{"id":721604,"date":"2025-01-29T14:21:06","date_gmt":"2025-01-29T13:21:06","guid":{"rendered":"https:\/\/www.hashtag.al\/?p=721604"},"modified":"2025-01-29T14:21:06","modified_gmt":"2025-01-29T13:21:06","slug":"pse-inteligjenca-artificiale-kineze-ka-mahnitur-boten","status":"publish","type":"post","link":"https:\/\/www.hashtag.al\/index.php\/2025\/01\/29\/pse-inteligjenca-artificiale-kineze-ka-mahnitur-boten\/","title":{"rendered":"Pse inteligjenca artificiale kineze ka mahnitur bot\u00ebn","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"

\"\"<\/span><\/p>\n

Modeli i par\u00eb i arsyetimit n\u00eb bot\u00eb, nj\u00eb form\u00eb e avancuar e inteligjenc\u00ebs artificiale, u publikua n\u00eb shtator nga OpenAI, nj\u00eb firm\u00eb amerikane. o1, si\u00e7 quhet, p\u00ebrdor nj\u00eb \u201czinxhir mendimi\u201d p\u00ebr t\u2019iu p\u00ebrgjigjur pyetjeve t\u00eb v\u00ebshtira n\u00eb shkenc\u00eb dhe matematik\u00eb, duke zb\u00ebrthyer problemet n\u00eb hapat e tyre p\u00ebrb\u00ebr\u00ebs dhe duke testuar qasje t\u00eb ndryshme, se t\u2019i paraqes\u00eb nj\u00eb p\u00ebrfundim p\u00ebrdoruesit. Zbulimi i saj nisi nj\u00eb gar\u00eb p\u00ebr t\u00eb kopjuar k\u00ebt\u00eb metod\u00eb. Google doli me nj\u00eb model arsyetimi t\u00eb quajtur \u201cGemini Flash Thinking\u201d n\u00eb dhjetor. OpenAI u p\u00ebrgjigj me o3, nj\u00eb p\u00ebrdit\u00ebsim t\u00eb o1, disa dit\u00eb m\u00eb von\u00eb.<\/span><\/p>\n

Por Google, me t\u00eb gjitha burimet e saj, nuk ishte n\u00eb fakt firma e par\u00eb q\u00eb imitoi OpenAI. M\u00eb pak se tre muaj pasi u publikua o1, Alibaba, gjigandi kinez i tregtis\u00eb elektronike, l\u00ebshoi nj\u00eb version t\u00eb ri t\u00eb chatbot-it t\u00eb tij Qwen, QwQ, me t\u00eb nj\u00ebjtat aft\u00ebsi \u201carsyetimi\u201d. \u201c\u00c7far\u00eb do t\u00eb thot\u00eb t\u00eb mendosh, t\u00eb pyes\u00ebsh, t\u00eb kuptosh?\u201d kompania pyeti n\u00eb nj\u00eb postim. Nj\u00eb tjet\u00ebr firm\u00eb kineze, DeepSeek, kishte publikuar nj\u00eb version t\u00eb nj\u00eb modeli arsyetimi, t\u00eb quajtur R1, nj\u00eb jav\u00eb m\u00eb par\u00eb. Pavar\u00ebsisht p\u00ebrpjekjeve t\u00eb qeveris\u00eb amerikane, p\u00ebr t\u00eb frenuar industrin\u00eb e inteligjenc\u00ebs artificiale t\u00eb Kin\u00ebs, dy firma kineze kishin reduktuar avantazhin teknologjik t\u00eb homolog\u00ebve t\u00eb tyre amerikan\u00eb n\u00eb disa jav\u00eb.<\/span><\/p>\n

Firmat kineze nuk jan\u00eb t\u00eb vetmet n\u00eb zhvillimin e modeleve t\u00eb arsyetimit: n\u00eb dhjetor, DeepSeek publikoi nj\u00eb model t\u00eb ri t\u00eb madh gjuh\u00ebsor (LLM), nj\u00eb form\u00eb e IA, q\u00eb analizon dhe gjeneron tekst. v3 ishte pothuajse 700 gigabajt, shum\u00eb i madh p\u00ebr t\u2019u ekzekutuar n\u00eb \u00e7do gj\u00eb, p\u00ebrve\u00e7 pajisjeve t\u00eb specializuara, dhe kishte 685 miliard\u00eb parametra, parime individuale q\u00eb kombinohen p\u00ebr t\u00eb formuar rrjetin e modelit. Kjo e b\u00ebri at\u00eb m\u00eb t\u00eb madh se \u00e7do gj\u00eb e publikuar m\u00eb par\u00eb p\u00ebr t\u2019u shkarkuar falas. Llama 3.1 LLM e Meta, e cila u publikua n\u00eb korrik, ka vet\u00ebm 405 miliard\u00eb parametra.<\/span><\/p>\n

