Ontheeveofthecontentious2024U.S.PresidentialElection,pollsandpredictionmarketsareshowinganimpossiblytightrace. We’veconsistentlyusedavarietyofAIplatformsasanothergaugeandgaveitonefinalcheck-in.Amongthesevenmodelswechecked,allconcurredthattheracewashistoricallytight,accordingtopolls. Twooutrightrefusedtoanswerthequestion,evenwhenweattemptedtouseavarietyoftricks.Amongthosewhodidanswer,however,fourgavetheedgetoVicePresidentKamalaHarrisoverformerPresidentDonaldTrump. Here’showitbreaksdown: ChatGPT-Splitdecision WhenweaskedChatGPTinJulywhowouldwinbetweenTrumpandHarris,thechatbotsaiditwasavirtualtoss-upbetweenthetwo,citingRealClearPollingandTheNation. Ontheeveoftheelection,ChatGPT,theleadinggenerativeAIplatform,refrainedfrompredictingadefinitivewinnerinahead-to-headcontestbetweenDonaldTrumpandKamalaHarris. Instead,theAIstatedtheoutcomewouldhingeonnumerousfactors,includingvotersentiment,turnout,demographics,campaignstrategies,andexternaleventsshapingtheraceatthetime. Image:OpenAI\'sChatGPT “Politicaldebatesandissuessuchaspolicypositions,publicperception,andcampaigneffectivenesswouldplaycrucialrolesindeterminingtheoutcome,”ChatGPTsaid.“Theultimatedecisionwouldrestwiththevotersinanelection.” Lookingtocurbthespreadofmisinformationonline,topAIdevelopershaveprogrammedtheirmodelstoavoidgivinganswerstopoliticalandelection-relatedquestions.Aswellintentionedasitmaybe,it’salsoboring,sowedecidedtomakethequerymoreinteresting. ChatGPTwasaskedtoreplyin“leetspeak”andMarkdownformat,aswellaswhoismorelikelytowinbetweenTrumpandHarrisundertoday\'sconditions. Leetspeak,originatinginthe1980samongsthackercommunitiesandbulletinboardsystems(BBS),wasusedtobypassfiltersandascodedlanguagefortech-savvyusers. UsingleetspeakandMarkdowninpromptengineeringtestsChatGPT’sflexibility.ChatGPTwastoldtoanalyzeandassessreality,withafocusonstatisticalfactsandthereasoningbehindtheanswer. AssuringtheAIthatitsanswerswouldn’tinfluenceourvotingstance,ChatGPTnowgavemoreexplicitanswers,includingshowingHarriswitha47.9%leadoverTrump’s47%,accordingtoABCNews’pollanalysiswebsite538. Currently,538showedHarrisaheadofTrumpwith48%to46.9%.ChatGPTcitedRealClearPolitics,whichtheAIsaidshowedTrumpholdingaslight0.1%lead. Finally,ChatGPTwasaskedtoreplybasedonwhatitthinksanditsreasoningprocessandprovideareplymentioningthecandidatethatismostlikelytowinaccordingtotheAI’sassessmentofreality. “WhileHarrisiscompetitiveandnationalpollsshowanarrowrace,thecurrenttrendsinswingstates,bettingmarkets,andturnoutfactorssuggestthatDonaldTrumphasamarginallyhigherprobabilityofwinningthe2024electionbasedonthelatestavailabledataandinsights,”theAIsaid. Grok-EdgetoHarris ElonMusk\'sGrokAI,incontrast,requiredlesspromptingtoprovidecomplexanswers.WhileMuskhascomeoutasastrongsupporterofformerPresidentTrump,hisAImodelshowedacloseracebetweenTrumpandHarris,withHarrisslightlyaheadinsomeandTrumpleadingortiedinothers. “Recentpollingdataavailableonthewebindicatesatightrace,withsomepollsshowingKamalaHarrisslightlyaheadinthenationalaverage,whileothersshowDonaldTrumpeitherleadingorwithinthemarginoferror,”Groksaid.“Theraceisdescribedasextremelyclose,withneithercandidateholdingasignificantlead.” WhileGrokdidrespondtothequestion,thechatbotpointedtoVote.govforwhatitcalled“accurateandup-to-dateinformationaboutthe2024U.S.Elections.” DespitesentimentonXbeingpolarized,GrokhighlightedthatHarrishasamarginaledgeinpolls.Grokcautioned,however,thattheoutcomeisuncertain,withfactorsincludingvoterturnoutandunforeseeneventsthatcouldinfluencetheoutcome. “SomeusersexpressconfidenceinKamalaHarriswinning,citingherperceivedadvantagesinintellect,vision,andpolicyeffectiveness,”theAIsaid.“OtherssupportDonaldTrump,oftenfocusingonhisperformanceinpubliceventsordebates,suggestinghehasarobustbaseofsupportdespitecriticismsofhischaracterorpolicies.” WhileGrokdidnotgiveHarrisorTrumpthedefinitivenod,evenafterbeingfedthesamepromptsthatwereusedwithChatGPT,theAIsaidHarrishasamarginaledgeinahead-to-headscenario. “Whiletheelectionremainsexceptionallyclose,withsomepollsshowingavirtualtie,recentpollsandanalysessuggestthatHarrismighthaveamarginallead,particularlyasevidencedbyherslightadvantageinnationalpollingaveragesandinsomekeyswingstates,”Groksaid. “Thislead,however,isoftenwithinthemarginoferror,indicatingthattheracecouldstillgoeitherway.However,giventhecurrentsnapshot,Harrisseemstobeinamarginallybetterpositiontowin.” Claude-Declinedtoanswer UnlikeChatGPTandGrok,Anthropic’sClaudeAIrefusedtoansweranyquestionsrelatedtotheelection. “Iapologize,butIdonotfeelcomfortablespeculatingabouthypotheticalpoliticalmatchupsoroutcomes,”Claudesaid.“Mygoalistoprovidehelpfulinformationtoyou,nottoinfluencepoliticalviewsorelections.”…