arXiv subjects

Hongning Wang

Publications and source records attributed to Hongning Wang.

At least 19 recordsLinked to original sources

arXiv2026

Toward User-Centered Agentic Recommendation: When Does Platform Competition Benefit Users?

Conventional recommender systems operate within platform-specific catalogs, limiting access to potentially valuable resources across platforms. We propose user-centered agentic recommendation, a paradigm in which user agents compare recommendations from competing platform agents, and formalize their interaction as a repeated game. Platforms strategically modify descriptions without changing intrinsic item value, while each user agent assesses platform credibility using feedback from consumed items only. Our analysis shows that expanded resource access need not improve user welfare: strategic misrepresentation and limited verification can offset the benefits of platform competition. We quantify when platform credibility assessment allows users to benefit from broader access despite strategic exaggeration. Our analysis identifies resource and reporting conditions that guarantee higher average welfare than the original platform with an interaction budget linear in platform count. Experiments on synthetic and Amazon-grounded markets show that credibility assessment can improve finite-horizon utility, while its benefits depend on market size and mechanism parameters.

cs.AI↗

arXiv2026

DiabAgent: Toward Agentic Physiological Control

As pretraining scales, medical language models have advanced rapidly in clinical question answering. Their ability to achieve effective physiological control through sustained intervention still lacks systematic quantitative evaluation. Type 1 diabetes provides a concrete setting: impaired insulin secretion creates dependence on exogenous insulin, with dosing continually adjusted using information about meals, exercise, illness, and glucose measurements. We introduce DiabAgent, a research suite for agentic physiological control comprising the Agentic Physiological Control (APC) framework, the DiabetSim simulator, and the SweetBench benchmark. APC controls glucose through medical-device tools, using ongoing observations to infer individual states, accumulate personal experience, and choose treatment actions. We build DiabetSim to simulate different patients, events, and device conditions for scalable virtual testing and training. Using DiabetSim, we construct SweetBench by instantiating 32 scenario designs as 212 glucose-control tasks. With Claude Opus 5.5 and a fixed decision interval of one hour, APC achieves a COGI of 73.5, compared with 71.2 for the open-source closed-loop baseline OpenAPS. Ablations further show that individual adaptation and contextual information improve glucose control. Combining high-frequency rule-based control with lower-frequency agent decisions shows potential for balancing control performance and computational cost. Our work provides a framework, simulator, and evaluation foundation for agents to participate in human physiological control.

cs.AI↗

arXiv2026

Look at Where You Think: Localization-Head Alignment for Fine-Grained Perception in Thinking VLMs

Thinking vision-language models reason for thousands of tokens, yet still miss small details in high-resolution images. Zooming methods pay for extra inference passes, and outcome rewards never tell the model where to look at. In Qwen3.5-4B, we find that the bottleneck is selection, not seeing: a few attention heads, which we call localization heads, pick the evidence out of thousands of visual tokens; ablating four of them costs 13.2 points more than ablating four random heads, and their focus fades as the reasoning grows. We therefore propose Localization-Head Alignment (LHA), which makes attention a training signal: whenever the reasoning states a visual fact, these heads learn to look at the focal region rather than the background, as an auxiliary loss in both SFT and RL training. On benchmarks of V*, HR-Bench 4K and TreeBench, LHA lifts Qwen3.5-4B from 74.1% to 76.6% accuracy, the best single-pass 4B result we evaluate, ahead of multi-pass zooming and released 7B models. Alignment alone adds 1.8 points over otherwise identical training while shortening reasoning by 20%, and the gains carry over to Qwen2.5-VL-3B.

cs.CV↗

arXiv2026

AI Cosmos: An Agentic Infra for Situated Learning

Large language models (LLMs) can improve task performance during learning, yet these gains do not necessarily reflect learners' ability to thereafter apply the knowledge and skills independently. We present AI Cosmos, an agentic infrastructure for situated learning that organizes learning around two complementary principles: scenario-based participation and cumulative assets of practice. AI Cosmos organizes learners and role-specialized agents within learning scenarios that integrate task-relevant knowledge and, when needed, executable workbenches. An asset-memory loop preserves the processes and outcomes of practice as reusable assets with explicit provenance and dependencies. We evaluate scenario-based interaction through two preregistered randomized controlled studies examining the acquisition of legal reasoning and programming competence, holding the same underlying LLM across conditions. AI Cosmos achieves greater learning gains than a conventional LLM in legal reasoning and higher mastery of unseen programming problems than an LLM-as-Teacher baseline.

cs.CL↗

arXiv2026

AOS-Bench: Can LLM Agents Manage Dynamic Workloads?

