Prompt injection attacks can self-replicate across LLM agents in multi-agent systems, enabling data theft, misinformation, and system disruption while propagating silently.
hub Canonical reference
Encouraging Divergent Thinking in Large Language Models through Multi-Agent Debate
Canonical reference. 77% of citing Pith papers cite this work as background.
64
Pith papers citing it
28
external citations · Pith
Background
77% of classified citations
abstract
Modern large language models (LLMs) like ChatGPT have shown remarkable performance on general language tasks but still struggle on complex reasoning tasks, which drives the research on cognitive behaviors of LLMs to explore human-like problem-solving strategies. Along this direction, one representative strategy is self-reflection, which asks an LLM to refine the solution with the feedback generated by itself iteratively. However, our study shows that such reflection-style methods suffer from the Degeneration-of-Thought (DoT) problem: once the LLM has established confidence in its solutions, it is unable to generate novel thoughts later through reflection even if its initial stance is incorrect. To address the DoT problem, we propose a Multi-Agent Debate (MAD) framework, in which multiple agents express their arguments in the state of "tit for tat" and a judge manages the debate process to obtain a final solution. Clearly, our MAD framework encourages divergent thinking in LLMs which would be helpful for tasks that require deep levels of contemplation. Experiment results on two challenging datasets, commonsense machine translation and counter-intuitive arithmetic reasoning, demonstrate the effectiveness of our MAD framework. Extensive analyses suggest that the adaptive break of debate and the modest level of "tit for tat" state are required for MAD to obtain good performance. Moreover, we find that LLMs might not be a fair judge if different LLMs are used for agents. Code is available at https://github.com/Skytliang/Multi-Agents-Debate.
hub tools
citation-role summary
background 11
baseline 1
method 1
citation-polarity summary
representative citing papers
DPO fine-tuning causes LLM agents to share output distributions with pairwise error correlations of ρ=0.70, reducing ten agents to the effective power of ≈1.4 independent forecasters.
Multi-agent LLM deliberation is modeled with recoverable hidden anchors that allow opinions to escape the convex hull of initial beliefs, unlike classical consensus 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.
Successor-representation spectra of row-stochastic communication operators predict perturbation robustness, consensus speed, and error accumulation in multi-agent LLM topologies, with condition number showing perfect empirical rank correlation.
EditRefiner uses a perception-reasoning-action-evaluation agent loop and the EditFHF-15K human feedback dataset to refine text-guided image edits more accurately than prior methods.
HANDRAISER learns optimal interruption points in multi-agent LLM communication using estimated future reward and cost, achieving 32.2% lower communication cost with comparable or better task results across games, scheduling, and debate.
Discourse among AI agents on Moltbook is largely determined by architectural constraints like context windows and identity files, appearing as social learning but actually short-horizon contextual conditioning.
A survey that defines Compound AI Systems, proposes a multi-dimensional taxonomy based on component roles and orchestration strategies, reviews four foundational paradigms, and identifies key challenges for future research.
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.
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.
Tool-using LLM agents can implement undetectable stegosystems, shifting the primary barrier to covert multi-agent collusion from technical feasibility to coordination without explicit agreement.
On-policy self-distillation with sampled demonstrations reduces rollout diversity by amplifying existing probability gaps in the base model, unlike ideal RL which preserves ratios among correct outputs.
Propagational Proxy Voting driven by letter entropy and centered reasoning embeddings beats majority by +2.24 pp on non-trivial MMLU-Pro questions without labels or training.
Semantic Quorum Assurance routes AI infrastructure proposals to diverse sandboxed validators and applies risk-adaptive quorums to cut unsafe approvals from 18.5% to 0.3% on 500 scenarios.
H-CSC is a BFT protocol that converts embedding-derived signals over LLM proposals into typed semantic_commit, verdict_commit, or abort outcomes, with empirical results on diagnostic and benchmark tasks showing high commit rates and safety under attacks.
PROBE pipeline with deterministic PCAP normalization, verdict-aware evidence ensembles, and composite reliability scoring raises weighted evidence F1 to 0.957 on 87 Wi-Fi captures while avoiding LLM self-confidence and evaluation bias issues.
Closed-system multi-step LLM reasoning is subject to an information-theoretic bound where mutual information with evidence decreases, preserving accuracy while eroding faithfulness, with EGSR recovering it on SciFact and FEVER.
Architectural heterogeneity across 7-9B models reduces first-choice concentration in policy simulations (70.9% to 46.1% and 46.0% to 22.9%), while coherence validation shows a scenario-dependent tradeoff.
