerine mal etmeleri anlam\u0131na geliyor. Dolay\u0131s\u0131yla RAIL gibi etik s\u0131n\u0131rlamalar getirme iddias\u0131nda olan lisanslar, etik zorluklar\u0131 ele almak iddias\u0131yla sunulmalar\u0131na ra\u011fmen kullan\u0131c\u0131lar\u0131n makine \u00f6\u011frenimiyle yap\u0131lan hesaplamalar\u0131 kontrol etmeleri i\u00e7in ger\u00e7ekten gerekli olan \u00f6zg\u00fcrl\u00fckleri de sa\u011flama konusunda yetersiz kal\u0131yorlar [10].<\/p>\n\n\n\n<h1 class=\"wp-block-heading\">Sonu\u00e7<\/h1>\n\n\n\n<p class=\"wp-block-paragraph\">A\u00e7\u0131k a\u011f\u0131rl\u0131klar, ne OSI&#8217;nin \u015feffafl\u0131k kriterlerini ne de FSF&#8217;nin \u00f6zg\u00fcrl\u00fck felsefesini tam anlam\u0131yla kar\u015f\u0131layan bir model olsa da, k\u00fcresel \u00f6l\u00e7ekte kabul g\u00f6rmeye devam ediyor. Hem ABD hem de \u00c7in merkezli stratejiler bu modeli desteklemeyi s\u00fcrd\u00fcr\u00fcyor.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Bu durum \u00f6zellikle \u00c7inli teknoloji akt\u00f6rleri i\u00e7in yeni olanaklar sunuyor. 16 Temmuz\u2019da \u00c7inli Moonshot AI, bir milyon belirte\u00e7lik ba\u011flam penceresine ve 2,8 trilyon parametreye sahip Kimi K3 modelini duyurdu. Moonshot, K3\u2019\u00fcn en g\u00fc\u00e7l\u00fc \u00f6zel m\u00fclkiyetli modellerin h\u00e2l\u00e2 bir ad\u0131m gerisinde kald\u0131\u011f\u0131n\u0131 kabul etmekle birlikte, kendi de\u011ferlendirme testlerinde modelin en ileri d\u00fczeyde (<em>frontier-level<\/em>) performans sergiledi\u011fini bildirdi. Model ilk olarak Kimi\u2019nin API\u2019si \u00fczerinden kullan\u0131ma sunuldu. 27 Temmuz\u2019da ise model a\u011f\u0131rl\u0131klar\u0131 Moonshot\u2019\u0131n resmi GitHub deposu \u00fczerinden eri\u015fime a\u00e7\u0131ld\u0131[11].<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u00c7inli \u015firketlerin geli\u015fmi\u015f donan\u0131mlara eri\u015fimini k\u0131s\u0131tlayan ABD ihracat kontrolleri, b\u00fcy\u00fck yapay zek\u00e2 modellerinin tek bir merkezde veya yerel altyap\u0131larda e\u011fitilip \u00e7al\u0131\u015ft\u0131r\u0131lmas\u0131n\u0131 zorla\u015ft\u0131r\u0131yor. Bu ko\u015fullar alt\u0131nda model a\u011f\u0131rl\u0131klar\u0131n\u0131n yay\u0131mlanmas\u0131, d\u00fcnya genelindeki bulut sa\u011flay\u0131c\u0131lar\u0131n\u0131n, bar\u0131nd\u0131rma platformlar\u0131n\u0131n ve i\u015fletmelerin modeli kendi donan\u0131mlar\u0131 \u00fczerinde \u00e7al\u0131\u015ft\u0131rmas\u0131na olanak tan\u0131yor. Moonshot bu sayede, modelin t\u00fcm isteklerini kendi veri merkezlerinden kar\u015f\u0131lamak zorunda kalmadan, \u00e7\u0131kar\u0131m (<em>inference<\/em>) y\u00fck\u00fcn\u00fc k\u00fcresel kullan\u0131c\u0131 taban\u0131na da\u011f\u0131t\u0131yor. B\u00f6ylelikle \u015firket, kendi sunucu ve hesaplama kapasitesini devasa yat\u0131r\u0131mlarla b\u00fcy\u00fctmek zorunda kalmadan modelin k\u00fcresel \u00f6l\u00e7ekteki kullan\u0131m kapasitesini ve