Pith. sign in

REVIEW 4 major objections 5 minor 92 references

Hallucination risk in multi-agent LLM systems can be scored before any agent speaks, by combining role–query misalignment with how errors would travel on the communication graph.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-30 19:37 UTC pith:KZXMIGVZ

load-bearing objection Solid pre-hoc MAS risk screen with real speed and localization numbers; the cascade story is only partly load-bearing because role–query mismatch already predicts a lot of the label. the 4 major comments →

arxiv 2607.26836 v1 pith:KZXMIGVZ submitted 2026-07-29 cs.CR

Before Agents Speak: Pre-hoc Failure Risk Inference in Multi-Agent Systems

Shi Lin , Chenpei Wang , Peng Qian , Dezhang Kong , Minghao Li , Yufeng Li , Xun Wang This is my paper
classification cs.CR
keywords multi-agent systemshallucination detectionpre-hoc risk inferenceNoisy-ORpropagation modelingLLM agentscommunication topology
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

When several language-model agents talk to each other, a single wrong answer can be copied and amplified until the whole team fails. Today’s defenses usually wait until answers already exist, which is too late once the damage has spread. This paper argues you can instead estimate each agent’s failure risk and the system-wide cascade risk before any message is sent. HalluProp scores an agent’s intrinsic risk from fine-grained mismatch between its role prompt and the user query, scores how likely a bad answer would jump along each communication link from role semantics plus simple topology features, then folds both into a differentiable Noisy-OR over the graph. On three benchmarks and four model backbones it localizes agents that later hallucinate with average AUROC 84.6%, runs in under a second (over 65× faster than post-hoc baselines), and, when used to prune or re-prompt the riskiest agent, yields post-intervention accuracy on par with or better than strong after-the-fact methods. A sympathetic reader cares because upstream screening can reshape roles and edges before unsafe content ever circulates.

Core claim

HalluProp shows that pre-execution, graph-level hallucination risk in multi-agent systems is predictable from two sources alone: intrinsic risk from query-conditioned competence versus susceptibility in each agent’s role embedding, and inter-agent propagation risk from semantic influence plus degree-based topology, combined by iterative differentiable Noisy-OR. That diagnosis localizes faulty agents at average AUROC 84.6% and supports interventions whose accuracy matches post-hoc baselines without observing any generation.

What carries the argument

Differentiable Noisy-OR fixed-point inference on the agent communication graph: each agent’s final risk is one minus the product of its intrinsic safety and the safety of every incoming link, iterated a few steps from the intrinsic prior so both node risks and edge transmission probabilities are learned end-to-end.

Load-bearing premise

Whether an agent will later output a wrong answer is largely readable from static role–query text misalignment and simple graph degrees, under the assumption that intrinsic failure and neighbor infections act as independent causes.

What would settle it

Hold roles, query, and topology fixed, run the multi-agent system, and check whether agents HalluProp ranked highest-risk before any message was sent are systematically not the ones whose outputs disagree with the reference answer—especially when failures come from decoding noise or multi-round drift rather than role–query mismatch.

Watch this falsifier — get emailed when new claim-graph text bears on it.

If this is right

  • MAS designers can rank and reassign or silence high-risk agents before the first round of dialogue.
  • Upstream screening can sit as a sub-second firewall in front of multi-round agent workflows without replacing runtime monitors.
  • Refining a risky agent’s prompt (making it cautious) can beat simply cutting it out of the graph while preserving collaboration.
  • The same Noisy-OR graph diagnosis applies across random, star, chain, and hierarchical topologies and across closed- and open-source backbones.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • If pre-hoc scores are reliable enough, role libraries and topology templates could be optimized offline for low cascade risk before deployment.
  • The competence–susceptibility split suggests adversarial or underspecified system prompts are a primary controllable lever, not only model choice.
  • Combining this upstream filter with a cheap post-hoc check on only the top-ranked agents could cut total verification cost while covering both static and drift-driven failures.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

4 major / 5 minor

Summary. The paper proposes HalluProp, a pre-hoc framework that estimates agent-level and system-level hallucination risk in LLM multi-agent systems before any inter-agent interaction. Intrinsic risk is obtained from query-conditioned competence/susceptibility decomposition of role and query embeddings (Eqs. 2–5); inter-agent propagation risk combines semantic compatibility with degree-based topology factors (Eqs. 6–7); these are aggregated by K-step differentiable Noisy-OR fixed-point iteration on the communication graph (Eqs. 8–9). Two-stage training uses node-level then graph-level binary cross-entropy (Eqs. 10–11). On MMLU, MATH, and MedQA with four LLM backbones, the method reports average AUROC 84.6%, Hit@1 83.9%, sub-second diagnosis (>65× vs post-hoc baselines), and post-intervention accuracy comparable to GUARDIAN (pruning 63.6%, refinement 67.6%). Ablations attribute large drops to removing intrinsic, topology, or propagation modules.

