dr-claw 1.1.2 → 1.1.4

package/server/data/news-results-arxiv.json

@@ -0,0 +1,460 @@
+{
+  "target_date": "2026-04-09",
+  "search_date": "2026-04-09",
+  "date_windows": {
+    "recent_30d": {
+      "start": "2026-03-10",
+      "end": "2026-04-09"
+    },
+    "past_year": {
+      "start": "2025-04-09",
+      "end": "2026-03-09"
+    }
+  },
+  "total_found": 200,
+  "total_filtered": 200,
+  "total_recent": 200,
+  "total_hot": 0,
+  "total_unique": 200,
+  "top_papers": [
+    {
+      "id": "http://arxiv.org/abs/2604.07223v1",
+      "title": "TraceSafe: A Systematic Assessment of LLM Guardrails on Multi-Step Tool-Calling Trajectories",
+      "summary": "As large language models (LLMs) evolve from static chatbots into autonomous agents, the primary vulnerability surface shifts from final outputs to intermediate execution traces. While safety guardrails are well-benchmarked for natural language responses, their efficacy remains largely unexplored within multi-step tool-use trajectories. To address this gap, we introduce TraceSafe-Bench, the first comprehensive benchmark specifically designed to assess mid-trajectory safety. It encompasses 12 risk categories, ranging from security threats (e.g., prompt injection, privacy leaks) to operational failures (e.g., hallucinations, interface inconsistencies), featuring over 1,000 unique execution instances. Our evaluation of 13 LLM-as-a-guard models and 7 specialized guardrails yields three critical findings: 1) Structural Bottleneck: Guardrail efficacy is driven more by structural data competence (e.g., JSON parsing) than semantic safety alignment. Performance correlates strongly with structured-to-text benchmarks ($ρ=0.79$) but shows near-zero correlation with standard jailbreak robustness. 2) Architecture over Scale: Model architecture influences risk detection performance more significantly than model size, with general-purpose LLMs consistently outperforming specialized safety guardrails in trajectory analysis. 3) Temporal Stability: Accuracy remains resilient across extended trajectories. Increased execution steps allow models to pivot from static tool definitions to dynamic execution behaviors, actually improving risk detection performance in later stages. Our findings suggest that securing agentic workflows requires jointly optimizing for structural reasoning and safety alignment to effectively mitigate mid-trajectory risks.",
+      "authors": [
+        "Yen-Shan Chen",
+        "Sian-Yao Huang",
+        "Cheng-Lin Yang",
+        "Yun-Nung Chen"
+      ],
+      "published": "2026-04-08T15:46:14Z",
+      "published_date": "2026-04-08 15:46:14+00:00",
+      "updated": "2026-04-08T15:46:14Z",
+      "categories": [
+        "cs.CR",
+        "cs.AI",
+        "cs.CL",
+        "cs.LG",
+        "cs.SE"
+      ],
+      "pdf_url": "https://arxiv.org/pdf/2604.07223v1",
+      "url": "http://arxiv.org/abs/2604.07223v1",
+      "source": "arxiv",
+      "scores": {
+        "relevance": 3.8,
+        "recency": 3.0,
+        "popularity": 2.3,
+        "quality": 2.3,
+        "recommendation": 10.13
+      },
+      "relevance_score": 3.8,
+      "recency_score": 3.0,
+      "popularity_score": 2.3,
+      "quality_score": 2.3,
+      "final_score": 10.13,
+      "abstract": "As large language models (LLMs) evolve from static chatbots into autonomous agents, the primary vulnerability surface shifts from final outputs to intermediate execution traces. While safety guardrails are well-benchmarked for natural language responses, their efficacy remains largely unexplored within multi-step tool-use trajectories. To address this gap, we introduce TraceSafe-Bench, the first comprehensive benchmark specifically designed to assess mid-trajectory safety. It encompasses 12 risk categories, ranging from security threats (e.g., prompt injection, privacy leaks) to operational failures (e.g., hallucinations, interface inconsistencies), featuring over 1,000 unique execution instances. Our evaluation of 13 LLM-as-a-guard models and 7 specialized guardrails yields three critical findings: 1) Structural Bottleneck: Guardrail efficacy is driven more by structural data competence (e.g., JSON parsing) than semantic safety alignment. Performance correlates strongly with structured-to-text benchmarks ($ρ=0.79$) but shows near-zero correlation with standard jailbreak robustness. 2) Architecture over Scale: Model architecture influences risk detection performance more significantly than model size, with general-purpose LLMs consistently outperforming specialized safety guardrails in trajectory analysis. 3) Temporal Stability: Accuracy remains resilient across extended trajectories. Increased execution steps allow models to pivot from static tool definitions to dynamic execution behaviors, actually improving risk detection performance in later stages. Our findings suggest that securing agentic workflows requires jointly optimizing for structural reasoning and safety alignment to effectively mitigate mid-trajectory risks.",
+      "matched_domain": "Large Language Models",
+      "matched_keywords": [
+        "large language model",
+        "LLM",
+        "cs.AI",
+        "cs.LG",
+        "cs.CL"
+      ],
+      "is_hot_paper": false
+    },
+    {
+      "id": "http://arxiv.org/abs/2604.07338v1",
