Kwenzekeni
I-MarkTechPost ibika ukuthi i-Keenable AI ikhiphe i-NEEDLE, ibhentshimakhi yomthombo ovulekile yokuhlola ama-API osesho lwewebhu asetshenziswa abasebenzeli be-AI. Amasethi emibuzo yawo avuselelwa kusukela emithonjeni emisha yomphakathi kunokuba alungiswe kusenesikhathi, ngemibuzo yezindaba eyakhiwa kabusha njalo ngehora nangezezimali, imibuzo yesazi, yezomthetho, nejulile eyakhiwa kabusha nsuku zonke.
I-MarkTechPost ibika ukuthi i-Keenable AI ine-NeEDLE yomthombo ovulekile, ogama layo libhekisele ku-News, Everyday, Expert, Deep-tail, kanye Nokuhlola Kwezomthetho. Ibhentshimakhi ihloselwe amasistimu okusesha asetshenziswa abasebenzeli be-AI, lapho imodeli ingase ikhiphe imibuzo ephindaphindiwe futhi incike ezihlokweni, amazwibela, namazinga ngaphambi kokunquma ukuthi izolanda ikhasi yini. Idizayini emaphakathi iwukugwema isethi yemibuzo engaguquki unomphela. Ngokombiko, imibuzo yezindaba ivuselelwa njalo ngehora kusukela kokuphakelayo kwe-RSS okukhethiwe okungu-124 kanye namathrendi we-Google. Imibuzo yezezimali, yesazi, yezomthetho, kanye neyebhizinisi eyivelakancane ivuselelwa nsuku zonke kusuka emithonjeni efaka i-SEC XBRL, Wikidata, GLEIF, arXiv, Europe PMC, CourtListener, eCFR, kanye nokukhishwa kwe-agent-trajectory yomphakathi.
I-MarkTechPost ibika ukuthi i-LLM iguqula izinto zomthombo zibe imibuzo, kuyilapho ukuhlola komshini kusetshenziswa ukuthola ukuthi umbuzo ngokwawo uyayinikeza impendulo. Ibhentshimakhi ihlanganisa izinhlobo ezinhlanu zemisebenzi ehlukene. Izindaba nokusesha okujulile kwehlulelwa ngekhwalithi yezinga; ihlola ezezimali ukuthi iqiniso eliceliwe liyavela yini phakathi kwamazwibela amahlanu aphezulu; futhi ukusesha kwesazi nomthetho kuthathwa njengemisebenzi eyaziwayo etholwe ukufaniswa kwezihlonzi. Isihloko sithi ihhanisi lomthombo ovulekile liyithuluzi lomyalo we-Python elikhiqiza futhi lisebenzise imiyalo engaphansi, lingasebenza kukhompuyutha ephathekayo noma ngokuhlanganisa okuqhubekayo, futhi lidinga ukhiye we-OpenRouter wokwahlulela kanye nokhiye we-API wenjini yokusesha ngayinye ehloliwe.
Embuzweni ngamunye, i-MarkTechPost ithi i-NEEDLE ithumela umbhalo wombuzo ofanayo kuma-API okusesha angu-15 ngaphansi kwephrothokholi eyodwa. Izingcingo zenziwa eyodwa ngesikhathi, okuvumela ukubambezeleka kwephesenti ukuthi kuqhathaniswe ngaphandle kokulayisha ngesikhathi esisodwa. Ukuhlola kusebenzisa izinga lenjini ngayinye, izihloko, namazwibela; amakhasi awalandiwe, futhi imiphumela ayilinganiswa kabusha. Ubufakazi bufinyezwe ezinhlamvu ezingu-2,000, futhi ijaji aliboniswa igama lenjini. Ngokombiko, Izenzo zomphakathi ze-GitHub zisebenza futhi ziqhutshwa njalo kudathasethi ye-Hugging Face ziyingxenye yendlela yokukhiqiza kabusha yephrojekthi, nakuba lezo zinsiza zingaqinisekiswanga ngokuzimela kumpahla enikeziwe.
Imininingwane yomthombo: marktechpost.com ↗
Kungani kubalulekile
Ibhentshimakhi yakhelwe ukubhekana nobuthakathaka ekuhlolweni kokusesha okumile: isistimu ye-AI ingase ibambe ngekhanda imibuzo nezimpendulo ezishicilelwe, noma ithole ukhiye wempendulo yomphakathi esikhundleni sokusesha ngempela. I-NEEDLE iphinda iqhathanise injini ngayinye nosilingi obonakalayo ngokusekelwe emiphumeleni ehlanganisiwe ebuyiswe yibo bonke abahlinzeki abahloliwe.
Ibhentshimakhi ibhekana nenkinga ebonakalayo ekulinganiseni usesho lwe-AI. Ukuhlola okuqinile kungase kungabi nolwazi oluncane lapho amamodeli ebambe imibuzo ngekhanda, lapho izimpendulo zishunyekwa kudatha yokuqeqeshwa, noma lapho umenzeli ekwazi ukufinyelela ifayela lempendulo elitholakala esidlangalaleni. Isethi yombuzo evuselelwe ngokuqhubekayo yenza lezo zinqamuleli zibe nzima, okungenani ngokomgomo. I-MarkTechPost iqhathanisa le ndlela nokunye ukuhlola okubukhoma okufana ne-LiveBench, i-LiveCodeBench, ne-SWE-bench-Live, kodwa i-athikili enikeziwe ayiqinisekisi ngokuzimela ukuthi indlela yokusebenza ye-NEEDLE iqhathaniswa kanjani nalawo maphrojekthi.
