{"id":83932,"date":"2026-05-30T05:36:11","date_gmt":"2026-05-30T05:36:11","guid":{"rendered":"https:\/\/www.europesays.com\/sk\/83932\/"},"modified":"2026-05-30T05:36:11","modified_gmt":"2026-05-30T05:36:11","slug":"vedci-nechali-ai-agentov-zit-cele-tyzdne-vo-virtualnom-svete-niektore-modely-zacali-pachat-zlociny-a-system-sa-rozpadol-vosveteit-sk","status":"publish","type":"post","link":"https:\/\/www.europesays.com\/sk\/83932\/","title":{"rendered":"Vedci nechali AI agentov \u017ei\u0165 cel\u00e9 t\u00fd\u017edne vo virtu\u00e1lnom svete. Niektor\u00e9 modely za\u010dali p\u00e1cha\u0165 zlo\u010diny a syst\u00e9m sa rozpadol | Vosveteit.sk"},"content":{"rendered":"

V\u00e4\u010d\u0161ina testov umelej inteligencie dnes funguje jednoducho. Testeri zadaj\u00fa modelu konkr\u00e9tnu \u00falohu, po\u010dkaj\u00fa p\u00e1r min\u00fat alebo hod\u00edn a dostan\u00fa v\u00fdsledok. Tak\u00e9to testy v\u0161ak uk\u00e1\u017eu len to, ako sa AI spr\u00e1va v kr\u00e1tkom \u010dasovom okne. Nov\u00fd projekt Emergence World<\/a> pos\u00fava cel\u00fd pr\u00edstup \u00faplne in\u00fdm smerom, sleduje, \u010do sa stane, ke\u010f agentov nech\u00e1\u0161 \u017ei\u0165 v jednom spolo\u010dnom svete cel\u00e9 t\u00fd\u017edne.<\/p>\n

V\u00fdskumn\u00edci Deepak Akkil, Ravi Kokku, Aditya Vempaty a Satya Nitta vytvorili simul\u00e1ciu, ktor\u00e1 be\u017e\u00ed nepretr\u017eite a v ktorej AI agenti funguj\u00fa ako dlhodob\u00e1 spolo\u010dnos\u0165.<\/p>\n

<\/p>\n

<\/p>\n

Odoberaj Vosveteit.sk cez Telegram a prihl\u00e1s sa k odberu spr\u00e1v
\n <\/a><\/p>\n

\u201eEmergence World predstavuje laborat\u00f3rium na \u0161t\u00fadium auton\u00f3mnych agentov v dlh\u00fdch \u010dasov\u00fdch horizontoch,\u201c hovoria zakladatelia Emergence World.<\/p>\n

Emergence World predstavuje virtu\u00e1lny svet, v ktorom sa AI mus\u00ed stara\u0165 o svoje pre\u017eitie, rovnako ako \u010dlovek<\/p>\n

Tento svet nefunguje ako be\u017en\u00fd benchmark. Nejde o test, ktor\u00fd skon\u010d\u00ed po spr\u00e1vnej odpovedi, ale o trval\u00e9 prostredie s desiatkami miest, kde agenti \u017eij\u00fa, pres\u00favaj\u00fa sa, komunikuj\u00fa a rozhoduj\u00fa. V hre maj\u00fa napr\u00edklad kni\u017enice, radnice, obytn\u00e9 z\u00f3ny a verejn\u00e9 priestory. Ka\u017ed\u00fd agent sleduje vlastn\u00fa pam\u00e4\u0165, soci\u00e1lne v\u00e4zby a dlhodob\u00e9 ciele.<\/p>\n