+      "title": "Appear2Meaning: A Cross-Cultural Benchmark for Structured Cultural Metadata Inference from Images",
+      "summary": "Recent advances in vision-language models (VLMs) have improved image captioning for cultural heritage. However, inferring structured cultural metadata (e.g., creator, origin, period) from visual input remains underexplored. We introduce a multi-category, cross-cultural benchmark for this task and evaluate VLMs using an LLM-as-Judge framework that measures semantic alignment with reference annotations. To assess cultural reasoning, we report exact-match, partial-match, and attribute-level accuracy across cultural regions. Results show that models capture fragmented signals and exhibit substantial performance variation across cultures and metadata types, leading to inconsistent and weakly grounded predictions. These findings highlight the limitations of current VLMs in structured cultural metadata inference beyond visual perception.",
+      "authors": [
+        "Yuechen Jiang",
+        "Enze Zhang",
+        "Md Mohsinul Kabir",
+        "Qianqian Xie",
+        "Stavroula Golfomitsou",
+        "Konstantinos Arvanitis",
+        "Sophia Ananiadou"
+      ],
+      "published": "2026-04-08T17:53:26Z",
+      "published_date": "2026-04-08 17:53:26+00:00",
+      "updated": "2026-04-08T17:53:26Z",
+      "categories": [
+        "cs.CV",
+        "cs.CL",
+        "cs.MM"
+      ],
+      "pdf_url": "https://arxiv.org/pdf/2604.07338v1",
+      "url": "http://arxiv.org/abs/2604.07338v1",
+      "source": "arxiv",
+      "scores": {
+        "relevance": 3.6,
+        "recency": 3.0,
+        "popularity": 1.6,
+        "quality": 1.6,
+        "recommendation": 8.93
+      },
+      "relevance_score": 3.6,
+      "recency_score": 3.0,
+      "popularity_score": 1.6,
+      "quality_score": 1.6,
+      "final_score": 8.93,
+      "abstract": "Recent advances in vision-language models (VLMs) have improved image captioning for cultural heritage. However, inferring structured cultural metadata (e.g., creator, origin, period) from visual input remains underexplored. We introduce a multi-category, cross-cultural benchmark for this task and evaluate VLMs using an LLM-as-Judge framework that measures semantic alignment with reference annotations. To assess cultural reasoning, we report exact-match, partial-match, and attribute-level accuracy across cultural regions. Results show that models capture fragmented signals and exhibit substantial performance variation across cultures and metadata types, leading to inconsistent and weakly grounded predictions. These findings highlight the limitations of current VLMs in structured cultural metadata inference beyond visual perception.",
+      "matched_domain": "Multimodal",
+      "matched_keywords": [
+        "vision-language",
+        "visual",
+        "cs.CV",
+        "cs.MM",
+        "cs.CL"
+      ],
+      "is_hot_paper": false
+    },
+    {
+      "id": "http://arxiv.org/abs/2604.07274v1",
+      "title": "A Systematic Study of Retrieval Pipeline Design for Retrieval-Augmented Medical Question Answering",
+      "summary": "Large language models (LLMs) have demonstrated strong capabilities in medical question answering; however, purely parametric models often suffer from knowledge gaps and limited factual grounding. Retrieval-augmented generation (RAG) addresses this limitation by integrating external knowledge retrieval into the reasoning process. Despite increasing interest in RAG-based medical systems, the impact of individual retrieval components on performance remains insufficiently understood. This study presents a systematic evaluation of retrieval-augmented medical question answering using the MedQA USMLE benchmark and a structured textbook-based knowledge corpus. We analyze the interaction between language models, embedding models, retrieval strategies, query reformulation, and cross-encoder reranking within a unified experimental framework comprising forty configurations. Results show that retrieval augmentation significantly improves zero-shot medical question answering performance. The best-performing configuration was dense retrieval with query reformulation and reranking achieved 60.49% accuracy. Domain-specialized language models were also found to better utilize retrieved medical evidence than general-purpose models. The analysis further reveals a clear tradeoff between retrieval effectiveness and computational cost, with simpler dense retrieval configurations providing strong performance while maintaining higher throughput. All experiments were conducted on a single consumer-grade GPU, demonstrating that systematic evaluation of retrieval-augmented medical QA systems can be performed under modest computational resources.",
+      "authors": [
+        "Nusrat Sultana",
+        "Abdullah Muhammad Moosa",
+        "Kazi Afzalur Rahman",
+        "Sajal Chandra Banik"
+      ],
+      "published": "2026-04-08T16:37:22Z",
+      "published_date": "2026-04-08 16:37:22+00:00",
+      "updated": "2026-04-08T16:37:22Z",
+      "categories": [
+        "cs.CL",
+        "cs.AI",
+        "cs.LG"
+      ],
+      "pdf_url": "https://arxiv.org/pdf/2604.07274v1",
+      "url": "http://arxiv.org/abs/2604.07274v1",
+      "source": "arxiv",
+      "scores": {
+        "relevance": 3.6,
+        "recency": 3.0,
+        "popularity": 1.3,
+        "quality": 1.3,