LLM i DeepSeek nuk \u00ebsht\u00eb vet\u00ebm m\u00eb i madh se shum\u00eb nga homolog\u00ebt e tij per\u00ebndimor\u00eb \u2013 ai \u00ebsht\u00eb gjithashtu m\u00eb i mir\u00eb, i krahasuar vet\u00ebm me modelet e Google dhe OpenAI. Paul Gauthier, themeluesi i Aider, nj\u00eb platform\u00eb kodimi IA, drejtoi modelin e ri DeepSeek p\u00ebrmes standardit t\u00eb tij t\u00eb kodimit dhe zbuloi se ai i tejkaloi t\u00eb gjith\u00eb rival\u00ebt e tij, p\u00ebrve\u00e7 vet\u00eb o1.<\/span><\/p>\n

Inteligjenca artificiale kineze \u00ebsht\u00eb tani aq af\u00ebr n\u00eb cil\u00ebsi me rival\u00ebt e saj amerikan\u00eb, saq\u00eb drejtuesi i OpenAI, Sam Altman, u ndje i detyruar t\u00eb shpjegonte ngusht\u00ebsin\u00eb e hendekut. Menj\u00ebher\u00eb pasi DeepSeek publikoi v3, ai shkroi n\u00eb Twitter: \u201c\u00cbsht\u00eb (relativisht) e leht\u00eb t\u00eb kopjosh di\u00e7ka, q\u00eb e di se funksionon. \u00cbsht\u00eb jasht\u00ebzakonisht e v\u00ebshtir\u00eb t\u00eb b\u00ebsh di\u00e7ka t\u00eb re, t\u00eb rrezikshme dhe t\u00eb v\u00ebshtir\u00eb, kur nuk e di n\u00ebse do t\u00eb funksionoj\u00eb.\u201d<\/span><\/p>\n

Industria e inteligjenc\u00ebs artificiale t\u00eb Kin\u00ebs, fillimisht ishte shfaqur \u2018e dor\u00ebs s\u00eb dyt\u00eb\u2019. Kjo mund t\u00eb jet\u00eb pjes\u00ebrisht sepse i \u00ebsht\u00eb dashur t\u00eb p\u00ebrballet me sanksionet amerikane. N\u00eb vitin 2022 Amerika ndaloi eksportin e \u00e7ipave t\u00eb avancuar n\u00eb Kin\u00eb. Nvidia, nj\u00eb prodhuesi kryesor i \u00e7ipave, i \u00ebsht\u00eb dashur t\u00eb hartoj\u00eb ulje t\u00eb ve\u00e7anta p\u00ebr produktet e saj p\u00ebr tregun kinez. Amerika ka k\u00ebrkuar gjithashtu t\u00eb parandaloj\u00eb Kin\u00ebn, q\u00eb t\u00eb zhvilloj\u00eb kapacitetin p\u00ebr t\u00eb prodhuar \u00e7ipa m\u00eb t\u00eb mir\u00eb brenda kufirit, duke ndaluar eksportet e pajisjeve t\u00eb nevojshme dhe duke k\u00ebrc\u00ebnuar me nd\u00ebshkime p\u00ebr firmat jo-amerikane, q\u00eb mund t\u00eb ndihmojn\u00eb.<\/span><\/p>\n

Nj\u00eb penges\u00eb tjet\u00ebr \u00ebsht\u00eb zhvillimi brenda kufirit. Firmat kineze erdh\u00ebn me vones\u00eb n\u00eb LLM, pjes\u00ebrisht p\u00ebr shkak t\u00eb shqet\u00ebsimeve rregullatore. Ata shqet\u00ebsoheshin se si censuruesit do t\u00eb reagonin ndaj modeleve, q\u00eb mund t\u00eb japin informacion t\u00eb pasakt\u00eb ose \u2013 m\u00eb keq \u2013 t\u00eb dilnin me deklarata politikisht t\u00eb rrezikshme. Gjigandi i k\u00ebrkimit Baidu, kishte eksperimentuar me LLM p\u00ebr vite me rradh\u00eb dhe kishte krijuar nj\u00eb t\u00eb quajtur \u201cERNIE\u201d, por hezitoi ta vint\u00eb n\u00eb p\u00ebrdorim at\u00eb p\u00ebr publikun.<\/span><\/p>\n

