Cabanga ngekusasa lapho i-AI ingakhiyelwanga ezindaweni ezingaphansi zezinkampani, kodwa yakhiwe endaweni evulekile, isitini ngesitini, umphakathi womhlaba wonke wabasunguli. Lapho ukubambisana, hhayi ukuncintisana, kubhebhezela intuthuko, futhi ukucatshangelwa kwezimiso zokuziphatha kunesisindo esilinganayo nokusebenza okungahluziwe. Lena akuyona inganekwane yesayensi, inguquko evulekile yomthombo ovulekile enhliziyweni yokuthuthukiswa kwe-AI. Kodwa iBig Tech ine-ajenda yayo, ifihla amamodeli akhawulelwe njengomthombo ovulekile kuyilapho izama ukuzuza izinzuzo zomphakathi ovuleke ngempela.
Ake sihlaziye izingqimba zekhodi futhi sembule iqiniso ngemuva kwale mizamo. Lokhu kuhlola ikusasa le-AI yomthombo ovulekile kuzohlakaza “abazenzisi” futhi kunqobe “abangempela” ekuthuthukisweni kwe-AI ukuze kwembule injini emisha eyisoftware yomthombo ovulekile evungama ngaphansi kwayo yonke into. Okubalulekile ukuthi i-AI yomthombo ovulekile izozala isitaki sedatha yomthombo ovulekile.
Isidingo
Isihloko sakamuva sikaMatteo Wong kuThe Atlantic, ' Akukaze Kwaba Nento Enjengokuthi 'Vula' i-AI ' ichaza inkambiso ekhulayo kwezemfundo kanye nomphakathi wesoftware womthombo ovulekile ngempela we-AI. “Umbono uwukwakha amamodeli asobala uma kuqhathaniswa umphakathi ongawasebenzisa kalula futhi kalula, uwafunde, futhi uwakhiqize kabusha, uzama ukwenza intando yeningi ubuchwepheshe obugxile kakhulu obungase bukwazi ukuguqula umsebenzi, amaphoyisa, ukuzijabulisa ngisho nenkolo.” Yona leyo Atlantic iphakamisa ukuthi izinkampani zeBig Tech ezifana ne-Meta zizama ukugcwalisa lesi sidingo emakethe 'ngokugeza ngokuvula' imikhiqizo yazo. Bathatha izimfanelo nedumela elihle lomphakathi womthombo ovulekile ngaphandle kokuvula umkhiqizo wabo ngempela. Kodwa, akukho okungathatha indawo yento yangempela. Lokhu kungenxa yokuthi isoftware yomthombo ovulekile weqiniso ishayela ukuqamba okusha nokusebenzisana: izimfanelo ezimbili ezidingeka kakhulu ukuze siqhubekele phambili nge-AI ngokuzibophezela.
Abazenzisi
I-LLaMA 2, imodeli yolimi enkulu eyakhiwe yi -Meta ekhululekile ukuyisebenzisela kokubili ucwaningo nokusetshenziswa kwezentengiso. Ukuholela abanye ekuphakamiseni i-LLaMA 2 ingumthombo ovulekile. Kodwa-ke, i-Meta isebenzise imikhawulo ethile eqinile ekusetshenzisweni kwemodeli yabo. Isibonelo, i-LLaMA 2 ayikwazi ukusetshenziselwa ukuthuthukisa noma iyiphi enye imodeli yolimi olukhulu. Isikhundla esiphambene nendabuko imodeli yokuqamba ehlanganisiwe yangasese yesofthiwe evulekile ekhuthaza ukwambulwa kwamahhala nokuvulekile kokuqanjwa kabusha ukuze kuzuze wonke umuntu emphakathini wesofthiwe.
I-Meta iphinde yakhubaza ukusetshenziswa kwemodeli yabo ngokungavumeli ukuhlanganiswa kwe-LLaMA 2 nemikhiqizo enabasebenzisi banyanga zonke abayizigidi ezingu-700 futhi ngokungavezi ukuthi iyiphi idatha imodeli yabo eqeqeshwe ngayo noma ikhodi abayisebenzise ukuyakha. Ngokungadaluli, i-Meta izivulela imibuzo yokuchema okungokwemvelo nokucwaswa ngephutha. Imodeli eqeqeshelwe idatha yokucwasa izoba nikeza izimpendulo ezicwasayo . Ngaphandle kokuthi umphakathi we-software uwonke ukwazi ukubuka ikhodi esetshenziswe ukwakha imodeli ukuze ubone ukuthi ingabe zikhona yini izivikelo ezakhelwe ngaphakathi noma idatha esetshenzisiwe ukuyiqeqesha, sishiywa ebumnyameni kule mibuzo yokuziphatha. Esikhathini lapho ucwaningo olushicilelwe nge-AI ukhathazeke kakhulu ngokusebenza kunobulungisa futhi ukuhlonipha lokhu kufiphazwa kuyaphazamisa kakhulu.
