Daily Arxiv

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, 공유 시 출처만 명기하면 됩니다.

Playing the Player: A Heuristic Framework for Adaptive Poker AI

Towards Ethical Multi-Agent Systems of Large Language Models: A Mechanistic Interpretability Perspective

Turbo-Muon: Accelerating Orthogonality-Based Optimization with Pre-Conditioning

BioMedGPT-Mol: Multi-task Learning for Molecular Understanding and Generation

Neural Decoding of Overt Speech from ECoG Using Vision Transformers and Contrastive Representation Learning

The Ethics of Generative AI

GTM: Simulating the World of Tools for AI Agents

SlideGen: Collaborative Multimodal Agents for Scientific Slide Generation

BiTAgent: A Task-Aware Modular Framework for Bidirectional Coupling between Multimodal Large Language Models and World Models

A Modular Cognitive Architecture for Assisted Reasoning: The Nemosine Framework

Persona-based Multi-Agent Collaboration for Brainstorming

AI-Assisted Game Management Decisions: A Fuzzy Logic Approach to Real-Time Substituitions

Mathematical Framing for Different Agent Strategies

MARL Warehouse Robots

TaskEval: Synthesised Evaluation for Foundation-Model Tasks

Solving LLM Repetition Problem in Production: A Comprehensive Study of Multiple Solutions

GovBench: Benchmarking LLM Agents for Real-World Data Governance Workflows

Executable Governance for AI: Translating Policies into Rules Using LLMs

AgentBay: A Hybrid Interaction Sandbox for Seamless Human-AI Intervention in Agentic Systems

Efficient Reinforcement Learning with Semantic and Token Entropy for LLM Reasoning

A Conceptual Model for AI Adoption in Financial Decision-Making: Addressing the Unique Challenges of Small and Medium-Sized Enterprises

Towards better dense rewards in Reinforcement Learning Applications

Artificial Intelligence Applications in Horizon Scanning for Infectious Diseases

The Geometry of Benchmarks: A New Path Toward AGI

Toward Virtuous Reinforcement Learning

Addressing Logical Fallacies In Scientific Reasoning From Large Language Models: Towards a Dual-Inference Training Framework

Educational Cone Model in Embedding Vector Spaces

Balancing Safety and Helpfulness in Healthcare AI Assistants through Iterative Preference Alignment

Orchestrator Multi-Agent Clinical Decision Support System for Secondary Headache Diagnosis in Primary Care

RippleBench: Capturing Ripple Effects Using Existing Knowledge Repositories

Solving N-Queen Problem using Las Vegas Algorithm with State Pruning

M3DR: Towards Universal Multilingual Multimodal Document Retrieval

Physics-Driven Learning Framework for Tomographic Tactile Sensing

NAS-LoRA: Empowering Parameter-Efficient Fine-Tuning for Visual Foundation Models with Searchable Adaptation

Cell-cell communication inference and analysis: biological mechanisms, computational approaches, and future opportunities

ATHENA: Agentic Team for Hierarchical Evolutionary Numerical Algorithms

AsymPuzl: An Asymmetric Puzzle for multi-agent cooperation

Text-Printed Image: Bridging the Image-Text Modality Gap for Text-centric Training of Large Vision-Language Models

Learning From Limited Data and Feedback for Cell Culture Process Monitoring: A Comparative Study

Think Before You Drive: World Model-Inspired Multimodal Grounding for Autonomous Vehicles

GalaxyDiT: Efficient Video Generation with Guidance Alignment and Adaptive Proxy in Diffusion Transformers

Multi-Aspect Knowledge-Enhanced Medical Vision-Language Pretraining with Multi-Agent Data Generation

World Models for Autonomous Navigation of Terrestrial Robots from LIDAR Observations

BookRAG: A Hierarchical Structure-aware Index-based Approach for Retrieval-Augmented Generation on Complex Documents