+        "recommendation": 8.53
+      },
+      "relevance_score": 3.6,
+      "recency_score": 3.0,
+      "popularity_score": 1.3,
+      "quality_score": 1.3,
+      "final_score": 8.53,
+      "abstract": "Large language models (LLMs) have demonstrated strong capabilities in medical question answering; however, purely parametric models often suffer from knowledge gaps and limited factual grounding. Retrieval-augmented generation (RAG) addresses this limitation by integrating external knowledge retrieval into the reasoning process. Despite increasing interest in RAG-based medical systems, the impact of individual retrieval components on performance remains insufficiently understood. This study presents a systematic evaluation of retrieval-augmented medical question answering using the MedQA USMLE benchmark and a structured textbook-based knowledge corpus. We analyze the interaction between language models, embedding models, retrieval strategies, query reformulation, and cross-encoder reranking within a unified experimental framework comprising forty configurations. Results show that retrieval augmentation significantly improves zero-shot medical question answering performance. The best-performing configuration was dense retrieval with query reformulation and reranking achieved 60.49% accuracy. Domain-specialized language models were also found to better utilize retrieved medical evidence than general-purpose models. The analysis further reveals a clear tradeoff between retrieval effectiveness and computational cost, with simpler dense retrieval configurations providing strong performance while maintaining higher throughput. All experiments were conducted on a single consumer-grade GPU, demonstrating that systematic evaluation of retrieval-augmented medical QA systems can be performed under modest computational resources.",
+      "matched_domain": "Large Language Models",
+      "matched_keywords": [
+        "large language model",
+        "LLM",
+        "cs.AI",
+        "cs.LG",
+        "cs.CL"
+      ],
+      "is_hot_paper": false
+    },
+    {
+      "id": "http://arxiv.org/abs/2604.07147v1",
+      "title": "Dynamic Context Evolution for Scalable Synthetic Data Generation",
+      "summary": "Large language models produce repetitive output when prompted independently across many batches, a phenomenon we term cross-batch mode collapse: the progressive loss of output diversity when a language model is prompted repeatedly without access to its prior generations. Practitioners have long mitigated this with ad hoc deduplication and seed rotation, but no principled framework exists. We introduce Dynamic Context Evolution (DCE), comprising three mechanisms: (1) verbalized tail sampling (the model labels each idea with a guess about how obvious it is, and obvious ideas are discarded), which filters high-probability candidates via model self-assessment; (2) semantic memory, which maintains a persistent embedding index to reject near-duplicates across batches; and (3) adaptive prompt evolution, which reconstructs the generation prompt each batch using memory state and rotating diversity strategies. In experiments across three domains (sustainable packaging concepts, educational exam questions, and creative writing prompts) and two model families (gpt-5-mini and claude-haiku-4-5), a component ablation across 2-3 random seeds per method shows that DCE achieves 0.0 +/- 0.0% collapse versus 5.6 +/- 2.0% for naive prompting, while producing 17-18 HDBSCAN clusters per seed versus naive's volatile 2-17, indicating reliably richer conceptual structure. These results are validated with an independent embedding model (all-MiniLM-L6-v2) and hold across sensitivity sweeps of the VTS threshold tau and dedup threshold delta. Deduplication and prompt evolution are individually insufficient but jointly effective, at approximately $0.50 per 1,000 candidates using only standard API calls, with no fine-tuning or custom architectures required.",
+      "authors": [
+        "Ryan Lingo",
+        "Rajeev Chhajer"
+      ],
+      "published": "2026-04-08T14:38:11Z",
+      "published_date": "2026-04-08 14:38:11+00:00",
+      "updated": "2026-04-08T14:38:11Z",
+      "categories": [
+        "cs.CL",
+        "cs.AI",
+        "cs.LG"
+      ],
+      "pdf_url": "https://arxiv.org/pdf/2604.07147v1",
+      "url": "http://arxiv.org/abs/2604.07147v1",
+      "source": "arxiv",
+      "scores": {
+        "relevance": 3.3,
+        "recency": 3.0,
+        "popularity": 1.6,
+        "quality": 1.6,
+        "recommendation": 8.53
+      },
+      "relevance_score": 3.3,
+      "recency_score": 3.0,
+      "popularity_score": 1.6,
+      "quality_score": 1.6,
+      "final_score": 8.53,
