| id,name,type,description,attributes | |
| zero-shot,Zero-Shot Prompting,Technique,,"{""input"": ""Query (instruction only)"", ""output"": ""Direct answer"", ""cost"": ""minimal"", ""llm_calls"": 1, ""requires_large_model"": false}" | |
| few-shot,Few-Shot Prompting,Technique,,"{""input"": ""Query + demonstration examples"", ""output"": ""Answer following demonstrated pattern"", ""cost"": ""low"", ""llm_calls"": 1, ""requires_large_model"": false}" | |
| cot,Chain-of-Thought (CoT) Prompting,Technique,,"{""input"": ""Query + exemplars with reasoning steps"", ""output"": ""Reasoning trace → final answer"", ""cost"": ""low"", ""llm_calls"": 1, ""requires_large_model"": true, ""note"": ""Emergent at ≥100B params (PaLM-540B)""}" | |
| auto-cot,Automatic Chain-of-Thought (Auto-CoT),Technique,,"{""input"": ""Question pool → clustered → auto-generated CoT exemplars + query"", ""output"": ""Reasoning trace → answer"", ""cost"": ""moderate"", ""llm_calls"": ""K+1 (K clusters)"", ""requires_large_model"": true}" | |
| self-consistency,Self-Consistency,Technique,,"{""input"": ""CoT prompt + query (sampled N times)"", ""output"": ""Set[reasoning traces] → majority-voted answer"", ""cost"": ""high"", ""llm_calls"": ""N (typically 5-40)"", ""requires_large_model"": true}" | |
| logicot,Logical CoT (LogiCoT) Prompting,Technique,,"{""input"": ""Query + reductio-ad-absurdum verification loop"", ""output"": ""Verified reasoning trace → answer"", ""cost"": ""high"", ""llm_calls"": ""iterative"", ""requires_large_model"": true}" | |
| cos,Chain-of-Symbol (CoS) Prompting,Technique,,"{""input"": ""Query + few-shot with symbolic spatial representations"", ""output"": ""Symbolic reasoning chain → answer"", ""cost"": ""low"", ""llm_calls"": 1, ""token_savings"": ""More compact than NL CoT""}" | |
| echo,Self-Harmonized CoT (ECHO),Technique,,"{""input"": ""Question pool → clustered → diverse rationales unified"", ""output"": ""Harmonized reasoning → answer"", ""cost"": ""moderate"", ""llm_calls"": ""K+2""}" | |
| logic-of-thought,Logic-of-Thought Prompting,Technique,,"{""input"": ""Query → extract propositions → apply logic laws → augmented prompt"", ""output"": ""Logic-augmented reasoning → answer"", ""cost"": ""moderate"", ""llm_calls"": 2}" | |
| cd-cot,Contrastive Denoising with Noisy CoT (CD-CoT),Technique,,"{""input"": ""Query + diverse noisy CoT samples"", ""output"": ""Denoised majority answer"", ""cost"": ""high"", ""llm_calls"": ""N samples""}" | |
| r-cot,Reverse Chain-of-Thought (R-CoT),Technique,,"{""input"": ""Query → generate answer → reverse-verify by reconstructing problem"", ""output"": ""Verified answer (reconstruction check)"", ""cost"": ""moderate"", ""llm_calls"": 2}" | |
| cod,Chain of Draft (CoD),Technique,,"{""input"": ""Query + instruction for minimal-draft reasoning"", ""output"": ""Concise reasoning tokens → answer"", ""cost"": ""minimal"", ""llm_calls"": 1, ""token_savings"": ""80% fewer tokens than CoT""}" | |
| complex-cot,Complex CoT,Technique,,"{""input"": ""Query + longest/most-complex available exemplars"", ""output"": ""N sampled complex chains → top-K voted answer"", ""cost"": ""high"", ""llm_calls"": ""N samples""}" | |
| ccot,Contrastive Chain-of-Thought (CCoT),Technique,,"{""input"": ""Query + valid AND invalid reasoning demonstrations"", ""output"": ""Improved reasoning trace → answer"", ""cost"": ""low"", ""llm_calls"": 1}" | |
| contrastive-sc,Contrastive Self-Consistency,Technique,,"{""input"": ""Contrastive CoT prompt → N samples"", ""output"": ""Majority voted answer from contrastive samples"", ""cost"": ""high"", ""llm_calls"": ""N samples""}" | |
| faithful-cot,Faithful CoT,Technique,,"{""input"": ""Query → NL decomposition → symbolic program (Python/Datalog)"", ""output"": ""Interpretable symbolic reasoning → executed answer"", ""cost"": ""moderate"", ""llm_calls"": ""1 + code exec""}" | |
| tot,Tree-of-Thoughts (ToT) Prompting,Technique,,"{""input"": ""Query → tree of partial solutions explored via BFS/DFS"", ""output"": ""Best evaluated path → answer"", ""cost"": ""very_high"", ""llm_calls"": ""O(b^d) branching"", ""requires_large_model"": true}" | |
| got,Graph-of-Thought (GoT) Prompting,Technique,,"{""input"": ""Query → DAG of interconnected thoughts"", ""output"": ""Synthesized solution from graph traversal"", ""cost"": ""very_high"", ""llm_calls"": ""many""}" | |
| xot,XoT (Everything of Thoughts),Technique,,"{""input"": ""Query → adaptive switching between linear/tree/graph modes"", ""output"": ""Flexible multi-path reasoning → answer"", ""cost"": ""very_high"", ""llm_calls"": ""adaptive""}" | |
| bot,Buffer of Thoughts (BoT),Technique,,"{""input"": ""Query → retrieve thought-template → instantiate reasoning"", ""output"": ""Template-guided reasoning → answer"", ""cost"": ""moderate"", ""llm_calls"": 2, ""note"": ""12% of ToT computational cost""}" | |
| lot-layer,Layer-of-Thoughts (LoT),Technique,,"{""input"": ""Query + hierarchical constraint layers (hard → soft)"", ""output"": ""Filtered candidates → answer"", ""cost"": ""moderate"", ""llm_calls"": ""layers""}" | |
| eedp,End-to-End DAG-Path (EEDP) Prompting,Technique,,"{""input"": ""Entity-linked DAG → backbone path extraction"", ""output"": ""Flattened path input → answer"", ""cost"": ""moderate"", ""llm_calls"": 1}" | |
| least-to-most,Least-to-Most Prompting,Technique,,"{""input"": ""Query → decomposition prompt → sub-Qs solved sequentially"", ""output"": ""Sequential sub-answers → final answer"", ""cost"": ""moderate"", ""llm_calls"": ""1 + N sub-problems""}" | |
| plan-and-solve,Plan-and-Solve (PS) Prompting,Technique,,"{""input"": ""Query + 'Let's first understand and devise a plan…'"", ""output"": ""Plan → step-by-step solution → answer"", ""cost"": ""low"", ""llm_calls"": 1}" | |
