Ngokombiko we-《商业内幕》 wangomhla ziyi-13 (namuhla), u-CEO we-Y Combinator u-Chen Jiaxing (Garry Tan unike osomabhizinisi abakhathazekile ngokusetshenziswa ngokweqile kwama-token e-AI iseluleko esicacile: wasebenzise, ungazigodli. Ngisho noma izindleko ziphezulu, osomabhizinisi kufanele bazimisele ukutshala imali eningi kuma-token e-AI.
Lapho ekhuluma ngama-agent e-AI, u-Chen Jiaxing wathi: “Kufanele ukhuphule amandla awo aze afike ezingeni eliphezulu. Ngokwesibonelo, uwunike ama-token angu-1 million ngesikhathi esisodwa, noma angu-800,000. Ngicabanga ukuthi uma usufike kulelo zinga, empeleni uzobe usuphila ku-2028 kusenesikhathi.”
U-Chen Jiaxing ukholelwa ukuthi ukusebenzisa ama-agent amaningi e-AI kungavumela osomabhizinisi ukuthi bathole kusenesikhathi amakhono e-AI abantu abaningi abazokwazi ukuwasebenzisa kuphela esikhathini esizayo. Nokho, le nzuzo yokuba phambili ibiza kakhulu. “Anginaso isiqiniseko, kodwa ukusebenzisa ama-agent ngamandla aphezulu kungabiza u-50,000-100,000 wamadola ngonyaka (inothi: ngokwesilinganiso samanje sokushintshanisa imali, cishe kungu-338,000-676,000 yuan). Izindleko ziphezulu ngendlela emangalisayo. Kodwa kuma-CEO noma osomabhizinisi, ukwenza kanjalo kunengqondo ngempela.”
U-Chen Jiaxing ukholelwa ukuthi ukuhlola okubizayo kungaguqulwa kancane kancane kube yinzuzo yesikhathi eside. Ngemva kokuba i-agent ye-AI iphothule ngempumelelo umsebenzi othile, osomabhizinisi kufanele baguqule yonke le nqubo ibe “ikhono”, bayihlele ibe imiyalelo engasetshenziswa ngokuphindaphindiwe, ukuze i-AI ikwazi ukuwuqedela ngokuzenzakalela umsebenzi ofanayo kaningi esikhathini esizayo.
U-Chen Jiaxing wafingqa le ndlela ngomusho owodwa: “Ifayela elilodwa le-Markdown lingumsebenzi oyedwa. Lo msebenzi angawuqeda kahle umsebenzi njalo, futhi ungamcela ukuthi awenze izikhathi eziningi ngendlela ofuna ngayo.”
Inothi: Ifayela le-Markdown lisho iqoqo lemiyalelo ebhaliwe etshela i-agent ye-AI ukuthi ingawuqeda kanjani umsebenzi othile.
U-Chen Jiaxing ungomunye wabaphathi abambalwa embonini yezobuchwepheshe abeseka esidlangalaleni i-“tokenmaxxing”. Ngaleso sikhathi, abanye baseSilicon Valley bakholelwa ukuthi ukusebenzisa ama-token amaningi kuwukumosha izinsiza.
Lapha, i-“tokenmaxxing” ihunyushwa ngokuqondile ngokuthi “ukugcwalisa ama-token”, okusho ukwenza ngamabomu i-agent ye-AI isebenzise ama-token amaningi, ngaphandle kokubeka ukunciphisa ukusetshenziswa kwama-token nokwehlisa izindleko njengomgomo oyinhloko.
Ngesonto eledlule, u-Praveen · Nepali · Nagar, ophethe ezobuchwepheshe kwa-Uber, wathi i-Uber ibone idatha ethile yezindleko ze-AI okufanele inakwe, futhi lokho “kungesinye isibonakaliso sokuthi inkathi ebizwa ngokuthi ‘tokenmaxxing’ isizophela.” “Esigabeni esilandelayo, kungakhathaliseki ukuthi ekugcineni sizobizwa ngani, okubalulekile ngeke kube ukuthi ubani osebenzisa ama-token amaningi, kodwa ukuthi ubani okwazi ukuwasebenzisa ngendlela ephumelela kakhulu.”
U-Scott · Wu, u-CEO wenkampani yokuhlela nge-AI i-Cognition, naye wathi esiqeshini se-podcast sangoJuni ukuthi ukukhuthaza “tokenmaxxing” ezinkampanini sekuqhele kancane emgomeni oyinhloko. “Umqondo jikelele awunaphutha, kodwa kwezinye izindawo abantu bawenze ngokweqile ngempela. Abanye bathi, ‘Sibabeka ezikhundleni ngokuthi onjiniyela basebenzise ama-token amangaki.’ Kungcono sibabeke ezikhundleni ngokuthi umuntu ngamunye ukhiqize imiphumela emingaki ngempela.”