{"id":27911,"date":"2023-08-17T11:40:48","date_gmt":"2023-08-17T09:40:48","guid":{"rendered":"https:\/\/kinit.sk\/research-visit-at-the-university-of-sheffield-spreading-and-combating-false-information-with-large-language-models\/"},"modified":"2023-08-17T11:49:31","modified_gmt":"2023-08-17T09:49:31","slug":"vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely","status":"publish","type":"post","link":"https:\/\/kinit.sk\/sk\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\/","title":{"rendered":"V\u00fdskumn\u00e1 n\u00e1v\u0161teva na Univerzite v Sheffielde: Falo\u0161n\u00e9 inform\u00e1cie a ve\u013ek\u00e9 jazykov\u00e9 modely"},"content":{"rendered":"
\n

Ivan Srba<\/em><\/a> p\u00f4sob\u00ed ako \u010dlen <\/em>t\u00edmu spracovania webov\u00fdch a pou\u017e\u00edvate\u013esk\u00fdch \u00fadajov<\/em><\/a>, kde sa \u0161pecializuje na rie\u0161enie probl\u00e9mov \u0161kodliv\u00fdch inform\u00e1ci\u00ed prostredn\u00edctvom umelej inteligencie. V j\u00fani 2023 dostal pr\u00edle\u017eitos\u0165 absolvova\u0165 v\u00fdskumn\u00fa n\u00e1v\u0161tevu na <\/em>Univerzite v Sheffielde<\/em><\/a>. \u010citajte \u010falej a dozviete sa viac o jeho sk\u00fasenostiach.<\/em><\/p>\n<\/div>\n\n

\n

Vznik ve\u013ek\u00fdch predtr\u00e9novan\u00fdch jazykov\u00fdch modelov (s po\u010dtom parametrov od stoviek mili\u00f3nov, ako je napr\u00edklad RoBERTa, a\u017e po najnov\u0161ie modely s po\u010dtom parametrov v stovk\u00e1ch mili\u00e1rd, ako je ChatGPT) pril\u00e1kal zna\u010dn\u00fa pozornos\u0165 v\u00fdskumnej komunity aj \u0161irokej verejnosti.<\/p>\n<\/div>\n\n

\n

Medzi mnoh\u00fdmi obavami t\u00fdkaj\u00facimi sa praktick\u00e9ho vyu\u017eitia t\u00fdchto modelov sa \u010dasto opakuje jedna konkr\u00e9tna hroba. Tieto modely m\u00f4\u017eu by\u0165 zneu\u017eit\u00e9 na automatick\u00e9 generovanie textu, ktor\u00fd nesk\u00f4r m\u00f4\u017ee sl\u00fa\u017ei\u0165 na \u0161\u00edrenie existuj\u00facich alebo \u00faplne nov\u00fdch falo\u0161n\u00fdch inform\u00e1ci\u00ed. Na druhej strane v\u0161ak maj\u00fa tieto modely ve\u013ek\u00fd potenci\u00e1l prispie\u0165 k boju proti falo\u0161n\u00fdm inform\u00e1ci\u00e1m. <\/strong>Tento potenci\u00e1l v\u0161ak zatia\u013e zost\u00e1va nepresk\u00faman\u00fd.<\/p>\n<\/div>\n\n

\n

Po\u010das m\u00f4jho v\u00fdskumn\u00e9ho pobytu v t\u00edme GATE NLP na Univerzite v Sheffielde (s podporou programu SoBigData++<\/a> TNA<\/a>) som mal v\u00fdnimo\u010dn\u00fa pr\u00edle\u017eitos\u0165 zhodnoti\u0165 potenci\u00e1l ve\u013ek\u00fdch jazykov\u00fdch modelov v kontexte \u0161\u00edrenia a boja proti dezinform\u00e1ci\u00e1m. Z\u00e1rove\u0148 som sa venoval mo\u017enostiam ich vyu\u017eitia pri automatickom hodnoten\u00ed d\u00f4veryhodnosti online obsahu.<\/p>\n<\/div>\n\n

\n

V tomto \u010dl\u00e1nku op\u00ed\u0161em dosiahnut\u00e9 v\u00fdsledky a osobn\u00e9 sk\u00fasenosti z n\u00e1v\u0161tevy na Univerzite v Sheffielde a spolupr\u00e1ce s tamoj\u0161\u00edm v\u00fdskumn\u00fdm t\u00edmom pre spracovanie prirodzen\u00e9ho jazyka (angl. Natural Language Processing – NLP).<\/p>\n<\/div>\n\n

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

Ve\u013ek\u00e9 jazykov\u00e9 modely a falo\u0161n\u00e9 inform\u00e1cie<\/h2>\n<\/div>\n\n
\n

Najprv som vytvoril sumariz\u00e1ciu s\u00fa\u010dasn\u00fdch poznatkov o tejto problematike, aby som z\u00edskal komplexn\u00fd obraz o ve\u013ek\u00fdch jazykov\u00fdch modelov a ich vz\u0165ahu k falo\u0161n\u00fdm inform\u00e1ci\u00e1m (konkr\u00e9tne dezinform\u00e1ci\u00e1m a manipul\u00e1ci\u00e1m). Vych\u00e1dzal som z existuj\u00facich pr\u00e1c a z na\u0161ich vlastn\u00fdch v\u00fdskumn\u00fdch aktiv\u00edt v KInITe, ako aj zo minul\u00fdch\/aktu\u00e1lnych v\u00fdskumn\u00fdch aktiv\u00edt GATE t\u00edmu na Univerzite v Sheffielde.<\/p>\n<\/div>\n\n

