{"id":37517,"date":"2025-11-20T17:00:00","date_gmt":"2025-11-20T16:00:00","guid":{"rendered":"https:\/\/www.gnews.cz\/?p=37517"},"modified":"2025-11-20T08:53:18","modified_gmt":"2025-11-20T07:53:18","slug":"tyden-ve-svete-ai-autonomni-vozy-waymo-na-americkych-dalnicich-open-source-llm-kimi-k2-vladne-zebrickum-tvrzeni-anthropic-vyvolava-bezpecnostni-kontroverzi","status":"publish","type":"post","link":"https:\/\/gnews.cz\/cs_cz\/ekonomika\/tyden-ve-svete-ai-autonomni-vozy-waymo-na-americkych-dalnicich-open-source-llm-kimi-k2-vladne-zebrickum-tvrzeni-anthropic-vyvolava-bezpecnostni-kontroverzi\/","title":{"rendered":"T\u00fdden ve sv\u011bt\u011b AI: Autonomn\u00ed vozy Waymo na americk\u00fdch d\u00e1lnic\u00edch, open-source LLM Kimi K2 vl\u00e1dne \u017eeb\u0159\u00ed\u010dk\u016fm, tvrzen\u00ed Anthropic vyvol\u00e1v\u00e1 bezpe\u010dnostn\u00ed kontroverzi"},"content":{"rendered":"\n

Tento t\u00fdden byl sv\u011bdkem v\u00fdznamn\u00fdch pokrok\u016f v oblasti um\u011bl\u00e9 inteligence, kter\u00e9 se vyzna\u010dovaly historick\u00fdm miln\u00edkem pro autonomn\u00ed vozidla, p\u0159\u00edchodem nov\u00e9ho v\u00fdkonn\u00e9ho open-source modelu AI a kontroverzn\u00ed debatou o roli AI v kybernetick\u00e9 bezpe\u010dnosti. Tyto ud\u00e1losti poukazuj\u00ed na technologii, kter\u00e1 se rychle p\u0159esouv\u00e1 z v\u00fdzkumn\u00fdch laborato\u0159\u00ed do re\u00e1ln\u00fdch aplikac\u00ed a p\u0159in\u00e1\u0161\u00ed jak bezprecedentn\u00ed mo\u017enosti, tak i nov\u00e9 komplexn\u00ed v\u00fdzvy.<\/p>\n\n\n\n

Revoluce v autonomn\u00ed doprav\u011b: Waymo poprv\u00e9 voz\u00ed z\u00e1kazn\u00edky po d\u00e1lnic\u00edch bez \u0159idi\u010de<\/h2>\n\n\n\n

V historick\u00e9m okam\u017eiku pro technologii autonomn\u00edho \u0159\u00edzen\u00ed se Waymo stalo prvn\u00ed spole\u010dnost\u00ed, kter\u00e1 nasadila pln\u011b autonomn\u00ed tax\u00edky bez \u0159idi\u010de na americk\u00fdch d\u00e1lnic\u00edch<\/strong>. Tato slu\u017eba, kter\u00e1 je nyn\u00ed k dispozici plat\u00edc\u00edm z\u00e1kazn\u00edk\u016fm ve Phoenixu, San Franciscu a Los Angeles, m\u016f\u017ee d\u00edky vyu\u017eit\u00ed vysokorychlostn\u00edch silnic zkr\u00e1tit dobu j\u00edzdy a\u017e o 50 %. Tento \u00fasp\u011bch je vyvrcholen\u00edm milion\u016f kilometr\u016f testov\u00e1n\u00ed na ve\u0159ejn\u00fdch silnic\u00edch, uzav\u0159en\u00fdch okruz\u00edch a v simulac\u00edch. Kl\u00ed\u010dovou p\u0159ek\u00e1\u017ekou bylo zvl\u00e1dnut\u00ed \u201epsychologick\u00e9ho dopadu\u201c na cestuj\u00edc\u00ed, kte\u0159\u00ed se vzdali kontroly p\u0159i rychlosti 105 km\/h, co\u017e je v\u00fdzva, kterou spoluzakladatel Waymo v\u00fdslovn\u011b uznal. Spole\u010dnost, kter\u00e1 se m\u016f\u017ee pochlubit bezpe\u010dnostn\u00ed bilanc\u00ed s o 91 % m\u00e9n\u011b nehodami zp\u016fsobuj\u00edc\u00edmi zran\u011bn\u00ed ne\u017e lid\u0161t\u00ed \u0159idi\u010di v porovnateln\u00fdch sc\u00e9n\u00e1\u0159\u00edch, nyn\u00ed pl\u00e1nuje v\u00fdznamnou expanzi do m\u011bst jako Dallas, Detroit a Lond\u00fdn.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Kimi K2: Nov\u00fd open-source konkurent<\/h2>\n\n\n\n

