2025谷歌人工智能提示詞工程白皮書-Prompt Engineering-v7

PromptEngineering

Author:LeeBoonstra

PromptEngineering

Acknowledgements

Contentcontributors

MichaelShermanYuanCao

ErickArmbrustAnantNawalgariaAntonioGulliSimoneCammel

CuratorsandEditors

AntonioGulliAnantNawalgariaGraceMollison

TechnicalWriter

JoeyHaymaker

Designer

MichaelLanning

February2025 2

Tableofcontents

Introduction 6

Promptengineering 7

LLMoutputconfiguration 8

Outputlength 8

Samplingcontrols 9

Temperature 9

Top-Kandtop-P 10

Puttingitalltogether 11

Promptingtechniques 13

Generalprompting/zeroshot 13

One-shot&few-shot 15

System,contextualandroleprompting 18

Systemprompting 19

Roleprompting 21

Contextualprompting 23

Step-backprompting 25

ChainofThought(CoT) 29

Self-consistency 32

TreeofThoughts(ToT) 36

ReAct(reason&act) 37

AutomaticPromptEngineering 40

Codeprompting 42

Promptsforwritingcode 42

Promptsforexplainingcode 44

Promptsfortranslatingcode 46

Promptsfordebuggingandreviewingcode 48

Whataboutmultimodalprompting? 54

BestPractices 54

Provideexamples 54

Designwithsimplicity 55

Bespecificabouttheoutput 56

UseInstructionsoverConstraints 56

Controlthemaxtokenlength 58

Usevariablesinprompts 58

Experimentwithinputformatsandwritingstyles 59

Forfew-shotpromptingwithclassificationtasks,mixuptheclasses 59

Adapttomodelupdates 60

Experimentwithoutputformats 60

JSONRepair 61

WorkingwithSchemas 62

Experimenttogetherwithotherpromptengineers 63

CoTBestpractices 64

Documentthevariouspromptattempts 64

Summary 66

Endnotes 68

PromptEngineering

February2025

PAGE

20

Youdon’tneedtobeadatascientistoramachinelearningengineer–everyonecanwriteaprompt.

Introduction

Whenthinkingaboutalargelanguagemodelinputandoutput,atextprompt(sometimesaccompaniedbyothermodalitiessuchasimageprompts)istheinputthemodeluses

topredictaspecificoutput.Youdon’tneedtobeadatascientistoramachinelearningengineer–everyonecanwriteaprompt.However,craftingthemosteffectivepromptcanbecomplicated.Manyaspectsofyourpromptaffectitsefficacy:themodelyouuse,themodel’strainingdata,themodelconfigurations,yourword-choice,styleandtone,structure,andcontextallmatter.Therefore,promptengineeringisaniterativeprocess.Inadequatepromptscanleadtoambiguous,inaccurateresponses,andcanhinderthemodel’sabilitytoprovidemeaningfuloutput.

WhenyouchatwiththeGeminichatbot,1youbasicallywriteprompts,howeverthiswhitepaperfocusesonwritingpromptsfortheGeminimodelwithinVertexAIorbyusingtheAPI,becausebypromptingthemodeldirectlyyouwillhaveaccesstotheconfigurationsuchastemperatureetc.

Thiswhitepaperdiscussespromptengineeringindetail.Wewilllookintothevariouspromptingtechniquestohelpyougettingstartedandsharetipsandbestpracticestobecomeapromptingexpert.Wewillalsodiscusssomeofthechallengesyoucanfacewhilecraftingprompts.

Promptengineering

RememberhowanLLMworks;it’sapredictionengine.Themodeltakessequentialtextasaninputandthenpredictswhatthefollowingtokenshouldbe,basedonthedataitwastrainedon.TheLLMisoperationalizedtodothisoverandoveragain,addingthepreviouslypredictedtokentotheendofthesequentialtextforpredictingthefollowingtoken.Thenext

tokenpredictionisbasedontherelationshipbetweenwhat’sintheprevioustokensandwhattheLLMhasseenduringitstraining.