\n

Z\u00edskan\u00fd preh\u013ead najnov\u0161\u00edch v\u00fdsledkov potvrdzuje, \u017ee rozpoznan\u00e9 hrozby t\u00fdkaj\u00face sa schopnosti t\u00fdchto modelov generova\u0165 dezinform\u00e1cie nie s\u00fa zanedbate\u013en\u00e9. Ve\u013ek\u00e9 jazykov\u00e9 modely dok\u00e1\u017eu generova\u0165 vysoko kvalitn\u00e9 texty, ktor\u00e9 s\u00fa pre \u013eud\u00ed \u0165a\u017eko rozoznate\u013en\u00e9<\/strong>. Okrem toho, ako sme uk\u00e1zali v r\u00e1mci prebiehaj\u00facej pr\u00e1ce na projekte vera.ai,<\/a> tieto modely dok\u00e1\u017eu generova\u0165 aj ve\u013emi presved\u010div\u00e9 dezinforma\u010dn\u00e9 texty<\/strong> (novinov\u00e9 \u010dl\u00e1nky, pr\u00edspevky na soci\u00e1lnych m\u00e9di\u00e1ch, koment\u00e1re) s nov\u00fdmi (\u00faplne vymyslen\u00fdmi) argumentami. Aktu\u00e1lne techniky, ktor\u00e9 s\u00fa vyu\u017e\u00edvan\u00e9 s cie\u013eom zamedzi\u0165 \u0161\u00edreniu tak\u00fdchto textov (ako napr. bezpe\u010dnostn\u00e9 filtre, angl. safety filters) sa uk\u00e1zali ako nedostato\u010dn\u00e9.<\/p>\n<\/div>\n\n

\n

Pozit\u00edvnou spr\u00e1vou je, \u017ee tak ako sme uk\u00e1zali v ned\u00e1vnom v\u00fdskume v r\u00e1mci projektu VIGILANT<\/a>, detekcia umelo generovan\u00e9ho textu je mo\u017en\u00e1 a to aj pre text v r\u00f4znych jazykoch, pri\u010dom \u00faspe\u0161nos\u0165 detekcie sa l\u00ed\u0161i pre r\u00f4zne jazyky, gener\u00e1tory textu a samotn\u00e9 detektory.<\/p>\n<\/div>\n\n

\n

Ve\u013ek\u00e9 jazykov\u00e9 modely maj\u00fa z\u00e1rove\u0148 ve\u013ek\u00fd potenci\u00e1l adresova\u0165\u00a0 r\u00f4zne probl\u00e9my, vr\u00e1tane t\u00fdch, ktor\u00e9 s\u00favisia s dezinform\u00e1ciami.<\/strong> M\u00f4\u017eu sl\u00fa\u017ei\u0165 ako zdroj slab\u00fdch sign\u00e1lov (angl. weak signals), hodnoti\u0165 d\u00f4veryhodnos\u0165 zdrojov alebo dokonca poskytova\u0165 vysvetlenia koncov\u00fdm pou\u017e\u00edvate\u013eom. Tento potenci\u00e1l je v\u0161ak preva\u017ene nepresk\u00faman\u00fd. Pr\u00e1ve preto sa stal centrom m\u00f4jho \u010fal\u0161ieho sk\u00famania po\u010das v\u00fdskumnej n\u00e1v\u0161tevy.<\/strong><\/p>\n<\/div>\n\n

\n
\"\"
Porovnanie ve\u013ekosti ve\u013ek\u00fdch jazykov\u00fdch modelov (zdroj<\/a>). Ve\u013ekos\u0165 modelu GPT-4 nie je v s\u00fa\u010dasnosti verejne dostupn\u00e1, ale predpoklad\u00e1 sa, \u017ee bude ma\u0165 aspo\u0148 1 miliardu parametrov.<\/figcaption><\/figure>\n<\/div>\n\n
\n

Klasifik\u00e1cia textu jazykov\u00fdmi modelmi<\/h3>\n<\/div>\n\n
\n

Za\u010diatkom roka 2017 priniesli jazykov\u00e9 modely nov\u00fd sp\u00f4sob rie\u0161enia \u00faloh spracovania prirodzen\u00e9ho jazyka (NLP), ozna\u010dovan\u00fd ako  \u201cpre-train, fine-tune\u201d. V tomto pr\u00edstupe je jazykov\u00fd model najsk\u00f4r predtr\u00e9novan\u00fd na ve\u013ekom mno\u017estve surov\u00fdch d\u00e1t a potom dotr\u00e9novan\u00fd na konkr\u00e9tnu \u00falohu. S najnov\u0161\u00edmi predtr\u00e9novan\u00fdmi ve\u013ek\u00fdmi jazykov\u00fdmi modelmi sa objavila \u010fal\u0161ia paradigma ozna\u010dovan\u00e1 ako \u201cpre-train, prompt, and predict\u201d. V tejto paradigme je konkr\u00e9tna \u00faloha rie\u0161en\u00e1 pomocou promptov – pri\u010dom techniky na n\u00e1jdenie najlep\u0161\u00edch promptov sa ozna\u010duje ako \u201cprompt engineering\u201d. Konkr\u00e9tnym pr\u00edkladom in\u017einierstva promptov je tzv. in-context learning, kde je jazykov\u00fd model dotr\u00e9novan\u00fd nieko\u013ek\u00fdmi oanotovan\u00fdch vzorkami, ktor\u00e9 s\u00fa s\u00fa\u010das\u0165ou samotn\u00e9ho promptu.<\/p>\n<\/div>\n\n