Significance. If the pre-hoc localization claim holds under realistic role and topology variation, the work offers a practically useful upstream screening layer for MAS reliability: risk scores can guide pruning or role refinement before costly multi-round execution, with clear efficiency gains over post-hoc detectors. The combination of a structured competence–susceptibility prior, topology-aware edge risks, and end-to-end differentiable Noisy-OR aggregation is a coherent modeling contribution relative to purely response-based or trajectory-based MAS failure attribution. Strengths include multi-backbone/multi-dataset evaluation, intervention utility against four post-hoc baselines, scalability and canonical-topology checks, and ablations that quantify module contributions. The appendix’s monotone-convergence and gradient analysis of Noisy-OR is a welcome formalization. The main significance risk is that predictive power may largely track static role–query mismatch (especially under deliberately noisy/mismatched roles) rather than true foresight of interaction-induced cascades; resolving that distinction would determine whether the result is a strong cascade-risk tool or primarily a role-fitness

major comments (4)
  1. [§4 Experimental Setup; Eqs. (10)–(11)] Experimental Setup and Eqs. 10–11: Ground-truth labels yi are defined by matching each agent’s output to the reference answer after three interaction rounds, and both y_node and y_graph are described under the same answer-match rule. How y_node (“before communication”) is obtained is not operationally specified—whether via isolated single-agent runs, a separate generation pass, or another proxy. Without a precise protocol, it is unclear that stage-1 supervision is truly pre-interaction, and the central “before agents speak” claim cannot be audited. Please define the exact generation and labeling procedure for y_node vs y_graph, including whether isolated runs use the same decoding settings and whether labels can leak multi-round context.
  2. [Appendix Role Construction; Table 4; Abstract] Role Construction (Appendix) deliberately injects domain-mismatched, underspecified, and adversarially perturbed roles. Under that design, high AUROC (Table 1) and a still-substantial 66.9 AUROC with propagation disabled (Table 4, “w/o Inter-Agent Propagation”) are consistent with ranking static role–query misfit rather than inferring cascading infection dynamics. The load-bearing narrative in the abstract and §1–3 (propagation-aware pre-hoc cascade diagnosis) therefore needs a control that holds: (i) all roles clean/well-specified, and/or (ii) fixed roles with only topology or query varied. Report AUROC/Hit and the intrinsic-vs-propagation ablation under those controls; if gains collapse when roles are clean, the cascade framing should be narrowed to role-risk screening plus a secondary graph aggregator.
  3. [Tables 1–2; Figure 3] Table 1–2 and Figure 3 report point estimates only (AUROC, Hit/Prec/Rec, post-intervention accuracy) with no standard errors, confidence intervals, or repeated seeds over topology/role sampling. Given randomized topologies (25–100% edge probability), role perturbations, and temperature 0.7 decoding for labels, variance is first-order for the average 84.6% AUROC and the pruning≈GUARDIAN parity claim (63.6% vs 63.5%). Please add multi-seed or bootstrap intervals and, where intervention gaps are small, a simple significance check so that “comparable to SOTA post-hoc” is statistically supported.
  4. [§3 System-Level Risk Inference; Eqs. (8)–(9); Appendix] Eqs. 8–9 and Appendix “Causal Independence”: system risk hi depends on neighbor hj under conditional independence of intrinsic and infection causes. Labels remain final answer-match after multi-round dialogue, so hi is trained as a correlate of eventual wrongness, not of an observed infection event or intermediate adoption of a neighbor’s error. To support the propagation mechanism as more than a flexible graph smoother, provide at least one diagnostic: e.g., correlation of learned tj→i with measured error adoption along edges, or accuracy of predicting who fails only after a known upstream hallucinator is introduced. Absent that, tone down causal language (“infects,” “cascading”) to predictive risk aggregation.
minor comments (5)
  1. [Figure 2; Eq. (9)] Figure 2 and Eq. (9): notation mixes h_i^{k+1}, h^{(k)}, and final h; a single consistent superscript convention would help.
  2. [Table 4] Table 4 caption says “Average AUROC” 84.8 while Table 1 average is 84.6; clarify whether the ablation subset or backbone differs.
  3. [§4 MAS Configuration] §4 MAS Configuration: “three rounds” of interaction is fixed; a brief sensitivity note on rounds would clarify label stability for post-hoc baselines and y_graph.
  4. [§5] Related Work is adequate but could more sharply separate single-agent uncertainty/consistency detectors from MAS topology methods when motivating pre-hoc vs post-hoc.
  5. [Abstract; §1] Minor typography/spacing issues appear in the abstract and early sections (e.g., missing spaces in “newsystemicrisk”, “cascading hallucinationsetting”); a pass for PDF text extraction artifacts would improve readability.