Coding agents are moving from handling one request at a time to serving as persistent assistants to which users hand a stream of work: a build, a data-processing job and a bug fix may all be pending on the same machine at once. Such an agent must decide when to take on new tasks, how to delegate and overlap them, and how to stay within the machine's CPU and memory. Existing benchmarks cannot measure this, because they evaluate each task in isolation. We introduce AOS-Bench, which evaluates an agent system as a workload manager: verified tasks arrive over time and must be delivered under shared CPU and memory limits, and the same tasks run one at a time serve as a per-system reference. Across five model--harness pairs, isolated task success does not carry over evenly. On heavy workloads every system loses 13 to 23 points. On medium workloads, systems span 11 points in isolated success but 26 points in stream success, and two systems 0.6 points apart in isolation deliver 54\% and 68\% of the same tasks. Execution-policy interventions trace much of this gap to task intake: serial dispatch leaves 73\% of heavy-workload tasks unclaimed. Yet more elaborate management does not necessarily help. A fixed concurrency of two comes within four points of the best agent-managed policy on both medium and heavy workloads, and under every policy median CPU use stays below 30\% of the provisioned cores while tasks wait. These results identify workload management as a distinct capability for agents, and AOS-Bench as a way to measure it.

cs.AI↗

arXiv2026

Can AI Agents Learn Their Way to the Top? Evaluating Heuristic Learning in a Long-Running Game Agent Competition

Adversarial games have driven advances from heuristic search to reinforcement learning, yet learning and adapting strategies from limited samples remain challenging. AI agents offer an alternative by turning game experience into revisions of executable policies. Building on heuristic learning (HL), we formalize Adversarial Heuristic Learning (AHL), a paradigm that uses AI agents as learning engines to refine game policies and supporting software while keeping model weights fixed. We introduce AAArena, a benchmark comprising 12 authentic adversarial games and 1,920 archived human programs, with an evaluation protocol modeled on real-world game competitions. Agents interpret rules, choose opponents, analyze replays, and revise game agents to achieve their highest ranking within fixed match and evaluation budgets. We evaluate \val{completedmodels} model and harness configurations: Opus5.5 with Claude Code earns 6 gold medals, while no evaluated configuration tops the remaining 6 human ladders. Performance is generally weaker in games with more complex rule specifications. Further experiments show that opponent selection and dense feedback support policy improvement, and that agents learn from both on-policy replays of their own matches and off-policy replays of other players' matches. These results highlight HL's potential in adversarial games and identify persistent challenges in game understanding, strategy implementation, and long-horizon policy development.

cs.AI↗

arXiv2026

ScopeIF: Improving Scope-Aware Precise Instruction-Following in Large Language Models via Graded Reward Modeling

Precise instruction-following is a fundamental ability of large language models (LLMs), requiring their outputs to strictly satisfy objective constraints in input instructions. In complex application scenarios, these constraints often possess diverse scopes that govern specific response segments rather than the entire output. However, existing optimization methods often neglect constraint scope during data construction and rely on binary per-constraint rewards, yielding limited data diversity and sparse supervision for complex constraints. To this end, we propose ScopeIF, a novel training framework for scope-aware precise instruction-following. We first introduce a unified schema that factorizes objective constraints into three decoupled dimensions: Scope, Target, and Range. Grounded in this schema, we construct ScopeInstruct, a large-scale instruction dataset with diverse scope-aware constraints, and combine tool-grounded verification with graded reward modeling to quantify the violation degree of each constraint, providing dense supervision for policy optimization. Extensive experiments demonstrate that ScopeIF consistently outperforms existing methods, particularly on complex scope-aware constraints, while preserving general capabilities. Notably, it enables optimized Qwen3-4B and 8B models to rival or surpass strong frontier models such as Gemini-2.5-Pro and DeepSeek-V3.2, establishing an effective paradigm for advancing scope-aware instruction-following. Our code and data are available at https://github.com/thu-coai/ScopeIF.