Isilinganiso esihlanganisiwe se-NEEDLE singaba wusizo ngoba amaphuzu aphansi angaba nezimbangela ezihlukile. I-MarkTechPost ibika ukuthi ibhentshimakhi ihlanganisa yonke imiphumela yezinjini ebuyisiwe yombuzo, ama-oda ahlanganisa iphuli ngokuhambisana, futhi iyisebenzisela ukudala uphahla oluqinile olubizwa ngokuthi "ukugcina." Igebe elikhulu phakathi kwenjini ngayinye kanye nalolo silingi lusikisela ukuthi impahla efanelekile itholwe okungenani umhlinzeki oyedwa kodwa ayizange ivele noma ibekwe kahle kuleyo njini. Uphahla olubuthakathaka luphakamisa ukuthi abahlinzeki abahloliwe bahlulekile ngokuhlanganyela ukubuyisa ubufakazi obuqinile. Lona umehluko wokuxilonga kunobufakazi bekhwalithi yosesho.
Imiphumela ebikiwe ikhombisa ukuthi ukusebenza kuyehluka kakhulu ngomsebenzi. Ewindini lezinsuku eziyisikhombisa eliphela ngo-Agasti 28, i-MarkTechPost ibika amaphuzu ezezimali angu-0.910 we-Exa, 0.872 okwe-Keenable, 0.871 okwe-Perplexity, kanye no-0.847 kwe-Google, ngokumelene nenani eliphakeme elingu-0.965. Amaphuzu e-Scholar asukela ku-0.774 ku-Keenable kuya ku-0.310 e-Tavily, uma kuqhathaniswa nokungu-0.869. I-Deep-tail ibikwe njengesigaba esinzima kunazo zonke: I-Exa ifinyelele ku-0.557 kokugcina, i-Keenable 0.470, kanye ne-Bing 0.199. Lezi zibalo ziyizicelo ezivela endaweni, hhayi izilinganiso eziqinisekisiwe ngokuzimela.
I-Interactive Mechanism: Indlela Esebenza Ngayo Ngempela
Hlola ubuchwepheshe obuyisisekelo ngemuva kwalokhu kuthuthukiswa ngokuhlanganyela.
crm_get_transaction(id='4092').An agent must create a draft calendar event for Tuesday at 2 p.m. Which evidence would establish the requested result?
Ongakubuka ngokulandelayo
Umthombo onikeziwe awuqinisekisi ngokuzimela inqolobane, ideshibhodi, ama-artifact esethi yedatha, noma izikolo ezibikiwe. Ukugijima kwesikhathi esizayo kufanele kubonise ukuthi ingabe inqubo yomthombo obukhoma webhentshimakhi ihlala ikwazi ukukhiqizwa kabusha, noma ngabe i-oracle yayo ehlanganisiwe ihlukanisa ngokunengqondo ukwehluleka kokubuyisa nokuhluleka kwezinga, nokuthi imiphumela ishintsha kanjani njengoba abahlinzeki bosesho bebuyekeza izinkomba zabo nama-API.
Inkinga yokuqala okufanele ibukwe ukuphindaphinda. I-MarkTechPost ibika ukuthi ikhodi ye-NEEDLE ikhishwa ngaphansi kwelayisensi ye-MIT nokuthi imibuzo esakazwayo ingaphinda idalwe, kodwa umthombo onikeziwe awufaki ukuhlolwa okuzimele kwekhodi, ishejuli yomthombo, ukuhlola ukuvuza, ukwaziswa kwejaji, noma ama-artifact ashicilelwe. Abacwaningi nabahlinzeki bosesho bazodinga ukuthola ukuthi ingabe ukugijima okusha kukhiqiza ukwakhiwa kombuzo ofanayo nokuthi ingabe izinguquko kokuphakelayo komphakathi noma okubhaliswa zirekhodwa ngokucacile ngokwanele ukuze kubuyekezwe kamuva.
Indaba yesibili ukuthi ingabe uphahla "lokugcina" luhunyushwa ngokucophelela. Ikala umphumela omuhle kakhulu otholwe izinjini ezibambe iqhaza, hhayi indawo yonke ephelele yamakhasi ewebhu afanelekile. Ngakho-ke ingadalula ukuphuthelwa okwabiwe kuphela phakathi kwabahlinzeki abahloliwe kanye newindi lomthombo. I-MarkTechPost iphinda ibike ukuthi u-opharetha we-NEEDLE, u-Keenable, uqhudelana kubhentshimakhi. Lokho kudala ukungqubuzana okungayisebenzisi ngokwayo indlela ngokwayo, kodwa kwenza ikhodi ekhanye, amalogi aphelele, ukwahlulela okungaboni, kanye nokuphinda okuzimele kubaluleke kakhulu.
Ukubambezeleka nakho kuzobaluleka kubathuthukisi be-ejenti. I-MarkTechPost ibika ukuthi i-Keenable-realtime ku-193 millisecond median kanye nama-millisecond angu-284 ku-95th percentile, uma kuqhathaniswa ne-Exa ku-1,876 kanye no-2,955 millisecond kanye ne-Bing ku-2,767 kanye ne-9,381 milliseconds ngewindi lezinsuku eziyisikhombisa ezifanayo. I-athikili ithi izingcingo ezihlulekile azibandakanyi kumasampula wokubambezeleka, ngakho ukuhlolwa okuzayo kufanele kuhlole izilinganiso zokungaphumeleli, imikhawulo yesilinganiso, izinguquko zekhava, kanye nokuhwebelana phakathi kwesivinini nekhwalithi yokubuyisa eduze kwalawa maphesenti.