Circularity Check

0 steps flagged

No significant circularity: HalluProp is standard supervised risk prediction from static features, not a by-construction tautology.

full rationale

The derivation chain does not reduce outputs to inputs by definition. Intrinsic risk π_i is a learned function of frozen encoder embeddings of role p_i and query q after query-conditioned gating and competence/susceptibility decomposition (Eqs. 2–5); propagation t_j→i is a learned function of those vectors plus degree features on existing edges (Eqs. 6–7); system-level h is obtained by K steps of differentiable Noisy-OR fixed-point iteration from (π, T) (Eqs. 8–9), with a standard causal-independence justification in the appendix—not a redefinition of the evaluation labels. Training uses BCE against behavioral answer-match labels y_node / y_graph (Eqs. 10–11); evaluation reports AUROC/Hit@k against held-out agent failure labels and post-intervention accuracy. That is ordinary discriminative ML: the target is observed after generation, but h is not algebraically identical to y, nor is a fitted scalar renamed as a prediction of itself. Self-citations (e.g., Kong/Lin security surveys) are background and not load-bearing uniqueness theorems. Ablations and encoder swaps further show the pipeline is empirically falsifiable rather than forced by construction. Concerns that predictive power may largely track static role–query mismatch rather than multi-hop cascades are validity/overclaim issues, not circularity under the stated criteria.

Axiom & Free-Parameter Ledger

7 free parameters · 5 axioms · 2 invented entities

The central claim rests on supervised learning over encoder features and a probabilistic graphical assumption (independent causes via Noisy-OR), plus many fitted neural parameters. No new physical entity is postulated; competence/susceptibility are learned latent splits. Background math is standard; the domain assumptions about what predicts LLM agent failure are the fragile load-bearing pieces.

free parameters (7)
  • W_gate, b_g (query-conditioned gate)
    Learned d×d gate mapping role⊙query features to soft masks; fitted in stage-1 BCE on node labels.
  • w_ρ, w_κ, b_π (intrinsic risk head)
    Projection weights turning susceptibility/competence into πi; fitted to y_node.
  • W_sem, b_t (semantic propagation)
    Learned bilinear compatibility for ℓsem_j→i and bias into tj→i; fitted in stage-2.
  • MLP ϕ for topology factor γ_j,i
    Maps normalized in/out degree to γ≥1; hand-chosen structural featurization with learned weights.
  • λ_anc (stage-2 anchor weight)
    Balances graph-level BCE against retaining intrinsic BCE; chosen hyperparameter affecting T vs π tradeoff.
  • K (Noisy-OR fixed-point depth) = 3 (operating point)
    Iteration count; paper selects K≈3 after sensitivity curve plateaus (~84.6 AUROC).
  • Text encoder choice and frozen embeddings = bert-base-uncased default
    Default bert-base-uncased (and alternatives in Table 5); representation backbone is a free modeling choice that shifts AUROC by several points.
axioms (5)
  • domain assumption Causal/conditional independence of intrinsic failure and each neighbor infection event so that reliability multiplies in a Noisy-OR (Eq. 8–9, Appendix).
    Required for the closed product form of hi; correlated multi-agent errors would violate the factorization.
  • domain assumption Static role text and query embeddings (plus degree features) carry enough signal to predict post-generation incorrectness before any token is produced.
    Core pre-hoc premise in §3 Intra-Agent and Problem Formulation; if false, AUROC cannot generalize beyond the constructed role noise.
  • domain assumption Hallucination label yi ∈ {0,1} equals mismatch of agent output to reference answer after three interaction rounds.
    Experimental Setup operational definition; equates factual/answer error with the paper’s “hallucination” construct.
  • standard math Monotone bounded Noisy-OR iteration from h(0)=π converges to a usable fixed point on cyclic MAS graphs for small K.
    Appendix invokes monotone convergence on [0,1]^N; standard for damped product updates when t_j→i∈(0,1).
  • ad hoc to paper Propagation only along existing edges (tj→i=0 if edge absent) and γ amplifies high out/in degree senders/receivers.
    Modeling choice in Inter-Agent Propagation; structural prior not derived from data-generating theory of LLMs.
invented entities (2)
  • Competence vector κi and susceptibility vector ρi no independent evidence
    purpose: Decompose gated role embeddings into task-aligned vs residual risk-inducing features driving πi and propagation.
    Paper-defined latent split (Eq. 4); useful modeling construct without independent measurement outside the trained heads.
  • HalluProp systemic risk vector h via differentiable Noisy-OR on the communication graph no independent evidence
    purpose: Aggregate intrinsic and multi-hop propagation risk into per-agent pre-hoc diagnosis scores.
    Named framework output (Eq. 1, 8–9); evaluated only against the paper’s own labeled MAS runs.

