Kwenzekeni
Ukuqaliswa kohlelo lwe-AI kuya ngokuya kusetshenziswa amamodeli esisindo esivulekile ukuze kwehliswe izindleko ezikhuphukayo ze-API kusukela ku-OpenAI kanye ne-Anthropic. U-Harvey, i-unicorn ye-AI esemthethweni, yehle imajini yayo yonke yafika ku--50% ngoJuni ngenxa yokusetshenziswa okuphezulu kwe-OpenAI's GPT-4. Ephendula, uHarvey ukhiphe imodeli okungeyakhe eyesekwa yiMoonshot AI's Kimi K3, ebuyisele amamajini amahle. Ezinye izinkampani, okuhlanganisa i-Abridge, Decagon, kanye neRamp, nazo zakha noma zenza amamodeli azo ngendlela oyifisayo, kanti ezinye zithuthela ku-80% wethrafikhi ziye kumasistimu okuzenzela. Lolu shintsho luqhutshwa amasu entengo anolaka avela kubahlinzeki abangamamodeli kanye nokusebenza okuthuthukisiwe kwezinye izindlela zomthombo ovulekile.
I-Harvey, isiqalo se-AI esisemthethweni esinenani lama-dollar ayizigidi eziyizinkulungwane ezingu-15.6, sihlangabezane nokwehla okuphawulekayo kwemajini iyonke isuka ku-50% yaya ku-50% ngoJuni ngenxa yezindleko eziphezulu ze-API kusukela ku-GPT-4 ye-OpenAI. Ukuze kubhekwane nalokhu, u-Harvey ukhiphe imodeli yomuntu siqu ngo-Agasti, esekelwa yi-Moonshot AI's Kimi K3, enikeza ukusebenza eduze nemikhiqizo ehamba phambili ye-Anthropic ngengxenye yezindleko, ibuyisela amamajini amahle.
Umkhuba usabalale ezimbonini eziningi. I-Abridge yakha imodeli eyisisekelo yezimo zomtholampilo isebenzisa imodeli yomthombo ovulekile ye-NVIDIA, kuyilapho i-Decagon icubungula u-80% wezicelo zayo zemibuzo ngemodeli yayo siqu. Ku-fintech, i-Ramp ne-Rogo ihlola ukuziqeqesha, futhi kumathuluzi okuhlela, i-Cursor ne-Cognition zikhiphe amamodeli enziwe ngokwezifiso.
I-OpenAI ne-Anthropic sezishintshele ekushajeni abasebenzisi bebhizinisi okwengeziwe ukuze zisetshenziswe imodeli, zijezisa amaphethini 'okukhulisa amathokheni' evolumu ephezulu. Ukwengeza, zombili izinkampani ziqasha ithalente kumathrekhi asemqoka wokuqalisa futhi zethula ama-plugin aqondene nomkhakha, zidala ukuncintisana okuqondile namakhasimende azo. Ubungozi bokumiswa kokufinyelela, njengoba bubonwa ku-Cursor ngemuva kokutholwa kwayo yi-SpaceX, buphinde bukhuthaze abaqalayo ukuthi bafune ukuzimela okuyimodeli.
Ngaphandle koshintsho, iziqalo eziningi azilindele ukushiya ngokuphelele i-OpenAI ne-Anthropic. U-Harvey usathembele ku-Anthropic's Claude Opus ngemisebenzi eyinkimbinkimbi, ekhombisa isu lokusebenzisa elixubile. Izinselelo zisekhona, okuhlanganisa izindleko eziphezulu zethalente zonjiniyela abalungisa kahle, isidingo sedatha yokuphathelene, nezindleko ezibalulekile zengqalasizinda zokuphatha amamodeli anesisindo esivulekile.
Imininingwane yomthombo: eu.36kr.com ↗
Kungani kubalulekile
Lo mkhuba ubonisa ushintsho lwesakhiwo kushintsho lwezindleko zemboni ye-AI, njengoba izinkampani zesendlalelo sohlelo lokusebenza zifuna ukunciphisa ukuncika kuma-API obunikazi abizayo. Ku-OpenAI kanye ne-Anthropic, ezilungiselela ama-IPO, ukuphuma kwamakhasimende ebhizinisi ukuya kumamodeli anesisindo esivulekile kungase kucekele phansi imali yawo eyinhloko. Lesi sinyathelo siphinde sigqamise ukukhula okukhulayo kwamamodeli emithombo evulekile njengendlela engabizi kakhulu, engase imise kabusha isimo sokuncintisana futhi iphoqelele abahlinzeki abangamamodeli ukuthi balungise amanani abo namasu ebhizinisi ukuze kugcinwe amakhasimende ebhizinisi.
