DSPy compiles short declarative programs into LM pipelines that self-optimize and outperform both standard few-shot prompting and expert-written chains on math, retrieval, and QA tasks.
super hub Mixed citations
Self-Refine: Iterative Refinement with Self-Feedback
Mixed citation behavior. Most common role is background (68%).
148
Pith papers citing it
Background
68% of classified citations
abstract
Like humans, large language models (LLMs) do not always generate the best output on their first try. Motivated by how humans refine their written text, we introduce Self-Refine, an approach for improving initial outputs from LLMs through iterative feedback and refinement. The main idea is to generate an initial output using an LLMs; then, the same LLMs provides feedback for its output and uses it to refine itself, iteratively. Self-Refine does not require any supervised training data, additional training, or reinforcement learning, and instead uses a single LLM as the generator, refiner, and feedback provider. We evaluate Self-Refine across 7 diverse tasks, ranging from dialog response generation to mathematical reasoning, using state-of-the-art (GPT-3.5, ChatGPT, and GPT-4) LLMs. Across all evaluated tasks, outputs generated with Self-Refine are preferred by humans and automatic metrics over those generated with the same LLM using conventional one-step generation, improving by ~20% absolute on average in task performance. Our work demonstrates that even state-of-the-art LLMs like GPT-4 can be further improved at test time using our simple, standalone approach.
hub tools
citation-role summary
background 23
baseline 2
method 2
dataset 1
citation-polarity summary
claims ledger
- abstract Like humans, large language models (LLMs) do not always generate the best output on their first try. Motivated by how humans refine their written text, we introduce Self-Refine, an approach for improving initial outputs from LLMs through iterative feedback and refinement. The main idea is to generate an initial output using an LLMs; then, the same LLMs provides feedback for its output and uses it to refine itself, iteratively. Self-Refine does not require any supervised training data, additional training, or reinforcement learning, and instead uses a single LLM as the generator, refiner, and f
authors
co-cited works
representative citing papers
Promptbreeder evolves both task prompts and the mutation prompts that improve them using LLMs, outperforming Chain-of-Thought and Plan-and-Solve on arithmetic and commonsense reasoning benchmarks.
Empirical study on production-scale clinical NLP shows direct learning from verifier rejections fails due to sparse data while fixed ontology and evidence-support filters succeed, with selectivity determined by matching verifier evidence.
Preregistered placebo-controlled decomposition shows external executable counterexamples drive self-repair gains in small code models more than re-exposure or self-critique.
AlgoBench creates traceable variants of competitive programming problems via constraint shifts that invalidate original algorithms, paired with complexity metrics that reveal LLMs often produce functionally correct but asymptotically unsuitable solutions.
Models delayed verification in multi-agent LLMs as graph consensus, derives stability thresholds (inverse golden ratio for delay two) via grounded Laplacian, and gives a supermodular greedy rule for corrector placement; experiments on five models confirm dose-delay oscillations.
LLM judges exhibit high stability under neutral re-evaluation but substantial reversibility under targeted post-decision challenges, quantified via a new Evaluation Robustness Score (ERS).
An automatic numeric-remapping attack generator reveals 12-26 point accuracy drops on GSM8K for three LLMs while MAWPS and MultiArith stay near 98%.
Six multi-agent architectures for LLM code generation on HumanEval form two complexity clusters separated by 50-130%, with no accuracy advantage for the complex cluster.
HIDBench unifies DARPA-E3, DARPA-E5, and NodLink datasets with a data pipeline to benchmark LLMs for host-based intrusion detection, showing high precision on simple logs but sharp drops in MCC and rises in false positives on complex noisy data.
Proposes Formal Skill as a programmable runtime abstraction for LLM agents, implemented in open-source FairyClaw, achieving competitive Harness-Bench scores with substantially fewer tokens.
DecisionBench supplies a fixed task suite, model pool, delegation interface, and multi-axis metrics to evaluate emergent delegation, showing similar quality across awareness conditions but 15-31 point headroom under perfect delegation.
Mismatched wrong drafts from Qwen2.5-Math-1.5B improve Mathstral-7B GRPO training, reaching 71.98% greedy pass@1 on MATH-500 and lifting AIME 2025/2026 pass@k over baselines and other draft variants.