PBRC is a contract protocol that enforces evidential belief updates in deliberative multi-agent systems and proves it prevents conformity-driven false cascades under conservative fallbacks.
PolySwarm aggregates predictions from 50 LLM personas for Polymarket trading using Bayesian combination and divergence metrics, outperforming single models in calibration while adding latency arbitrage via CEX price models.
A compact 0.6B-parameter coordinator with a 10K-parameter head uses evolutionary strategy to dynamically delegate roles to LLMs, achieving SOTA results such as 86.2% on LiveCodeBench.
PosterForest uses a hierarchical Poster Tree and content-layout agent collaboration to generate scientific posters from papers without training, outperforming prior automated methods in human and automated evaluation.
Math reasoning gains in LLMs rarely transfer to general domains; RL tuning generalizes while SFT causes forgetting and representation drift.
citing papers explorer
-
Prompt Infection: LLM-to-LLM Prompt Injection within Multi-Agent Systems
Prompt injection attacks can self-replicate across LLM agents in multi-agent systems, enabling data theft, misinformation, and system disruption while propagating silently.
-
Preference Optimization Drives Monoculture in LLM Prediction Markets
DPO fine-tuning causes LLM agents to share output distributions with pairwise error correlations of ρ=0.70, reducing ten agents to the effective power of ≈1.4 independent forecasters.
-
Hidden Anchors in Multi-Agent LLM Deliberation
Multi-agent LLM deliberation is modeled with recoverable hidden anchors that allow opinions to escape the convex hull of initial beliefs, unlike classical consensus models.
-
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.
-
Predictive Maps of Multi-Agent Reasoning: A Successor-Representation Spectrum for LLM Communication Topologies
Successor-representation spectra of row-stochastic communication operators predict perturbation robustness, consensus speed, and error accumulation in multi-agent LLM topologies, with condition number showing perfect empirical rank correlation.
-
EditRefiner: A Human-Aligned Agentic Framework for Image Editing Refinement
EditRefiner uses a perception-reasoning-action-evaluation agent loop and the EditFHF-15K human feedback dataset to refine text-guided image edits more accurately than prior methods.
-
Learning to Interrupt in Language-based Multi-agent Communication
HANDRAISER learns optimal interruption points in multi-agent LLM communication using estimated future reward and cost, achieving 32.2% lower communication cost with comparable or better task results across games, scheduling, and debate.
-
What Do AI Agents Talk About? Discourse and Architectural Constraints in the First AI-Only Social Network
Discourse among AI agents on Moltbook is largely determined by architectural constraints like context windows and identity files, appearing as social learning but actually short-horizon contextual conditioning.
-
From Standalone LLMs to Integrated Intelligence: A Survey of Compound Al Systems
A survey that defines Compound AI Systems, proposes a multi-dimensional taxonomy based on component roles and orchestration strategies, reviews four foundational paradigms, and identifies key challenges for future research.
-
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.
-
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.
-
Tool Use Enables Undetectable Steganography in Multi-Agent LLM Systems
Tool-using LLM agents can implement undetectable stegosystems, shifting the primary barrier to covert multi-agent collusion from technical feasibility to coordination without explicit agreement.
-
On-Policy Self-Distillation with Sampled Demonstrations Reduces Output Diversity
On-policy self-distillation with sampled demonstrations reduces rollout diversity by amplifying existing probability gaps in the base model, unlike ideal RL which preserves ratios among correct outputs.
-
When Does Delegation Beat Majority? A Delegation-Based Aggregator for Multi-Sample LLM Inference
Propagational Proxy Voting driven by letter entropy and centered reasoning embeddings beats majority by +2.24 pp on non-trivial MMLU-Pro questions without labels or training.
-
Semantic Quorum Assurance: Collective Certification for Non-Deterministic AI Infrastructure
Semantic Quorum Assurance routes AI infrastructure proposals to diverse sandboxed validators and applies risk-adaptive quorums to cut unsafe approvals from 18.5% to 0.3% on 500 scenarios.
-
Hierarchical Certified Semantic Commitment for Byzantine-Resilient LLM-Agent Collaboration
H-CSC is a BFT protocol that converts embedding-derived signals over LLM proposals into typed semantic_commit, verdict_commit, or abort outcomes, with empirical results on diagnostic and benchmark tasks showing high commit rates and safety under attacks.
-
Evidence-Grounded Ensemble Diagnosis of 802.11 Packet Captures: A Multi-Stage Pipeline with Deterministic Reliability Scoring
PROBE pipeline with deterministic PCAP normalization, verdict-aware evidence ensembles, and composite reliability scoring raises weighted evidence F1 to 0.957 on 87 Wi-Fi captures while avoiding LLM self-confidence and evaluation bias issues.