Ukwamukelwa kwamamodeli anesisindo esivulekile ngabaqalisi abakhulu be-AI kumele isilinganiso esibalulekile sokulondoloza izindleko esingashintsha amandla emali engenayo ye-OpenAI ne-Anthropic, ikakhulukazi njengoba belungiselela ama-IPO. Lolu shintsho lungase luphoqe abahlinzeki bemodeli ukuthi bacabangele kabusha amasu abo entengo kanye neminikelo yebhizinisi ukuze kugcinwe ukwethembeka kwamakhasimende.
Ukusebenza okuthuthukisiwe kwamamodeli omthombo ovulekile, njenge-Moonshot AI's Kimi K3, kuwenza abe enye indlela esebenzayo yama-API wobunikazi ezimweni eziningi zokusetshenziswa. Lokhu kuthuthukiswa kungasheshisa ukwenziwa kwentando yeningi kwamakhono e-AI futhi kunciphise umgoqo ekungeneni kwezicelo ezintsha ze-AI, kugqugquzele ukuqamba okusha emikhakheni efana nezomthetho, ezempilo, kanye nezezimali.
Lo mkhuba futhi ugqamisa ukubaluleka okukhulayo kwedatha nengqalasizinda ekuthuthukisweni kwe-AI. Ama-Startups kufanele atshale kudatha yokuphathelene nezinsiza zekhompuyutha ukuze aqeqeshe amamodeli asebenzayo, okungase kuvuleke amathuba amasha kubahlinzeki bedatha nezinkampani zengqalasizinda yamafu kuyilapho kubeka izinselele kumafemu amancane anezinsiza ezilinganiselwe.
I-Interactive Mechanism: Indlela Esebenza Ngayo Ngempela
Hlola ubuchwepheshe obuyisisekelo ngemuva kwalokhu kuthuthukiswa ngokuhlanganyela.
In AI, what are a model's "parameters"?
Ongakubuka ngokulandelayo
Gada ukusebenza kwezimali kwe-OpenAI ne-Anthropic njengoba besondela kuma-IPO abo, ikakhulukazi noma yiziphi izinguquko kumanani e-API noma amasu okugcina ibhizinisi. Landelela izilinganiso zokutholwa kwamamodeli anesisindo esivulekile phakathi kweziqalo ezinkulu zohlelo lwe-AI kanye nokuthuthukiswa kwamamodeli amasha okungewakho emikhakheni efana nokunakekelwa kwezempilo, ezezimali, nezinsizakalo zomthetho. Futhi, bheka umthelela wethalente nezindleko zengqalasizinda ekubeni nokwenzeka kwamamodeli okuziqeqesha kwabaqalayo abancane.
Buka noma yiziphi izinguquko ze-OpenAI ne-Anthropic's API's API noma izinkontileka zebhizinisi ukuze uphendule ekutholweni okukhulayo kwamamodeli anesisindo esivulekile. Amandla abo okugcina amakhasimende ebhizinisi azobaluleka ekulinganisweni kwawo kwe-IPO kanye nokusimama kwesikhathi eside.
Gada ukusebenza nokwamukelwa kwamamodeli amasha okungewakho kusukela ekuqaleni njenge-Harvey, i-Abridge, ne-Decagon. Impumelelo kulezi zindawo ingaqinisekisa indlela yomthombo ovulekile futhi ikhuthaze ukufuduka okuqhubekayo kude nama-API obunikazi.
Bheka umthelela wethalente nezindleko zengqalasizinda ekubeni nokwenzeka kwamamodeli okuziqeqesha. Uma lezi zithiyo zihlala ziphakeme, iziqalo ezincane zingaqhubeka zithembele ekuxubeni kwamamodeli wemithombo evulekile nevaliwe, kuyilapho amafemu amakhulu engazibophezela ngokugcwele ezixazululweni ezingezakho.