arXiv cs.AI
When to Re-Plan: Subgoal Persistence in Hierarchical Latent Reasoning
The paper titled 'When to Re-Plan: Subgoal Persistence in Hierarchical Latent Reasoning' explores the balance between…
Hub / Ai Research
ai-research · WeSearch
Ai Research news.
Page 4 of Ai Research headlines on WeSearch — deduped and updated continuously from 10+ editorial sources.
arXiv cs.AI
Unveiling the Structure of Do-Calculus Reasoning via Derivation Graphs
The paper introduces derivation graphs to enhance the understanding of do-calculus reasoning. These graphs help in…
arXiv cs.AI
Code-on-Graph: Iterative Programmatic Reasoning via Large Language Models on Knowledge Graphs
The paper introduces Code-on-Graph (CoG), a new framework for integrating Large Language Models (LLMs) with Knowledge…
arXiv cs.AI
Dynamic Objective Selection with Safeguards and LLM Oversight for Financial Decision-Making
The paper introduces Dynamic Objective Selection with Safeguards (DOSS) for financial decision-making. DOSS aims to…
arXiv cs.AI
SkillPyramid: A Hierarchical Skill Consolidation Framework for Self-Evolving Agents
The article introduces SkillPyramid, a framework designed to enhance the skill consolidation of self-evolving AI…
arXiv cs.AI
The DeepSpeak-Agentic Dataset
The DeepSpeak-Agentic dataset consists of over 37 hours of semi-structured conversations between humans and AI agents.…
arXiv cs.AI
EvoDrive: Pareto Evolution for Safety-Critical Autonomous Driving via Self-Improving LLM Agents
EvoDrive is a new framework designed for generating safety-critical scenarios in autonomous driving systems. It…
arXiv cs.AI
From Answers to States: Verifiable Process-Level Evaluation of Chemical Reasoning in Large Language Models
The paper introduces ChemCoTBench-V2, a benchmark designed for evaluating chemical reasoning in large language models.…
arXiv cs.AI
Diagnosing Knowledge Gaps in LLM Tool Use: An Agentic Benchmark for Novel API Acquisition
The paper introduces NovelAPIBench, a dynamic benchmark designed to evaluate large language models' ability to use…
arXiv cs.AI
Towards Non-Monotonic Entailment in Propositional Defeasible Standpoint Logic
The paper discusses advancements in propositional defeasible standpoint logic, focusing on non-monotonic entailment.…
arXiv cs.AI
Gender-Dependent Diagnostic Substitution in LLM Medical Triage: Same Symptoms, Unequal Urgency
A recent study investigates gender-dependent disparities in medical triage recommendations made by large language…
arXiv cs.AI
TSQAgent: Rating Time Series Data Quality via Dedicated Agentic Reasoning
The paper introduces TSQAgent, a framework designed to improve the assessment of time series data quality using large…
arXiv cs.AI
Bridging Auxiliary Constraints to Resolve Instruction Following in Large Reasoning Models
The paper introduces a framework to improve instruction following in Large Reasoning Models (LRMs) by addressing the…
arXiv cs.AI
Cross-Lingual Token Arbitrage: Optimizing Code Agent Context Windows via Local LLM Preprocessing
The paper discusses a new approach to optimize coding agents by reducing input-token costs. It introduces a middleware…
arXiv cs.AI
From Prompt to Service: An SLM-Based Agent Orchestration Gateway for AI-Driven Virtual Worlds
The paper introduces an SLM-based Agent Orchestration Gateway designed for AI-driven virtual worlds. This gateway…
arXiv cs.AI
SAGE: A Quantitative Evaluation of Socialized Evolution in Agent Ecosystems
The paper introduces SAGE, a framework for evaluating socialized evolution in agent ecosystems. It compares two…
arXiv cs.AI
Overlaying Governance: A Compositional Authorization Framework for Delegation and Scope in Agentic AI
The paper presents a new compositional authorization framework for managing delegation and scope in agentic AI…
arXiv cs.AI
ThoughtFold: Folding Reasoning Chains via Introspective Preference Learning
The paper introduces ThoughtFold, a framework designed to improve the efficiency of Large Reasoning Models (LRMs) by…
arXiv cs.AI
A formal definition and meta-model for a machine theory of mind
The paper presents a formal definition and meta-model for the Machine Theory of Mind. It integrates insights from…
arXiv cs.AI
StepFinder: A Temporal Semantic Framework for Failure Attribution in Multi-Agent Systems
StepFinder is a new framework designed for failure attribution in multi-agent systems. It aims to improve the…
arXiv cs.AI
DMF: A Deterministic Memory Framework for Conversational AI Agents
The Deterministic Memory Framework (DMF) aims to enhance memory systems for conversational AI agents. It replaces…
arXiv cs.AI
What Makes Interaction Trajectories Effective for Training Terminal Agents?
