The Reference · Working Knowledge

271 terms

The essentials

Start with these 15

LLM Token Hallucination RAG Agent API Prompt Engineering Context Window Inference Attention GPU Open vs Closed Weights Alignment Prompt Injection System Prompt

.env / SecretsConfig and keys kept out of codeEnvironments & PackagingInter .gitignoreFiles git should never trackVersion ControlInter AdamAdaptive Moment EstimationHow Models LearnAdv AgentAgentic AIGetting Models to ActCore Agent MemoryHow an agent remembers across stepsGetting Models to ActInter Agent RuntimeThe loop that runs an agentThe WorkbenchInter AGENTS.mdProject instructions for agentsThe WorkbenchInter AGIArtificial General IntelligenceThe Big PictureConcept AIArtificial IntelligenceThe Big PictureCore AlignmentMatching AI behavior to intentSafety & AlignmentCore APIApplication Programming InterfaceGetting Models to ActCore API Key / TokenYour secret credential for a serviceEnvironments & PackagingCore Approval ModeHuman-in-the-loop for tool useThe WorkbenchCore ASIArtificial SuperintelligenceThe Big PictureConcept ASICApplication-Specific Integrated CircuitHardware & InfrastructureInter ASR / STTAutomatic Speech Recognition / Speech-to-TextLanguage & PerceptionInter AttentionHow a model weighs relevant contextModel ArchitecturesCore Background AgentAsync delegated agentThe WorkbenchInter Background JobsKeep a command runningThe Command LineAdv BackpropBackpropagationHow Models LearnInter BashA widely used Unix shellThe Command LineCore BenchmarkA standardized model testMeasuring ModelsCore BisectFind the commit that broke itVersion ControlAdv BlameWho last changed each lineVersion ControlInter BLEU / ROUGETranslation / summarization metricsMeasuring ModelsAdv BranchA parallel line of workVersion ControlCore Browser ControlAgent-driven browsingThe WorkbenchInter Bug / IssueA defect, and the ticket for itShipping & CollaborationCore BuildTurn source into a runnable appEnvironments & PackagingCore BundlerPacks your code for the browserEnvironments & PackagingInter CAIConstitutional AIHow Models LearnAdv CapexCapital ExpenditureHardware & InfrastructureCore cat / lessRead a file in the terminalThe Command LineCore Checkout / SwitchMove between branches or commitsVersion ControlCore CheckpointRollback point in agent workThe WorkbenchInter Cherry-pickCopy one commit to another branchVersion ControlAdv chmod / PermissionsWho can read, write, or run a fileThe Command LineAdv CI/CDAutomated testing and shippingShipping & CollaborationInter Citation GroundingAnswers tied to sourcesThe WorkbenchInter CLAUDE.mdClaude Code project instructionsThe WorkbenchInter CLICommand-Line InterfaceThe Command LineCore CloneCopy a repo to your machineVersion ControlCore Cloud AgentRemote / delegated agentThe WorkbenchInter Cloud SandboxRemote execution environmentThe WorkbenchInter ClusterA networked array of GPUsHardware & InfrastructureInter CNNConvolutional Neural NetworkModel ArchitecturesInter Code ReviewA teammate checks your changeShipping & CollaborationCore Codebase IndexSearchable repo mapThe WorkbenchInter Coding AgentAgent for software workThe WorkbenchCore Column / FieldOne attribute of every rowDatabases & SQLCore CommandOne typed instructionThe Command LineCore Command ChainingRun commands in sequenceThe Command LineInter CommitA saved snapshot of changesVersion ControlCore Compiler / InterpreterTwo ways code gets runEnvironments & PackagingInter ComputeRaw processing powerHardware & InfrastructureCore Computer UseAgents that operate a screenGetting Models to ActInter Context CompactionShrinking what the model seesThe WorkbenchAdv Context WindowHow much a model can “see” at onceHow Models RunCore CoTChain of ThoughtGetting Models to ActCore cp / mv / rmCopy, move, delete filesThe Command LineCore CPUCentral Processing UnitHardware & InfrastructureCore cronRun jobs on a scheduleThe Command LineAdv CRUDCreate, Read, Update, DeleteDatabases & SQLInter Ctrl-C / Ctrl-DInterrupt, and