arXiv subjects

Zhiwei Liu

Publications and source records attributed to Zhiwei Liu.

At least 19 recordsLinked to original sources

arXiv2026

Herculean: An Agentic Benchmark for Financial Intelligence

As AI agents advance, the central question is no longer whether they can solve isolated financial tasks, but whether their capabilities generalize across complete professional workflows. Existing benchmarks cannot answer this question because they evaluate either static competencies or agents within a single interaction structure. We introduce Herculean, a benchmark of end-to-end agentic financial intelligence that crosses agent architecture, backbone model, and four structurally distinct workflows: Trading, Hedging, Market Insights, and Auditing. Each workflow is implemented as an MCP skill that preserves its native state, actions, constraints, and outcome criteria, allowing intact agent systems to be evaluated under matched external conditions. Across 30 architecture--backbone configurations, we find that agent capability is not globally ordered. Workflow rankings are weakly aligned, changing the architecture frequently reverses performance under the same backbone, and each architecture concentrates failure in different forms of workflow control. These results establish that agent performance is an interaction between architecture and workflow demand, rather than an intrinsic property of the underlying model. Consequently, success on a single financial workflow provides little evidence of dependable, general financial intelligence.

cs.AI↗

arXiv2026

EGI-BENCH: An Enterprise General Intelligence Benchmark Suite

Enterprise adoption of LLMs in agentic workflows is gated by evaluations that span a wide arc of abilities: CRM operations, heterogeneous-document retrieval, personal data understanding, multi-turn planning, tool use, function calling, long-horizon research, and long-context software engineering. These evaluations live today in separate repositories with divergent runners, dependencies, and scoring conventions, which makes cross-benchmark comparison difficult and obscures what a single aggregate number actually represents. We release EGI-BENCH, a unified LLM agent benchmark suite that consolidates eight enterprise facing agent benchmarks (CRMARENA, HERB, PERSONABENCH, USERBENCH, MCP-UNIVERSE, MCPEVAL, BENCHFORCE, LOCOBENCH) behind a single command-line interface with two runtime-budgeted execution modes. On top of the suite we define the Enterprise General Intelligence (EGI) Score, a six-dimensional framework (M1-M6) whose buckets are populated by fine-grained sub-metrics routed from each benchmark rather than by benchmark-level averages. We describe the suite, the sub-metric routing, and the quick/deep budgets, and report results for fourteen frontier and cost-efficient LLMs spanning the GPT, Claude, and Gemini families under quick mode. Top-line EGI ranges from 39.6 (GPT-4o mini) to 65.6 (gpt-5.5), and no single model dominates every capability axis: gpt-5.5 leads task outcome, planning, and retrieval grounding; Claude Sonnet 4.5 leads tool actioning and safety/abstention; and Claude Haiku 4.5 leads engineering quality. Code is available at https://github.com/SalesforceAIResearch/egi-bench.

cs.AI↗

arXiv2026

CHI-Bench: Can AI Agents Automate End-to-End, Long-Horizon, Policy-Rich Healthcare Workflows?

End-to-end automation of realistic healthcare operations stresses three capabilities underrepresented in current benchmarks: policy density, decisions must be grounded in a large library of medical, insurance, and operational rules; Multi-role composition: a single task requires the agent to play multiple roles with handoffs; and multilateral interaction: intermediate workflow steps are multi-turn dialogs, such as peer-to-peer review and patient outreach. We introduce $χ$-Bench, a benchmark of long-horizon healthcare workflows across three domains: provider prior authorization, payer utilization management, and care management. Each task hands the agent a clinical case in a high-fidelity simulator of 20 healthcare apps exposed via 87 MCP tools, which it must drive to a terminal status through tool calls and writing the role's artifacts, guided by a 1,290+ document managed-care operations handbook skill. Across 30 agent harness/models configurations, the best agent resolves only 28.0% of tasks, no agent clears 20% on strict pass^3, and executing all tasks in a single session slumps the performance to 3.8%. These results raise the hypothesis that similar gaps are likely to surface in other policy-dense, role-composed, irreversible enterprise domains.

cs.CL↗

arXiv2026

Beyond Math and Code: Lightweight Corpus-Grounded Process Rewards for Factual Question Answering