Test-Time Hinting trains a hint generator to prepend contextual guidance to VLM prompts, improving accuracy on natural-image VQA benchmarks with generalization to unseen tasks and models.
Fast-Slow Training uses context optimization as fast weights alongside parameter updates as slow weights to achieve up to 3x better sample efficiency, higher performance, and less catastrophic forgetting than standard RL in continual LLM learning.
In 30-step recursive LLM loops, append-mode persistent escape from source basins reaches 50% near 400 tokens under full history but plateaus below 50% under tail-clip memory policy, while replace-mode switching largely reflects state reset.
AuDisAgent reformulates multimodal controversy detection as a dynamic audience dissemination process using screening, panel discussion, and arbitration agents, plus comment bootstrapping, and reports outperforming prior static methods on a public dataset.
GSAR is a grounding-evaluation framework for multi-agent LLMs that uses a four-way claim typology, evidence-weighted asymmetric scoring, and tiered recovery decisions to detect and mitigate hallucinations.
A new 7x4 taxonomy organizes agentic AI security threats by architectural layer and persistence timescale, revealing under-explored upper layers and missing defenses after surveying 116 papers.
AgentFlow uses a typed graph DSL covering roles, prompts, tools, topology and protocol plus a runtime-signal feedback loop to optimize multi-agent harnesses, reaching 84.3% on TerminalBench-2 and discovering ten new zero-days in Chrome including two critical sandbox escapes.
HiPO improves LLM reasoning performance by optimizing preferences separately on response segments rather than entire outputs.
Refute-or-Promote applies adversarial multi-agent review with kill gates and empirical verification to filter LLM defect candidates, killing 79-83% before disclosure and yielding 4 CVEs plus multiple accepted fixes across libraries, C++ standard, and compilers.
CACM improves language-based drug discovery agents by 36.4% via protocol auditing, a grounded diagnostician, and compressed static/dynamic/corrective memory channels that localize failures and bias corrections.
ConFixAgent repairs diverse concurrency bugs end-to-end by using Static Happens-Before graphs to extract relevant code context for LLMs, outperforming prior tools in benchmarks.
citing papers explorer
-
DSPy: Compiling Declarative Language Model Calls into Self-Improving Pipelines
DSPy compiles short declarative programs into LM pipelines that self-optimize and outperform both standard few-shot prompting and expert-written chains on math, retrieval, and QA tasks.
-
Promptbreeder: Self-Referential Self-Improvement Via Prompt Evolution
Promptbreeder evolves both task prompts and the mutation prompts that improve them using LLMs, outperforming Chain-of-Thought and Plan-and-Solve on arithmetic and commonsense reasoning benchmarks.
-
Dynamic Bidirectional Pattern Memory: A Production-Scale Empirical Characterisation of Inference-Time Gating in Clinical NLP
Empirical study on production-scale clinical NLP shows direct learning from verifier rejections fails due to sparse data while fixed ontology and evidence-support filters succeed, with selectivity determined by matching verifier evidence.
-
Falsification, Not Exposure: An Internally Preregistered Placebo-Controlled Decomposition of Self-Repair Feedback in Frozen Small Code Models
Preregistered placebo-controlled decomposition shows external executable counterexamples drive self-repair gains in small code models more than re-exposure or self-critique.
-
AlgoBench: Benchmarking Algorithmic Adaptation in Code Generation
AlgoBench creates traceable variants of competitive programming problems via constraint shifts that invalidate original algorithms, paired with complexity metrics that reveal LLMs often produce functionally correct but asymptotically unsuitable solutions.
-
Delayed Verification Destabilizes Multi-Agent LLM Belief: Instability Thresholds and Optimal Corrector Placement
Models delayed verification in multi-agent LLMs as graph consensus, derives stability thresholds (inverse golden ratio for delay two) via grounded Laplacian, and gives a supermodular greedy rule for corrector placement; experiments on five models confirm dose-delay oscillations.