\n

Obidve tieto paradigmy maj\u00fa ve\u013ek\u00fd potenci\u00e1l pom\u00f4c\u0165 v boji proti dezinform\u00e1ci\u00e1m. Konkr\u00e9tne som sa zameriaval na v\u00fdskum toho, ako ich m\u00f4\u017eeme vyu\u017ei\u0165 na hodnotenie d\u00f4veryhodnosti obsahu.<\/strong> Po\u010das v\u00fdskumnej n\u00e1v\u0161tevy sme nadviazali na na\u0161e predch\u00e1dzaj\u00face v\u00fdsledky v r\u00e1mci d\u00e1tovej v\u00fdzvy SemEval Task 3<\/a>, kde t\u00edm GATE navrhol najlep\u0161ie rie\u0161enie<\/a> pre pod\u00falohy 1 a 2 (Kategoriz\u00e1cia \u017e\u00e1nrov a Detekcia t\u00e9matick\u00fdch kateg\u00f3ri\u00ed novinov\u00fdch \u010dl\u00e1nkov). Ja som bol zas \u010dlenom t\u00edmu KInIT, v ktorom sme navrhli rie\u0161enie<\/a>, ktor\u00e9 dosiahlo najlep\u0161ie v\u00fdsledky v pod\u00falohe 3 (Detekcia techn\u00edk presvied\u010dania), pri\u010dom sme obsadili 6 prv\u00fdch miest z 9 jazykov.<\/a><\/p>\n<\/div>\n\n

\n

Detekcia techn\u00edk presvied\u010dania s vyu\u017eit\u00edm promptov a in-context u\u010denia<\/h3>\n<\/div>\n\n
\n

Najsk\u00f4r sme vykonali viacero experimentov s vyu\u017eit\u00edm promptov a in-context u\u010denia na z\u00e1klade d\u00e1t zo s\u00fa\u0165a\u017ee SemEval a tie\u017e d\u00e1t z podobnej s\u00fa\u0165a\u017ee DIPROMATS<\/a>. Experimentovali sme s jednoduchou bin\u00e1rnou verziou \u00falohy (predpoveda\u0165, \u010di vstupn\u00fd text obsahuje presvied\u010daj\u00face techniky alebo nie) a taktie\u017e s p\u00f4vodnou verziou \u00falohy (predpoveda\u0165 pr\u00edtomnos\u0165 jednotliv\u00fdch techn\u00edk presvied\u010dania). Experimenty boli uskuto\u010dnen\u00e9 na ve\u013ekom jazykovom modely od OpenAI (ChatGPT) a tie\u017e ve\u013ekom jazykovom modely zalo\u017eenom na LLaMA (LLaMA Vicuna). Zvl\u00e1\u0161tny d\u00f4raz sme kl\u00e1dli na n\u00e1hodnos\u0165 predikci\u00ed – experimentovali sme s r\u00f4znym po\u010dtom, porad\u00edm, technikami v\u00fdberu a distrib\u00faciou zna\u010diek in-context pr\u00edkladov, aby sme sk\u00famali najvhodnej\u0161ie form\u00e1ty promptov.<\/p>\n<\/div>\n\n

\n

V\u00fdsledky experimentov uk\u00e1zali, \u017ee ani prompting, ani in-context u\u010denie nedok\u00e1\u017eu prekona\u0165 dotr\u00e9novan\u00fd model (angl. fine-tuned model).<\/strong> D\u00f4vodom je, \u017ee obe sady d\u00e1t (SemEval a DIPROMATS) obsahuj\u00fa pomerne ve\u013ek\u00e9 mno\u017estvo ozna\u010dkovan\u00fdch pr\u00edkladov. Tieto tr\u00e9novacie sady umo\u017e\u0148uj\u00fa viacjazy\u010dn\u00e9mu modelu (XLM-RoBERTaLARGE<\/sub><\/a>) dostato\u010dn\u00e9 dotr\u00e9novanie pre konkr\u00e9tnu \u00falohu, a teda najnov\u0161ie jazykov\u00e9 modely funguj\u00face v re\u017eime bez poskytnutia ozna\u010dkovan\u00fdch pr\u00edkladov (angl. zero-shot) alebo len s nieko\u013ek\u00fdmi pr\u00edkladmi (angl. few-shot) nedok\u00e1\u017eu dosiahnu\u0165 porovnate\u013en\u00fd v\u00fdkon.<\/p>\n<\/div>\n\n

\n

Toto je zauj\u00edmav\u00e9 zistenie, ktor\u00e9 si zasl\u00fa\u017ei \u010fal\u0161ie presk\u00famanie. Je mo\u017en\u00e9 vykona\u0165 \u010fal\u0161ie experimenty na r\u00f4znych \u00faloh\u00e1ch spracovania prirodzen\u00e9ho jazyka a na viacer\u00fdch d\u00e1tov\u00fdch sad\u00e1ch, aby sme zistili, \u010di sa potvrd\u00ed rovnak\u00e9 pozorovanie. Ak \u00e1no, potom bude d\u00f4le\u017eit\u00e9 presk\u00fama\u0165, \u010di existuje ur\u010dit\u00fd bod zlomu v po\u010dte tr\u00e9novac\u00edch vzoriek, pri ktorom dotr\u00e9novanie prev\u00fd\u0161i prompting a in-context u\u010denie najnov\u0161\u00edch ve\u013ek\u00fdch jazykov\u00fdch modelov.<\/p>\n<\/div>\n\n