| mathprompter,MathPrompter,Technique,,"{""input"": ""Query → algebraic expression with variables → Python function → multi-run validation"", ""output"": ""Verified numerical answer"", ""cost"": ""moderate"", ""llm_calls"": ""2 + N validations + code exec""}" | |
| thot,Thread of Thought (ThoT) Prompting,Technique,,"{""input"": ""Long chaotic context + query → segment analysis → answer"", ""output"": ""Section summaries → final answer"", ""cost"": ""moderate"", ""llm_calls"": 2}" | |
| not,Narrative-of-Thought (NoT) Prompting,Technique,,"{""input"": ""Query → temporally ordered narrative construction"", ""output"": ""Grounded narrative → answer"", ""cost"": ""low"", ""llm_calls"": 1}" | |
| iap,Instance-Adaptive Prompting (IAP),Technique,,"{""input"": ""Query → saliency/uncertainty scoring → adapted prompt"", ""output"": ""Instance-tuned reasoning → answer"", ""cost"": ""moderate"", ""llm_calls"": 2}" | |
| take-step-back,Take a Step Back Prompting,Technique,,"{""input"": ""Query → step-back question → abstract answer → grounded answer"", ""output"": ""High-level principle → detailed answer"", ""cost"": ""moderate"", ""llm_calls"": ""2-3 (may include retrieval)""}" | |
| analogical-reasoning,Analogical Reasoning,Technique,,"{""input"": ""Query → LLM generates similar solved examples → solves original"", ""output"": ""Self-generated analogues + solution → answer"", ""cost"": ""moderate"", ""llm_calls"": ""2-3""}" | |
| synthetic-prompting,Synthetic Prompting,Technique,,"{""input"": ""Backward (LLM generates Q from chain) + forward (chain for Q) → select by complexity"", ""output"": ""Augmented exemplars → CoT answer"", ""cost"": ""high"", ""llm_calls"": ""many (synthesis + selection)""}" | |
| maieutic,Maieutic Prompting,Technique,,"{""input"": ""Query → recursive hypothesis generation → tree of propositions → belief scoring"", ""output"": ""Contradiction-free answer"", ""cost"": ""very_high"", ""llm_calls"": ""recursive (many)""}" | |
| metacognitive,Metacognitive Prompting (MP),Technique,,"{""input"": ""Query → 5 stages: understand → judge → evaluate → decide → confidence"", ""output"": ""Multi-stage self-reflective answer + confidence"", ""cost"": ""moderate"", ""llm_calls"": ""1 (structured) or 5""}" | |
| decomp,Decomposed Prompting (DecomP),Technique,,"{""input"": ""Query → hierarchical/recursive decomposition → sub-LM solvers"", ""output"": ""Aggregated sub-answers → final answer"", ""cost"": ""high"", ""llm_calls"": ""1 + N sub-solvers + tools""}" | |
| thor,Three-Hop Reasoning (THOR),Technique,,"{""input"": ""Query → identify aspect → determine opinion → infer sentiment"", ""output"": ""Aspect → Opinion → Sentiment label"", ""cost"": ""moderate"", ""llm_calls"": 3}" | |
| coe,Chain-of-Event (CoE),Technique,,"{""input"": ""Document → event extraction → generalization → filtering → chronological integration"", ""output"": ""Concise event-based summary"", ""cost"": ""moderate"", ""llm_calls"": 4}" | |
| rag,Retrieval Augmented Generation (RAG),Technique,,"{""input"": ""Query → retriever fetches top-k docs → query + docs"", ""output"": ""Knowledge-grounded answer"", ""cost"": ""moderate"", ""llm_calls"": ""1 + retrieval"", ""requires_external"": ""retriever + knowledge base""}" | |
| react,ReAct Prompting,Technique,,"{""input"": ""Query → interleaved Thought / Action / Observation traces"", ""output"": ""Action results + reasoning → final answer"", ""cost"": ""high"", ""llm_calls"": ""iterative + tool calls"", ""requires_external"": ""tools / envs""}" | |
| cove,Chain-of-Verification (CoVe),Technique,,"{""input"": ""Query → draft response → verification Qs → verified revision"", ""output"": ""Hallucination-reduced answer"", ""cost"": ""high"", ""llm_calls"": ""4+ (generate, plan, verify, revise)""}" | |
| con,Chain-of-Note (CoN) Prompting,Technique,,"{""input"": ""Query + retrieved docs → systematic note-taking scoring relevance"", ""output"": ""Relevance-filtered grounded answer"", ""cost"": ""moderate"", ""llm_calls"": 2}" | |
| cok,Chain-of-Knowledge (CoK) Prompting,Technique,,"{""input"": ""Query → reasoning prep → dynamic knowledge adaptation from heterogeneous sources"", ""output"": ""Knowledge-corrected rationale → answer"", ""cost"": ""high"", ""llm_calls"": ""3+ (prep, adapt, consolidate)""}" | |
| implicit-rag,Implicit RAG,Technique,,"{""input"": ""Long context + query → LLM self-retrieves relevant sections"", ""output"": ""Self-retrieved sections → answer"", ""cost"": ""low"", ""llm_calls"": ""1-2""}" | |
| graph-rag,GraphRAG,Technique,,"{""input"": ""Query → KG-based retrieval + community summarization"", ""output"": ""KG-grounded answer"", ""cost"": ""high"", ""llm_calls"": ""1 + KG traversal""}" | |
| gnn-rag,GNN-RAG,Technique,,"{""input"": ""Query → GNN reasons over KG → LLM generates from subgraph"", ""output"": ""GNN-extracted answer candidates → NL answer"", ""cost"": ""high"", ""llm_calls"": ""1 + GNN inference""}" | |
| agentic-rag,Agentic RAG,Technique,,"{""input"": ""Query → autonomous agent with retrieval/tool access"", ""output"": ""Goal-directed, multi-step grounded answer"", ""cost"": ""very_high"", ""llm_calls"": ""iterative + tools""}" | |
| pot,Program of Thoughts (PoT) Prompting,Technique,,"{""input"": ""Query → LLM generates Python program expressing reasoning"", ""output"": ""Executed code output → numerical answer"", ""cost"": ""moderate"", ""llm_calls"": ""1 + code exec"", ""requires_external"": ""Python interpreter""}" | |
| coc,Chain of Code (CoC) Prompting,Technique,,"{""input"": ""Query → LLM generates code + LMulator simulates non-executable parts"", ""output"": ""Interleaved execution + simulation → answer"", ""cost"": ""moderate"", ""llm_calls"": ""2+ (gen + simulate)"", ""note"": ""84% on BIG-Bench Hard (+12%)""}" | |
| scot,Structured Chain-of-Thought (SCoT),Technique,,"{""input"": ""Query → program-structured reasoning (seq/branch/loop) → code"", ""output"": ""Structured pseudocode reasoning → generated source code"", ""cost"": ""low"", ""llm_calls"": 1}" | |
