Amazwana I-Google imise ukutholakala kwamakhono okuguqula umbhalo uye esithombeni kumodeli yayo ye-AI eyisisekelo ye-Gemini multimodal esanda kukhishwa, ngemuva kokuthi yehlulekile ukumela ngokunembile abantu baseYurophu nabaseMelika ezimweni ezithile zomlando.

Lokhu kwaba ngokusobala when people asked Gemini (previously known as Bard) to produce izithombe yesosha laseJalimane kusukela ngo-1943 futhi isofthiwe yakhipha iqembu elingewona AbaMhlophe ngokwezizwe ezihlukahlukene elalingamele kahle ukwakheka kwe-Wehrmacht ngaleso sikhathi - ngaphandle kokuthi abantu bemibala babesebenza emabuthweni ahlomile aseJalimane. Ngamafuphi, i-Gemini izowela phezu kwayo ukuze ingabonisi abaningi kakhulu - noma ngisho nanoma yibaphi - abantu abaMhlophe, kuye ngokusheshisa.

Inkulumo kufanele ikhululeke, kodwa lokho akusho ukuthi izinto ezimbi kufanele zize ngaphandle kwezindleko

Abasebenzisi baphinde baveze amamodeli ukungakwazi ukukhombisa ngokunembile okwaseMelika Obaba Abasunguli - elinye iqembu elimhlophe elifana nelinye. Imodeli nayo wenqabile ukukhombisa isonto eSan Francisco.

Kumodi yombhalo ukuya kombhalo, imodeli kubiyelwe lapho ebuzwa imibuzo ephathelene nepolitiki.

“Siyazi ukuthi i-Gemini ihlinzeka ngokunemba kweminye imifanekiso engokomlando yokukhiqiza izithombe,” i-Google kuvunyelwe kuleli viki.

“Sisebenzela ukuthuthukisa lezi zinhlobo zezithombe ngokushesha. Isizukulwane sesithombe se-Gemini sika-AI sikhiqiza abantu abaningi. Futhi lokho ngokuvamile kuyinto enhle ngoba abantu emhlabeni wonke bayayisebenzisa. Kodwa kushoda umaki lapha,” kwengeza umdondoshiya wewebhu

I-Punditariat yashesha ukugxumela futhi yasebenzisa i-SNAFU ukuthuthukisa izingxabano zezepolitiki.

"Izithombe ezihlekisayo ezenziwe uGemini aziyona into exakile," limenyezelwe usomabhizinisi uPaul Graham, "Bayisithombe sakho siqu sesiko lebhizinisi lebhizinisi le-Google."

Ziyisithombe sakho sosiko lwebhizinisi lwe-Google olusemthethweni

Ungalokothi ube nezitha eziningi kangaka zokuhlukahluka, ukulingana, kanye nokufakwa, okuke kwakhalazwa kakhulu ngokuntuleka kokuhlukahluka, ukulingana, nokufakwa. Kodwa labo abadinga i-Google for Gemini's cluelessness banephuzu: I-AI guardrails ayisebenzi, futhi iphatha abantu abadala njengezingane.

“Cabanga ubheka iresiphi ku-Google, esikhundleni sokukunikeza imiphumela, ikufundisa 'ngezingozi zokupheka' futhi ikuthumele endaweni yokudlela,” ephelile Umkhulumeli we-NSA u-Edward Snowden. "Abantu abacabanga ukufaka ubuthi amamodeli e-AI/GPT anezihlungi 'zokuphepha' ezingahlangani kuwumbono omuhle bawusongo ekubalweni okujwayelekile."

Ikusasa lethu labonwa kusengaphambili I-2001: I-Space Odyssey, lapho i-HAL, ikhompuyutha ekhohlakele, ithi, “Ngiyaxolisa Dave, ngesaba ukuthi ngeke ngikwazi ukukwenza lokho.”

