サインイン

Daily Arxiv

新規
전 세계에서 발간되는 인공지능 관련 논문을 정리하는 페이지 입니다. 본 페이지는 Google Gemini를 활용해 요약 정리하며, 비영리로 운영 됩니다. 논문에 대한 저작권은 저자 및 해당 기관에 있으며, 요약본 공유 시 출처만 명기하면 됩니다. This service is supported by Google Gemini.
SpecMAGIC: Specification-Guided Multi-Task Generalization in Multi-Agent Reinforcement Learning
Secure Agentic AI and Multimodal Intelligence in 6G Communication Systems
RoleFlow: Self-Evolving Multi-Role Agentic Collaboration via Structural Adaptation
Soft Ontology Priors over Latent Concept Posteriors for Weakly Supervised Neuro-Symbolic Learning
Route-and-Generate: KL-Regularized Router Improvement for Reasoning Routing in LLMs
Breaking the high-degree barrier in Lehmer's problem with a human-AI discovery loop
Feedback That Backfires in Games: Do LLM Players Repeat the Move They Just Watched Fail?
When Stable and Preferred Semantics Agree: A Path-Based Approach
Neural Networks with Differentiable Belief-Function Layers
Accelerating Wildfire Hazard Mapping Across Canada with an AI Surrogate Modelling Framework
MCP vs CLI for AI Agents: A Comparative Analysis of Tool Integration Strategies for Autonomous Systems
The Cost of Saying Hello: An Empirical Study of Handshake Overhead in AI Agent Communication Protocols
To Manipulate or Not: Opinion Dynamics in LLM Interactions
Speedrunning Relational Foundation Model Pretraining
Sketch2Sim: an end-to-end modular framework for hand-drawn circuit analysis
Decision language models as affordable and rapid alternatives to LLM judges of clinical generative AI
Forging More Tasks into Better Harnesses: Agentic Harness Evolution via Successive Scale Racing
Automating ML for Science: Can Frontier Agents Climb Scientific Hills in the Wild?
SpecEvo: Speculative Evolution with Large Language Models for Cost-Efficient Scientific Discovery
Scaling in Games: Continued Pre-Training For Embodied Agents In Diverse Virtual Worlds
A Small Model Asks Better: Fine-Tuning Query Clarification for Retrieval-Augmented Generation
Omni-Woodpecker: Hallucination Correction via Evidence-Prior Rebalancing for Omni LLMs
A general-purpose System One decision model for clinical records
ReuF2F: Benchmarking Coding Agents on Formalized Undergraduate Research Problems
Physical AI: Concepts, Techniques, and Challenges
M-Plan: Comprehensive Planning For LLM Agents with Memory Arrangement
Teaching AI Agent to Play Poker Strategically Through Preference Optimization
Does Multi-Agent Collaboration Reduce Hallucination in Clinical Tasks?
SPEM: Structure-Aware Procedural Memory for Efficient Skill Use in LLM Agents
When Planning Problems Become Hard for LLMs: Separating Structural Complexity from Semantic Interpretability
LightCRS: Lightweight Multi-Domain Conversational Recommendation System
MAMath-Bench: Process-Aware and Robust Evaluation of Multimodal Multi-Agent Mathematical Reasoning
Fix, Not Just Resample: Bounded Compiler-Guided Repair in Lean 4
Industrial Agentic AI: LLM-Driven New-Generation Agents for Smart Manufacturing
DataPraxis: Scaling Analytics Tasks via Verified Workspace Evolution
MedPAO-Assist: Edge-Deployable, Modality-Aware Agentic AI for Radiology Report Structuring
Pareto-Optimal RTL Code Generation via Multi-Objective Reinforcement Learning with Large Language Models
Large language model and reinforcement learning based autonomous construction and progress updates
Navalia: Advancing Cowork Agents through Verifiable Task Synthesis and Long-Horizon Post-Training
Can Compact Open-Weight Large Language Models Recover Mobility Events from Incomplete Trajectories? A Controlled Evaluation Using Air Trips
AM-Bench: A Unified Taxonomy and Evaluation Suite for Agentic Misalignment
Cortex: A Fixed-Point Theory of Governed Coding Agents - Technical Report
Probing Ambiguity: Adversarial Intent Resolution for Robust Text-to-SQL
Training-Free Prompt Selection for Efficient Reinforcement Learning With Partial Rollouts
Distillation, Self-Training, and the Fragility of Learned Chemical Reasoning
Agents' Last Doubt: Knowing When to Act and When to Step Back in Tool-Using Agents
Self-Improving Video-to-Action Models
TorchTPU - A Hybrid Eager/Compiled Architecture for Native PyTorch Execution on TPU
A Conditional No-Hair Theorem for Transformers: How Language Models Forget Their Prompt
