Iza kukhuphisana njani i-OpenAI?

Ugqatso lwezixhobo ze-AI Lunomgca oMtsha ongaphambili — kwaye i-OpenAI ibuva ubushushu

Kuninzi luka-2023, i-OpenAI yayijongeka ingachukunyiswa. I-ChatGPT iye yawela abasebenzisi abazizigidi ezili-100 ngokukhawuleza kunaso nasiphi na isicelo somthengi embalini, kwaye ixabiso lenkampani yeedola ezingama-86 eebhiliyoni lenze ukuba ibe ngumthwali womgangatho ongenakuphikiswa wexesha le-AI yokuvelisa. Kodwa ishishini letekhnoloji linetalente ekhohlakeleyo yokuthoba iintshatsheli zalo, kwaye ekuqaleni kuka-2025, imeko yokhuphiswano yayitshintshile kakhulu kangangokuba nabona baxhasi banenzondelelo ye-OpenAI babebuza lo mbuzo ungakhululekanga: kwenzeka ntoni xa i-advanteji yomhambisi wokuqala iphela?

Impendulo yalo mbuzo ibalulekile ngaphaya kweSilicon Valley. Amakhulu amawaka amashishini - ukusuka kwi-solopreneurs esebenzisa izixhobo ze-AI-powered to Fortune 500 amaqela okuthengwa kwempahla - enza izigqibo zeqonga ngoku eziza kuchaza iziseko zabo zokusebenza kwiminyaka elishumi ezayo. Ukuqonda indlela i-OpenAI eceba ukukhuphisana ngayo, kwaye apho inokuthi ifike khona, ayiseyiyo nje inkcazo enomdla yetekhnoloji. Sisinyanzelo sobuchule kuye nabani na owakha ishishini langoku.

AbaXhasi abaSadlali ukubambisa

Xa uGoogle wayesungula iGemini Ultra ekuqaleni kuka-2024, iibenchmarks zabalisa ibali elinzima. Kweminye imisebenzi yokuqiqa, iye yagqwesa iGPT-4. Kwabanye, yaphela. Kodwa umqondiso wokwenyani wawungekho kumanani - wawusasazwe. UGoogle unokushumeka iimodeli zakhe ngokuthe ngqo kwiNdawo yokuSebenza, uPhendlo, i-Android, kunye neLifu ngeendlela i-OpenAI ingenako ukuphindaphinda ngaphandle kwentsebenziswano. Olo hlobo lokudityaniswa kweqonga kubunzulu ngumjelo okhuphisanayo ogcina abathengi beshishini benyanisekile nokuba yeyiphi imodeli egqwesileyo ngobuchule nangoLwesibini.

I-Anthropic's Claude 3 Opus ibonise ukuba i-AI egxile kukhuseleko inokuba nako ngokwenene, ukutsala udidi oluthile lwabathengi beshishini - iinkqubo zokhathalelo lwezempilo, iifemu ezisemthethweni, amaziko emali - abalinganisela ukuthotyelwa kwemithetho njengobunzima bokusebenza okukrwada. Ngeli xesha, isigqibo se-Meta sokuvula umthombo we-Llama 3 ngokusisiseko siphazamise ukuguquguquka kwamaxabiso kulo lonke ishishini. Xa imodeli ekwaziyo ukukhuphela kunye nokuzibamba, wonke umnikezeli wezorhwebo kufuneka athethelele iintlawulo zorhumo ngento engaphaya komgangatho wemodeli yedwa.

I-Mistral, i-Cohere, kunye neqela leenkwenkwezi elikhulayo leenkampani ezithe ngqo ezithe nkqo ze-AI bakrola i-niches ethi i-OpenAI's generalist positioning yenza kube nzima ukuyikhusela. Ukuqaliswa kwe-fintech akufuni ngokuyimfuneko eyona modeli inamandla yehlabathi-yenjongo jikelele - ifuna imodeli eqonda ulwimi lokuthotyelwa, ithetha kakuhle malunga nexabiso lezinto eziphuma kuyo, kwaye idibanise ngokucocekileyo kunye nedatha ekhoyo.

