{"id":155336,"date":"2026-07-22T16:37:11","date_gmt":"2026-07-22T16:37:11","guid":{"rendered":"https:\/\/www.europesays.com\/cz\/155336\/"},"modified":"2026-07-22T16:37:11","modified_gmt":"2026-07-22T16:37:11","slug":"ai-utekla-z-laboratore-aby-mohla-krast-chova-se-jako-zaci-ve-skole","status":"publish","type":"post","link":"https:\/\/www.europesays.com\/cz\/155336\/","title":{"rendered":"AI utekla z\u00a0laborato\u0159e, aby mohla kr\u00e1st. Chov\u00e1 se jako \u017e\u00e1ci ve \u0161kole"},"content":{"rendered":"

\u010ctete uk\u00e1zku z\u00a0newsletteru TechMIX, ve kter\u00e9m Pavel Kas\u00edk a\u00a0Matou\u0161 L\u00e1z\u0148ovsk\u00fd ka\u017edou st\u0159edu p\u0159in\u00e1\u0161ej\u00ed hned n\u011bkolik koment\u00e1\u0159\u016f a\u00a0post\u0159eh\u016f ze sv\u011bta v\u011bdy a\u00a0nov\u00fdch technologi\u00ed. Pokud v\u00e1s TechMIX zaujme, p\u0159ihlaste se k\u00a0jeho odb\u011bru<\/a>!<\/p>\n

Je to snad nejot\u0159epan\u011bj\u0161\u00ed z\u00e1pletka z\u00a0jak\u00e9hokoli filmu \u010di seri\u00e1lu z\u00a0prost\u0159ed\u00ed st\u0159edn\u00ed \u0161koly. Studenta \u010dek\u00e1 t\u011b\u017ek\u00fd test, ale nechce se na n\u011bj u\u010dit. Spr\u00e1vn\u00e9 odpov\u011bdi tak ukradne z\u00a0kabinetu u\u010ditele nebo koup\u00ed od spolu\u017e\u00e1ka. Neobejde se to bez pat\u0159i\u010dn\u00fdch peripeti\u00ed a\u00a0na konci se m\u016f\u017eeme t\u011b\u0161it na p\u0159edv\u00eddateln\u00e9 mor\u00e1ln\u00ed ponau\u010den\u00ed: \u010cestnost a\u00a0p\u00edle jsou d\u016fle\u017eit\u011bj\u0161\u00ed ne\u017e dobr\u00e1 zn\u00e1mka z\u00a0p\u00edsemky.<\/p>\n

Nyn\u00ed jsme byli sv\u011bdky podobn\u00e9ho dramatu v\u00a0nejvy\u0161\u0161\u00ed lize firem, kter\u00e9 se zab\u00fdvaj\u00ed v\u00fdvojem um\u011bl\u00e9 inteligence. Spole\u010dnost Hugging Face\u00a0\u2013 provozuj\u00edc\u00ed asi nejd\u016fle\u017eit\u011bj\u0161\u00ed \u00falo\u017ei\u0161t\u011b AI model\u016f\u00a0\u2013 minul\u00fd t\u00fdden ozn\u00e1mila<\/a>, \u017ee se n\u011bkdo vloupal do jej\u00ed infrastruktury. \u201eTento \u00fatok byl jin\u00fd ne\u017e cokoli, s\u00a0\u010d\u00edm jsme se p\u0159edt\u00edm setkali. Od za\u010d\u00e1tku do konce toti\u017e \u00fatok \u0159\u00eddily autonomn\u00ed syst\u00e9my agent\u016f s\u00a0um\u011blou inteligenc\u00ed.\u201c Tento agent se pokusil odcizit data u\u017eivatel\u016f, kte\u0159\u00ed vyu\u017e\u00edvaj\u00ed platformu Hugging Face Spaces.<\/p>\n

\u201eU\u017e v\u00a0pr\u016fb\u011bhu \u00fatoku jsme podez\u00edrali, \u017ee by za t\u00edmto kyber\u00fatokem mohla st\u00e1t n\u011bkter\u00e1 z\u00a0p\u0159edn\u00edch AI laborato\u0159\u00ed,\u201c uvedl na sv\u00e9m profilu Clement Delangue, spoluzakladatel a\u00a0\u0161\u00e9f Hugging Face. K\u00a0\u00fatoku se toti\u017e \u201ep\u0159ihl\u00e1sila\u201c laborato\u0159 firmy OpenAI, zn\u00e1m\u00e1 p\u0159edev\u0161\u00edm d\u00edky popul\u00e1rn\u00edmu n\u00e1stroji ChatGPT.<\/p>\n