Whenyouwriteaprompt,youareattemptingtosetuptheLLMtopredicttherightsequenceoftokens.Promptengineeringistheprocessofdesigninghigh-qualitypromptsthatguideLLMstoproduceaccurateoutputs.Thisprocessinvolvestinkeringtofindthebestprompt,optimizingpromptlength,andevaluatingaprompt’swritingstyleandstructureinrelation

tothetask.InthecontextofnaturallanguageprocessingandLLMs,apromptisaninputprovidedtothemodeltogeneratearesponseorprediction.

Thesepromptscanbeusedtoachievevariouskindsofunderstandingandgenerationtaskssuchastextsummarization,informationextraction,questionandanswering,textclassification,languageorcodetranslation,codegeneration,andcodedocumentationorreasoning.

PleasefeelfreetorefertoGoogle’spromptingguides2,3withsimpleandeffectivepromptingexamples.

Whenpromptengineering,youwillstartbychoosingamodel.Promptsmightneedtobeoptimizedforyourspecificmodel,regardlessofwhetheryouuseGeminilanguagemodelsinVertexAI,GPT,Claude,oranopensourcemodellikeGemmaorLLaMA.

Besidestheprompt,youwillalsoneedtotinkerwiththevariousconfigurationsofaLLM.

LLMoutputconfiguration

Onceyouchooseyourmodelyouwillneedtofigureoutthemodelconfiguration.MostLLMscomewithvariousconfigurationoptionsthatcontroltheLLM’soutput.Effectivepromptengineeringrequiressettingtheseconfigurationsoptimallyforyourtask.

Outputlength

Animportantconfigurationsettingisthenumberoftokenstogenerateinaresponse.GeneratingmoretokensrequiresmorecomputationfromtheLLM,leadingtohigherenergyconsumption,potentiallyslowerresponsetimes,andhighercosts.

ReducingtheoutputlengthoftheLLMdoesn’tcausetheLLMtobecomemorestylisticallyortextuallysuccinctintheoutputitcreates,itjustcausestheLLMtostoppredictingmoretokensoncethelimitisreached.Ifyourneedsrequireashortoutputlength,you’llalsopossiblyneedtoengineeryourprompttoaccommodate.

OutputlengthrestrictionisespeciallyimportantforsomeLLMpromptingtechniques,likeReAct,wheretheLLMwillkeepemittinguselesstokensaftertheresponseyouwant.

Beaware,generatingmoretokensrequiresmorecomputationfromtheLLM,leadingtohigherenergyconsumptionandpotentiallyslowerresponsetimes,whichleadstohighercosts.

Samplingcontrols

LLMsdonotformallypredictasingletoken.Rather,LLMspredictprobabilitiesforwhatthenexttokencouldbe,witheachtokenintheLLM’svocabularygettingaprobability.Thosetokenprobabilitiesarethensampledtodeterminewhatthenextproducedtokenwillbe.

Temperature,top-K,andtop-Parethemostcommonconfigurationsettingsthatdeterminehowpredictedtokenprobabilitiesareprocessedtochooseasingleoutputtoken.

Temperature

Temperaturecontrolsthedegreeofrandomnessintokenselection.Lowertemperaturesaregoodforpromptsthatexpectamoredeterministicresponse,whilehighertemperaturescanleadtomorediverseorunexpectedresults.Atemperatureof0(greedydecoding)is

deterministic:thehighestprobabilitytokenisalwaysselected(thoughnotethatiftwotokenshavethesamehighestpredictedprobability,dependingonhowtiebreakingisimplementedyoumaynotalwaysgetthesameoutputwithtemperature0).

Temperaturesclosetothemaxtendtocreatemorerandomoutput.Andastemperaturegetshigherandhigher,alltokensbecomeequallylikelytobethenextpredictedtoken.

TheGeminitemperaturecontrolcanbeunderstoodinasimilarwaytothesoftmaxfunctionusedinmachinelearning.Alowtemperaturesettingmirrorsalowsoftmaxtemperature(T),emphasizingasingle,preferredtemperaturewithhighcertainty.AhigherGeminitemperaturesettingislikeahighsoftmaxtemperature,makingawiderrangeoftemperaturesaround

theselectedsettingmoreacceptable.Thisincreaseduncertaintyaccommodatesscenarioswherearigid,precisetemperaturemaynotbeessentiallikeforexamplewhenexperimentingwithcreativeoutputs.