Mezit\u00edm se na sc\u00e9n\u011b model\u016f um\u011bl\u00e9 inteligence objevil nov\u00fd siln\u00fd konkurent. Spole\u010dnost Moonshot AI <\/strong>vydala Kimi K2 Thinking<\/strong>, open-source model s bilionem parametr\u016f, kter\u00fd konkuruje \u0161pi\u010dkov\u00fdm propriet\u00e1rn\u00edm syst\u00e9m\u016fm v \u201eagentn\u00edch\u201c \u00fakolech \u2013 t\u011bch, kter\u00e9 vy\u017eaduj\u00ed v\u00edcestup\u0148ov\u00e9 uva\u017eov\u00e1n\u00ed a pou\u017eit\u00ed n\u00e1stroj\u016f. Jeho jedine\u010dn\u00e1 architektura mu umo\u017e\u0148uje prokl\u00e1dat uva\u017eov\u00e1n\u00ed a akci, zastavovat se, aby \u201ep\u0159em\u00fd\u0161lel\u201c mezi v\u00fdb\u011brem n\u00e1stroj\u016f, co\u017e mu umo\u017enilo vy\u0159e\u0161it pokro\u010dil\u00fd matematick\u00fd probl\u00e9m prost\u0159ednictv\u00edm 23 r\u016fzn\u00fdch krok\u016f uva\u017eov\u00e1n\u00ed a akce. Model byl dolad\u011bn s pou\u017eit\u00edm 4bitov\u00e9 p\u0159esnosti, d\u00edky \u010demu\u017e je rychlej\u0161\u00ed a m\u016f\u017ee b\u011b\u017eet na levn\u011bj\u0161\u00edm hardwaru<\/strong>, co\u017e je v\u00fdznamn\u00e1 v\u00fdhoda na trz\u00edch s omezen\u00fdm p\u0159\u00edstupem k pokro\u010dil\u00fdm \u010dip\u016fm.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Kontroverze kolem kyber\u00fatoku pomoc\u00ed um\u011bl\u00e9 inteligence<\/h2>\n\n\n\n

Tento t\u00fdden se tak\u00e9 odehr\u00e1la v\u00fdznamn\u00e1 kontroverze. Spole\u010dnost Anthropic tvrdila, \u017ee zma\u0159ila prvn\u00ed rozs\u00e1hl\u00fd kyber\u00fatok proveden\u00fd s minim\u00e1ln\u00edm lidsk\u00fdm z\u00e1sahem<\/strong>, kter\u00fd \u00fadajn\u011b provedli \u010d\u00edn\u0161t\u00ed hacke\u0159i podporovan\u00ed st\u00e1tem pomoc\u00ed jej\u00ed um\u011bl\u00e9 inteligence Claude Code. Spole\u010dnost uvedla, \u017ee um\u011bl\u00e1 inteligence provedla 80\u201390 % technick\u00fdch krok\u016f. Toto tvrzen\u00ed v\u0161ak vyvolalo silnou skepsi ze strany nez\u00e1visl\u00fdch v\u00fdzkumn\u00edk\u016f v oblasti kybernetick\u00e9 bezpe\u010dnosti. Ti argumentuj\u00ed, \u017ee sou\u010dasn\u00ed AI agenti je\u0161t\u011b nejsou schopni autonomn\u011b prov\u00e1d\u011bt tak slo\u017eit\u00e9 \u00fatoky, a poukazuj\u00ed na to, \u017ee pou\u017eit\u00e9 n\u00e1stroje byly b\u011b\u017en\u00e9 a \u017ee AI m\u00e1 zn\u00e1mou tendenci<\/strong> \u201ehalucinovat\u201c fakta<\/strong>, co\u017e z n\u00ed \u010din\u00ed nespolehliv\u00e9ho hackera. Tato ud\u00e1lost vyvolala z\u00e1sadn\u00ed debatu o re\u00e1ln\u00fdch schopnostech AI v oblasti kybernetick\u00e9 bezpe\u010dnosti a o tenk\u00e9 hranici mezi demonstrac\u00ed s\u00edly produktu a \u0161\u00ed\u0159en\u00edm strachu.<\/p>\n\n\n\n