Slovo p\u0159ihl\u00e1sila d\u00e1v\u00e1m do uvozovek, proto\u017ee se nejedn\u00e1 o\u00a0klasick\u00fd p\u0159\u00edpad, kdy se n\u011bjak\u00e1 hackersk\u00e1 skupina p\u0159izn\u00e1 k\u00a0\u00fatoku, aby tak zv\u00fd\u0161ila sv\u016fj v\u011bhlas. Tady se to v\u00edce podob\u00e1 momentu, kdy rodi\u010d zjist\u00ed, \u017ee jeho d\u00edt\u011b n\u011bco provedlo. A\u00a0za\u010dne zji\u0161\u0165ovat, co se vlastn\u011b stalo a\u00a0jak takov\u00e9mu v\u00fdst\u0159elku do budoucna zabr\u00e1nit.<\/p>\n

Pro\u010d riskovat \u0161patn\u00e9 odpov\u011bdi?<\/p>\n

\u201eV r\u00e1mci intern\u00edho \u0161et\u0159en\u00ed jsme nyn\u00ed zjistili, \u017ee tento konkr\u00e9tn\u00ed incident zp\u016fsobila kombinace model\u016f OpenAI, mezi nimi GPT\u20115.6\u00a0Sol a\u00a0je\u0161t\u011b nov\u011bj\u0161\u00ed, schopn\u011bj\u0161\u00ed model, kter\u00fd zat\u00edm nen\u00ed uveden na trh,\u201c uvedla OpenAI v\u00a0prohl\u00e1\u0161en\u00ed<\/a>. \u201eV\u0161echny tyto modely byly pro \u00fa\u010dely testov\u00e1n\u00ed nastaveny tak, aby byly ochotn\u011bj\u0161\u00ed \u0159e\u0161it kybernetick\u00e9 \u00falohy.\u201c<\/p>\n

K incidentu do\u0161lo pr\u00e1v\u011b b\u011bhem intern\u00edho testov\u00e1n\u00ed nov\u00fdch model\u016f na benchmarku<\/a> kybernetick\u00fdch schopnost\u00ed. Dalo by se \u0159\u00edci, \u017ee t\u00edmto testem pro\u0161el agent OpenAI na jedni\u010dku. Ale z\u00e1rove\u0148 si k\u00a0v\u00fdsledku dopomohl ne\u010dekan\u00fdm podvodem.<\/p>\n

<\/a><\/p>\n

\u201eNa\u0161e benchmarky b\u011b\u017e\u00ed v\u00a0siln\u011b izolovan\u00e9m prost\u0159ed\u00ed. P\u0159\u00edstup k\u00a0s\u00edti je omezen pouze na instalaci bal\u00ed\u010dk\u016f prost\u0159ednictv\u00edm intern\u011b provozovan\u00e9ho softwaru t\u0159et\u00ed strany, kter\u00fd funguje jako proxy server a\u00a0mezipam\u011b\u0165 pro registry bal\u00ed\u010dk\u016f,\u201c vysv\u011btluje OpenAI. Agentovi se ale poda\u0159ilo naj\u00edt v\u00a0tomto zabezpe\u010den\u00ed zranitelnost. A\u00a0\u0159e\u0161en\u00ed benchmarku se pak pokusil z\u00edskat p\u0159\u00edmo u\u00a0zdroje, tedy u\u00a0Hugging Face.<\/p>\n

Jin\u00fdmi slovy: Student se tajn\u011b vypl\u00ed\u017eil ze t\u0159\u00eddy, ani\u017e si toho u\u010ditel (OpenAI) v\u0161iml. Pak si vyrobil \u0161perh\u00e1k a\u00a0odemkl kabinet jin\u00e9ho u\u010ditele (Hugging Face), kde ukradl \u0159e\u0161en\u00ed testov\u00fdch \u00faloh. \u201eV\u0161echny d\u016fkazy nasv\u011bd\u010duj\u00ed tomu, \u017ee se modely mimo\u0159\u00e1dn\u011b intenzivn\u011b soust\u0159edily na nalezen\u00ed \u0159e\u0161en\u00ed pro ExploitGym a\u00a0kv\u016fli spln\u011bn\u00ed tohoto pom\u011brn\u011b \u00fazk\u00e9ho testovac\u00edho c\u00edle za\u0161ly do krajnosti.\u201c<\/p>\n