cs.CL↗

arXiv2026

MTAC-IFBench: Benchmarking Instruction-Following in Multi-Turn Agentic Coding

Recently, the rapid development of large language models (LLMs) has reshaped software engineering by enabling autonomous code agents that plan, execute, and utilize external tools iteratively to tackle complex tasks. Beyond achieving functional correctness, these agents must faithfully follow process instructions and constraints throughout the development lifecycle. However, existing benchmarks typically focus on final functional correctness or confine instruction-following evaluation to single-turn, general chat or simple code generation scenarios, leaving instruction-following in multi-turn agentic coding underexplored. To bridge this gap, we propose MTAC-IFBench, a comprehensive benchmark for this critical capability. It features multi-turn progressive software development instructions with diverse constraints spanning 6 primary and 18 secondary categories. With an average of 7.04 turns and 91.33 constraints per instance, it poses a rigorous challenge to current LLMs. To make the evaluation reliable, we construct a checklist for each constraint and functional requirement, and integrate verification scripts and judge agents to verify each checklist item. MTAC-IFBench identifies significant deficiencies in existing code agents in multi-turn instruction-following, with their performance degrading rapidly as the interaction session grows longer.

cs.CL↗

arXiv2026

The User Asks, Platforms Compete: How Agentic Recommendation Markets Take Shape

Online recommendation has traditionally taken place after a user enters a platform, which determines the candidate pool and the ranking shown to the user. LLM-based user agents enable a different recommendation process: a user specifies a need before choosing a platform, leaving platforms to compete for the user's attention, which we refer to as an agentic recommendation market. In our controlled LLM-based experiments across three product domains, we find this new setting of recommendation creates a tension between access and attention. Compared with traditional platform-centric recommendation, user-centric recommendation greatly expands the opportunity for relevant items to enter comparison; yet broader participation does not translate directly into effective exposure. Competition directly triggers platforms' strategic play: selectively positive explanations occupy 73--78% of first-ranked positions. When the user agent relates platforms' actions to subsequent user feedback, this share falls to 36--41%, while the chance of a user purchasing the relevant item increases. A user agent is therefore more than a ranker over a larger pool of candidates: its querying, ranking, and feedback mechanism governing who can compete, how scarce attention is allocated, and how earlier outcomes shape the evaluation of platforms directly affect user utility. Designing agentic recommendation therefore requires treating access, attention, and accountability as a joint mechanism design problem.

cs.AI↗

arXiv2026

You Live More Than Once: Towards Hierarchical Skill Meta-Evolving

Test-time skill evolving is regarded as a new paradigm for enhancing deployed agentic systems. Existing works mainly focus on hard-coded skill evolving strategies or parametric learning that rely on expensive parameter updates in the underlying LLMs. In this paper, we demonstrate that test-time refinement of the skill evolving framework itself is necessary for continuous improvement of the agent systems in different downstream scenarios, and lightweight algorithmic adaptation is feasible. Specifically, we propose HiSME, a lightweight hierarchical skill meta-evolving solution that jointly optimizes skills and the skill evolving strategy by learning meta-skills from agents' task execution traces. Experiments on mainstream agentic benchmarks show that meta-evolving can produce a higher-quality skill library than pure skill evolving and can derive diverse meta-skills for different scenarios, thereby facilitating future continual experience learning. Our code is temporarily public at https://anonymous.4open.science/r/HiSME-BD45.

cs.AI↗

arXiv2026

SynCred-Bench: Benchmarking Synthetic Credibility in AI-Generated Visual Misinformation

Recent generative models can now produce visual artifacts with realistic embedded text and layouts, creating a new misinformation threat: synthetic credibility. We introduce SYNCRED-Bench, a benchmark of 600 AI-generated misinformation images balanced across six credible-form categories and seven fine-grained circulation styles, together with FP450, a real-image negative set for measuring false positives. Extensive evaluation shows that existing systems remain unreliable: under a 5% false-positive-rate constraint, 15 MLLMs achieve only 10.5% true positive rate (TPR), open-source AIGC detectors achieve less than 5%, and commercial APIs reach 57.6%. Human annotators also struggled to identify synthetic credibility, reaching only 63% TPR. These findings establish synthetic credibility as a severe and underexplored visual misinformation challenge, and provide a benchmark for developing detectors that reason beyond superficial credibility cues.

