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.
hub Canonical reference
Teaching Large Language Models to Self-Debug
Canonical reference. 82% of citing Pith papers cite this work as background.
75
Pith papers citing it
74
external citations · Pith
Background
82% of classified citations
abstract
Large language models (LLMs) have achieved impressive performance on code generation. However, for complex programming tasks, generating the correct solution in one go becomes challenging, thus some prior works have designed program repair approaches to improve code generation performance. In this work, we propose Self-Debugging, which teaches a large language model to debug its predicted program via few-shot demonstrations. In particular, we demonstrate that Self-Debugging can teach the large language model to perform rubber duck debugging; i.e., without any human feedback on the code correctness or error messages, the model is able to identify its mistakes by investigating the execution results and explaining the generated code in natural language. Self-Debugging achieves the state-of-the-art performance on several code generation benchmarks, including the Spider dataset for text-to-SQL generation, TransCoder for C++-to-Python translation, and MBPP for text-to-Python generation. On the Spider benchmark where there are no unit tests to verify the correctness of predictions, Self-Debugging with code explanation consistently improves the baseline by 2-3%, and improves the prediction accuracy on problems of the hardest level by 9%. On TransCoder and MBPP where unit tests are available, Self-Debugging improves the baseline accuracy by up to 12%. Meanwhile, by leveraging feedback messages and reusing failed predictions, Self-Debugging notably improves sample efficiency, and can match or outperform baseline models that generate more than 10x candidate programs.
hub tools
citation-role summary
background 15
baseline 1
method 1
citation-polarity summary
claims ledger
- abstract Large language models (LLMs) have achieved impressive performance on code generation. However, for complex programming tasks, generating the correct solution in one go becomes challenging, thus some prior works have designed program repair approaches to improve code generation performance. In this work, we propose Self-Debugging, which teaches a large language model to debug its predicted program via few-shot demonstrations. In particular, we demonstrate that Self-Debugging can teach the large language model to perform rubber duck debugging; i.e., without any human feedback on the code correct
co-cited works
representative citing papers
Preregistered placebo-controlled decomposition shows external executable counterexamples drive self-repair gains in small code models more than re-exposure or self-critique.
Introduces OPT* tasks and two training regimes (solver-guided online policy optimization with rank-based reward shaping and search-based offline RL) plus a theoretical link between search success and information extraction per budget unit, showing empirical gains in optimization-like reasoning.
3DCodeBench is a new benchmark evaluating 12 VLMs on translating multimodal prompts into procedural 3D modeling code, paired with 3DCodeArena for human preference rankings.
PowerCodeBench and a boundary-aware intervention raise LLM accuracy on power-system code generation by 32-56 points across ten open-weight models and four commercial APIs on a 2,000-task benchmark.
Constrained Diffusion for Code (CDC) integrates constraint satisfaction into the reverse denoising process of discrete diffusion models via constraint-aware operators that use optimization and program analysis to steer generation toward feasible programs.
RubricRefine is a training-free pre-execution method that creates rubrics to score and fix inter-tool contract violations in agent code, reaching 0.86 average on M3ToolEval across seven models with zero executions and lower latency.
A new Intent Fidelity Score and refinement loop verify that LLM-generated simulation code matches the intended PDEs, improving performance on a 220-case benchmark where execution alone fails to ensure correctness.
LEAF-SQL uses level-wise exploration with adaptive fine-graining and dual agents to generate diverse SQL skeletons, reaching 71.6% execution accuracy on the BIRD benchmark and outperforming prior search- and skeleton-based methods.
SkillGuard extracts executable environment contracts from LLM skill documents to detect only relevant drifts, reporting zero false positives on 599 cases, 100% precision in known-drift tests, and raising one-round repair success from 10% to 78%.
CP-SynC uses coordinated LLM agents to generate, validate via synthesized checkers, and select MiniZinc models from natural language, substantially outperforming baselines on a 100-problem benchmark.
ADI equips AI debugging agents with function-level interaction via a new execution trace structure, raising SWE-bench Verified resolution to 63.8% at $1.28 per task and delivering 6-18% gains when added to existing agents.
PhysCodeBench benchmark and SMRF multi-agent framework enable better AI generation of physically accurate 3D simulation code, boosting performance by 31 points over baselines.
PlayCoder raises the rate of LLM-generated GUI apps that can be played end-to-end without logic errors from near zero to 20.3% Play@3 by adding repository-aware generation, agent-driven testing, and iterative repair.
