Khawube nomfanekiso wekamva apho i-AI ingatshixwanga kumagumbi angaphantsi komhlaba, kodwa yakhiwe kwindawo evulekileyo, isitena ngesitena, luluntu lwehlabathi lwabavelisi. Apho intsebenziswano, hayi ukhuphiswano, ibasa ukuqhubela phambili, kunye neengqwalaselo zokuziphatha zinobunzima obulinganayo kunye nokusebenza okukrwada. Le asiyontsomi yenzululwazi, luguqulo oluvulekileyo lwemithombo evulekileyo entliziyweni yophuhliso lwe-AI. Kodwa iBig Tech ine-ajenda yayo, ifihla iimodeli ezithintelweyo njengomthombo ovulekileyo ngelixa uzama ukuvuna izibonelelo zoluntu oluvuleleke ngokwenene.
Masiyixovule emva iileya zekhowudi kwaye sityhile inyani emva kwezi nzame. Olu phononongo lwekamva lomthombo ovulekileyo we-AI luya kucazulula "abazenzisi" kwaye lutshatshelise "abokwenyani" kuphuhliso lwe-AI ukufumanisa i-injini yokuvelisa eyisoftware evulekileyo encwinayo phantsi kwayo yonke into. Eyona nto iphambili kukuba i-AI yomthombo ovulekileyo iya kuzala isitaki sedatha evulekileyo.
Isidingo
Inqaku lakutshanje nguMatteo Wong kwiAtlantiki, ' Akuzange kubekho nto inje ngokuba 'Vula' i-AI ' ichaza indlela ekhulayo kwizifundiswa kunye noluntu lwesoftware yomthombo ovuleleke ngokwenene we-AI. "Umbono kukudala imodeli ecacileyo nenokuthi uluntu luyisebenzise ngokulula nangexabiso eliphantsi, lufunde, kwaye luvelise kwakhona, ukuzama ukwenza idemokhrasi itekhnoloji egxile kakhulu enokuba namandla okuguqula umsebenzi, amapolisa, ulonwabo kunye nonqulo." Kwaloo Atlantic inye icebisa ukuba iinkampani zeBig Tech ezifana neMeta zizama ukuzalisa le mfuno kwintengiso 'ngokuhlamba evulekileyo' iimveliso zazo. Bathatha iimpawu kunye nodumo oluhle loluntu oluvulelekileyo ngaphandle kokuvula ngokwenene imveliso yabo. Kodwa, akukho nto ithatha indawo yento yokwenyani. Oku kungenxa yokuba isoftware yomthombo ovulekileyo wokwenyani iqhuba ukuqamba izinto ezintsha kunye nentsebenziswano: iimpawu ezimbini ezifuneka ngamandla ukuqhubela phambili nge-AI ngenkathalo.
Abazenzisi
I-LLaMA 2, imodeli yolwimi enkulu eyenziwe yi -Meta ekhululekile ukuyisebenzisela zombini uphando kunye nokusetyenziswa kwezorhwebo. Ukukhokelela abanye ukuba bacebise i-LLaMA 2 ngumthombo ovulekileyo. Nangona kunjalo, iMeta iphumeze izithintelo ezithile eziqatha ekusebenziseni imodeli yazo. Umzekelo, i-LLaMA 2 ayinakusetyenziswa ekuphuculeni nayiphi na enye imodeli yolwimi olukhulu. Isikhundla esichasene nesithethe imodeli entsha edibeneyo yabucala yesoftware evulekileyo ekhuthaza ukutyhilwa kwasimahla kunye nokuvulelekileyo kokusungula izinto ezintsha ukuze kuxhamle wonke umntu kuluntu lwesoftware.
I-Meta iphinde yakhubaza ukusetyenziswa kwemodeli yabo ngokungavumeli ukudityaniswa kwe-LLaMA 2 kunye neemveliso ezinabasebenzisi abazizigidi ezingama-700 ngenyanga kwaye ngokungavezi ukuba yeyiphi idatha imodeli yabo eqeqeshelwe kuyo okanye ikhowudi abayisebenzisileyo ukuyakha. Ngokungabhengezi, iMeta izivulela imibuzo yocalucalulo lwendalo kunye nocalucalulo ngengozi. Imodeli eqeqeshwe kwidatha yocalucalulo iya banike iimpendulo zocalucalulo . Ngaphandle koluntu lwesoftware ngokubanzi lukwazi ukujonga nokuba ikhowudi esetyenziselwa ukwakha imodeli ukuze ibone ukuba kukho naziphi na izikhuselo ezakhiweyo okanye idatha esetyenziselwa ukuyiqeqesha, sishiywe ebumnyameni kule mibuzo yokuziphatha. Ngexesha apho uphando olupapashiweyo kwi-AI ixhalabele kakhulu ukusebenza kunobulungisa kwaye intlonipho le nto ithethwayo iphazamisa kakhulu.