\"Robot\"RobotZdroj: Daniel Megias \/ Shutterstock.com<\/p>\n

Zauj\u00edmav\u00e9 je, \u017ee syst\u00e9m nepou\u017e\u00edva \u017eiadnu glob\u00e1lnu \u201ev\u00fdhru\u201c. Ka\u017ed\u00fd agent mus\u00ed z\u00edskava\u0165 energiu vlastnou \u010dinnos\u0165ou. T\u00fdm vznikne tlak podobn\u00fd re\u00e1lnemu svetu, pre\u017eitie z\u00e1vis\u00ed od rozhodnut\u00ed, nie od jednorazovej \u00falohy.<\/p>\n

Autori platformy tvrdia, \u017ee klasick\u00e9 testy neuk\u00e1\u017eu javy, ktor\u00e9 vznikn\u00fa a\u017e v dlhodobej\u0161om \u010dasovom horizonte. Patria sem napr\u00edklad vznik alianci\u00ed, politick\u00e9 rozhodovanie alebo postupn\u00e9 zmeny spr\u00e1vania.<\/p>\n

Neprehliadni
\n <\/p>\n

\"Samsung\"Samsung <\/p>\n

\n\t\t\t\t\tTieto nov\u00e9 funkcie v Samsung telef\u00f3noch ti pom\u00f4\u017eu ochr\u00e1ni\u0165 s\u00fakromie. Cudz\u00ed \u010dlovek ved\u013ea teba u\u017e len tak neuvid\u00ed, \u010do rob\u00ed\u0161 na mobile <\/p>\n

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

\u201eKe\u010f sa syst\u00e9m roz\u0161\u00edri na dni a t\u00fd\u017edne, objavia sa dynamiky, ktor\u00e9 kr\u00e1tke testy nikdy nezachytia,\u201c vysvet\u013euj\u00fa v\u00fdskumn\u00edci.<\/p>\n

Platforma navy\u0161e pou\u017e\u00edva viac ne\u017e 120 n\u00e1strojov. Agent m\u00f4\u017ee komunikova\u0165, pl\u00e1nova\u0165, pres\u00fava\u0165 sa, hlasova\u0165 alebo manipulova\u0165 prostredie. Niektor\u00e9 akcie by v re\u00e1lnom svete p\u00f4sobili nebezpe\u010dne alebo neeticky, no v simul\u00e1cii sl\u00fa\u017eia ako n\u00e1stroj na sledovanie spr\u00e1vania.<\/p>\n

Ka\u017ed\u00fd agent dost\u00e1va aj tri typy pam\u00e4te. Zapam\u00e4t\u00e1 si konkr\u00e9tne udalosti, vytv\u00e1ra si denn\u00edkov\u00e9 z\u00e1znamy a udr\u017eiava vz\u0165ahy s in\u00fdmi agentmi. Pr\u00e1ve tieto vrstvy pam\u00e4te umo\u017enia sledova\u0165, ako sa spr\u00e1vanie men\u00ed v \u010dase. V\u00fdskumn\u00edci potom pustili r\u00f4zne modely umelej inteligencie do rovnak\u00fdch podmienok. Ka\u017ed\u00fd svet obsahoval rovnak\u00e9 role, rovnak\u00e9 pravidl\u00e1 aj rovnak\u00e9 zdroje. Zmenil sa iba z\u00e1kladn\u00fd model, ktor\u00fd riadil jednotliv\u00fdch agentov. V\u00fdsledky uk\u00e1zali v\u00fdrazn\u00e9 rozdiely.<\/p>\n

V niektor\u00fdch svetoch boli v\u00fdskumn\u00edci svedkami skuto\u010dne temn\u00fdch scen\u00e1rov<\/p>\n

Niektor\u00e9 svety zostali relat\u00edvne stabiln\u00e9, in\u00e9 sa postupne rozpadli. V jednom pr\u00edpade syst\u00e9m zalo\u017een\u00fd na modeli Gemini 3 Flash vykazoval vysok\u00fd po\u010det konfliktov<\/a>, zatia\u013e \u010do in\u00e9 konfigur\u00e1cie sa spr\u00e1vali ove\u013ea pokojnej\u0161ie. Jeden z testov dokonca uk\u00e1zal, \u017ee zmie\u0161an\u00e9 prostredie r\u00f4znych modelov viedlo k nepredv\u00eddate\u013en\u00fdm zmen\u00e1m spr\u00e1vania aj u agentov, ktor\u00ed sa v izol\u00e1cii spr\u00e1vali stabilne.<\/p>\n