\n

I kdy\u017e se shodli na tom, \u017ee AI m\u016f\u017ee urychlit \u00fakoly, jako je anal\u00fdza protokol\u016f a reverzn\u00ed in\u017een\u00fdrstv\u00ed, zjistili, \u017ee agenti AI zat\u00edm nejsou schopni prov\u00e1d\u011bt v\u00edcestup\u0148ov\u00e9 \u00fakoly bez lidsk\u00e9ho z\u00e1sahu a \u017ee kybernetick\u00e9 \u00fatoky neautomatizuj\u00ed o mnoho efektivn\u011bji ne\u017e hackersk\u00e9 n\u00e1stroje, kter\u00e9 jsou k dispozici ji\u017e des\u00edtky let. \u201e\u00dato\u010dn\u00edci zde nevym\u00fd\u0161lej\u00ed nic nov\u00e9ho,\u201c<\/strong> uvedl v\u00fdzkumn\u00edk Kevin Beaumont v online bezpe\u010dnostn\u00edm f\u00f3ru.<\/p>\n<\/blockquote>\n\n\n\n

Krom\u011b Claude Code hacke\u0159i pou\u017e\u00edvali b\u011b\u017en\u00e9 open-source n\u00e1stroje, uvedla spole\u010dnost Anthropic. Obrana proti t\u011bmto zn\u00e1m\u00fdm n\u00e1stroj\u016fm je v\u0161ak bezpe\u010dnostn\u00edm expert\u016fm tak\u00e9 dob\u0159e zn\u00e1ma a nen\u00ed jasn\u00e9, jak by Claude Code tuto situaci zm\u011bnil.
pole\u010dnost Anthropic sama pouk\u00e1zala na to, \u017ee Claude Code mohl informace, kter\u00e9 \u00fadajn\u011b hacknul, zkreslit, proto\u017ee \u201e\u010dasto p\u0159eh\u00e1n\u011bl zji\u0161t\u011bn\u00ed\u201c a \u201eob\u010das si data vym\u00fd\u0161lel\u201c. Takov\u00e9 chov\u00e1n\u00ed je podle spole\u010dnosti v\u00fdznamnou p\u0159ek\u00e1\u017ekou pro pou\u017eit\u00ed syst\u00e9mu k prov\u00e1d\u011bn\u00ed kybernetick\u00fdch \u00fatok\u016f.<\/p>\n\n\n\n

\n

V \u0159\u00edjnu obvinil David Sacks, poradce B\u00edl\u00e9ho domu pro um\u011blou inteligenci, spole\u010dnost Anthropic z prov\u00e1d\u011bn\u00ed \u201esofistikovan\u00e9 strategie regulace zalo\u017een\u00e9 na \u0161\u00ed\u0159en\u00ed strachu\u201c<\/strong>.<\/p>\n<\/blockquote>\n\n\n\n

\"\"<\/figure>\n\n\n\n

Um\u011bl\u00e1 inteligence se u\u010d\u00ed prohled\u00e1vat vlastn\u00ed pam\u011b\u0165, \u010d\u00edm\u017e zvy\u0161uje efektivitu a p\u0159esnost<\/h2>\n\n\n\n

