I-TechCrunch ibuyekeze uhlu lwamagama e-AI ngokufaka i-opaque recurrence ye-OpenAI Astra

Okushintshile

I-TechCrunch ibuyekeze uhlu lwayo oluqhubeka lubuyekezwa lwamagama e-AI ngoSepthemba 7, yagqamisa i-“opaque recurrence”, eyichaza njengendlela yokucabanga esetshenziswa kumodeli entsha ye-OpenAI, i-Astra, esikhathaze abacwaningi bokuphepha kwe-AI. Lolu hlu luhlanganisa namagama asejwayelekile kodwa aphikisanwayo, okuhlanganisa i-AGI, ama-ejenti e-AI kanye nama-API endpoint. TechCrunch

Lo mbiko ukhombisa ukuthi kungani izincazelo zibalulekile. Isikhulu esiphezulu se-OpenAI, uSam Altman, sichaze i-AGI njengosebenza naye womuntu osezingeni eliphakathi; umqulu wezinjongo ze-OpenAI uyichaza njengezinhlelo ezizimele kakhulu ezidlula abantu emisebenzini eminingi ebaluleke ngokwezomnotho; i-Google DeepMind yona iyibheka njenge-AI enekhono elingekho ngaphansi kwelabantu emisebenzini eminingi yokucabanga. I-TechCrunch nayo ithi i-ejenti ye-AI ngokuvamile yenza imisebenzi edinga izinyathelo eziningi, njengokufaka izicelo zezindleko, ukubhukha amathikithi noma ukugcina ikhodi, futhi ingase isebenzise izinhlelo eziningi ze-AI; kodwa iyaphawula ukuthi leli gama alinayo incazelo efanayo kuwo wonke umuntu. Ama-API endpoint ayizindawo zokuxhumana kusofthiwe lapho izinhlelo zokusebenza zishintshisana khona ngedatha noma zenza izenzo, okuhlanganisa nokuvumela ama-ejenti ukuba asebenzisane nezinsiza zezinkampani zangaphandle. TechCrunch

Kungani lokhu kubalulekile

Lesi sibuyekezo sibeka igama elisha elihlobene ne-Astra eduze kwenkinga ebanzi: amalebula emikhiqizo ye-AI angasitha umehluko osebenzayo phakathi kwe-chatbot, uhlelo oluzenzakalelayo olwenza imisebenzi ngezinyathelo eziningi, kanye nesofthiwe enegunya lokwenza izenzo ngokuhlanganiswa nezinsiza zangaphandle.

Mayelana ne-Astra, i-TechCrunch ayichazi ukuthi i-opaque recurrence isebenza kanjani noma ayiqinisekisi imiphumela yayo ekuphepheni. Lokho kushiya ukukhathazeka okubikiwe kugxile endleleni yokucabanga engakachazwa, kuyilapho kwenza ukuhlolwa kobuchwepheshe kubaluleke ngaphambi kokuba kuhlolwe indima nokuziphatha kwayo. TechCrunch

Ukuhlolwa komthelela

Abaqhubi bemikhiqizo ye-AI bangase basuse izingxoxo ezisheshayo ezimangalweni ezibanzi zama-“ejenti” bazise ekuqedweni kwemisebenzi ethile, ekufinyeleleni amathuluzi nasekudluliseleni umsebenzi kubantu. Lokho kungathuthukisa izinqumo zokusebenzisa, kodwa kuphela uma amaqembu ezibekela imingcele yawo yokusebenza kunokuba aphathe ilebula njengencazelo yobuchwepheshe.

Amaqembu ezokuthenga ezinkampanini angase abe sengcupheni enkulu emasontweni azayo. Lapho i-ejenti yethulwa njengengasebenzisa ama-API endpoint ezinsizeni zezinkampani zangaphandle, abathengi bangase kudingeke bahlole ukuhamba komsebenzi kwangempela, ukuhlanganiswa, izimvume nezilawuli zokugunyaza ngaphambi kokwabela izinsiza zokuhlola nokuhlanganisa. Imibhalo ehambisanayo yabathengisi nezidingo zabathengi ezisekelwe ebufakazini kunganciphisa leyo ngozi.

Abacwaningi bokuphepha kwe-AI bangase bayibhekisise kakhulu indlela ebikiwe ye-Astra ye-opaque recurrence. Lowo mthelela uncike ekutheni i-OpenAI inikeze imininingwane eyanele yobuchwepheshe ukuze kuhlolwe ukuthi le ndlela ikuthinta kanjani ukuziphatha kwemodeli.

Izimo ezingase zenzeke

Umbono wethu wesikhathi esizayo (ukuqagela okunolwazi):

Okungenzeka kakhulu: Uma abathengisi beqhubeka nokumaketha amakhono ama-ejenti kuyilapho izincazelo zihlala zingafani, amaqembu emikhiqizo nabathengi bazoguqula kakhulu izimangalo zama-“ejenti” zibe yizidingo ezicacisa ububanzi bemisebenzi, ukuhlanganiswa nezimvume zabantu ezinyangeni ezintathu kuya kweziyisithupha ezizayo. Lena yindlela engenzeka kakhulu ngoba i-TechCrunch ichaza ama-ejenti njengezinhlelo ezenza imisebenzi ngezinyathelo eziningi, kuyilapho ivuma ukuthi izincazelo ziyahlukana. Imibhalo ehlukanisa imisebenzi yengxoxo nezenzo zangaphandle izoqinisa lo mkhuba; incazelo efanayo yokusebenza eyamukelwa abahlinzeki abakhulu ingawenza buthaka.

Okuhle kakhulu: Lapho abaqhubi bekhawulela amathuluzi azimele ekuhambeni komsebenzi okuchazwe ngokucacile nangokunqunywa, bese besebenzisa ukufinyelela kwe-API ngenhloso, ukusetshenziswa kungaba nokuhleleka okukhulu ezinyangeni eziyisithupha kuya kweziyi-12. Imikhiqizo izoshicilela imingcele ecacile yemisebenzi, yokuhlanganiswa neyokugunyaza, kanti abathengi bazoqhathanisa ukuhamba komsebenzi okubhaliwe kunokuba baqhathanise izimangalo ezibanzi zokuzimela. Lokho kunganciphisa igebe phakathi kwemiboniso nezinhlelo ezisebenza ngempela.

Okubi kakhulu: Uma imikhiqizo eqhudelanayo iqhubeka nokusebenzisa ilebula efanayo ye-ejenti kumazinga ahluke kakhulu okuzimela nawokufinyelela kuma-endpoint, abathengi bangase baqondise isabelomali sokuhlola nomsebenzi wokuhlanganisa emathuluzini angahambisani nokulindelwe ezinyangeni eziyisithupha kuya kweziyi-12. Ukuphinda kwenziwe umsebenzi okubangelwa imingcele engadalulwanga kungaqinisa lowo mphumela; ukudalulwa okujwayelekile kwezenzo, izimvume nemingcele yokusebenza kungawenza buthaka.

Okufanele kubhekwe ngokulandelayo

• Ukuthi i-OpenAI iyayikhipha yini incazelo yobuchwepheshe echaza i-opaque recurrence nendima yayo ku-Astra emasontweni noma ezinyangeni ezizayo.
• Ukuthi abathengisi abaningi be-AI bayayicacisa yini imisebenzi edinga izinyathelo eziningi, ukuhlanganiswa nemingcele yokugunyaza esekela izimangalo zabo zama-ejenti.