pith-pipeline@v1.2.0-daily-grok45 · 22451 in / 3886 out tokens · 90124 ms · 2026-07-30T19:37:47.778799+00:00 · methodology

0 comments
read the original abstract

LLM-based multi-agent systems (MAS) have exhibited remarkable capabilities in collaborative reasoning and decision-making, yet their interconnected communications introduce new systemic risk: localized hallucinations can propagate along agent communication chain, amplify through interactions, and ultimately trigger cascading failures. Existing countermeasures predominantly follow a post-hoc paradigm, identifying failures only after unsafe behaviors emerge, by which time harmful effects may have already spread throughout the agent network. To tackle this problem, we investigate a complementary pre-hoc approach and propose HalluProp, a Propagation-aware Hallucination inference framework that estimates individual agent failures and emergent system-level hallucination risks before inter-agent interaction. First, we model intrinsic hallucination risks by identifying fine-grained semantic misalignment between agent roles and task queries. We then characterize inter-agent risk propagation by modeling both semantic influence and communication topology. Finally, we integrate these two risks via a differentiable Noisy-OR inference mechanism to derive a systemic diagnosis. Extensive experiments show that HalluProp accurately localizes faulty agents, achieving an average AUROC of 84.6%, while enabling sub-second diagnosis with over $65\times$ speedup over post-hoc methods. By facilitating early intervention through upstream screening, HalluProp effectively complements post-hoc methods, highlighting the potential of pre-hoc risk inference for building more reliable multi-agent systems.

Figures

Figures reproduced from arXiv: 2607.26836 by Chenpei Wang, Dezhang Kong, Minghao Li, Peng Qian, Shi Lin, Xun Wang, Yufeng Li.

Figure 1
Figure 1. Figure 1: Post-hoc methods detect hallucinations from ob [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: The overall structure of HalluProp comprises three components: (1) [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Figure 3: Robustness of HalluProp with different MAS con [PITH_FULL_IMAGE:figures/full_fig_p006_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Sensitivity to iteration depth K. 15%. This result highlights that the structural properties of the interaction graph (e.g., node degree and connectivity patterns) amplify risk diffusion, and removing topological informa￾tion weakens the identification of high-impact broadcasters, reducing the system-level hallucination diagnosis accuracy. Inter-Agent Propagation Module. To evaluate the necessity of modeli… view at source ↗
Figure 5
Figure 5. Figure 5: A comparative case study that illustrates hallucination propagation in MAS and how HalluProp effectively addresses it. [PITH_FULL_IMAGE:figures/full_fig_p012_5.png] view at source ↗

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

92 extracted references · 14 linked inside Pith

  1. [3]

    Findings of the

    Web fraud attacks against llm-driven multi-agent systems , author=. Findings of the

  2. [4]

    Findings of the Association for Computational Linguistics: ACL 2025 , pages=

    Direct Behavior Optimization: Unlocking the Potential of Lightweight LLMs , author=. Findings of the Association for Computational Linguistics: ACL 2025 , pages=

    2025

  3. [8]

    Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 2) , year=

    Measuring Mathematical Problem Solving With the MATH Dataset , author=. Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 2) , year=

  4. [9]

    Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , pages=

    Minicheck: Efficient fact-checking of llms on grounding documents , author=. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing , pages=

    2024

  5. [10]

    Advances in Neural Information Processing Systems , volume=

    Llm-check: Investigating detection of hallucinations in large language models , author=. Advances in Neural Information Processing Systems , volume=

  6. [11]

    A probabilistic framework for llm hallucination detection via belief tree propagation , author=. Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers) , pages=

    2025

  7. [13]

    ACM Transactions on Information Systems , volume=

    A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions , author=. ACM Transactions on Information Systems , volume=. 2025 , publisher=

    2025

  8. [14]

    2024 , booktitle=

    On the Resilience of Multi-Agent Systems with Malicious Agents , author=. 2024 , booktitle=

    2024

  9. [15]

    Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Don’t hallucinate, abstain: Identifying llm knowledge gaps via multi-llm collaboration , author=. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  10. [16]

    Findings of the Association for Computational Linguistics: EACL 2024 , pages=

    Improving Grounded Language Understanding in a Collaborative Environment by Interacting with Agents Through Help Feedback , author=. Findings of the Association for Computational Linguistics: EACL 2024 , pages=

    2024

  11. [17]

    Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages=

    Understanding the information propagation effects of communication topologies in llm-based multi-agent systems , author=. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing , pages=