Gjigant\u00ebt e tjer\u00eb t\u00eb internetit t\u00eb Kin\u00ebs, p\u00ebrfshir\u00eb Tencent dhe Huawei, po nd\u00ebrtojn\u00eb modelet e tyre. Por DeepSeek ka origjin\u00eb t\u00eb ndryshme. Nuk ekzistonte as kur Alibaba nxori modelin e par\u00eb Qwen. Ai rrjedh nga High-Flyer, nj\u00eb fond mbrojt\u00ebs i krijuar n\u00eb 2015 p\u00ebr t\u00eb p\u00ebrdorur IA, p\u00ebr t\u00eb fituar nj\u00eb avantazh n\u00eb tregtimin e aksioneve. Kryerja e k\u00ebrkimeve themelore ndihmoi High-Flyer t\u00eb b\u00ebhej nj\u00eb nga fondet sasiore m\u00eb t\u00eb m\u00ebdha n\u00eb vend.<\/span><\/p>\n

Ashtu si OpenAI, DeepSeek premtoi t\u00eb zhvillonte IA p\u00ebr t\u00eb mir\u00ebn publike. Kompania do t\u00eb b\u00ebj\u00eb publike shumic\u00ebn e rezultateve t\u00eb saj t\u00eb trajnimit, n\u00eb p\u00ebrpjekje p\u00ebr t\u00eb parandaluar \u201cmonopolizimin\u201d e teknologjis\u00eb nga vet\u00ebm disa individ\u00eb ose firma. Ndryshe nga OpenAI, i cili u detyrua t\u00eb k\u00ebrkonte fonde private p\u00ebr t\u00eb mbuluar kostot e larta t\u00eb trajnimit, DeepSeek ka pasur gjithmon\u00eb akses n\u00eb rezervat e m\u00ebdha t\u00eb fuqis\u00eb kompjuterike t\u00eb High-Flyer.<\/span><\/p>\n

Jo vet\u00ebm, q\u00eb modeli ishte krijuar me \u00e7mim t\u00eb lir\u00eb, por edhe funksionimi i tij kushton m\u00eb pak. DeepSeek i ndan detyrat mbi \u00e7ipa t\u00eb shumt\u00eb n\u00eb m\u00ebnyr\u00eb m\u00eb efikase se koleg\u00ebt e tij dhe fillon hapin tjet\u00ebr t\u00eb nj\u00eb procesi, p\u00ebrpara se t\u00eb p\u00ebrfundoj\u00eb ai i m\u00ebparshmi. Kjo e lejon at\u00eb t\u00eb mbaj\u00eb \u00e7ipat t\u00eb punojn\u00eb me kapacitet t\u00eb plot\u00eb me pak tepric\u00eb.<\/span><\/p>\n

K\u00ebrkimi i DeepSeek p\u00ebr efikasitet nuk ka ndalur k\u00ebtu. K\u00ebt\u00eb jav\u00eb, edhe pse publikoi R1 t\u00eb plot\u00eb, kompania publikoi gjithashtu nj\u00eb grup variantesh m\u00eb t\u00eb vogla, m\u00eb t\u00eb lira dhe m\u00eb t\u00eb shpejta, t\u00eb cilat jan\u00eb pothuajse po aq t\u00eb fuqishme sa modeli m\u00eb i madh.<\/span><\/p>\n

Alibaba dhe DeepSeek sfidojn\u00eb laborator\u00ebt m\u00eb t\u00eb avancuar per\u00ebndimor\u00eb edhe n\u00eb nj\u00eb m\u00ebnyr\u00eb tjet\u00ebr. Ndryshe nga OpenAI dhe Google, laborator\u00ebt kinez\u00eb ndjekin drejtimin e Meta dhe i b\u00ebjn\u00eb sistemet e tyre t\u00eb disponueshme n\u00ebn nj\u00eb licenc\u00eb me burim t\u00eb hapur. N\u00ebse d\u00ebshironi t\u00eb shkarkoni nj\u00eb Qwen AI dhe t\u00eb nd\u00ebrtoni programimin tuaj, mundeni\u2014nuk nevojitet leje specifike.<\/span><\/p>\n

Jo vet\u00ebm an\u00eb pozitive<\/strong><\/span><\/p>\n

Puna n\u00eb Kin\u00eb ka edhe an\u00ebt e saj negative. Pyete DeepSeek v3 p\u00ebr Tajvanin, p\u00ebr shembull, dhe modelja me g\u00ebzim fillon t\u00eb shpjegoj\u00eb se \u00ebsht\u00eb nj\u00eb ishull n\u00eb Azin\u00eb Lindore \u201ci njohur zyrtarisht si Republika e Kin\u00ebs\u201d. Por, pasi t\u00eb ket\u00eb kompozuar disa fjali n\u00eb k\u00ebto rreshta, ai ndalet, fshin p\u00ebrgjigjen e tij fillestare dhe n\u00eb vend t\u00eb k\u00ebsaj sugjeron shkurt: \u201cLe t\u00eb flasim p\u00ebr di\u00e7ka tjet\u00ebr\u201d.<\/span><\/p>\n