| pal,Program-Aided Language Models (PAL),Technique,,"{""input"": ""Query → LLM generates interleaved NL + Python code"", ""output"": ""Code execution → answer"", ""cost"": ""moderate"", ""llm_calls"": ""1 + code exec"", ""requires_external"": ""Python interpreter""}" | |
| code-prompting,Code Prompting,Technique,,"{""input"": ""Query → LLM generates code representation of conditional logic"", ""output"": ""Code-structured reasoning → answer"", ""cost"": ""low"", ""llm_calls"": 1}" | |
| scratchpad,Scratchpad Prompting,Technique,,"{""input"": ""Query + instruction to emit intermediate tokens on scratchpad"", ""output"": ""Scratchpad trace + final answer"", ""cost"": ""low"", ""llm_calls"": 1, ""note"": ""Fixed 512-token context window""}" | |
| binder,Binder,Technique,,"{""input"": ""Query → parsed to program binding LLM API as function → executed"", ""output"": ""Program output (neural-symbolic)"", ""cost"": ""moderate"", ""llm_calls"": ""1 + code exec""}" | |
| dater,DATER,Technique,,"{""input"": ""Large table → decompose to sub-tables → SQL sub-queries → few-shot reasoning"", ""output"": ""Sub-table + sub-query answers → final answer"", ""cost"": ""high"", ""llm_calls"": ""3+ + SQL exec""}" | |
| art,Automatic Reasoning and Tool-Use (ART),Technique,,"{""input"": ""Query → task library retrieval → structured program → tool execution"", ""output"": ""Multi-step tool-augmented answer"", ""cost"": ""high"", ""llm_calls"": ""1 + tools"", ""requires_external"": ""task library + tools""}" | |
| talm,TALM (Tool Augmented Language Models),Technique,,"{""input"": ""Query → LLM emits tool-call tokens → API called → result appended"", ""output"": ""Tool-augmented text"", ""cost"": ""moderate"", ""llm_calls"": ""1 + API calls""}" | |
| toolformer,Toolformer,Technique,,"{""input"": ""Query → self-supervised tool-call embedding in prompt"", ""output"": ""Tool-augmented response"", ""cost"": ""moderate"", ""llm_calls"": ""1 + tool calls""}" | |
| mrkl,MRKL System,Technique,,"{""input"": ""Query → routed to specialized module APIs → integrated result"", ""output"": ""Multi-module integrated answer"", ""cost"": ""high"", ""llm_calls"": ""1 router + K module calls""}" | |
| chameleon,Chameleon,Technique,,"{""input"": ""Query → LLM plans tool sequence → orchestrates execution"", ""output"": ""Multi-tool composed answer"", ""cost"": ""very_high"", ""llm_calls"": ""planner + tools""}" | |
| gorilla,Gorilla,Technique,,"{""input"": ""Query → LLM selects/generates correct API call from large catalog"", ""output"": ""Correct API invocation + result"", ""cost"": ""moderate"", ""llm_calls"": 1}" | |
| hugginggpt,HuggingGPT,Technique,,"{""input"": ""Query → task planning → model selection → execution → response integration"", ""output"": ""Multi-model orchestrated answer"", ""cost"": ""very_high"", ""llm_calls"": ""planner + N model calls""}" | |
| toolalpaca,ToolAlpaca,Technique,,"{""input"": ""Query → simulated tool-use environment → generalized tool interaction"", ""output"": ""Tool-assisted answer"", ""cost"": ""moderate"", ""llm_calls"": ""1 + tools""}" | |
| ape,Automatic Prompt Engineer (APE),Technique,,"{""input"": ""Input-output examples → LLM generates candidate prompts → scored → selected"", ""output"": ""Optimized prompt instruction"", ""cost"": ""very_high"", ""llm_calls"": ""many (generation + scoring + search)""}" | |
| opro,Optimization by Prompting (OPRO),Technique,,"{""input"": ""Meta-prompt with optimization trajectory (solution,score pairs) → LLM proposes new solutions"", ""output"": ""Optimized prompt / solution"", ""cost"": ""very_high"", ""llm_calls"": ""iterative (many rounds)""}" | |
| active-prompt,Active-Prompt,Technique,,"{""input"": ""Training set → k answers/question → uncertainty ranked → top-n human-annotated → few-shot"", ""output"": ""Uncertainty-informed CoT answer"", ""cost"": ""high"", ""llm_calls"": ""k × |train| + 1"", ""requires_external"": ""human annotators""}" | |
| self-refine,Self-Refine Prompting,Technique,,"{""input"": ""Query → draft → self-critique → refined draft (iterative loop)"", ""output"": ""Iteratively refined output"", ""cost"": ""high"", ""llm_calls"": ""2+ per iteration""}" | |
| self-contrast,Self-Contrast,Technique,,"{""input"": ""Query → generate multiple diverse perspectives → compare → reflect"", ""output"": ""Cross-perspective refined answer"", ""cost"": ""moderate"", ""llm_calls"": ""N perspectives + 1 reflection""}" | |
| critic,CRITIC Framework,Technique,,"{""input"": ""Query → draft → external tools evaluate → LLM refines → iterate"", ""output"": ""Externally validated refined answer"", ""cost"": ""high"", ""llm_calls"": ""iterative + tool calls""}" | |
| verify-and-edit,Verify-and-Edit (VE),Technique,,"{""input"": ""CoT chains → SC finds uncertain → edit chains with external knowledge → re-generate"", ""output"": ""Fact-corrected rationale → answer"", ""cost"": ""high"", ""llm_calls"": ""SC + edit + generation + retrieval""}" | |
| ensemble-refinement,Ensemble Refinement (ER),Technique,,"{""input"": ""CoT prompt → N generations → concatenate → M refinements → vote"", ""output"": ""Refined majority-voted answer"", ""cost"": ""high"", ""llm_calls"": ""N + M + voting""}" | |
| diverse,DiVeRSe,Technique,,"{""input"": ""Multiple diverse prompts → completions → step-aware verifier → vote"", ""output"": ""Step-verified voted answer"", ""cost"": ""very_high"", ""llm_calls"": ""K prompts × N samples + verifier""}" | |
| ama,Ask Me Anything (AMA),Technique,,"{""input"": ""Query → reformulated as multiple open-ended questions → answered → aggregated"", ""output"": ""Aggregated multi-perspective answer"", ""cost"": ""high"", ""llm_calls"": ""Q reformulations + Q answers + aggregation""}" | |
| prompt-boosting,PromptBoosting,Technique,,"{""input"": ""Prepared prompt set → AdaBoost-style iterative selection/weighting"", ""output"": ""Ensemble-boosted classification"", ""cost"": ""high"", ""llm_calls"": ""10+ per instance""}" | |
