arXiv.org
Information Discernment in Large Language Models
Do they weigh information appropriately -- updating more for reliable sources (source discernment) and more when…
Hub / Ai Research
ai-research · WeSearch
Ai Research news.
The latest AI and machine-learning research — new papers, model architectures, transformers, reinforcement learning, benchmarks, and lab announcements.
Lead story
arXiv.org
NEXUS: Structured Runtime Safety for Tool-Using LLM Agents
Researchers have introduced NEXUS, a structured runtime safety monitor for tool-using LLM agents, which applies a formal intervention policy to ensure safe execution of high-impact actions. NEXUS combines deterministic…
Latest in Ai Research
arXiv.org
Benchmarking Confidential GPU Inference on NVIDIA H100 under Intel TDX
However, the performance cost of enabling confidential execution for GPU-accelerated large language model serving…
arXiv.org
FormulaSPIN: Self-Play Fine-Tuning for Natural Language to Spreadsheet Formula Generation
Existing approaches rely on static supervised data, which quickly saturates on limited annotations. In this paper, we…
arXiv.org
OpenEvoShield: Dual Non-Stationary Continual Defense for Open-World Multi-Agent System Attacks
Unlike static threats, these attacks are doubly dynamic: adversaries refine injection strategies against deployed…
arXiv.org
Hybrid LSTM-Graph Neural Framework for Robust Financial Fraud Detection and Adversarial Resilience
This paper proposes FraudShield AI, a hybrid framework that integrates Long Short-Term Memory (LSTM) networks with…
arXiv.org
FineServe: A Fine-Grained Dataset and Characterization of Global LLM Serving Workloads
Achieving low latency and high throughput under volatile demand requires deep understanding of real-world serving…
Google News
How AI Is Helping States Cut Through Decades of Red Tape - Stanford HAI
How AI Is Helping States Cut Through Decades of Red Tape Stanford HAI
arXiv.org
Spectral-LSH: Sub-Quadratic Prompt Compression via Krylov-Projected Locality-Sensitive Hashing
We propose Spectral-LSH, a training-free prompt compression method that operates before the prompt enters the language…
arXiv.org
Rethinking Uncertainty Evaluation in Large Language Models
What we actually need is for LLM confidence estimates to satisfy the conditions required of coherent probabilistic…
arXiv.org
Geometry-Guided Constraint Learning for LLM Safety Classification
We show that sparse autoencoder (SAE) feature extraction resolves this: K=2 becomes optimal for 12/14 categories on…
arXiv.org
Logic-Guided Data Extraction with Answer Set Programming and Large Language Models
This paper proposes a logic-guided data extraction framework combining LLM-based extraction with Answer Set…
arXiv.org
Statistically Grounded Sparse-Feature Interventions for Activation-Space Control in Large Language Models
Computer Science > Artificial Intelligence arXiv:2607.19364 (cs) [Submitted on 5 Jun 2026] Title:Statistically…
arXiv.org
AdaRoPE: Not All Attention Heads Should Rotate and Scale Equally
Using simplified retrieval tasks and length generalization scenarios, we show -- both empirically and theoretically --…
arXiv.org
GraphContainer: A Unified Platform for Comparing and Debugging Graph RAG Methods
However, existing approaches remain highly fragmented and incompatible. The structural heterogeneity of graph formats…
arXiv.org
Lifted Representation Hypothesis in Language Models
However, it remains unclear how these structures are stored, selected, and revised. To study this process, we propose…
arXiv.org
Euclean: Automated Geometry Problem Formalization with Unified Verification in Lean
This split increases the trusted computing base and hinders unified model development. Existing geometry-in-Lean…
arXiv.org
Mitigating Scaffolding Collapse in Socratic Tutors via Representation Alignment
Prior defenses primarily constrain observable responses through prompting, preference optimization, or filtering,…
arXiv.org
Beyond Tracking or Shortcut: Composition-Bounded Predictive States in Poker Autoregressive Models
This paper studies this ambiguity in a no-range Limit Hold'em autoregressive model trained only on action and value…
arXiv.org
Profile-Graph Memory for LLM Agents: Implicit Cross-Entity Traversal through Narrative Profiles
First, we introduce MemHop, a multi-hop memory benchmark of 1,000 questions at hop depths 1-5 across 10 social-network…
arXiv.org
LISA: Linear-Indexed Sparse Attention for Efficient Long-Context Reasoning
However, the O(n^2) computational complexity of standard self-attention causes inference costs to grow sharply with…
arXiv.org
Stochastic Primal-Dual Decoding for Multiobjective Generative Recommender Systems
In practice, recommendation often involves constructing slates -- ordered lists of items -- that must satisfy multiple…
arXiv.org
OriginBlame: Record- and Token-Level Data Provenance for AI Training Datasets
Existing provenance systems operate at file or dataset level, forcing catastrophic over-deletion. We present ob, a…
arXiv.org
ProofCouncil: An LLM Agent for Solving Open Mathematical Problems
However, their performance can be further improved through agentic workflows tailored to real-world mathematical…
arXiv.org
Communication-Efficient Digital-Twin Coordination for Heterogeneous LLM Embodied Agents over Computing Power Networks
Computer Science > Artificial Intelligence arXiv:2607.09330 (cs) [Submitted on 10 Jul 2026]…
arXiv.org
