{"id":1551,"date":"2024-12-22T06:56:47","date_gmt":"2024-12-22T06:56:47","guid":{"rendered":"https:\/\/europaskolos.lt\/index.php\/2024\/12\/22\/pranesama-kad-openai-gpt-5-nepateisino-lukesciu\/"},"modified":"2024-12-22T06:56:47","modified_gmt":"2024-12-22T06:56:47","slug":"pranesama-kad-openai-gpt-5-nepateisino-lukesciu","status":"publish","type":"post","link":"https:\/\/europaskolos.lt\/index.php\/2024\/12\/22\/pranesama-kad-openai-gpt-5-nepateisino-lukesciu\/","title":{"rendered":"Prane\u0161ama, kad OpenAI GPT-5 nepateisino l\u016bkes\u010di\u0173"},"content":{"rendered":"


\n<\/p>\n

\n

Remiantis nauja \u201eThe Wall Street Journal\u201c ataskaita, \u201eOpenAI\u201c pastangos sukurti kit\u0105 pagrindin\u012f model\u012f GPT-5 atsilieka nuo grafiko, o rezultatai dar nepateisina mil\u017eini\u0161k\u0173 i\u0161laid\u0173.<\/p>\n

Tai pakartoja ankstesn\u0119 \u201eThe Information\u201c ataskait\u0105, kurioje teigiama, kad \u201eOpenAI\u201c ie\u0161ko nauj\u0173 strategij\u0173, nes GPT-5 gali b\u016bti ne toks didelis \u0161uolis \u012f priek\u012f, kaip ankstesni modeliai. Ta\u010diau WSJ istorijoje yra papildomos informacijos apie 18 m\u0117nesi\u0173 trukus\u012f GPT-5, kodiniu pavadinimu Orion, k\u016brim\u0105.<\/p>\n

Prane\u0161ama, kad \u201eOpenAI\u201c baig\u0117 ma\u017eiausiai du didelius mokymus, kuri\u0173 tikslas \u2013 patobulinti model\u012f, mokant j\u012f naudojant did\u017eiulius duomen\u0173 kiekius. Pradinis treniruo\u010di\u0173 b\u0117gimas vyko l\u0117\u010diau nei tik\u0117tasi, o tai rodo, kad didesnis b\u0117gimas u\u017etrukt\u0173 ir brangiai. Ir nors teigiama, kad GPT-5 gali veikti geriau nei jo pirmtakai, jis dar n\u0117ra pakankamai pa\u017eeng\u0119s, kad pateisint\u0173 modelio eksploatavimo i\u0161laidas.<\/p>\n

WSJ taip pat prane\u0161a, kad \u201eOpenAI\u201c taip pat pasamd\u0117 \u017emones, kad jie sukurt\u0173 naujus duomenis ra\u0161ant kod\u0105 ar sprend\u017eiant matematikos problemas, o ne tik pasikliaujant vie\u0161ai prieinamais duomenimis ir licencijavimo sandoriais. Jis taip pat naudoja sintetinius duomenis, sukurtus kito modelio, o1.<\/p>\n

\u201eOpenAI\u201c i\u0161 karto neatsak\u0117 \u012f pra\u0161ym\u0105 pakomentuoti. Bendrov\u0117 anks\u010diau teig\u0117, kad \u0161iais metais nei\u0161leis modelio kodiniu pavadinimu \u201eOrion\u201c.<\/p>\n<\/div>\n


\n
Source link <\/a><\/p>\n","protected":false},"excerpt":{"rendered":"

Remiantis nauja \u201eThe Wall Street Journal\u201c ataskaita, \u201eOpenAI\u201c pastangos sukurti kit\u0105 pagrindin\u012f model\u012f GPT-5 atsilieka nuo grafiko, o rezultatai dar nepateisina mil\u017eini\u0161k\u0173 i\u0161laid\u0173. Tai pakartoja ankstesn\u0119 \u201eThe Information\u201c ataskait\u0105, kurioje teigiama, kad \u201eOpenAI\u201c ie\u0161ko nauj\u0173 strategij\u0173, nes GPT-5 gali b\u016bti ne toks didelis \u0161uolis \u012f priek\u012f, kaip ankstesni modeliai. Ta\u010diau WSJ istorijoje yra papildomos informacijos […]<\/p>\n","protected":false},"author":1,"featured_media":1552,"comment_status":"","ping_status":"","sticky":false,"template":"","format":"standard","meta":{"om_disable_all_campaigns":false,"_monsterinsights_skip_tracking":false,"_monsterinsights_sitenote_active":false,"_monsterinsights_sitenote_note":"","_monsterinsights_sitenote_category":0,"footnotes":""},"categories":[5],"tags":[2700,24,2702,2701,589,114],"class_list":["post-1551","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-technologijos-programine-iranga-prietaisai","tag-gpt5","tag-kad","tag-lukesciu","tag-nepateisino","tag-openai","tag-pranesama"],"aioseo_notices":[],"aioseo_head":"\n\t\t\n\t\n\t\n\t\n\t\n\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t\n\t\t