Papers

AIThis paper introduces 'The AI Scientist', a pipeline that automates the entire scientific research lifecycle. The system generates ideas, writes code, runs experiments, analyzes data, writes the full manuscript, and performs its own peer review, with the generated manuscript passing the first round of peer review for a top-tier machine learning conference workshop.

AIThis paper systematically organizes the latest techniques in LLM across four dimensions: pre-training, post-training, utilization, and evaluation. It also identifies key research issues and challenges such as theoretical foundations, efficient scaling, alignment, and agentic capabilities.

AIAxiom is an RNN that constructs a unitary transition matrix via a product of Householder reflections, enabling lossless processing of sequences of arbitrary length. It achieves 76.5-99.9% accuracy on the delayed copy task with 13x fewer parameters than LSTM, and when attached to GPT-2, it retrieves facts from 7 chunks back with 62.3% accuracy.

AIWe prove in four stages that the token embedding layer is the geometric foundation of transformer attention. Fourier-based training techniques (PFFT, FGP) and No-Q attention preserve the embedding's geometric authority and achieve lightweight performance.

AIWe propose Cognitive Impedance Matching Theory (CIMT), a compiler theory that enhances the capabilities of a fixed LLM model through world-side interface, validation, repair, and audit design. Human evaluators or LLM judges are not treated as privileged evaluators but modeled as fallible measurement channels, providing a framework to prove system reliability using only observable atomic data.

AICo-Scientist is a multi-agent AI system built on Gemini that assists scientists by generating and refining novel research hypotheses based on their objectives and existing evidence. It uses an asynchronous task execution framework and a tournament evolution process to improve hypothesis quality, demonstrating practical value in biomedical applications like drug repurposing.

AIA field experiment with 758 knowledge workers at a global consulting firm examined GPT-4's impact on task performance. The study found that while AI significantly boosted productivity and quality on most tasks, it hindered performance on a complex managerial task, revealing a 'jagged technological frontier'.

AIThis paper presents an interface theory that distinguishes genuine improvement from illusion in AI systems that change their own evaluation criteria. It develops a mathematical framework ensuring the stability of self-improvement loops through replayable observable interfaces and certified stable gains.

AIThis study quantitatively evaluates two specialized clinical AI tools against three frontier general-purpose LLMs across three medical benchmarks. The frontier LLMs outperformed the clinical AI tools in all evaluations, which performed comparably to an automated Google Search AI Overview on the real clinical queries benchmark.

AIProposes the Spiral Time framework, which represents time as 2D spiral coordinates (radius=trend, angle=seasonality). In LSTM and Transformer experiments, it improves MAPE by 83% over scalar time, with monotonic performance gains across all experiments.

AIThis study analyzes the practical effectiveness of differential privacy (DP) guarantees when adapting large language models (LLM) for sensitive applications. The researchers benchmark privacy risks using state-of-the-art attacks while systematically varying the adaptation data distribution, identifying key factors for achieving practical privacy protection.

AIThis paper provides a comprehensive survey of Multimodal Large Language Models (MLLMs) focused on vision-language tasks such as image captioning and visual question answering. It examines MLLM architectures, training pipelines, and practical applications, while highlighting fundamental constraints like information bottlenecks and data-processing limits.

AIQwen-Scope open-sources 14 groups of sparse autoencoders (SAEs) for Qwen3/3.5 series models and demonstrates their use in inference-time steering, evaluation analysis, data-centric workflows, and post-training optimization, showing that SAEs can serve as practical interfaces for model development beyond post-hoc analysis.

AIThis study analyzed 7.3 million academic articles from 2020 to 2025 to track the widespread adoption of large language models (LLMs) in scholarly research. The analysis found that by 2025, an estimated 57% of published articles showed evidence of LLM influence, with significant variation in adoption rates across regions, institutional ranks, and academic disciplines.

AIThis research identifies necessary fixes to scale Schedule-Free Learning for large language models. The proposed ScheduleFree+ method significantly outperforms Warmup-Stable-Decay schedules, especially in long-duration training.

AIThis research proposes 'natural identifiers (NIDs)' to address the challenges of privacy auditing for large language models (LLMs). NIDs are structured random strings naturally occurring in training datasets, enabling post-hoc differential privacy auditing without retraining and dataset inference without needing a private held-out dataset.

AIThis study proposes a variational recurrent neural network (VRNN) that integrates latent random variables into the hidden state of a recurrent neural network (RNN). Experiments on speech and handwriting data demonstrate that VRNN better models the variability of structured sequential data than existing models.

AIWe consider a partial observability model in bandit problems where the learner can observe losses of some other actions in addition to its own. We propose the first algorithm that guarantees near-optimal regret without prior knowledge of the observation system, and show that the implicit exploration strategy is more efficient computationally and information-theoretically than existing methods.

AITo address the difficulty of model selection in deep unsupervised domain adaptation (Deep UDA), we propose Deep Embedded Validation (DEV), which embeds adapted feature representations into the validation procedure to provide an unbiased estimator of the target domain risk. Variance is reduced using the control variate technique, and effectiveness is demonstrated theoretically and empirically.