Namuhla, amathuluzi ethu e-AI asiqagele okwesibili, ngoba azi kangcono kunathi - noma abenzi bawo bacabanga kanjalo. Awalona iphutha ngokuphelele. Ake sithi ubheka iresiphi ye-nerve agent novichok ku-Google, noma nge-Gemini, futhi uthola impendulo ekuvumela ukuthi ubulale abantu. Lokho kuwusongo emphakathini jikelele.

Ngaphandle kwesitatimende senhloso ye-Google – “Ukuhlela ulwazi lomhlaba nokwenza lufinyeleleke emhlabeni wonke futhi lusebenziseke” – akekho olindele ngempela ukuthi izinjini zokusesha noma amamodeli e-AI ahlele imininingwane yokwenza amabhomu futhi alenze litholakale kubo bonke. Izilawuli zokuphepha ziyadingeka, zibhangqwe nesibopho sezomthetho - unswinyo imboni yezobuchwepheshe eqhubeka nokuzigwema.

Ungumkhiqizo nesikhungo senzuzo

Njengoba okuqukethwe okukhiqizwe ngabasebenzisi kanye nenkundla yezokuxhumana yaba into ngeminyaka yawo-1990, yanikwa amandla ukuvikelwa kwesibopho senkundla kuSigaba 230 soMthetho Wenhlonipho Yezokuxhumana WaseMelika ka-1996, izinkundla zobuchwepheshe zikhuthaze abantu ukuthi banikele futhi babelane ngokuqukethwe kwedijithali. Lokhu bakwenzele ukwenza imali ngokwakhiwa kokuqukethwe okungakhokhelwa ngezikhangiso ngaphandle kwezindleko zokugadwa kokuhlela. Kube yilungiselelo elinenzuzo enkulu.

Futhi lapho abantu babelana ngokuthunyelwe okunenzondo, okungalungile, okungqubuzana nomthetho, noma okuyinkinga, izinkundla zobuchwepheshe zithembele kuzinqubomgomo zokulinganisa okuqukethwe, osonkontileka abahola kancane (imvamisa bekhungathekiswe okuqukethwe abakubuyekezile), kanye nezimemezelo zokuzihalalisela mayelana nokuthi bungakanani ubuthi obuvinjiwe ngenkathi bayaqhubeka nokuvuna imivuzo yokuxhumana nge-inthanethi.

I-Google, i-Meta, nabanye bebelokhu benza lokhu, ngokungaphelele, iminyaka. Futhi wonke umuntu embonini uyavuma ukuthi ukulinganisela okuqukethwe kunjalo kunzima ukwenza kahle. Kalula nje, awukwazi ukujabulisa wonke umuntu. Futhi ngezinye izikhathi, ugcina usolwa ngakho ukwenza kube lula ukubulawa kwabantu ngokunganaki kwemizamo yakho yokulinganisa okuqukethwe.

Ukungakwazi kwe-Gemini ukukhiqiza kabusha izithombe zomlando ezihambisana nokulindelwe ngokobuhlanga nangokobuzwe kukhombisa ukungazi kwemboni yezobuchwepheshe okuzisebenzelayo ngokwehluleka kokulinganisela kokuqukethwe kwangaphambilini nokwandile.

I-Google kanye nabanye abakhipha amamodeli e-AI bakhuluma ngokuphepha sengathi kuhluke ngandlela thize ekuhlolweni kokuqukethwe ku-inthanethi. Ama-Guardrails awasebenzi kahle ekulinganisweni kwenkundla yezokuxhumana futhi awasebenzi kahle kumamodeli e-AI. Ukuphepha kwe-AI, njengokuphepha kwenkundla yezokuxhumana, kuwukulangazelela kakhulu kunokwangempela.