The paper investigates the effectiveness of interaction trajectories in training terminal agents. It reveals that…
arXiv cs.AI
CP-Agent: Context-Aware Multimodal Reasoning for Cellular Morphological Profiling under Chemical Perturbations
The article introduces CP-Agent, a multimodal large language model designed for cellular morphological profiling under…
arXiv cs.AI
InfoMem: Training Long-Context Memory Agents with Answer-Conditioned Information Gain
The paper introduces InfoMem, a new reward mechanism designed for training long-context memory agents in artificial…
arXiv cs.AI
The Violation Situation Pattern: A Knowledge-Graph Pattern for Compliance Violations
The Violation Situation Pattern (VSP) is a new knowledge-graph pattern designed to improve compliance violation…
arXiv cs.AI
The Reliability Gap in Benchmark Auditing: Distribution Shift and Scale as Failure Modes of Contamination Detection
The article discusses the challenges of benchmark auditing in artificial intelligence, particularly regarding…
arXiv cs.AI
LEAP: Supercharging LLMs for Formal Mathematics with Agentic Frameworks
The article discusses LEAP, a new framework designed to enhance the capabilities of Large Language Models (LLMs) in…
arXiv cs.AI
A Negative Result on Cross-Model Activation Transfer in a Pythia Multi-Hop Setting
The paper presents a negative result regarding cross-model activation transfer in a multi-hop reasoning setting using…
arXiv cs.AI
Distilling Answer-Set Programming Rules from LLMs for Neurosymbolic Visual Question Answering
The article discusses a novel approach for enhancing Visual Question Answering (VQA) by distilling rules from Large…
arXiv cs.AI
Do Real-World Datasets Contain Natural Experiments? An Empirical Study Using Causal Feature Selection
The study investigates whether real-world datasets contain natural experiments, which are implicit interventions…
arXiv cs.AI
Solipsistic Superintelligence is Unlikely to be Cooperative
A recent paper argues that superintelligence developed from a solipsistic approach to AI design is unlikely to be…
arXiv cs.AI
Perceive Before Reasoning: A Pre-Reasoning Perception Framework for Efficient and Reliable Proactive Mobile Agents
The article discusses a new framework called the Pre-Reasoning Perception Framework (PRPF) designed to enhance…
arXiv cs.AI
Effect of Demographic Bias on Skin Lesion Classification
The study investigates the impact of demographic bias on skin lesion classification using ResNet-based models. It…
arXiv cs.AI
MedCUA-Bench: A Screenshot-Only Benchmark for Clinical Computer-Use Agents
MedCUA-Bench is a newly introduced benchmark designed specifically for clinical computer-use agents. It aims to…
arXiv cs.AI
ClinicalMC: A Benchmark for Multi-Course Clinical Decision-Making with Large Language Models
The article introduces ClinicalMC, a benchmark designed for evaluating large language models in multi-course clinical…
arXiv cs.AI
GTBench: A Curriculum-Grounded Benchmark for Evaluating LLMs as Mathematical Research Assistants in Graph Theory
The article introduces GTBench, a benchmark designed to evaluate large language models (LLMs) as mathematical research…
arXiv cs.AI
Think-Before-Speak: From Internal Evaluation to Public Expression in Multi-Agent Social Simulation
The paper introduces a new framework called Think-Before-Speak (TBS) for multi-agent social simulation. TBS separates…
arXiv cs.AI
Uncertainty-Aware Clarification in LLM Agents with Information Gain
The article discusses a new framework for Large Language Model (LLM) agents that aims to improve their performance in…
arXiv cs.AI
EvoTrainer: Co-Evolving LLM Policies and Training Harnesses for Autonomous Agentic Reinforcement Learning
EvoTrainer is a new autonomous training framework designed for co-evolving LLM policies and training harnesses. It…
arXiv cs.AI
DeskCraft: Benchmarking Desktop Agents on Professional Workflows and Human-in-the-Loop Collaboration
The paper titled 'DeskCraft' introduces a new benchmark for evaluating desktop agents in professional workflows that…
arXiv cs.AI
From Long News to Accurate Forecast: Importance-Aware Fusion and PRM-Guided Reflection for Time Series Forecasting
A new framework has been developed to enhance time series forecasting by integrating news articles. This approach…
arXiv cs.AI
Decomposing how prompting steers behavior
The paper titled 'Decomposing how prompting steers behavior' explores how prompting influences the internal…
arXiv cs.AI
The Shadow Price of Reasoning: Economic Perspective on Optimal Budget Allocation for LLMs
The article discusses a new approach to budget allocation for Large Language Models (LLMs) based on economic…
arXiv cs.AI
DELTAMEM: Incremental Experience Memory for LLM Agents via Residual Trees
The paper introduces DeltaMem, a framework designed to enhance memory management in Large Language Model (LLM) agents.…
arXiv cs.AI
CORE: Conflict-Oriented Reasoning for General Multimodal Manipulation Detection
The article discusses a new framework called CORE, which stands for Conflict-Oriented Reasoning, designed to enhance…
arXiv cs.AI
SkillDAG: Self-Evolving Typed Skill Graphs for LLM Skill Selection at Scale
The paper introduces SkillDAG, a novel approach for selecting skills in large language models (LLMs) by modeling…
arXiv cs.AI
ToolGate: Token-Efficient Pre-Call Control for Tool-Augmented Vision-Language Agents
The paper introduces ToolGate, a system designed to improve the efficiency of tool-augmented vision-language agents.…
arXiv cs.AI
RelGT-AC: A Relational Graph Transformer for Autocomplete Tasks in Relational Databases
The paper introduces RelGT-AC, a new model designed for autocomplete tasks in relational databases. It enhances the…
arXiv cs.AI
TriEval: A Resource-Efficient Pipeline for LLM Bias, Toxicity, and Truthfulness Assessment
TriEval is a new pipeline designed to assess bias, toxicity, and truthfulness in large language models (LLMs)…
arXiv cs.AI
AUDITFLOW: Executable Symbolic Environments for Structured Financial Reporting Verification
The article discusses AuditFlow, a new framework designed for structured financial reporting verification. It utilizes…