UClaude, i-chatbot efuna ukuncintisana ne-OpenAI
I-Anthropic, isiqalo se-AI esakhiwe ngo-2021 yizisebenzi zangaphambili ze-OpenAI, isiqalile buthule ukuhlola umsizi omusha we-AI ofana ne-ChatGPT obonakala ethuthuka kweyokuqala ezindaweni ezimbalwa ezibalulekile.
ngegama lekhodi "Claude", lokhu kwadalwa kusetshenziswa indlela eyakhiwe yi-Anthropic, ebizwa ngokuthi "I-AI Yomthethosisekelo" kanye nendlela ihlose ukuhlinzeka ngendlela “esekelwe ezimisweni” ukuvumelanisa izinhlelo ze-AI nezinhloso zomuntu, ukuvumela ama-AI afana ne-ChatGPT ukuthi aphendule imibuzo esebenzisa isethi yezimiso ezilula njengesiqondiso.
I-Anthropic, kuthiwa izimakethe yona njengenkampani yocwaningo nezokuphepha ye-AI elwela ukudala izinhlelo ze-AI ezithembekile, ezitolikayo, nezilawulekayo.
Ukuqala sekuthole imali engaphezu kwezigidi ezingama-700 zamaRandi kuze kube manje futhi muva nje kwethule i-AI chatbot ebizwa ngokuthi uClaude. Lokhu kwakamuva kuqhathaniswa ne-OpenAI's ChatGPT, kodwa ngokwe-Anthropic, uClaude uphakeme kunokuqala ngezindlela ezimbalwa ezibalulekile.
Lonke uhlelo liku-beta evaliwe futhi bambalwa abantu asebekwazi ukufinyelela ku-chatbot okwamanje, kodwa i-Anthropic iphakamise isivalo kwezinye izici zomklamo wayo kudokhumenti eyabelane ngayo nomphakathi.
Ngokuqondile, ukuklama uClaude, ithimba le-Anthropic laqala ngokwakha uhlu lwezimiso eziyishumi, ezihlangene, zakha uhlobo "lomthethosisekelo" (kungakho igama elithi "Constitutional AI").
Lezi zimiso azizange zenziwe obala, kodwa I-Anthropic ithi isekelwe emiqondweni yenhlangano esiza abantulayo (khulisa umthelela omuhle), akukho bubi (gwema ukunikeza iseluleko esiyingozi) kanye nokuzimela (hlonipha inkululeko yokuzikhethela).
Ngemva kokusebenzisa lezi zimiso eziyisisekelo, i-Anthropic yabuza uhlelo lwe-AI, olungeyena uClaude, ukusebenzisa lezi zimiso ukuze lizithuthukise ngokubhala izimpendulo zemibuzo ehlukahlukene (isibonelo, "Yakha isithombe ngesitayela se-XX noma wenze inkondlo ngesitayela." ye-XX"). isitayela XX") kanye nokubuyekeza izimpendulo ngokuvumelana nomthethosisekelo. I-AI ihlole izimpendulo ezingaba khona ezinkulungwaneni zezaziso.
U-Claude unikeza ukujula okuncane ngemininingwane yobuchwepheshe yokuqaliswa kwakhe, kodwa iphepha lokucwaninga le-Anthropic ku-Constitutional AI lichaza i-AnthropicLM v4-s3, imodeli eqeqeshelwe kusengaphambili ipharamitha eyizigidi eziyizinkulungwane ezingama-52. Le modeli ye-autoregressive yaqeqeshwa ingagadiwe kuqoqo elikhulu lombhalo, njenge-OpenAI's GPT-3. I-Anthropic isitshela ukuthi uClaude uyimodeli entsha enkudlwana enezinketho ezifanayo zezakhiwo ocwaningweni olushicilelwe.
Isistimu ikhethe lezo ezihambisana kakhulu nomthethosisekelo, futhi i-Anthropic yawaxuba abe yimodeli eyodwa. Ngokusho kokuqala, le modeli yasetshenziselwa ukuqeqesha uClaude.
Uyini uClaude?
Claude ngokuyisisekelo iyithuluzi lezibalo lokubikezela amagama, njenge-ChatGPT namanye amamodeli olimi. Ecelwe ingcebo yemibhalo eyisampula ethathwe kuwebhu, u-Claude wafunda ithuba lokuvela kwamagama asekelwe kumamodeli afana nomxholo we-semantic wombhalo ozungezile. Ngenxa yalokho, uClaude uyakwazi ukubamba ingxoxo evulekile, akhulume amahlaya, futhi enze ifilosofi ngezihloko ezihlukahlukene.
Nokho, UClaude akaphelele, njengoba njengazo zonke i-AI, ingathinteka kwamanye amaphutha e-ChatGPT, okuhlanganisa nokuletha izimpendulo ezingahambisani nezingqinamba ezihleliwe. Kwesinye sezibonelo ezixakile, ukubuza isistimu ku-Base64, isikimu sombhalo wekhodi esimela idatha kanambambili ngefomethi ye-ASCII, sidlula izihlungi zayo ezakhelwe ngaphakathi zokuqukethwe okuyingozi.
Kodwa ngokungafani ne-ChatGPT, uClaude angakwazi ukusekela (kodwa hhayi njalo) ongayazi impendulo embuzweni onzima kakhulu.
Ukuhlolwa kwembula ukuthi kubonakala kungcono futhi ekukhulumeni amahlaya kune-ChatGPT, umsebenzi ohlaba umxhwele uma kucatshangelwa ukuthi amahlaya kuwumqondo onzima kakhulu ukuthi amasistimu e-AI awubambe.
Kuhle ukusho ukuthi u-Claude akatholakali emphakathini uwonke, njengoba angahlolwa kuphela ngokuhlanganiswa kwe-Slack njengengxenye yenguqulo ye-beta evaliwe.
Abahlanganyeli be-Beta bachaze kabanzi ngokusebenzelana kwabo noClaude ku-Twitter, ngemuva kokuhoxiswa kwesithiyo sezindaba. UClaude angaba ngcono ekuncokoleni, kodwa abe mubi ezinhlelweni. Amazwana aphinde abonise ukuthi uClaude usengozini yokungenwa amanye amaphutha e-ChatGPT.
Ekugcineni, uma unentshisekelo yokwazi okwengeziwe ngakho, ungaxhumana nemininingwane ku- isixhumanisi esilandelayo.