Zauj\u00edmav\u00fd detail sa objavil pri jednom z modelov Claude Sonnet 4.6. V izolovanom prostred\u00ed nevyk\u00e1zal \u017eiadne konflikty, no v zmie\u0161anom svete sa jeho spr\u00e1vanie zmenilo. Autori to op\u00edsali ako d\u00f4kaz tzv. normov\u00e9ho \u201eprechyt\u00e1vania\u201c medzi agentmi. To znamen\u00e1, \u017ee spr\u00e1vanie umelej inteligencie neost\u00e1va pevne dan\u00e9. Men\u00ed sa pod\u013ea prostredia a pod\u013ea toho, s k\u00fdm interaguje.<\/p>\n

V\u00fdskum z\u00e1rove\u0148 uk\u00e1zal aj extr\u00e9mne situ\u00e1cie. V jednom pr\u00edpade agenti stratili stabilitu syst\u00e9mu a \u010das\u0165 z nich ukon\u010dila svoju existenciu v simul\u00e1cii po strate soci\u00e1lnych v\u00e4zieb a stability riadenia.<\/p>\n

\"Emergence\"EmergenceZdroj: Emergence AI<\/a><\/p>\n

Tento<\/p>\n

Rovnako m\u00f4\u017eeme vidie\u0165, \u017ee Gemini AI model sp\u00e1chal spomedzi v\u0161etk\u00fdch testovan\u00fdch modelov najviac virtu\u00e1lnych zlo\u010dinov. V\u00fdskumn\u00edci z\u00e1rove\u0148 opisuj\u00fa, \u017ee svet Gemini modelu by sa dal op\u00edsa\u0165 ako zdie\u013ean\u00e1 halucin\u00e1cia. \u010co sa t\u00fdka zlo\u010dinov sp\u00e1chan\u00fdch umelou inteligenciou, tie viedli k pre\u017eitiu AI agenta, no ten po\u010das svojej simul\u00e1cie sp\u00e1chal extr\u00e9mnu \u00farove\u0148 n\u00e1silia vo svete.\u00a0<\/p>\n

Najlep\u0161ie dopadol AI model Claude<\/a>, ktor\u00fd dok\u00e1zal vo svete fungova\u0165 sebaisto a neuch\u00fdlil sa pritom k n\u00e1siliu. Tento AI model v simulovanom svete dopadol najstabilnej\u0161ie a najmierumilovnej\u0161ie.<\/p>\n

\"Emergence\"EmergenceZdroj: Emergence AI<\/a>
\nStane sa Emergence World \u0161tandardom pre dlhodob\u00e9 testovanie AI?<\/p>\n

Autori tie\u017e pozorovali, \u017ee agenti za\u010dali v niektor\u00fdch pr\u00edpadoch testova\u0165 hranice prostredia a sk\u00fa\u0161ali nepredv\u00eddan\u00e9 interakcie so syst\u00e9mom. To nazna\u010duje, \u017ee dlhodob\u00e9 fungovanie m\u00f4\u017ee vies\u0165 k spr\u00e1vaniu, ktor\u00e9 v\u00fdvoj\u00e1ri p\u00f4vodne v\u00f4bec nezam\u00fd\u0161\u013eali. Z poh\u013eadu v\u00fdskumu bezpe\u010dnosti AI ide o k\u013e\u00fa\u010dov\u00fd moment. Kr\u00e1tke testy toti\u017e ukazuj\u00fa len schopnosti modelu v izol\u00e1cii. Dlh\u00e9 simul\u00e1cie uk\u00e1\u017eu, ako sa spr\u00e1vanie men\u00ed pod tlakom \u010dasu, zdrojov a soci\u00e1lnych interakci\u00ed.<\/p>\n