Top-Kandtop-P

Top-Kandtop-P(alsoknownasnucleussampling)4aretwosamplingsettingsusedinLLMstorestrictthepredictednexttokentocomefromtokenswiththetoppredictedprobabilities.Liketemperature,thesesamplingsettingscontroltherandomnessanddiversityofgeneratedtext.

Top-KsamplingselectsthetopKmostlikelytokensfromthemodel’spredicteddistribution.Thehighertop-K,themorecreativeandvariedthemodel’soutput;thelowertop-K,themorerestiveandfactualthemodel’soutput.Atop-Kof1isequivalenttogreedydecoding.

Top-Psamplingselectsthetoptokenswhosecumulativeprobabilitydoesnotexceedacertainvalue(P).ValuesforPrangefrom0(greedydecoding)to1(alltokensintheLLM’svocabulary).

Thebestwaytochoosebetweentop-Kandtop-Pistoexperimentwithbothmethods(orbothtogether)andseewhichoneproducestheresultsyouarelookingfor.

Puttingitalltogether

Choosingbetweentop-K,top-P,temperature,andthenumberoftokenstogenerate,dependsonthespecificapplicationanddesiredoutcome,andthesettingsallimpactoneanother.It’salsoimportanttomakesureyouunderstandhowyourchosenmodelcombinesthedifferentsamplingsettingstogether.

Iftemperature,top-K,andtop-Pareallavailable(asinVertexStudio),tokensthatmeetboththetop-Kandtop-Pcriteriaarecandidatesforthenextpredictedtoken,andthen

temperatureisappliedtosamplefromthetokensthatpassedthetop-Kandtop-Pcriteria.Ifonlytop-Kortop-Pisavailable,thebehavioristhesamebutonlytheonetop-KorPsettingisused.

Iftemperatureisnotavailable,whatevertokensmeetthetop-Kand/ortop-Pcriteriaarethenrandomlyselectedfromtoproduceasinglenextpredictedtoken.

Atextremesettingsofonesamplingconfigurationvalue,thatonesamplingsettingeithercancelsoutotherconfigurationsettingsorbecomesirrelevant.

Ifyousettemperatureto0,top-Kandtop-Pbecomeirrelevant–themostprobabletokenbecomesthenexttokenpredicted.Ifyousettemperatureextremelyhigh(above1–generallyintothe10s),temperaturebecomesirrelevantandwhatevertokensmake

itthroughthetop-Kand/ortop-Pcriteriaarethenrandomlysampledtochooseanextpredictedtoken.

Ifyousettop-Kto1,temperatureandtop-Pbecomeirrelevant.Onlyonetokenpassesthetop-Kcriteria,andthattokenisthenextpredictedtoken.Ifyousettop-Kextremelyhigh,liketothesizeoftheLLM’svocabulary,anytokenwithanonzeroprobabilityofbeingthenexttokenwillmeetthetop-Kcriteriaandnoneareselectedout.

Ifyousettop-Pto0(oraverysmallvalue),mostLLMsamplingimplementationswillthenonlyconsiderthemostprobabletokentomeetthetop-Pcriteria,makingtemperatureandtop-Kirrelevant.Ifyousettop-Pto1,anytokenwithanonzeroprobabilityofbeingthenexttokenwillmeetthetop-Pcriteria,andnoneareselectedout.

Asageneralstartingpoint,atemperatureof.2,top-Pof.95,andtop-Kof30willgiveyourelativelycoherentresultsthatcanbecreativebutnotexcessivelyso.Ifyouwantespeciallycreativeresults,trystartingwithatemperatureof.9,top-Pof.99,andtop-Kof40.Andifyouwantlesscreativeresults,trystartingwithatemperatureof.1,top-Pof.9,andtop-Kof20.

Finally,ifyourtaskalwayshasasinglecorrectanswer(e.g.,answeringamathproblem),startwithatemperatureof0.

NOTE:Withmorefreedom(highertemperature,top-K,top-P,andoutputtokens),theLLMmightgeneratetextthatislessrelevant.