    2025

  12. [18]

    Forty-first International Conference on Machine Learning , year=

    Improving factuality and reasoning in language models through multiagent debate , author=. Forty-first International Conference on Machine Learning , year=

  13. [19]

    The Twelfth International Conference on Learning Representations , year=

    MetaGPT: Meta programming for a multi-agent collaborative framework , author=. The Twelfth International Conference on Learning Representations , year=

  14. [20]

    Advances in Neural Information Processing Systems , volume=

    Swe-agent: Agent-computer interfaces enable automated software engineering , author=. Advances in Neural Information Processing Systems , volume=

  15. [21]

    Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Psysafe: A comprehensive framework for psychological-based attack, defense, and evaluation of multi-agent system safety , author=. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  16. [24]

    Proceedings of the 2023 conference on empirical methods in natural language processing , pages=

    Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models , author=. Proceedings of the 2023 conference on empirical methods in natural language processing , pages=

    2023

  17. [25]

    The 2023 Conference on Empirical Methods in Natural Language Processing , year=

    Halueval: A large-scale hallucination evaluation benchmark for large language models , author=. The 2023 Conference on Empirical Methods in Natural Language Processing , year=

    2023

  18. [26]

    Advances in Neural Information Processing Systems , volume=

    Alleviating hallucinations in large language models through multi-model contrastive decoding and dynamic hallucination detection , author=. Advances in Neural Information Processing Systems , volume=

  19. [27]

    International Conference on Learning Representations , year=

    Measuring Massive Multitask Language Understanding , author=. International Conference on Learning Representations , year=

  20. [28]

    arXiv preprint arXiv:2110.14168 , year=

    Training verifiers to solve math word problems , author=. arXiv preprint arXiv:2110.14168 , year=

    Pith/arXiv arXiv

  21. [29]

    Applied Sciences , volume=

    What disease does this patient have? a large-scale open domain question answering dataset from medical exams , author=. Applied Sciences , volume=. 2021 , publisher=

    2021

  22. [30]

    Advances in Neural Information Processing Systems , volume=

    Guardian: Safeguarding llm multi-agent collaborations with temporal graph modeling , author=. Advances in Neural Information Processing Systems , volume=

  23. [31]

    International Conference on Learning Representations , volume=

    Scaling large language model-based multi-agent collaboration , author=. International Conference on Learning Representations , volume=

  24. [32]

    , author=

    Large Language Model based Multi-Agents: A Survey of Progress and Challenges. , author=. 33rd International Joint Conference on Artificial Intelligence (IJCAI 2024) , year=

    2024

  25. [33]

    Findings of the Association for Computational Linguistics: EMNLP 2025 , year=

    Debunc: Mitigating hallucinations in large language model agent communication with uncertainty estimations , author=. Findings of the Association for Computational Linguistics: EMNLP 2025 , year=

    2025

  26. [34]

    Proceedings of the 41st International Conference on Machine Learning , pages=

    How language model hallucinations can snowball , author=. Proceedings of the 41st International Conference on Machine Learning , pages=

  27. [35]

    arXiv preprint arXiv:2310.13548 , year=

    Towards understanding sycophancy in language models , author=. arXiv preprint arXiv:2310.13548 , year=

    Pith/arXiv arXiv

  28. [36]

    Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

    Conformity in large language models , author=. Proceedings of the 63rd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers) , pages=

  29. [37]

    Computational Linguistics , pages=

    �� Siren’s Song in the AI Ocean: A Survey on Hallucination in Large Language Models , author=. Computational Linguistics , pages=. 2025 , publisher=

    2025

  30. [38]

    Nature , volume=

    Detecting hallucinations in large language models using semantic entropy , author=. Nature , volume=. 2024 , publisher=

    2024

  31. [39]

    arXiv preprint arXiv:2307.13528 , year=

    FacTool: Factuality Detection in Generative AI--A Tool Augmented Framework for Multi-Task and Multi-Domain Scenarios , author=. arXiv preprint arXiv:2307.13528 , year=

    Pith/arXiv arXiv

  32. [40]

    arXiv preprint arXiv:2502.11127 , year=

    G-safeguard: A topology-guided security lens and treatment on llm-based multi-agent systems , author=. arXiv preprint arXiv:2502.11127 , year=

    Pith/arXiv arXiv

  33. [41]

    Proceedings of the AAAI conference on artificial intelligence , volume=

    Graph of thoughts: Solving elaborate problems with large language models , author=. Proceedings of the AAAI conference on artificial intelligence , volume=

  34. [42]

    arXiv preprint arXiv:2310.02170 , year=

    Dynamic llm-agent network: An llm-agent collaboration framework with agent team optimization , author=. arXiv preprint arXiv:2310.02170 , year=

