SparksofArtificialGeneralIntelligence:EarlyexperimentswithGPT-4S´ebastienBubeckVarunChandrasekaranRonenEldanJohannesGehrkeEricHorvitzEceKamarPeterLeeYinTatLeeYuanzhiLiScottLundbergHarshaNoriHamidPalangiMarcoTulioRibeiroYiZhangMicrosoftResearchAbstractArtificialintelligence(AI)researchershavebeendevelopingandrefininglargelanguagemodels(LLMs)thatexhibitremarkablecapabilitiesacrossavarietyofdomainsandtasks,challengingourunderstandingoflearningandcognition.ThelatestmodeldevelopedbyOpenAI,GPT-4[Ope23],wastrainedusinganunprecedentedscaleofcomputeanddata.Inthispaper,wereportonourinvestigationofanearlyversionofGPT-4,whenitwasstillinactivedevelopmentbyOpenAI.Weco...
发表评论取消回复