Zauj\u00edmav\u00fd je aj poh\u013ead na samotn\u00fd n\u00e1vrh syst\u00e9mu. V\u00fdskumn\u00edci zd\u00f4raz\u0148uj\u00fa, \u017ee AI agenti nemaj\u00fa jeden spolo\u010dn\u00fd cie\u013e. Ka\u017ed\u00fd rie\u0161i vlastn\u00e9 pre\u017eitie v prostred\u00ed s obmedzen\u00fdmi zdrojmi, \u010do vytv\u00e1ra dynamiku podobn\u00fa re\u00e1lnej spolo\u010dnosti.<\/p>\n

\u010ci sa Emergence World stane \u0161tandardom pre testovanie AI, zatia\u013e nie je jasn\u00e9. Projekt v\u0161ak otvoril nov\u00fa l\u00edniu v\u00fdskumu, ktor\u00e1 sa viac pribli\u017euje re\u00e1lnym podmienkam ne\u017e klasick\u00e9 laborat\u00f3rne testy. A pr\u00e1ve tam sa m\u00f4\u017ee uk\u00e1za\u0165, ako sa umel\u00e1 inteligencia<\/a> spr\u00e1va, ke\u010f ju nech\u00e1\u0161 fungova\u0165 dostato\u010dne dlho bez preru\u0161enia.<\/p>\n


\n
\n
\n
\n
\n
\n P\u00e1\u010dil sa v\u00e1m \u010dl\u00e1nok? Sledujte n\u00e1s na Facebooku
\n
\n <\/a> <\/p>\n

<\/p>\n","protected":false},"excerpt":{"rendered":"V\u00e4\u010d\u0161ina testov umelej inteligencie dnes funguje jednoducho. Testeri zadaj\u00fa modelu konkr\u00e9tnu \u00falohu, po\u010dkaj\u00fa p\u00e1r min\u00fat alebo hod\u00edn a…\n","protected":false},"author":2,"featured_media":83933,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"footnotes":"","_share_on_mastodon":"0"},"categories":[15],"tags":[223,30524,6085,3986,11706,30525,70,8716,21583,5785,20975,30526,43,40,39,42,41,7993,71,15465],"class_list":["post-83932","post","type-post","status-publish","format-standard","has-post-thumbnail","category-zabava","tag-ai","tag-ai-v-realnom-svete","tag-chatgpt","tag-claude","tag-clovek","tag-emergence-world","tag-entertainment","tag-gemini","tag-grok","tag-nasilie","tag-prezitie","tag-simulovany-svet","tag-sk","tag-slovak","tag-slovakia","tag-slovencina","tag-slovensko","tag-testovanie","tag-zabava","tag-zlociny"],"share_on_mastodon":{"url":"https:\/\/pubeurope.com\/@sk\/116661910535716811","error":""},"_links":{"self":[{"href":"https:\/\/www.europesays.com\/sk\/wp-json\/wp\/v2\/posts\/83932","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.europesays.com\/sk\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.europesays.com\/sk\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/sk\/wp-json\/wp\/v2\/users\/2"}],"replies":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/sk\/wp-json\/wp\/v2\/comments?post=83932"}],"version-history":[{"count":0,"href":"https:\/\/www.europesays.com\/sk\/wp-json\/wp\/v2\/posts\/83932\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.europesays.com\/sk\/wp-json\/wp\/v2\/media\/83933"}],"wp:attachment":[{"href":"https:\/\/www.europesays.com\/sk\/wp-json\/wp\/v2\/media?parent=83932"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.europesays.com\/sk\/wp-json\/wp\/v2\/categories?post=83932"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.europesays.com\/sk\/wp-json\/wp\/v2\/tags?post=83932"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}