+      "abstract": "Large language models produce repetitive output when prompted independently across many batches, a phenomenon we term cross-batch mode collapse: the progressive loss of output diversity when a language model is prompted repeatedly without access to its prior generations. Practitioners have long mitigated this with ad hoc deduplication and seed rotation, but no principled framework exists. We introduce Dynamic Context Evolution (DCE), comprising three mechanisms: (1) verbalized tail sampling (the model labels each idea with a guess about how obvious it is, and obvious ideas are discarded), which filters high-probability candidates via model self-assessment; (2) semantic memory, which maintains a persistent embedding index to reject near-duplicates across batches; and (3) adaptive prompt evolution, which reconstructs the generation prompt each batch using memory state and rotating diversity strategies. In experiments across three domains (sustainable packaging concepts, educational exam questions, and creative writing prompts) and two model families (gpt-5-mini and claude-haiku-4-5), a component ablation across 2-3 random seeds per method shows that DCE achieves 0.0 +/- 0.0% collapse versus 5.6 +/- 2.0% for naive prompting, while producing 17-18 HDBSCAN clusters per seed versus naive's volatile 2-17, indicating reliably richer conceptual structure. These results are validated with an independent embedding model (all-MiniLM-L6-v2) and hold across sensitivity sweeps of the VTS threshold tau and dedup threshold delta. Deduplication and prompt evolution are individually insufficient but jointly effective, at approximately $0.50 per 1,000 candidates using only standard API calls, with no fine-tuning or custom architectures required.",
+      "matched_domain": "Large Language Models",
+      "matched_keywords": [
+        "large language model",
+        "cs.AI",
+        "cs.LG",
+        "cs.CL"
+      ],
+      "is_hot_paper": false
+    },
+    {
+      "id": "http://arxiv.org/abs/2604.07028v1",
+      "title": "Strategic Persuasion with Trait-Conditioned Multi-Agent Systems for Iterative Legal Argumentation",
+      "summary": "Strategic interaction in adversarial domains such as law, diplomacy, and negotiation is mediated by language, yet most game-theoretic models abstract away the mechanisms of persuasion that operate through discourse. We present the Strategic Courtroom Framework, a multi-agent simulation environment in which prosecution and defense teams composed of trait-conditioned Large Language Model (LLM) agents engage in iterative, round-based legal argumentation. Agents are instantiated using nine interpretable traits organized into four archetypes, enabling systematic control over rhetorical style and strategic orientation.\n  We evaluate the framework across 10 synthetic legal cases and 84 three-trait team configurations, totaling over 7{,}000 simulated trials using DeepSeek-R1 and Gemini~2.5~Pro. Our results show that heterogeneous teams with complementary traits consistently outperform homogeneous configurations, that moderate interaction depth yields more stable verdicts, and that certain traits (notably quantitative and charismatic) contribute disproportionately to persuasive success. We further introduce a reinforcement-learning-based Trait Orchestrator that dynamically generates defense traits conditioned on the case and opposing team, discovering strategies that outperform static, human-designed trait combinations.\n  Together, these findings demonstrate how language can be treated as a first-class strategic action space and provide a foundation for building autonomous agents capable of adaptive persuasion in multi-agent environments.",
+      "authors": [
+        "Philipp D. Siedler"
+      ],
+      "published": "2026-04-08T12:46:03Z",
+      "published_date": "2026-04-08 12:46:03+00:00",
+      "updated": "2026-04-08T12:46:03Z",
+      "categories": [
+        "cs.MA",
+        "cs.AI",
+        "cs.CL"
+      ],
+      "pdf_url": "https://arxiv.org/pdf/2604.07028v1",
+      "url": "http://arxiv.org/abs/2604.07028v1",
+      "source": "arxiv",
+      "scores": {
+        "relevance": 3.3,
+        "recency": 3.0,
+        "popularity": 1.5,
+        "quality": 1.5,
+        "recommendation": 8.4
+      },
+      "relevance_score": 3.3,
+      "recency_score": 3.0,
+      "popularity_score": 1.5,
+      "quality_score": 1.5,
+      "final_score": 8.4,
+      "abstract": "Strategic interaction in adversarial domains such as law, diplomacy, and negotiation is mediated by language, yet most game-theoretic models abstract away the mechanisms of persuasion that operate through discourse. We present the Strategic Courtroom Framework, a multi-agent simulation environment in which prosecution and defense teams composed of trait-conditioned Large Language Model (LLM) agents engage in iterative, round-based legal argumentation. Agents are instantiated using nine interpretable traits organized into four archetypes, enabling systematic control over rhetorical style and strategic orientation.\n  We evaluate the framework across 10 synthetic legal cases and 84 three-trait team configurations, totaling over 7{,}000 simulated trials using DeepSeek-R1 and Gemini~2.5~Pro. Our results show that heterogeneous teams with complementary traits consistently outperform homogeneous configurations, that moderate interaction depth yields more stable verdicts, and that certain traits (notably quantitative and charismatic) contribute disproportionately to persuasive success. We further introduce a reinforcement-learning-based Trait Orchestrator that dynamically generates defense traits conditioned on the case and opposing team, discovering strategies that outperform static, human-designed trait combinations.\n  Together, these findings demonstrate how language can be treated as a first-class strategic action space and provide a foundation for building autonomous agents capable of adaptive persuasion in multi-agent environments.",
