Kelly Mears

KnowledgeDomain 03

Open map

Agents & Language Models

How systems built on language models actually work, and how they fail.

Notes
58
Links within
272
Links outward
200
Knowledge graph: 166 notes, 472 links.Each dot is a note, sized by how many links it has and colored by topic. Focus or hover a dot to isolate it and its neighbors; activate it to open the note. The graph is a single tab stop: use the arrow keys to move between notes, Home and End for the most and least linked, Enter to open the focused note, and Escape to leave the graph.Ground TruthSilent FailurePlausible MechanismDeterminismGoodhart's LawFalsifiabilityAccessibilityProvenanceAnchoring EffectCode ReviewLarge Language ModelLeast PrivilegeRace ConditionIdempotenceContainerizationSecret ManagementObservabilityProcedural GenerationCharacter EncodingCache InvalidationFlaky TestHash FunctionDocumentation RotCode SigningContext WindowFail FastTool UseVacuous TruthPull RequestScreen ReaderSeeded RandomnessNondeterminismSupply Chain SecurityExhaustive ClaimFirewallPrompt EngineeringSchema ValidationTokenHallucinationSensitivity and SpecificityStreaming ResponseToken BudgetWrite-Ahead LoggingAssertionBase Rate FallacyCronEvaluation HarnessPrompt InjectionResource StarvationTest DoubleConfidence TrickGuardrailKnowledge GraphLinguistic RelativityAgentic LoopAutomatic Speech RecognitionControl GroupsDefensive DefaultHeisenbugPersonal DataSchema DriftSystem CallSystem PromptTest FixtureWebhookWikiGit WorktreeLoad BalancingMechanistic InterpretabilityService VirtualizationSQL InjectionAgent SandboxAutocompleteBacklinkChangelogData CompressionJSON SchemaLatency and JitterPrecision and RecallPromptRecord and Replay TestingServer-Side Request ForgeryStructured DataAdversarial ReviewAutomation BiasBackground AgentLeast Recently Used CacheLinux NamespacesMalaphorStructured OutputTail LatencyTime as ResourceVirtual MemoryZettelkastenCopy-on-WriteDesign by ContractFair-Play MysteryFingerprintHuman in the LoopLLM-as-JudgeMarkov ChainMondegreenPrompt CachingRetrieval-Augmented GenerationTrieType SystemAgent MemoryConstrained DecodingFan-Out and Fan-InForm ValidationGraphics Processing UnitMemoizationModel Context ProtocolModel RoutingMonadMulti-Agent OrchestrationOpenAPIOverfittingShibbolethSubagentTest OracleUnreachable CodeKV CachePrototype PollutionQuantizationSegmented SleepSnowcloneSuperuserType InferenceAgent SkillDisco ElysiumFine-TuningIn-Context LearningLexical AnalysisOccam's RazorRerankingSchema EvolutionSubtitlingWeb Search EngineXMLByte-Pair EncodingChrootDuck TypingEntropyFilesystem SnapshotTemperature SamplingTemporal UpscalingThreat ModelingTypeScriptEmbeddingMixture of ExpertsOnline ReviewOptical Character RecognitionReasoning ModelRecurrent Neural NetworkStemming and LemmatizationText-to-SpeechVector DatabaseChain-of-Thought PromptingDraft and PublishedWake WordModel CardRegularizationAblation StudyHeat MapObject Detection

Drag to pan · hold ⌘/ctrl and scroll — or pinch — to zoom.

01

Where it reaches

02

The hubs

Large Language Model

A neural network trained to predict text, used as a general-purpose instruction-following system.

Agents & Language Models38 connections

Context Window

The bounded span of tokens a model can attend to in a single request.

Agents & Language Models27 connections

Tool Use

Giving a model a set of callable functions so it can act on the world rather than only describe it.

Agents & Language Models27 connections

Nondeterminism

The property that identical inputs may produce different outputs, and what it costs to test around.

Agents & Language Models24 connections

03

Every note

A

B

C

E

F

G

H

I

J

K

L

M

N

O

P

Q

R

S

T

V

W