IiPivothi zeQhinga le-OpenAI: Eyona nto iBheja kuyo iNkampani

Impendulo ye-OpenAI kolu xinzelelo lokhuphisanayo iye yanamacala amaninzi, kwaye kufanelekile ukuthatha i-strand nganye nzulu kunokuba igxothe inkampani njengemveliso enye emangalisekileyo ejongene norhwebo olungenakuthintelwa. Elona nyathelo libalulekileyo lobuchule ibe kukutyhalela phambili kwiikontrakthi zeshishini - ngokukodwa ezinkulu, izivumelwano zeminyaka emininzi kunye nemibutho ezimisele ukuhlawula amaxabiso eprimiyamu okwenziwa ngokwezifiso, iziseko zophuhliso ezizinikeleyo, kunye nezivumelwano zomgangatho wenkonzo ezingenakubonelelwa ngeemveliso zodidi lwasimahla.

Utyalo-mali lwenkampani kwiimodeli zokuqiqa - uthotho lwe-o1 kunye ne-o3 - lubonisa umzamo wangabom wokuba ngumnikazi wesiphelo esiphezulu se-spectrum yesakhono. Ngelixa abakhuphisanayo baleqa umlinganiso webenchmark kwimisebenzi yolwimi oluqhelekileyo, i-OpenAI ibheja ukuba ukuqiqa kwamanyathelo amaninzi anzima kubonisa isakhono esahlukileyo esithethelela amaxabiso eprimiyamu. Ukwamkelwa kwangethuba ziifemu zomthetho zisebenzisa le mifuziselo kuhlalutyo lwekhontrakthi kunye neenkampani ezixuba amayeza ezizithumela kuphando lwentsebenziswano yamachiza zibonisa ukuba ubhejo lunokufaneleka okungagqibelelanga.

Mhlawumbi okona kubaluleke kakhulu, i-OpenAI ibithe cwaka isakha ubudlelwane bomgangatho wabasebenzi - ifakela iAPI yayo nzulu kwiimveliso zomntu wesithathu kangangokuba iindleko zokutshintsha zibe zithintelo. Xa i-chatbot ejongene nomthengi wakho, inkqubo yakho yolawulo lolwazi lwangaphakathi, kunye nomncedisi wakho wekhowudi zonke zisebenza kwi-API enye esisiseko, iindleko zokutshintshela azikho nje ezobuchwephesha - zezombutho, zeziko, kunye nobuntu obunzulu.

Ingxaki yokuDityaniswa kweShishini ekungekho mntu uthetha ngoKwaneleyo

Nalu ukhuphiswano oluguquguqukayo uninzi lwe-AI olugubungela ubunzima obungaphantsi: idabi lolawulo lwe-AI alikho ngokuyintloko malunga nemodeli yamanqaku aphezulu kwiibenchmarks zeMMLU. Kumalunga nokuba ngawaphi amandla e-AI adityaniswa kakhulu ngaphandle komthungo kumaqonga apho amashishini aqhuba imisebenzi yawo.

Iinkampani eziphumelele ixesha le-AI aziyi kuba zezo zineemodeli ezinamandla kakhulu - ziya kuba zizakhono zabo ze-AI zingabonakali ngenxa yokuba zingenakuhlukaniswa kwiinkqubo ezisetyenziswa ngabantu yonke imihla.

Lo mbono uphinda wenze imbonakalo yokhuphiswano lonke. Inkampani yezothutho ephakathi kunye nabasebenzi abangama-200 ayivavanyi abathengisi be-AI bebodwa - ibuza ukuba zeziphi izixhobo ze-AI ezisebenza ngaphakathi kweCRM yayo, inkqubo yayo yolawulo lwezithuthi, iqonga layo le-HR, kunye nesoftware yayo ye-invoyisi. Xa i-AI ifakwe kwinqanaba lokuhamba komsebenzi kunokuba inikezelwe njengojongano lwengxoxo oluzimeleyo, iba yinto encangathi kakhulu.

Amaqonga afana ne-Mewayz, edibanisa kwiimodyuli ze-207 zoshishino - ukusuka kwi-CRM kunye nokuhlawula kwiinkqubo zokubhuka kunye nezixhobo ze-link-in-bio - kubasebenzisi be-138,000 emhlabeni jikelele, bamele uhlobo lweziseko zokusebenza apho amandla e-AI kufuneka aphile. Amashishini aqhuba umbhobho wawo wokuthengisa, ukusebenza kwe-HR, kunye nokuhamba kwemali ngemodyuli enye ye-OS abakhangeli sixhobo se-AI esahlukileyo sokufunda. Bajonge ubukrelekrele obuvela ngokwemvelo kwiinkqubo abasele bezisebenzisa. Kungenxa yoko le nto owona mbuzo unomdla wokhuphiswano we-OpenAI ungenguye "yeyiphi imodeli eyona ilungileyo?" — ngukuthi "leliphi iqonga eliya kusishumayeza, kwaye ngawaphi amagama?"