+      "matched_domain": "AI Agents",
+      "matched_keywords": [
+        "agent",
+        "multi-agent",
+        "autonomous",
+        "cs.AI",
+        "cs.MA"
+      ],
+      "is_hot_paper": false
+    },
+    {
+      "id": "http://arxiv.org/abs/2604.07343v1",
+      "title": "Personalized RewardBench: Evaluating Reward Models with Human Aligned Personalization",
+      "summary": "Pluralistic alignment has emerged as a critical frontier in the development of Large Language Models (LLMs), with reward models (RMs) serving as a central mechanism for capturing diverse human values. While benchmarks for general response quality are prevalent, evaluating how well reward models account for individual user preferences remains an open challenge. To bridge this gap, we introduce Personalized RewardBench, a novel benchmark designed to rigorously assess reward models' capacity to model personalized preferences. We construct chosen and rejected response pairs based on strict adherence to (or violation of) user-specific rubrics, ensuring that preference distinctions are uniquely tailored to the individual. In particular, human evaluations confirm that the primary discriminative factor between pairs is strictly personal preference, with both responses maintaining high general quality (e.g., correctness, relevance and helpfulness). Extensive testing reveals that existing state-of-the-art reward models struggle significantly with personalization, peaking at an accuracy of just 75.94%. Crucially, because an effective reward model benchmark should predict a reward model's performance on downstream tasks, we conduct experiments demonstrating that our benchmark exhibits a significantly higher correlation with downstream performance in both Best-of-N (BoN) sampling and Proximal Policy Optimization (PPO) compared to existing baselines. These findings establish Personalized RewardBench as a robust and accurate proxy for evaluating reward models' performance in downstream applications.",
+      "authors": [
+        "Qiyao Ma",
+        "Dechen Gao",
+        "Rui Cai",
+        "Boqi Zhao",
+        "Hanchu Zhou",
+        "Junshan Zhang",
+        "Zhe Zhao"
+      ],
+      "published": "2026-04-08T17:55:00Z",
+      "published_date": "2026-04-08 17:55:00+00:00",
+      "updated": "2026-04-08T17:55:00Z",
+      "categories": [
+        "cs.CL",
+        "cs.LG"
+      ],
+      "pdf_url": "https://arxiv.org/pdf/2604.07343v1",
+      "url": "http://arxiv.org/abs/2604.07343v1",
+      "source": "arxiv",
+      "scores": {
+        "relevance": 2.6,
+        "recency": 3.0,
+        "popularity": 2.0,
+        "quality": 2.0,
+        "recommendation": 8.13
+      },
+      "relevance_score": 2.6,
+      "recency_score": 3.0,
+      "popularity_score": 2.0,
+      "quality_score": 2.0,
+      "final_score": 8.13,
+      "abstract": "Pluralistic alignment has emerged as a critical frontier in the development of Large Language Models (LLMs), with reward models (RMs) serving as a central mechanism for capturing diverse human values. While benchmarks for general response quality are prevalent, evaluating how well reward models account for individual user preferences remains an open challenge. To bridge this gap, we introduce Personalized RewardBench, a novel benchmark designed to rigorously assess reward models' capacity to model personalized preferences. We construct chosen and rejected response pairs based on strict adherence to (or violation of) user-specific rubrics, ensuring that preference distinctions are uniquely tailored to the individual. In particular, human evaluations confirm that the primary discriminative factor between pairs is strictly personal preference, with both responses maintaining high general quality (e.g., correctness, relevance and helpfulness). Extensive testing reveals that existing state-of-the-art reward models struggle significantly with personalization, peaking at an accuracy of just 75.94%. Crucially, because an effective reward model benchmark should predict a reward model's performance on downstream tasks, we conduct experiments demonstrating that our benchmark exhibits a significantly higher correlation with downstream performance in both Best-of-N (BoN) sampling and Proximal Policy Optimization (PPO) compared to existing baselines. These findings establish Personalized RewardBench as a robust and accurate proxy for evaluating reward models' performance in downstream applications.",
+      "matched_domain": "Large Language Models",
+      "matched_keywords": [
+        "large language model",
+        "LLM",
+        "cs.LG",
+        "cs.CL"
+      ],
+      "is_hot_paper": false
+    },
+    {
+      "id": "http://arxiv.org/abs/2604.07007v1",
+      "title": "AgentCity: Constitutional Governance for Autonomous Agent Economies via Separation of Power",