pazar pay\u0131n\u0131 art\u0131rabiliyor. Daha fazla geli\u015ftirici ve kurum bu modeli kendi altyap\u0131s\u0131nda \u00e7al\u0131\u015ft\u0131rd\u0131k\u00e7a model standartla\u015fma yolunda ilerleyebilecek. Bu da Moonshot\u2019a geni\u015f bir kullan\u0131c\u0131 ekosistemi, marka g\u00f6r\u00fcn\u00fcrl\u00fc\u011f\u00fc ve gelecekte ticarile\u015febilecek yeni pazar olanaklar\u0131 yaratabilir. Ancak lisans s\u00f6zle\u015fmesinin ikinci maddesi a\u00e7\u0131k a\u011f\u0131rl\u0131klar\u0131n baz\u0131 k\u0131s\u0131tlamalarla beraber sunulabilece\u011fini g\u00f6steriyor[12]:<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Lisans Alan veya ba\u011fl\u0131 kurulu\u015flar\u0131ndan herhangi biri &#8220;Model as a Service&#8221; (Hizmet Olarak Model) i\u015fi y\u00fcr\u00fct\u00fcyorsa ve Lisans Alan ile ba\u011fl\u0131 kurulu\u015flar\u0131n\u0131n toplam geliri, birbirini takip eden herhangi bir 12 ayl\u0131k d\u00f6nemde toplamda 20 milyon ABD dolar\u0131n\u0131 (veya di\u011fer para birimlerindeki kar\u015f\u0131l\u0131\u011f\u0131n\u0131) a\u015f\u0131yorsa, Lisans Alan, Yaz\u0131l\u0131m\u0131 veya ondan t\u00fcretilen \u00e7al\u0131\u015fmalar\u0131 herhangi bir ticari ama\u00e7la kullanmadan \u00f6nce Moonshot AI ile ayr\u0131 bir s\u00f6zle\u015fme yapmal\u0131d\u0131r.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">ABD\u2019li \u015firketlerin yay\u0131mlad\u0131\u011f\u0131 bildiride[3] ise g\u00fc\u00e7l\u00fc bir yapay zek\u00e2 ekosistemi yaratman\u0131n \u00f6nemine dikkat \u00e7ekiliyor. Bildiri, a\u00e7\u0131k kaynak yaz\u0131l\u0131m\u0131n d\u00fcnyan\u0131n d\u00f6rt bir yan\u0131ndaki geli\u015ftiricilerin kodu inceleyebilece\u011fi, de\u011fi\u015ftirebilece\u011fi ve geli\u015ftirebilece\u011fi \u015feffaf bir ekosistem kurdu\u011fu vurgusuyla ba\u015fl\u0131yor. Devam\u0131nda, ABD\u2019nin her sekt\u00f6re yay\u0131lan g\u00fc\u00e7l\u00fc ve a\u00e7\u0131k bir ekosistem in\u015fa etmesinin sa\u011flayaca\u011f\u0131 stratejik avantajlar s\u0131ralan\u0131yor. Ancak b\u00f6yle bir yap\u0131n\u0131n kendili\u011finden olu\u015famayaca\u011f\u0131, politika yap\u0131c\u0131lar\u0131n (karar vericilerin) derhal harekete ge\u00e7mesi gerekti\u011fi \u00f6nemle belirtiliyor.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Son zamanlarda a\u00e7\u0131k a\u011f\u0131rl\u0131klar etraf\u0131nda d\u00f6nen tart\u0131\u015fmalar\u0131 ve a\u00e7\u0131k a\u011f\u0131rl\u0131klara dayal\u0131 stratejilerin sonu\u00e7lar\u0131n\u0131 de\u011ferlendirirken \u00fc\u00e7 konuyu g\u00f6z \u00f6n\u00fcnde bulundurman\u0131n \u00f6nemli oldu\u011funu d\u00fc\u015f\u00fcn\u00fcyorum.