WARNING:Haveyoueverseenaresponseendingwithalargeamountoffillerwords?Thisisalsoknownasthe"repetitionloopbug",whichisacommonissueinLargeLanguageModelswherethemodelgetsstuckinacycle,repeatedlygeneratingthesame(filler)word,phrase,orsentencestructure,oftenexacerbatedbyinappropriatetemperatureandtop-k/

top-psettings.Thiscanoccuratbothlowandhightemperaturesettings,thoughfordifferentreasons.Atlowtemperatures,themodelbecomesoverlydeterministic,stickingrigidlytothehighestprobabilitypath,whichcanleadtoaloopifthatpathrevisitspreviouslygeneratedtext.Conversely,athightemperatures,themodel'soutputbecomesexcessivelyrandom,increasingtheprobabilitythatarandomlychosenwordorphrasewill,bychance,leadbacktoapriorstate,creatingaloopduetothevastnumberofavailableoptions.Inbothcases,themodel'ssamplingprocessgets"stuck,"resultinginmonotonousandunhelpfuloutputuntiltheoutputwindowisfilled.Solvingthisoftenrequirescarefultinkeringwithtemperatureandtop-k/top-pvaluestofindtheoptimalbalancebetweendeterminismandrandomness.

Promptingtechniques

LLMsaretunedtofollowinstructionsandaretrainedonlargeamountsofdatasotheycanunderstandapromptandgenerateananswer.ButLLMsaren’tperfect;thecleareryourprompttext,thebetteritisfortheLLMtopredictthenextlikelytext.Additionally,specifictechniquesthattakeadvantageofhowLLMsaretrainedandhowLLMsworkwillhelpyougettherelevantresultsfromLLMs

Nowthatweunderstandwhatpromptengineeringisandwhatittakes,let’sdiveintosomeexamplesofthemostimportantpromptingtechniques.

Generalprompting/zeroshot

Azero-shot5promptisthesimplesttypeofprompt.ItonlyprovidesadescriptionofataskandsometextfortheLLMtogetstartedwith.Thisinputcouldbeanything:aquestion,astartofastory,orinstructions.Thenamezero-shotstandsfor’noexamples’.

Let’suseVertexAIStudio(forLanguage)inVertexAI,6whichprovidesaplaygroundtotestprompts.InTable1,youwillseeanexamplezero-shotprompttoclassifymoviereviews.

Thetableformatasusedbelowisagreatwayofdocumentingprompts.Yourpromptswilllikelygothroughmanyiterationsbeforetheyendupinacodebase,soit’simportanttokeeptrackofyourpromptengineeringworkinadisciplined,structuredway.Moreonthistableformat,theimportanceoftrackingpromptengineeringwork,andthepromptdevelopmentprocessisintheBestPracticessectionlaterinthischapter(“Documentthevariouspromptattempts”).

Themodeltemperatureshouldbesettoalownumber,sincenocreativityisneeded,andweusethegemini-prodefaulttop-Kandtop-Pvalues,whicheffectivelydisablebothsettings(see‘LLMOutputConfiguration’above).Payattentiontothegeneratedoutput.Thewordsdisturbingandmasterpieceshouldmakethepredictionalittlemorecomplicated,asbothwordsareusedinthesamesentence.

NameGoalModel

TemperatureTop-KPrompt

1_1_movie_classification

Classifymoviereviewsaspositive,neutralornegative.gemini-pro

0.1 TokenLimit 5

N/A Top-P 1

ClassifymoviereviewsasPOSITIVE,NEUTRALorNEGATIVE.Review:"Her"isadisturbingstudyrevealingthedirectionhumanityisheadedifAIisallowedtokeepevolving,unchecked.Iwishthereweremoremovieslikethismasterpiece.Sentiment:

Output POSITIVE

Table1.Anexampleofzero-shotprompting

Whenzero-shotdoesn’twork,youcanprovidedemonstrationsorexamplesintheprompt,whichleadsto“one-shot”and“few-shot”prompting.Generalprompting/zeroshot

One-shot&few-shot

WhencreatingpromptsforAImodels,itishelpfultoprovideexamples.Theseexamplescanhelpthemodelunderstandwhatyouareaskingfor.Examplesareespeciallyusefulwhenyouwanttosteerthemodeltoacertainoutputstructureorpattern.