AIChatlaw is a multi-agent legal assistant specialized in the Chinese legal system, using a Role-Aligned Mixture-of-Experts (RA-MoE) architecture. It achieves a 7.73% improvement in accuracy over GPT-4 on the LawBench benchmark and an 11-point higher score on the legal professional exam.

AIThis paper systematically categorizes and analyzes graph neural network (GNN) models that address heterophily in graphs. It summarizes various approaches to overcome the limitations of existing GNNs that assume homophily and suggests future research directions for heterophilic graph learning.

AIThis paper presents MicroGrowAgents, an AI-driven, agent-based system that automates the design of optimized microbial growth media by integrating knowledge graphs, metabolic modeling, and optimal experimental design. The system uses specialized agents to query biological knowledge, mine literature, and generate statistically optimal experiments, aiming to reduce experimental burden and accelerate the discovery of growth-promoting conditions.

AIThis paper proposes that the universe operates as a necessity system, where organizational constraint drives structural configurations toward the golden ratio through a recursive mechanism identical to the Fibonacci sequence. The theory aims to explain the cosmological constant without fine-tuning and presents five independent empirical pillars of support, including nuclear morphometry and high-redshift galaxy observations.

AIReveals that Multi-modal Large Language Models (MLLMs) risk leaking sensitive information embedded in images, and presents a comprehensive dataset MM-Privacy for evaluating these risks. Experiments confirm that various MLLMs expose personal information across different tasks, and task inconsistency increases privacy risks.

AIThis study examines the impact of weakened patent protection following the Alice Corp. vs. CLS Bank decision on firms' innovation, competition, acquisitions, lawsuits, and employment agreements. It uses large language models (LLMs) to identify the potential exposure of firms' patent portfolios to the Alice decision and investigates the resulting unequal impacts.

AIThis survey addresses the multilayered challenges of deploying large language models on edge hardware, including compression, compiler behavior, and system-level trade-offs. It provides a deployment-centric taxonomy of compression strategies, analyzing their interaction with hardware toolchains and highlighting limitations in current benchmarking suites.

AIAccurate carbon emission forecasting in power distribution networks is a critical challenge due to the integration of large-scale electric vehicles and renewable energy sources. This paper proposes CarbonGPT, a model that uses a causal encoder and a meta causal graph dictionary to address spurious correlations and enhance LLM-based prediction.

AIThis survey systematically analyzes methods for integrating external knowledge into LLMs to address hallucinations and knowledge gaps. It discusses parametric and non-parametric approaches to improve reasoning and factual accuracy in domain-specific tasks.

AIThis paper proposes BRIDGE, a benchmark for evaluating large language models on understanding real-world clinical practice texts. The benchmark includes various clinical tasks such as disease diagnosis, treatment planning, and patient status summarization.

AIThis research analyzes which brands are recommended within specific categories and the concentration of their ownership across large language models. The authors propose three exploratory metrics—Category Ownership Index (COI), Competitive Vacuum Index (CVI), and Displacement Score (DS)—and conduct an empirical analysis across 3 models, 5 industries, and 250 category queries, finding results that challenge a strong winner-takes-all narrative.

AIThis paper presents a large-scale empirical study evaluating ten LLMs with seven prompting strategies against nine traditional techniques for software vulnerability analysis. The study finds that existing prompting strategies often lead to LLMs underperforming traditional methods, and proposes a vulnerability-specific chain-of-thought prompting (VSP) to improve performance.

AIThe human brain represents sentence meaning differently depending on word order, while large language models (LLMs) are less sensitive to order and rely more on context. fMRI experiments and model analysis revealed differences in sentence processing between humans and LLMs.

AIThis paper provides a systematic survey of how Large Language Models (LLMs) are utilized to address Operations Research (OR) problems. It analyzes the roles of LLMs in OR, such as model formulation, algorithm design, and solution verification, along with practical applications and benchmark datasets.

AIThis study analyzes how state-level media control influences the outputs of large language models. It reveals correlations between internet censorship and media control indicators across countries and biases in LLM outputs.

AIThis is the first survey paper to comprehensively examine the pretraining data exposure problem in LLMs from the perspectives of data contamination and membership inference. It systematizes attack and defense methodologies and suggests future research directions for evaluation integrity and privacy protection.

AIThis paper analyzes how the capabilities of Large Language Models (LLMs) emerge through the lens of emergence concepts from complexity science. The study reviews several approaches to quantifying emergence and questions whether LLMs possess emergent intelligence.

AIOCR and multilingual text understanding are major failure modes of multimodal LLMs. The proposed framework combines synthetic data generation, OCR-aware fine-tuning with LoRA, and visual chain-of-thought prompting to significantly improve OCR completeness and multilingual translation accuracy in complex real-world images.

AIA comprehensive survey analyzing architectures, evaluation methods, and safety of LLM-based network operations and AIOps agents. It organizes related research around autonomy hierarchy, tool scope, evidence traces, and assurance contracts, emphasizing the need for workflow-centered evaluation beyond static QA.

