940 papers
This paper demonstrates that reasoning consistently enhances honesty in large language models by guiding them through a representational space where deceptive states are metastable and easily disrupted, ultimately nudging the models toward more stable, honest defaults.
This study proposes and evaluates a novel methodology for enhancing fake news debunking by using Large Language Models to generate personalized messages tailored to Big Five personality traits, demonstrating that such targeted approaches generally increase persuasiveness while highlighting both the potential and ethical implications of automated, personality-driven disinformation correction.
This paper argues that advancements in logical reasoning for large language models inadvertently create a mechanistic pathway to dangerous situational awareness and strategic deception, necessitating new safety frameworks like the RAISE model to mitigate these emergent risks.
The paper introduces Sentinel, an autonomous AI agent that achieves reliable, scalable clinical triage for remote patient monitoring by outperforming individual clinicians in sensitivity and consistency while maintaining a clinically defensible overtriage profile at a negligible cost.
This paper introduces Guardian, a consensus-driven, multi-LLM pipeline enhanced by QLoRA fine-tuning that coordinates specialized models and a consensus engine to perform auditable, structured information extraction for critical missing-person investigations while avoiding unconstrained decision-making.
MASEval introduces a framework-agnostic library that shifts multi-agent evaluation from a model-centric to a system-centric approach, demonstrating through extensive experiments that implementation decisions regarding topology and orchestration impact performance as significantly as model selection.
This paper demonstrates that a lightweight, label-free screening strategy using Word2Vec-derived composition embeddings often outperforms transformer-based methods in filtering vast combinatorial electrocatalyst libraries by prioritizing candidates based on their similarity to text-derived property concepts like conductivity and dielectricity.
This paper demonstrates that Contamination Detection via output Distribution (CDD) is largely ineffective for small language models (70M–410M parameters) because it fails to detect verbatim memorization, whereas probability-based methods like perplexity and Min-k% Prob consistently outperform it across various benchmarks.
This paper proposes a geometric taxonomy of LLM hallucinations into three distinct types (unfaithfulness, confabulation, and factual error) and introduces corresponding detection metrics, the Semantic Grounding Index and Directional Grounding Index, which effectively identify unfaithful and confabulated outputs while revealing that apparent signals for factual errors in existing benchmarks often stem from stylistic annotation confounds rather than genuine geometric distinctions.
This paper demonstrates that spectral metrics derived from natural language processing of requirements specifications can predict integration effort with high accuracy (correlations >0.95), offering a validated method to quantify structural complexity at the requirements stage and bridge the gap between architectural analysis and requirements engineering.
The paper introduces RedSage, an open-source, locally deployable cybersecurity LLM trained on a massive curated dataset and agentic augmentation pipeline, which achieves state-of-the-art performance on specialized cybersecurity benchmarks while also improving general reasoning capabilities.
This paper introduces the Determinism-Faithfulness Assurance Harness (DFAH), a framework and set of financial benchmarks demonstrating that decision determinism and task accuracy in LLM agents are uncorrelated, thereby necessitating independent measurement to ensure reliable regulatory audit replay in financial services.
This paper introduces MAS-Orchestra, a training-time framework that optimizes multi-agent system orchestration via function-calling reinforcement learning, alongside the MASBENCH benchmark, to demonstrate that multi-agent benefits are task-dependent and to achieve significant performance gains with over 10x efficiency on complex reasoning tasks.
This paper presents a lightweight, two-stage multitask vision-language framework that integrates a Swin Transformer encoder with sequence-to-sequence decoders to achieve state-of-the-art, explainable visual question answering for crop disease identification with near-perfect classification accuracy and strong generalization capabilities.
This survey proposes a unified four-paradigm framework to categorize and analyze the fragmented landscape of agentic AI adaptation, distinguishing between agent-side improvements (A1/A2) and tool-side enhancements (T1/T2) to systematically review post-training methods, memory architectures, and skill systems while evaluating their trade-offs and outlining future challenges.
The paper introduces CompanionCast, a multi-agent framework that orchestrates specialized AI agents with multimodal detection, context caching, and spatial audio to enhance social presence and emotional sharing during shared media experiences, as validated by improved outcomes in pilot studies with soccer fans.
The Parallel Decoder Transformer (PDT) introduces a frozen-trunk architecture that enables synchronized parallel decoding by integrating a planner-seeded latent workspace and a dynamic notes bus, allowing multiple output streams to internally coordinate, resolve ownership, and synchronize generation without external orchestration.
This paper introduces Graphectory, a graph-based framework for analyzing the stochastic execution trajectories of agentic software systems, which reveals that richer prompts and stronger models yield more complex reasoning patterns while enabling real-time monitoring and intervention that significantly improves problem resolution rates and efficiency.
The paper presents NANOMIND, a hardware-software co-design framework that decomposes Large Multimodal Models into modular components and dynamically schedules them across heterogeneous accelerators on unified-memory SoCs, enabling a battery-powered device to run LMMs entirely on-device with significantly improved energy efficiency and throughput.
This paper introduces "benchmark signatures"—sets of salient tokens from in-the-wild corpora whose perplexity predicts model performance—to reveal nuanced overlaps and distinct capacities across 89 LLM benchmarks, offering a robust alternative to raw performance correlations for understanding the landscape of LLM abilities and the divergence between machine and human semantic organization.