AbaBonene
I-Mistral AI ifumene ukuqondwa kwiimodeli zayo zeelwimi ezinkulu ezinomthombo ovulekileyo, ngakumbi iMistral 7B kunye neMixtral 8x7B. Inkampani izama ukuqinisekisa ukufikeleleka okubanzi kwiimodeli zayo ze-AI, ukukhuthaza uphononongo, ukuguqulwa, kunye nokusetyenziswa kwakhona luluntu lwesoftware evulekileyo.
vLLM imele "vectorized low-latency model serving" kwaye lithala leencwadi elivulelekileyo eliyilelwe ngokukodwa ukukhawulezisa kunye nokwandisa iimodeli zolwimi ezinkulu (LLMs). Sisixhobo esinamandla esinokuphucula kakhulu ukusebenza kunye nokusetyenziswa kweeLLMs. Oku kuyenza i-asethi exabisekileyo kubaphuhlisi abasebenza kwiintlobo ezahlukeneyo zezicelo ze-AI, ukusuka kwii-chatbots kunye nabancedisi benyani ekudalweni komxholo kunye nokuveliswa kwekhowudi. Kangangokuba, iMistral icebisa ukuba kusetyenziswe i-vLLM njengomncedisi we-inference yeemodeli ze-7B kunye ne-8x7B.
EleutherAI yilebhu yophando ye-AI engenzi nzuzo ekhule ukusuka kwiseva yeDiscord ukuxoxa nge-GPT-3 ukuya kumbutho ophambili wophando ongenzi nzuzo. Eli qela laziwa ngomsebenzi walo wokuqeqesha nokukhuthaza izithethe ezivulelekileyo zenzululwazi kwiNatural Language Processing. Bakhuphe iimodeli zolwimi ezinkulu ezinomthombo ovulekileyo kwaye babandakanyeka kwiiprojekthi zophando ezinxulumene nolungelelwaniso lwe-AI kunye nokutolika. Yabo LM-Ukubopha Iprojekthi mhlawumbi sesona sixhobo siphambili sovavanyo lwemithombo evulekileyo yemizekelo yolwimi.
Phi-2 yi-LLM yeMicrosoft ebetha ngaphezulu kobunzima bayo. Uqeqeshelwe kumdibaniso wemibhalo eyenziweyo kunye neewebhusayithi ezihluziweyo, le modeli incinci, kodwa inamandla iyagqwesa kwimisebenzi efana nokuphendula imibuzo, ukushwankathela, kunye nokuguqulela. Eyona nto ibeka i-Phi-2 ngokwahlukileyo kugxininiso lwayo ekuqiqeni nasekuqondeni ulwimi, okukhokelela ekusebenzeni okuchukumisayo nangaphandle kobuchule bolungelelwaniso obuphambili.
Iimodeli ezininzi ezifanelekileyo zokuzinzisa umthombo ovulekileyo zomeleza indawo evulelekileyo ye-AI yokuvelisa. Ezi ziimeko zangoku zobuchule bomthombo ovulekileyo kwaye zibandakanya I-UAE-Enkulu-V1 kwaye iilwimi ezininzi-e5-ezinkulu .
Baninzi ngakumbi kule ntsimi ikhula rhoqo. Olu luhlu lulinganiselweyo sisiqalo nje.
Open Source Drives Innovation
Ukwamkela intanda-bulumko yokuvula ngokugqithisileyo okuvulelekileyo, iinkampani ezithatha inxaxheba ngokwenene kuphuhliso lwesoftware evulelekileyo zicela umngeni kwiimbono zemveli zenzuzo yokhuphiswano ngokuvuma ukuba ayizizo zonke iikhowudi ezilungileyo okanye iingcamango ezintle ezihlala ngaphakathi kwintlangano yabo . Olu tshintsho luxhasa i ingxoxo ekwabelwana ngezinto ezintsha ngaphakathi kwenkqubo yendalo evulelekileyo ikhokelele ekukhuleni ngokukhawuleza kwemarike, ibonelela ngeefemu ezincinci zesoftware ezinemali yeR&D enyiniweyo. ithuba lokuxhamla ukusuka kwi-R&D spillovers ekhoyo kwisoftware evulelekileyo. Oku kungenxa yokuba, ngokuchaseneyo nokukhutshwa kwemveli, ukuqaliswa okuvulekileyo iphucula izibonelelo zangaphakathi ngokusebenzisa ubukrelekrele obuhlangeneyo boluntu, ngaphandle kokunciphisa iinzame zeR&D zangaphakathi. Oku kuthetha ukuba iinkampani zesoftware enomthombo ovulekileyo akufuneki zincame uhlahlo lwabiwo-mali ukuze zilandele ubunkokeli bengcinga kunye nekhowudi ngaphandle kombutho wazo.