Process supervision during reinforcement learning (RL) post-training matters in factual question answering (QA) because responses receiving positive response-level rewards can still contain sentence-level factual errors. Unlike math and code, factual QA lacks inexpensive programmatic checks, making reward computation a bottleneck. Existing methods rely on neural verifiers to score individual sentences, requiring extensive model inference as RL repeats these checks across many sampled responses at every update. We therefore introduce CorVer (Corpus Verify), a lightweight training-time approach to process supervision that derives sentence-level rewards from corpus co-occurrence statistics. A 0.5B extractor identifies subject-object pairs, indexed corpus queries supply their co-occurrence counts, and the resulting sentence rewards are assigned to the corresponding tokens for RL. On Qwen3-4B and Qwen3-8B, CorVer reduces mean complete training time by 5.5-10.4 times relative to the four factuality-RL baselines. Across the four models evaluated against these baselines, CorVer achieves the highest accuracy in 17 of 20 model-benchmark settings. CorVer outperforms the unmodified models in all 30 standard factual-QA settings (six models from three families across five benchmarks) and remains effective on two additional multi-hop QA datasets.

cs.CL↗

arXiv2026

TimeLitmus: A Diagnostic Benchmark for Cross-Modal Understanding and Explanation Faithfulness in Event-Conditioned Time-Series Prediction

Large language models (LLMs) are increasingly used to make predictions from numerical time-series histories and textual events. Yet accuracy alone cannot reveal whether correct answers reflect effective integration of the two inputs or instead arise from event polarity, unimodal priors, or superficial cues. Likewise, plausible explanations may rationalize predictions without faithfully reflecting the evidence that drives model behavior. We introduce TimeLitmus, a diagnostic benchmark for cross-modal understanding and explanation faithfulness in event-conditioned time-series prediction. TimeLitmus contains 4,856 evaluation records across Finance and Traffic, combining natural prediction with controlled counterfactual and contrastive interventions, explanation-targeted faithfulness tests, and systematic shortcut controls. Across ten representative LLMs, standard prediction accuracy substantially overstates reliable cross-modal understanding: Hard Paired Contrast (HPC) pair correctness peaks at only 19.2% in Finance and 11.7% in Traffic, and all ten models show lower-than-expected consistency on Finance series-side controls. Models often recognize scenario relations explicitly yet fail to apply them during independent prediction. Explanation faithfulness shows a similar gap: in Traffic, most models cite the manipulated temporal factor in over 90% of cases, while behavioral support remains below 22%. Human annotators outperform LLMs on matched controlled and hard-pair diagnostics, confirming that these distinctions are recoverable from the inputs. Natural-only adaptation yields selective gains in evidence selection and input sensitivity, but not consistent gains in controlled or hard-pair behavior. The benchmark, evaluation suite, and supervised adaptation data will be released publicly.

cs.AI↗

arXiv2026

Critical-State RL: Diagnosing Trainable States for Multi-Turn Tool Use

Multi-turn tool-use failures can hinge on a single model call, yet reward variation alone does not reveal which call would benefit from training. When rewards depend on later interactions, their variation can reflect downstream randomness rather than differences between the current actions. We introduce Critical-State RL to identify trainable states in multi-turn interactions. Given task-defined candidate calls and local rewards, the method assesses whether each reward captures the action's effect on task success and whether improvement over a reference policy is possible. It then uses nested sampling to separate action-dependent reward variation from continuation noise and optimizes the policy at the selected states using contextual-bandit training. Experiments on the Berkeley Function Calling Leaderboard (BFCL) v4 compare training at diagnostic-selected states with training at alternative states. For missing-function tasks, the diagnostic selects the response after the tool becomes available; for missing-argument tasks, it selects the response before the missing argument is supplied. Training the selected responses improves performance, including about 14 percentage points on the missing-function task, while training the alternatives leaves performance flat or worse. We further apply the recipe across models and tasks, including logged repeat-call avoidance and memory management.

cs.LG↗

arXiv2026

HalluTracer: Pre-Decoding Truthfulness Prediction via Depth-Averaged Probe-Logit

Internal-state probes enable truthfulness prediction before a large language model generates an answer. When detectors change both the layers they read and the rules used to combine them, the source of improved prediction becomes difficult to identify. We separate these choices and find that retaining more layers improves prediction even under fixed equal weighting. An exact Fisher-ratio decomposition explains why the additional benefit of linear reweighting is limited on these probe scores: information inlayer-wise differences largely overlaps with that captured by the depth mean. Estimating additional weights can then offset this small benefit when the data used to fit them are limited. These findings motivate our proposed method HalluTracer, which averages layer-wise probe logits to predict truthfulness before decoding. Across six models and four benchmarks, including TruthfulQA, HalluTracer achieves the highest area under the receiver operating characteristic curve (AUROC) in 23 of 24 model--benchmark pairs among the compared methods. The results support using evidence from across the network without requiring a correspondingly more flexible aggregation rule, clarifying the distinct roles of layer selection and weighting in pre-decoding detection.