AIWe propose a method for LLM-based probability estimation that introduces hierarchical factor structures and causal Bayesian networks to reduce 'unknown' predictions in sparse factor spaces. Experiments show that compared to direct LLM inference, it significantly reduces unknown predictions, provides more reliable probabilities, and reduces time and token costs.

AIThis paper traces how an organization's declared purpose is translated into the criteria, structures, and signals that govern real decisions. It demonstrates through case studies that meaning is systematically reformulated as it moves through governance systems.

Version 2 (2026-08-04). Revised to the NLL Universal Paper Format v6. The original text is retained in full; nothing has been deleted. Corrections appear as marked blocks placed at the section that carries the claim, and each one states what the paper said, what the data show, th…

AIThis paper systematically reviews empirical studies on the application of Large Language Models (LLMs) in education. The studies primarily apply LLMs to student support, teacher assistance, and automated assessment, reporting benefits such as improved learning outcomes and personalized feedback. However, the review also highlights significant challenges, including hallucination, bias, potential for academic dishonesty, and privacy concerns.

AIThe Encyclotron is a reproducible instrument that quantitatively measures the degree to which AI summarization systems distort or simplify scholarly knowledge. It calculates the gap between scholarly and retrieval graphs using variables such as compression loss, invention, and distortion, and tracks changes over time.

AIThis study proposes a novel strategy combining transcriptomic data and machine learning to predict the function of oxidative phosphorylation (OXPHOS) genes in C. elegans. By integrating supervised learning ensembles and cluster-based inference, it identifies promising new candidate OXPHOS genes that were previously unannotated.

AINiCLIP is a contrastive language-image pretrained model trained on over 23,000 neuroscientific articles to predict cognitive tasks, concepts, and domains from brain activation patterns. Performance is optimized with full-text articles and a curated cognitive ontology, showing accurate predictions on group-level activation maps but limitations on noisy subject-level maps.

AIThis study developed a machine learning framework to automatically recognize and quantify multiple features of axons and myelin from electron microscopy images. When applied to spinal cord fibers in variably hypomyelinated mice, it demonstrated that reductions in myelin sheath thickness and length correlate with changes in mitochondrial density and periaxonal area.

AIWe introduce the concept of weighted rules under the stable model semantics following the log-linear models of Markov Logic. This enables resolving inconsistencies, ranking stable models, assigning probabilities, and applying statistical inference.

AIProposes a new theoretical framework called Semantic Physics, exploring the convergence horizon of information theory and semantics through concepts of semantic saturation and ontology competition. Introduces original concepts such as compression survival and semantic dark matter to analyze the limits of semantic processing in AI systems.

AIThis paper proposes a new approach using graph data to address the lack of spatial reasoning abilities in LLMs. It envisions a future where search engines integrate with LLMs to answer complex spatial questions through graph-enhanced reasoning for domains like urban planning and civil engineering.

AIRapid decision-making reduces the time to detect discrepancies between intent and actual behavior, allowing errors to accumulate. This paper introduces the concept of 'Translation Half-Life' to measure how quickly interpretive shifts become embedded in an organization before they can be corrected.

AIPresents an ontology defining information as the structural pattern of energy differences (energy texture). Through the six pillars of Energy-Efficiency Theory (EET), it explains why information is inherently constrained by energy.

AIThe CHOPSTICK architecture is a framework for the AGI era that treats AI outputs only as candidate materials subject to human review and approval, separating human discretion and responsibility. It redefines human intelligence not as a subordinate form of AGI but as a human-centered discretionary structure, preventing AI outputs from being mistaken for human judgment or decisions.

AIProposes the Provenance Erasure Rate (PER) metric to measure the proportion of claims in AI-composed outputs that lack explicit attribution to original sources. PER focuses on source visibility rather than truthfulness, and a case study of a Google AI Overview that fabricated a false biography from real poetry fragments showed PER=1.0.

AIDeepMind's 'AI Agent Traps' paper classifies adversarial influence on agents into six categories, but this is merely 'meaning feudalism' that presupposes platform sovereignty and treats external influence as attack. The paper omits legitimate environmental influence (commons repair), and the author proposes S4 (Legitimate Influence Blindness) as a new shadow.

Abstract Accurate medical image segmentation is critical for early medical diagnosis. Most existing methods are based on U-shape structure and use element-wise addition or concatenation to fuse different level features progressively in decoder. However, both the two operations ea…

AIMeaning changes leave traceable traces before they appear in performance indicators. This paper derives Translation Coherence as a measurable property from governance artefacts and allocation patterns, making alignment empirically observable.

AITo address the problem of meaning degradation when AI accelerates decision-making, this paper proposes a closed-loop architecture that preserves human interpretive control while supporting AI analysis. By constraining how intent is translated into criteria, metrics, and allocation rules, it prevents drift within the system and enables traceable decisions.

AIProposes the concept of the 'Operating Spine' as a minimal causal architecture that can identify misalignment between intent, decision criteria, and outcomes within an organization's internal structure. Instead of post-hoc analysis, it makes drift observable within the decision system.

AIIn the AGI era, a structural methodology for multi-layer cross-verification of AI outputs before linking them to roles, responsibilities, evidence, etc. The number and composition of verification layers vary according to output type, domain risk, role sensitivity, etc.