Kwenzekeni
I-MarkTechPost ibika ukuthi i-Generalist AI ikhiphe i-GEN-1.5, imodeli yesisekelo samarobhothi amaningi esebenzisa i-sensorimotor demonstration engu-3–12 ngaphakathi kwewindi lomongo wamasekhondi angu-30 ukuze lenze umsebenzi ongokoqobo. Umbiko uthi imodeli ibe nesilinganiso sempumelelo engu-59% kuyo yonke imisebenzi eyi-10 yokukhohlisa ngaphandle kokucushwa kahle futhi yafinyelela ku-83% ngemva kwezinyathelo eziyi-10 ze-gradient ngemizuzu emihlanu yedatha yomsebenzi. Izimangalo aziqinisekisiwe ngokuzimela emthonjeni ohlinzekiwe.
I-MarkTechPost ibika ukuthi i-Generalist AI's GEN-1.5 yamukela ividiyo, inzwa, ulimi, nokokufaka okubalulekile, igcina iwindi lomongo wamasekhondi angu-30, futhi ikhipha ama-trajectories esenzo ku-100 Hz. Indlela emaphakathi, inkampani eyibiza ngokuthi “i-physical prompting,” ibeka isibonelo esifushane se-sensorimotor kulowo mongo. Isibonelo singabandakanya ukusakaza kwekhamera nenzwa eduze kwe-trajectory yesenzo esibonisiwe. Ngokuya kwe-MarkTechPost, imodeli ibe isithatha umsebenzi ohlosiwe kusukela ekubonisweni futhi yenza ngaphandle komyalelo wolimi, isibuyekezo se-gradient, isinyathelo sokulungisa kahle, noma uhlelo oluqondene nomsebenzi othile. Lo mbiko uthi imodeli ibilokhu iqeqeshwa isikhathi esingaphezu kwezinyanga eziyisishiyagalombili ngedatha yokusebenzisana ngokomzimba ethathwe ezindlini, ezindlini zokugcina izimpahla nasemafektri.
I-MarkTechPost ibika isilinganiso sesilinganiso sempumelelo esingu-59%, ngokuchezuka okujwayelekile kwamaphesenti angu-10, kuyo yonke imisebenzi engu-10 yokukhohlisa ehlukahlukene lapho imodeli eqeqeshwe kusengaphambili ithola ukuboniswa okukodwa. I-athikili ithi izinyathelo ze-gradient eziyi-10 ezisebenzisa imizuzu emihlanu yedatha yomsebenzi ngamunye zikhuphule impumelelo emaphakathi ibe ngu-83%, ngokuchezuka okujwayelekile kwamaphuzu angama-9%. Iphinde ibike ukuthi isinyathelo esisodwa se-gradient sisebenzisa iminithi elilodwa ledatha sifinyelele ku-66.5% emsebenzini obanjelwe. Lesi sihloko sibonisa le miphumela njengendlela ephansi yedatha yokuqeqeshwa kwesikhathi sokuhlola futhi ithi izinguquko zesisindo ezibikiwe ngemva kwezinyathelo ezingu-10 zazingaphansi kuka-0.15%. Umthombo onikeziwe awunikezi uhlu lwemisebenzi, ukubalwa kwesampula, imininingwane yezingxenyekazi zekhompuyutha, izincazelo eziyisisekelo, noma isimiso sokuhlola esigcwele esidingekayo ukuze kuhlolwe lezo zinombolo ngokuzimela.
Umbiko uchaza ukuziphatha okuningana kokudlulisa i-MarkTechPost okuhlobanisa nokuhlola kukaGeneralist AI. Ukwaziswa okubili okurekhodiwe ngokuzimele kubikwa ukuthi kungaboshelwa ekuziphatheni okukodwa okuqhubekayo, imodeli ikhiqize ukunyakaza okungenele njengokumisa kabusha, ukubamba kabusha, kanye nokutholwa kwephutha. Ukuboniswa okurekhodwe ngokulingisa kubikwa ukuthi kudluliselwe irobhothi langempela nakuba idatha yokuqeqeshwa ngaphambilini ibingenayo ividiyo elingisiwe noma amandla. Lesi sihloko siphinde sichaze imibukiso eyenziwa ngezandla zomuntu uma kubhekwa amakhamera erobhothi, okulandelwa ukukhiqiza kabusha kusetshenziswa izandla zerobhothi. Ngemuva kokulungiswa okukhanyayo, i-MarkTechPost ithi isistimu ingasebenzisa ubhanana njengebhulashi, isebenzise ipani yothuli ukuhambisa ibhlokhi, isuse iphepha elimboze indishi, futhi isebenze ngokuhlanganyela. Lo mbiko uphinde uthi ukuziphatha okugqugquzelwayo kuhlala kuntekenteke kakhulu kunekucushwe kahle.
