Yeqa uye kokuyinhloko
Ama-modeli kanye nobuchwepheshe

I-Google ikhiphe imodeli yokuguqula inkulumo ibe umbhalo i-Gemini 3.5 Transcribe, ekwazi ukususa ngokuzenzakalela amagama okugcwalisa

Le modeli ye-AI ingasusa ngokuzenzakalela amagama okugcwalisa enkulumweni, ilungise okuqukethwe okungalungile, futhi inyuse isivinini sokubhala inkulumo sibe cishe ngo-70%. Izinga lamaphutha enkulumweni yesikhathi sangempela lehle laya ku-5.5%, futhi isekela nohlu lwamagama olwenziwe ngokwezifiso. Nokho, ukulungiswa okuhlakaniphile kungashintsha amagama okuqala, ngakho ayifanele izimo lapho kudingeka kugcinwe amazwi oqobo. #ImodeliYe-AIGoogle#

I-Google ikhiphe i-Gemini 3.5 Transcribe, imodeli yokuguqula inkulumo ibe umbhalo esusa amagama okugcwalisa

Ngesikhathi umhlaba usalinde ukukhishwa kwe-Gemini 3.5 Pro, i-Google iqale yakhipha enye imodeli ochungechungeni lwe-3.5. Le nkampani ikhiphe namuhla imodeli ye-Gemini 3.5 Transcribe yokuguqula inkulumo ibe umbhalo. Le modeli ye-AI ihloselwe ukuthuthukisa ulwazi lokufaka ngenkulumo, futhi ingasusa ngokuzenzakalela amagama okugcwalisa afana nokuthi “嗯” no-“呃”, kanye nezingxenye umuntu azilungisa ngokwakhe ngemva kokukhuluma kabi, bese ikhiqiza umbhalo oshelelayo nohlelekile.

I-Google ikhiphe imodeli ye-Gemini 3.5 Transcribe yokuguqula inkulumo ibe umbhalo, ekwazi ukususa amagama okugcwalisa

Le modeli isivele isekela umsebenzi we-“Rambler” wekhibhodi ye-Gboard ku-Pixel 11, futhi maduze izongena kancane kancane kweminye imikhiqizo namasevisi e-Google.

I-Google ithe uma iqhathaniswa nenjini yangaphambilini yokuguqula inkulumo ibe umbhalo ebibizwa nge-Chirp 3, i-Gemini 3.5 Transcribe ithuthuke kokubili ngesivinini nangokunemba. Kusukela umsebenzisi eqala ukukhuluma kuze kukhiqizwe umbhalo wokuloba wokugcina, isivinini sonke kulindeleke ukuthi sikhuphuke cishe ngo-70%; ezimweni zenkulumo yesikhathi sangempela, izinga lamaphutha lehle laya ku-5.5%.

I-Google ikhiphe imodeli ye-Gemini 3.5 Transcribe yokuguqula inkulumo ibe umbhalo, ekwazi ukususa amagama okugcwalisa

Nokho, ukwanda kokunemba akukukhulu kakhulu. Ngokwemininingwane ye-Google, izinga lamaphutha le-Chirp 3 lingu-7.32%. Kodwa kubasebenzisi abavame ukusebenzisa ukufaka ngenkulumo, ngisho nokwenza amaphutha ambalwa okubhala kusho ukuthi kuncipha okuqukethwe okuzodinga ukulungiswa ngesandla kamuva, ngakho kusewukuthuthuka okuwusizo.

Amandla e-Gemini 3.5 Transcribe awagcini nje “ngokuzwa kahle” lokho okushiwo umsebenzisi; iphinde izame ukuqonda lokho umsebenzisi afuna ukukusho ngempela. Ngesikhathi sokukhuluma, imodeli ingasusa ngokuzenzakalela amagama okugcwalisa afana nokuthi “嗯” no-“呃” athikameza ukugeleza komusho. Uma umsebenzisi ekhuluma kabi bese ezilungisa ngokushesha, imodeli ingakwazi nokulungisa umbhalo osuvele ukhiqiziwe ngesikhathi sangempela. Ngaphezu kwalokho, abasebenzisi bangahlinzeka ngohlu lwamagama olwenziwe ngokwezifiso ukuze imodeli ibone kangcono amagama obuchwepheshe nawezinkambu ezithile.

Le modeli okwamanje isekela izilimi ezingu-85, futhi ingacubungula umsindo oqoshiwe ngaphambilini onabakhulumi abangafika kwabangu-3.

Yiqiniso, lo msebenzi nawo unenkinga ecacile: umsebenzisi kufanele athembe ukuthi i-AI ingakuqonda ngokunembile lokho afuna ukukusho ngempela. Okungenani embhalweni omfushane, ngokwasesimweni sokusebenzisa umsebenzi we-Rambler, imodeli ibonakala ikwazi ukuhlanza kahle izinkulumo ezingahambisani phakathi kwazo kanye nokungingiza ngesikhathi sokukhuluma, yenze umbhalo ubukeke ungokwemvelo.

Kodwa ngesikhathi esifanayo, i-Gemini 3.5 Transcribe ayiguquli nje inkulumo ibe umbhalo. Ngesikhathi ihlela okuqukethwe, i-AI ingashintsha amagama asetshenziswe umsebenzisi ekuqaleni. Ngakho-ke, ezimweni lapho kudingeka kugcinwe ngokuqinile amazwi oqobo futhi kungavunyelwe ukuguqulwa ngokuthanda, lokhu “kulungiswa okuhlakaniphile” kungase kungafaneleki.