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Birincisi, a\u00e7\u0131k a\u011f\u0131rl\u0131klar (en az\u0131ndan ilk etapta) \u00e7ok say\u0131da kamu kurumu ve \u015firket taraf\u0131ndan yapay zek\u00e2 kullan\u0131m maliyetlerini d\u00fc\u015f\u00fcrmenin bir arac\u0131 olarak g\u00f6r\u00fclecek ve bu y\u00f6nde \u00e7e\u015fitli giri\u015fimler olacakt\u0131r. \u00c7\u00fcnk\u00fc API \u00fczerinden al\u0131nan yapay zek\u00e2 hizmetlerinin kullan\u0131m maliyetleri ciddi bir gider kalemi h\u00e2line gelmi\u015ftir. A\u00e7\u0131k a\u011f\u0131rl\u0131klar sayesinde \u015firketler, modeli kendi altyap\u0131lar\u0131nda \u00e7al\u0131\u015ft\u0131rarak bu maliyetleri azaltabilir. Ancak bu, sadece gerekli donan\u0131m ve kalifiye i\u015fg\u00fcc\u00fcne eri\u015fim olana\u011f\u0131 bulunanlar i\u00e7in ge\u00e7erli bir se\u00e7enek olacakt\u0131r.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Ancak a\u00e7\u0131k a\u011f\u0131rl\u0131klar, kullan\u0131c\u0131lara tam anlam\u0131yla bir \u00f6zg\u00fcrl\u00fck sa\u011flamasa da yeni sorumluluklar getirmektedir. API \u00fczerinden eri\u015fim sa\u011flayan bir model sa\u011flay\u0131c\u0131s\u0131, diledi\u011fi zaman g\u00fcvenlik \u00f6nlemlerini g\u00fcncelleyebilir, g\u00fcvenlik a\u00e7\u0131klar\u0131n\u0131 yamalayabilir veya bar\u0131nd\u0131rd\u0131\u011f\u0131 modele eri\u015fimi sonland\u0131rabilir. A\u011f\u0131rl\u0131klar dola\u015f\u0131ma girdikten sonra ise sa\u011flay\u0131c\u0131n\u0131n bu kopyalar \u00fczerindeki kontrol\u00fc b\u00fcy\u00fck \u00f6l\u00e7\u00fcde ortadan kalkar; kopyalar\u0131n sa\u011flay\u0131c\u0131n\u0131n kontrol\u00fc d\u0131\u015f\u0131nda \u00e7o\u011falt\u0131lmas\u0131 ve kullan\u0131lmaya devam etmesi m\u00fcmk\u00fcn h\u00e2le gelir. Kullan\u0131c\u0131lar da modelin g\u00fcvenlik \u00f6nlemlerini ince ayar yoluyla de\u011fi\u015ftirebilir veya tamamen kald\u0131rabilir. Dolay\u0131s\u0131yla bir modeli indiren kullan\u0131c\u0131, model sa\u011flay\u0131c\u0131s\u0131ndan ba\u011f\u0131ms\u0131zl\u0131k kazan\u0131rken; sa\u011flay\u0131c\u0131 da modelin bundan sonra nas\u0131l kullan\u0131laca\u011f\u0131n\u0131 kontrol etme yetene\u011finin \u00f6nemli bir b\u00f6l\u00fcm\u00fcn\u00fc kaybeder. Bu durum ayn\u0131 zamanda reg\u00fclasyonlar kar\u015f\u0131s\u0131nda sa\u011flay\u0131c\u0131n\u0131n sorumlulu\u011funun azalaca\u011f\u0131, a\u00e7\u0131k a\u011f\u0131rl\u0131klarla modeli \u00e7al\u0131\u015ft\u0131ran tarafa ise daha fazla sorumluluk y\u00fcklenece\u011fi anlam\u0131na geliyor.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u0130kincisi, tekelle\u015fmenin oda\u011f\u0131n\u0131n API \u00fczerinden hizmet veren teknoloji \u015firketlerinden ekosistem sa\u011flay\u0131c\u0131 \u015firketlere kayacak olmas\u0131. Cecilia Rikap\u2019\u0131n belirtti\u011fi gibi ekosistemler olu\u015fturmak, teknoloji end\u00fcstrisinin y\u0131llard\u0131r ba\u015far\u0131yla uygulad\u0131\u011f\u0131 bir modeldir[13]. \u0130stikrarl\u0131 bir \u00e7ekirdek (platform) ve hareketli bir \u00e7evre (uygulamalar) ile karakterize edilen bu modelde kazanan daima platformun ger\u00e7ek sahipleri olmaktad\u0131r. Olu\u015fturduklar\u0131 ekosistemler sayesinde B\u00fcy\u00fck Teknoloji \u015firketleri, di\u011fer kurulu\u015flar taraf\u0131ndan \u00fcretilen bilgiye do\u011frudan m\u00fclkiyetine sahip olmadan eri\u015febilmekte, onu y\u00f6nlendirebilmekte ve \u00fczerinden y\u00fcksek k\u00e2rlar elde edebilmektedir.