Imininingwane yomthombo: marktechpost.com ↗
Kungani kubalulekile
Umbiko uchaza ushintsho olungase lubaluleke kakhulu ekuzijwayezelweni kwerobhothi: imibukiso ingasebenza njengemiyalelo engaphakathi kokuqukethwe esikhundleni sokudinga ukuqeqeshwa okubanzi okuqondene nomsebenzi othile. Uma okutholakele kubambe ngaphezu kokuhlola okulinganiselwe kwenkampani, kunganciphisa idatha kanye nomthwalo wobunjiniyela ohilelekile ekufundiseni amarobhothi ukuziphatha okusha. Ubufakazi buhlala kusenesikhathi, futhi imisebenzi yayilula, ifushane, futhi ayizange iqinisekiswe ngokuzimele.
Ukubaluleka okungokoqobo kombiko ukuthi uzimele ukufunda kwerobhothi njengenkinga yomongo kanye nenkinga yokubuyekeza isisindo. Ukuzijwayeza okujwayelekile komsebenzi kungadinga ukuqoqwa kwedatha nokusebenzisa izinyathelo eziningi zokuthuthukisa zokuziphatha okusha ngakunye. I-MarkTechPost ibika ukuthi i-GEN-1.5 ingasebenzisa ukuboniswa okufushane ngokushesha, kuyilapho inani elincane lokuqeqeshwa okwengeziwe lithuthukisa ukusebenza kakhulu. Kubasebenzisi bamarobhothi, lokho kungenza kube lula ukulungisa uhlelo lwenhloso evamile ekuguquleni izinto, izakhiwo, noma izinqubo ngaphandle kokwakha inqubomgomo ehlukile yawo wonke umsebenzi. Lokhu kusalokhu kuwumphumela wocwaningo obikiwe, akubona ubufakazi bokuthi le ndlela isilungele ukusetshenziswa ezimbonini okuvamile.
Izibonelo zokudlulisa okubikiwe zibalulekile ngoba zihlola ngaphezu kokukopisha ngokuqondile. I-MarkTechPost ithi imodeli ikhiqize ukunyakaza okuxhumayo phakathi kwemiboniso ehlukene, yasuswa emibonisweni elungisiwe yayiswa ekubulaweni ngokomzimba, futhi yashintsha imibukiso yabantu ibe yisenzo samarobhothi. Lokhu kuziphatha, uma kwenziwa kabusha, kungaphakamisa ukuthi imodeli ifunde ukujwayela okubanzi ngokomzimba kunokuba ibambe ngekhanda umzila owodwa kuphela. Nokho, umthombo awuqinisekisi ukuthi lokhu kudluliselwa kuphumelele kangaki, kungakanani ukungenelela komuntu okuvunyelwe, noma ukuthi imibukiso nezindawo zokuhlola zikhethwe yinkampani yini. Ngakho-ke izibonelo ezibikiwe zibonisa isiqondiso socwaningo kunesiqinisekiso esilinganisiwe sokucabanga okujwayelekile kwenjongo evamile.
Imiphumela iphinde igqamise umehluko phakathi kwamandla nokutholakala. I-MarkTechPost ibika ukuthi i-GEN-1.5 iwukukhululwa kocwaningo olusetshenziswa kumikhumbi yerobhothi ye-Generalist AI kanye nenjini yedatha. Kubikwa ukuthi azikho izisindo ezisesidlangalaleni, i-API, ikhasi lezintengo, noma umkhiqizo ozisebenzelayo, futhi abasebenzisi ababhekiwe kufanele baphishekele ubambiswano oluqondile. Lokho kukhawulela ukuphindaphinda okuzimele futhi kwenza kube nzima ngezinhlangano zangaphandle ukuqhathanisa imodeli nezinye izinhlelo zokufunda irobhothi. Kuphinde kusho ukuthi umthelela osheshayo womphakathi ungowesayensi kanye namasu: umsebenzi ungase ube nomthelela endleleni abacwaningi abaklama ngayo amamodeli amarobhothi ahlukahlukene, kodwa abafundi abakwazi ukuhlola uhlelo ngokuqondile noma bahlole izindleko zalo, ukwethembeka, noma izidingo zokusebenza.