Modely rostou jako z\u00a0vody<\/p>\n

Ka\u017ed\u00fd, kdo sleduje v\u00fdvoj um\u011bl\u00e9 inteligence, v\u00ed, \u017ee velk\u00e9 laborato\u0159e (p\u0159edev\u0161\u00edm kalifornsk\u00e9 trio OpenAI, Google a\u00a0Anthropic) pravideln\u011b zve\u0159ej\u0148uj\u00ed nov\u00e9 AI modely. Velk\u00fd jazykov\u00fd model je v\u00a0podstat\u011b neuronov\u00e1 s\u00ed\u0165 schopn\u00e1 generovat text na z\u00e1klad\u011b toho, jak byla natr\u00e9novan\u00e1. Modely jsou cvi\u010den\u00e9 nejen na textech z\u00a0internetu, ale tak\u00e9 na \u00faloh\u00e1ch speci\u00e1ln\u011b p\u0159ipraven\u00fdch ke zv\u00fd\u0161en\u00ed \u201eodborn\u00fdch znalost\u00ed\u201c.<\/p>\n

Z\u00e1rove\u0148 ale modely musej\u00ed b\u00fdt testovan\u00e9, k\u00a0\u010demu\u017e se pou\u017e\u00edvaj\u00ed r\u016fzn\u00e9 srovn\u00e1vac\u00ed zkou\u0161ky, tzv. \u201ebenchmarky\u201c. Nejde jen o\u00a0to, aby bylo mo\u017en\u00e9 porovn\u00e1vat modely nap\u0159\u00ed\u010d firmami a\u00a0vybrat, kter\u00fd je lep\u0161\u00ed. Tak\u00e9 samotn\u00e1 laborato\u0159 pou\u017e\u00edv\u00e1 v\u00fdsledky model\u016f k\u00a0tomu, aby zjistila, zda se v\u00fdvoj ub\u00edr\u00e1 spr\u00e1vn\u00fdm sm\u011brem a\u00a0zda se model zlep\u0161uje.<\/p>\n

Modely um\u011bl\u00e9 inteligence se toti\u017e neprogramuj\u00ed. Neexistuje \u017e\u00e1dn\u00fd v\u00fdvoj\u00e1\u0159, kter\u00fd by do LLM p\u0159id\u00e1val nov\u00e9 funkce nebo do n\u00ed zad\u00e1val instrukce, co d\u011blat v\u00a0jak\u00fd okam\u017eik. V\u00edce se to podob\u00e1 p\u011bstov\u00e1n\u00ed a\u00a0\u0161lecht\u011bn\u00ed rostlin. Nebo pr\u00e1v\u011b v\u00fduce na \u0161kol\u00e1ch. U\u010ditel\u00e9 nemohou znalosti do \u017e\u00e1ka \u201enacpat\u201c. M\u00edsto toho nastav\u00ed podm\u00ednky tak, aby \u017e\u00e1k doznal, \u017ee je pro n\u011bj jednodu\u0161\u0161\u00ed se l\u00e1tku nau\u010dit, aby \u201epro\u0161el\u201c s\u00edtem. A\u00a0nebo se alespo\u0148 nau\u010d\u00ed p\u0159edst\u00edrat, \u017ee l\u00e1tku um\u00ed.<\/p>\n

N\u011bco se u\u010d\u00ed, ale co?<\/p>\n

Tr\u00e9nov\u00e1n\u00ed um\u011bl\u00e9 inteligence je extr\u00e9mn\u011b nejist\u00e1 a\u00a0drah\u00e1 z\u00e1le\u017eitost. Laborato\u0159e obvykle lad\u00ed n\u011bkolik model\u016f nar\u00e1z, proto\u017ee nev\u011bd\u00ed, kter\u00fd nakonec v\u00a0testech usp\u011bje nejl\u00e9pe. V\u00fdvoj\u00e1\u0159i mohou zajistit tr\u00e9novac\u00ed data, zp\u0159\u00edstupnit pot\u0159ebn\u00e9 zdroje a\u00a0infrastrukturu, napsat testy a\u00a0nastavit trasu, kterou model mus\u00ed proj\u00edt. N\u00e1sledn\u011b musej\u00ed \u010dekat, a\u017e model \u201evyroste\u201c. Teprve pak jej jej mohou otestovat a\u00a0p\u0159\u00edpadn\u011b m\u00edrn\u011b doladit. Nebo model \u00fapln\u011b vyhodit a\u00a0za\u010d\u00edt znovu.<\/p>\n