Abangempela
I-Mistral AI ithole ukuqashelwa ngezinhlobo zayo zezilimi ezinkulu zomthombo ovulekile, ikakhulukazi i-Mistral 7B ne-Mistral 8x7B. Inkampani ilwela ukuqinisekisa ukufinyeleleka okubanzi kumamodeli ayo e-AI, ukubuyekezwa okukhuthazayo, ukuguqulwa, nokusebenzisa kabusha umphakathi wesoftware evulekile.
i-vLLM imele "vectorized low-latency model serving" futhi ingumtapo wolwazi ovulekile oklanyelwe ngokukhethekile ukusheshisa nokuthuthukisa amamodeli olimi amakhulu (LLMs). Kuyithuluzi elinamandla elingathuthukisa kakhulu ukusebenza nokusebenziseka kwama-LLM. Lokhu kuyenza ibe yimpahla ebalulekile yonjiniyela abasebenza kuzinhlelo zokusebenza ezihlukene ze-AI, kusukela kuma-chatbots nabasizi ababonakalayo kuye ekudaleni okuqukethwe kanye nokwenziwa kwamakhodi. Kangangoba, iMistral incoma ukusebenzisa i-vLLM njengeseva ye-inference yamamodeli we-7B kanye ne-8x7B.
EleutherAI iyilebhu yocwaningo ye-AI engenzi nzuzo ekhule isuka kuseva ye-Discord ukuze ixoxe nge-GPT-3 ukuya enhlanganweni ehamba phambili yocwaningo engenzi nzuzo. Iqembu laziwa ngomsebenzi walo wokuqeqesha kanye nokukhuthaza izinkambiso zesayensi evulekile ku-Natural Language Processing. Bakhiphe izinhlobo ezahlukene zezilimi ezinkulu ezitholakala emithonjeni evulekile futhi babambe iqhaza kumaphrojekthi ocwaningo ahlobene nokuqondanisa kwe-AI nokutolika. Yabo I-LM-Harness iphrojekthi cishe iyithuluzi elihamba phambili lokuhlola imithombo evulekile yamamodeli olimi.
Phi-2 yi-LLM yakwaMicrosoft eshaya ngaphezu kwesisindo sayo. Uqeqeshwe ngenhlanganisela yemibhalo yokwenziwa namawebhusayithi ahlungiwe, le modeli encane, kodwa enamandla idlula imisebenzi efana nokuphendula imibuzo, ukufingqa, nokuhumusha. Okwenza ngempela i-Phi-2 ihluke ukugxila kwayo ekucabangeni nasekuqondeni kolimi, okuholela ekusebenzeni okumangazayo ngisho nangaphandle kwamasu okuqondanisa athuthukile.
Amamodeli amaningi anekhono okushumeka umthombo ovulekile aqinisa isikhala somthombo ovulekile we-AI okhiqizayo. Lezi yisimo samanje somthombo ovulekile futhi zifaka I-UAE-Large-V1 futhi izilimi eziningi-e5-ezinkulu .
Baningi abengeziwe kulo mkhakha okhula njalo. Lolu hlu olulinganiselwe luyisiqalo nje.
I-Open Source Drives Innovation
Ukwamukela ifilosofi yokuqamba okusha okuvuleleke ngokwedlulele, izinkampani ezibamba iqhaza ngempela ekuthuthukisweni kwesoftware yomthombo ovulekile ziphonsela inselelo imibono yendabuko yenzuzo yokuncintisana ngokuvuma ukuthi akuwona wonke amakhodi amahle noma imibono emihle ehlala ngaphakathi kwenhlangano yabo . Lolu shintsho lusekela i- ukuphikisana okwabelane ngokusungula izinto ezintsha ngaphakathi kwe-open-source ecosystem kuholela ekukhuleni kwemakethe ngokushesha, ukuhlinzeka amafemu ama-software amancane kakhulu ngezimali ezilinganiselwe ze-R&D ithuba lokuhlomula kusuka kuma-spillovers e-R&D akhona kusofthiwe yomthombo ovulekile. Lokhu kungenxa yokuthi, ngokungafani nokukhishwa kwendabuko, ukusungula okusha okuvulekile ithuthukisa izinsiza zangaphakathi ngokusebenzisa ubuhlakani obuhlangene bomphakathi, ngaphandle kokunciphisa imizamo yangaphakathi ye-R&D. Okusho ukuthi izinkampani zesoftware enomthombo ovulekile akudingeki zidele amabhajethi azo ukuze ziphishekele ubuholi bemicabango kanye nekhodi ngaphandle kwenhlangano yazo.