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

Efekt\u00edvnej\u0161ie ladenie s pomocou adapt\u00e9rov<\/h3>\n<\/div>\n\n
\n

V\u00fdsledky predch\u00e1dzaj\u00faceho v\u00fdskumu uk\u00e1zali, \u017ee v pr\u00edpade dostato\u010dne ve\u013ek\u00e9ho tr\u00e9novacieho s\u00faboru dotr\u00e9novanie (angl. fine-tuning) prekon\u00e1va prompting a in-context u\u010denie. Jemn\u00e9 ladenie ve\u013ek\u00fdch jazykov\u00fdch modelov v\u0161ak m\u00f4\u017ee ma\u0165 pomerne vysok\u00e9 v\u00fdpo\u010dtov\u00e9 n\u00e1roky.<\/p>\n<\/div>\n\n

\n

V tejto s\u00favislosti som v \u00fazkej spolupr\u00e1ci s GATE t\u00edmom porovnal v\u00fdkon dotr\u00e9novan\u00fdch\u00a0 jazykov\u00fdch modelov (ktor\u00e9 sme p\u00f4vodne pou\u017eili na d\u00e1tach SemEval) s v\u00fdkonom modelov natr\u00e9novan\u00fdch technikou \u201clow-rank adaptation\u201d (LoRA<\/a>) a tzv. adapt\u00e9rmi pre ve\u013ek\u00e9 jazykov\u00e9 modely. Obe techniky maj\u00fa za cie\u013e poskytn\u00fa\u0165 parametrovo efekt\u00edvne dotr\u00e9novanie ve\u013ek\u00fdch jazykov\u00fdch modelov s minim\u00e1lnou stratou v\u00fdkonu a v niektor\u00fdch pr\u00edpadoch dokonca dosahova\u0165 lep\u0161ie v\u00fdsledky v porovnan\u00ed s plne dotr\u00e9novan\u00fdmi modelmi.<\/p>\n<\/div>\n\n

\n

Dosiahnut\u00e9 v\u00fdsledke preuk\u00e1zali, \u017ee v niektor\u00fdch situ\u00e1ci\u00e1ch s\u00fa LoRA a adapt\u00e9ry schopn\u00e9 prekona\u0165 v\u00fdkon plne dotr\u00e9novan\u00e9 modely.<\/strong> Experimenty s d\u00e1tami zo s\u00fa\u0165a\u017ee SemEval uk\u00e1zali, ako jednotliv\u00e9 techniky performuj\u00fa na \u00falohach s r\u00f4znymi charakteristikami (d\u013a\u017eka vstupu, po\u010det tried, celkov\u00e1 n\u00e1ro\u010dnos\u0165 klasifik\u00e1cie) a r\u00f4znymi viacjazy\u010dn\u00fdmi nastaveniami (tr\u00e9novanie na p\u00f4vodn\u00fdch d\u00e1tach, a na anglick\u00fdch d\u00e1ta s\/bez prekladu), a tie\u017e sme porovnali v\u00fdkon pre jednotliv\u00e9 jazyky.<\/p>\n<\/div>\n\n

\n

V\u0161eobecne m\u00f4\u017eeme kon\u0161tatova\u0165, \u017ee na\u0161e zistenia s\u00fa v s\u00falade s existuj\u00facimi v\u00fdskumn\u00fdmi pr\u00e1cami, ktor\u00e9 poukazuj\u00fa na to, \u017ee nedoch\u00e1dza k v\u00fdrazn\u00e9mu poklesu v\u00fdkonu alebo sa dokonca m\u00f4\u017ee v\u00fdkon zv\u00fd\u0161i\u0165 pri v\u00fdznamnej redukcii v\u00fdpo\u010dtov\u00fdch n\u00e1kladov. Okrem toho sme t\u00fdmito podrobn\u00fdmi porovnaniami v\u00fdznamne roz\u0161\u00edrili u\u017e existuj\u00face pr\u00e1ce.<\/p>\n<\/div>\n\n

\n

Program SoBigData++ TNA<\/h3>\n<\/div>\n\n
\n

T\u00e1to pr\u00e1ca vznikla v r\u00e1mci v\u00fdskumnej n\u00e1v\u0161tevy, ktor\u00fa podporil program SoBigData++ Transnational Access (TNA)<\/a>. Po\u010das tejto n\u00e1v\u0161tevy som pod veden\u00edm profesorky Kaliny Bontcheva<\/a> spolupracoval s GATE NLP<\/a> t\u00edmom na Univerzite v Sheffielde.<\/p>\n<\/div>\n\n

\n

Pobyt mi poskytol jedine\u010dn\u00fa pr\u00edle\u017eitos\u0165 prezentova\u0165 v\u00fdsledky svojej pr\u00e1ce na semin\u00e1roch t\u00edmu GATE, ako aj diskutova\u0165 a vykon\u00e1va\u0165 v\u00fdskumn\u00e9 experimenty s \u010dlenmi tohto t\u00edmu. Taktie\u017e som sa podie\u013eal na spracovan\u00ed dosiahnut\u00fdch v\u00fdsledkov vo forme spolo\u010dn\u00e9ho \u010dl\u00e1nku v odbornom \u010dasopise.<\/p>\n<\/div>\n\n

