{"id":682482,"date":"2026-03-18T03:16:04","date_gmt":"2026-03-18T03:16:04","guid":{"rendered":"https:\/\/microscopemedia.com\/?p=682482"},"modified":"2026-03-18T03:16:04","modified_gmt":"2026-03-18T03:16:04","slug":"un-studiu-arata-ca-chatgpt-greseste-in-domeniul-stiintific-mai-des-decat-ai-crede","status":"publish","type":"post","link":"https:\/\/microscopemedia.com\/?p=682482","title":{"rendered":"Un studiu arat\u0103 c\u0103 ChatGPT gre\u0219e\u0219te \u00een domeniul \u0219tiin\u021bific mai des dec\u00e2t ai crede"},"content":{"rendered":"
<\/div>\n

Echipa de cercet\u0103tori a evaluat peste 700 de ipoteze \u0219i a pus aceea\u0219i \u00eentrebare de 10 ori pentru fiecare dintre ele, pentru a m\u0103sura gradul de consecven\u021b\u0103, arat\u0103 ScienceDaily<\/a>.<\/p>\n

C\u00e2nd experimentul a fost realizat pentru prima dat\u0103 \u00een 2024, ChatGPT a r\u0103spuns corect \u00een 76,5% din cazuri. \u00centr-un test de urm\u0103rire din 2025, precizia a crescut u\u0219or la 80%.<\/p>\n

Cu toate acestea, odat\u0103 ce cercet\u0103torii au ajustat rezultatele pentru a \u021bine cont de ghicirea aleatorie, rezultatele au p\u0103rut mult mai pu\u021bin impresionante. Inteligen\u021ba artificial\u0103 (IA) a avut o performan\u021b\u0103 cu doar aproximativ 60% mai bun\u0103 dec\u00e2t \u0219ansa, un nivel mai apropiat de un D sc\u0103zut dec\u00e2t de o fiabilitate puternic\u0103.<\/p>\n

Sistemul a \u00eent\u00e2mpinat cele mai mari dificult\u0103\u021bi \u00een identificarea afirma\u021biilor false, etichet\u00e2ndu-le corect doar \u00een 16,4% din cazuri.<\/p>\n

De asemenea, a prezentat o inconsisten\u021b\u0103 notabil\u0103. Chiar \u0219i atunci c\u00e2nd i s-a dat exact aceea\u0219i solicitare de 10 ori, ChatGPT a produs r\u0103spunsuri consecvente doar \u00een aproximativ 73% din cazuri.<\/p>\n

R\u0103spunsurile inconsistente ridic\u0103 \u00eengrijor\u0103ri<\/h2>\n

\u201eNu vorbim doar despre acurate\u021be, vorbim despre inconsisten\u021b\u0103, pentru c\u0103 dac\u0103 pui aceea\u0219i \u00eentrebare iar \u0219i iar, ob\u021bii r\u0103spunsuri diferite\u201d, a spus Cicek, profesor asociat la Departamentul de Marketing \u0219i Afaceri Interna\u021bionale din cadrul Carson College of Business al WSU \u0219i autor principal al noii publica\u021bii.<\/p>\n

\u201eAm folosit 10 solicit\u0103ri cu exact aceea\u0219i \u00eentrebare. Totul era identic. R\u0103spundea c\u0103 este adev\u0103rat. Apoi, spune c\u0103 este fals. Este adev\u0103rat, este fals, fals, adev\u0103rat. Au existat mai multe cazuri \u00een care au fost cinci r\u0103spunsuri adev\u0103rate \u0219i cinci false\u201d.<\/p>\n

Concluziile subliniaz\u0103 importan\u021ba pruden\u021bei atunci c\u00e2nd ne baz\u0103m pe IA pentru decizii importante, \u00een special cele care necesit\u0103 un ra\u021bionament nuan\u021bat sau complex. De\u0219i IA generativ\u0103 poate produce un limbaj fluent \u0219i conving\u0103tor, aceasta nu demonstreaz\u0103 \u00eenc\u0103 acela\u0219i nivel de \u00een\u021belegere conceptual\u0103.<\/p>\n

De ce exper\u021bii \u00eendeamn\u0103 la pruden\u021b\u0103 \u00een ceea ce prive\u0219te IA<\/h2>\n

Pe baza acestor constat\u0103ri, cercet\u0103torii recomand\u0103 liderilor de afaceri s\u0103 verifice informa\u021biile generate de IA \u0219i s\u0103 le abordeze cu scepticism. De asemenea, ei subliniaz\u0103 necesitatea form\u0103rii pentru a \u00een\u021belege mai bine ce pot \u0219i ce nu pot face eficient sistemele de IA.<\/p>\n

De\u0219i acest studiu s-a concentrat \u00een mod specific pe ChatGPT, Cicek a remarcat c\u0103 experimente similare cu alte instrumente de IA au produs rezultate comparabile. Lucrarea se bazeaz\u0103, de asemenea, pe cercet\u0103ri anterioare care indic\u0103 pruden\u021b\u0103 \u00een ceea ce prive\u0219te entuziasmul excesiv legat de IA. Un sondaj na\u021bional din 2024 a constatat c\u0103 consumatorii erau mai pu\u021bin dispu\u0219i s\u0103 cumpere produse atunci c\u00e2nd acestea erau comercializate cu accent pe IA.<\/p>\n","protected":false},"excerpt":{"rendered":"

Echipa de cercet\u0103tori a evaluat peste 700 de ipoteze \u0219i a pus aceea\u0219i \u00eentrebare de 10 ori pentru fiecare dintre ele, pentru a m\u0103sura gradul de consecven\u021b\u0103, arat\u0103 ScienceDaily. C\u00e2nd … Read More<\/a><\/p>\n","protected":false},"author":1,"featured_media":682483,"comment_status":"closed","ping_status":"closed","sticky":false,"template":"Default","format":"standard","meta":[],"categories":[1],"tags":[],"_links":{"self":[{"href":"https:\/\/microscopemedia.com\/index.php?rest_route=\/wp\/v2\/posts\/682482"}],"collection":[{"href":"https:\/\/microscopemedia.com\/index.php?rest_route=\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/microscopemedia.com\/index.php?rest_route=\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/microscopemedia.com\/index.php?rest_route=\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/microscopemedia.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcomments&post=682482"}],"version-history":[{"count":0,"href":"https:\/\/microscopemedia.com\/index.php?rest_route=\/wp\/v2\/posts\/682482\/revisions"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/microscopemedia.com\/index.php?rest_route=\/wp\/v2\/media\/682483"}],"wp:attachment":[{"href":"https:\/\/microscopemedia.com\/index.php?rest_route=%2Fwp%2Fv2%2Fmedia&parent=682482"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/microscopemedia.com\/index.php?rest_route=%2Fwp%2Fv2%2Fcategories&post=682482"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/microscopemedia.com\/index.php?rest_route=%2Fwp%2Fv2%2Ftags&post=682482"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}