| detox-chain,Detox-Chain,Technique,,"{""input"": ""Toxic text → identify toxic elements → substitute → generate safe output"", ""output"": ""Detoxified prompt/response"", ""cost"": ""moderate"", ""llm_calls"": ""2-3""}" | |
| emotion-prompting,Emotion Prompting,Technique,,"{""input"": ""Query + emotional stimulus sentences appended"", ""output"": ""Emotionally enhanced response"", ""cost"": ""minimal"", ""llm_calls"": 1}" | |
| rar,Rephrase and Respond (RaR),Technique,,"{""input"": ""Query + instruction to rephrase before answering"", ""output"": ""Rephrased question + improved answer"", ""cost"": ""minimal"", ""llm_calls"": ""1 (single-turn) or 2""}" | |
| s2a,System 2 Attention (S2A) Prompting,Technique,,"{""input"": ""Query + context → LLM filters irrelevant context → answers from filtered"", ""output"": ""Attention-filtered answer"", ""cost"": ""moderate"", ""llm_calls"": 2}" | |
| chain-of-table,Chain of Table Prompting,Technique,,"{""input"": ""Table + query → iterative table operations (add_col, sort, filter) → answer"", ""output"": ""Evolved table representation → answer"", ""cost"": ""high"", ""llm_calls"": ""iterative (plan + execute per step)""}" | |
| instructed-prompting,Instructed Prompting,Technique,,"{""input"": ""Query + explicit instruction to ignore irrelevant info"", ""output"": ""Direct answer (distraction-free)"", ""cost"": ""minimal"", ""llm_calls"": 1}" | |
| fed-sp-dp,Federated SP/DP Self-Consistency/CoT,Technique,,"{""input"": ""Query paraphrased (same/diff params) → answered → SC/CoT federated"", ""output"": ""Majority-voted or hint-guided answer"", ""cost"": ""high"", ""llm_calls"": ""K paraphrases + K answers + voting""}" | |
| hyde,HyDE (Hypothetical Document Embeddings),Technique,,"{""input"": ""Query → LLM generates hypothetical doc → used as retrieval query"", ""output"": ""Better-aligned retrieved documents"", ""cost"": ""moderate"", ""llm_calls"": ""1 + retrieval""}" | |
| query2doc,Query2Doc,Technique,,"{""input"": ""Query → LLM generates pseudo-document → concat with query for retrieval"", ""output"": ""Expanded query → improved retrieval"", ""cost"": ""moderate"", ""llm_calls"": ""1 + retrieval""}" | |
| selfmem,Selfmem,Technique,,"{""input"": ""Iterative: memory pool of prior LLM outputs → selector picks best → next round"", ""output"": ""Progressively improved generation"", ""cost"": ""high"", ""llm_calls"": ""iterative""}" | |
| meta-rag,MetaRAG,Technique,,"{""input"": ""Retrieved info + metacognitive monitoring/evaluation/planning"", ""output"": ""Metacognitively refined answer"", ""cost"": ""high"", ""llm_calls"": ""3+ (monitor + evaluate + plan)""}" | |
| minds-eye,Mind's Eye,Technique,,"{""input"": ""Physical question → text-to-code → MuJoCo simulation → NL answer"", ""output"": ""Physically grounded answer"", ""cost"": ""high"", ""llm_calls"": ""2 + physics sim"", ""requires_external"": ""MuJoCo physics engine""}" | |
| multibot,MultiPoT,Technique,,"{""input"": ""Query → solutions in multiple PLs → SC selects best"", ""output"": ""Best cross-PL executed answer"", ""cost"": ""high"", ""llm_calls"": ""K languages + code exec + voting""}" | |
| hag,HAG (Hyperparameter-Aware Generation),Technique,,"{""input"": ""Query → LLM generates optimal decoding hyperparameters → generates with them"", ""output"": ""Self-configured optimally decoded response"", ""cost"": ""moderate"", ""llm_calls"": 2}" | |
| reflexion,Reflexion,Technique,,"{""input"": ""Query → actor generates trajectory → evaluator scores → reflector critiques → memory updated → retry"", ""output"": ""Iteratively improved action trace → answer"", ""cost"": ""very_high"", ""llm_calls"": ""3+ per iteration (actor + evaluator + reflector)"", ""requires_external"": ""environment / evaluator""}" | |
| self-discover,Self-Discover,Technique,,"{""input"": ""Task → select relevant reasoning modules → adapt to task → compose into JSON reasoning structure → solve"", ""output"": ""Self-composed reasoning plan → structured solution"", ""cost"": ""high"", ""llm_calls"": ""3+ (select → adapt → implement)"", ""note"": ""Up to +42% over CoT, 10-40x less compute than CoT-SC""}" | |
| skeleton-of-thought,Skeleton-of-Thought (SoT),Technique,,"{""input"": ""Query → generate skeleton outline → parallel expansion of each point → assembly"", ""output"": ""Structured, parallel-generated response"", ""cost"": ""moderate"", ""llm_calls"": ""1 + N parallel expansions"", ""note"": ""ICLR 2024, 2x+ speedup, 60% quality maintained/improved""}" | |
| cumulative-reasoning,Cumulative Reasoning,Technique,,"{""input"": ""Query → proposer suggests steps → verifier validates → accumulate → reporter synthesizes"", ""output"": ""DAG of verified propositions → final answer"", ""cost"": ""very_high"", ""llm_calls"": ""many (propose → verify → report cycles)""}" | |
| multi-agent-debate,Multi-Agent Debate (MAD),Technique,,"{""input"": ""Query → N agents debate iteratively → judge selects best answer"", ""output"": ""Consensus answer from multi-agent deliberation"", ""cost"": ""very_high"", ""llm_calls"": ""N agents × K rounds + judge""}" | |
| alg-majority-voting,Majority Voting / Answer Aggregation,AlgorithmicComponent,Selecting the most frequent/consistent answer across paths,"{""subtype"": ""voting""}" | |
| alg-tree-search,Tree Search (BFS/DFS),AlgorithmicComponent,Exploring branching paths through a tree of partial solutions,"{""subtype"": ""search""}" | |
| alg-backtracking,Backtracking,AlgorithmicComponent,Reverting to a previous state when a path is unpromising,"{""subtype"": ""search""}" | |
| alg-code-execution,External Code Execution,AlgorithmicComponent,"Running generated code via an interpreter for computation (Python, SQL)","{""subtype"": ""execution""}" | |
| alg-retrieval,External Knowledge Retrieval,AlgorithmicComponent,Fetching documents/passages from external knowledge sources,"{""subtype"": ""retrieval""}" | |