Joint, Decoupled, or Agent: When Each Multimodal Alignment Paradigm Wins
Cache2LoRA: Understanding and Improving Context Internalization
Persistent Memory and Iterative Improvement in an LLM Agent for Chinese-Character Drawing
World of SWECraft: Agent Collaboration at Scale via Simulated Software Worlds
LogicVal:A Logic-Formalized Benchmark for Constraint Satisfaction in Text-to-Image Generation
Unified Multiparadigm Fusion Intelligence for Rare Disease Diagnosis
Beyond Explicit Intent: Discovering and Shaping Artistic Intent through Human-AI Co-Creation
What Social Media Teaches Us About Accumulative AI Risks
A Controlled LLM Workflow for Evidence-Grounded Diagnosis of Risk Score Differences
Doing Less with More: A First-Principles Exploration of the Suitability of Agentic Computing over Alternative Architectural Choices
GardenBench: A lightweight, daily evaluation of LLM capabilities
FlexUMM: Flexible Representation Learning for Unified Multimodal Models
Not All Graphs Are Maps: A Task-Agnostic Exploration Framework for Mobile GUI Navigation
From Relevant Memories to Sufficient Evidence: Rethinking Memory Utilization in LLM Agents
At the Limits of Computability: Steps Towards Autonomous AI Discovery on the Busy Beaver Frontier
Repairing LLM-Generated Specifications with Deterministic Diagnosis
FRACTAL: A Self-Similar Multi-Agent Orchestration Architecture with Formal Communication, Convergence, and Depth Bounds
Are Large Reasoning Models Good Everyday Reasoners? A Comparison with Non-Reasoning Models on Logical Reasoning Tasks
Testing the Planning Claims of LLM-Based Agents against Formal Semantics
Harness-Delta Attribution: Dissecting Gains From Harness Evolution
Author-in-the-loop Agentic Reproducibility Auditing for a Machine Learning Conference
Functional Welfare for AI Safety
Learnable Splines for Unstructured Neural Operators
Null Controls Are Necessary but Not Sufficient: Evaluating Acceptance Guards for Scientific Discovery and Interpretability
From Natural Language Intent to Executable Control: Hierarchical Feedback-Guided LLM Planning for Multi-component Cyber-Physical Systems
PRIME: Towards Verifiable Evidence-Grounded Reasoning in LLM Agents
Sensitivity Analysis of Decision Rules in an Agent-Based Agro-Ecological Model
ECHOES: Seeing Beyond the Current Image via Recursive History Compression and Privileged On-Policy Distillation
From Events to Reasoning: Hierarchical Rule Composition over Explicit States
Rethinking User Encoders: Routing Diffusion States for Scenario-Adaptive User Representation
Structured Authority Injection: Persistent Human Control for LLM Agents via Typed MCP Overrides
Defining and Generating Compositional Structure
A Multi-Tiered AI Accessibility Model (MAIAM): Proportionate Access and Capability Limits for Comprehensive Frontier AI Safety
Anticipatory Commonsense with Moving Objects: An ASP-Based Framework for Predictive Reasoning with Interacting Trajectories
Evolving Together, Failing Apart: The Skill-Policy Co-evolution Trap in Self-Improving Agents
Confidence-Calibrated Inference Expansion for Evaluator-Guided Test-Time Reasoning
Temporal Specification Guidance for Text-to-Video Generation
A Self-Improving Agent that Grows its Own DSL: Oracle-Free Operator Induction for Exact Question Answering
When Reasoning Traces Conflict with Evidence: How Answer Configuration Shapes Evidence Following
EvolvingBrain: An Adaptive Skill System for Multi-modal Agents Self-Evolution in Open Worlds
Open-world AI: a Tale of Three Requirements
ARIA: Runtime Inspection and Amendment for Multi-Agent Systems
When Reality Outpaces Cognition: A New Scaling Principle for Reliable Intelligence
L2G-SR: Learning to Guide LLM-based Symbolic Regression
IneqAgent: Agentic Inequality Proving in Lean with Symbolic Computation
Counterfactual Neighborhood Risk Fields: Evidence-Grounded Selective Safety for Vision-Language Models
When Correctness Feedback Erases Evidence: Phase-Separated Completeness Auditing in Verus
AlphaDiana: A Measurement System for Agents in Verifiable Reasoning Tasks
Value-Informed Schedule Trimming for Accelerated Sequential Monte Carlo Guidance in Discrete Diffusion
Evaluating Model Capabilities After Benchmark Saturation
The World Won't Wait: Evaluating and Teaching Agents Under a Live Clock
新しい