How Does Bayesian Causal Discovery Fail? Characterising Structural Consequences in Linear Gaussian Networks under Latent Confounding
The paper investigates how Bayesian causal discovery behaves when latent confounding is present in linear Gaussian…
arXiv.org
LongMedBench: Benchmarking Medical Agents for Long-Horizon Clinical Decision-Making
Prior evaluations of LLM-based medical agents have largely emphasized short-context knowledge QA and tool use.…
arXiv.org
Fictional Worldbuilding: Multi-Agent LLM Collaboration with Hierarchical Context Compression and Iterative Review
Two experiments across 20 diverse worldbuilding tasks, using GPT-OSS 120B and DeepSeek v3.2 as LLM backends,…
arXiv.org
OpenProver: Agentic and Interactive Theorem Proving with Lean 4
OpenProver integrates a Planner-Worker-Verifier architecture inspired by recent ATP agentic systems such as Aletheia.…
arXiv.org
Toward Auditable AI Scientists: A Hypothesis Evolution Protocol for LLM Agents
Equipped with broad knowledge, flexible reasoning, and tool use, they have the potential to autonomously explore and…
arXiv.org
Scoped Verification for Reliable Long-Horizon Agentic Context Evolution under Distribution Shift
Computer Science > Artificial Intelligence arXiv:2607.09175 (cs) [Submitted on 10 Jul 2026] Title:Scoped Verification…
arXiv.org
KV-PRM: Efficient Process Reward Modeling via KV-Cache Transfer for Multi-Agent Test-Time Scaling
However, existing PRMs are text-based: they re-encode the entire trajectory text from scratch. In long multi-agent…
arXiv.org
Neuro-Agentic Control: A Deep Learning-based LLM-Powered Agentic AI Framework for Controlling Security Controls
While Large Language Models (LLMs) have strong semantic reasoning abilities to assist in decision support, their…
arXiv.org
ARCANA: A Reflective Multi-Agent Program Synthesis Framework for ARC-AGI-2 Reasoning
ARCANA decomposes each task into iterative perception, hypothesis generation, symbolic execution, and reflective…
arXiv.org
L-MAD: A Systematic Evaluation of Multi-Agent Debate Structures in Legal Reasoning
The paper presents the Legal Multi-Agent Debate (L-MAD) framework for evaluating debate structures in legal textual…
arXiv.org
MedRealMM: A Real-World Multimodal Benchmark for Chinese Online Medical Consultation
Many rely on synthetic conversations or patient simulators, omit patient-uploaded medical images, or evaluate…
arXiv.org
A Formalization of the Mean-Field Derivation of the Vlasov Equation: AI-Assisted Lean Formalization as a Strategy Game
Computer Science > Artificial Intelligence arXiv:2607.08986 (cs) [Submitted on 9 Jul 2026] Title:A Formalization of…
arXiv.org
Long-Horizon-Terminal-Bench: Testing the Limits of Agents on Long-Horizon Terminal Tasks with Dense Reward-Based Grading
However, existing terminal benchmarks largely focus on simple problems that finish within minutes and are evaluated…
arXiv.org
GATS: Graph-Augmented Tree Search with Layered World Models for Efficient Agent Planning
We present \textbf{GATS} (Graph-Augmented Tree Search), a planning framework that combines systematic UCB1-based tree…
arXiv.org
CogniConsole: Externalizing Inference-Time Control as a Formal Abstraction for Reliable LLM Interactions
We challenge this by demonstrating that reliability is significantly influenced by \emph{inference-time control} --…
arXiv.org
Interval Certifications for Multilayered Perceptrons via Lattice Traversal
In particular, we show that the adversarial robustness problem can be reduced to a lattice traversal problem. Each…
arXiv cs.AI
Augmenting Fundamental Analysis with Large Language Models: A RAG-Based System for Generating Investor Briefs
Securities and Exchange Commission (SEC) which can be found in EDGAR. We were preprocessing those data and than…
arXiv cs.AI
Event Stream based Multi-Modal Video Anomaly Detection: A Benchmark Dataset and Algorithms
To address these limitations, we propose EVAD, an event enhanced VAD framework that jointly exploits conventional…
arXiv cs.AI
Integrating Large Language Models and Graph Convolutional Networks for Semi-Supervised Image Classification
Therefore, semi-supervised approaches such as Graph Convolutional Networks (GCNs), which learn from both labeled and…
arXiv cs.AI
Beyond Metadata: CAPRA for Hidden Subgroup Analysis under Missing Metadata in Medical Imaging
Once those metadata disappear, clinically critical failure modes can be masked by strong aggregate performance, and…
arXiv cs.AI
A Coreset Selection Framework with Ensemble Aggregation for Image Classification
Selecting representative training subsets, however, remains challenging: individual sample contributions are unclear,…
arXiv cs.AI
PRecG: Legal Precedent Retrieval with Graph Neural Networks and Rhetorical Role Segmentation
Current approaches for automatic precedent retrieval map legal documents to a low-dimensional semantic space and…
arXiv cs.AI
OmniMapBench: Benchmarking Visual-Centric Reasoning on Diverse Map Documents
A critical limitation is identified in many document understanding benchmarks: visual content is often reducible to…
arXiv cs.AI
Inside the Skill Market: From Software Engineering Activities to Reusable Agent Skills
SE) has continuously evolved through increasingly powerful forms of reuse, from source code and libraries to…
arXiv cs.AI
On Locality and Length Generalization in Visual Reasoning
This makes human vision distinctly different from most popular computer vision models in use today, which input images…