    Pith/arXiv arXiv

  35. [43]

    Findings of the Association for Computational Linguistics: EMNLP 2021 , pages=

    Retrieval Augmentation Reduces Hallucination in Conversation , author=. Findings of the Association for Computational Linguistics: EMNLP 2021 , pages=

    2021

  36. [44]

    Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing , pages=

    FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation , author=. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing , pages=

    2023

  37. [45]

    Findings of the Association for Computational Linguistics: EMNLP 2023 , pages=

    WikiChat: Stopping the hallucination of large language model chatbots by few-shot grounding on Wikipedia , author=. Findings of the Association for Computational Linguistics: EMNLP 2023 , pages=

    2023

  38. [46]

    arXiv preprint arXiv:2306.13781 , year=

    Retrieving supporting evidence for llms generated answers , author=. arXiv preprint arXiv:2306.13781 , year=

    Pith/arXiv arXiv

  39. [47]

    ICLR 2025 Workshop on Foundation Models in the Wild , year=

    KnowHalu: Hallucination Detection via Multi-Form Knowledge Based Factual Checking , author=. ICLR 2025 Workshop on Foundation Models in the Wild , year=

    2025

  40. [48]

    Thirty-seventh Conference on Neural Information Processing Systems Datasets and Benchmarks Track , year=

    FELM: Benchmarking Factuality Evaluation of Large Language Models , author=. Thirty-seventh Conference on Neural Information Processing Systems Datasets and Benchmarks Track , year=

  41. [49]

    Findings of the association for computational linguistics: ACL 2024 , pages=

    Chain-of-verification reduces hallucination in large language models , author=. Findings of the association for computational linguistics: ACL 2024 , pages=

    2024

  42. [50]

    Advances in neural information processing systems , volume=

    Chain-of-thought prompting elicits reasoning in large language models , author=. Advances in neural information processing systems , volume=

  43. [52]

    Findings of the Association for Computational Linguistics: ACL 2024 , pages=

    Fact-checking the output of large language models via token-level uncertainty quantification , author=. Findings of the Association for Computational Linguistics: ACL 2024 , pages=

    2024

  44. [53]

    The 2023 Conference on Empirical Methods in Natural Language Processing , year=

    The Internal State of an LLM Knows When It's Lying , author=. The 2023 Conference on Empirical Methods in Natural Language Processing , year=

    2023

  45. [54]

    The Twelfth International Conference on Learning Representations , year=

    INSIDE: LLMs' Internal States Retain the Power of Hallucination Detection , author=. The Twelfth International Conference on Learning Representations , year=

  46. [55]

    Findings of the Association for Computational Linguistics: ACL 2024 , pages=

    Unsupervised real-time hallucination detection based on the internal states of large language models , author=. Findings of the Association for Computational Linguistics: ACL 2024 , pages=

    2024

  47. [57]

    The Twelfth International Conference on Learning Representations , year=

    Self-contradictory Hallucinations of Large Language Models: Evaluation, Detection and Mitigation , author=. The Twelfth International Conference on Learning Representations , year=

  48. [58]

    L.; Almeida, D.; Altenschmidt, J.; Altman, S.; Anadkat, S.; et al

    Achiam, J.; Adler, S.; Agarwal, S.; Ahmad, L.; Akkaya, I.; Aleman, F. L.; Almeida, D.; Altenschmidt, J.; Altman, S.; Anadkat, S.; et al. 2023. Gpt-4 technical report. arXiv preprint arXiv:2303.08774

    Pith/arXiv arXiv 2023

  49. [59]

    Azaria, A.; and Mitchell, T. 2023. The Internal State of an LLM Knows When It's Lying. In The 2023 Conference on Empirical Methods in Natural Language Processing

    2023

  50. [60]

    Cao, Z.; Yang, Y.; and Zhao, H. 2023. Autohall: Automated hallucination dataset generation for large language models. arXiv preprint arXiv:2310.00259

    arXiv 2023

  51. [61]

    Chen, C.; Liu, K.; Chen, Z.; Gu, Y.; Wu, Y.; Tao, M.; Fu, Z.; and Ye, J. 2023 a . INSIDE: LLMs' Internal States Retain the Power of Hallucination Detection. In The Twelfth International Conference on Learning Representations

    2023

  52. [62]

    Chen, K.; Zhen, T.; Wang, H.; Liu, K.; Li, X.; Huo, J.; Yang, T.; Xu, J.; Dong, W.; and Gao, Y. 2025. MedSentry: Understanding and Mitigating Safety Risks in Medical LLM Multi-Agent Systems. arXiv preprint arXiv:2505.20824

    Pith/arXiv arXiv 2025

  53. [63]