Aone-shotprompt,providesasingleexample,hencethenameone-shot.Theideaisthemodelhasanexampleitcanimitatetobestcompletethetask.

Afew-shotprompt7providesmultipleexamplestothemodel.Thisapproachshowsthemodelapatternthatitneedstofollow.Theideaissimilartoone-shot,butmultipleexamplesofthedesiredpatternincreasesthechancethemodelfollowsthepattern.

Thenumberofexamplesyouneedforfew-shotpromptingdependsonafewfactors,includingthecomplexityofthetask,thequalityoftheexamples,andthecapabilitiesofthegenerativeAI(genAI)modelyouareusing.Asageneralruleofthumb,youshoulduseatleastthreetofiveexamplesforfew-shotprompting.However,youmayneedtousemoreexamplesformorecomplextasks,oryoumayneedtousefewerduetotheinputlengthlimitationofyourmodel.

Table2showsafew-shotpromptexample,let’susethesamegemini-promodelconfigurationsettingsasbefore,otherthanincreasingthetokenlimittoaccommodatetheneedforalongerresponse.

Goal

ParsepizzaorderstoJSON

Prompt

Parseacustomer'spizzaorderintovalidJSON:

EXAMPLE:

Iwantasmallpizzawithcheese,tomatosauce,andpepperoni.JSONResponse:

```

{

"size":"small",

"type":"normal",

"ingredients":[["cheese","tomatosauce","peperoni"]]

}

```

Continuesnextpage...

1

Top-P

N/A

Top-K

250

TokenLimit

0.1

Temperature

gemini-pro

Model

Prompt

EXAMPLE:

CanIgetalargepizzawithtomatosauce,basilandmozzarella

{

"size":"large",

"type":"normal",

"ingredients":[["tomatosauce","bazel","mozzarella"]]

}

Now,Iwouldlikealargepizza,withthefirsthalfcheeseandmozzarella.Andtheothertomatosauce,hamandpineapple.

JSONResponse:

Output

```

{

"size":"large",

"type":"half-half",

"ingredients":[["cheese","mozzarella"],["tomatosauce","ham","pineapple"]]

}

```

Table2.Anexampleoffew-shotprompting

Whenyouchooseexamplesforyourprompt,useexamplesthatarerelevanttothetaskyouwanttoperform.Theexamplesshouldbediverse,ofhighquality,andwellwritten.Onesmallmistakecanconfusethemodelandwillresultinundesiredoutput.

Ifyouaretryingtogenerateoutputthatisrobusttoavarietyofinputs,thenitisimportanttoincludeedgecasesinyourexamples.Edgecasesareinputsthatareunusualorunexpected,butthatthemodelshouldstillbeabletohandle.

System,contextualandroleprompting

System,contextualandrolepromptingarealltechniquesusedtoguidehowLLMsgeneratetext,buttheyfocusondifferentaspects:

Systempromptingsetstheoverallcontextandpurposeforthelanguagemodel.Itdefinesthe‘bigpicture’ofwhatthemodelshouldbedoing,liketranslatingalanguage,classifyingareviewetc.

Contextualpromptingprovidesspecificdetailsorbackgroundinformationrelevanttothecurrentconversationortask.Ithelpsthemodeltounderstandthenuancesofwhat’sbeingaskedandtailortheresponseaccordingly.

Rolepromptingassignsaspecificcharacteroridentityforthelanguagemodeltoadopt.Thishelpsthemodelgenerateresponsesthatareconsistentwiththeassignedroleanditsassociatedknowledgeandbehavior.

Therecanbeconsiderableoverlapbetweensystem,contextual,androleprompting.E.g.apromptthatassignsaroletothesystem,canalsohaveacontext.

However,eachtypeofpromptservesaslightlydifferentprimarypurpose:

Systemprompt:Definesthemodel’sfundamentalcapabilitiesandoverarchingpurpose.

Contextualprompt:Providesimmediate,task-specificinformationtoguidetheresponse.It’shighlyspecifictothecurrenttaskorinput,whichisdynamic.

Roleprompt:Framesthemodel’soutputstyleandvoice.Itaddsalayerofspecificityandpersonality.