Structural dependency graphs and staged pre-execution verification raise LLM-based EDA code pass rates to 82.5% (single-step) and 70-84% (multi-step) while halving tool calls by catching dependency violations before runtime.
Cross-modal agreement between chain-of-thought and program-of-thought reasoning enables self-consistency with only two LLM samples, reducing sampling cost by 9.3x while improving accuracy.
FORGE uses a reasoning-action-observation loop and Dynamic Forest of Agents to perform scalable LLM-based binary analysis, finding 1,274 vulnerabilities across 591 of 3,457 real-world firmware binaries at 72.3% precision and broader coverage than prior methods.
BACE reformulates LLM code synthesis as Bayesian co-evolution of code and test populations anchored on minimal public examples, achieving superior performance on LiveCodeBench v6.
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.
SelfEvolve achieves 92.7% Pass@1 success on 11 runtime self-extension tasks and outperforms baselines like AutoGen by 61.8% with statistical significance.
DynaFix iteratively feeds execution-level dynamic information such as variable states and control flows into LLM prompts to repair 186 bugs on Defects4J, a 10% gain over baselines including 38 previously unrepaired cases.
SWE-EVO shows GPT-5.4 with OpenHands reaching only 25% success on complex multi-file evolution tasks versus 72.8% on SWE-Bench Verified, and introduces Fix Rate as a partial-progress metric.
CodeCureAgent achieves 96.8% plausible fixes and 86.3% correct fixes for 1,000 SonarQube warnings across 106 Java projects using an agentic LLM framework.
ImProver is an LLM agent using Chain-of-States, error-correction, and retrieval to rewrite Lean proofs for arbitrary user-defined optimization criteria like shortness and readability.
citing papers explorer
-
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.
-
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.
-
Step-by-Step Optimization-like Reasoning in LLMs over Expanding Search Spaces
Introduces OPT* tasks and two training regimes (solver-guided online policy optimization with rank-based reward shaping and search-based offline RL) plus a theoretical link between search success and information extraction per budget unit, showing empirical gains in optimization-like reasoning.
-
3DCodeBench: Benchmarking Agentic Procedural 3D Modeling Via Code
3DCodeBench is a new benchmark evaluating 12 VLMs on translating multimodal prompts into procedural 3D modeling code, paired with 3DCodeArena for human preference rankings.
-
Knowledge Boundary Probing and Demand-Guided Intervention for LLM-Based Power System Code Generation
PowerCodeBench and a boundary-aware intervention raise LLM accuracy on power-system code generation by 32-56 points across ten open-weight models and four commercial APIs on a 2,000-task benchmark.
-
Constrained Code Generation with Discrete Diffusion
Constrained Diffusion for Code (CDC) integrates constraint satisfaction into the reverse denoising process of discrete diffusion models via constraint-aware operators that use optimization and program analysis to steer generation toward feasible programs.
-
RubricRefine: Improving Tool-Use Agent Reliability with Training-Free Pre-Execution Refinement
RubricRefine is a training-free pre-execution method that creates rubrics to score and fix inter-tool contract violations in agent code, reaching 0.86 average on M3ToolEval across seven models with zero executions and lower latency.
-
Your Simulation Runs but Solves the Wrong Physics: PDE-Grounded Intent Verification for LLM-Generated Multiphysics Simulation Code
A new Intent Fidelity Score and refinement loop verify that LLM-generated simulation code matches the intended PDEs, improving performance on a 220-case benchmark where execution alone fails to ensure correctness.
-
LEAF-SQL: Level-wise Exploration with Adaptive Fine-graining for Text-to-SQL Skeleton Prediction
LEAF-SQL uses level-wise exploration with adaptive fine-graining and dual agents to generate diverse SQL skeletons, reaching 71.6% execution accuracy on the BIRD benchmark and outperforming prior search- and skeleton-based methods.
-
Skill Drift Is Contract Violation: Proactive Maintenance for LLM Agent Skill Libraries
SkillGuard extracts executable environment contracts from LLM skill documents to detect only relevant drifts, reporting zero false positives on 599 cases, 100% precision in known-drift tests, and raising one-round repair success from 10% to 78%.
-
CP-SynC: Multi-Agent Zero-Shot Constraint Modeling in MiniZinc with Synthesized Checkers
CP-SynC uses coordinated LLM agents to generate, validate via synthesized checkers, and select MiniZinc models from natural language, substantially outperforming baselines on a 100-problem benchmark.