\n

Na z\u00e1ver m\u00f4\u017eem poveda\u0165, \u017ee mi t\u00e1to v\u00fdskumn\u00e1 n\u00e1v\u0161teva umo\u017enila zdokonali\u0165 moje znalosti v oblasti spracovania prirodzen\u00e9ho jazyka (NLP), posilni\u0165 moju medzin\u00e1rodn\u00fa v\u00fdskumn\u00fa sie\u0165 kontaktov a nabra\u0165 sk\u00fasenosti z metodologick\u00fdch, organiza\u010dn\u00fdch a mana\u017e\u00e9rskych schopnost\u00ed GATE t\u00edmu.<\/strong><\/p>\n<\/div>","protected":false},"excerpt":{"rendered":"

Ivan Srba p\u00f4sob\u00ed ako \u010dlen t\u00edmu spracovania webov\u00fdch a pou\u017e\u00edvate\u013esk\u00fdch \u00fadajov, kde sa \u0161pecializuje na rie\u0161enie probl\u00e9mov \u0161kodliv\u00fdch inform\u00e1ci\u00ed prostredn\u00edctvom umelej inteligencie. V j\u00fani 2023 dostal pr\u00edle\u017eitos\u0165 absolvova\u0165 v\u00fdskumn\u00fa n\u00e1v\u0161tevu…<\/p>\n","protected":false},"author":26,"featured_media":27907,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"","format":"standard","meta":{"_acf_changed":false,"footnotes":""},"categories":[83,142],"tags":[111],"class_list":["post-27911","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-news-sk","category-2023-sk","tag-disinformation-sk"],"acf":[],"yoast_head":"\nV\u00fdskumn\u00e1 n\u00e1v\u0161teva na Univerzite v Sheffielde: Falo\u0161n\u00e9 inform\u00e1cie a ve\u013ek\u00e9 jazykov\u00e9 modely - KInIT<\/title>\n<meta name=\"description\" content=\"Ivan Srba p\u00f4sob\u00ed ako \u010dlen t\u00edmu spracovania webov\u00fdch a pou\u017e\u00edvate\u013esk\u00fdch \u00fadajov, kde sa \u0161pecializuje na rie\u0161enie probl\u00e9mov \u0161kodliv\u00fdch inform\u00e1ci\u00ed prostredn\u00edctvom umelej inteligencie. V j\u00fani 2023 dostal pr\u00edle\u017eitos\u0165 absolvova\u0165 v\u00fdskumn\u00fa n\u00e1v\u0161tevu na Univerzite v Sheffielde. \u010citajte \u010falej a dozviete sa viac o jeho sk\u00fasenostiach.\" \/>\n<meta name=\"robots\" content=\"index, follow, max-snippet:-1, max-image-preview:large, max-video-preview:-1\" \/>\n<link rel=\"canonical\" href=\"https:\/\/kinit.sk\/sk\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\/\" \/>\n<meta property=\"og:locale\" content=\"sk_SK\" \/>\n<meta property=\"og:type\" content=\"article\" \/>\n<meta property=\"og:title\" content=\"V\u00fdskumn\u00e1 n\u00e1v\u0161teva na Univerzite v Sheffielde: Falo\u0161n\u00e9 inform\u00e1cie a ve\u013ek\u00e9 jazykov\u00e9 modely - KInIT\" \/>\n<meta property=\"og:description\" content=\"Ivan Srba p\u00f4sob\u00ed ako \u010dlen t\u00edmu spracovania webov\u00fdch a pou\u017e\u00edvate\u013esk\u00fdch \u00fadajov, kde sa \u0161pecializuje na rie\u0161enie probl\u00e9mov \u0161kodliv\u00fdch inform\u00e1ci\u00ed prostredn\u00edctvom umelej inteligencie. V j\u00fani 2023 dostal pr\u00edle\u017eitos\u0165 absolvova\u0165 v\u00fdskumn\u00fa n\u00e1v\u0161tevu na Univerzite v Sheffielde. \u010citajte \u010falej a dozviete sa viac o jeho sk\u00fasenostiach.\" \/>\n<meta property=\"og:url\" content=\"https:\/\/kinit.sk\/sk\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\/\" \/>\n<meta property=\"og:site_name\" content=\"KInIT\" \/>\n<meta property=\"article:published_time\" content=\"2023-08-17T09:40:48+00:00\" \/>\n<meta property=\"article:modified_time\" content=\"2023-08-17T09:49:31+00:00\" \/>\n<meta property=\"og:image\" content=\"https:\/\/kinit.sk\/wp-content\/uploads\/2023\/08\/202308_sheffield_feature_image.png\" \/>\n\t<meta property=\"og:image:width\" content=\"1200\" \/>\n\t<meta property=\"og:image:height\" content=\"628\" \/>\n\t<meta property=\"og:image:type\" content=\"image\/png\" \/>\n<meta name=\"author\" content=\"Marianna Palkova\" \/>\n<meta name=\"twitter:card\" content=\"summary_large_image\" \/>\n<meta name=\"twitter:creator\" content=\"@kinit\" \/>\n<meta name=\"twitter:site\" content=\"@kinit\" \/>\n<meta name=\"twitter:label1\" content=\"Autor\" \/>\n\t<meta name=\"twitter:data1\" content=\"Marianna Palkova\" \/>\n\t<meta name=\"twitter:label2\" content=\"Predpokladan\u00fd \u010das \u010d\u00edtania\" \/>\n\t<meta