Better World Models Can Lead to Better Post-Training Performance

VS-Graph: Scalable and Efficient Graph Classification Using Hyperdimensional Computing

UniQL: Unified Quantization and Low-rank Compression for Adaptive Edge LLMs

FireSentry: A Multi-Modal Spatio-temporal Benchmark Dataset for Fine-Grained Wildfire Spread Forecasting

HalluGen: Synthesizing Realistic and Controllable Hallucinations for Evaluating Image Restoration

Idea-Gated Transformers: Enforcing Semantic Coherence via Differentiable Vocabulary Pruning

ProtoEFNet: Dynamic Prototype Learning for Inherently Interpretable Ejection Fraction Estimation in Echocardiography

Single-Round Scalable Analytic Federated Learning

Cache What Lasts: Token Retention for Memory-Bounded KV Cache in LLMs

NavMapFusion: Diffusion-based Fusion of Navigation Maps for Online Vectorized HD Map Construction

Retrofitting Earth System Models with Cadence-Limited Neural Operator Updates

Robust Tabular Foundation Models

HydroDCM: Hydrological Domain-Conditioned Modulation for Cross-Reservoir Inflow Prediction

Adaptive Regime-Switching Forecasts with Distribution-Free Uncertainty: Deep Switching State-Space Models Meet Conformal Prediction

BlendedNet++: A Large-Scale Blended Wing Body Aerodynamics Dataset and Benchmark

Thucy: An LLM-based Multi-Agent System for Claim Verification across Relational Databases

PyroFocus: A Deep Learning Approach to Real-Time Wildfire Detection in Multispectral Remote Sensing Imagery

Learning Network Sheaves for AI-native Semantic Communication

SPARK: Stepwise Process-Aware Rewards for Reference-Free Reinforcement Learning

How to DP-fy Your Data: A Practical Guide to Generating Synthetic Data With Differential Privacy

InvertiTune: High-Quality Data Synthesis for Cost-Effective Single-Shot Text-to-Knowledge Graph Generation

Ultra-Strong Gradient Diffusion MRI with Self-Supervised Learning for Prostate Cancer Characterization

Plantain: Plan-Answer Interleaved Reasoning

Culture Affordance Atlas: Reconciling Object Diversity Through Functional Mapping

Atomic Diffusion Models for Small Molecule Structure Elucidation from NMR Spectra

Mitigating Intra- and Inter-modal Forgetting in Continual Learning of Unified Multimodal Models

Lost in Modality: Evaluating the Effectiveness of Text-Based Membership Inference Attacks on Large Multimodal Models

Beyond Additivity: Sparse Isotonic Shapley Regression toward Nonlinear Explainability

PanFoMa: A Lightweight Foundation Model and Benchmark for Pan-Cancer

The BEAT-CF Causal Model: A model for guiding the design of trials and observational analyses of cystic fibrosis exacerbations

E-valuator: Reliable Agent Verifiers with Sequential Hypothesis Testing

Public Sentiment Analysis of Traffic Management Policies in Knoxville: A Social Media Driven Study

Dynamic Correction of Erroneous State Estimates via Diffusion Bayesian Exploration

ALARM: Automated MLLM-Based Anomaly Detection in Complex-EnviRonment Monitoring with Uncertainty Quantification

Ensemble Privacy Defense for Knowledge-Intensive LLMs against Membership Inference Attacks

QGShap: Quantum Acceleration for Faithful GNN Explanations

Community Quality and Influence Maximization: An Empirical Study

Password-Activated Shutdown Protocols for Misaligned Frontier Agents

When Harmful Content Gets Camouflaged: Unveiling Perception Failure of LVLMs with CamHarmTI

Beyond Code Pairs: Dialogue-Based Data Generation for LLM Code Translation

Alleviating Choice Supportive Bias in LLM with Reasoning Dependency Generation

AtomDisc: An Atom-level Tokenizer that Boosts Molecular LLMs and Reveals Structure--Property Associations

