N\u00eb mes t\u00eb gar\u00ebs globale p\u00ebr zhvillimin e Inteligjenc\u00ebs Artificiale (AI), nj\u00eb val\u00eb tensioni e re po shfaqet mes Shteteve t\u00eb Bashkuara dhe Kin\u00ebs. Kompania amerikane Anthropic paralajm\u00ebron se disa firma kineze po vjedhin teknologji q\u00eb mund t\u00eb vendosin se kush do t\u00eb ket\u00eb avantazhin n\u00eb fush\u00ebn e AI-s\u00eb. Sipas denoncimeve, DeepSeek, Moonshot AI dhe MiniMax krijuan fshehurazi mbi 16 milion\u00eb biseda me chatbotin Claude t\u00eb Anthropic, duke p\u00ebrdorur m\u00eb shum\u00eb se 24,000 llogari false p\u00ebr t\u00eb \u201cthithur\u201d inteligjenc\u00ebn e tij dhe p\u00ebr t\u00eb trajnuar modelet e tyre konkurruese.<\/p>\n\n\n\n
\u201cKjo praktik\u00eb, e njohur si \u2018distillation\u2019, p\u00ebrdoret p\u00ebr t\u00eb marr\u00eb p\u00ebrgjigjet e nj\u00eb modeli t\u00eb madh AI dhe p\u00ebr t\u2019i p\u00ebrdorur ato p\u00ebr t\u00eb trajnuar nj\u00eb model m\u00eb t\u00eb vog\u00ebl, m\u00eb t\u00eb shpejt\u00eb dhe m\u00eb t\u00eb lir\u00eb,\u201d shpjegon Anthropic.<\/p>\n\n\n\n
N\u00eb k\u00ebt\u00eb m\u00ebnyr\u00eb, sipas kompanis\u00eb, modelet e vogla mund t\u00eb zhvillohen shum\u00eb m\u00eb shpejt dhe me kosto shum\u00eb m\u00eb t\u00eb ul\u00ebt, duke shmangur or\u00ebt e k\u00ebrkimit dhe investimin q\u00eb k\u00ebrkohet p\u00ebr t\u00eb nd\u00ebrtuar nj\u00eb AI t\u00eb fort\u00eb nga fillimi.<\/p>\n\n\n\n
Rreziqet komb\u00ebtare dhe siguria<\/strong> \u201cModelet e tilla nuk do t\u00eb ken\u00eb sigurin\u00eb e nevojshme p\u00ebr t\u00eb parandaluar p\u00ebrdorimin nga aktor\u00eb shtet\u00ebror\u00eb ose jo-shtet\u00ebror\u00eb,\u201d theksoi Anthropic. Nga ana tjet\u00ebr, Google thekson se p\u00ebrdoruesit e zakonsh\u00ebm t\u00eb AI-s\u00eb nuk rrezikohen nga k\u00ebto sulme, pasi ato nuk c\u00ebnojn\u00eb konfidencialitetin, disponueshm\u00ebrin\u00eb ose integritetin e sh\u00ebrbimeve t\u00eb AI-s\u00eb.<\/p>\n\n\n\n Si funksionon \u201ddistillation\u201d?<\/strong> \u201cDeepSeek k\u00ebrkoi nga Claude t\u00eb shpjegonte hap pas hapi se si arriti tek nj\u00eb p\u00ebrgjigje, duke krijuar t\u00eb dh\u00ebna t\u00eb m\u00ebdha p\u00ebr \u2018chain-of-thought\u2019,\u201d tha Anthropic. Google gjithashtu raporton se Gemini, chatbot-i i tyre, po abuzohet p\u00ebr kodim, skriptim dhe mbledhjen e informacionit t\u00eb ndjesh\u00ebm si kredencialet e llogarive dhe adresat e emailit. Anthropic pohon se ka vendosur masa p\u00ebr t\u00eb zbuluar k\u00ebto sulme n\u00eb koh\u00eb reale, por thekson se asnj\u00eb kompani AI nuk mund t\u00eb zgjidh\u00eb problemin e vetme.<\/p>\n","protected":false},"excerpt":{"rendered":" N\u00eb mes t\u00eb gar\u00ebs globale p\u00ebr zhvillimin e Inteligjenc\u00ebs Artificiale (AI), nj\u00eb val\u00eb tensioni e re po shfaqet mes Shteteve t\u00eb Bashkuara dhe Kin\u00ebs. Kompania amerikane Anthropic paralajm\u00ebron se disa firma kineze po vjedhin teknologji q\u00eb mund t\u00eb vendosin se kush do t\u00eb ket\u00eb avantazhin n\u00eb fush\u00ebn e AI-s\u00eb. Sipas denoncimeve, DeepSeek, Moonshot AI dhe<\/p>\n","protected":false},"author":4,"featured_media":26185,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[349],"tags":[40],"class_list":{"0":"post-26184","1":"post","2":"type-post","3":"status-publish","4":"format-standard","5":"has-post-thumbnail","7":"category-artificial-intelligence","8":"tag-featured"},"yoast_head":"\n
Google dhe Anthropic paralajm\u00ebrojn\u00eb se modelet e krijuara p\u00ebrmes k\u00ebsaj praktike mund t\u00eb p\u00ebrb\u00ebjn\u00eb rreziqe t\u00eb m\u00ebdha p\u00ebr sigurin\u00eb komb\u00ebtare, sepse mungojn\u00eb masat mbrojt\u00ebse q\u00eb do t\u00eb pengonin p\u00ebrdorimin e AI-s\u00eb p\u00ebr arm\u00eb biologjike ose sulme kibernetike.<\/p>\n\n\n\n
Kompani kineze, sipas denoncimeve, p\u00ebrdor\u00ebn rrjete proxy<\/em> dhe nj\u00eb \u201chidra network<\/em>\u201d t\u00eb gjer\u00eb me llogari false p\u00ebr t\u00eb marr\u00eb qasje n\u00eb Claude, i cili \u00ebsht\u00eb i ndaluar n\u00eb Kin\u00eb. Ata krijuan volumin e madh t\u00eb komandave p\u00ebr t\u00eb mbledhur p\u00ebrgjigje cil\u00ebsore p\u00ebr trajnimin e modeleve t\u00eb tyre, p\u00ebrfshir\u00eb edhe m\u00ebnyra p\u00ebr t\u00eb shmangur temat e ndjeshme politikisht.<\/p>\n\n\n\n