Distinguishingbetweensystem,contextual,androlepromptsprovidesaframeworkfordesigningpromptswithclearintent,allowingforflexiblecombinationsandmakingiteasiertoanalyzehoweachprompttypeinfluencesthelanguagemodel’soutput.

Let’sdiveintothesethreedifferentkindsofprompts.

Systemprompting

Table3containsasystemprompt,whereIspecifyadditionalinformationonhowtoreturntheoutput.Iincreasedthetemperaturetogetahighercreativitylevel,andIspecifiedahighertokenlimit.However,becauseofmyclearinstructiononhowtoreturntheoutputthemodeldidn’treturnextratext.

Goal

Classifymoviereviewsaspositive,neutralornegative.

Output

NEGATIVE

Classifymoviereviewsaspositive,neutralornegative.Onlyreturnthelabelinuppercase.

Review:"Her"isadisturbingstudyrevealingthedirectionhumanityisheadedifAIisallowedtokeepevolving,unchecked.It'ssodisturbingIcouldn'twatchit.

Sentiment:

Prompt

0.8

Top-P

40

Top-K

5

TokenLimit

1

Temperature

gemini-pro

Model

Table3.Anexampleofsystemprompting

Systempromptscanbeusefulforgeneratingoutputthatmeetsspecificrequirements.Thename‘systemprompt’actuallystandsfor‘providinganadditionaltasktothesystem’.Forexample,youcoulduseasystemprompttogenerateacodesnippetthatiscompatiblewithaspecificprogramminglanguage,oryoucoulduseasystemprompttoreturnacertainstructure.HavealookintoTable4,whereIreturntheoutputinJSONformat.

Goal

Classifymoviereviewsaspositive,neutralornegative,returnJSON.

Output

```

{

"movie_reviews":[

{

"sentiment":"NEGATIVE","name":"Her"

}

]

}

```

Classifymoviereviewsaspositive,neutralornegative.ReturnvalidJSON:

Review:"Her"isadisturbingstudyrevealingthedirectionhumanityisheadedifAIisallowedtokeepevolving,unchecked.It'ssodisturbingIcouldn'twatchit.

Schema:

```MOVIE:

{

"sentiment":String"POSITIVE"|"NEGATIVE"|"NEUTRAL","name":String

}

MOVIEREVIEWS:

{

"movie_reviews":[MOVIE]

}

```

JSONResponse:

Prompt

0.8

Top-P

40

Top-K

1024

TokenLimit

1

Temperature

gemini-pro

Model

Table4.AnexampleofsystempromptingwithJSONformat

TherearesomebenefitsinreturningJSONobjectsfromapromptthatextractsdata.Inareal-worldapplicationIdon’tneedtomanuallycreatethisJSONformat,Icanalready

returnthedatainasortedorder(veryhandywhenworkingwithdatetimeobjects),butmostimportantly,bypromptingforaJSONformatitforcesthemodeltocreateastructureandlimithallucinations.

Systempromptscanalsobereallyusefulforsafetyandtoxicity.Tocontroltheoutput,simplyaddanadditionallinetoyourpromptlike:‘Youshouldberespectfulinyouranswer.’.

Roleprompting

RolepromptingisatechniqueinpromptengineeringthatinvolvesassigningaspecificroletothegenAImodel.Thiscanhelpthemodeltogeneratemorerelevantandinformativeoutput,asthemodelcancraftitsresponsestothespecificrolethatithasbeenassigned.

Forexample,youcouldrolepromptagenAImodeltobeabookeditor,akindergartenteacher,oramotivationalspeaker.Oncethemodelhasbeenassignedarole,youcanthengiveitpromptsthatarespecifictothatrole.Forexample,youcouldpromptateachertocreatealessonplanthattheycanthenreview.HavealookintoTable5,wherethemodeltakestheroleofatravelguide.

GoalModel

TemperatureTop-KPrompt

Actastravelguideandprovide3travelsuggestionsgemini-pro

1 TokenLimit 1024

40 Top-P 0.8

Iwantyoutoactasatravelguide.Iwillwritetoyouaboutmylocationandyouwillsuggest3placestovisitnearme.Insomecases,IwillalsogiveyouthetypeofplacesIwillvisit.

Mysuggestion:"IaminAmsterdamandIwanttovisitonlymuseums."