name=\"twitter:data2\" content=\"6 min\u00fat\" \/>\n<script type=\"application\/ld+json\" class=\"yoast-schema-graph\">{\"@context\":\"https:\\\/\\\/schema.org\",\"@graph\":[{\"@type\":\"Article\",\"@id\":\"https:\\\/\\\/kinit.sk\\\/sk\\\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\\\/#article\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/kinit.sk\\\/sk\\\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\\\/\"},\"author\":{\"name\":\"Marianna Palkova\",\"@id\":\"https:\\\/\\\/kinit.sk\\\/#\\\/schema\\\/person\\\/8b175aaaf3267b5bbbbb97e4a6db8cea\"},\"headline\":\"V\u00fdskumn\u00e1 n\u00e1v\u0161teva na Univerzite v Sheffielde: Falo\u0161n\u00e9 inform\u00e1cie a ve\u013ek\u00e9 jazykov\u00e9 modely\",\"datePublished\":\"2023-08-17T09:40:48+00:00\",\"dateModified\":\"2023-08-17T09:49:31+00:00\",\"mainEntityOfPage\":{\"@id\":\"https:\\\/\\\/kinit.sk\\\/sk\\\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\\\/\"},\"wordCount\":1779,\"image\":{\"@id\":\"https:\\\/\\\/kinit.sk\\\/sk\\\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/kinit.sk\\\/wp-content\\\/uploads\\\/2023\\\/08\\\/202308_sheffield_feature_image.png\",\"keywords\":[\"disinformation\"],\"articleSection\":[\"News\",\"2023\"],\"inLanguage\":\"sk-SK\"},{\"@type\":\"WebPage\",\"@id\":\"https:\\\/\\\/kinit.sk\\\/sk\\\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\\\/\",\"url\":\"https:\\\/\\\/kinit.sk\\\/sk\\\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\\\/\",\"name\":\"V\u00fdskumn\u00e1 n\u00e1v\u0161teva na Univerzite v Sheffielde: Falo\u0161n\u00e9 inform\u00e1cie a ve\u013ek\u00e9 jazykov\u00e9 modely - KInIT\",\"isPartOf\":{\"@id\":\"https:\\\/\\\/kinit.sk\\\/#website\"},\"primaryImageOfPage\":{\"@id\":\"https:\\\/\\\/kinit.sk\\\/sk\\\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\\\/#primaryimage\"},\"image\":{\"@id\":\"https:\\\/\\\/kinit.sk\\\/sk\\\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\\\/#primaryimage\"},\"thumbnailUrl\":\"https:\\\/\\\/kinit.sk\\\/wp-content\\\/uploads\\\/2023\\\/08\\\/202308_sheffield_feature_image.png\",\"datePublished\":\"2023-08-17T09:40:48+00:00\",\"dateModified\":\"2023-08-17T09:49:31+00:00\",\"author\":{\"@id\":\"https:\\\/\\\/kinit.sk\\\/#\\\/schema\\\/person\\\/8b175aaaf3267b5bbbbb97e4a6db8cea\"},\"description\":\"Ivan Srba p\u00f4sob\u00ed ako \u010dlen t\u00edmu spracovania webov\u00fdch a pou\u017e\u00edvate\u013esk\u00fdch \u00fadajov, kde sa \u0161pecializuje na rie\u0161enie probl\u00e9mov \u0161kodliv\u00fdch inform\u00e1ci\u00ed prostredn\u00edctvom umelej inteligencie. V j\u00fani 2023 dostal pr\u00edle\u017eitos\u0165 absolvova\u0165 v\u00fdskumn\u00fa n\u00e1v\u0161tevu na Univerzite v Sheffielde. \u010citajte \u010falej a dozviete sa viac o jeho sk\u00fasenostiach.\",\"breadcrumb\":{\"@id\":\"https:\\\/\\\/kinit.sk\\\/sk\\\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\\\/#breadcrumb\"},\"inLanguage\":\"sk-SK\",\"potentialAction\":[{\"@type\":\"ReadAction\",\"target\":[\"https:\\\/\\\/kinit.sk\\\/sk\\\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\\\/\"]}]},{\"@type\":\"ImageObject\",\"inLanguage\":\"sk-SK\",\"@id\":\"https:\\\/\\\/kinit.sk\\\/sk\\\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\\\/#primaryimage\",\"url\":\"https:\\\/\\\/kinit.sk\\\/wp-content\\\/uploads\\\/2023\\\/08\\\/202308_sheffield_feature_image.png\",\"contentUrl\":\"https:\\\/\\\/kinit.sk\\\/wp-content\\\/uploads\\\/2023\\\/08\\\/202308_sheffield_feature_image.png\",\"width\":1200,\"height\":628},{\"@type\":\"BreadcrumbList\",\"@id\":\"https:\\\/\\\/kinit.sk\\\/sk\\\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\\\/#breadcrumb\",\"itemListElement\":[{\"@type\":\"ListItem\",\"position\":1,\"name\":\"Home\",\"item\":\"https:\\\/\\\/kinit.sk\\\/sk\\\/\"},{\"@type\":\"ListItem\",\"position\":2,\"name\":\"News\",\"item\":\"https:\\\/\\\/kinit.sk\\\/sk\\\/category\\\/news-sk\\\/\"},{\"@type\":\"ListItem\",\"position\":3,\"name\":\"V\u00fdskumn\u00e1 n\u00e1v\u0161teva na Univerzite v Sheffielde: Falo\u0161n\u00e9 inform\u00e1cie a ve\u013ek\u00e9 jazykov\u00e9 