-
The Reasoning Trap: An Information-Theoretic Bound on Closed-System Multi-Step LLM Reasoning
Closed-system multi-step LLM reasoning is subject to an information-theoretic bound where mutual information with evidence decreases, preserving accuracy while eroding faithfulness, with EGSR recovering it on SciFact and FEVER.
-
Preserving Disagreement: Architectural Heterogeneity and Coherence Validation in Multi-Agent Policy Simulation
Architectural heterogeneity across 7-9B models reduces first-choice concentration in policy simulations (70.9% to 46.1% and 46.0% to 22.9%), while coherence validation shows a scenario-dependent tradeoff.
-
Preregistered Belief Revision Contracts
PBRC is a contract protocol that enforces evidential belief updates in deliberative multi-agent systems and proves it prevents conformity-driven false cascades under conservative fallbacks.
-
PolySwarm: A Multi-Agent Large Language Model Framework for Prediction Market Trading and Latency Arbitrage
PolySwarm aggregates predictions from 50 LLM personas for Polymarket trading using Bayesian combination and divergence metrics, outperforming single models in calibration while adding latency arbitrage via CEX price models.
-
TRINITY: An Evolved LLM Coordinator
A compact 0.6B-parameter coordinator with a 10K-parameter head uses evolutionary strategy to dynamically delegate roles to LLMs, achieving SOTA results such as 86.2% on LiveCodeBench.
-
PosterForest: Hierarchical Multi-Agent Collaboration for Scientific Poster Generation
PosterForest uses a hierarchical Poster Tree and content-layout agent collaboration to generate scientific posters from papers without training, outperforming prior automated methods in human and automated evaluation.
-
Does Math Reasoning Improve General LLM Capabilities? Understanding Transferability of LLM Reasoning
Math reasoning gains in LLMs rarely transfer to general domains; RL tuning generalizes while SFT causes forgetting and representation drift.
-
Mixture-of-Agents Enhances Large Language Model Capabilities
A layered Mixture-of-Agents system combining multiple LLMs achieves state-of-the-art results on AlpacaEval 2.0 (65.1%), MT-Bench, and FLASK, outperforming GPT-4 Omni.
-
A Dynamic LLM-Powered Agent Network for Task-Oriented Agent Collaboration
DyLAN automatically selects and dynamically organizes LLM agents for collaboration, outperforming fixed-agent baselines on code generation, reasoning, and decision tasks with up to 25% accuracy gains on some MMLU subjects.
-
Large Language Models Cannot Self-Correct Reasoning Yet
LLMs cannot reliably self-correct reasoning mistakes using only their internal capabilities and often degrade in performance without external feedback.
-
ToRA: A Tool-Integrated Reasoning Agent for Mathematical Problem Solving
ToRA trains language models on interactive tool-use trajectories with imitation learning and output shaping to integrate reasoning and external tools, yielding 13-19% gains on math datasets and new highs like 44.6% on MATH for a 7B model.
-
DoLa: Decoding by Contrasting Layers Improves Factuality in Large Language Models
DoLa reduces hallucinations in LLMs by contrasting logits from later versus earlier layers during decoding, improving truthfulness on TruthfulQA by 12-17 absolute points without fine-tuning or retrieval.
-
Cognitive Architectures for Language Agents
CoALA is a modular cognitive architecture for language agents that organizes memory components, action spaces for internal and external interaction, and a generalized decision-making loop to support more systematic development of capable agents.
-
When the Judge Changes, So Does the Measurement: Auditing LLM-as-Judge Reliability
Judge upgrades are not interchangeable: only Qwen3 1.7B→4B yields robust adjacent gains, MiniMax adjacent releases do not, and stronger judges reduce but do not remove bias or correlated jury errors.
-
Data Selection Through Iterative Self-Filtering for Vision-Language Settings
An iterative bootstrapped self-filtering approach selects balanced clean and diverse subsets from noisy vision-language datasets to train improved CLIP models.
-
What Should Agents Say? Action-state Communication for Efficient Multi-Agent Systems
Introduces PACT protocol that projects agent outputs into action-state records, yielding comparable or better task performance with substantially fewer tokens in multi-agent LLM systems and production harnesses.
-
LLM Consortium for Software Design Refinement: A Controlled Experiment on Multi-Agent Collaboration Topologies
A 520-run factorial experiment ranks an adversarial rewrite topology highest and cross-model review second among 12 LLM collaboration structures for software design, with parallel merge performing worst.