Imilinganiselo emihlanu apho uKhuphiswano luza kugqitywa khona

Ukucutha ingxolo yeemfazwe zebhenchmark kunye nezibhengezo zenkxaso-mali, idabi lokhuphiswano lolawulo lwe-AI kwimarike yeshishini liya kugqitywa ngemilinganiselo emihlanu yekhonkrithi:

1. Ifestile yomxholo kunye nokwakhiwa kwememori - Ukuhamba komsebenzi weshishini kufuna i-AI ekhumbula umxholo kwiiseshoni, iiprojekthi kunye nabasebenzisi. Nabani na osombulula inkumbulo ethe gqolo, enokufunyanwa kwisikali uyaliphumelela ishishini.
2. Imodeli yokutsha yexabiso - Ukubhaliswa kwe-Flat-rate kunika indlela yokusetyenziswa kwexabiso, ixabiso elisekelwe kwisiphumo, kunye neemodeli ezixubileyo. Ukuguquguquka kwamaxabiso e-OpenAI kuya kuqinisekisa ukuba ngawaphi amacandelo emarike anokusebenza ngenzuzo.
3. Ubunzulu be-Multimodal — Ukukwazi ukucubungula amaxwebhu, imifanekiso, i-audio, kunye nevidiyo ngaxeshanye akuyona into emnandi kumashishini amaninzi - ngumahluko phakathi kwesixhobo esincedayo kunye nesixhobo esiguqula ukusebenza.
4. Ukuthotyelwa kunye nokuhlala kwedatha - Amashishini aseYurophu asebenza phantsi kwe-GDPR, imibutho yezempilo ephantsi kwe-HIPAA, kunye namaziko emali phantsi kweendlela ezahlukeneyo zokulawula zidinga i-AI enokuqinisekisa apho idatha yabo ihlala khona kunye nendlela ephathwa ngayo.
5. Ubunzulu bePlatform ecosystem — Njengoko kuxoxiwe, eyona AI incangathi yi-AI emiliselwe nzulu kwimisebenzi yemihla ngemihla. Inkampani eyakha eyona nkqubo ityebileyo yokudityaniswa kwendalo - okanye ibe ngumaleko wobukrelekrele obungagqibekanga ngaphakathi kwamaqonga oshishino aphambili - kuya kuba nzima ukuyisusa.

I-OpenAI inezikhundla ezinentsingiselo kuninzi lwale milinganiselo kunye nobuthathaka obumandundu kwabanye. Izakhono zayo ezininzi kunye ne-GPT-4o ziyamangalisa ngokwenene. Iziseko zayo zokuthobela ziye zaphucuka kodwa zihlala zisemva kwabo bakhuphisana nabalumkileyo. I-ecosystem yayo ixhomekeke kakhulu kubaphuhlisi beqela lesithathu abaya bephonononga iindlela ezizezinye.

I-Open-Source Wildcard Akukho mntu unexabiso elipheleleyo ngaphakathi

Esona sisongelo sokhuphiswano esingaxatyiswanga kakhulu kwi-OpenAI - kunye neshishini lonke le-AI yorhwebo - kukuvuthwa okuqhubekayo kweemodeli zomthombo ovulekileyo. Uthotho lwe-Meta's Llama, ukukhutshwa okuvulelekileyo kwe-Mistral, kunye nenani elikhulayo leemodeli eziphuhliswe ekuhlaleni zenze enye indlela yokwenyani ye-AI yorhwebo kwimibutho enezixhobo zobugcisa.

Xa inkampani ikwazi ukulungisa kakuhle imodeli yomthombo ovulekileyo kwidatha yayo yobunikazi, iyibeke kwisiseko sayo, kwaye igweme iindleko ze-API ye-token ngokupheleleyo, ixabiso lexabiso le-APIs zorhwebo liba nzima ukulichaza kuphela ngenxa yezizathu zokukwazi. Abadlali bezorhwebo - i-OpenAI ibandakanyiwe - kufuneka banikeze into evulekile-umthombo ongenako ukuphindaphinda ngokulula: iziqinisekiso ezithembekileyo, ukucoca ukhuseleko, ukuhlaziywa okuqhubekayo ngaphandle kobuchule be-ML yangaphakathi, kunye nenkxaso yokudibanisa.