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Google\u2019\u0131n Android ve Apple\u2019\u0131n iOS platformlar\u0131 bunun ba\u015fl\u0131ca \u00f6rnekleridir. Geli\u015ftiriciler, geni\u015f t\u00fcketici kitlelerine ula\u015fabilmek ad\u0131na bu platformlara entegre olmak zorunda kalan uygulamalar in\u015fa ederler. S\u00f6z konusu platformlarda yer alan uygulama geli\u015ftiricileri; \u00fcrettikleri gelirin \u00f6nemli bir k\u0131sm\u0131n\u0131 payla\u015fmak ve platform sahipleri taraf\u0131ndan dayat\u0131lan kurallara s\u0131k\u0131 s\u0131k\u0131ya uymak durumunda kal\u0131rlar.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">G\u00fcn\u00fcm\u00fcz\u00fcn dev teknoloji \u015firketleri aras\u0131nda Amazon, Microsoft ve Google; yapay zek\u00e2 ve bulut ekosistemleri in\u015fa etme konusunda a\u00e7\u0131k ara \u00f6ndedir. Bu \u015firketler yaln\u0131zca en b\u00fcy\u00fck ekosistemleri y\u00f6netmekle kalmamakta ayn\u0131 zamanda kendi denetimleri alt\u0131nda kalan ve aralar\u0131nda aral\u0131ks\u0131z veri\/bilgi ak\u0131\u015f\u0131n\u0131n ger\u00e7ekle\u015fti\u011fi, birbiriyle entegre pek \u00e7ok alt ekosistemi b\u00fcnyelerinde bar\u0131nd\u0131rmaktad\u0131r. Giderek daha fazla say\u0131da geli\u015ftiricinin bu \u015firketlerin sundu\u011fu ara\u00e7 setlerine al\u0131\u015fmas\u0131 ve bulut altyap\u0131lar\u0131na ba\u011f\u0131ml\u0131 h\u00e2le gelmesiyle birlikte a\u011f etkileri zirveye ula\u015fmakta; olu\u015fturulan ekosistemler \u015firketlerin pazar h\u00e2kimiyetini sars\u0131lmaz k\u0131lan birer bariyere d\u00f6n\u00fc\u015fmektedir. ABD\u2019li \u015firketlerin yay\u0131mlad\u0131\u011f\u0131 ortak bildiride vurgulanan &#8220;ekosistem yaratma&#8221; hedefi de tam olarak bu stratejik h\u00e2kimiyet aray\u0131\u015f\u0131n\u0131 d\u0131\u015fa vurmaktad\u0131r.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u00dc\u00e7\u00fcnc\u00fcs\u00fc, \u00f6zg\u00fcr\/a\u00e7\u0131k kaynak yaz\u0131l\u0131m\u0131n ba\u015far\u0131s\u0131n\u0131 ne tek ba\u015f\u0131na bireysel yaz\u0131l\u0131m geli\u015ftiricilerinin di\u011ferkaml\u0131\u011f\u0131yla ne de \u015firketlerin hay\u0131rseverli\u011fiyle a\u00e7\u0131klayabiliriz. Di\u011ferkam geli\u015ftiricilerin eme\u011fini kamusalla\u015ft\u0131ran ve \u015firketlerin \u00f6zg\u00fcr\/a\u00e7\u0131k kaynak projelere yakla\u015f\u0131m\u0131n\u0131 belirleyen as\u0131l unsur, ba\u015fta GNU GPL (GNU General Public Licenses) olmak \u00fczere telif hukukuna dayanan g\u00fc\u00e7l\u00fc lisanslama mekanizmalar\u0131d\u0131r. \u015eirketler ise do\u011falar\u0131 gere\u011fi para kazanmak i\u00e7in kurulmu\u015flard\u0131r; hay\u0131r i\u015fleri yapmak i\u00e7in de\u011fil.