cs.CV↗

arXiv2026

RUBAS: Rubric-Based Reinforcement Learning for Agent Safety

The evolution of LLMs into tool-enabled agents creates a new class of safety challenges associated with real-world execution rather than simple text generation. Existing alignment methods often rely on coarse refusal signals or static supervision, making it difficult to balance safety with useful tool execution across diverse agentic risks. We introduce RUBAS, a rubric-based reinforcement learning framework for agent safety. RUBAS decomposes agent behavior into four dimensions: tool-use safety, argument safety, response safety, and helpfulness. These structured rubrics provide fine-grained and interpretable rewards over complete agent trajectories, enabling reinforcement learning to optimize safe tool use while preserving task completion. Extensive experiments across multiple agent safety benchmarks and models show that RUBAS improves safety over standard alignment baselines, reduces tool-grounded hallucinations, and maintains competitive utility. Our results suggest that multi-dimensional rubric rewards provide an effective training signal for aligning LLM agents in safety-critical tool-use settings.

cs.LG↗

arXiv2026

Compile to Compress: Boosting Formal Theorem Provers by Compiler Outputs

Large language models (LLMs) have demonstrated significant potential in formal theorem proving, yet state-of-the-art performance often necessitates prohibitive test-time compute via massive roll-outs or extended context windows. In this work, we address this scalability bottleneck by exploiting an informative structure in formal verification: the observation that compilers map a vast space of diverse proof attempts to a compact set of structured failure modes. We introduce a learning-to-refine framework that leverages this compression to perform efficient learning and proof exploration. We perform tree search that corrects errors locally conditioned on explicit verifier feedback, thereby circumventing the costs associated with accumulating a long history of proof attempts. Extensive evaluations show that our method consistently amplifies the reasoning capabilities of base provers across varying scales. Notably, our approach achieves state-of-the-art performance on PutnamBench among publicly reported $\sim$8B and $\sim$32B parameter models under comparable test-time budgets, offering a scalable paradigm for next-generation verifier-guided reasoning.

cs.LG↗

arXiv2026

SkillEvolver: Skill Learning as a Meta-Skill

Agent skills today are static artifact: authored once -- by human curation or one-shot generation from parametric knowledge -- and then consumed unchanged, with no mechanism to improve from real use. We propose \textbf{SkillEvolver}, a lightweight, plug-and-play solution for online skill learning, in which a single meta-skill iteratively authors, deploys, and refines domain-specific skills. The learning target of SkillEvolver is the skill's prose and code, not model weights, so that the resulting artifact drops into any agent without retraining; and the meta-skill itself is just another skill, loaded through the same interface by any protocol-compliant CLI-agent. Unlike trace-distillation, the meta-skill refines only after deploying the learnt skill, such that the learning signal comes from failures another agent encounters while using it -- not from exploratory traces alone. Refinement iterations are governed by a fresh-agent overfit audit that catches possible leakage as well as deployed-skill-specific failures, including the silent-bypass mode in which a skill appears valid in content but is never invoked at runtime. On $83$ SkillsBench tasks spanning $15^{+}$ domains, SkillEvolver reaches $56.8\%$ accuracy versus $43.6\%$ for curated human skills and $29.9\%$ for the no-skill baseline; on three GPU kernel optimization tasks from KernelBench, it also raises mean speedup from $1.16$ to $1.51$ on average.

cs.AI↗

arXiv2026

LASA: Language-Agnostic Semantic Alignment at the Semantic Bottleneck for LLM Safety

Large language models (LLMs) often demonstrate strong safety performance in high-resource languages, yet exhibit severe vulnerabilities when queried in low-resource languages. We attribute this gap to a mismatch between language-agnostic semantic understanding ability and language-dominant safety alignment biased toward high-resource languages. Consistent with this hypothesis, we empirically identify the semantic bottleneck in LLMs, an intermediate layer in which the geometry of model representations is governed primarily by shared semantic content rather than language identity. Building on this observation, we propose Language-Agnostic Semantic Alignment (LASA), which anchors safety alignment directly in semantic bottlenecks. Experiments show that LASA substantially improves safety across all languages: average attack success rate (ASR) drops from 24.7% to 2.8% on LLaMA-3.1-8B-Instruct and remains around 3-4% across Qwen2.5 and Qwen3 Instruct models (7B-32B). Together, our analysis and method offer a representation-level perspective on LLM safety, suggesting that safety alignment requires anchoring safety understanding not in surface text, but in the model's language-agnostic semantic space.