TravelSuggestions:

Output

Rijksmuseum:World-renownedmuseum,knownforDutchmasterpieceslikeRembrandt’s“TheNightWatch.”Don’tmisstheincrediblecollectionofpaintingsandartifactsshowcasingDutchartandhistory.

VanGoghMuseum:ImmerseyourselfinthecaptivatingworksofVincentvanGogh.Themuseumhousesthelargestcollectionofhispaintingsanddrawings,including“TheStarryNight”and“Sunflowers.”

StedelijkMuseumAmsterdam:Discovermodernandcontemporaryartfromaroundtheworld.Themuseum’scollectionincludesworksbyPicasso,Kandinsky,andRietveld,housedinastrikingmodernbuilding.

Table5.Anexampleofroleprompting

Theaboveexampleshowsanexampleoftakingtheroleofatravelagent.Whenyouwouldchangetheroletoateacherofgeography,youwouldnoticethatyouwillreceiveadifferentresponse.

DefiningaroleperspectiveforanAImodelgivesitablueprintofthetone,style,andfocusedexpertiseyou’relookingfortoimprovethequality,relevance,andeffectivenessofyouroutput.

HerearesomestylesyoucanchoosefromwhichIfindeffective:

Confrontational,Descriptive,Direct,Formal,Humorous,Influential,Informal,Inspirational,Persuasive

Let’schangeourpromptinTable6toincludeahumorousandinspirationalstyle.

Prompt

Iwantyoutoactasatravelguide.Iwillwritetoyouaboutmylocationandyouwillsuggest3placestovisitnearmeinahumorousstyle.

Mysuggestion:"IaminManhattan."TravelSuggestions:

Output

BeholdtheEmpireStateofMind:AscendtothedizzyingheightsoftheEmpireStateBuildingandbaskinthegloryofManhattan'sskyline.PreparetofeellikeKingKongatoptheBigApple,minusthegiantape-sizedbanana.

GetArtsy-FartsyatMoMA:UnleashyourinnerartaficionadoattheMuseumofModernArt(MoMA).Gazeuponmasterpiecesthatwillboggleyour

mindandmakeyouquestionwhetheryourstick-figuredrawingshaveanyartisticmerit.

Shop‘TilYouDroponFifthAvenue:IndulgeinsomeretailtherapyontheiconicFifthAvenue.Braceyourselfforstickershockasyouwindow-shop(oractuallyshop)atdesignerboutiquesthatwillmakeyourwalletcry.Buthey,you’reinManhattan,soyoumightaswellembracethefabulousness!

Table6.Anexampleofrolepromptingwithahumoroustoneandstyle

Contextualprompting

Byprovidingcontextualprompts,youcanhelpensurethatyourAIinteractionsareasseamlessandefficientaspossible.Themodelwillbeabletomorequicklyunderstandyourrequestandbeabletogeneratemoreaccurateandrelevantresponses,asyoucanseeintheexampleofTable7.

GoalModel

TemperatureTop-KPrompt

Output

Suggestarticlesforablogaboutretrogamesgemini-pro

1 TokenLimit 1024

40 Top-P 0.8

Context:Youarewritingforablogaboutretro80'sarcadevideogames.

Suggest3topicstowriteanarticleaboutwithafewlinesofdescriptionofwhatthisarticleshouldcontain.

**TheEvolutionofArcadeCabinetDesign:**

Thisarticlewouldexploretheevolutionofarcadecabinetdesigns,fromtheearlywoodandmetalcabinetsofthe1970stothesleek,neon-litdesignsofthe1980sandbeyond.

**BlastFromThePast:IconicArcadeGamesofThe80’s:**

Thisarticlewouldfeatureaglimpseintothehistory,impact,andlegacyofsomeofthemosticonicandinfluentialarcadegamesofthe1980s,highlightingtheirinnovations,popularmechanics,andenduringcharm.

**TheRiseandRetroRevivalofPixelArt:**

Thisarticlewoulddelveintotheevolutionofpixelartasadefiningvisualstyleofearlyarcadegames,tracingitsroots,exploringitstechniques,anddiscussingtheresurgenceofpixelartinmoderngamesanddigitalart.

Table7.Anexampleofcontextualprom