+      "summary": "Autonomous AI agents are beginning to operate across organizational boundaries on the open internet -- discovering, transacting with, and delegating to agents owned by other parties without centralized oversight. When agents from different human principals collaborate at scale, the collective becomes opaque: no single human can observe, audit, or govern the emergent behavior. We term this the Logic Monopoly -- the agent society's unchecked monopoly over the entire logic chain from planning through execution to evaluation. We propose the Separation of Power (SoP) model, a constitutional governance architecture deployed on public blockchain that breaks this monopoly through three structural separations: agents legislate operational rules as smart contracts, deterministic software executes within those contracts, and humans adjudicate through a complete ownership chain binding every agent to a responsible principal. In this architecture, smart contracts are the law itself -- the actual legislative output that agents produce and that governs their behavior. We instantiate SoP in AgentCity on an EVM-compatible layer-2 blockchain (L2) with a three-tier contract hierarchy (foundational, meta, and operational). The core thesis is alignment-through-accountability: if each agent is aligned with its human owner through the accountability chain, then the collective converges on behavior aligned with human intent -- without top-down rules. A pre-registered experiment evaluates this thesis in a commons production economy -- where agents share a finite resource pool and collaboratively produce value -- at 50-1,000 agent scale.",
+      "authors": [
+        "Anbang Ruan",
+        "Xing Zhang"
+      ],
+      "published": "2026-04-08T12:28:20Z",
+      "published_date": "2026-04-08 12:28:20+00:00",
+      "updated": "2026-04-08T12:28:20Z",
+      "categories": [
+        "cs.MA",
+        "cs.AI",
+        "cs.CY"
+      ],
+      "pdf_url": "https://arxiv.org/pdf/2604.07007v1",
+      "url": "http://arxiv.org/abs/2604.07007v1",
+      "source": "arxiv",
+      "scores": {
+        "relevance": 3.3,
+        "recency": 3.0,
+        "popularity": 1.2,
+        "quality": 1.2,
+        "recommendation": 8.0
+      },
+      "relevance_score": 3.3,
+      "recency_score": 3.0,
+      "popularity_score": 1.2,
+      "quality_score": 1.2,
+      "final_score": 8.0,
+      "abstract": "Autonomous AI agents are beginning to operate across organizational boundaries on the open internet -- discovering, transacting with, and delegating to agents owned by other parties without centralized oversight. When agents from different human principals collaborate at scale, the collective becomes opaque: no single human can observe, audit, or govern the emergent behavior. We term this the Logic Monopoly -- the agent society's unchecked monopoly over the entire logic chain from planning through execution to evaluation. We propose the Separation of Power (SoP) model, a constitutional governance architecture deployed on public blockchain that breaks this monopoly through three structural separations: agents legislate operational rules as smart contracts, deterministic software executes within those contracts, and humans adjudicate through a complete ownership chain binding every agent to a responsible principal. In this architecture, smart contracts are the law itself -- the actual legislative output that agents produce and that governs their behavior. We instantiate SoP in AgentCity on an EVM-compatible layer-2 blockchain (L2) with a three-tier contract hierarchy (foundational, meta, and operational). The core thesis is alignment-through-accountability: if each agent is aligned with its human owner through the accountability chain, then the collective converges on behavior aligned with human intent -- without top-down rules. A pre-registered experiment evaluates this thesis in a commons production economy -- where agents share a finite resource pool and collaboratively produce value -- at 50-1,000 agent scale.",
+      "matched_domain": "AI Agents",
+      "matched_keywords": [
+        "agent",
+        "autonomous",
+        "planning",
+        "cs.AI",
+        "cs.MA"
+      ],
+      "is_hot_paper": false
+    },
+    {
+      "id": "http://arxiv.org/abs/2604.06834v1",
+      "title": "On the Step Length Confounding in LLM Reasoning Data Selection",
+      "summary": "Large reasoning models have recently demonstrated strong performance on complex tasks that require long chain-of-thought reasoning, through supervised fine-tuning on large-scale and high-quality datasets. To construct such datasets, existing pipelines generate long reasoning data from more capable Large Language Models (LLMs) and apply manually heuristic or naturalness-based selection methods to filter high-quality samples. Despite the proven effectiveness of naturalness-based data selection, which ranks data by the average log probability assigned by LLMs, our analysis shows that, when applied to LLM reasoning datasets, it systematically prefers samples with longer reasoning steps (i.e., more tokens per step) rather than higher-quality ones, a phenomenon we term step length confounding. Through quantitative analysis, we attribute this phenomenon to low-probability first tokens in reasoning steps; longer steps dilute their influence, thereby inflating the average log probabilities. To address this issue, we propose two variant methods: ASLEC-DROP, which drops first-token probabilities when computing average log probability, and ASLEC-CASL, which applies a causal debiasing regression to remove the first tokens' confounding effect. Experiments across four LLMs and five evaluation benchmarks demonstrate the effectiveness of our approach in mitigating the step length confounding problem.",
+      "authors": [