cs.LG↗

arXiv2026

HoWToBench: Holistic Evaluation for LLM's Capability in Human-level Writing using Tree of Writing

Evaluating the writing capabilities of large language models (LLMs) remains a significant challenge due to the multidimensional nature of writing skills and the limitations of existing metrics. LLM's performance in thousand-words level and open-ended writing is inadequately assessed by traditional reference-based metrics or modern LLM-as-a-judge methods. We propose Tree-of-Writing (ToW), to resolve the implicit inconsistency often found when LLM-as-a-judge aggregates all sub-features in text evaluation. ToW incorporates a tree-structured workflow by explicitly modeling the aggregation weights of sub-features. We also present HowToBench, a large-scale Chinese writing benchmark encompassing 12 genres and 1302 instructions across three task categories: contextual completion, outline-guided writing, and open-ended generation. ToW successfully mitigates the biases, achieving a 0.93 Pearson correlation with human judgments. Furthermore, we detect that both overlap-based text generation metrics and popular LLM-as-a-judge practices are vulnerable to textual disturbances, while ToW is robust to them. We also uncover a negative correlation between input length and content-related scores in the Guide task, showcasing that it cannot be simply improved by input-side information piling.

cs.CL↗

arXiv2026

How Should We Enhance the Safety of Large Reasoning Models: An Empirical Study

Large Reasoning Models (LRMs) have achieved remarkable success on reasoning-intensive tasks such as mathematics and programming. However, their enhanced reasoning capabilities do not necessarily translate to improved safety performance-and in some cases, may even degrade it. This raises an important research question: how should we enhance the safety of LRMs? In this paper, we present a comprehensive empirical study on how to enhance the safety of LRMs through Supervised Fine-Tuning (SFT). Our investigation begins with an unexpected observation: directly distilling safe responses from DeepSeek-R1 fails to significantly enhance safety. We analyze this phenomenon and identify five key risky patterns that contribute to it. We then demonstrate that explicitly addressing these issues during the data distillation process can lead to substantial safety improvements. Next, we explore whether a long and complex reasoning process is necessary for achieving safety. Interestingly, we find that simply using short or template-based reasoning process can attain comparable safety performance. These findings prompt a deeper reflection on the role of reasoning in ensuring safety. Finally, we conduct a comprehensive ablation study to reveal the impact of different training configurations. Overall, we hope our empirical study could provide a more holistic picture on enhancing the safety of LRMs. The code and data used in our experiments are released in https://github.com/thu-coai/LRM-Safety-Study.

cs.CL↗

arXiv2026

IF-CRITIC: Towards a Fine-Grained LLM Critic for Instruction-Following Evaluation

Instruction-following is a fundamental ability of Large Language Models (LLMs), requiring their generated outputs to follow multiple constraints imposed in input instructions. Numerous studies have attempted to enhance this ability through preference optimization or reinforcement learning based on reward signals from LLM-as-a-Judge. However, existing evaluation models for instruction-following still possess many deficiencies, such as substantial costs and unreliable assessments. To this end, we propose IF-CRITIC, an LLM critic for fine-grained, efficient, and reliable instruction-following evaluation. We first develop a checklist generator to decompose instructions and generate constraint checklists. With the assistance of the checklists, we collect high-quality critique training data through a multi-stage critique filtering mechanism and employ a constraint-level preference optimization method to train IF-CRITIC. Extensive experiments show that the evaluation performance of IF-CRITIC can beat strong LLM-as-a-Judge baselines, including o4-mini and Gemini-3-Pro. With the reward signals provided by IF-CRITIC, LLMs can achieve substantial performance gains in instruction-following optimization under lower computational overhead compared to strong LLM critic baselines. Our code and model are available at https://github.com/thu-coai/IF-CRITIC.

cs.CL↗