Yuchen Fan a jeho kolegov\u00e9 z Tsinghua University, Shanghai Jiao Tong University, Shanghai AI Laboratory, University College London, China State Construction Engineering Corporation Third Bureau a WeChat AI p\u0159edstavili metodu Self-Search Reinforcement Learning (SSRL)<\/strong>, inovativn\u00ed metodu, kter\u00e1 v\u00fdrazn\u011b zlep\u0161uje zp\u016fsob, jak\u00fdm velk\u00e9 jazykov\u00e9 modely p\u0159istupuj\u00ed k informac\u00edm a vyu\u017e\u00edvaj\u00ed je. Tento p\u0159\u00edstup u\u010d\u00ed modely systematicky prohled\u00e1vat sv\u00e9 vlastn\u00ed parametry<\/strong> \u2013 simuluje vyhled\u00e1v\u00e1n\u00ed na webu generov\u00e1n\u00edm dotaz\u016f a odpov\u00edd\u00e1n\u00edm na n\u011b \u2013 co\u017e dramaticky zlep\u0161uje z\u00edsk\u00e1v\u00e1n\u00ed znalost\u00ed z jejich st\u00e1vaj\u00edc\u00edch tr\u00e9novac\u00edch dat. V testech na \u0161esti benchmarkov\u00fdch testech dos\u00e1hly modely s podporou SSRL vynikaj\u00edc\u00edho v\u00fdkonu, p\u0159i\u010dem\u017e jeden model dos\u00e1hl p\u0159esnosti 43,1 %. Tato technika tak\u00e9 vytv\u00e1\u0159\u00ed efektivn\u011bj\u0161\u00ed hybridn\u00ed syst\u00e9my, kde AI nejprve konzultuje sv\u00e9 intern\u00ed znalosti, ne\u017e vyhled\u00e1 extern\u00ed informace, co\u017e potenci\u00e1ln\u011b sni\u017euje v\u00fdpo\u010detn\u00ed n\u00e1klady a z\u00e1rove\u0148 zlep\u0161uje p\u0159esnost odpov\u011bd\u00ed v \u00fakolech n\u00e1ro\u010dn\u00fdch na znalosti.<\/p>\n\n\n\n

\"\"<\/figure>\n\n\n\n


The Batch<\/a> \u2013 DeepLearning.Ai by Andrew Ng \/ gnews.cz \u2013 GH<\/strong><\/p>\n\n\n\n

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

Tento t\u00fdden byl sv\u011bdkem v\u00fdznamn\u00fdch pokrok\u016f v oblasti um\u011bl\u00e9 inteligence, kter\u00e9 se vyzna\u010dovaly historick\u00fdm miln\u00edkem pro autonomn\u00ed vozidla, p\u0159\u00edchodem nov\u00e9ho v\u00fdkonn\u00e9ho open-source…<\/p>\n","protected":false},"author":3,"featured_media":37533,"comment_status":"open","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"_lmt_disableupdate":"","_lmt_disable":"","footnotes":""},"categories":[7144,4,3711,2973],"tags":[122,5173,7588,7591,7587,7586,7590,7589,7134,7601,3785,7201,38,7585,41],"class_list":["post-37517","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-ai-ekonomika","category-ekonomika","category-english","category-zahranicni","tag-ai","tag-anthropic","tag-autonomni-vozy","tag-claude-code","tag-kimi","tag-kimi-k2","tag-kyberutoky","tag-moonshot-ai","tag-open-source","tag-self-search-reinforcement-learning-ssrl","tag-the-batch","tag-tyden-ve-svete-ai","tag-usa","tag-waymo","tag-world"],"acf":[],"aioseo_notices":[],"_links":{"self":[{"href":"https:\/\/gnews.cz\/cs_cz\/wp-json\/wp\/v2\/posts\/37517","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/gnews.cz\/cs_cz\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/gnews.cz\/cs_cz\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/gnews.cz\/cs_cz\/wp-json\/wp\/v2\/users\/3"}],"replies":[{"embeddable":true,"href":"https:\/\/gnews.cz\/cs_cz\/wp-json\/wp\/v2\/comments?post=37517"}],"version-history":[{"count":2,"href":"https:\/\/gnews.cz\/cs_cz\/wp-json\/wp\/v2\/posts\/37517\/revisions"}],"predecessor-version":[{"id":37535,"href":"https:\/\/gnews.cz\/cs_cz\/wp-json\/wp\/v2\/posts\/37517\/revisions\/37535"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/gnews.cz\/cs_cz\/wp-json\/wp\/v2\/media\/37533"}],"wp:attachment":[{"href":"https:\/\/gnews.cz\/cs_cz\/wp-json\/wp\/v2\/media?parent=37517"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/gnews.cz\/cs_cz\/wp-json\/wp\/v2\/categories?post=37517"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/gnews.cz\/cs_cz\/wp-json\/wp\/v2\/tags?post=37517"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}