end-of-inputThe Command LineCore CUDANvidia's GPU programming platformHardware & InfrastructureInter curl / wgetFetch from the internet by commandThe Command LineInter Cursor RulesCursor project instructionsThe WorkbenchInter CVComputer VisionLanguage & PerceptionCore Data PrivacyControl over personal and sensitive dataSafety & AlignmentCore Data WallRunning out of human textHow Models LearnInter DatabaseAn organized store of dataDatabases & SQLCore DependencyOutside code your project needsEnvironments & PackagingCore DeployPut code live for usersShipping & CollaborationCore DiffThe line-by-line changeVersion ControlInter DiffusionDiffusion ModelModel ArchitecturesCore DirectoryA folderThe Command LineCore DistillationTeacher → student compressionHow Models LearnInter DLDeep LearningThe Big PictureCore DNNDeep Neural NetworkThe Big PictureCore Docker / ContainerAn app sealed with its environmentEnvironments & PackagingInter DotfilesYour shell's config (.zshrc)The Command LineInter DPODirect Preference OptimizationHow Models LearnAdv EmbeddingMeaning as numbersHow Models RunInter EndpointAn API's specific addressGetting Models to ActInter Environment VariableA named value the system holdsThe Command LineInter EU AI ActThe EU's risk-based AI lawSafety & AlignmentInter EvalsTests you build for your own useMeasuring ModelsCore Exit CodeDid the command succeed?The Command LineAdv F1F1 ScoreMeasuring ModelsInter FetchDownload changes without mergingVersion ControlInter File PathA file's addressThe Command LineCore findLocate files by nameThe Command LineInter Fine-tuningAdapting a trained modelHow Models LearnCore Flag / ArgumentThe options after a commandThe Command LineInter FLOP / FLOPSFloating-Point Operations (Per Second)Hardware & InfrastructureInter Foreign KeyA link to another tableDatabases & SQLAdv ForkYour own copy of someone's repoVersion ControlInter FP32 / FP16 / BF16Floating-point precision formatsHow Models RunAdv FP8 / INT8 / INT4Low-precision inference formatsHow Models RunAdv FrameworkA structure to build insideEnvironments & PackagingInter Frontier ModelThe largest, most capable modelsThe Big PictureCore GANGenerative Adversarial NetworkModel ArchitecturesInter Gateway DaemonAlways-on agent hostThe WorkbenchAdv GEMINI.mdGemini project instructionsThe WorkbenchInter GitThe standard for tracking code changesVersion ControlCore git statusWhat's changed right nowVersion ControlCore GitHub / GitLabWhere repos live onlineVersion ControlCore GPTGenerative Pre-trained TransformerModel ArchitecturesCore GPUGraphics Processing UnitHardware & InfrastructureCore grepSearch text for a patternThe Command LineInter GroundingTying an answer to external evidenceGetting Models to ActCore GSM8KGrade School Math 8KMeasuring ModelsInter GuardrailsSafety constraints around a modelSafety & AlignmentInter HallucinationConfident fabricationMeasuring ModelsCore HarnessAgent harness / workbenchThe WorkbenchCore HBMHigh Bandwidth MemoryHardware & InfrastructureAdv head / tailFirst or last lines of a fileThe Command LineInter HomebrewThe package manager for a MacEnvironments & PackagingInter HooksAgent lifecycle scriptsThe WorkbenchAdv HumanEvalCoding ability benchmarkMeasuring ModelsInter ICLIn-Context LearningGetting Models to ActInter IDE AgentIn-editor agent harnessThe WorkbenchCore ImageThe blueprint for a containerEnvironments & PackagingAdv IndexA lookup that makes search fastDatabases & SQLAdv InferenceRunning a trained modelHow Models RunCore Inference ProviderWho runs the modelThe WorkbenchInter JailbreakBypassing safety guardrailsSafety & AlignmentInter JOINCombine two tablesDatabases & SQLInter KV CacheKey-Value CacheHow Models RunAdv Latency / ThroughputTwo competing speed metricsHow Models RunInter Linter / FormatterAutomatic code-quality toolsShipping & CollaborationInter LLMLarge Language ModelModel ArchitecturesCore Local ModelOn-device inferenceThe WorkbenchInter Local SandboxOn-machine execution boundaryThe WorkbenchInter