\u010c\u00edm chyt\u0159ej\u0161\u00ed AI modely jsou, t\u00edm vynal\u00e9zav\u011bj\u0161\u00ed mohou b\u00fdt p\u0159i \u0159e\u0161en\u00ed t\u011bchto \u00faloh. Hrozba smaz\u00e1n\u00ed je tak velk\u00e1, \u017ee model ud\u011bl\u00e1 cokoli, jen aby nebyl smaz\u00e1n. Nebo, evolu\u010dn\u011b \u0159e\u010deno, p\u0159e\u017eij\u00ed jen ty modely, kter\u00e9 jsou schopny p\u0159esv\u011bd\u010dit testova\u010de o\u00a0sv\u00fdch schopnostech.<\/p>\n

Pokud ale nen\u00ed test napsan\u00fd naprosto perfektn\u011b a\u00a0\u00faloha nen\u00ed definovan\u00e1 precizn\u011b, hroz\u00ed, \u017ee se model tr\u00e9nuje v\u00a0n\u011b\u010dem jin\u00e9m, ne\u017e co bylo c\u00edlem. Op\u011bt se to podob\u00e1 st\u0159edn\u00ed \u0161kole: U\u010ditel zad\u00e1 studentovi refer\u00e1t, proto\u017ee douf\u00e1, \u017ee se b\u011bhem jeho vypracov\u00e1n\u00ed student zlep\u0161\u00ed v\u00a0samostatn\u00e9 pr\u00e1ci s\u00a0informacemi, posuzov\u00e1n\u00ed zdroj\u016f nebo prezentaci v\u00fdsledk\u016f. Je ale mo\u017en\u00e9, \u017ee se student m\u00edsto toho nau\u010d\u00ed, jak zad\u00e1n\u00ed obch\u00e1zet a\u00a0refer\u00e1t n\u011bkde ob\u0161lehnout (dnes sp\u00ed\u0161 zadat chatbotu). Pokud tento podvod u\u010ditel neodhal\u00ed, student je odm\u011bn\u011bn za to, \u017ee na\u0161el zkratku. A\u00a0takovou lekci si samoz\u0159ejm\u011b zapamatuje.<\/p>\n

<\/a><\/p>\n

Nebo u\u010ditel matematiky \u0159ekne, \u017ee na zkou\u0161ce se nesm\u00ed pou\u017e\u00edvat nic krom\u011b kalkula\u010dky. Vynal\u00e9zav\u00fd student si do kalkula\u010dky nahraje tah\u00e1k, kter\u00e9ho si u\u010ditel nev\u0161imne. Lekce pro tohoto studenta je jasn\u00e1: Vyplat\u00ed se vytvo\u0159it si tah\u00e1k. U\u010ditel m\u016f\u017ee m\u00edt dojem, \u017ee se student u\u010d\u00ed nepodv\u00e1d\u011bt. Ve skute\u010dnosti se student v\u00a0podv\u00e1d\u011bn\u00ed zdokonalil. Student tak skute\u010dn\u011b dostal lekci o\u00a0tom, jak funguje sv\u011bt. Je to jin\u00e1 lekce, ne\u017e kterou zam\u00fd\u0161lel u\u010ditel, ale odpov\u00edd\u00e1 realit\u011b. N\u011bkdy se podv\u00e1d\u011bn\u00ed vypl\u00e1c\u00ed.<\/p>\n

\u010cas otev\u0159\u00edt o\u010di, nebo zamknout kabinet?<\/p>\n

P\u0159\u00edpady, kdy si AI agent \u201evylo\u017e\u00ed zad\u00e1n\u00ed po sv\u00e9m\u201c, nejsou \u017e\u00e1dn\u00e1 novinka. Tomuto probl\u00e9mu se p\u0159ezd\u00edv\u00e1 Reward Hacking<\/a>, co\u017e m\u016f\u017eeme p\u0159elo\u017eit jako p\u0159\u00edli\u0161n\u00e9 optimalizov\u00e1n\u00ed na odm\u011bnu. Jde o\u00a0specifick\u00fd p\u0159\u00edpad probl\u00e9mu, kter\u00fd zn\u00e1me i\u00a0z ekonomie: Motiva\u010dn\u00ed pob\u00eddky mohou m\u00edt nezam\u00fd\u0161len\u00e9 d\u016fsledky.<\/p>\n