SPIN-Bench: How Well Do LLMs Plan Strategically and Reason Socially?

작성자
Haebom
카테고리
未設定

저자

Jianzhu Yao, Kevin Wang, Ryan Hsieh, Haisu Zhou, Tianqing Zou, Zerui Cheng, Zhangyang Wang, Pramod Viswanath

개요

본 논문은 전략적 계획, 상호작용 및 협상(SPIN-Bench)이라는 새로운 다중 도메인 평가 기준을 제시합니다. 기존 벤치마크들이 좁은 범위의 계획이나 단일 에이전트 추론에 집중하는 것과 달리, SPIN-Bench는 고전적인 PDDL 작업, 경쟁 보드 게임, 협력 카드 게임 및 다중 에이전트 협상 시나리오를 하나의 통합된 프레임워크로 결합합니다. 다양한 사회적 설정을 시뮬레이션하고 AI 에이전트의 추론 및 전략적 행동을 평가하기 위한 벤치마크와 아레나를 모두 포함합니다. 행동 공간, 상태 복잡성 및 상호 작용하는 에이전트의 수를 체계적으로 변경하여 성공이 단계별 의사 결정뿐만 아니라 다른 (적대적 또는 협력적인) 참가자에 대한 개념적 추론에도 의존하는 다양한 사회적 설정을 시뮬레이션합니다. 실험 결과, 최신 LLMs는 기본적인 사실 검색 및 단거리 계획을 상당히 잘 처리하지만, 대규모 상태 공간에 대한 심층 다단계 추론과 불확실성 하에서의 사회적으로 능숙한 조정이 필요한 작업에서는 상당한 성능 병목 현상을 겪는다는 것을 보여줍니다. SPIN-Bench는 강력한 다중 에이전트 계획, 사회적 추론 및 인간-AI 협력에 대한 미래 연구의 촉매제가 될 것으로 기대됩니다.

시사점, 한계점

•
시사점:
◦
다중 에이전트 상호작용, 전략적 추론 및 협상 능력을 평가하는 새로운 벤치마크 SPIN-Bench 제시.
◦
기존 벤치마크의 한계를 극복하고, 다양한 사회적 상황을 포괄적으로 평가 가능.
◦
최신 LLMs의 사회적 추론 및 다단계 추론 능력의 한계를 밝히고, 향후 연구 방향 제시.
◦
인간-AI 협력 연구에 기여할 잠재력을 가짐.
•
한계점:
◦
SPIN-Bench의 다양한 과제들이 LLMs의 모든 측면을 포괄적으로 평가하는지는 추가 연구 필요.
◦
현재 벤치마크의 규모 및 복잡도가 향후 더욱 발전된 AI 모델에 적합할지에 대한 검토 필요.
◦
인간의 사회적 추론과 비교 분석을 위한 추가 연구 필요.
👍
Slashpageで作成