{"id":2201,"date":"2024-05-27T07:30:49","date_gmt":"2024-05-27T07:30:49","guid":{"rendered":"https:\/\/howtogeek.blog\/cs\/?p=2201"},"modified":"2024-05-27T07:30:49","modified_gmt":"2024-05-27T07:30:49","slug":"googles-ai-search-tells-users-to-eat-rocks-and-put-glue-on-pizza-cs","status":"publish","type":"post","link":"https:\/\/howtogeek.blog\/cs\/googles-ai-search-tells-users-to-eat-rocks-and-put-glue-on-pizza-cs\/","title":{"rendered":"Vyhled\u00e1v\u00e1n\u00ed AI od Googlu \u0159\u00edk\u00e1 u\u017eivatel\u016fm, aby jedli kameny a lepili pizzu"},"content":{"rendered":"

Technologick\u00e9 spole\u010dnosti se na AI zam\u011b\u0159uj\u00ed v\u00edce ne\u017e kdy p\u0159edt\u00edm, co\u017e vedlo k tomu, \u017ee se modely AI neust\u00e1le zlep\u0161uj\u00ed a zlep\u0161uj\u00ed v r\u016fzn\u00fdch \u00fakolech. Nap\u0159\u00edklad OpenAI ned\u00e1vno p\u0159edstavila sv\u016fj nejnov\u011bj\u0161\u00ed vlajkov\u00fd model GPT-4o, kter\u00fd dokonce dok\u00e1\u017ee detekovat a simulovat lidsk\u00e9 emoce. Nov\u00fd multimod\u00e1ln\u00ed model m\u016f\u017ee krom\u011b textov\u00fdch funkc\u00ed shroma\u017e\u010fovat data a poskytovat informace pomoc\u00ed vizu\u00e1ln\u00edch a zvukov\u00fdch funkc\u00ed.<\/p>\n

Ne v\u0161ude je to v\u0161ak dobr\u00e1 zpr\u00e1va, proto\u017ee tyto modely mohou st\u00e1le d\u011blat chyby a nab\u00edzet \u0161patn\u00e9 informace a n\u00e1vrhy. Nejnov\u011bj\u0161\u00edm p\u0159\u00edkladem toho jsou p\u0159ehledy AI spole\u010dnosti Google, kter\u00e9 vyhled\u00e1vac\u00ed gigant p\u0159edstavil za\u010d\u00e1tkem tohoto m\u011bs\u00edce. C\u00edlem bylo poskytnout souhrny informac\u00ed, kter\u00e9 u\u017eivatel\u00e9 hledali, generovan\u00e9 um\u011blou inteligenc\u00ed. P\u0159i skute\u010dn\u00e9m pou\u017eit\u00ed se tato funkce ukazuje jako nespolehliv\u00e1, proto\u017ee AI u\u017eivatel\u016fm nab\u00edz\u00ed fakticky nespr\u00e1vn\u00e9 odpov\u011bdi a podivn\u00e9 n\u00e1vrhy, kter\u00e9 ned\u00e1vaj\u00ed smysl.<\/p>\n

P\u0159ehledy AI nab\u00edzej\u00ed podivn\u00e9 n\u00e1vrhy \u2013 Internet je rozd\u011blen\u00fd<\/h2>\n

P\u0159ehledy AI spole\u010dnosti Google jsou navr\u017eeny tak, aby sn\u00ed\u017eily \u00fasil\u00ed spojen\u00e9 s vyhled\u00e1v\u00e1n\u00edm informac\u00ed t\u00edm, \u017ee poskytuj\u00ed souhrny informac\u00ed shrom\u00e1\u017ed\u011bn\u00fdch z r\u016fzn\u00fdch str\u00e1nek generovan\u00e9 um\u011blou inteligenc\u00ed. Probl\u00e9m je v tom, \u017ee AI v sou\u010dasn\u00e9 dob\u011b nedok\u00e1\u017ee ur\u010dit, kter\u00fd zdroj poskytuje v\u011brohodn\u00e9 a p\u0159esn\u00e9 informace, co\u017e znamen\u00e1, \u017ee m\u016f\u017ee vytv\u00e1\u0159et souhrny pomoc\u00ed fale\u0161n\u00fdch dat.<\/p>\n