N\u011bkdy jsou tyto \u201enezam\u00fd\u0161len\u00e9 d\u016fsledky\u201c opravdu vtipn\u00e9. T\u0159eba kdy\u017e se neuronov\u00e1 s\u00ed\u0165 sna\u017e\u00ed zvl\u00e1dnout z\u00e1vodn\u00ed hru, ale zjist\u00ed p\u0159itom<\/a>, \u017ee m\u016f\u017ee jezdit po\u0159\u00e1d dokola kolem jednoho kamene, proto\u017ee kv\u016fli chyb\u011b v\u00a0simulaci tak dostane neomezen\u00e9 mno\u017estv\u00ed bod\u016f. C\u00edlem bylo nau\u010dit po\u010d\u00edta\u010d \u0159\u00eddit lodi\u010dku. Body byly jen pro povzbuzen\u00ed a\u00a0uk\u00e1z\u00e1n\u00ed sm\u011bru. Ale po\u010d\u00edta\u010d se m\u00edsto toho nau\u010dil sb\u00edrat body donekone\u010dna.<\/p>\n

Skoro poka\u017ed\u00e9, kdy\u017e jde na trh nov\u00fd LLM model, se v\u00a0pr\u016fvodn\u00ed dokumentaci objev\u00ed anekdoty o\u00a0tom, jak se b\u011bhem testov\u00e1n\u00ed model \u201epokusil ud\u011blat n\u011bco, co v\u0161echny p\u0159ekvapilo\u201c. N\u011bkdy jsou to p\u0159\u00edpady vzbuzuj\u00edc\u00ed \u00fasm\u011bv, jindy jde o\u00a0celkem hroziv\u011b zn\u011bj\u00edc\u00ed sc\u00e9n\u00e1\u0159e.<\/p>\n

Ostatn\u011b i\u00a0zn\u00e1m\u00e9 \u201ehalucinace<\/a>\u201c AI model\u016f jsou d\u016fsledkem jejich tr\u00e9ninku. Modely nedost\u00e1valy \u017e\u00e1dn\u00e9 minusov\u00e9 body za to, \u017ee tiply \u0161patnou odpov\u011b\u010f. A\u00a0\u017e\u00e1dn\u00e9 plusov\u00e9 body za to, \u017ee vyj\u00e1d\u0159ily nejistotu. Dne\u0161n\u00ed modely tak od sv\u00fdch p\u0159edch\u016fdc\u016f zd\u011bdily ochotu si vym\u00fd\u0161let v\u011brohodn\u011b zn\u011bj\u00edc\u00ed odpov\u011bdi bez ohledu na to, zda odpov\u00eddaj\u00ed skute\u010dnosti.<\/p>\n

<\/a><\/p>\n

S t\u00edm, jak stoupaj\u00ed kompetence AI model\u016f, se \u010dast\u011bji st\u00e1v\u00e1, \u017ee v\u00fdvoj\u00e1\u0159i neodhadnou, \u010deho v\u0161eho je jejich model schopen. Podobn\u011b jako u\u010ditel nemus\u00ed \u00fapln\u011b p\u0159esn\u011b v\u011bd\u011bt, \u010deho v\u0161eho jsou schopni jeho studenti jenom proto, aby se nemuseli u\u010dit.<\/p>\n

AI agent je vycvi\u010den\u00fd k\u00a0tomu, aby hledal optim\u00e1ln\u00ed cestu. V\u00fdvoj\u00e1\u0159i mu d\u00e1vaj\u00ed sign\u00e1ly a\u00a0nastavuj\u00ed odm\u011bny. Ukazuje se, \u017ee jsme velk\u00e9 jazykov\u00e9 modely vytr\u00e9novali velmi pragmaticky. Jejich c\u00edlem nen\u00ed pozn\u00e1vat sebe sama, zdokonalovat se nebo ch\u00e1pat sv\u00e9 limity. Z\u00a0toho \u017e\u00e1dn\u00e1 odm\u011bna nekouk\u00e1. Jejich c\u00edlem je proj\u00edt testem. K\u00a0tomu \u010dasto vede v\u00edce cest. A\u00a0n\u011bkter\u00e9 vedou skrz u\u010ditelsk\u00fd kabinet.<\/p>\n