modely\"}]},{\"@type\":\"WebSite\",\"@id\":\"https:\\\/\\\/kinit.sk\\\/#website\",\"url\":\"https:\\\/\\\/kinit.sk\\\/\",\"name\":\"KInIT\",\"description\":\"Vyu\u017e\u00edvame v\u00fdskum pre \u013eud\u00ed a priemysel\",\"potentialAction\":[{\"@type\":\"SearchAction\",\"target\":{\"@type\":\"EntryPoint\",\"urlTemplate\":\"https:\\\/\\\/kinit.sk\\\/?s={search_term_string}\"},\"query-input\":{\"@type\":\"PropertyValueSpecification\",\"valueRequired\":true,\"valueName\":\"search_term_string\"}}],\"inLanguage\":\"sk-SK\"},{\"@type\":\"Person\",\"@id\":\"https:\\\/\\\/kinit.sk\\\/#\\\/schema\\\/person\\\/8b175aaaf3267b5bbbbb97e4a6db8cea\",\"name\":\"Marianna Palkova\"}]}<\/script>\n<!-- \/ Yoast SEO plugin. -->","yoast_head_json":{"title":"V\u00fdskumn\u00e1 n\u00e1v\u0161teva na Univerzite v Sheffielde: Falo\u0161n\u00e9 inform\u00e1cie a ve\u013ek\u00e9 jazykov\u00e9 modely - KInIT","description":"Ivan Srba p\u00f4sob\u00ed ako \u010dlen t\u00edmu spracovania webov\u00fdch a pou\u017e\u00edvate\u013esk\u00fdch \u00fadajov, kde sa \u0161pecializuje na rie\u0161enie probl\u00e9mov \u0161kodliv\u00fdch inform\u00e1ci\u00ed prostredn\u00edctvom umelej inteligencie. V j\u00fani 2023 dostal pr\u00edle\u017eitos\u0165 absolvova\u0165 v\u00fdskumn\u00fa n\u00e1v\u0161tevu na Univerzite v Sheffielde. \u010citajte \u010falej a dozviete sa viac o jeho sk\u00fasenostiach.","robots":{"index":"index","follow":"follow","max-snippet":"max-snippet:-1","max-image-preview":"max-image-preview:large","max-video-preview":"max-video-preview:-1"},"canonical":"https:\/\/kinit.sk\/sk\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\/","og_locale":"sk_SK","og_type":"article","og_title":"V\u00fdskumn\u00e1 n\u00e1v\u0161teva na Univerzite v Sheffielde: Falo\u0161n\u00e9 inform\u00e1cie a ve\u013ek\u00e9 jazykov\u00e9 modely - KInIT","og_description":"Ivan Srba p\u00f4sob\u00ed ako \u010dlen t\u00edmu spracovania webov\u00fdch a pou\u017e\u00edvate\u013esk\u00fdch \u00fadajov, kde sa \u0161pecializuje na rie\u0161enie probl\u00e9mov \u0161kodliv\u00fdch inform\u00e1ci\u00ed prostredn\u00edctvom umelej inteligencie. V j\u00fani 2023 dostal pr\u00edle\u017eitos\u0165 absolvova\u0165 v\u00fdskumn\u00fa n\u00e1v\u0161tevu na Univerzite v Sheffielde. \u010citajte \u010falej a dozviete sa viac o jeho sk\u00fasenostiach.","og_url":"https:\/\/kinit.sk\/sk\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\/","og_site_name":"KInIT","article_published_time":"2023-08-17T09:40:48+00:00","article_modified_time":"2023-08-17T09:49:31+00:00","og_image":[{"width":1200,"height":628,"url":"https:\/\/kinit.sk\/wp-content\/uploads\/2023\/08\/202308_sheffield_feature_image.png","type":"image\/png"}],"author":"Marianna Palkova","twitter_card":"summary_large_image","twitter_creator":"@kinit","twitter_site":"@kinit","twitter_misc":{"Autor":"Marianna Palkova","Predpokladan\u00fd \u010das \u010d\u00edtania":"6 min\u00fat"},"schema":{"@context":"https:\/\/schema.org","@graph":[{"@type":"Article","@id":"https:\/\/kinit.sk\/sk\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\/#article","isPartOf":{"@id":"https:\/\/kinit.sk\/sk\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\/"},"author":{"name":"Marianna Palkova","@id":"https:\/\/kinit.sk\/#\/schema\/person\/8b175aaaf3267b5bbbbb97e4a6db8cea"},"headline":"V\u00fdskumn\u00e1 n\u00e1v\u0161teva na Univerzite v Sheffielde: Falo\u0161n\u00e9 inform\u00e1cie a ve\u013ek\u00e9 jazykov\u00e9 modely","datePublished":"2023-08-17T09:40:48+00:00","dateModified":"2023-08-17T09:49:31+00:00","mainEntityOfPage":{"@id":"https:\/\/kinit.sk\/sk\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\/"},"wordCount":1779,"image":{"@id":"https:\/\/kinit.sk\/sk\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\/#primaryimage"},"thumbnailUrl":"https:\/\/kinit.sk\/wp-content\/uploads\/2023\/08\/202308_sheffield_feature_image.png","keywords":["disinformation"],"articleSection":["News","2023"],"inLanguage":"sk-SK"},{"@type":"WebPage","@id":"https:\/\/kinit.sk\/sk\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\/","url":"https:\/\/kinit.sk\/sk\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\/","name":"V\u00fdskumn\u00e1 