-
Stability vs. Manipulability: Evaluating Robustness Under Post-Decision Interaction in LLM Judges
LLM judges exhibit high stability under neutral re-evaluation but substantial reversibility under targeted post-decision challenges, quantified via a new Evaluation Robustness Score (ERS).
-
Testing LLM Arithmetic Reasoning Generalization with Automatic Numeric-Remapping Attacks
An automatic numeric-remapping attack generator reveals 12-26 point accuracy drops on GSM8K for three LLMs while MAWPS and MultiArith stay near 98%.
-
How Generation Architecture Shapes Code Complexity in Multi-Agent LLM Systems: A Paired Study on HumanEval
Six multi-agent architectures for LLM code generation on HumanEval form two complexity clusters separated by 50-130%, with no accuracy advantage for the complex cluster.
-
HIDBench: Benchmarking Large Language Models for Host-Based Intrusion Detection
HIDBench unifies DARPA-E3, DARPA-E5, and NodLink datasets with a data pipeline to benchmark LLMs for host-based intrusion detection, showing high precision on simple logs but sharp drops in MCC and rises in false positives on complex noisy data.
-
Formal Skill: Programmable Runtime Skills for Efficient and Accurate LLM Agents
Proposes Formal Skill as a programmable runtime abstraction for LLM agents, implemented in open-source FairyClaw, achieving competitive Harness-Bench scores with substantially fewer tokens.
-
DecisionBench: A Benchmark for Emergent Delegation in Long-Horizon Agentic Workflows
DecisionBench supplies a fixed task suite, model pool, delegation interface, and multi-axis metrics to evaluate emergent delegation, showing similar quality across awareness conditions but 15-31 point headroom under perfect delegation.
-
Weak-to-Strong Elicitation via Mismatched Wrong Drafts
Mismatched wrong drafts from Qwen2.5-Math-1.5B improve Mathstral-7B GRPO training, reaching 71.98% greedy pass@1 on MATH-500 and lifting AIME 2025/2026 pass@k over baselines and other draft variants.
-
Test-Time Hinting for Black-Box Vision-Language Models
Test-Time Hinting trains a hint generator to prepend contextual guidance to VLM prompts, improving accuracy on natural-image VQA benchmarks with generalization to unseen tasks and models.
-
Learning, Fast and Slow: Towards LLMs That Adapt Continually
Fast-Slow Training uses context optimization as fast weights alongside parameter updates as slow weights to achieve up to 3x better sample efficiency, higher performance, and less catastrophic forgetting than standard RL in continual LLM learning.
-
Perturbation Dose Responses in Recursive LLM Loops: Raw Switching, Stochastic Floors, and Persistent Escape under Append, Replace, and Dialog Updates
In 30-step recursive LLM loops, append-mode persistent escape from source basins reaches 50% near 400 tokens under full history but plateaus below 50% under tail-clip memory policy, while replace-mode switching largely reflects state reset.
-
From Static Analysis to Audience Dissemination: A Training-Free Multimodal Controversy Detection Multi-Agent Framework
AuDisAgent reformulates multimodal controversy detection as a dynamic audience dissemination process using screening, panel discussion, and arbitration agents, plus comment bootstrapping, and reports outperforming prior static methods on a public dataset.
-
GSAR: Typed Grounding for Hallucination Detection and Recovery in Multi-Agent LLMs
GSAR is a grounding-evaluation framework for multi-agent LLMs that uses a four-way claim typology, evidence-weighted asymmetric scoring, and tiered recovery decisions to detect and mitigate hallucinations.
-
A Systematic Survey of Security Threats and Defenses in LLM-Based AI Agents: A Layered Attack Surface Framework
A new 7x4 taxonomy organizes agentic AI security threats by architectural layer and persistence timescale, revealing under-explored upper layers and missing defenses after surveying 116 papers.
-
Synthesizing Multi-Agent Harnesses for Vulnerability Discovery
AgentFlow uses a typed graph DSL covering roles, prompts, tools, topology and protocol plus a runtime-signal feedback loop to optimize multi-agent harnesses, reaching 84.3% on TerminalBench-2 and discovering ten new zero-days in Chrome including two critical sandbox escapes.