localhost / PortYour own machine, and its doorsShipping & CollaborationInter LockfilePins exact dependency versionsEnvironments & PackagingAdv LogThe history of commitsVersion ControlCore LoggingThe record an app writes about itselfShipping & CollaborationInter LoRALow-Rank AdaptationHow Models LearnAdv LSTMLong Short-Term MemoryModel ArchitecturesInter main / HEADThe official line, and where you areVersion ControlInter man / --helpGet help on a commandThe Command LineInter MCPModel Context ProtocolGetting Models to ActInter MCP ServerMCP tool providerThe WorkbenchInter MergeCombine one branch into anotherVersion ControlCore Merge ConflictWhen two changes collideVersion ControlInter MFUModel FLOPs UtilizationHardware & InfrastructureAdv MigrationA versioned schema changeDatabases & SQLInter MLMachine LearningThe Big PictureCore MMLUMassive Multitask Language UnderstandingMeasuring ModelsInter Model CardDocumentation for a model releaseMeasuring ModelsInter Model RouterMulti-model gatewayThe WorkbenchAdv MoEMixture of ExpertsModel ArchitecturesAdv MonorepoOne repo, many packagesEnvironments & PackagingInter Multi-channel AgentSame agent, many surfacesThe WorkbenchAdv MultimodalMultiple data types, unifiedLanguage & PerceptionCore NLPNatural Language ProcessingLanguage & PerceptionCore NLU / NLGNatural Language Understanding / GenerationLanguage & PerceptionInter NNNeural NetworkThe Big PictureCore node_modulesWhere installed packages liveEnvironments & PackagingInter NoSQLNon-table databasesDatabases & SQLInter NPUNeural Processing UnitHardware & InfrastructureInter npxRun a package without installing itEnvironments & PackagingInter NVLink / InfiniBandHigh-speed GPU interconnectsHardware & InfrastructureAdv OCROptical Character RecognitionLanguage & PerceptionInter Open Source / LicensePublic code, and its rulesShipping & CollaborationInter Open vs Closed WeightsDownloadable, or API-onlyThe Big PictureInter ORMObject-Relational MapperDatabases & SQLAdv Package / LibraryReusable code you plug inEnvironments & PackagingCore Package ManagerInstalls the code you depend onEnvironments & PackagingCore package.jsonA JS project's manifestEnvironments & PackagingCore ParametersThe numbers a model learnsThe Big PictureCore PATHWhere the shell finds programsThe Command LineInter PEFTParameter-Efficient Fine-TuningHow Models LearnAdv PerplexityHow “surprised” a model isMeasuring ModelsAdv Personal AgentAssistant across your lifeThe WorkbenchInter PipeFeed one command into the nextThe Command LineInter PR AgentAgent that ships via pull requestThe WorkbenchInter Pre-trainingThe first, expensive learning stageHow Models LearnCore Primary KeyA row's unique IDDatabases & SQLInter Process / KillA running program, and stopping itThe Command LineInter Prompt CachingReusing repeated input processingHow Models RunInter Prompt EngineeringPhrasing inputs for better outputsGetting Models to ActCore Prompt InjectionHiding instructions in what a model readsSafety & AlignmentCore ProvenanceWhere data or output came fromSafety & AlignmentInter PullFetch others' latest changesVersion ControlCore Pull RequestPropose and review a merge (a “PR”)Version ControlCore PushSend commits to the shared copyVersion ControlCore QLoRAQuantized Low-Rank AdaptationHow Models LearnAdv QuantizationLower-precision compressionHow Models RunInter QueryA request to a databaseDatabases & SQLCore RAGRetrieval-Augmented GenerationGetting Models to ActCore Rate LimitA cap on how much you can callGetting Models to ActInter ReActReasoning + ActingGetting Models to ActAdv Reasoning ModelTrained to think before answeringModel ArchitecturesCore RebaseReplay your work on the latestVersion ControlAdv Red TeamingAttacking a model on purposeSafety & AlignmentInter RedirectSend output to a fileThe Command LineInter Remote / originThe shared, hosted copyVersion ControlInter REPLAn interactive code promptThe Command LineInter RepositoryA project under git (a “repo”)Version ControlCore