Nav\u00edc se nezd\u00e1 b\u00fdt nejlep\u0161\u00ed p\u0159i ur\u010dov\u00e1n\u00ed z\u00e1m\u011bru u\u017eivatele. Nap\u0159\u00edklad navrhl, aby geologov\u00e9 na dotaz u\u017eivatele doporu\u010dovali sn\u00edst jeden mal\u00fd k\u00e1men denn\u011b, p\u0159i\u010dem\u017e svou odpov\u011b\u010f zjevn\u011b zakl\u00e1dali na informac\u00edch z humorn\u00e9ho webu The Onion.<\/p>\n

D\u00e1le bylo uvedeno, \u017ee kameny jsou \u017eivotn\u011b d\u016fle\u017eit\u00fdm zdrojem miner\u00e1l\u016f a vitam\u00edn\u016f pot\u0159ebn\u00fdch pro lep\u0161\u00ed tr\u00e1vic\u00ed zdrav\u00ed, a dokonce se navrhlo skr\u00fdt kameny a obl\u00e1zky v potravin\u00e1ch, jako je zmrzlina. Podobn\u011b na dal\u0161\u00ed ot\u00e1zku t\u00fdkaj\u00edc\u00ed se toho, jak s\u00fdr l\u00e9pe p\u0159ilnout k pizze, AI spole\u010dnosti Google navrhlo p\u0159idat lepidlo, aby se zv\u00fd\u0161ila lepivost.<\/p>\n

\"\"
Zdroj: X<\/figcaption><\/figure>\n

Kdy\u017e byl dot\u00e1z\u00e1n na po\u010det muslimsk\u00fdch prezident\u016f, kter\u00e9 USA m\u011bly, uvedlo, \u017ee Barack Obama byl jedin\u00fdm muslimsk\u00fdm prezidentem, co\u017e je fakticky nespr\u00e1vn\u00e9, proto\u017ee je k\u0159es\u0165an. A na ot\u00e1zku t\u00fdkaj\u00edc\u00ed se proch\u00e1zej\u00edc\u00edch ledvinov\u00fdch kamen\u016f AI odpov\u011bd\u011bla, \u017ee se doporu\u010duje vyp\u00edt dva litry mo\u010di ka\u017ed\u00fdch 24 hodin.<\/p>\n

Um\u011bl\u00e1 inteligence spole\u010dnosti Google tak\u00e9 uvedla, \u017ee pes hr\u00e1l v NBA a dokonce se mu poda\u0159ilo vytvo\u0159it novou formu m\u011b\u0159en\u00ed nazvanou \u201ekilotomato\u201c, kdy\u017e byl dot\u00e1z\u00e1n na tato t\u00e9mata. Existuje n\u011bkolik dal\u0161\u00edch p\u0159\u00edklad\u016f, kter\u00e9 ilustruj\u00ed, jak p\u0159ehledy AI mohou a poskytovaly nespr\u00e1vn\u00e9 informace.<\/p>\n

\"\"
Zdroj: Sherr\/CNET<\/figcaption><\/figure>\n

Zat\u00edmco internet za\u017e\u00edv\u00e1 poledne s vesel\u00fdmi odpov\u011b\u010fmi, tyto odpov\u011bdi mohou b\u00fdt tak\u00e9 nebezpe\u010dn\u00e9. P\u0159ehledy Google AI b\u011bhem okam\u017eiku poskytly u\u017eivateli dotaz na depresi sk\u00e1kaj\u00edc\u00ed z Golden Gate Bridge jako jeden z mo\u017en\u00fdch n\u00e1vrh\u016f.<\/p>\n

