I-DeepMind kaGoogle isombulula iingxaki zemathematika ngobukrelekrele bokwenziwa

Inkqubela phambili yamva nje kwiimodeli zeelwimi ezinkulu (LLMs) yenze i-AI iguquguquke ngakumbi, kodwa oku kuza necala elisezantsi: iimpazamo.

I-Generative AI ithambekele ekwenzeni izinto, kodwa i-Google DeepMind ize ne-LLM entsha enamathela kwiinyaniso zemathematika.

IFunSearch yenkampani inokusombulula iingxaki zezibalo ezintsonkothileyo.

Ngommangaliso, izisombululo eziveliswayo azichanekanga kuphela; zizicombululo ezitsha ngokupheleleyo ekungekho mntu wakha wazifumana.

Ixesha lokufunda eliqikelelweyo: 4 imizuzu

IFunSearch ibizwa ngolo hlobo kuba ikhangela imisebenzi yemathematika, hayi kuba iyonwabisa. Nangona kunjalo, abanye abantu banokuthatha ingxaki ye-cap set njenge-hoot: iingcali zezibalo azikwazi nokuvumelana ngeyona ndlela yokusombulula ngayo, ziyenza ibe yimfihlelo yamanani yokwenyani. DeepMind sele yenze inkqubela phambili kubukrelekrele bokwenziwa ngemifuziselo yayo yeAlpha efana neAlphaFold (ukusonga iprotein), iAlphaStar (StarCraft), kunye neAlphaGo (edlala iGo). Ezi nkqubo zazingasekelwanga kwi-LLM, kodwa ziveze iikhonsepthi ezintsha zemathematika.

NgeFunSearch, DeepMind iqale ngemo yolwimi olukhulu, uguqulelo lweGoogle PaLM 2 ebizwa Codey. Kukho inqanaba le-LLM yesibini emsebenzini, ehlalutya imveliso yeCodey kwaye isusa ulwazi olungachanekanga. Iqela elisemva kwalo msebenzi lalingazi ukuba le ndlela iya kusebenza kwaye ayiqinisekanga ukuba kutheni, ngokutsho komphandi DeepMind Alhussein Fawzi.

Ukuqala, iinjineli kwi DeepMind baye benza umboniso wePython wengxaki yokuseta umnqwazi, kodwa bashiya imigca echaza isisombululo. Umsebenzi weCodey yayikukongeza imigca eyisombulule ngokuchanekileyo ingxaki. Impazamo yokujonga umaleko emva koko ifumana izisombululo zeCodey ukubona ukuba zichanekile na. Kwimathematika ekwinqanaba eliphezulu, iiequations zinokuba nesisombululo esingaphezulu kwesinye, kodwa ayizizo zonke ezithathwa njengezilungileyo ngokulinganayo. Ngokuhamba kwexesha, i-algorithm ichonga ezona zisombululo zeCodey kwaye izifake kwakhona kwimodeli.

I-DeepMind ivumela i-FunSearch iqhube iintsuku ezininzi, ixesha elide ngokwaneleyo ukuvelisa izigidi zezisombululo ezinokwenzeka. Oku kuvumele iFunSearch ukuba icokise ikhowudi kwaye ivelise iziphumo ezingcono. Ngokophando olusanda kupapashwa, L 'ingqiqo ifumene isisombululo esingaziwayo ngaphambili kodwa esichanekileyo kwiseti yengxaki. DeepMind kwakhona ikhulule iFunSearch kwenye ingxaki yezibalo enzima ebizwa ngokuba yingxaki yokupakisha isikhongozeli, i-algorithm echaza eyona ndlela isebenzayo yokupakisha izikhongozeli. IFunSearch ifumene isisombululo ngokukhawuleza kunezo zibalwe ngabantu.

Iingcali zezibalo zisazama ukudibanisa itekhnoloji yeLLM kumsebenzi wabo kunye nomsebenzi we DeepMind ibonisa indlela enokwenzeka emayilandelwe. Iqela likholelwa ukuba le ndlela inokubakho ngenxa yokuba ivelisa ikhowudi yekhompyutheni kunokuba isisombululo. Oku kuhlala kulula ukukuqonda nokuqinisekisa kuneziphumo zemathematika ekrwada.

Ufundo olunxulumeneyo

• Akusekho luck nje: ibhola yefantasy iba ngumzekelo wohlalutyo. UFantalgoritmo wazalwa, xa ihlabathi lebhola ekhatywayo lidibana nemathematika
• Gato, DeepMind kunye nogqatso olusingise kubukrelekrele bokwenziwa ngokubanzi
• Yintoni i-Predictive Analytics, kwaye kutheni ibalulekile kwishishini
• IZAKHIWO ZORhwebo OLUZINZILEYO: NGOLAWULO LWE-ENEJI KUNYE NOBUCHULE OBUGCINWEYO, UKUSETYENZISWA KWAMANDLA KWIHLABATHI KUNOKUNCHISWA NGE-30% KUNYE NOKHUSELEKO lwe-CO37 ngama-2%.
• Innovation kunye nobudlelwane abathengi - amava kunye nengqondo-deductive ubukrelekrele

BlogInnovazione.it