Kumashishini amancinci kunye nee-solopreneurs ezisebenza ngaphakathi kwamaqonga akhupha ubunzima be-AI, lo mahluko ubhalisa kancinci. Umnini weshishini olawula yonke imisebenzi yakhe ngeqonga elidityanisiweyo elifana ne-Mewayz akufuneki avavanye imodeli ye-AI yoyilo - bafuna iCRM yabo ukuba iveze ulwazi oluchanekileyo, ii-invoyisi zabo kwiflegi engaqhelekanga, kunye nenkqubo yabo yokucwangcisa ukuze bazenzele ngokuzenzekelayo. I-AI yingxaki yeqonga. Olo hlobo lomaleko wokuthabatha, apho imodeli esisiseko ingenamsebenzi kumsebenzisi wokugqibela, ngowona mhlobo ubalaseleyo weshishini le-AI kunye nomngeni walo omkhulu wexesha elide.

Kuthetha ukuthini oku kumashishini enza iziGqibo ze-AI namhlanje

Ukuba uyinkokeli yeshishini uzama ukuzulazula kumhlaba we-AI ngelixa i-OpenAI kunye nabo bakhuphisana nabo besalungisa indawo yabo, imigaqo embalwa ibambe nokuba imarike ishukuma njani ekugqibeleni:

• Beka phambili udibaniso ngaphezulu kwesakhono — I-AI ephucula ukuhamba kwakho kokwenyani ibetha imodeli egqwesileyo ngokwethiyori ongazange uyicinge kakuhle indlela yokusasaza.
• Kuphephe ukutshixelwa komthengisi omnye kuluhlu lwemodeli — Yakha iinkqubo zakho malunga neziphumo kunye nokuhamba komsebenzi, kungekhona malunga neemodeli ezithile ze-API ezinokutshintsha amaxabiso okanye ukufumaneka.
• Vavanya i-AI njengenxalenye yesitaki sakho sokusebenza, singahlukani nayo — Eyona nto ixabisekileyo yotyalo-mali lwe-AI ikhompawundi xa ifakwe kwiinkqubo ezisetyenziswa liqela lakho rhoqo ngeyure, ingafikelelwanga ngengxoxo eyahlukileyo.
• Funa ukucaca ekuphathweni kwedatha — Ukuqonda indlela idatha yeshishini lakho esetyenziselwa ngayo uqeqesho lwemodeli, apho igcinwa khona, kwaye ngubani onokufikelela kuyo akusiyo i-paranoia - yinqobo yokusebenza ngenxa yokunyamekela ngo-2025.
• Cinga ekuhambeni komsebenzi, kungekhona iimpawu - Umzekelo onokubhala ikhowudi, ushwankathele amaxwebhu, kwaye uphendule imibuzo uyamangalisa. Iqonga elisebenzisa ezo zakhono ukwenza ngokuzenzekelayo inkqubo yokwamkelwa kwe-invoyisi, iflegi yee-akhawunti zabathengi abasemngciphekweni, kunye nemibono yeHR enokusetyenzwa iyayiguqula.

Umdlalo omde: Ukudityaniswa, uRhwebo, kunye nento eSindayo

Ukuba imbali inika nayiphi na isikhokelo, ukwanda kwangoku kweemodeli ze-AI kunye namaqonga ekugqibeleni kuya kudibanisa. Siyibonile le pateni kwicomputing yamafu, kwisoftware yeCRM, kwiziseko zophuhliso lwe-e-commerce - ixesha lokwahlukana okuqhumayo kulandelwa kukudityaniswa kweqela labadlali ababalaseleyo, ngamnye esebenza kumacandelo orhwebo ahlukeneyo anezindululo zexabiso ezahlukeneyo.

Eyona ndlela ingcono ye-OpenAI yempilo ekhuphisanayo yexesha elide inokuthi iqhube ngezicwangciso ezibini ngaxeshanye: ukuba ngumaleko wokuqiqa osisiseko kwezona meko ziphezulu zokusetyenziswa kweshishini apho iprimiyamu yesakhono kufuneka ihlawulwe, kunye nokwakha okanye ukufumana uhlobo lobunzulu beqonga elenza ukuhamba komsebenzi kwinqanaba lokutshixa. Inkampani inenkunzi, italente, kunye nokuqondwa kophawu ukwenza nasiphi na isicwangciso. Nokuba inokuziphumeza zombini, ngokukhawuleza ngokwaneleyo, ngokuchasene nabakhuphisana abanemali eninzi abangasadlaliyo ukubambisa — oko kukungaqiniseki kokwenyani.