\"\"
Zdroj: Reddit<\/figcaption><\/figure>\n

Jak\u00e1 je odpov\u011b\u010f spole\u010dnosti Google?<\/h2>\n

Aby se Google vypo\u0159\u00e1dal se slo\u017eitou situac\u00ed, kterou vytvo\u0159ily jeho p\u0159ehledy AI, podnik\u00e1 rychl\u00e9 kroky k n\u00e1prav\u011b chyb ve faktick\u00fdch informac\u00edch. Tyto instance tak\u00e9 vyu\u017eije ke zlep\u0161en\u00ed sv\u00e9 AI, aby se sn\u00ed\u017eila mo\u017enost takov\u00fdch incident\u016f.<\/p>\n

To znamen\u00e1, \u017ee spole\u010dnost tvrd\u00ed, \u017ee AI funguje do zna\u010dn\u00e9 m\u00edry tak, jak by m\u011bla, a k t\u011bmto nespr\u00e1vn\u00fdm odpov\u011bd\u00edm do\u0161lo kv\u016fli poru\u0161en\u00ed z\u00e1sad a velmi neobvykl\u00fdm u\u017eivatelsk\u00fdm dotaz\u016fm. Nep\u0159edstavuj\u00ed zku\u0161enost, kterou m\u00e1 v\u011bt\u0161ina lid\u00ed s p\u0159ehledy AI. Google tak\u00e9 uvedl, \u017ee mnoho p\u0159\u00edklad\u016f ukazuj\u00edc\u00edch nespr\u00e1vn\u00e9 nebo podivn\u00e9 odpov\u011bdi bylo upraveno a p\u0159i intern\u00edm testov\u00e1n\u00ed AI nemohl reprodukovat podobn\u00e9 v\u00fdsledky.<\/p>\n

Omezen\u00ed um\u011bl\u00e9 inteligence<\/h2>\n

Od spu\u0161t\u011bn\u00ed ChatGPT od OpenAI p\u0159ed dv\u011bma lety u\u0161la AI a jej\u00ed p\u0159\u00edbuzn\u00e9 technologie dlouhou cestu. Zlep\u0161ili se v ur\u010dov\u00e1n\u00ed toho, co u\u017eivatel\u00e9 hledaj\u00ed, a poskytuj\u00ed relevantn\u011bj\u0161\u00ed odpov\u011bdi. V\u00fdsledkem je, \u017ee s touto technologi\u00ed je nyn\u00ed integrov\u00e1no st\u00e1le v\u00edce produkt\u016f orientovan\u00fdch na spot\u0159ebitele.<\/p>\n

I kdy\u017e to m\u016f\u017ee b\u00fdt u\u017eite\u010dn\u00e9 pro \u00fasporu \u010dasu a \u00fasil\u00ed p\u0159i hled\u00e1n\u00ed informac\u00ed nebo vytv\u00e1\u0159en\u00ed obsahu, je d\u016fle\u017eit\u00e9 pochopit, \u017ee AI m\u00e1 st\u00e1le ur\u010dit\u00e1 omezen\u00ed. Za prv\u00e9 a p\u0159edev\u0161\u00edm, modely um\u011bl\u00e9 inteligence maj\u00ed st\u00e1le tendenci halucinovat, co\u017e znamen\u00e1, \u017ee si ve snaze odpov\u011bd\u011bt na dotaz u\u017eivatele mohou vym\u00fd\u0161let fakta a data, kter\u00e1 nejsou pravdiv\u00e1.<\/p>\n

Nav\u00edc, jak je uvedeno v\u00fd\u0161e, i v p\u0159\u00edpadech, kdy si AI nevym\u00fd\u0161l\u00ed vlastn\u00ed fakta, m\u016f\u017ee z\u00edsk\u00e1vat sv\u00e9 informace odn\u011bkud, kter\u00e9 nejsou d\u016fv\u011bryhodn\u00e9. Op\u011bt to m\u016f\u017ee ovlivnit informace o u\u017eivateli, kdy\u017e se ve v\u00fdsledc\u00edch vyhled\u00e1v\u00e1n\u00ed objev\u00ed nepravdiv\u00e9 informace jako spr\u00e1vn\u00e1 odpov\u011b\u010f. Proto nyn\u00ed t\u00e9m\u011b\u0159 ka\u017ed\u00e1 spole\u010dnost ve sv\u00fdch n\u00e1stroj\u00edch AI zobrazuje varov\u00e1n\u00ed, \u017ee informace, kter\u00e9 AI poskytuje, nemus\u00ed b\u00fdt pravdiv\u00e9.<\/p>\n