| alg-tool-use,External Tool Integration,AlgorithmicComponent,"Calling external tools (search, calculators, APIs) during reasoning","{""subtype"": ""execution""}" | |
| alg-clustering,Clustering (K-Means / Sentence-BERT),AlgorithmicComponent,Grouping questions/examples by similarity for representative sampling,"{""subtype"": ""analysis""}" | |
| alg-multiple-sampling,Multiple Path Sampling,AlgorithmicComponent,Generating multiple distinct reasoning chains via temperature/nucleus sampling,"{""subtype"": ""sampling""}" | |
| alg-uncertainty-estimation,Uncertainty Estimation,AlgorithmicComponent,Measuring model uncertainty to select examples or prompts,"{""subtype"": ""analysis""}" | |
| alg-auto-prompt-gen,Automated Prompt Generation / Optimization,AlgorithmicComponent,"Automatically generating, scoring, and selecting candidate prompts","{""subtype"": ""optimization""}" | |
| alg-kg-reasoning,Knowledge Graph / Subgraph Reasoning,AlgorithmicComponent,Reasoning over knowledge graph structures for retrieval or inference,"{""subtype"": ""retrieval""}" | |
| alg-thought-evaluation,Thought / State Evaluation,AlgorithmicComponent,Assessing quality or promise of intermediate reasoning states (heuristic or LLM),"{""subtype"": ""analysis""}" | |
| alg-boosting,AdaBoost-Style Prompt Boosting,AlgorithmicComponent,Iteratively selecting/weighting prompts using boosting algorithms,"{""subtype"": ""optimization""}" | |
| alg-physics-simulation,Physics Engine Simulation,AlgorithmicComponent,"Running physical simulations (e.g., MuJoCo) to ground reasoning","{""subtype"": ""execution""}" | |
| alg-lmulator,LMulator (LM-Emulated Execution),AlgorithmicComponent,Using the LLM itself to simulate code execution for non-executable parts,"{""subtype"": ""execution""}" | |
| pc-intermediate-reasoning,Intermediate Reasoning Steps,PromptComponent,Generating explicit step-by-step reasoning in natural language,"{""subtype"": ""formatting""}" | |
| pc-symbolic-representation,Symbolic / Formal Representation,PromptComponent,"Replacing NL with condensed symbols, formal logic, or algebraic variables","{""subtype"": ""formatting""}" | |
| pc-contrastive-examples,Contrastive Examples (Positive + Negative),PromptComponent,Providing both correct and incorrect reasoning demonstrations,"{""subtype"": ""examples""}" | |
| pc-emotional-stimulus,Emotional Stimulus / Affective Cues,PromptComponent,Appending emotionally charged phrases to enhance performance,"{""subtype"": ""stimulus""}" | |
| pc-question-rephrasing,Question Rephrasing / Clarification,PromptComponent,Having the LLM rephrase/expand a question for better comprehension,"{""subtype"": ""formatting""}" | |
| pc-concise-reasoning,Concise / Draft Reasoning (Token Minimization),PromptComponent,"Generating minimal, information-dense reasoning steps","{""subtype"": ""constraint""}" | |
| pc-analogical-examples,Self-Generated Analogical Examples,PromptComponent,LLM generates similar solved examples before tackling the target,"{""subtype"": ""examples""}" | |
| pc-complexity-selection,Complexity-Based Example Selection,PromptComponent,Selecting exemplars with more reasoning steps as demonstrations,"{""subtype"": ""examples""}" | |
| pc-role-assignment,Role / Persona Assignment,PromptComponent,Assigning a specific role or persona to guide LLM behaviour,"{""subtype"": ""instruction""}" | |
| pc-task-instruction,Explicit Task Instruction,PromptComponent,"Clear specification of intent, domain, and output format","{""subtype"": ""instruction""}" | |
| pc-propositional-logic,Propositional Logic Extraction,PromptComponent,Extracting logical propositions and applying formal logic laws,"{""subtype"": ""formatting""}" | |
| pc-narrative-construction,Narrative Construction (Temporal Ordering),PromptComponent,Building temporally grounded narratives to structure reasoning,"{""subtype"": ""formatting""}" | |
| pc-code-generation,Code Generation (Prompt Pattern),PromptComponent,Expressing reasoning as executable code / pseudocode within the prompt,"{""subtype"": ""formatting""}" | |
| pc-template-reuse,Thought-Template Reuse,PromptComponent,Retrieving and instantiating high-level reusable reasoning patterns,"{""subtype"": ""examples""}" | |
| pc-context-filtering,Context Regeneration / Irrelevant Info Filtering,PromptComponent,Rewriting the input context to remove distracting information,"{""subtype"": ""constraint""}" | |
| pc-role-assignment,Role Assignment / Persona,PromptComponent,Assigning a specific role or persona to the LLM to guide its behavior,"{""subtype"": ""instruction""}" | |
| pc-task-instruction,Explicit Task Instruction,PromptComponent,"Clear, explicit instruction describing the task to be performed","{""subtype"": ""instruction""}" | |
| pc-format-constraints,Output Format Constraints,PromptComponent,"Specifying structured output format (outline, bullet points, JSON)","{""subtype"": ""formatting""}" | |
| pc-algebraic-abstraction,Algebraic Abstraction + Multi-Run Validation,PromptComponent,"Replacing numerics with variables, solving symbolically, validating with multiple inputs","{""subtype"": ""formatting""}" | |
| df-decomposition,Problem Decomposition,DataFlow,Breaking a complex problem into smaller sub-problems,"{""subtype"": ""decomposition""}" | |
| df-sequential-chaining,Sequential Sub-Problem Chaining,DataFlow,Solving sub-problems in order where each answer feeds the next,"{""subtype"": ""chaining""}" | |
| df-verification,Verification / Self-Verification,DataFlow,Checking correctness of generated reasoning or answers,"{""subtype"": ""verification""}" | |
| df-self-reflection,Self-Reflection / Self-Critique,DataFlow,Model critiques its own output to identify flaws,"{""subtype"": ""feedback""}" | |
| df-iterative-refinement,Iterative Refinement / Feedback Loop,DataFlow,Repeatedly improving output through generate-critique-revise cycles,"{""subtype"": ""feedback""}" | |