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">\u00d6rne\u011fin GNU GPL, \u015firketlerin \u00f6zg\u00fcr yaz\u0131l\u0131mlardan yararlanmas\u0131n\u0131 engellemez; ancak \u015firketlerin bu yaz\u0131l\u0131mlar\u0131 \u00f6zel m\u00fclkiyetlerine ge\u00e7irmesini imk\u00e2ns\u0131z k\u0131lar. GNU GPL, copyleft mekanizmas\u0131 sayesinde telif hakk\u0131n\u0131 (copyright) yok saymam\u0131\u015f, aksine telif hukukunun g\u00fcc\u00fcn\u00fc \u00f6zg\u00fcrl\u00fc\u011f\u00fc garanti alt\u0131na almak i\u00e7in kullanm\u0131\u015ft\u0131r. Nitekim bir yaz\u0131l\u0131m yaz\u0131ld\u0131\u011f\u0131 andan itibaren telif hakk\u0131 yasalar\u0131yla koruma alt\u0131na girer; hak sahibi bir ihlal durumunda do\u011frudan yarg\u0131ya ba\u015fvurabilir. Dolay\u0131s\u0131yla patentte oldu\u011fu gibi resmi bir kurumun tesciline gerek yoktur. Yaz\u0131l\u0131m\u0131n\u0131 GPL ile lisanslayan bir geli\u015ftirici, telif yasalar\u0131n\u0131n korumas\u0131 alt\u0131ndad\u0131r; fakat klasik copyright anlay\u0131\u015f\u0131n\u0131n aksine bu korumay\u0131 copyleft prensibiyle i\u015fletir. Yaz\u0131l\u0131m geli\u015ftirici k\u0131saca \u015funu s\u00f6yler: &#8220;Yaz\u0131l\u0131m\u0131m\u0131 diledi\u011finiz gibi kullan\u0131n, de\u011fi\u015ftirin, payla\u015f\u0131n. Ancak payla\u015f\u0131rken benim size tan\u0131d\u0131\u011f\u0131m bu \u00f6zg\u00fcrl\u00fckleri aynen koruyun ve sonraki kullan\u0131c\u0131lar\u0131n haklar\u0131n\u0131 k\u0131s\u0131tlamay\u0131n. Aksi takdirde, bu \u015fartlara uymad\u0131\u011f\u0131n\u0131z an varsay\u0131lan telif hakk\u0131 yasalar\u0131 devreye girer ve kullan\u0131m hakk\u0131n\u0131z\u0131 kaybedersiniz.&#8221;<\/p>\n\n\n\n<p class=\"wp-block-paragraph\">Bunun sonucunda \u00f6zg\u00fcrl\u00fck, \u015firketlerin l\u00fctfetti\u011fi bir ayr\u0131cal\u0131k olmaktan \u00e7\u0131kar. Yaz\u0131l\u0131m \u00fczerinde yap\u0131lan de\u011fi\u015fiklikler kullan\u0131c\u0131lara da\u011f\u0131t\u0131ld\u0131\u011f\u0131nda yeniden GPL ile sunulmak zorunda oldu\u011fundan, \u00f6zg\u00fcrl\u00fc\u011f\u00fcn s\u00fcreklili\u011fi bir tercih ya da hay\u0131rseverlik de\u011fil, hukuki bir zorunluluk h\u00e2line gelir. A\u00e7\u0131k kaynak yaz\u0131l\u0131m anlat\u0131lar\u0131nda s\u0131kl\u0131kla Linux \u00e7ekirde\u011finin geli\u015ftirilmesindeki i\u015f birli\u011finden ve Linus Torvalds\u2019\u0131n ba\u015far\u0131l\u0131 y\u00f6netiminden, a\u00e7\u0131k kaynak yaz\u0131l\u0131mlardaki kodun binlerce g\u00f6z taraf\u0131ndan denetlenip \u015feffafla\u015fmas\u0131ndan ve hatalar\u0131n h\u0131zl\u0131ca giderilmesinden \u00f6vg\u00fcyle bahsedilir. Oysa hakiki \u00f6zg\u00fcrl\u00fck, s\u00fcreklilik ve g\u00fcvence ister. Bir yaz\u0131l\u0131m\u0131n a\u00e7\u0131k kalmas\u0131 veya aniden kapat\u0131lmas\u0131 tek bir \u015firketin inisiyatifine ba\u011fl\u0131ysa, orada ger\u00e7ek anlamda bir \u00f6zg\u00fcrl\u00fckten s\u00f6z edilemez.