{"id":1122,"date":"2025-04-10T14:06:39","date_gmt":"2025-04-10T12:06:39","guid":{"rendered":"https:\/\/octodeep.com\/blog\/?p=1122"},"modified":"2025-05-22T14:16:21","modified_gmt":"2025-05-22T12:16:21","slug":"gpt-5-na-obzoru-nova-era-umele-inteligence","status":"publish","type":"post","link":"https:\/\/octodeep.com\/blog\/gpt-5-na-obzoru-nova-era-umele-inteligence\/","title":{"rendered":"GPT-5 na obzoru: Nov\u00e1 \u00e9ra nebo jen dal\u0161\u00ed update?\u00a0"},"content":{"rendered":"\n

Je\u0161t\u011b p\u0159ed dv\u011bma lety by se konverzace s um\u011blou inteligenc\u00ed dala popsat jako fascinuj\u00edc\u00ed experiment. Dnes u\u017e je pro miliony lid\u00ed b\u011b\u017en\u00fdm n\u00e1strojem v pr\u00e1ci, ve \u0161kole i p\u0159i osobn\u00edch projektech. A zat\u00edmco sou\u010dasn\u00fd, ve\u0159ejn\u011b dostupn\u00fd model GPT-4o je st\u00e1le na prvn\u00edch p\u0159\u00ed\u010dk\u00e1ch, co se t\u00fd\u010de v\u00fdkonnosti, pozornost se pomalu p\u0159esouv\u00e1 k jeho n\u00e1stupci GPT-5<\/a>.<\/strong><\/p>\n\n\n\n

OpenAI<\/a> zat\u00edm \u017e\u00e1dn\u00e9 ofici\u00e1ln\u00ed datum vyd\u00e1n\u00ed ChatGPT-5<\/a> neozn\u00e1mila<\/strong>, av\u0161ak spole\u010dnost  potvrdila, \u017ee na dal\u0161\u00ed generaci modelu<\/a> u\u017e pracuje, sb\u00edr\u00e1 tr\u00e9ninkov\u00e1 data a vyv\u00edj\u00ed pokro\u010dilej\u0161\u00ed architekturu.<\/p>\n\n\n\n

M\u00edsto p\u011btky byl v roce 2024 p\u0159edstaven GPT-4<\/a> Turbo, <\/strong>co\u017e je rychlej\u0161\u00ed, levn\u011bj\u0161\u00ed a schopn\u011bj\u0161\u00ed verze modelu GPT-4<\/strong><\/a>.<\/strong> I p\u0159esto se o\u010dek\u00e1v\u00e1, \u017ee GPT-5 <\/strong>p\u0159inese z\u00e1sadn\u00ed posun nejen ve schopnostech porozum\u011bn\u00ed lidsk\u00e9 \u0159e\u010di, ale i v kontextu pam\u011bti, u\u010den\u00ed a samostatn\u00e9ho rozhodov\u00e1n\u00ed. <\/p>\n\n\n\n

Na release si budeme muset je\u0161t\u011b po\u010dkat, ale pokud bude GPT-5 stejn\u011b ambici\u00f3zn\u00ed, jako jeho p\u0159edch\u016fdci, m\u016f\u017eeme o\u010dek\u00e1vat dal\u0161\u00ed velk\u00fd pr\u016flom ve sv\u011bt\u011b um\u011bl\u00e9 inteligence<\/a>. Co o modelu v\u00edme a co jen tu\u0161\u00edme? <\/p>\n\n\n\n

GPT-5 v n\u00e1znac\u00edch: Co prozradil Sam Altman<\/h2>\n\n\n\n

Zat\u00edmco OpenAI ofici\u00e1ln\u011b ml\u010d\u00ed o detailech, CEO spole\u010dnosti Sam Altman<\/strong> u\u017e n\u011bkolikr\u00e1t nazna\u010dil na sv\u00e9m \u00fa\u010dtu na platform\u011b X, \u017ee se dal\u0161\u00ed verze modelu aktivn\u011b vyv\u00edj\u00ed. Podle jeho vyj\u00e1d\u0159en\u00ed bude GPT-5 znamenat sjednocen\u00ed schopnost\u00ed jazykov\u00e9ho modelu (LLM<\/a>) s tzv. \u201e<\/strong>reasoning\u201d<\/em><\/strong>, co\u017e je pokro\u010dil\u00e9 logick\u00e9 a analytick\u00e9 my\u0161len\u00ed, kter\u00e9 se objevilo u model\u016f ze s\u00e9rie GPT-4o (Omni). <\/p>\n\n\n\n

12. \u00fanora Sam Altman na sv\u00e9m \u00fa\u010dtu na platform\u011b X napsal:<\/p>\n\n\n\n

 \u201eAKTUALIZACE PL\u00c1NU OPENAI PRO GPT-4.5 A GPT-5:<\/em><\/p>\n\n\n\n

Chceme odv\u00e1d\u011bt lep\u0161\u00ed pr\u00e1ci v\u202ftom, jak sd\u00edl\u00edme na\u0161i pl\u00e1novanou roadmapu, a mnohem lep\u0161\u00ed pr\u00e1ci v\u202ftom, jak zjednodu\u0161ujeme na\u0161i nab\u00eddku produkt\u016f.<\/em><\/p>\n\n\n\n

Chceme, aby um\u011bl\u00e1 inteligence<\/a> pro v\u00e1s prost\u011b fungovala; uv\u011bdomujeme si, jak slo\u017eit\u00e1 se na\u0161e nab\u00eddka model\u016f a produkt\u016f stala.<\/em><\/p>\n\n\n\n

Nesn\u00e1\u0161\u00edme  ru\u010dn\u00ed v\u00fdb\u011br modelu stejn\u011b jako vy a chceme se vr\u00e1tit ke kouzlu jednotn\u00e9 inteligence.<\/em><\/p>\n\n\n\n