-
Empowering Autonomous Debugging Agents with Efficient Dynamic Analysis
ADI equips AI debugging agents with function-level interaction via a new execution trace structure, raising SWE-bench Verified resolution to 63.8% at $1.28 per task and delivering 6-18% gains when added to existing agents.
-
PhysCodeBench: Benchmarking Physics-Aware Symbolic Simulation of 3D Scenes via Self-Corrective Multi-Agent Refinement
PhysCodeBench benchmark and SMRF multi-agent framework enable better AI generation of physically accurate 3D simulation code, boosting performance by 31 points over baselines.
-
PlayCoder: Making LLM-Generated GUI Code Playable
PlayCoder raises the rate of LLM-generated GUI apps that can be played end-to-end without logic errors from near zero to 20.3% Play@3 by adding repository-aware generation, agent-driven testing, and iterative repair.
-
Structural Verification for Reliable EDA Code Generation without Tool-in-the-Loop Debugging
Structural dependency graphs and staged pre-execution verification raise LLM-based EDA code pass rates to 82.5% (single-step) and 70-84% (multi-step) while halving tool calls by catching dependency violations before runtime.
-
Self-Consistency from Only Two Samples: CoT-PoT Ensembling for Efficient LLM Reasoning
Cross-modal agreement between chain-of-thought and program-of-thought reasoning enables self-consistency with only two LLM samples, reducing sampling cost by 9.3x while improving accuracy.
-
Feedback-Driven Execution for LLM-Based Binary Analysis
FORGE uses a reasoning-action-observation loop and Dynamic Forest of Agents to perform scalable LLM-based binary analysis, finding 1,274 vulnerabilities across 591 of 3,457 real-world firmware binaries at 72.3% precision and broader coverage than prior methods.
-
BACE: LLM-based Code Generation through Bayesian Anchored Co-Evolution of Code and Test Populations
BACE reformulates LLM code synthesis as Bayesian co-evolution of code and test populations anchored on minimal public examples, achieving superior performance on LiveCodeBench v6.
-
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.
-
DynaFix: Iterative Automated Program Repair Driven by Execution-Level Dynamic Information
DynaFix iteratively feeds execution-level dynamic information such as variable states and control flows into LLM prompts to repair 186 bugs on Defects4J, a 10% gain over baselines including 38 previously unrepaired cases.
-
SWE-EVO: Benchmarking Coding Agents in Long-Horizon Software Evolution Scenarios
SWE-EVO shows GPT-5.4 with OpenHands reaching only 25% success on complex multi-file evolution tasks versus 72.8% on SWE-Bench Verified, and introduces Fix Rate as a partial-progress metric.
-
CodeCureAgent: Automatic Classification and Repair of Static Analysis Warnings
CodeCureAgent achieves 96.8% plausible fixes and 86.3% correct fixes for 1,000 SonarQube warnings across 106 Java projects using an agentic LLM framework.
-
ImProver: Agent-Based Automated Proof Optimization
ImProver is an LLM agent using Chain-of-States, error-correction, and retrieval to rewrite Lean proofs for arbitrary user-defined optimization criteria like shortness and readability.
-
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.
-
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.
-
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.
-
Sequential Planning via Anchored Robotic Keypoints
SPARK reaches 43.7% success on six LIBERO-PRO cells by LLM-generated typed behavior trees plus multi-prompt perception and recovery, more than doubling CaP-Agent0 and VLA baselines.
-
E-TTS: A New Embodied Test-Time Scaling Framework for Robotic Manipulation
E-TTS introduces a plug-and-play test-time scaling method for embodied tasks that unifies reasoning-action sampling with history buffers and closed-loop refinement to improve performance on manipulation benchmarks.
-
AutoPDE: Reliable Agentic PDE Solving via Explicitly Represented Solver Strategies
AutoPDE maintains an explicit solver strategy through PDE analysis, numerical method selection, and adaptive tuning, achieving 54.5% pass rate on PDE Agent Bench, 14.2 points above the strongest baseline.
-
Auto-Configuring Scientific Simulators with Lightweight Coding-Agent Adapters
SIGA is a coding-agent adapter using retrieval, procedural memory, and validation gates that raises success rate on GEOS from 0.720 to 0.789 while cutting variance 16x and matching expert quality in minutes instead of hours.
