pro·to·logue noun
- In biological nomenclature, the original description published together with a new name, against which every later use of that name is checked.
- This site: the reference record for 230 prompting and large language model techniques, each with a short citable definition, its relations to other techniques, and the paper that introduced it.
Foundations
Core concepts of prompting — the parts of a prompt, how models consume it, and the basic zero-shot and few-shot paradigms.
- Chat TemplatePTL-0107
- Context WindowPTL-0004
- DelimitersPTL-0014
- Diffusion Language ModelPTL-0228
- Emergent AbilitiesPTL-0184
- Few-shot PromptingPTL-0009
- Fine-TuningPTL-0105
- Knowledge DistillationPTL-0183
- Foundation ModelPTL-0103
- Large Language ModelPTL-0102
- Base ModelPTL-0104
- Large Language ModelPTL-0102
- Greedy DecodingPTL-0109
- Instruction FollowingPTL-0113
- Instruction TuningPTL-0010
- Knowledge CutoffPTL-0112
- KV CachePTL-0182
- Log ProbabilitiesPTL-0110
- Min-p SamplingPTL-0218
- Mixture of ExpertsPTL-0229
- PrefillPTL-0015
- PromptPTL-0001
- Multimodal PromptingPTL-0111
- Prompt TemplatePTL-0013
- System PromptPTL-0003
- Prompt EngineeringPTL-0002
- Reinforcement Learning from Human FeedbackPTL-0011
- Direct Preference OptimizationPTL-0012
- Reward ModelPTL-0106
- Role PromptingPTL-0016
- Speculative DecodingPTL-0217
- System One ModelPTL-0224
- TemperaturePTL-0006
- TokenPTL-0005
- Top-k SamplingPTL-0108
- Top-p SamplingPTL-0007
- Zero-shot PromptingPTL-0008
Exemplars & In-Context Learning
How demonstrations inside a prompt are chosen, ordered, scaled, and calibrated, and what they actually teach the model.
- Exemplar OrderingPTL-0020
- Exemplar SelectionPTL-0019
- Active PromptingPTL-0023
- Few-shot CalibrationPTL-0021
- In-Context LearningPTL-0017
- Demonstration Label SensitivityPTL-0018
- Many-shot In-Context LearningPTL-0022
Reasoning Elicitation
Techniques that get a model to produce intermediate reasoning, decompose problems, or explore multiple solution paths before answering.
- Analogical PromptingPTL-0036
- Chain-of-Thought PromptingPTL-0024
- Automatic Chain-of-ThoughtPTL-0027
- Chain of DraftPTL-0116
- Complexity-Based PromptingPTL-0115
- Contrastive Chain-of-ThoughtPTL-0038
- Self-ConsistencyPTL-0028
- Universal Self-ConsistencyPTL-0029
- Thread of ThoughtPTL-0039
- Tree of ThoughtsPTL-0033
- Graph of ThoughtsPTL-0034
- Zero-shot Chain-of-ThoughtPTL-0025
- Plan-and-Solve PromptingPTL-0031
- Decomposed PromptingPTL-0114
- Generated Knowledge PromptingPTL-0044
- Least-to-Most PromptingPTL-0030
- Maieutic PromptingPTL-0043
- Program of ThoughtsPTL-0046
- Program-Aided Language ModelsPTL-0045
- Rephrase and RespondPTL-0041
- ScratchpadPTL-0026
- Self-AskPTL-0042
- Self-DiscoverPTL-0037
- Skeleton-of-ThoughtPTL-0035
- Step-Back PromptingPTL-0032
- System 2 AttentionPTL-0040
Self-Critique & Verification
Techniques in which a model, or a set of models, checks, critiques, votes on, or revises outputs.
- Best-of-N SamplingPTL-0054
- Chain-of-VerificationPTL-0049
- Confidence CalibrationPTL-0121
- LLM-as-a-JudgePTL-0050
- Mixture-of-AgentsPTL-0052
- Multi-Agent DebatePTL-0051
- Process Reward ModelPTL-0053
- ReflexionPTL-0048
- Self-CorrectionPTL-0117
- CRITICPTL-0119
- Self-RefinePTL-0047
- Chain of DensityPTL-0120
- Self-VerificationPTL-0118
- SelfCheckGPTPTL-0122
Retrieval & Tool Use
Grounding generation in external information and letting models call functions, search, and other tools.
- BM25PTL-0127
- ChunkingPTL-0125
- Citation GenerationPTL-0139
- ColBERTPTL-0130
- Contextual RetrievalPTL-0132
- Dense RetrievalPTL-0126
- EmbeddingPTL-0123
- Function CallingPTL-0060
- Code Execution ToolPTL-0138
- Programmatic Tool CallingPTL-0223
- Tool SearchPTL-0222
- Web Search ToolPTL-0137
- Hybrid SearchPTL-0128
- Model Context ProtocolPTL-0061
- Query RewritingPTL-0131
- ReActPTL-0058
- RerankingPTL-0129
- Retrieval-Augmented GenerationPTL-0055
- Active RetrievalPTL-0136
- Agentic RAGPTL-0134
- Corrective RAGPTL-0135
- GraphRAGPTL-0133
- Hypothetical Document EmbeddingsPTL-0056
- Self-RAGPTL-0057
- Structured OutputsPTL-0062
- ToolformerPTL-0059
- Vector DatabasePTL-0124
Agents & Orchestration
Patterns for composing multiple model calls, tools, and memory into workflows and autonomous agents.