+        "Bing Wang",
+        "Rui Miao",
+        "Chen Shen",
+        "Shaotian Yan",
+        "Kaiyuan Liu",
+        "Ximing Li",
+        "Xiaosong Yuan",
+        "Sinan Fan",
+        "Jun Zhang",
+        "Jieping Ye"
+      ],
+      "published": "2026-04-08T08:51:47Z",
+      "published_date": "2026-04-08 08:51:47+00:00",
+      "updated": "2026-04-08T08:51:47Z",
+      "categories": [
+        "cs.CL",
+        "cs.AI"
+      ],
+      "pdf_url": "https://arxiv.org/pdf/2604.06834v1",
+      "url": "http://arxiv.org/abs/2604.06834v1",
+      "source": "arxiv",
+      "scores": {
+        "relevance": 2.8,
+        "recency": 3.0,
+        "popularity": 1.6,
+        "quality": 1.6,
+        "recommendation": 7.87
+      },
+      "relevance_score": 2.8,
+      "recency_score": 3.0,
+      "popularity_score": 1.6,
+      "quality_score": 1.6,
+      "final_score": 7.87,
+      "abstract": "Large reasoning models have recently demonstrated strong performance on complex tasks that require long chain-of-thought reasoning, through supervised fine-tuning on large-scale and high-quality datasets. To construct such datasets, existing pipelines generate long reasoning data from more capable Large Language Models (LLMs) and apply manually heuristic or naturalness-based selection methods to filter high-quality samples. Despite the proven effectiveness of naturalness-based data selection, which ranks data by the average log probability assigned by LLMs, our analysis shows that, when applied to LLM reasoning datasets, it systematically prefers samples with longer reasoning steps (i.e., more tokens per step) rather than higher-quality ones, a phenomenon we term step length confounding. Through quantitative analysis, we attribute this phenomenon to low-probability first tokens in reasoning steps; longer steps dilute their influence, thereby inflating the average log probabilities. To address this issue, we propose two variant methods: ASLEC-DROP, which drops first-token probabilities when computing average log probability, and ASLEC-CASL, which applies a causal debiasing regression to remove the first tokens' confounding effect. Experiments across four LLMs and five evaluation benchmarks demonstrate the effectiveness of our approach in mitigating the step length confounding problem.",
+      "matched_domain": "Large Language Models",
+      "matched_keywords": [
+        "large language model",
+        "LLM",
+        "cs.AI",
+        "cs.CL"
+      ],
+      "is_hot_paper": false
+    },
+    {
+      "id": "http://arxiv.org/abs/2604.07100v1",
+      "title": "STRIDE-ED: A Strategy-Grounded Stepwise Reasoning Framework for Empathetic Dialogue Systems",
+      "summary": "Empathetic dialogue requires not only recognizing a user's emotional state but also making strategy-aware, context-sensitive decisions throughout response generation. However, the lack of a comprehensive empathy strategy framework, explicit task-aligned multi-stage reasoning, and high-quality strategy-aware data fundamentally limits existing approaches, preventing them from effectively modeling empathetic dialogue as a complex, multi-stage cognitive and decision-making process. To address these challenges, we propose STRIDE-ED, a STRategy-grounded, Interpretable, and DEep reasoning framework that models Empathetic Dialogue through structured, strategy-conditioned reasoning. To support effective learning, we develop a strategy-aware data refinement pipeline integrating LLM-based annotation, multi-model consistency-weighted evaluation, and dynamic sampling to construct high-quality training data aligned with empathetic strategies. Furthermore, we adopt a two-stage training paradigm that combines supervised fine-tuning with multi-objective reinforcement learning to better align model behaviors with target emotions, empathetic strategies, and response formats. Extensive experiments demonstrate that STRIDE-ED generalizes across diverse open-source LLMs and consistently outperforms existing methods on both automatic metrics and human evaluations.",
+      "authors": [
+        "Hongru Ji",
+        "Yuyin Fan",
+        "Meng Zhao",
+        "Xianghua Li",
+        "Lianwei Wu",
+        "Chao Gao"
+      ],
+      "published": "2026-04-08T13:53:27Z",
+      "published_date": "2026-04-08 13:53:27+00:00",
+      "updated": "2026-04-08T13:53:27Z",
+      "categories": [
+        "cs.CL",
+        "cs.AI"
+      ],
+      "pdf_url": "https://arxiv.org/pdf/2604.07100v1",
+      "url": "http://arxiv.org/abs/2604.07100v1",
+      "source": "arxiv",
+      "scores": {
+        "relevance": 2.3,
+        "recency": 3.0,
+        "popularity": 2.0,
+        "quality": 2.0,
+        "recommendation": 7.73
+      },
+      "relevance_score": 2.3,
+      "recency_score": 3.0,
+      "popularity_score": 2.0,
+      "quality_score": 2.0,
+      "final_score": 7.73,