| df-context-augmentation,Context Augmentation,DataFlow,Enriching the prompt with retrieved / external information before generation,"{""subtype"": ""aggregation""}" | |
| df-action-observation,Action-Observation Loop,DataFlow,Interleaving LLM reasoning actions with environment observations,"{""subtype"": ""chaining""}" | |
| df-document-scoring,Document Relevance Scoring / Note-Taking,DataFlow,Systematically evaluating relevance of retrieved documents,"{""subtype"": ""filtering""}" | |
| df-knowledge-adaptation,Knowledge Domain ID + Dynamic Adaptation,DataFlow,Identifying relevant knowledge domains and adapting reasoning from them,"{""subtype"": ""aggregation""}" | |
| df-graph-reasoning,Graph-Based Reasoning Flow,DataFlow,Modeling reasoning as a directed graph with merging/splitting paths,"{""subtype"": ""chaining""}" | |
| df-tabular-reasoning,Tabular Reasoning / Table Operations,DataFlow,"Operating on structured tables (add columns, sort, filter, group_by)","{""subtype"": ""chaining""}" | |
| df-external-memory,External Memory / Scratchpad,DataFlow,Using an external workspace to store and retrieve intermediate results,"{""subtype"": ""storage""}" | |
| df-reductio-ad-absurdum,Reductio ad Absurdum Verification,DataFlow,Verifying a step by checking if its negation leads to contradiction,"{""subtype"": ""verification""}" | |
| df-event-extraction,Event Extraction + Generalization + Filtering,DataFlow,"Extracting events, generalizing, chronologically filtering for coverage","{""subtype"": ""filtering""}" | |
| df-aspect-opinion-sentiment,Aspect → Opinion → Sentiment Chain,DataFlow,"Multi-hop reasoning through aspect identification, opinion determination, sentiment inference","{""subtype"": ""chaining""}" | |
| df-reasoning-harmonization,Reasoning Unification / Harmonization,DataFlow,Iteratively refining diverse rationales into a unified coherent pattern,"{""subtype"": ""aggregation""}" | |
| df-hierarchical-filtering,Hierarchical Constraint Filtering,DataFlow,Applying layered constraints (hard → soft) to filter candidates,"{""subtype"": ""filtering""}" | |
| df-dag-extraction,DAG Path Extraction,DataFlow,Extracting backbone paths between endpoints in directed acyclic graphs,"{""subtype"": ""chaining""}" | |
| df-human-annotation,Human Annotation / Active Learning Loop,DataFlow,Humans annotate most uncertain examples for improved prompting,"{""subtype"": ""feedback""}" | |
| df-query-paraphrasing,Paraphrasing / Crowd-Sourced Query Variants,DataFlow,Using synonymous reformulations of the query for diversity,"{""subtype"": ""decomposition""}" | |
| df-demo-library,Cross-Task Demonstration Library,DataFlow,Maintaining a library of demos from related tasks for transfer,"{""subtype"": ""storage""}" | |
| df-representative-selection,Representative Example Selection,DataFlow,Choosing representative examples (from clusters or pools) as demonstrations,"{""subtype"": ""filtering""}" | |
| df-saliency-analysis,Saliency / Attention Flow Analysis,DataFlow,Analyzing information flow through attention layers to select prompts,"{""subtype"": ""filtering""}" | |
| df-parallel-decomposition,Parallel Decomposition,DataFlow,Decomposing a problem into independent parts and processing them in parallel,"{""subtype"": ""decomposition""}" | |
| df-feedback-loop,Feedback Loop,DataFlow,"Iterative cycle of generation, evaluation, and refinement using self-generated feedback","{""subtype"": ""feedback""}" | |
| cap-arithmetic,Arithmetic Computation,CognitiveCapability,"Performing numerical calculations, basic math operations", | |
| cap-symbolic,Symbolic Reasoning,CognitiveCapability,"Manipulating abstract symbols, variables, and formal expressions", | |
| cap-spatial,Spatial Reasoning,CognitiveCapability,"Understanding physical/spatial relationships, navigation, object positioning", | |
| cap-temporal,Temporal Reasoning,CognitiveCapability,"Understanding time sequences, temporal constraints, chronological ordering", | |
| cap-deductive,Deductive Logic,CognitiveCapability,Drawing necessary conclusions from premises via logical rules, | |
| cap-inductive,Inductive Reasoning,CognitiveCapability,Generalizing from specific examples or observations, | |
| cap-causal,Causal Inference,CognitiveCapability,Understanding cause-and-effect relationships and mechanisms, | |
| cap-analogical,Analogical Thinking,CognitiveCapability,"Reasoning by analogy, transferring knowledge from similar domains", | |
| cap-planning,Multi-Step Planning,CognitiveCapability,Formulating and executing multi-step goal-directed strategies, | |
| cap-abstraction,Abstraction,CognitiveCapability,Extracting high-level concepts or principles from specific instances, | |
| cap-self-monitoring,Self-Monitoring / Metacognition,CognitiveCapability,"Evaluating own reasoning quality, detecting errors or uncertainties", | |
| cap-compositional,Compositional Generalization,CognitiveCapability,Combining learned patterns in novel ways to solve new problems, | |
| cap-multi-hop,Multi-Hop Inference,CognitiveCapability,Chaining multiple reasoning steps over different information sources, | |
| cap-retrieval-integration,Knowledge Integration,CognitiveCapability,Incorporating and synthesizing information from external knowledge sources, | |
| cap-code-synthesis,Code Synthesis,CognitiveCapability,Generating executable code from natural language specifications, | |
| cap-creative-generation,Creative Generation,CognitiveCapability,"Producing novel, creative text: stories, poetry, open-ended writing", | |
| cap-emotional-understanding,Emotional Understanding,CognitiveCapability,Comprehending and responding to emotional/affective cues, | |
| math-word-problems,"Math Word Problems (GSM8K, SVAMP)",Task,, | |
| math-algebraic,"Algebraic Math (AQuA, MATH)",Task,, | |
| math-tabular,"Tabular Math (FinQA, TabMWP)",Task,, | |