-
HiPO: Hierarchical Preference Optimization for Adaptive Reasoning in LLMs
HiPO improves LLM reasoning performance by optimizing preferences separately on response segments rather than entire outputs.
-
Refute-or-Promote: An Adversarial Stage-Gated Multi-Agent Review Methodology for High-Precision LLM-Assisted Defect Discovery
Refute-or-Promote applies adversarial multi-agent review with kill gates and empirical verification to filter LLM defect candidates, killing 79-83% before disclosure and yielding 4 CVEs plus multiple accepted fixes across libraries, C++ standard, and compilers.
-
Constraint-Aware Corrective Memory for Language-Based Drug Discovery Agents
CACM improves language-based drug discovery agents by 36.4% via protocol auditing, a grounded diagnostician, and compressed static/dynamic/corrective memory channels that localize failures and bias corrections.
-
An End-to-End Approach for Fixing Concurrency Bugs via SHB-Based Context Extractor
ConFixAgent repairs diverse concurrency bugs end-to-end by using Static Happens-Before graphs to extract relevant code context for LLMs, outperforming prior tools in benchmarks.
-
Debug2Fix: Can Interactive Debugging Help Coding Agents Fix More Bugs?
Debug2Fix integrates interactive debugging via subagents into coding agents, delivering >20% gains on GitBug-Java and SWE-Bench-Live while enabling weaker models to match stronger ones.
-
Software Self-Extension with SelfEvolve: an Agentic Architecture for Runtime Code Generation
SelfEvolve achieves 92.7% Pass@1 success on 11 runtime self-extension tasks and outperforms baselines like AutoGen by 61.8% with statistical significance.
-
Compass vs Railway Tracks: Unpacking User Mental Models for Communicating Long-Horizon Work to Humans vs. AI
Users treat human delegation for long tasks as a flexible compass but AI delegation as rigid railway tracks due to perceived AI limitations in inference and judgment.
-
L1: Controlling How Long A Reasoning Model Thinks With Reinforcement Learning
LCPO trains L1 reasoning models to adhere to prompt-specified CoT lengths, supporting accuracy-compute trade-offs and yielding short reasoning models that outperform larger baselines at matched lengths.
-
Medusa: Simple LLM Inference Acceleration Framework with Multiple Decoding Heads
Medusa augments LLMs with multiple decoding heads and tree-based attention to predict and verify several tokens in parallel, yielding 2.2-3.6x inference speedup via two fine-tuning regimes.
-
Self-Rewarding Language Models
Iterative self-rewarding via LLM-as-Judge in DPO training on Llama 2 70B improves instruction following and self-evaluation, outperforming GPT-4 on AlpacaEval 2.0.
-
AgentCoder: Multi-Agent-based Code Generation with Iterative Testing and Optimisation
A three-agent loop of code generation, test creation, and execution feedback lifts pass@1 to 96.3% on HumanEval and 91.8% on MBPP for GPT-4 while using roughly half the tokens of prior state-of-the-art.
-
Large Language Models as Optimizers
Large language models can optimize by being prompted with histories of past solutions and scores to propose better ones, producing prompts that raise accuracy up to 8% on GSM8K and 50% on Big-Bench Hard over human-designed baselines.
-
Reflexion: Language Agents with Verbal Reinforcement Learning
Reflexion lets LLM agents improve via stored verbal reflections on task feedback, reaching 91% pass@1 on HumanEval and outperforming prior GPT-4 results.
-
DICE: Entropy-Regularized Equilibrium Selection for Stable Multi-Agent LLM Coordination
HQRE entropy regularization makes multi-agent LLM coordination well-posed, yielding unique equilibria, linear mirror convergence, bounded Bayesian regret, and DICE gains of 4.3–8.5 pp on reasoning/planning tasks.
-
Recursive Self-Improvement in AI: From Bounded Self-Refinement to Autonomous Research Loops
A survey of 1,250 papers organizes AI self-improvement along two axes—what is improved and loop closure—finding that demonstrated self-improvement strength tracks a verification hierarchy from formal verifiers down to intrinsic self-assessment.