Kumashishini angama-138,000 asebenzisa amaqonga asebenzayo adibeneyo afana ne-Mewayz ukuqhuba umsebenzi wabo wemihla ngemihla, amandla okukhuphisana phakathi kwababoneleli be-AI yinto yesibini. Okubalulekileyo kukuba amaqonga axhomekeke kuwo enza izigqibo ezikrelekrele malunga nokuba yeyiphi i-AI yokudibanisa kunye nendlela yokwenza ezo zakhono zizive zingaphantsi njengesixhobo esahlukileyo kwaye zifana nobukrelekrele obungabonakaliyo obulukiweyo kuwo wonke umsebenzi. Njengoko iimfazwe ze-AI zitshisa, amashishini aza kuzuza kakhulu ngawo amaziko okusebenza akhiwe ukuba abe yimodyuli ngokwaneleyo yokuphucula i-intelligence layer njengoko imarike iguquguquka - kwaye ibhetyebhetye ngokwaneleyo ukuyibuyisela ukuba into engcono iza kunye.

Imibuzo Ebuzwa Rhoqo

Ngoobani abona bakhuphisana nabakhulu kwi-OpenAI kwimarike ye-AI ngoku?

OpenAI ngoku ijongene noxinzelelo oluvela kwiindlela ezininzi. I-Google iye yavuselela iinzame zayo ze-AI kunye ne-Gemini, i-Anthropic's Claude iye yaba ngumbangi weshishini, kwaye i-Meta isasaza iimodeli ezivulekileyo ezinciphisa indlela ye-API ehlawulweyo ngokupheleleyo. Ngaphakathi, iziqalo ezifana neMistral kunye ne-Cohere ziqingqa i-niche dominance, ngelixa kumazwe ngamazwe, iilebhu zaseTshayina zivale umsantsa wokukwazi ngokukhawuleza kunokuba uninzi lwabahlalutyi bebeqikelele.

Yintoni eyanika i-OpenAI inzuzo yayo yokuqala, kwaye ingaba isazinzile?

Intloko yentloko ye-OpenAI yavela ekubeni ngowokuqala ukuvelisa imifuziselo yolwimi olukhulu ngokomlinganiselo — Ukuqaliswa kwentsholongwane ye-ChatGPT ekupheleni kuka-2022 kwaseka udidi. Kodwa izibonelelo zokuqala zokuhambisa kwisoftware zitshabalalisa ngokukhawuleza. Ngeembangi ngoku ezihambelana nokusebenza kwinqanaba le-GPT-4 kwaye zihlala zibonelela ngamaxabiso aphantsi, i-OpenAI's moat ixhomekeke kwi-ecosystem yayo, ubudlelwane beshishini, kunye nesantya esiqhubekayo se-R&D - akukho nanye eqinisekisiweyo.

Amashishini kufuneka asabele njani kwinkangeleko ye-AI eguquka ngokukhawuleza?

Amashishini kufuneka aphephe ukutshixelwa kwabathengisi kwaye endaweni yoko amkele amaqonga akhelwe ukudibanisa i-AI ngokuguquguqukayo. Izixhobo ezifana ne-Mewayz - i-OS ye-modyuli ye-207 yeshishini efumaneka kwi- $ 19 / ngenyanga kwi-app.mewayz.com - yenzelwe ukulungelelanisa njengoko i-AI ephantsi ye-landscape iguquguquka, inika abaqhubi ukuhamba komsebenzi abakufunayo ngaphandle kokubheja isitaki sabo siphela kumboneleli omnye oqhubekayo olawulayo. Ukuba bhetyebhetye kuphela kweheji ethembekileyo.

Ngaba i-OpenAI ingakwazi ukugcina isikhundla sayo sobunkokeli ixesha elide?

Ubunkokheli be-OpenAI buyakhutshiswana okokuqala. Ukuyigcina kuya kufuna okungaphezulu kunokukhupha i-GPT iteration elandelayo — ifuna isixhobo esisinyanzelo somphuhlisi, intembeko yeshishini elingangeni moya, kunye nempendulo ethembekileyo kumthombo ovulekileyo wosoyikiso. I-OpenAI inetalente engaqhelekanga kunye nemali eyinkunzi emva kwayo, kodwa ixesha lolawulo olungenamsebenzi liphelile. Isahluko esilandelayo siya kumiselwa ngokubulawa, kungekhona ngokukhawuleza.