    Chen, S.; Zhao, Y.; Zhang, J.; Chern, I.-C.; Gao, S.; Liu, P.; and He, J. 2023 b . FELM: Benchmarking Factuality Evaluation of Large Language Models. In Thirty-seventh Conference on Neural Information Processing Systems Datasets and Benchmarks Track

    2023

  54. [64]

    Dhuliawala, S.; Komeili, M.; Xu, J.; Raileanu, R.; Li, X.; Celikyilmaz, A.; and Weston, J. 2024. Chain-of-verification reduces hallucination in large language models. In Findings of the association for computational linguistics: ACL 2024, 3563--3578

    2024

  55. [65]

    B.; and Mordatch, I

    Du, Y.; Li, S.; Torralba, A.; Tenenbaum, J. B.; and Mordatch, I. 2023. Improving factuality and reasoning in language models through multiagent debate. In Forty-first International Conference on Machine Learning

    2023

  56. [66]

    Fadeeva, E.; Rubashevskii, A.; Shelmanov, A.; Petrakov, S.; Li, H.; Mubarak, H.; Tsymbalov, E.; Kuzmin, G.; Panchenko, A.; Baldwin, T.; et al. 2024. Fact-checking the output of large language models via token-level uncertainty quantification. In Findings of the Association for Computational Linguistics: ACL 2024, 9367--9385

    2024

  57. [67]

    Feng, S.; Shi, W.; Wang, Y.; Ding, W.; Balachandran, V.; and Tsvetkov, Y. 2024. Don’t hallucinate, abstain: Identifying llm knowledge gaps via multi-llm collaboration. In Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), 14664--14690

    2024

  58. [68]

    Grattafiori, A.; Dubey, A.; Jauhri, A.; Pandey, A.; Kadian, A.; Al-Dahle, A.; Letman, A.; Mathur, A.; Schelten, A.; Vaughan, A.; et al. 2024. The llama 3 herd of models. arXiv preprint arXiv:2407.21783

    Pith/arXiv arXiv 2024

  59. [69]

    Guo, T.; Chen, X.; Wang, Y.; Chang, R.; Pei, S.; Chawla, N.; Wiest, O.; and Zhang, X. 2024. Large Language Model based Multi-Agents: A Survey of Progress and Challenges. In 33rd International Joint Conference on Artificial Intelligence (IJCAI 2024). IJCAI; Cornell arxiv

    2024

  60. [70]

    Hendrycks, D.; Burns, C.; Kadavath, S.; Arora, A.; Basart, S.; Tang, E.; Song, D.; and Steinhardt, J. 2021. Measuring Mathematical Problem Solving With the MATH Dataset. In Thirty-fifth Conference on Neural Information Processing Systems Datasets and Benchmarks Track (Round 2)

    2021

  61. [71]

    Hong, S.; Zhuge, M.; Chen, J.; Zheng, X.; Cheng, Y.; Wang, J.; Zhang, C.; Wang, Z.; Yau, S. K. S.; Lin, Z.; et al. 2023. MetaGPT: Meta programming for a multi-agent collaborative framework. In The Twelfth International Conference on Learning Representations

    2023

  62. [72]

    Hou, B.; Zhang, Y.; Andreas, J.; and Chang, S. 2025. A probabilistic framework for llm hallucination detection via belief tree propagation. In Proceedings of the 2025 Conference of the Nations of the Americas Chapter of the Association for Computational Linguistics: Human Language Technologies (Volume 1: Long Papers), 3076--3099

    2025

  63. [73]

    R.; Li, J.-J.; et al

    Huang, J.-t.; Zhou, J.; Jin, T.; Zhou, X.; Chen, Z.; Wang, W.; Yuan, Y.; Sap, M.; Lyu, M. R.; Li, J.-J.; et al. 2024. On the Resilience of Multi-Agent Systems with Malicious Agents. In ICML

    2024

  64. [74]

    Huang, L.; Yu, W.; Ma, W.; Zhong, W.; Feng, Z.; Wang, H.; Chen, Q.; Peng, W.; Feng, X.; Qin, B.; et al. 2025. A survey on hallucination in large language models: Principles, taxonomy, challenges, and open questions. ACM Transactions on Information Systems, 43(2): 1--55

    2025

  65. [75]

    Jin, D.; Pan, E.; Oufattole, N.; Weng, W.-H.; Fang, H.; and Szolovits, P. 2021. What disease does this patient have? a large-scale open domain question answering dataset from medical exams. Applied Sciences, 11(14): 6421

    2021

  66. [76]