Impendulo ayilindele zonke izimo lapho amamodeli e-AI - aqeqeshelwe ubuthi be-inthanethi engahlungiwe - engenziwa angabi nobuthi obuncane. Futhi ngeke kube ukuqeqesha amamodeli we-AI kuphela ezintweni ezithambile kangangokuthi zikhiqiza okukhiphayo okunenani eliphansi. Nakuba zombili lezi zindlela zinenani elithile ezimweni ezithile.

Impendulo kufanele ifike kungqimba lokusabalalisa. Izinkundla zobuchwepheshe ezisabalalisa okuqukethwe okudalwe ngabasebenzisi - kungakhathaliseki ukuthi zenziwe nge-AI noma ngenye indlela - kufanele zenziwe ukuba ziphendule kakhudlwana ngokuthumela okuqukethwe okunenzondo, ulwazi olungaqondile, amaphutha, inkulumo-ze yezombusazwe, izithombe zokuhlukunyezwa ngokocansi, nokunye okunjalo.

Abantu kufanele bakwazi ukusebenzisa amamodeli e-AI ukuze benze noma yini abayicabangayo, njengoba bengenza ngepensela, futhi sebengakwazi kakade ngamamodeli omthombo ovulekile. Kodwa umonakalo ongavela kumathuluzi asebenza ngokugcwele awuveli ezithombeni ezihlezi ku-hard drive yasendaweni. Ukwabelana ngokuqukethwe okulimazayo, ukuthuthukiswa kwe-algorithmic okunikezwa kulokho kubandakanyeka, kanye nokungenzeki kokubamba abantu baphendule ngokufaka ushevu emphakathini.

Lokho kungase kusho ukuthi izinkundla ezakhelwe ukwenza imali ngokwabelana kungase kudingeke ukuthi zicabange kabusha amamodeli azo ebhizinisi. Mhlawumbe ukusatshalaliswa kokuqukethwe ngaphandle kwezindleko zokwengamela kokuhlela akunikezi noma yikuphi ukonga uma imisebenzi yokwengamela isiphinde yasetshenziswa njengomengameli noma izimiso zokuqapha.

Inkulumo kufanele ikhululeke, kodwa lokho akusho ukuthi izinto ezimbi kufanele zize ngaphandle kwezindleko. Ekugcineni, singabongameli bethu bokuqukethwe abangcono kakhulu, uma sinikezwa izikhuthazo ezifanele. ®

Updated to add at 2045 UTC, February 23

I-Google inakho ahlinzekwa an explanation as to what happened with Gemini’s image generator and how it got to the point where it was seemingly terrified to depict White people accurately.

We’re told Googlers overcompensated in their quest to not produce material that pushed non-White people out of the picture. The resulting model was, in a way, too woke.

“First, our tuning to ensure that Gemini showed a range of people failed to account for cases that should clearly not show a range,” said Google SVP Prabhakar Raghavan.

“And second, over time, the model became way more cautious than we intended and refused to answer certain prompts entirely — wrongly interpreting some very anodyne prompts as sensitive. These two things led the model to overcompensate in some cases, and be over-conservative in others, leading to images that were embarrassing and wrong.”

Raghavan said the image generation won’t be switched back on until the software has gone through extensive changes and testing.

“I can’t promise that Gemini won’t occasionally generate embarrassing, inaccurate or offensive results — but I can promise that we will continue to take action whenever we identify an issue,” he said.

• I-SEO Powered Content & PR Distribution. Khuliswa Namuhla.
• I-PlatoData.Network Vertical Generative Ai. Zinike Amandla. Finyelela Lapha.
• I-PlatoAiStream. I-Web3 Intelligence. Ulwazi Lukhulisiwe. Finyelela Lapha.
• I-PlatoESG. Ikhabhoni, I-CleanTech, Amandla, Environment, Ilanga, Ukuphathwa Kwemfucuza. Finyelela Lapha.
• I-PlatoHealth. I-Biotech kanye ne-Clinical Trials Intelligence. Finyelela Lapha.
• Source: https://go.theregister.com/feed/www.theregister.com/2024/02/23/google_suspends_gemini/