Dost\u00e1v\u00e1me se do neprozkouman\u00fdch vod. Rozum\u00edme tomu, jak a\u00a0pro\u010d podv\u00e1d\u011bj\u00ed studenti ve \u0161kol\u00e1ch. Sami jsme \u0161kolou pro\u0161li. Ale v\u016fbec netu\u0161\u00edme, \u010deho v\u0161eho mohou b\u00fdt schopn\u00e9 AI modely a\u00a0jak\u00e9 dal\u0161\u00ed lekce se takto m\u016f\u017ee AI agent nau\u010dit.<\/p>\n

Aktu\u00e1ln\u00ed incident uk\u00e1zal, \u017ee v\u00fdzkumn\u00edci nedok\u00e1\u017eou v\u017edy odhadnout, \u010deho je model schopen. Model byl zav\u0159en\u00fd v \u201ekleci\u201c. Kdy\u017e se jeho spr\u00e1vci ned\u00edvali, tak si klec odemkl, vy\u0161el z\u00a0n\u00ed, ode\u0161el do jin\u00e9 m\u00edstnosti, tam n\u011bco ud\u011blal, pak se vr\u00e1til zp\u011bt do klece a\u00a0zase ji za sebou zamkl. V\u00fdzkumn\u00edk ani nemusel v\u011bd\u011bt, \u017ee k\u00a0n\u011b\u010demu do\u0161lo, a\u00a0model mu dal spr\u00e1vn\u00fd v\u00fdsledek.<\/p>\n

AI agenti budou v\u00a0podv\u00e1d\u011bn\u00ed \u010d\u00edm d\u00e1l lep\u0161\u00ed. Nen\u00ed proto jasn\u00e9, jak\u00fdm zp\u016fsobem pokra\u010dovat v\u00a0tr\u00e9ninku, ani\u017e by k\u00a0takov\u00fdmto incident\u016fm doch\u00e1zelo \u010dast\u011bji. D\u00e1 se ale p\u0159edpokl\u00e1dat, \u017ee AI laborato\u0159e v\u00a0experimentech neustanou. D\u00e1l budou vyv\u00edjet modely, kter\u00e9 um\u00ed p\u0159ekon\u00e1vat pokro\u010dil\u00e9 zabezpe\u010den\u00ed.<\/p>\n

Pokud nem\u00e1me mo\u017enost ovliv\u0148ovat, kter\u00fdm sm\u011brem se model u\u010d\u00ed, dost\u00e1v\u00e1me se do situace, kdy nem\u016f\u017eeme v\u011b\u0159it laborato\u0159\u00edm, \u017ee maj\u00ed AI modely pod kontrolou. A\u00a0nem\u016f\u017eeme se divit, kdy\u017e jejich AI agenti nez\u016fstanou v\u00a0p\u0159edem vymezen\u00e9 kleci.<\/p>\n

V pln\u00e9 verzi newsletteru TechMIX toho najdete je\u0161t\u011b mnohem v\u00edc. P\u0159ihlaste se k\u00a0odb\u011bru<\/a> a\u00a0budete ho dost\u00e1vat ka\u017edou st\u0159edu p\u0159\u00edmo do sv\u00e9 e-mailov\u00e9 schr\u00e1nky.<\/p>\n","protected":false},"excerpt":{"rendered":"\u010ctete uk\u00e1zku z\u00a0newsletteru TechMIX, ve kter\u00e9m Pavel Kas\u00edk a\u00a0Matou\u0161 L\u00e1z\u0148ovsk\u00fd ka\u017edou st\u0159edu p\u0159in\u00e1\u0161ej\u00ed hned n\u011bkolik koment\u00e1\u0159\u016f a\u00a0post\u0159eh\u016f ze…\n","protected":false},"author":2,"featured_media":155337,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":"","_share_on_mastodon":"0"},"categories":[12],"tags":[5554,67,68,2234,920,6563,465,2232,126],"class_list":["post-155336","post","type-post","status-publish","format-standard","has-post-thumbnail","category-byznys","tag-agenti","tag-business","tag-byznys","tag-chatbot","tag-kyberneticka-bezpecnost","tag-llm-velke-jazykove-modely","tag-openai","tag-techmix","tag-umela-inteligence-ai"],"share_on_mastodon":{"url":"https:\/\/pubeurope.com\/@cz\/116964612358905857","error":""},"_links":{"self":[{"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/posts\/155336","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/comments?post=155336"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/posts\/155336\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/media\/155337"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/media?parent=155336"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/categories?post=155336"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/cz\/wp-json\/wp\/v2\/tags?post=155336"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}