Zat\u00edmco podivn\u00e9 odpov\u011bdi poskytovan\u00e9 p\u0159ehledy AI spole\u010dnosti Google mohou b\u00fdt z\u00e1bavn\u00e9 \u010d\u00edst, vyvol\u00e1v\u00e1 to v\u00e1\u017enou ot\u00e1zku o spolehlivosti model\u016f AI obecn\u011b. M\u016f\u017ee potenci\u00e1ln\u011b v\u00e9st k z\u00e1va\u017en\u00fdm chyb\u00e1m, pokud se osoba spol\u00e9h\u00e1 na nespr\u00e1vn\u00e9 informace poskytnut\u00e9 um\u011blou inteligenc\u00ed a nem\u016f\u017ee ur\u010dit, zda jsou spr\u00e1vn\u00e9 nebo ne.<\/p>\n

Krom\u011b toho Google u\u017eivatel\u016fm nyn\u00ed neumo\u017e\u0148uje \u00fapln\u011b vypnout p\u0159ehledy AI, tak\u017ee tato funkce zde z\u016fstane, co\u017e je dal\u0161\u00ed \u010d\u00e1st probl\u00e9mu. M\u016f\u017eete v\u0161ak p\u0159ej\u00edt do nastaven\u00ed sv\u00e9ho \u00fa\u010dtu Google a deaktivovat jej v Laborato\u0159i, jak je vysv\u011btleno v pr\u016fvodci n\u00ed\u017ee. Hled\u00e1n\u00ed odpov\u011bd\u00ed na r\u016fzn\u00fdch str\u00e1nk\u00e1ch m\u016f\u017ee b\u00fdt pomalej\u0161\u00ed, ale je m\u00e9n\u011b pravd\u011bpodobn\u00e9, \u017ee naraz\u00edte na vymy\u0161len\u00e1 fakta a podivn\u00e9 n\u00e1vrhy.<\/p>\n","protected":false},"excerpt":{"rendered":"

Technologick\u00e9 spole\u010dnosti se na AI zam\u011b\u0159uj\u00ed v\u00edce ne\u017e kdy p\u0159edt\u00edm, co\u017e vedlo k tomu, \u017ee se modely AI neust\u00e1le zlep\u0161uj\u00ed a zlep\u0161uj\u00ed v r\u016fzn\u00fdch \u00fakolech. Nap\u0159\u00edklad OpenAI ned\u00e1vno p\u0159edstavila sv\u016fj nejnov\u011bj\u0161\u00ed vlajkov\u00fd model GPT-4o, kter\u00fd dokonce dok\u00e1\u017ee detekovat a simulovat lidsk\u00e9 emoce. Nov\u00fd multimod\u00e1ln\u00ed model m\u016f\u017ee krom\u011b textov\u00fdch funkc\u00ed shroma\u017e\u010fovat data a poskytovat informace pomoc\u00ed […]<\/p>\n","protected":false},"author":1,"featured_media":0,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[1],"tags":[179,21,126],"class_list":["post-2201","post","type-post","status-publish","format-standard","hentry","category-how-to","tag-artificial-intelligence","tag-google","tag-microsoft"],"acf":[],"_links":{"self":[{"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/posts\/2201","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/comments?post=2201"}],"version-history":[{"count":1,"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/posts\/2201\/revisions"}],"predecessor-version":[{"id":2202,"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/posts\/2201\/revisions\/2202"}],"wp:attachment":[{"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/media?parent=2201"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/categories?post=2201"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/howtogeek.blog\/cs\/wp-json\/wp\/v2\/tags?post=2201"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}