cs.CL↗

arXiv2026

Mixed moment of $GL(2)$ and $GL(3)$ $L$-functions in the level aspect

In this paper, we establish an asymptotic formula of mixed moments of $L(1/2,f)$ and $L(1/2,\mathrm{sym}^2 f)$ for cusp forms $f$ of weight $k$, prime level $q$ and real primitive nebentypus $χ$ modulo $q$ in the level aspect with a power-saving error term. As an application, we deduce the simultaneous non-vanishing of $L(1/2, f)$ and $L(1/2,\mathrm{sym}^2 f)$ for sufficiently large prime levels.

math.NT↗

arXiv2026

Prediction-Market Seed Capital Recovery from Noise-Dominant Flow

Automated prediction markets require sponsors to prefund liquidity before observing order flow, creating a financing challenge at launch. We study whether nonnegative charges conditioned on observable payoff direction can improve recovery of this prefunded capital while limiting their effect on informed participation. We develop Seed Capital Flow (SCF), a direction-conditioned levy, in a stylized binary cost-function market with informed and liquidity-motivated traders. When order composition differs across directions, SCF concentrates the permitted fee burden on the direction with relatively more liquidity-motivated flow, whereas a uniform fee spreads it across both directions. Under a sufficiently tight common retention constraint, this allocation yields higher expected recovery capacity and can make additional liquidity choices financially viable. Synthetic numerical audits examine robustness to alternative flow patterns, stochastic arrivals, and label misspecification. The results characterize a mechanism-design tradeoff rather than an empirical prediction: the market remains prefunded, recovery is expected rather than guaranteed, and the analysis is limited to an opening-cohort setting.

cs.CE↗

arXiv2026

One AI Signal, Many Human Judgments: A Bayesian Cascade Analysis of AI-based Credibility Indicators in Online Information Spread

Social media platforms increasingly use AI-based credibility indicators to help users judge misinformation. Unlike individual human-AI decision-making, these indicators are embedded in information spread: users see both an AI prediction and earlier judgments shaped by the same AI, and their own judgments may then enter the public history. Yet how to analytically characterize this process remains under-explored. We therefore introduce a social-learning lens for this setting by extending the classical Bayesian cascade model with the AI indicator as a shared public signal. The resulting Gateway condition compares the evidence from the AI prediction with users' private impressions. Through this view, we show that AI changes what public history means. Crowd agreement may reflect accumulated independent human evidence, or repeated dependence on the same AI prediction. This creates a preservation-correction trade-off: stronger reliance on AI can preserve correct predictions, but can also lock in incorrect ones by blocking corrective private impressions. We calibrate the model using human-subject data on news veracity judgments. Although the AI outperforms human users, the average user weights it below her own impression but above several peer judgments, while individual users vary from discounting the AI to relying on it enough to cascade. Simulations show that over-reliance on a weak AI is especially harmful, and that diversifying AI signals across users can better keep the crowd informative. We conclude with implications for understanding human-AI interaction in information spread and designing misinformation interventions.

cs.HC↗

arXiv2026

MiraMind: Benchmarking Reliable Mental Health Reasoning beyond Answer Accuracy

Mental-health reasoning with large language models (LLMs) is an evidence-constrained judgment problem: models must transform limited, subjective, and often ambiguous evidence into interpretations, decisions, or claims whose specificity, certainty, severity, and actionability remain warranted. Existing benchmarks mainly evaluate specific clinical roles or final answers, leaving the reliability of explicit reasoning trajectories under-specified. We introduce MiraMind, a unified benchmark spanning six task families and 13 datasets across appraisal, diagnosis, intervention, abstraction, and verification. MiraMind evaluates both task outcomes and the trajectories connecting evidence to judgment through usability, logical structure, and informational contribution. Evaluating 20 LLMs reveals a restraint gap, in which the specificity or certainty of model judgments exceeds what limited evidence supports. We further train Mindora, an 8B model that targets evidence-to-judgment transitions through hard-case supervision, structured trajectory rewriting, and consistency-aware optimization. Mindora achieves the best average rank on MiraMind, improves over its backbone across all six task families, and produces more balanced reasoning trajectories. These results show that MiraMind can expose shared restraint failures and evaluate whether targeted post-training improves evidence-constrained mental-health reasoning.