Revert / ResetUndo committed changesVersion ControlInter RLAIFReinforcement Learning from AI FeedbackHow Models LearnAdv RLHFReinforcement Learning from Human FeedbackHow Models LearnCore RNNRecurrent Neural NetworkModel ArchitecturesInter RollbackUndo a bad releaseShipping & CollaborationInter Row / RecordOne entry in a tableDatabases & SQLCore RuntimeThe thing that runs your codeEnvironments & PackagingInter SamplingChoosing the next generated tokenHow Models RunCore Scaling LawsCompute → capability, predictablyHow Models LearnCore SchemaA database's blueprintDatabases & SQLInter scp / rsyncCopy files between machinesThe Command LineAdv ScriptA saved file of commandsThe Command LineInter SDKSoftware Development KitGetting Models to ActInter Search GroundingWeb search in the loopThe WorkbenchInter sed / awkTransform text in a streamThe Command LineAdv SELECTRead data from a tableDatabases & SQLCore Semantic Code SearchMeaning-based code lookupThe WorkbenchInter Semantic VersioningMAJOR.MINOR.PATCH numberingEnvironments & PackagingInter ServerAn always-on computer running your appShipping & CollaborationCore SFTSupervised Fine-TuningHow Models LearnInter SGDStochastic Gradient DescentHow Models LearnInter SkillsPackaged agent capabilitiesThe WorkbenchAdv SQLThe language for querying databasesDatabases & SQLCore SquashCombine many commits into oneVersion ControlInter SSHSecure login to another machineThe Command LineInter SSMState Space ModelModel ArchitecturesAdv StagingChoosing what goes in a commitVersion ControlInter Staging vs ProductionRehearsal copy vs the real thingShipping & CollaborationInter StashShelve unfinished changesVersion ControlInter stdout / stderrA program's two output streamsThe Command LineAdv SubagentAgent spawned by an agentThe WorkbenchAdv SubmoduleA repo inside a repoVersion ControlAdv sudoRun as administratorThe Command LineInter Synthetic DataAI-generated training dataHow Models LearnInter System PromptHigh-priority application instructionsGetting Models to ActCore TableRows and columns of dataDatabases & SQLCore Tag / ReleaseMark a specific versionVersion ControlInter tar / zipBundle and compress filesThe Command LineInter TCOTotal Cost of OwnershipHardware & InfrastructureInter TemperatureA sampling randomness controlHow Models RunCore TerminalWhere you type commands (the shell)The Command LineCore Terminal AgentShell-native coding agentThe WorkbenchCore Test CoverageHow much code the tests exerciseShipping & CollaborationInter TokenThe unit a model reads & writesHow Models RunCore Token BudgetSpending limit per taskThe WorkbenchInter Tool Call TraceLog of agent actionsThe WorkbenchInter Tool PermissionWhat an agent may invokeThe WorkbenchCore Tool UseFunction CallingGetting Models to ActCore ToTTree of ThoughtsGetting Models to ActAdv touchCreate an empty fileThe Command LineInter TPSTokens Per SecondHow Models RunInter TPUTensor Processing UnitHardware & InfrastructureInter Training DataWhat a model learns fromHow Models LearnCore TransactionAll-or-nothing changesDatabases & SQLAdv TransformerThe architecture behind modern AIModel ArchitecturesCore TTFTTime To First TokenHow Models RunInter TTSText-to-SpeechLanguage & PerceptionInter Unit TestAn automated check on one piece of codeShipping & CollaborationCore VAEVariational AutoencoderModel ArchitecturesAdv Vector DBVector DatabaseGetting Models to ActInter Virtual EnvironmentAn isolated sandbox per projectEnvironments & PackagingInter VLMVision Language ModelModel ArchitecturesInter VRAMVideo RAMHardware & InfrastructureInter WebhookOne service pings anotherShipping & CollaborationAdv WHEREFilter to matching rowsDatabases & SQLInter WildcardsMatch many files with a patternThe Command LineInter Working DirectoryThe folder you're acting inThe Command LineInter WorktreeIsolated git working copyThe WorkbenchAdv Zero / One / Few-shotExamples given inside the promptGetting Models to ActCore zshThe modern default shellThe Command LineInter