| math-search-based,Search-based Math (Game of 24),Task,, | |
| logical-reasoning,Logical Reasoning,Task,, | |
| commonsense-reasoning,Commonsense Reasoning,Task,, | |
| multi-hop-reasoning,Multi-Hop Reasoning,Task,, | |
| cf-qa,Context-Free Question Answering,Task,, | |
| contextual-qa,Contextual Question Answering,Task,, | |
| conversational-qa,Conversational Contextual QA,Task,, | |
| spatial-qa,Spatial Question Answering,Task,, | |
| table-math,Table-Based Math Problem Solving,Task,, | |
| table-qa,Table-Based Question Answering,Task,, | |
| table-truth,Table-Based Truthfulness,Task,, | |
| truthfulness,Truthfulness / Factual Accuracy,Task,, | |
| free-response,Free Response Generation,Task,, | |
| code-gen,Code Generation,Task,, | |
| dialogue,Dialogue Systems,Task,, | |
| social-reasoning,Social Reasoning,Task,, | |
| causal-reasoning,Causal Reasoning,Task,, | |
| text-classification,Multilabel Text Classification,Task,, | |
| task-completion,Language-Based Task Completion,Task,, | |
| relation-extraction,Relation Extraction,Task,, | |
| nli,Natural Language Inference,Task,, | |
| stance-detection,Stance Detection,Task,, | |
| paraphrasing,Paraphrasing,Task,, | |
| summarization,Summarization,Task,, | |
| wsd,Word Sense Disambiguation,Task,, | |
| ner,Named Entity Recognition,Task,, | |
| machine-translation,Machine Translation,Task,, | |
| sentiment,Emotion / Sentiment Understanding,Task,, | |
| recommender,Recommender Systems,Task,, | |
| physical-reasoning,Physical Reasoning / Simulation,Task,, | |
| hallucination-reduction,Hallucination Reduction,Task,, | |
| br-cot-gsm8k,cot on GSM8K,BenchmarkResult,,"{""technique"": ""cot"", ""model"": ""palm-540b"", ""dataset"": ""GSM8K"", ""metric"": ""accuracy"", ""value"": ""90.2%"", ""paper"": ""sahoo-2025""}" | |
| br-sc-gsm8k,self-consistency on GSM8K,BenchmarkResult,,"{""technique"": ""self-consistency"", ""model"": ""palm-540b"", ""dataset"": ""GSM8K"", ""metric"": ""accuracy_improvement"", ""value"": ""+17.9%"", ""paper"": ""sahoo-2025""}" | |
| br-tot-game24,tot on Game of 24,BenchmarkResult,,"{""technique"": ""tot"", ""model"": ""gpt-4"", ""dataset"": ""Game of 24"", ""metric"": ""success_rate"", ""value"": ""74%"", ""comparison"": ""CoT: 4%"", ""paper"": ""sahoo-2025""}" | |
| br-cod-tokens,cod on Various,BenchmarkResult,,"{""technique"": ""cod"", ""model"": ""gpt-4o"", ""dataset"": ""Various"", ""metric"": ""token_reduction"", ""value"": ""80%"", ""paper"": ""sahoo-2025""}" | |
| br-bot-cost,bot on Various,BenchmarkResult,,"{""technique"": ""bot"", ""model"": ""llama-3"", ""dataset"": ""Various"", ""metric"": ""cost_vs_tot"", ""value"": ""12% of ToT cost"", ""paper"": ""sahoo-2025""}" | |
| br-con-noisy,con on Open-domain QA (noisy),BenchmarkResult,,"{""technique"": ""con"", ""model"": ""llama-2"", ""dataset"": ""Open-domain QA (noisy)"", ""metric"": ""EM_improvement"", ""value"": ""+7.9"", ""paper"": ""sahoo-2025""}" | |
| br-emotion-bigbench,emotion-prompting on BIG-Bench,BenchmarkResult,,"{""technique"": ""emotion-prompting"", ""model"": ""gpt-4"", ""dataset"": ""BIG-Bench"", ""metric"": ""accuracy_improvement"", ""value"": ""+115%"", ""paper"": ""sahoo-2025""}" | |
| br-emotion-induction,emotion-prompting on Instruction Induction,BenchmarkResult,,"{""technique"": ""emotion-prompting"", ""model"": ""gpt-4"", ""dataset"": ""Instruction Induction"", ""metric"": ""accuracy_improvement"", ""value"": ""+8%"", ""paper"": ""sahoo-2025""}" | |
| br-pot-math,pot on Math word problems,BenchmarkResult,,"{""technique"": ""pot"", ""model"": ""codex"", ""dataset"": ""Math word problems"", ""metric"": ""accuracy_improvement"", ""value"": ""+12% avg over CoT"", ""paper"": ""sahoo-2025""}" | |
| br-scot-humaneval,scot on HumanEval/MBPP/MBCPP,BenchmarkResult,,"{""technique"": ""scot"", ""model"": ""chatgpt"", ""dataset"": ""HumanEval/MBPP/MBCPP"", ""metric"": ""pass@k_improvement"", ""value"": ""+13.79%"", ""paper"": ""sahoo-2025""}" | |
| br-coc-bbh,coc on BIG-Bench Hard,BenchmarkResult,,"{""technique"": ""coc"", ""model"": ""gpt-3.5"", ""dataset"": ""BIG-Bench Hard"", ""metric"": ""accuracy"", ""value"": ""84%"", ""comparison"": ""+12% over CoT"", ""paper"": ""sahoo-2025""}" | |
| br-opro-gsm8k,opro on GSM8K,BenchmarkResult,,"{""technique"": ""opro"", ""model"": ""palm-2"", ""dataset"": ""GSM8K"", ""metric"": ""accuracy_improvement"", ""value"": ""+8% over human prompts"", ""paper"": ""sahoo-2025""}" | |
| br-stepback-timeqa,take-step-back on TimeQA,BenchmarkResult,,"{""technique"": ""take-step-back"", ""model"": ""palm-2"", ""dataset"": ""TimeQA"", ""metric"": ""accuracy_improvement"", ""value"": ""+27%"", ""paper"": ""sahoo-2025""}" | |
| br-stepback-musique,take-step-back on MuSiQue,BenchmarkResult,,"{""technique"": ""take-step-back"", ""model"": ""palm-2"", ""dataset"": ""MuSiQue"", ""metric"": ""accuracy_improvement"", ""value"": ""+7%"", ""paper"": ""sahoo-2025""}" | |
| br-active-reasoning,active-prompt on 8 complex reasoning tasks,BenchmarkResult,,"{""technique"": ""active-prompt"", ""model"": ""gpt-3"", ""dataset"": ""8 complex reasoning tasks"", ""metric"": ""accuracy_improvement"", ""value"": ""+7.0% over SC"", ""paper"": ""sahoo-2025""}" | |
| br-scratchpad-mbpp,scratchpad on MBPP-aug,BenchmarkResult,,"{""technique"": ""scratchpad"", ""model"": ""gpt-3"", ""dataset"": ""MBPP-aug"", ""metric"": ""success_rate"", ""value"": ""46.8%"", ""paper"": ""sahoo-2025""}" | |
| br-s2a-triviaqa,s2a on Modified TriviaQA,BenchmarkResult,,"{""technique"": ""s2a"", ""model"": ""llama-2"", ""dataset"": ""Modified TriviaQA"", ""metric"": ""accuracy"", ""value"": ""80.3%"", ""comparison"": ""Baseline: 62.8%"", ""paper"": ""debnath-2025""}" | |
| br-s2a-gsmic,s2a on GSM-IC,BenchmarkResult,,"{""technique"": ""s2a"", ""model"": ""llama-2"", ""dataset"": ""GSM-IC"", ""metric"": ""accuracy"", ""value"": ""61.3%"", ""comparison"": ""Baseline: 51.7%"", ""paper"": ""debnath-2025""}" | |