-
Cost-Effective Agent Harnesses for Abstract Reasoning and Generalization on ARC-AGI-1
Explorer-Definer and Reflective Orchestrator harnesses raise DeepSeek V3.2 from 15.5% to 67.25% pass@2 on ARC-AGI-1 public eval at $0.25–$0.62 per task without ARC-specific training.
-
AI Native Games: A Survey and Roadmap
AI-native games require runtime generative AI as a non-substitutable core-loop mechanism; a 53-game corpus clusters in language-forward narrative and epistemic designs.
-
TRIAGE: Role-Typed Credit Assignment for Agentic Reinforcement Learning
Role-typed, judge-assigned segment labels (decisive/exploration/no-progress/regression) added to GRPO advantages improve agentic RL success rates on three benchmarks, primarily by withholding positive credit from regressive actions in successful trajectories.
-
ReGRPO: Reflection-Augmented Policy Optimization for Tool-Using Agents
ReGRPO augments group-relative policy optimization with a reflective data engine that generates ErrorType-Evidence-FixPlan triplets from near-miss tool actions to improve recovery in multimodal agents.
-
ISM:Self-Improving Strategy Memory for Continual Mathematical Reasoning
ISM maintains a self-refined bank of verified strategy schemas to enable continual mathematical reasoning in frozen LLMs, outperforming baselines on MATH-Hard and OlympiadBench while using 64-86% fewer schemas.
-
Stop Hand-Holding Your Coding Agent: Engineering the Loops that Replace Step-by-Step Prompting
Introduces loop engineering as a distinct practice layer for coding agents, supplies a taxonomy and verification ladder, and analyzes a hand-coded corpus of fifty real loops.
-
Agent vs. Parametric World Models: Hybrid Planning for Reliable Language Agents
A small parametric transition model plus a Jaccard consistency gate grounds LLM agent state deltas, cutting hallucinated-state rate ~80% and raising success from 0.668 to 0.838 on graph planning benchmarks.
-
MetaPS: Adaptive Programmatic Strategy Selection for Market Agents
MetaPS trains models via simulation rollouts to select from programmatic strategy libraries for market agents, yielding better performance than fixed or direct LLM baselines across model sizes.
-
Answer Engineering: Local Trajectory Editing for Protocol-Constrained Decision Making in Large Language Models
Answer Engineering uses local trajectory editing during autoregressive generation to raise protocol compliance on a clinical SSNHL benchmark from 25.1% to 83.5% and balanced accuracy from 42.0% to 80.7%.
-
Building Agent Harnesses for Scientific Curation from Multimodal Sources
An agent harness combining staged task decomposition, multimodal evidence tooling, and artifact-grounded self-improvement scores 81.0 GRAS on multimodal scientific curation, 22.4 points above the strongest baseline — with the caveat that 8 of 23 evaluation papers were used for optimization.
-
Self-Preference Is Weak or Absent in Verifiable Instruction-Following Revision: A Four-Model Test Under Genuine Authorship
No detectable self-preference in LLMs accepting verified fixes to their own IFEval drafts versus neutral judges across four models and 85 comparisons (gap -5.1 pp, CI crossing zero).
-
REVES: REvision and VErification--Augmented Training for Test-Time Scaling
REVES augments LLM post-training by decoupling revision and verification signals from successful multi-step trajectories, reporting +6.5 point gains on LiveCodeBench over RL baselines.
-
Detect, Remask, Repair: Diffusion Editing for Faithful Summarization of Evolving Contexts
Diffusion-based localized editing framework for faithful summarization of evolving contexts, introducing the StreamSum benchmark and showing tradeoffs in faithfulness, speed, and preservation.
-
When Video Misreads: Closed-Loop Distillation of Reading Heuristics for Exploratory Manipulation Trace QA
Closed-Loop Trace Distillation distills one-line natural-language prompts from labeled training traces to improve VLM accuracy on predicting minimal-success action chains in Exploratory Manipulation Trace QA by 0.38-0.47 across simulator and real-robot tasks.
-
The Self-Correction Illusion: Role Relabeling Gates Explicit Error Flagging in Large Language Models
LLMs flag byte-identical errors 23-93 points more often when the error is presented under an external chat role than inside their own <thought> block, so role labeling, not content, gates explicit self-correction.