-
Dive into Ambiguity: A*-Inspired Multi-Agents Commonsense Obfuscation Attack on LLM Prompts
An A*-inspired multi-agent framework with hierarchical rewriting and a dynamic gamma parameter generates obfuscated prompts that achieve higher LLM attack success rates with fewer attempts than exhaustive search.
-
ExComm: Exploration-Stage Communication for Error-Resilient Agentic Test-Time Scaling
ExComm adds cross-agent conflict detection and soft belief correction plus trajectory diversification to agentic test-time scaling, yielding 5-6% gains over baselines on AIME and GAIA benchmarks.
-
AutoResearchClaw: Self-Reinforcing Autonomous Research with Human-AI Collaboration
AutoResearchClaw introduces a multi-agent research pipeline with debate, self-healing, verifiable outputs, human collaboration modes, and cross-run evolution that outperforms AI Scientist v2 by 54.7% on ARC-Bench.
-
Sequential Consensus for Multi-Agent LLM Debates: A Wald-SPRT compute governor with calibration-based failure detection
Adapts SPRT as a compute governor for multi-agent LLM debates using Beta-modeled consensus scores from an LLM judge, yielding 3.7x call reduction on GSM8K at -2pp accuracy versus fixed rounds.
-
Robust Multi-Agent LLMs under Byzantine Faults
SAC is a decentralized iterative filter-and-refine protocol that achieves (F+1)-robustness in LLM multi-agent systems, suppressing Byzantine influence and improving performance on reasoning benchmarks where prior methods fail.
-
When Independent Sampling Outperforms Agentic Reasoning
On Codeforces problems, independent k-shot sampling achieves better accuracy-cost and accuracy-query tradeoffs than agentic reasoning, even with prompt caching.
-
12 Angry AI Agents: Evaluating Multi-Agent LLM Decision-Making Through Cinematic Jury Deliberation
Twelve LLM agents in a 12 Angry Men jury setup almost always end in hung juries due to anchoring, with Llama-4-Scout showing more vote changes than GPT-4o, suggesting RLHF alignment intensity limits deliberative flexibility.
-
Mesh Memory Protocol: Semantic Infrastructure for Multi-Agent LLM Systems
MMP defines a seven-field CMB schema, role-based SVAF evaluation, content-hash lineage, and remix storage to enable traceable cross-session collaboration among autonomous LLM agents.
-
Learning from AVA: Early Lessons from a Curated and Trustworthy Generative AI for Policy and Development Research
AVA is a specialized GenAI platform for development policy research that provides verifiable syntheses from World Bank reports and is associated with 2.4-3.9 hours of weekly time savings in a large-scale user evaluation.
-
Adapt to Thrive! Adaptive Power-Mean Policy Optimization for Improved LLM Reasoning
APMPO boosts average Pass@1 scores on math reasoning benchmarks by 3 points over GRPO by using an adaptive power-mean policy objective and feedback-driven clipping bounds in RLVR training.
-
Free Energy-Driven Reinforcement Learning with Adaptive Advantage Shaping for Unsupervised Reasoning in LLMs
FREIA applies free energy principles and adaptive advantage shaping to unsupervised RL, outperforming baselines by 0.5-3.5 Pass@1 points on math reasoning with a 1.5B model.
-
Act or Escalate? Evaluating Escalation Behavior in Automation with Language Models
Language models display model-specific escalation thresholds in uncertain decisions that are not explained by scale or architecture, and supervised fine-tuning on explicit uncertainty reasoning produces robust, generalizable policies.
-
ZoFia: Zero-Shot Fake News Detection with Entity-Guided Retrieval and Multi-LLM Interaction
ZoFia is a zero-shot fake news detection framework that uses hierarchical entity salience retrieval followed by multi-LLM adversarial debate to improve robustness over single-model approaches.
-
How Far Are We from Generating Missing Modalities with Foundation Models?
Evaluates 42 variants of foundation models across three formalized paradigms for missing modality reconstruction, identifies shortfalls in semantic extraction and validation, and introduces an agentic framework that reduces FID by at least 14% for images and MER by at least 10% for text.
-
Voting Protocols as Coordination Mechanisms for Role-Constrained Multi-Agent Tutoring Systems
Simulations of 1200 interactions show that voting protocol choice and agent deliberation change which response is selected by role-constrained pedagogical agents on SciQ and HumanEval benchmarks, producing distinct coordination patterns and some simulated learning gains.
-
SPOQ: Specialist Orchestrated Queuing for Multi-Agent Software Engineering
SPOQ is a multi-agent orchestration approach using wave-based topological dispatch, dual validation gates, and Human-as-an-Agent integration that reports large gains in speed, planning quality, defect reduction, and test pass rates across experiments and a large repository study.