Ukongeza, iinkampani zesoftware enomthombo ovulekileyo ziqhuba ngobuchule ukuqamba izinto ezintsha ukukhupha ikhowudi kwangoko kwaye rhoqo , ukuqaphela imo eyongezelekayo yenkqubo yokuveliswa kwezinto ezintsha kuluntu lwesoftware. Yonke into yokuthetha into abaninzi abasele beyibonile: Isoftware yoMthombo ovulekileyo iqhuba izinto ezintsha.
Umthombo ovulekileyo ukhuthaza iNtsebenziswano
Ngokusebenzisa uthungelwano kuluntu lwesoftware evulekileyo, oosomashishini bayakwazi ukuzalisekisa iinjongo zexesha elifutshane kunye nexesha elide. Iinjongo zengeniso yexesha elifutshane zakha iinkampani kunye neenjongo zenzuzo yexesha elide ziyazigcina. Kwangaxeshanye, lo mzamo wothungelwano uziqhubela wona uthungelwano ngokwalo - lukhulisela usomashishini olandelayo. Kuyaziwa ukuba iiplatifomu ezivulekileyo zibonelela ngokufikelela kwikhowudi yomthombo, okwenza abaphuhlisi benze ukuphuculwa, iiplagi kunye nezinye iziqwenga zesoftware kwaye basebenzise ngokweemfuno zabo. Olu hlobo lwentsebenziswano lufumene i-boom ngokwamkelwa okubanzi kwe-Kubernetes luluntu olubanzi lwesoftware. Ngoku kunanini na ngaphambili, ubugcisa bezi mini busebenza kunye ngaphandle kongquzulwano oluncinane kakhulu yaye zinokubakho ngemizuzu kunye phantse naphi na.
Iinkampani zeBig Tech ziyayivuma le ntsebenziswano inzulu ehambelana noluntu oluvulelekileyo xa zikhupha ngokukhululekileyo izikhokelo, amathala eencwadi, kunye neelwimi ezizenzileyo ukugcina nokuphuhlisa izixhobo zangaphakathi. Ukwenza njalo kukhulisa iqela labaphuhlisi abakwaziyo ukusebenza kwiimveliso zabo kwaye baqalise ukuseta umgangatho wendlela ubugcisa obufanayo ekufuneka busebenze ngayo. Kwa elo nqaku le-Atlantiki licaphula umseki weMeta uMark Zuckerberg esithi "ixabiseke kakhulu kuthi ukubonelela ngokuba ngoku bonke abona baphuhlisi babalaseleyo kulo lonke ishishini basebenzisa izixhobo esizisebenzisayo ngaphakathi".
Umthombo oVulekileyo uvelisa uMthombo oVulekileyo
Ezi zizizathu zokuba kutheni sihlala sibona intsebenziswano phakathi kweenkampani ezinomthombo ovulekileyo. Umthombo ovulekileyo we-AI kunye neenkampani ze-ML ziya kuphuhlisa izisombululo ngokwemvelo kunye nezinye iimveliso ezivulekileyo ezivela kwiimveliso ezisisiseko ezifana nokugcinwa kwento ukuya kuyo yonke i-stack ukuya kwizixhobo zokubonisa. Xa enye inkampani enomthombo ovulekileyo ihambela phambili, sonke senza njalo. Le ndlela yokubambisana kunye nedibeneyo mhlawumbi yeyona nto ingcono yokubheja kwi-AI ethatha indlela egxile emntwini. La mandla endalo ahambelana nemfuno yemarike yomthombo ovulekileyo we-AI edityaniswe neempawu zesoftware yomthombo ovulekileyo wobuchule obutsha kunye nentsebenziswano iya kuqhuba umthombo ovulekileyo wesitaki sedatha ye-AI.
Nceda ujoyine kwaye ube negalelo kule ncoko kunye noluntu lwethu ngokuthumela i-imeyile ku [email protected] okanye usithumelele umyalezo kwijelo lethu leSlack .
Kwakhona ipapashwe apha .