{"id":120455,"date":"2024-11-02T13:09:07","date_gmt":"2024-11-02T11:09:07","guid":{"rendered":"https:\/\/prohoster.info\/blog\/novosti-interneta\/google-ispolzoval-bolshuyu-yazykovuyu-model-dlya-vyyavleniya-uyazvimosti-v-sqlite"},"modified":"2024-11-02T13:09:07","modified_gmt":"2024-11-02T11:09:07","slug":"google-ispolzoval-bolshuyu-yazykovuyu-model-dlya-vyyavleniya-uyazvimosti-v-sqlite","status":"publish","type":"post","link":"https:\/\/prohoster.info\/sk\/blog\/news\/google-ispolzoval-bolshuyu-yazykovuyu-model-dlya-vyyavleniya-uyazvimosti-v-sqlite","title":{"rendered":"Google pou\u017eil ve\u013ek\u00fd jazykov\u00fd model na identifik\u00e1ciu zranite\u013enosti v SQLite","gt_translate_keys":[{"key":"rendered","format":"text"}]},"content":{"rendered":"
Vedci z oddelen\u00ed Google Project Zero a Google DeepMind zverejnili spr\u00e1vu o v\u00fdvoji AI syst\u00e9mu Big Sleep, ktor\u00fd je postaven\u00fd na ve\u013ekom jazykovom modeli Gemini 1.5 Pro a je ur\u010den\u00fd na identifik\u00e1ciu zranite\u013enost\u00ed v zdrojovom k\u00f3de. Dosiahnut\u00fdm m\u00ed\u013enikom projektu je pomocou Big Sleep identifikovanie prvej prev\u00e1dzkyschopnej a predt\u00fdm nezn\u00e1mej zranite\u013enosti v existuj\u00facom projekte. Zranite\u013enos\u0165 bola identifikovan\u00e1 tokom kontroly k\u00f3dovej z\u00e1kladne DBMS SQLite AI syst\u00e9mom a vedie k prete\u010deniu dolnej hranice bufferu na z\u00e1sobn\u00edku (buffer underflow). Probl\u00e9m bol zisten\u00fd v ned\u00e1vno prijatom k\u00f3de a bol odstr\u00e1nen\u00fd pred jeho zaraden\u00edm do kone\u010dn\u00e9ho vydania SQLite 3.47.0. <\/p>\n
Model m\u00f4\u017ee by\u0165 pou\u017e\u00edvan\u00fd ako pomocn\u00fd n\u00e1stroj v oblastiach, kde je potrebn\u00e9 vykon\u00e1va\u0165 pr\u00e1cne manu\u00e1lne recenzovanie, ako aj na organiz\u00e1ciu automatickej kontroly nov\u00e9ho k\u00f3du na identifik\u00e1ciu zranite\u013enost\u00ed v rann\u00fdch f\u00e1zach v\u00fdvoja (predt\u00fdm, ako problematick\u00fd k\u00f3d prejde do kone\u010dn\u00fdch vydan\u00ed). Predpoklad\u00e1 sa, \u017ee vyv\u00edjan\u00fd AI model umo\u017en\u00ed identifikova\u0165 bezpe\u010dnostn\u00e9 probl\u00e9my v k\u00f3de, ktor\u00e9 je problematick\u00e9 ur\u010di\u0165 prostredn\u00edctvom fuzzing testovania. <\/p>\n
Dodatkovo mo\u017eno poznamena\u0165 vyhl\u00e1senie gener\u00e1lneho riadite\u013ea Google, pod\u013ea ktor\u00e9ho v s\u00fa\u010dasnosti viac ako \u0161tvrtina (25 %) cel\u00e9ho k\u00f3du vytv\u00e1ran\u00e9ho v spolo\u010dnosti sa generuje pomocou ve\u013ek\u00fdch jazykov\u00fdch modelov Gemini, po \u010dom je tento k\u00f3d recenzovan\u00fd a prij\u00edman\u00fd in\u017einiermi. Uv\u00e1dza sa, \u017ee tak\u00e9to vyu\u017eitie AI v\u00fdrazne ur\u00fdchlilo proces v\u00fdvoja produktov. \n Zdroj: opennet.ru<\/a> <\/p>","protected":false,"gt_translate_keys":[{"key":"rendered","format":"html"}]},"excerpt":{"rendered":"