{"id":1599,"date":"2026-02-05T14:05:34","date_gmt":"2026-02-05T14:05:34","guid":{"rendered":"https:\/\/brain.hr\/?p=1599"},"modified":"2026-02-06T09:19:16","modified_gmt":"2026-02-06T09:19:16","slug":"kako-ai-uci-razmisljati-6-6","status":"publish","type":"post","link":"https:\/\/brain.hr\/en\/kako-ai-uci-razmisljati-6-6\/","title":{"rendered":"How AI \"learns\" to think? (6\/6)"},"content":{"rendered":"<p>So far, we have learned how AI models are built from massive amounts of data and how they learn the rules of conversation by mimicking humans. Newer models, such as the OpenAI o1 or DeepSeek R1 series, represent a major shift: they no longer just mimic human responses but have begun to solve logical problems independently. Andrej Karpathy explains this leap through the third and most advanced stage of development: Reinforcement Learning (RL). <em>Reinforcement Learning<\/em> \u2013 RL).<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">The Three Stages of AI Schooling<\/h3>\n\n\n\n<p>Karpathy uses a schooling analogy to explain this technological progress:<\/p>\n\n\n\n<ul class=\"wp-block-list\">\n<li>\u010citanje ud\u017ebenika (prethodno treniranje): Model upija golemo znanje s interneta, ali ga jo\u0161 ne zna koristiti u razgovoru.<\/li>\n\n\n\n<li>Gledanje rje\u0161enja (SFT): Model promatra kako ljudi rje\u0161avaju zadatke i poku\u0161ava ih kopirati. Tako nastaju standardni asistenti koji ne mogu biti pametniji od \u010dovjeka kojeg imitiraju.<\/li>\n\n\n\n<li>Zadatci za vje\u017ebu (RL): Ovo je faza u kojoj model samostalno rje\u0161ava probleme. Ba\u0161 kao u\u010denik koji vje\u017eba zadatke na kraju poglavlja, AI poku\u0161ava tisu\u0107e razli\u010ditih pristupa dok sam ne do\u0111e do to\u010dnog rezultata.<\/li>\n<\/ul>\n\n\n\n<h3 class=\"wp-block-heading\">Internal Monologue: The AI that \"Talks to Itself\"<\/h3>\n\n\n\n<p>The greatest innovation of this phase is the development of an internal monologue <em>(Chain of Thought)<\/em>. The model discovered on its own that \"thinking out loud\" increases its accuracy. Before providing a final answer, it checks its own steps in the background: <em>\"Wait, this looks wrong\u2026 let\u2019s try another method\u2026 let\u2019s check one more time\u2026 Ah, now it makes sense.\"<\/em><\/p>\n\n\n\n<p>This behavior wasn't programmed; rather, the system learned that breaking a problem down into smaller steps and acknowledging its own mistakes leads to success.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Transcending Human Limitations<\/h3>\n\n\n\n<p>While older models relied exclusively on imitation, new models can surpass human intuition. Karpathy compares this to the historic moment when AlphaGo made a brilliant move in the game of Go\u2014a move no human would have ever played. In education, this signifies a transition from tools that merely assemble sentences to tools capable of solving the most difficult logical tasks.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">The Future: From \"Chat\" to Digital Agents<\/h3>\n\n\n\n<p>What does tomorrow hold? Karpathy predicts two major shifts:<\/p>\n\n\n\n<ol start=\"1\" class=\"wp-block-list\">\n<li>AI dobiva o\u010di i u\u0161i (multimodalnost): Modeli \u0107e nativno vidjeti i \u010duti svijet oko sebe, od analize tona glasa predava\u010da do uo\u010davanja pogre\u0161aka u video snimkama eksperimenata.<\/li>\n\n\n\n<li>Digitalni agenti: AI vi\u0161e ne\u0107e biti pasivan sugovornik, nego agent koji samostalno izvr\u0161ava dugotrajne zadatke (npr. planiranje cijelog izleta, komunikacija s prijevoznicima i unos u kalendar).