| br-ccot-gsm8k,ccot on GSM-8K,BenchmarkResult,,"{""technique"": ""ccot"", ""model"": ""gpt-3.5"", ""dataset"": ""GSM-8K"", ""metric"": ""accuracy_improvement"", ""value"": ""+9.8 points over CoT"", ""paper"": ""debnath-2025""}" | |
| br-ccot-bamboogle,ccot on Bamboogle,BenchmarkResult,,"{""technique"": ""ccot"", ""model"": ""gpt-3.5"", ""dataset"": ""Bamboogle"", ""metric"": ""accuracy_improvement"", ""value"": ""+16.0 points over CoT"", ""paper"": ""debnath-2025""}" | |
| br-query2doc-msmarco,query2doc on MSMARCO,BenchmarkResult,,"{""technique"": ""query2doc"", ""model"": ""gpt-3.5"", ""dataset"": ""MSMARCO"", ""metric"": ""relative_gain"", ""value"": ""+3%"", ""paper"": ""liu-2025""}" | |
| br-query2doc-trecdl,query2doc on TREC DL,BenchmarkResult,,"{""technique"": ""query2doc"", ""model"": ""gpt-3.5"", ""dataset"": ""TREC DL"", ""metric"": ""relative_gain"", ""value"": ""+15%"", ""paper"": ""liu-2025""}" | |
| br-metarag-2wiki,meta-rag on 2WikiMultihopQA,BenchmarkResult,,"{""technique"": ""meta-rag"", ""model"": ""gpt-4"", ""dataset"": ""2WikiMultihopQA"", ""metric"": ""accuracy_improvement"", ""value"": ""+34.6%"", ""paper"": ""liu-2025""}" | |
| br-metarag-hotpot,meta-rag on HotpotQA,BenchmarkResult,,"{""technique"": ""meta-rag"", ""model"": ""gpt-4"", ""dataset"": ""HotpotQA"", ""metric"": ""accuracy_improvement"", ""value"": ""+26%"", ""paper"": ""liu-2025""}" | |
| br-selfrefine-various,self-refine on Various,BenchmarkResult,,"{""technique"": ""self-refine"", ""model"": ""gpt-3.5"", ""dataset"": ""Various"", ""metric"": ""absolute_improvement"", ""value"": ""5–40%"", ""paper"": ""liu-2025""}" | |
| br-selfrefine-code,self-refine on Code generation,BenchmarkResult,,"{""technique"": ""self-refine"", ""model"": ""codex"", ""dataset"": ""Code generation"", ""metric"": ""improvement"", ""value"": ""+13%"", ""paper"": ""liu-2025""}" | |
| br-cot-math-vatsal,cot on Math Problem Solving pool,BenchmarkResult,,"{""technique"": ""cot"", ""model"": ""gpt-4"", ""dataset"": ""Math Problem Solving pool"", ""metric"": ""improvement_over_basic"", ""value"": ""+39%"", ""paper"": ""vatsal-2024""}" | |
| br-cot-commonsense-vatsal,cot on Commonsense Reasoning pool,BenchmarkResult,,"{""technique"": ""cot"", ""model"": ""gpt-4"", ""dataset"": ""Commonsense Reasoning pool"", ""metric"": ""improvement_over_basic"", ""value"": ""+26%"", ""paper"": ""vatsal-2024""}" | |
| br-sc-math-vatsal,self-consistency on Math Problem Solving pool,BenchmarkResult,,"{""technique"": ""self-consistency"", ""model"": ""gpt-4"", ""dataset"": ""Math Problem Solving pool"", ""metric"": ""improvement_over_cot"", ""value"": ""+11%"", ""paper"": ""vatsal-2024""}" | |
| br-cos-spatial-vatsal,cos on Spatial QA (Brick World etc.),BenchmarkResult,,"{""technique"": ""cos"", ""model"": ""chatgpt"", ""dataset"": ""Spatial QA (Brick World etc.)"", ""metric"": ""accuracy_improvement"", ""value"": ""+60.8%"", ""paper"": ""vatsal-2024""}" | |
| br-tot-game24-vatsal,tot on Game of 24,BenchmarkResult,,"{""technique"": ""tot"", ""model"": ""gpt-4"", ""dataset"": ""Game of 24"", ""metric"": ""success_rate_improvement"", ""value"": ""+65% over CoT"", ""paper"": ""vatsal-2024""}" | |
| br-lot-commonsense,logicot on Commonsense Reasoning,BenchmarkResult,,"{""technique"": ""logicot"", ""model"": ""gpt-4"", ""dataset"": ""Commonsense Reasoning"", ""metric"": ""improvement_over_cot"", ""value"": ""+16.2%"", ""paper"": ""vatsal-2024""}" | |
| br-lot-causal,logicot on Causal Reasoning,BenchmarkResult,,"{""technique"": ""logicot"", ""model"": ""gpt-4"", ""dataset"": ""Causal Reasoning"", ""metric"": ""improvement_over_cot"", ""value"": ""+15.8%"", ""paper"": ""vatsal-2024""}" | |
| br-maieutic-commonsense,maieutic on CSQA 2.0 / Com2Sense / Creak,BenchmarkResult,,"{""technique"": ""maieutic"", ""model"": ""gpt-3"", ""dataset"": ""CSQA 2.0 / Com2Sense / Creak"", ""metric"": ""improvement_over_cot"", ""value"": ""+20%"", ""paper"": ""vatsal-2024""}" | |
| br-decomp-commonsense,decomp on Commonsense (EM),BenchmarkResult,,"{""technique"": ""decomp"", ""model"": ""gpt-3"", ""dataset"": ""Commonsense (EM)"", ""metric"": ""improvement_over_cot"", ""value"": ""+25% avg"", ""paper"": ""vatsal-2024""}" | |
| br-cove-qa,cove on CF-QA / Contextual-QA / Free Response,BenchmarkResult,,"{""technique"": ""cove"", ""model"": ""llama-2"", ""dataset"": ""CF-QA / Contextual-QA / Free Response"", ""metric"": ""improvement_over_cot"", ""value"": ""+10%"", ""paper"": ""vatsal-2024""}" | |
| br-react-taskcompletion,react on ALFWorld / WebShop,BenchmarkResult,,"{""technique"": ""react"", ""model"": ""palm-540b"", ""dataset"": ""ALFWorld / WebShop"", ""metric"": ""improvement_over_rl"", ""value"": ""+10%"", ""paper"": ""vatsal-2024""}" | |
| br-mathprompter-math,mathprompter on Math dataset,BenchmarkResult,,"{""technique"": ""mathprompter"", ""model"": ""gpt-3.5"", ""dataset"": ""Math dataset"", ""metric"": ""accuracy"", ""value"": ""92.5% (from 78.7%)"", ""paper"": ""vatsal-2024""}" | |
| br-mp-various,"metacognitive on 10 NLP tasks (BoolQ, WiC, etc.)",BenchmarkResult,,"{""technique"": ""metacognitive"", ""model"": ""gpt-4"", ""dataset"": ""10 NLP tasks (BoolQ, WiC, etc.)"", ""metric"": ""comparison"", ""value"": ""Beats CoT and PS consistently"", ""paper"": ""vatsal-2024""}" | |
| br-synthetic-multi,synthetic-prompting on Various reasoning,BenchmarkResult,,"{""technique"": ""synthetic-prompting"", ""model"": ""gpt-3.5"", ""dataset"": ""Various reasoning"", ""metric"": ""absolute_improvement"", ""value"": ""up to +15.6%"", ""paper"": ""vatsal-2024""}" | |
| br-instructed-truth,instructed-prompting on Truthfulness benchmarks,BenchmarkResult,,"{""technique"": ""instructed-prompting"", ""model"": ""gpt-4"", ""dataset"": ""Truthfulness benchmarks"", ""metric"": ""normalized_micro_accuracy"", ""value"": ""88.2"", ""paper"": ""vatsal-2024""}" | |