    A.; Yan, Y.; Zhang, Z.; Zhuang, Y.; Malhotra, M.; et al

    Kim, Y.; Gu, K.; Park, C.; Park, C.; Schmidgall, S.; Heydari, A. A.; Yan, Y.; Zhang, Z.; Zhuang, Y.; Malhotra, M.; et al. 2025. Towards a science of scaling agent systems. arXiv preprint arXiv:2512.08296

    Pith/arXiv arXiv 2025

  67. [77]

    Kong, D.; Lin, S.; Xu, Z.; Wang, Z.; Li, M.; Li, Y.; Zhang, Y.; Peng, H.; Chen, X.; Sha, Z.; et al. 2025. A survey of llm-driven ai agent communication: Protocols, security risks, and defense countermeasures. arXiv preprint arXiv:2506.19676

    arXiv 2025

  68. [78]

    Kong, D.; Peng, H.; Zhang, Y.; Zhao, L.; Xu, Z.; Lin, S.; Lin, C.; and Han, M. 2026. Web fraud attacks against llm-driven multi-agent systems. In Findings of the A ssociation for Computational Linguistics: ACL 2026"

    2026

  69. [79]

    Li, J.; Cheng, X.; Zhao, X.; Nie, J.-Y.; and Wen, J.-R. 2023. Halueval: A large-scale hallucination evaluation benchmark for large language models. In The 2023 Conference on Empirical Methods in Natural Language Processing

    2023

  70. [80]

    Lin, S.; Yang, H.; Li, R.; Wang, X.; Lin, C.; Xing, W.; and Han, M. 2024. Llms can be dangerous reasoners: Analyzing-based jailbreak attack on large language models. arXiv preprint arXiv:2407.16205

    Pith/arXiv arXiv 2024

  71. [81]

    Manakul, P.; Liusie, A.; and Gales, M. 2023. Selfcheckgpt: Zero-resource black-box hallucination detection for generative large language models. In Proceedings of the 2023 conference on empirical methods in natural language processing, 9004--9017

    2023

  72. [82]

    F.; Awadallah, A.; and Kiseleva, J

    Mehta, N.; Teruel, M.; Deng, X.; Sanz, S. F.; Awadallah, A.; and Kiseleva, J. 2024. Improving Grounded Language Understanding in a Collaborative Environment by Interacting with Agents Through Help Feedback. In Findings of the Association for Computational Linguistics: EACL 2024, 1306--1321

    2024

  73. [83]

    Min, S.; Krishna, K.; Lyu, X.; Lewis, M.; Yih, W.-t.; Koh, P.; Iyyer, M.; Zettlemoyer, L.; and Hajishirzi, H. 2023. FActScore: Fine-grained Atomic Evaluation of Factual Precision in Long Form Text Generation. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, 12076--12100

    2023

  74. [84]

    M \"u ndler, N.; He, J.; Jenko, S.; and Vechev, M. 2023. Self-contradictory Hallucinations of Large Language Models: Evaluation, Detection and Mitigation. In The Twelfth International Conference on Learning Representations

    2023

  75. [85]

    Qian, C.; Xie, Z.; Wang, Y.; Liu, W.; Zhu, K.; Xia, H.; Dang, Y.; Du, Z.; Chen, W.; Yang, C.; et al. 2025. Scaling large language model-based multi-agent collaboration. In International Conference on Learning Representations, volume 2025, 41488--41505

    2025

  76. [86]

    Semnani, S.; Yao, V.; Zhang, H.; and Lam, M. 2023. WikiChat: Stopping the hallucination of large language model chatbots by few-shot grounding on Wikipedia. In Findings of the Association for Computational Linguistics: EMNLP 2023, 2387--2413

    2023

  77. [87]

    Shen, X.; Liu, Y.; Dai, Y.; Wang, Y.; Miao, R.; Tan, Y.; Pan, S.; and Wang, X. 2025. Understanding the information propagation effects of communication topologies in llm-based multi-agent systems. In Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing, 12358--12372

    2025

  78. [88]

    S.; Saha, S.; Kattakinda, P.; and Feizi, S

    Sriramanan, G.; Bharti, S.; Sadasivan, V. S.; Saha, S.; Kattakinda, P.; and Feizi, S. 2024. Llm-check: Investigating detection of hallucinations in large language models. Advances in Neural Information Processing Systems, 37: 34188--34216

    2024

  79. [89]

    Su, W.; Wang, C.; Ai, Q.; Hu, Y.; Wu, Z.; Zhou, Y.; and Liu, Y. 2024. Unsupervised real-time hallucination detection based on the internal states of large language models. In Findings of the Association for Computational Linguistics: ACL 2024, 14379--14391

    2024

  80. [90]

    Tang, L.; Laban, P.; and Durrett, G. 2024. Minicheck: Efficient fact-checking of llms on grounding documents. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, 8818--8847

    2024

Showing first 80 references.