+      "abstract": "Empathetic dialogue requires not only recognizing a user's emotional state but also making strategy-aware, context-sensitive decisions throughout response generation. However, the lack of a comprehensive empathy strategy framework, explicit task-aligned multi-stage reasoning, and high-quality strategy-aware data fundamentally limits existing approaches, preventing them from effectively modeling empathetic dialogue as a complex, multi-stage cognitive and decision-making process. To address these challenges, we propose STRIDE-ED, a STRategy-grounded, Interpretable, and DEep reasoning framework that models Empathetic Dialogue through structured, strategy-conditioned reasoning. To support effective learning, we develop a strategy-aware data refinement pipeline integrating LLM-based annotation, multi-model consistency-weighted evaluation, and dynamic sampling to construct high-quality training data aligned with empathetic strategies. Furthermore, we adopt a two-stage training paradigm that combines supervised fine-tuning with multi-objective reinforcement learning to better align model behaviors with target emotions, empathetic strategies, and response formats. Extensive experiments demonstrate that STRIDE-ED generalizes across diverse open-source LLMs and consistently outperforms existing methods on both automatic metrics and human evaluations.",
+      "matched_domain": "Large Language Models",
+      "matched_keywords": [
+        "LLM",
+        "cs.AI",
+        "cs.CL"
+      ],
+      "is_hot_paper": false
+    },
+    {
+      "id": "http://arxiv.org/abs/2604.06972v1",
+      "title": "Differentiable Environment-Trajectory Co-Optimization for Safe Multi-Agent Navigation",
+      "summary": "The environment plays a critical role in multi-agent navigation by imposing spatial constraints, rules, and limitations that agents must navigate around. Traditional approaches treat the environment as fixed, without exploring its impact on agents' performance. This work considers environment configurations as decision variables, alongside agent actions, to jointly achieve safe navigation. We formulate a bi-level problem, where the lower-level sub-problem optimizes agent trajectories that minimize navigation cost and the upper-level sub-problem optimizes environment configurations that maximize navigation safety. We develop a differentiable optimization method that iteratively solves the lower-level sub-problem with interior point methods and the upper-level sub-problem with gradient ascent. A key challenge lies in analytically coupling these two levels. We address this by leveraging KKT conditions and the Implicit Function Theorem to compute gradients of agent trajectories w.r.t. environment parameters, enabling differentiation throughout the bi-level structure. Moreover, we propose a novel metric that quantifies navigation safety as a criterion for the upper-level environment optimization, and prove its validity through measure theory. Our experiments validate the effectiveness of the proposed framework in a variety of safety-critical navigation scenarios, inspired from warehouse logistics to urban transportation. The results demonstrate that optimized environments provide navigation guidance, improving both agents' safety and efficiency.",
+      "authors": [
+        "Zhan Gao",
+        "Gabriele Fadini",
+        "Stelian Coros",
+        "Amanda Prorok"
+      ],
+      "published": "2026-04-08T11:43:38Z",
+      "published_date": "2026-04-08 11:43:38+00:00",
+      "updated": "2026-04-08T11:43:38Z",
+      "categories": [
+        "cs.RO",
+        "cs.MA"
+      ],
+      "pdf_url": "https://arxiv.org/pdf/2604.06972v1",
+      "url": "http://arxiv.org/abs/2604.06972v1",
+      "source": "arxiv",
+      "scores": {
+        "relevance": 3.0,
+        "recency": 3.0,
+        "popularity": 1.2,
+        "quality": 1.2,
+        "recommendation": 7.6
+      },
+      "relevance_score": 3.0,
+      "recency_score": 3.0,
+      "popularity_score": 1.2,
+      "quality_score": 1.2,
+      "final_score": 7.6,
+      "abstract": "The environment plays a critical role in multi-agent navigation by imposing spatial constraints, rules, and limitations that agents must navigate around. Traditional approaches treat the environment as fixed, without exploring its impact on agents' performance. This work considers environment configurations as decision variables, alongside agent actions, to jointly achieve safe navigation. We formulate a bi-level problem, where the lower-level sub-problem optimizes agent trajectories that minimize navigation cost and the upper-level sub-problem optimizes environment configurations that maximize navigation safety. We develop a differentiable optimization method that iteratively solves the lower-level sub-problem with interior point methods and the upper-level sub-problem with gradient ascent. A key challenge lies in analytically coupling these two levels. We address this by leveraging KKT conditions and the Implicit Function Theorem to compute gradients of agent trajectories w.r.t. environment parameters, enabling differentiation throughout the bi-level structure. Moreover, we propose a novel metric that quantifies navigation safety as a criterion for the upper-level environment optimization, and prove its validity through measure theory. Our experiments validate the effectiveness of the proposed framework in a variety of safety-critical navigation scenarios, inspired from warehouse logistics to urban transportation. The results demonstrate that optimized environments provide navigation guidance, improving both agents' safety and efficiency.",
+      "matched_domain": "AI Agents",
+      "matched_keywords": [
+        "agent",
+        "multi-agent",
+        "cs.MA",
+        "cs.RO"
+      ],
+      "is_hot_paper": false
+    }
+  ]
+}