Jako dal\u0161\u00ed uvedeme GPT-4.5, model<\/strong>, kter\u00fd jsme intern\u011b naz\u00fdvali Orion, a kter\u00fd bude na\u0161\u00edm posledn\u00edm modelem bez \u201echain-of-thought\u201c.<\/em><\/p>\n\n\n\n

Pot\u00e9 bude na\u0161\u00edm hlavn\u00edm c\u00edlem sjednotit modely<\/a> \u201eo-series\u201d \u0159ady GPT t\u00edm, \u017ee vytvo\u0159\u00edme syst\u00e9my, kter\u00e9 budou um\u011bt vyu\u017e\u00edt v\u0161echny na\u0161e n\u00e1stroje<\/a>, v\u011bd\u011bt, kdy maj\u00ed p\u0159em\u00fd\u0161let d\u00e9le \u010di ne, a obecn\u011b budou u\u017eite\u010dn\u00e9 pro velmi \u0161irokou \u0161k\u00e1lu \u00fakol\u016f.<\/em><\/p>\n\n\n\n

V r\u00e1mci ChatGPT<\/a> i na\u0161eho API uvedeme GPT-5 jako syst\u00e9m, kter\u00fd integruje velkou \u010d\u00e1st na\u0161\u00ed technologie<\/a> \u2013 v\u010detn\u011b modelu o3.<\/em> <\/em><\/strong>Model o3 ji\u017e nebudeme nad\u00e1le uv\u00e1d\u011bt jako samostatn\u00fd.<\/em><\/p>\n\n\n\n

Bezplatn\u00e1 verze ChatGPT<\/a> z\u00edsk\u00e1 neomezen\u00fd p\u0159\u00edstup k GPT-5 v b\u011b\u017en\u00e9m re\u017eimu inteligence (!!)<\/em><\/strong>, s omezen\u00edmi proti zneu\u017eit\u00ed.<\/em><\/p>\n\n\n\n

U\u017eivatel\u00e9 verze Plus budou moci spou\u0161t\u011bt GPT-5 na vy\u0161\u0161\u00ed \u00farovni inteligence<\/em><\/strong>, a u\u017eivatel\u00e9 verze Pro na je\u0161t\u011b vy\u0161\u0161\u00ed. <\/em>Tyto modely budou zahrnovat hlasov\u00e9 rozhran\u00ed, canvas (vizu\u00e1ln\u00ed pl\u00e1tno), vyhled\u00e1v\u00e1n\u00ed, pokro\u010dil\u00fd v\u00fdzkum a dal\u0161\u00ed funkce.<\/em><\/strong>\u201c<\/em><\/p>\n\n\n\n

Nejnov\u011bj\u0161\u00ed update, kter\u00fd Altman na sv\u00e9 soci\u00e1ln\u00ed s\u00edti zve\u0159ejnil je ze \u010dtvrt\u00e9ho dubna, kde \u0159\u00edk\u00e1: <\/p>\n\n\n\n

\u201eZm\u011bna pl\u00e1nu: <\/em>modely o3 a o4-mini p\u0159ece jen vyd\u00e1me<\/em><\/strong> \u2013 pravd\u011bpodobn\u011b b\u011bhem n\u011bkolika t\u00fddn\u016f \u2013 a pot\u00e9, o n\u011bkolik m\u011bs\u00edc\u016f pozd\u011bji, uvedeme GPT-5.<\/em><\/p>\n\n\n\n

D\u016fvod\u016f je cel\u00e1 \u0159ada, ale ten nejzaj\u00edmav\u011bj\u0161\u00ed je, \u017ee GPT-5 d\u00edky tomu budeme schopni ud\u011blat <\/em>mnohem lep\u0161\u00ed<\/em><\/strong>, ne\u017e jsme p\u016fvodn\u011b o\u010dek\u00e1vali\u201d<\/em><\/p>\n\n\n\n

\"Screenshot
Zdroj: @s<\/a>ama<\/a><\/figcaption><\/figure>\n\n\n\n

Tato jeho ozn\u00e1men\u00ed potvrzuj\u00ed, \u017ee si na GPT-5 budeme muset je\u0161t\u011b n\u011bjakou chv\u00edli po\u010dkat,<\/strong> nicm\u00e9n\u011b n\u00e1m nazna\u010duj\u00ed, \u017ee v\u00fdsledek by mohl st\u00e1t opravdu za to. OpenAi si d\u00e1v\u00e1 z\u00e1le\u017eet nejen na samotn\u00e9 technologii<\/a>, ale i na tom, aby nov\u00e1 generace um\u011bl\u00e9 inteligence byla stabiln\u00ed, \u0161k\u00e1lovateln\u00e1 a skute\u010dn\u011b p\u0159ipraven\u00e1 na vysok\u00e9 nasazen\u00ed v praxi. <\/strong><\/p>\n\n\n\n

Co nov\u00e9ho m\u016f\u017ee GPT-5 p\u0159in\u00e9st do sv\u011bta um\u011bl\u00e9 inteligence<\/h2>\n\n\n\n

Jak u\u017e to b\u00fdv\u00e1, tak s ka\u017edou novou generac\u00ed ChatGPT p\u0159ich\u00e1z\u00ed vylep\u0161en\u00e9 a nov\u00e9 funkce. P\u0159esto\u017ee zat\u00edm nev\u00edme, co p\u0159esn\u011b se chyst\u00e1, m\u016f\u017eeme z p\u0159\u00edsp\u011bvk\u016f Sama Altmana vytu\u0161it, \u017ee p\u016fjde o velk\u00fd krok vp\u0159ed v AI sv\u011bt\u011b. <\/p>\n\n\n\n