22
Agentic Neural Architecture Search
Researchers propose a framework that combines large language model generation with neural architecture search to automate the design of neural networks in open-ended spaces.
Top stories in Research
22
WCog-VLA: A Dual-Level World-Cognitive Vision-Language-Action Model for End-to-End Autonomous Driving
The WCog-VLA framework integrates semantic world forecasting and generative world evolution to enable proactive decision-making in end-to-end autonomous driving.
22
Bridging Modal Isolation in Interleaved Thinking: Supervising Modality Transitions via Stepwise Reinforcement
This paper proposes a stepwise reinforcement learning approach to prevent text and image modalities from diverging during complex, interleaved multimodal reasoning tasks.
22
LoKA: Low-precision Kernel Applications for Recommendation Models At Scale
The LoKA framework introduces low-precision FP8 kernels tailored to the numerical sensitivities and communication demands of large-scale recommendation models.
22
The Illusion of Equivalency: Statistical Characterization of Quantization Effects in LLMs
A new study proposes a decision-level metric called correctness agreement to better capture the behavioral changes in large language models caused by post-training quantization.
22
Infinity-Parser2 Technical Report
Researchers have introduced Infinity-Parser2, a multimodal document parsing model trained using a controllable data-synthesis pipeline and reinforcement learning.
22
Theoria: Rewrite-Acceptability Verification over Informal Reasoning States
Researchers have developed Theoria, a verification architecture that audits large language model outputs by rewriting solutions into typed state transitions to ensure correctness.
22
Where do LLMs Fall Short in CBT-Guided Affective Reasoning?
This study evaluates the limitations of large language models in applying Cognitive Behavioral Therapy, finding that they struggle to transition from theoretical knowledge to practical, user-specific guidance.
22
When Thinking Hurts: Epistemic Signals in the Reasoning Chains of Visual Language Models
An empirical study of reasoning-capable vision-language models reveals distinct patterns in how their internal uncertainty behaves during multi-step thinking processes.
22
RetailBench: Evaluating Long-Horizon Autonomous Decision-Making and Strategy Stability of LLM Agents in Realistic Retail Environments
Researchers have introduced RetailBench, a simulation benchmark designed to evaluate the long-horizon decision-making and strategic stability of language model agents in a virtual retail environment.
22
A Reliability Assessment of LALM Audio Judges for Full-Duplex Voice Agents
Researchers evaluated the reliability of Gemini models acting as audio judges to score full-duplex voice agent conversations directly from raw waveforms.
22
PolyWorkBench: Benchmarking Multilingual Long-Horizon LLM Agents
Researchers have introduced PolyWorkBench, a benchmark designed to evaluate the performance of large language model agents on long-horizon tasks in multilingual environments.
22
A safety-oriented hypothetico-deductive framework for AI-assisted differential diagnosis
AegisDx is a new clinical reasoning framework that coordinates multiple specialized language models to reduce diagnostic errors and verify medical reasoning.
22
Multi-Agent Firewall Architecture for Privacy Protection of Sensitive Data in Interactions with Language Models
A new open-source multi-agent firewall architecture secures user interactions with large language models by intercepting and filtering sensitive data across web and programmatic traffic.
22
AUTOPILOT VQA: Benchmarking Vision-Language Models for Incident-Centric Dashcam Understanding
Researchers have developed AUTOPILOT-VQA, a benchmark designed to evaluate how well vision-language models can reason about safety-critical traffic incidents using dashcam footage.
22
The Phasor Transformer: Resolving Attention Bottlenecks on the Unit Circle
The Phasor Transformer block is introduced as a phase-native alternative to standard self-attention, representing sequence states on a unit-circle manifold to bypass quadratic bottlenecks.
22
The complexities of patient-centred conversational artificial intelligence
An analysis of over two thousand real patient-chatbot interactions led to the development of a patient simulator that models clinical, emotional, and conversational variations.
21
ContextSniper: AntTrail's Token-Efficient Code Memory for Repository-Level Program Repair
The paper introduces ContextSniper, a token-efficient code memory module designed to select precise evidence for repository-level program repair using language model agents.
21
Game Theory Driven Multi-Agent Framework Mitigates Language Model Hallucination
Researchers introduced G-Frame, a multi-agent framework based on game theory principles designed to reduce hallucinations in lightweight language models applied to scientific domains.
21
DeepTutor: Towards Agentic Personalized Tutoring
DeepTutor is an open-source agentic framework that combines citation-grounded tutoring with calibrated feedback to personalize educational interactions.
21
MentalHospital: A Virtual Environment for Evaluating Psychiatric Clinical Encounters
MentalHospital is a virtual simulation environment designed to evaluate how effectively large language models handle complete psychiatric clinical encounters.
21
From Legacy Documentation to OSCAL: An MCP-Based Agent Pipeline for Threat-Informed Continuous Compliance in Critical Infrastructure
Researchers developed a multi-agent pipeline based on the Model Context Protocol to convert natural-language documentation into standardized formats for continuous compliance in critical infrastructure.
21
SLORR: Simple and Efficient In-Training Low-Rank Regularization
Researchers developed SLORR, an in-training low-rank regularization method designed to make neural networks more compressible without requiring complex matrix decompositions.
21
Who Broke the System? Failure Localization in LLM-Based Multi-Agent Systems
The paper introduces a method for identifying and localizing failures within complex, multi-agent systems powered by large language models.