-
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.
-
CYGNET: Cypher Gate for Neural Execution Triage and Cost Containment
A multi-backend structural gate for LLM-generated Cypher queries achieves 100% detection of parse, constraint, and schema errors at zero false positives on 1135 queries while preserving model accuracy and adding a cost planner.
-
See Before You Code: Learning Visual Priors for Spatially Aware Educational Animation Generation
OmniManim improves render quality in educational animation code generation by using a Vision Agent with coarse-to-fine bounding-box denoising and interpolation-aware optimization on new datasets.
-
Learning from Failures: Correction-Oriented Policy Optimization with Verifiable Rewards
CIPO jointly optimizes standard RLVR rewards with correction samples derived from the model's own failed attempts, yielding better reasoning and self-correction on math and code benchmarks.
-
Verification Mirage: Mapping the Reliability Boundary of Self-Verification in Medical VQA
Self-verification in medical VQA creates a verification mirage where verifiers exhibit high error and agreement bias on wrong answers, with reliability strongly conditioned on task type.
-
Debugging the Debuggers: Failure-Anchored Structured Recovery for Software Engineering Agents
PROBE turns runtime telemetry from failed software engineering agent runs into evidence-grounded diagnoses and actionable recovery guidance, achieving 65.37% diagnosis accuracy and 21.79% recovery rate on 257 cases.
-
PaT: Planning-after-Trial for Efficient Test-Time Code Generation
PaT defers planning until after failed trials in LLM code generation, enabling heterogeneous cheap-plus-powerful model setups that match large-model performance at roughly 69% lower cost.
-
EGREFINE: An Execution-Grounded Optimization Framework for Text-to-SQL Schema Refinement
EGRefine optimizes column renamings via execution-grounded verification and view materialization to recover Text-to-SQL accuracy lost to schema naming issues while guaranteeing query equivalence.
-
WebGen-R1: Incentivizing Large Language Models to Generate Functional and Aesthetic Websites with Reinforcement Learning
WebGen-R1 uses end-to-end RL with scaffold-driven generation and cascaded rewards for structure, function, and aesthetics to transform a 7B model into a generator of deployable multi-page websites that rivals much larger models.
-
CityRAG: Stepping Into a City via Spatially-Grounded Video Generation
PlayCoder combines a repository-aware coding agent with a vision-based GUI testing agent and an automated program repair loop to detect and fix silent logic errors in LLM-generated interactive application code.
-
PV-SQL: Synergizing Database Probing and Rule-based Verification for Text-to-SQL Agents
PV-SQL boosts Text-to-SQL execution accuracy by 5% and valid efficiency by 20.8% on BIRD benchmarks via database probing and rule-based SQL verification while using fewer tokens.
-
Beyond Fixed Tests: Repository-Level Issue Resolution as Coevolution of Code and Behavioral Constraints
Agent-CoEvo is a multi-agent LLM framework that coevolves code patches and test patches to resolve repository-level issues, outperforming fixed-test baselines on SWE-bench Lite and SWT-bench Lite.
-
Ensemble-Based Uncertainty Estimation for Code Correctness Estimation
Ensemble Semantic Entropy improves correlation with code correctness over single-model methods and powers a cascading scaling system that cuts FLOPs by 64.9% while preserving performance on LiveCodeBench.
-
Self-Execution Simulation Improves Coding Models
Training LLMs to simulate code execution enables self-verification and self-fixing that improves competitive programming performance.
-
GenoMAS: A Multi-Agent Framework for Scientific Discovery via Code-Driven Gene Expression Analysis
GenoMAS deploys six specialized LLM agents with guided planning to preprocess transcriptomic data and identify genes, reaching 89.13% composite similarity and 60.48% F1 on the GenoTEX benchmark while outperforming prior methods.
-
Effective LLM Code Refinement via Property-Oriented and Structurally Minimal Feedback
PGS generates property-oriented, structurally minimal feedback from high-level program properties to refine LLM code, yielding up to 13.4% pass@1 gains and 1.4-1.6x higher bug-fix rates than prior TDD and debugging baselines.
-
SysLLMatic: Large Language Models are Software System Optimizers
SysLLMatic integrates LLMs with performance diagnostics and a 43-pattern catalog to optimize complex software, reporting 1.54x latency and 1.24x energy gains over compilers on large Java systems where prior LLM methods did not scale.