Ukwengeza, izinkampani zesoftware yemithombo evulekile zishayela ngamasu ukuqamba okusha ukukhipha ikhodi kusenesikhathi futhi kaningi , ibona imvelo eqongelelekayo yenqubo yokuqamba kabusha emphakathini wesoftware. Konke okusho okuthile abaningi asebekuqaphela kakade: I-Open Source Software ishayela emisha.
Ukusebenzisana kwe-Open Source Fosters
Ngokusebenzisa inethiwekhi emphakathini wesofthiwe yemithombo evulekile, osomabhizinisi bayakwazi ukufeza imigomo yesikhathi esifushane neyesikhathi eside. Imigomo yenzuzo yesikhathi esifushane yakha izinkampani kanye nemigomo yenzuzo yesikhathi eside iyazisekela. Ngesikhathi esifanayo, lo mzamo wokuxhumana uziqhubekisela phambili inethiwekhi ngokwayo - ukuyikhulisa kusomabhizinisi olandelayo. Kuyaziwa ukuthi amapulatifomu avulekile ahlinzeka ngokufinyelela ikhodi yomthombo, okwenza abathuthukisi bakwazi ukudala ukuthuthukiswa, ama-plug-ins nezinye izingcezu zesofthiwe futhi basebenzise ngokuvumelana nezidingo zabo. Lolu hlobo oluthile lokusebenzisana lube ne-boom ngokwamukelwa okubanzi kwe-Kubernetes umphakathi obanzi we-software. Manje kunanini ngaphambili, ubuchwepheshe besimanje busebenza ndawonye ngokungqubuzana okuncane kakhulu futhi kungaba ngemizuzu ndawonye cishe noma kuphi.
Izinkampani ze-Big Tech zikwazisa lokhu kubambisana okujulile okungokwemvelo emphakathini womthombo ovulekile lapho zikhulula ngokukhululekile izinhlaka, imitapo yolwazi, nezilimi ezizidalile ukuze zinakekele futhi zithuthukise amathuluzi angaphakathi. Ukwenza kanjalo kujulisa inqwaba yonjiniyela abakwazi ukusebenza emikhiqizweni yabo futhi kuqale ukusetha izinga lokuthi ubuchwepheshe obufanayo kufanele busebenze kanjani. Yona leyo ndatshana ye-Atlantic icaphuna umsunguli we-Meta u-Mark Zuckerberg ethi "kube yigugu kakhulu kithi ukuthi sinikeze lokho ngoba manje bonke onjiniyela abahamba phambili kuyo yonke imboni basebenzisa amathuluzi esiwasebenzisayo ngaphakathi".
Umthombo Ovulekile Uzala Umthombo Ovulekile
Lezi yizici zokuthi kungani sivame ukubona ukusebenzisana phakathi kwezinkampani ezinomthombo ovulekile. Izinkampani ze-Open-source AI kanye ne-ML zizothuthukisa ngokwemvelo izixazululo ngeminye imikhiqizo yomthombo ovulekile kusukela kumikhiqizo eyisisekelo efana nokugcinwa kwento kuye phezulu esitaki kumathuluzi okubuka. Lapho inkampani eyodwa enomthombo ovulekile iqhubekela phambili, sonke siyahamba. Le ndlela ehlangene nehlanganisiwe cishe iwukubheja kwethu okuhle kakhulu ekuthuthukiseni i-AI ethatha indlela egxile kumuntu. Lawa mandla emvelo atholakala kumakethe adinga umthombo ovulekile we-AI kuhlanganiswe nezimfanelo zesofthiwe yomthombo ovulekile wokuqamba izinto ezintsha nokusebenzisana kuzoshayela umthombo ovulekile wesitaki sedatha ye-AI.
Sicela ujoyine futhi ufake isandla kule ngxoxo nasemphakathini wethu ngokusithumelela i-imeyili kokuthi [email protected] noma usithumelele umlayezo esiteshini sethu se-Slack .
Iphinde ishicilelwe lapha .