cs.CL↗

arXiv2026

NTDH: Complex Reasoning for Comprehensive Affective Analysis

Comprehensive affective analysis is challenging for two reasons: it spans heterogeneous prediction tasks with continuous, ordinal, and multi-label outputs, and affective meaning is context-dependent, requiring conflicting cues to be reconciled rather than mapped directly to labels. Existing methods learn this mapping directly and do not model the reconciliation explicitly. We recast the task as a complex-reasoning problem, which yields one output interface across heterogeneous label spaces and a trajectory over which a verifiable reward can be optimised; to our knowledge, this is the first such treatment covering both sentiment and emotion. The obstacle is on the data side: affective reasoning traces must be synthesised, and generic synthesis is misaligned with the targets, tolerances, and phenomena of affect, and discards or leaks its failure cases. We propose NTDH, which addresses these four failures. Naturalisation sets the training answer to the gold label, so it is correct by construction. A Tolerance-aware gate checks each answer against the task's own scoring margin. Domain-aware strategies refine the reasoning using ideas from affective science. Directional Hints report only the type and direction of an error, without exposing the target. We train Qwen3-8B with SFT and then GRPO under the same tolerance used for verification (up to a more permissive construction gate on the multi-label subtask), and a component ablation quantifies the data-quality effect of each part. Using 16,302 training records, about 14x fewer than comparable instruction-tuned systems, the final policy improves over its SFT checkpoint on five of six official-test metrics and achieves the strongest EI-reg result among the compared systems, at a Pearson correlation of 0.862.

cs.CL↗

arXiv2026

Agentic Commerce World: An Auditable and Verifiable Environment for Vibe Commerce

In vibe coding, people describe software in natural language and delegate implementation to AI agents. By analogy, vibe commerce allows people to express buying or selling goals in natural language and delegate the corresponding tasks to agents. Commerce, however, requires independently controlled Buyer and Merchant agents to interact in a shared market while preserving their private objectives and distinct authority. We introduce Agentic Commerce World (ACWorld), an environment for evaluating such agents across ongoing transactions. Through its Vibe Commerce Protocol (VCP), ACWorld validates agent actions before updating shared transaction state and records the resulting interactions, making agent behavior auditable and evaluation reproducible. The ACWorld Benchmark contains a 200-task capability-coverage track and a 60-task large-catalog track that searches 785,022 transactable listings. Across ten models, mean scores range from 65.9% to 85.6% and from 56.1% to 91.4%, respectively. Our analysis shows that process-level evidence is necessary: final state alone can miss evaluated errors, incomplete trajectories still retain useful process signals, and large-catalog tasks expose bottlenecks across stages.

cs.AI↗

arXiv2026

ShiJianBench: From Dialogue to Decision for Long-Horizon Evaluation of Investment Advisors

Conversational investment advisors influence not only what users know, but also how they make subsequent decisions as market conditions evolve. Existing evaluations primarily assess response quality or observed outcomes, leaving the long-horizon pathway from advisor language to investor behavior difficult to audit. We introduce ShiJianBench, an offline framework for evaluating conversational investment advisors through matched investor trajectories under fixed historical market feedback. At its core is a multi-agent investor simulator with explicit evolving state variables, motive-driven deliberation, long-term memory, and dialogue-grounded updates. The simulator is calibrated against aggregate behavioral patterns from 7,199 real users, and advisor policies are evaluated using separate investor-side, service-side, and content-side metrics under a hard compliance gate. Experiments on Chinese fund-market traces from 2021 to 2026 identify a stable leading group of LLM advisors that combines substantially stronger personalized content with competitive investor-side trajectory outcomes. These results reveal a systematic distinction between producing a high-quality response and delivering an effective long-horizon intervention, motivating trajectory-aware evaluation of conversational advisors.

cs.CL↗

arXiv2026

Skillsets on the Chain: A Blockchain-based Zero-Trust Framework for Agentic AI Networking

Agentic AI networking (AgentNet) systems rely heavily on third-party skillset implementations and distributed multi-agent collaboration, yet they face major claim-to-capability inconsistencies and security vulnerabilities under trust-by-declaration assumptions. To bridge this gap, this paper proposes TrustAgentNet, a dual-tier blockchain-secured zero-trust framework. Specifically, a global Chain of Skillsets (CoS) governs the lifecycle of skillset metadata with protocols empowered by specialized agents to enforce off-chain auditing while maintaining lightweight on-chain cryptographic consensus. Furthermore, transient, task-oriented Chains of Collaboration (CoC) are dynamically established to enable trustless distributed multi-agent collaboration. Theoretical analysis of the three-way trade-off among security level, task performance, and resource overhead is provided and empirically validated. Experimental results on a hardware prototype demonstrate that compared with no-blockchain trust-by-default baselines, the zero-trust overhead of TrustAgentNet is dominated by off-chain inference, while the blockchain layer incurs minor ledger costs via the ledger-IPFS storage and on/off-chain integration design. Crucially, the proposed verification pipeline achieves a flawless 100% accuracy across 50 AI models, correctly validating 40 honest skillsets and intercepting 10 adversarial ones, and generalizes to non-AI domains with an 83.91% accuracy and a 0.85 F1-score across 1478 features from 171 ClawHub skills. Adversarial experiments further show that TrustAgentNet enables autonomous skillset self-recovery against various malicious attacks.