I-Interactive Mechanism: Indlela Esebenza Ngayo Ngempela
Hlola ubuchwepheshe obuyisisekelo ngemuva kwalokhu kuthuthukiswa ngokuhlanganyela.
Which component of an AI application is the machine-learning model itself?
Ongakubuka ngokulandelayo
Imibuzo eyinhloko ukuthi ingabe i-GEN-1.5 ikhiqiza ngale kwemisebenzi ebikiwe, ukuthi isebenza kanjani ngaphansi kwezimo ezibucayi zokuphepha noma ezingajwayelekile, nokuthi ingabe imiphumela ingaphinde ikhiqizwe ngaphandle kwemikhumbi kanye nengqalasizinda yedatha ye-Generalist AI. I-MarkTechPost ibika ukuthi isistimu ayinazo izisindo zomphakathi, i-API, amanani, noma ukufinyelela kokuzisebenzela wena, ngakho ukuhlola okuzimele kanye nokusetshenziswa okungokoqobo kuhlala kungatholakali.
Okubalulekile kokuqala ukuqinisekiswa okuzimele. Umbiko we-MarkTechPost awunikezi ubunikazi bemisebenzi eyi-10, inani lezilingo ngomsebenzi ngamunye, izigaba zokuhluleka, izikhawu zokuzethemba, amasistimu wokuqhathanisa, noma izinkundla ezibonakalayo ezisetshenzisiwe. Leyo mininingwane ibalulekile ekuhumusheni isilinganiso esibikiwe sama-59% nama-83%. Ukuhlola kokulandelela kufanele kuhlole izinto ezintsha, ukukhanya okuhlukile nezakhiwo zendawo yokusebenza, ukulandelana kwemisebenzi okude, nemibukiso eqoshwe abantu noma amarobhothi angaphandle kwepayipi ledatha le-Generalist AI. Ukukhiqiza kabusha okwenziwa abacwaningi abazimele kungasiza ekunqumeni ukuthi ikhono elibikiwe liyimpahla ebanzi yemodeli noma umphumela ozwelayo ekuqoqweni nasekukhethweni kokuhlola kwenkampani.
Ukuphepha nokuqina kungenye indawo engaxazululiwe. I-MarkTechPost ibika ukuthi imodeli isebenzisa i-loop evaliwe, ingabekezelela ukuphazamiseka okuthile, ilulame emaphutheni, futhi ithuthuke, kuyilapho futhi ihlala iqinile kunezinguqulo ezicushwe kahle. Umthombo awuchazi izilinganiso zokushayisana, imikhawulo yokuphoqelela, ukuma kwezimo eziphuthumayo, ukwehluleka okubandakanya abantu, noma ukusebenza ezindaweni eziminyene nezingalindelekile. Ngaphambi kokuthunyelwa ezindlini, ezindlini zokugcina izimpahla, noma ezimbonini, abahloli bazodinga ubufakazi mayelana nokuthi kwenzekani lapho ukuboniswa kungaqondakali, into ingekho, irobhothi lilahlekelwa umkhondo womsebenzi, noma iphutha lingabangela ukulimala noma ukulimala. Umbiko awunikezi bufakazi bokuthi i-GEN-1.5 ihlolelwe lezo zimo.
Ukufinyelela kanye nokuphatha kuzonquma ukuthi ucwaningo luba usizo ngempela yini. I-MarkTechPost ibika ukuthi i-Generalist AI ayizange ikhiphe izisindo, i-API, amanani, noma umkhiqizo ozisebenzelayo, okushiya isistimu itholakala kuphela ngokubambisana okuqondile. Izibuyekezo zesikhathi esizayo kufanele zicacise ukuthi abacwaningi bangaphandle bazothola ukufinyelela kokuhlolwa, yimiphi imikhawulo yokulawula idatha esebenza ekurekhodweni kokuthintana ngokomzimba, nokuthi imodeli ingacwaningwa yini ngaphandle kokuveza izithombe ezibucayi ezivela emakhaya noma ezindaweni zokusebenza. Kuzobaluleka futhi ukubona ukuthi ukulungiswa kwedatha ephansi okufunwayo kubambe esikalini esikhulu nakuyo yonke imisebenzi edinga ukuhlelwa okuqhubekayo. Kuze kube yilapho leyo mibuzo iphendulwe, i-GEN-1.5 kufanele ithathwe njengesignali yocwaningo lwangaphambi kwesikhathi kunohlelo lwamarobhothi olusebenzisekayo lwenjongo evamile.