Yeqa uye kokuyinhloko
Ukhuselelo NokuBusa

I-Google yethule imodeli i-Gemini 3.8 Flash Cyber, esisetshenziswe ngokuphelele ngaphakathi ukuvikela ukuphepha kwekhodi

Ngo-2 Septhemba ngesikhathi sendawo, i-Google imemezele ukwethulwa kwemodeli i-Gemini 3.8 Flash Cyber, evulelwa amaqembu okuqala athembekile okuvikela ukuphepha ngohlelo i-Fairwind.

I-Google yethule imodeli i-Gemini 3.8 Flash Cyber ukuvikela ukuphepha kwekhodi

Ngo-2 Septhemba ngesikhathi sendawo, i-Google imemezele ukwethulwa kwemodeli i-Gemini 3.8 Flash Cyber, evulelwa amaqembu okuqala athembekile okuvikela ukuphepha ngohlelo i-Fairwind.

Isikhulu sithe ekuhlolweni okujwayelekile okuhlonishwayo embonini kokutholwa kwezikhala, i-CyberGym, i-Gemini 3.8 Flash Cyber ikhombise ikhono elisezingeni eliphezulu lokuthola izikhala ngokuzenzakalelayo. Ayigcinanga nje ngokudlula eyandulelayo i-3.5 Flash Cyber, kodwa iphinde yashiya ngemuva amamodeli amaningi aphambili amakhulu kakhulu ngosayizi.

I-Google yethule imodeli i-Gemini 3.8 Flash Cyber ukuvikela ukuphepha kwekhodi

I-Google ihlole i-Gemini 3.8 Flash Cyber ekuhlolweni kwayo okubanzi kwangaphakathi. Lokhu kuhlola kudinga imodeli ukuthi ithole izinhlobo ezahlukene zezikhala zokuphepha kumaphrojekthi ekhodi ayinkimbinkimbi ahlanganisa izilimi zokuhlela ezingu-20. Imiphumela ikhombise ukuthi imodeli yenze igxathu elikhulu uma iqhathaniswa namamodeli angaphambilini e-Google, futhi izinga lokuphumelela ekutholeni izikhala ledlula ku-70%.

I-Google yethule imodeli i-Gemini 3.8 Flash Cyber ukuvikela ukuphepha kwekhodi

I-Google ithe ngesikhathi ithuthukisa i-Gemini 3.8 Flash Cyber, inhloso yayo eyinhloko kwakuwukunikeza abavikeli bezokuphepha amakhono asezingeni lochwepheshe, ukuze bakwazi ukubusa lapho bebhekene nabahlaseli. Yingakho i-Google yagxila kusukela ekuqaleni ekulungiseni izikhala, yabeka leli khono ngaphezu kwezindlela zokuhlasela ezifana nokusebenzisa izikhala.

Ekuhlolweni kwe-CWE-Bench, i-Gemini 3.8 Flash Cyber ihlezi kuPareto Frontier: izinga layo lokunemba lokuqala (Pass@1) lifinyelele ku-47.2%, lisondela kumodeli ephambili engu-47.8%; ngesikhathi esifanayo, izindleko zokuyisebenzisa ziphansi kakhulu, okwenza kube nebhalansi enhle kakhulu phakathi kokusebenza nezindleko.

I-Google yethule imodeli i-Gemini 3.8 Flash Cyber ukuvikela ukuphepha kwekhodi

I-Google ithe i-Gemini 3.8 Flash inezindlela eziphelele zokuvikela ukuphepha ngaphakathi, esekela ukusetshenziswa okuhambisana nemithetho nokuhle, kuyilapho ivikela ngokuqinile ubungozi bokusetshenziswa kabi kwemodeli emkhakheni wamakhemikhali, wezinto eziphilayo, wemisebe kanye wezikhali zenuzi (CBRN), kanye nasekuhlaselweni kwe-inthanethi.

Ngokuphambene nalokho, i-Gemini 3.8 Flash Cyber iyehlise ngokusesilinganisweni imikhawulo yokuvimba amasu emkhakheni wokuphepha kwe-inthanethi, ngakho ivulelwa kuphela amaqembu okuphepha ahambisana nemithetho adinga amakhono aphelele kakhulu okuzivikela nokuhlasela ku-inthanethi.

Njengamanje, i-Google isisebenzise ngokuphelele ngaphakathi i-Gemini 3.8 Flash Cyber ukuvikela ukuphepha kwekhodi. Nazi izibonelo ezithile zokusetshenziswa kwayo:

Ithimba lezokuphepha le-Chrome lithole ekuhloleni ukuthi, ezikhaleni zokuphepha kusiphequluli se-Chrome, i-3.8 Flash Cyber ikhiqize amapheshana okulungisa asebenzayo aphindwe ngo-2.6 kunamamodeli ezentengiso aphambili, kanti lawo mamodeli ezentengiso aphambili anosayizi wamapharamitha omkhulu kakhulu kunawo.

Ukuhlolwa kwesikhungo sezokuphepha i-Wiz kukhombise ukuthi ekuhlolweni kwaso kwangaphakathi, uma kuqhathaniswa namanye amamodeli aphambili, i-Gemini 3.8 Flash Cyber yenyuse izinga lokuthola okuthile ngo-7.5%-9.7%, ngesikhathi esifanayo yehlisa kakhulu izindleko zokuyisebenzisa zafinyelela ku-1/5.2 kuya ku-1/2.3 wezindleko zangaphambilini.

Ithimba labacwaningi bezikhala efwini le-Google lisebenzise imodeli i-3.8 Flash Cyber ukuthola ngempumelelo isikhala esibucayi esakhiweni esiyisisekelo ngaphansi kwamahora angu-2; ngaphambi kwalokho, ucwaningo nokuphenya lolu hlobo lwezikhala kwakuvame ukuthatha izinyanga eziningi.