<\/p>\n\n\n\n<p class=\"wp-block-paragraph\"><a><\/a>Bug\u00fcn teknoloji devlerini ne tam anlam\u0131yla a\u00e7\u0131k kaynak yapay zek\u00e2 sunmaya ne de a\u00e7\u0131k a\u011f\u0131rl\u0131klar\u0131 yay\u0131mlamaya zorlayan hukuki bir dayanak vard\u0131r. Ne de onlar\u0131 kullan\u0131c\u0131 haklar\u0131na ve \u00f6zg\u00fcrl\u00fcklerine sayg\u0131 duymaya mecbur k\u0131lan bir yapt\u0131r\u0131m&#8230;<\/p>\n\n\n\n<h1 class=\"wp-block-heading\">Notlar<\/h1>\n\n\n\n<ol class=\"wp-block-list\">\n<li><a href=\"https:\/\/www.nytimes.com\/2026\/07\/28\/technology\/mark-zuckerberg-meta-ai.html\">https:\/\/www.nytimes.com\/2026\/07\/28\/technology\/mark-zuckerberg-meta-ai.html<\/a><\/li>\n<\/ol>\n\n\n\n<ol class=\"wp-block-list\">\n<li><a href=\"https:\/\/www.nytimes.com\/2026\/08\/10\/technology\/meta-ai-open-source.html\">https:\/\/www.nytimes.com\/2026\/08\/10\/technology\/meta-ai-open-source.html<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/images.nvidia.com\/pdf\/Open-Weights-and-American-AI-Leadership.pdf\">https:\/\/images.nvidia.com\/pdf\/Open-Weights-and-American-AI-Leadership.pdf<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/www.fsf.org\/news\/fsf-is-working-on-freedom-in-machine-learning-applications\">https:\/\/www.fsf.org\/news\/fsf-is-working-on-freedom-in-machine-learning-applications<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/opensource.org\/ai\/open-source-ai-definition\">https:\/\/opensource.org\/ai\/open-source-ai-definition<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/opensource.org\/ai\/open-weights\">https:\/\/opensource.org\/ai\/open-weights<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/archive.fosdem.org\/2025\/schedule\/event\/fosdem-2025-4818-fsf-s-criteria-for-free-machine-learning-applications\/\">https:\/\/archive.fosdem.org\/2025\/schedule\/event\/fosdem-2025-4818-fsf-s-criteria-for-free-machine-learning-applications\/<\/a><\/li>\n\n\n\n<li>Yaz\u0131 boyunca s\u00fcrekli kullan\u0131c\u0131lardan bahsediyorum. Fakat burada kastedilen kullan\u0131c\u0131lar; masa\u00fcst\u00fc ya da ak\u0131ll\u0131 telefon kullan\u0131c\u0131lar\u0131ndan ziyade, s\u00f6z konusu sistemleri yeniden \u00fcretebilecek teknik yeteneklere ve maddi olanaklara sahip kamu kurumlar\u0131 ile \u015firketlerdir.<\/li>\n\n\n\n<li>FSF\u2019nin b\u00fcy\u00fck dil modelleri \u00f6zelinde bir tart\u0131\u015fma y\u00fcr\u00fctmedi\u011fini, do\u011frudan makine \u00f6\u011frenimine odakland\u0131\u011f\u0131n\u0131 hat\u0131rlatmak isterim.