Laborator\u00ebt kinez\u00eb jan\u00eb m\u00eb transparent\u00eb se qeveria e tyre, pjes\u00ebrisht sepse duan t\u00eb krijojn\u00eb nj\u00eb ekosistem firmash t\u00eb p\u00ebrqendruar n\u00eb IA-n e tyre. Kjo ka nj\u00ebfar\u00eb vlere komerciale, n\u00eb at\u00eb q\u00eb kompanit\u00eb, q\u00eb nd\u00ebrtojn\u00eb mbi modelet me burim t\u00eb hapur, mund t\u00eb binden p\u00ebrfundimisht t\u00eb blejn\u00eb produkte ose sh\u00ebrbime nga krijuesit e tyre. Ajo gjithashtu sjell nj\u00eb p\u00ebrfitim strategjik p\u00ebr Kin\u00ebn, n\u00eb at\u00eb q\u00eb krijon aleat\u00eb n\u00eb konfliktin e saj me Amerik\u00ebn mbi IA.<\/span><\/p>\n

Kjo nuk do t\u00eb thot\u00eb domosdoshm\u00ebrisht se modelet kineze do t\u00eb pushtojn\u00eb bot\u00ebn. Inteligjenca artificiale amerikane ka ende aft\u00ebsi, q\u00eb rival\u00ebt e saj kinez\u00eb nuk mund t\u2019i p\u00ebrballojn\u00eb ende.<\/span><\/p>\n

M\u00eb shum\u00eb risi jan\u00eb n\u00eb zhvillim, sipas Sam Altman, i cili pritet t\u00eb njoftoj\u00eb s\u00eb shpejti se OpenAI ka nd\u00ebrtuar \u201csuper-agjent\u00eb t\u00eb nivelit t\u00eb doktoratur\u00ebs\u201d, t\u00eb cil\u00ebt jan\u00eb po aq t\u00eb aft\u00eb sa ekspert\u00ebt njer\u00ebzor\u00eb, p\u00ebr nj\u00eb s\u00ebr\u00eb detyrash intelektuale.\/Monitor\u00a0<\/strong><\/span><\/a><\/span><\/p>\n","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"

Modeli i par\u00eb i arsyetimit n\u00eb bot\u00eb, nj\u00eb form\u00eb e avancuar e inteligjenc\u00ebs artificiale, u publikua n\u00eb shtator nga OpenAI, nj\u00eb firm\u00eb amerikane. o1, si\u00e7 quhet, p\u00ebrdor nj\u00eb \u201czinxhir mendimi\u201d p\u00ebr t\u2019iu p\u00ebrgjigjur pyetjeve t\u00eb v\u00ebshtira n\u00eb shkenc\u00eb dhe matematik\u00eb, duke zb\u00ebrthyer problemet n\u00eb hapat e tyre p\u00ebrb\u00ebr\u00ebs dhe duke testuar qasje t\u00eb ndryshme, se […]<\/p>\n","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"author":49,"featured_media":721607,"comment_status":"closed","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"footnotes":""},"categories":[53],"tags":[],"class_list":["post-721604","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-fast-foodi"],"gt_translate_keys":[{"key":"link","format":"url"}],"_links":{"self":[{"href":"https:\/\/www.hashtag.al\/index.php\/wp-json\/wp\/v2\/posts\/721604","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/www.hashtag.al\/index.php\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/www.hashtag.al\/index.php\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/www.hashtag.al\/index.php\/wp-json\/wp\/v2\/users\/49"}],"replies":[{"embeddable":true,"href":"https:\/\/www.hashtag.al\/index.php\/wp-json\/wp\/v2\/comments?post=721604"}],"version-history":[{"count":99999,"href":"https:\/\/www.hashtag.al\/index.php\/wp-json\/wp\/v2\/posts\/721604\/revisions"}],"predecessor-version":[{"id":721608,"href":"https:\/\/www.hashtag.al\/index.php\/wp-json\/wp\/v2\/posts\/721604\/revisions\/721608"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/www.hashtag.al\/index.php\/wp-json\/wp\/v2\/media\/721607"}],"wp:attachment":[{"href":"https:\/\/www.hashtag.al\/index.php\/wp-json\/wp\/v2\/media?parent=721604"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/www.hashtag.al\/index.php\/wp-json\/wp\/v2\/categories?post=721604"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/www.hashtag.al\/index.php\/wp-json\/wp\/v2\/tags?post=721604"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}