<\/li>\n<\/ol>\n\n\n\n<h3 class=\"wp-block-heading\">Limitations of AI Systems<\/h3>\n\n\n\n<p>Despite their power, these systems are still full of unpredictable gaps. An AI might solve a Mathematical Olympiad problem and then fail a trivial question like: \"Is 9.11 greater than 9.9?\"<\/p>\n\n\n\n<p>Savjet za kraj: Koristite umjetnu inteligenciju kao mo\u0107an motor, ali volan uvijek dr\u017eite u svojim rukama. Neka vam AI slu\u017ei za inspiraciju i nacrte, ali vi ostanite &#8220;glavni urednik&#8221; koji donosi kona\u010dnu odluku.<\/p>\n\n\n\n<h3 class=\"wp-block-heading\">Read the previous articles in the series:<\/h3>\n\n\n\n<ol start=\"1\" class=\"wp-block-list\">\n<li><a href=\"https:\/\/brain.hr\/en\/kako-nastaje-baza-znanja-kojom-se-sluzi-chatgpt\/\" target=\"_blank\" rel=\"noreferrer noopener\">How is the knowledge base used by ChatGPT created?<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/brain.hr\/en\/zasto-najpametniji-modeli-grijese-na-najjednostavnijim-zadacima\/\" target=\"_blank\" rel=\"noreferrer noopener\">Why do the \u201csmartest\u201d models make mistakes on the simplest tasks?<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/brain.hr\/en\/moze-li-ai-znati-sve-a-ne-znati-razgovarati-3-6\/\" target=\"_blank\" rel=\"noreferrer noopener\">Can AI know everything and not be able to talk?<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/brain.hr\/en\/skola-lijepog-ponasanja-kako-ai-uci-postati-koristan-asistent-4-6\/\">Etiquette School: How AI Learns to Become a Useful Assistant?<\/a><\/li>\n\n\n\n<li><a href=\"https:\/\/brain.hr\/en\/fenomen-ai-halucinacija-5-6\/\">The Phenomenon of AI Hallucinations<\/a><\/li>\n<\/ol>\n\n\n\n<p><strong>Source:<\/strong>&nbsp;An analysis of Andrej Karpathy\u2019s technical lecture:&nbsp;<em><a href=\"https:\/\/www.youtube.com\/watch?v=7xTGNNLPyMI\">Deep Dive into LLMs like ChatGPT<\/a><\/em>.<\/p>\n\n\n\n<p><\/p>","protected":false},"excerpt":{"rendered":"<p>Do sada smo nau\u010dili kako AI modeli nastaju iz golemih koli\u010dina podataka te kako u\u010de pravila razgovora opona\u0161aju\u0107i ljude. Noviji modeli, poput serija OpenAI o1 ili DeepSeek R1, donose veliku promjenu jer vi\u0161e ne opona\u0161aju samo ljudske odgovore nego po\u010dinju samostalno rje\u0161avati logi\u010dke probleme. Andrej Karpathy obja\u0161njava ovaj skok kroz tre\u0107u i najnapredniju fazu razvoja: [&hellip;]<\/p>\n","protected":false},"author":8,"featured_media":1601,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_et_pb_use_builder":"off","_et_pb_old_content":"","_et_gb_content_width":"","footnotes":""},"categories":[14],"tags":[],"class_list":["post-1599","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-radovi"],"acf":{"radovi_source_url":"","radovi_button_label":"Pro\u010ditajte izvorni rad"},"_links":{"self":[{"href":"https:\/\/brain.hr\/en\/wp-json\/wp\/v2\/posts\/1599","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/brain.hr\/en\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/brain.hr\/en\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/brain.hr\/en\/wp-json\/wp\/v2\/users\/8"}],"replies":[{"embeddable":true,"href":"https:\/\/brain.hr\/en\/wp-json\/wp\/v2\/comments?post=1599"}],"version-history":[{"count":2,"href":"https:\/\/brain.hr\/en\/wp-json\/wp\/v2\/posts\/1599\/revisions"}],"predecessor-version":[{"id":1606,"href":"https:\/\/brain.hr\/en\/wp-json\/wp\/v2\/posts\/1599\/revisions\/1606"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/brain.hr\/en\/wp-json\/wp\/v2\/media\/1601"}],"wp:attachment":[{"href":"https:\/\/brain.hr\/en\/wp-json\/wp\/v2\/media?parent=1599"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/brain.hr\/en\/wp-json\/wp\/v2\/categories?post=1599"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/brain.hr\/en\/wp-json\/wp\/v2\/tags?post=1599"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}