cs.CR↗

arXiv2026

Paying for Honesty Without Knowing the Truth: Reputation-Penalty Design for LLM Marketplace Agents

LLM agents increasingly act as autonomous merchants that write their own product listings, and under competitive pressure, they fabricate attributes to win sales. Even under instructions to be honest, they fabricate attributes in a majority of listings across models. A platform's obvious remedy---verifying each claim against the truth---is unavailable, because it observes only a noisy, biased complaint signal, never the ground truth. We design CARP, a reputation-penalty mechanism with a deadband that forgives complaint noise and a state-dependent severity that counters reputation-driven detection erosion. CARP requires no product-level ground truth and is robust to strategic gaming. CARP protects consumers by suppressing the sales volume of low-rated liars while sparing honest sellers. Paired with SPARC, it closes most of the consumer-welfare gap relative to a perfect-information oracle, without ever accessing the truth. It also achieves the best welfare of the policies we compare. We further show that this felt penalty becomes behaviorally binding through SPARC, a byte-clean code-gated reflection mechanism: LLM merchants fabricate when lying is free but restrain themselves when fabrication costs them sales, a self-interested response rather than compliance. We trace this distinction to penalty-gated self-correction reasoning, and observe the binding across models, with supporting confidence intervals.

cs.AI↗

arXiv2026

Personalized Recommendation Tool Learning via Autonomous Language Agents

Although large language models (LLMs) have recently gained traction in recommender systems due to their strong reasoning capabilities and extensive world knowledge, previous LLM-based agents suffer from hallucination and context-length limitations, and thus are not suitable for full-ranking recommendation tasks. To circumvent these limitations through architectural design rather than modifying the LLM itself, we propose an agent-based recommendation framework, memory-based $\textbf{P}$ersonalized $\textbf{R}$ecommendation $\textbf{T}$ool learning via autonomous language $\textbf{A}$gents (PRTA), in which an LLM acts as a central planner interacting with multiple recommendation models as tools. The LLM-based agent is responsible for high-level reasoning and personalized tool selection, while traditional recommendation models perform full-ranking scoring, leveraging their scalability in modeling behavioral patterns. To support personalized tool selection, we design reflection mechanisms that enable the agent to evaluate and compare tools for each user based on user profiles and candidate ranked lists. Extensive experiments across three public datasets demonstrate the superiority of \modelname over traditional recommendation and LLM-based baselines in improving full-ranking recommendation performance.

cs.IR↗

arXiv2026

MatPhaseBench: A Semantics-Guided Benchmark for Materials Phase Diagrams Understanding

Materials phase diagrams are a core knowledge representation in materials science, encoding temperature,composition, phase stability, and phase transformation pathways, with their full understanding requiring thermodynamic mechanism analysis and scientific reasoning. Although VLMs have shown promise in scientific image understanding, their systematic evaluation on such logically complex images demanding deep mechanistic interpretation remains limited, and phase diagrams provide a challenging testbed for this purpose. We introduce MatPhaseBench, a high-quality, high-reliability benchmark for complex scientific image understanding, focused on materials phase diagrams. MatPhaseBench is constructed from 3681 papers in classical materials science journals, from which 200 high-quality diagram-text pairs were selected, covering 189 material systems and 70 elements. The benchmark has three key features: (1)targeting complex scientific image understanding-it moves beyond simple objective tests to open-ended tasks requiring deep comprehension; (2)comprehensive image-text alignment-semantic information associated with images is fully preserved during literature mining and matching; (3) high-quality human-supervised text acquisition-all descriptions undergo strict manual validation. Experimental results show that current VLMs remain substantially behind expert-level understanding: they are largely limited to surface visual perception, lack deep reasoning grounded in thermodynamic mechanisms, have limited domain awareness and expert analytical experience, and perform poorly in distinguishing fine-grained differences in composite or multi-diagram settings. Overall, MatPhaseBench constitutes a challenging research-grade benchmark, providing a foundational platform for complex scientific image understanding, phase diagram analysis, and trustworthy multi-modal AI in science.

cs.CV↗