Riaan 
Benj Edwards / Snap, Inc.
NgoMsombuluko, i-Snapchat imemezele i-chatbot yengxoxo enamandla ye-AI ebizwa ngokuthi “I-AI yami,” inikwa amandla ubuchwepheshe besitayela se-ChatGPT obuvela kwa-OpenAI. I-AI yami izotholakala ngo-$3.99 ngenyanga kwababhalisile be-Snapchat+ futhi iyaphuma “kuleli sonto,” ngokusho kwe- okuthunyelwe kwezindaba kusuka ku-Snap, Inc.
Abasebenzisi bazokwazi ukwenza i-AI bot ibe ngeyakho ngokuyinika igama langokwezifiso. Izingxoxo nemodeli ye-AI zizokwenzeka ngendlela efanayo nengxoxo evamile nomuntu. “Umbono omkhulu ukuthi ngaphezu kokukhuluma nabangane nomndeni wethu nsuku zonke, sizokhuluma ne-AI nsuku zonke,” isikhulu esiphezulu se-Snap u-Evan Spiegel. etshela uThe Verge.
Kepha njengabazala bayo abanamandla e-GPT, I-ChatGPT futhi Ingxoxo ye-Bing, U-Snap uthi i-My AI ijwayele “ekuboneni izinto ezingekho,” okungamanga angalindelekile akhiqizwa imodeli ye-AI. Kuleli phuzu, i-Snap ifaka isitatimende eside ku-My AI okuthunyelwe kwesimemezelo:
“Njengawo wonke ama-chatbot anamandla e-AI, i-My AI ijwayele ukubona izinto ezingekho futhi ingakhohliswa ukuthi isho noma yini nje. Sicela uqaphele ukushiyeka kwayo okuningi futhi uxolise kusengaphambili! Zonke izingxoxo ne-My AI zizogcinwa futhi zingabuyekezwa ukuze thuthukisa ulwazi lomkhiqizo. Sicela ungabelani nganoma yiziphi izimfihlo ne-My AI futhi unganciki kuyo ukuze uthole iseluleko.”
Phakathi kwabacwaningi bokufunda ngomshini, “i-hallucination” i ithemu echaza lapho imodeli ye-AI yenza iziqondiso ezingalungile mayelana nesihloko noma isimo esingafakwanga kusethi yayo yedatha yokuqeqeshwa. Ukuhlehla okwaziwayo kwezinhlobo zezilimi zamanje ezinkulu njenge-ChatGPT, ezingakha kalula amanga azwakalayo azwakalayo, njenge amaphepha emfundo ukuthi azikho futhi ama-biography angalungile.
Ngaphandle kokuphika okuqinile kuka-Snap mayelana nokuthambekela kwe-My AI yokwenza izinto, ifemu ithi i-bot yayo entsha ye-Snapchat izobekwa ngenhla kwezingxoxo nabangane kuthebhu yayo kuhlelo lokusebenza lwe-Snapchat futhi “izoncoma imibono yesipho sosuku lokuzalwa le-BFF yakho, ihlele uhambo lokuqwala izintaba. ngempelasonto ende, phakamisa iresiphi yesidlo sakusihlwa, noma ubhale i-haiku mayelana noshizi kumngane wakho othathekile nge-cheddar.”
I-Snap ayivumelani ukuthi i-bot efanayo engakwazi ukuba “rel[ied] ukuze uthole iseluleko” angaphinde ahlele “uhambo lokuqwala izintaba olunembile noluphephile lwempelasonto ende.” Abagxeki bokukhishwa ngokushesha kwe-generative AI baye kubanjwe kuqhubeke lolu hlobo lwe-dissonance ukukhombisa ukuthi mhlawumbe lawa ma-chatbots awakulungele ukusetshenziswa kabanzi, ikakhulukazi uma ethulwa njengesithenjwa.
Ngenkathi abantu benze okuthile komdlalo wokuzama ukweqa I-ChatGPT futhi Ingxoxo ye-Bing Kubikwa ukuthi i-Snap iqeqeshe imodeli yayo ye-GPT ukuthi ingaxoxi ngocansi, inhlamba, udlame, noma imibono yezepolitiki. Leyo mikhawulo ingase idingeke ngokukhethekile ukuze ugweme ukuziphatha okungaguquki sobone nge-Bing Chat emasontweni ambalwa edlule.
Futhi kabili, ngoba i-“AI Yami” ingase ibe nokuthile okunamandla ngaphansi kwe-hood: Imodeli yolimi enkulu yesizukulwane esilandelayo se-OpenAI. Ngokusho kweThe Verge, i-Snap isebenzisa uhlelo olusha lwebhizinisi le-OpenAI olubizwa ngokuthi “I-Foundry” i-OpenAI eyethulwe buthule ekuqaleni kwale nyanga. Inikeza izinkampani ezizinikele ukufinyelela kwamafu kumamodeli we-OpenAI we-GPT-3.5 kanye ne-“DV”. Ochwepheshe abambalwa be-AI baye kucatshangelwe ukuthi i-“DV” ingase ilingane ne-GPT-4, ukulandelela okunamandla amakhulu okunamahemuhemu okuthi GPT-3.
Ngamanye amazwi, “i-hallucinations” Snap eshiwo ekukhishweni kwayo kwezindaba ingase ize ngokushesha futhi ibe nemininingwane eminingi kune-ChatGPT. Futhi ngokucabangela imvelo ekholisayo kakhulu yamanye amamodeli e-GPT, abantu bangase bakholwe, naphezu kwezixwayiso. Yinto okumele uyibuke ezinyangeni ezizayo njengoba izinsiza zezentengiselwano ezisebenza ngamandla e-GPT ziza ku-inthanethi.