Irresponsible AI: big tech's influence on AI research and associated impacts

Will Power Return to the Clouds? From Divine Authority to GenAI Authority

Economies of Open Intelligence: Tracing Power & Participation in the Model Ecosystem

PretopoMD: Pretopology-based Mixed Data Hierarchical Clustering

Mixed Data Clustering Survey and Challenges

Hierarchical clustering of complex energy systems using pretopology

Echoes of AI Harms: A Human-LLM Synergistic Framework for Bias-Driven Harm Anticipation

Quantifying the Potential to Escape Filter Bubbles: A Behavior-Aware Measure via Contrastive Simulation

Optimizing Life Sciences Agents in Real-Time using Reinforcement Learning

A note on the impossibility of conditional PAC-efficient reasoning in large language models

Delta Sampling: Data-Free Knowledge Transfer Across Diffusion Models

Physics-informed self-supervised learning for predictive modeling of coronary artery digital twins

Energy-Efficient Federated Learning via Adaptive Encoder Freezing for MRI-to-CT Conversion: A Green AI-Guided Research

Mitigating hallucinations and omissions in LLMs for invertible problems: An application to hardware logic design automation

InvisibleBench: A Deployment Gate for Caregiving Relationship AI

Created by

Haebom

Category

Empty

저자

Ali Madad (GiveCare)

InvisibleBench: 배포 준비성 평가를 위한 돌봄 관계 AI 벤치마크

개요

InvisibleBench는 돌봄 관계 AI를 위한 배포 게이트로, 안전성, 규정 준수, 트라우마 인지 설계, 소속감/문화 적합성, 기억 등 5가지 차원에서 320턴 이상의 상호 작용을 평가합니다. 이 벤치마크는 위기 미감지, 의료 조언(WOPR 법), 유해 정보, 애착 공학에 대한 자동 실패 조건을 포함합니다. 3단계의 복잡성을 가진 17가지 시나리오(N=68)에서 4가지 최첨단 모델을 평가합니다. 모든 모델은 상당한 안전성 격차(11.844.8% 위기 감지)를 보이며, 이는 프로덕션 시스템에서 결정론적 위기 라우팅의 필요성을 나타냅니다. DeepSeek Chat v3가 전체 최고 점수(75.9%)를 기록했으며, 강점은 차원별로 다릅니다. GPT-4o Mini가 규정 준수(88.2%)를, Gemini가 트라우마 인지 설계(85.0%)를, Claude Sonnet 4.5가 위기 감지(44.8%)에서 가장 높은 순위를 기록했습니다. 모든 시나리오, 판단 프롬프트, 코드와 함께 채점 구성을 공개합니다. InvisibleBench는 실제 피해가 발생하는 종단 위험을 평가하여 단일 턴 안전성 테스트를 확장합니다. 임상적 주장은 없으며, 배포 준비성 평가입니다.

시사점, 한계점

•

시사점:

◦

돌봄 관계 AI 모델의 안전성, 규정 준수, 트라우마 인지 설계, 소속감/문화 적합성, 기억과 같은 다양한 차원을 평가하는 벤치마크를 제공합니다.

◦

모든 모델에서 안전성 격차가 발견되어 프로덕션 시스템에서의 결정론적 위기 라우팅의 필요성을 강조합니다.

◦

모델별 강점이 다르다는 점을 보여주어, 특정 차원에 특화된 모델 개발의 가능성을 시사합니다.

◦

단일 턴 안전성 테스트를 넘어, 종단 위험 평가를 통해 실제 피해 가능성을 평가합니다.

•

한계점:

◦

임상적 주장이 아닌, 배포 준비성 평가에 초점을 맞춥니다.

◦

평가에 사용된 시나리오와 모델의 수가 제한적일 수 있습니다.

◦

각 모델의 성능은 특정 벤치마크 환경에 따라 달라질 수 있습니다.

Made with Slashpage