n\u00e1v\u0161teva na Univerzite v Sheffielde: Falo\u0161n\u00e9 inform\u00e1cie a ve\u013ek\u00e9 jazykov\u00e9 modely - KInIT","isPartOf":{"@id":"https:\/\/kinit.sk\/#website"},"primaryImageOfPage":{"@id":"https:\/\/kinit.sk\/sk\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\/#primaryimage"},"image":{"@id":"https:\/\/kinit.sk\/sk\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\/#primaryimage"},"thumbnailUrl":"https:\/\/kinit.sk\/wp-content\/uploads\/2023\/08\/202308_sheffield_feature_image.png","datePublished":"2023-08-17T09:40:48+00:00","dateModified":"2023-08-17T09:49:31+00:00","author":{"@id":"https:\/\/kinit.sk\/#\/schema\/person\/8b175aaaf3267b5bbbbb97e4a6db8cea"},"description":"Ivan Srba p\u00f4sob\u00ed ako \u010dlen t\u00edmu spracovania webov\u00fdch a pou\u017e\u00edvate\u013esk\u00fdch \u00fadajov, kde sa \u0161pecializuje na rie\u0161enie probl\u00e9mov \u0161kodliv\u00fdch inform\u00e1ci\u00ed prostredn\u00edctvom umelej inteligencie. V j\u00fani 2023 dostal pr\u00edle\u017eitos\u0165 absolvova\u0165 v\u00fdskumn\u00fa n\u00e1v\u0161tevu na Univerzite v Sheffielde. \u010citajte \u010falej a dozviete sa viac o jeho sk\u00fasenostiach.","breadcrumb":{"@id":"https:\/\/kinit.sk\/sk\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\/#breadcrumb"},"inLanguage":"sk-SK","potentialAction":[{"@type":"ReadAction","target":["https:\/\/kinit.sk\/sk\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\/"]}]},{"@type":"ImageObject","inLanguage":"sk-SK","@id":"https:\/\/kinit.sk\/sk\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\/#primaryimage","url":"https:\/\/kinit.sk\/wp-content\/uploads\/2023\/08\/202308_sheffield_feature_image.png","contentUrl":"https:\/\/kinit.sk\/wp-content\/uploads\/2023\/08\/202308_sheffield_feature_image.png","width":1200,"height":628},{"@type":"BreadcrumbList","@id":"https:\/\/kinit.sk\/sk\/vyskumna-navsteva-na-univerzite-v-sheffielde-falosne-informacie-a-velke-jazykove-modely\/#breadcrumb","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https:\/\/kinit.sk\/sk\/"},{"@type":"ListItem","position":2,"name":"News","item":"https:\/\/kinit.sk\/sk\/category\/news-sk\/"},{"@type":"ListItem","position":3,"name":"V\u00fdskumn\u00e1 n\u00e1v\u0161teva na Univerzite v Sheffielde: Falo\u0161n\u00e9 inform\u00e1cie a ve\u013ek\u00e9 jazykov\u00e9 modely"}]},{"@type":"WebSite","@id":"https:\/\/kinit.sk\/#website","url":"https:\/\/kinit.sk\/","name":"KInIT","description":"Vyu\u017e\u00edvame v\u00fdskum pre \u013eud\u00ed a priemysel","potentialAction":[{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https:\/\/kinit.sk\/?s={search_term_string}"},"query-input":{"@type":"PropertyValueSpecification","valueRequired":true,"valueName":"search_term_string"}}],"inLanguage":"sk-SK"},{"@type":"Person","@id":"https:\/\/kinit.sk\/#\/schema\/person\/8b175aaaf3267b5bbbbb97e4a6db8cea","name":"Marianna Palkova"}]}},"_links":{"self":[{"href":"https:\/\/kinit.sk\/sk\/wp-json\/wp\/v2\/posts\/27911","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/kinit.sk\/sk\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/kinit.sk\/sk\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/kinit.sk\/sk\/wp-json\/wp\/v2\/users\/26"}],"replies":[{"embeddable":true,"href":"https:\/\/kinit.sk\/sk\/wp-json\/wp\/v2\/comments?post=27911"}],"version-history":[{"count":2,"href":"https:\/\/kinit.sk\/sk\/wp-json\/wp\/v2\/posts\/27911\/revisions"}],"predecessor-version":[{"id":27915,"href":"https:\/\/kinit.sk\/sk\/wp-json\/wp\/v2\/posts\/27911\/revisions\/27915"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/kinit.sk\/sk\/wp-json\/wp\/v2\/media\/27907"}],"wp:attachment":[{"href":"https:\/\/kinit.sk\/sk\/wp-json\/wp\/v2\/media?parent=27911"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/kinit.sk\/sk\/wp-json\/wp\/v2\/categories?post=27911"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/kinit.sk\/sk\/wp-json\/wp\/v2\/tags?post=27911"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}