<\/li>\n\n\n\n<li><a href=\"https:\/\/www.fsf.org\/blogs\/licensing\/rail-are-nonfree-and-unethical\">https:\/\/www.fsf.org\/blogs\/licensing\/rail-are-nonfree-and-unethical<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/www.lawfaremedia.org\/article\/open-weight-diplomacy\u2013how-china-s-ai-models-are-rerunning-the-digital-silk-road\">https:\/\/www.lawfaremedia.org\/article\/open-weight-diplomacy\u2013how-china-s-ai-models-are-rerunning-the-digital-silk-road<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/huggingface.co\/moonshotai\/Kimi-K3\/blob\/main\/LICENSE\">https:\/\/huggingface.co\/moonshotai\/Kimi-K3\/blob\/main\/LICENSE<\/a><\/li>\n\n\n\n<li>Rikap, C. (2026). The Rulers: Corporate Power in the Age of AI and the Cloud. Verso Books.<\/li>\n<\/ol>\n","protected":false},"excerpt":{"rendered":"<p>A\u00e7\u0131k a\u011f\u0131rl\u0131kl\u0131 modeller, yapay zek\u00e2y\u0131 \u00f6zg\u00fcrle\u015ftirmekten ziyade sorumlulu\u011fun ve m\u00fclkiyetin bi\u00e7im de\u011fi\u015ftirdi\u011fi yeni bir d\u00f6neme i\u015faret ediyor. A\u00e7\u0131k a\u011f\u0131rl\u0131klar, hem \u015firketlerin keyfi kararlar\u0131na ba\u011fl\u0131 hem de OSI ve FSF standartlar\u0131 a\u00e7\u0131s\u0131ndan hakiki bir \u00f6zg\u00fcrl\u00fckten uzak. \u015eirketler i\u00e7in a\u011f\u0131rl\u0131klar\u0131 yay\u0131mlamak; altyap\u0131 maliyetlerini d\u0131\u015fsalla\u015ft\u0131rman\u0131n, reg\u00fclatif riskleri devretmenin ve ekosistem tekelle\u015fmesini peki\u015ftirmenin bir yolu. Daha da \u00f6nemlisi, tamamen \u015firketlerin inisiyatifinde y\u00fcr\u00fcyen bu strateji, \u015firketlerin yeni a\u011f\u0131rl\u0131klar\u0131 yay\u0131mlamaktan veya a\u00e7\u0131k a\u011f\u0131rl\u0131k politikas\u0131n\u0131 s\u00fcrd\u00fcrmekten vazge\u00e7mesiyle yar\u0131n sona erebilir.<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[3,116,298],"tags":[583,179,584,585,586,557],"class_list":["post-1078","post","type-post","status-publish","format-standard","hentry","category-ozgur_yazilim","category-teknoloji-tarihi","category-yapay-zeka","tag-acik-agirliklar","tag-acik-kaynak","tag-acik-kaynak-yapay-zeka","tag-fsf","tag-osi","tag-regulasyon"],"views":58,"_links":{"self":[{"href":"https:\/\/yarimada.gen.tr\/index.php?rest_route=\/wp\/v2\/posts\/1078","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/yarimada.gen.tr\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/yarimada.gen.tr\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/yarimada.gen.tr\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/yarimada.gen.tr\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=1078"}],"version-history":[{"count":2,"href":"https:\/\/yarimada.gen.tr\/index.php?rest_route=\/wp\/v2\/posts\/1078\/revisions"}],"predecessor-version":[{"id":1081,"href":"https:\/\/yarimada.gen.tr\/index.php?rest_route=\/wp\/v2\/posts\/1078\/revisions\/1081"}],"wp:attachment":[{"href":"https:\/\/yarimada.gen.tr\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=1078"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/yarimada.gen.tr\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=1078"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/yarimada.gen.tr\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=1078"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}