- Agent MemoryPTL-0071
- Agent SkillsPTL-0148
- Agent2Agent ProtocolPTL-0147
- Agentic WorkflowPTL-0064
- Evaluator-OptimizerPTL-0069
- Orchestrator-WorkersPTL-0068
- ParallelizationPTL-0067
- Plan-and-ExecutePTL-0141
- Prompt ChainingPTL-0065
- RoutingPTL-0066
- AI AgentPTL-0063
- Coding AgentPTL-0145
- Computer-Use AgentPTL-0144
- Context CompactionPTL-0149
- Context EngineeringPTL-0072
- Dark FactoryPTL-0225
- HandoffPTL-0142
- Harness EngineeringPTL-0212
- Human-in-the-LoopPTL-0143
- Meta-PromptingPTL-0070
- Multi-Agent SystemPTL-0140
- Spec-Driven DevelopmentPTL-0226
- Vibe CodingPTL-0146
Prompt Optimization
Automatic and learned methods that search for, compress, or train better prompts.
- Activation SteeringPTL-0219
- Automatic Prompt EngineerPTL-0073
- Directional Stimulus PromptingPTL-0079
- DSPyPTL-0075
- Emotion PromptingPTL-0082
- Low-Rank AdaptationPTL-0078
- Model CascadePTL-0155
- Optimization by PromptingPTL-0074
- Prefix TuningPTL-0077
- Prompt CachingPTL-0081
- Prompt CompressionPTL-0080
- Prompt OptimizationPTL-0150
- Agentic Context EngineeringPTL-0214
- EvoPromptPTL-0154
- PromptbreederPTL-0153
- ProTeGiPTL-0151
- TextGradPTL-0152
- Prompt TuningPTL-0076
- QuantizationPTL-0230
Reasoning Models & Test-Time Compute
Models trained to reason at length, and methods that improve answers by spending more computation at inference time.
- Chain-of-Thought MonitoringPTL-0158
- Group Relative Policy OptimizationPTL-0216
- Reasoning ModelPTL-0083
- Extended ThinkingPTL-0085
- Reinforcement Learning with Verifiable RewardsPTL-0156
- Self-Taught ReasonerPTL-0086
- Test-Time Compute ScalingPTL-0084
- Budget ForcingPTL-0157
- Recursive Language ModelPTL-0213
- Sleep-time ComputePTL-0215
Security & Adversarial Prompting
Attacks that subvert a model's instructions and the defenses designed to resist them.
- Constitutional AIPTL-0095
- Deliberative AlignmentPTL-0167
- Dual LLM PatternPTL-0161
- Excessive AgencyPTL-0166
- GuardrailsPTL-0160
- Constitutional ClassifiersPTL-0168
- Instruction HierarchyPTL-0093
- JailbreakPTL-0090
- Adversarial SuffixPTL-0091
- Crescendo AttackPTL-0164
- Many-shot JailbreakingPTL-0092
- PAIRPTL-0165
- Lethal TrifectaPTL-0163
- Prompt InjectionPTL-0087
- Indirect Prompt InjectionPTL-0088
- Tool PoisoningPTL-0162
- Prompt LeakingPTL-0089
- Indirect Prompt InjectionPTL-0088
- Red TeamingPTL-0159
- SpotlightingPTL-0094
Failure Modes & Evaluation
Systematic ways prompted models go wrong, and the evaluations used to measure them.
- Alignment FakingPTL-0221
- Context RotPTL-0169
- Data ContaminationPTL-0178
- Emergent MisalignmentPTL-0220
- EvalsPTL-0179
- pass@kPTL-0180
- RAG EvaluationPTL-0181
- Evaluation AwarenessPTL-0227
- HallucinationPTL-0096
- Irrelevant Context DistractionPTL-0173
- LLM Judge BiasPTL-0177
- Lost in the MiddlePTL-0098
- Multi-Turn DegradationPTL-0174
- Needle in a HaystackPTL-0099
- Over-RefusalPTL-0176
- OverthinkingPTL-0170
- Prompt SensitivityPTL-0100
- Reversal CursePTL-0171
- Reward HackingPTL-0172
- SycophancyPTL-0097
- Text DegenerationPTL-0175
- Unfaithful Chain-of-ThoughtPTL-0101
Benchmarks
Standard test sets used to measure and compare language models, what each one measures, how it is scored, the prompting setup scores are reported under, and its known limitations.
- BenchmarkPTL-0185
- AIMEPTL-0199
- ARC-AGIPTL-0192
- Benchmark SaturationPTL-0186
- BIG-Bench HardPTL-0191
- BrowseCompPTL-0207
- Chatbot ArenaPTL-0210
- GAIAPTL-0206
- GPQAPTL-0189
- GSM8KPTL-0197
- HellaSwagPTL-0195
- HumanEvalPTL-0200
- Humanity's Last ExamPTL-0190
- IFEvalPTL-0208
- LiveCodeBenchPTL-0201
- MATHPTL-0198
- MMLUPTL-0187
- MMLU-ProPTL-0188
- MMMUPTL-0196
- MT-BenchPTL-0209
- OSWorldPTL-0205
- RULERPTL-0211
- SimpleQAPTL-0193
- SWE-benchPTL-0202
- Terminal-BenchPTL-0203
- TruthfulQAPTL-0194
- τ-benchPTL-0204
New terms, as they're coined
A weekly email listing the prompting and LLM techniques that appeared that week, each with a definition you can cite. Nothing else.
Delivered through Beehiiv. Unsubscribe from any issue.