1071 papers
FlashPrefill is a novel framework that achieves ultra-fast long-context prefilling by combining instantaneous block-searching for dynamic sparse patterns with a thresholding mechanism to eliminate long-tail attention scores, delivering up to a 27.78x speedup on 256K sequences while maintaining efficiency on shorter contexts.
The paper proposes SPOT, a framework that enhances the efficiency and interpretability of large language model reasoning by compressing explicit Chain-of-Thought into latent pause tokens through span-level semantic alignment and a frozen-head decoding constraint, achieving higher accuracy with significantly fewer generated tokens.
This paper presents a multilingual audit revealing that while contemporary large language models generally align with public opinion on broad social issues across Asian regions, they consistently fail to accurately represent diverse religious viewpoints—particularly those of minority groups—and often amplify negative stereotypes, a problem that persists despite lightweight prompting interventions and remains undetected by standard bias benchmarks.
The paper introduces EpisTwin, a neuro-symbolic framework that overcomes data fragmentation in Personal AI by grounding generative reasoning in a verifiable Personal Knowledge Graph derived from multimodal data, utilizing an agentic coordinator for complex reasoning and visual refinement, and validated by a new synthetic benchmark called PersonalQA-71-100.
This paper presents an empirical study on adapting speech foundation models to low-resource Pacific Indigenous languages, revealing that while strategies like Low-Rank Adaptation offer initial success, they ultimately struggle with catastrophic forgetting and internal representational drift during sequential learning, highlighting the urgent need for robust adaptation frameworks that balance plasticity and stability.
This study evaluates ChatGPT's ability to generate AI pastiches of contemporary artworks by combining human feedback from twelve international artists with computational analysis, revealing a significant gap between superficial visual similarities and the lack of conceptual depth, dimensionality, and emotional resonance in the AI-generated results, thereby advocating for a multi-metric "style transfer dashboard" for more comprehensive evaluation.
The paper introduces SAHOO, a practical framework that employs a Goal Drift Index, constraint preservation checks, and regression-risk quantification to effectively monitor and control alignment drift while significantly improving performance in recursive self-improving systems across code, reasoning, and truthfulness tasks.
This paper proposes a transparent AI framework for Mathematical Entity Relation Extraction (MERE) that utilizes BERT to achieve 99.39% accuracy in identifying mathematical relationships and incorporates SHAP-based explainability to enhance model trust and interpretability for applications in automated problem solving and education.
This study introduces a new deontic-focused Wason Selection Task dataset to demonstrate that large language models, like humans, exhibit superior reasoning with deontic rules and display matching-bias-like error patterns, highlighting the domain-specific nature of their reasoning capabilities.
This paper presents a comprehensive comparative study of four major LLM-based paradigms for Automated Essay Scoring on IELTS Writing Task 2, revealing that a configuration combining k-SFT and RAG achieves the strongest overall performance with a 93% F1-Score while highlighting critical accuracy-cost-robustness trade-offs.
This paper investigates the accuracy and potential biases of Large Language Models in abductive reasoning by converting a syllogistic dataset to evaluate their ability to draw tentative conclusions from limited information, aiming to bridge the gap between machine and human cognition beyond formal deduction.
PONTE is a human-in-the-loop framework that enhances the reliability and personalization of Explainable AI narratives by modeling user adaptation as a closed-loop process involving preference modeling, grounded generation, and iterative verification to overcome the limitations of one-size-fits-all approaches and Large Language Model hallucinations.
The paper introduces NOBLE, a pretraining architecture that permanently augments transformer linear layers with learnable nonlinear low-rank branches (specifically using CosNet activation), achieving significant training efficiency and speedups across various models with minimal parameter and time overhead, though its benefits may be hindered by certain stochastic data augmentations.
COLD-Steer is a training-free framework that steers large language models by approximating the representational changes of gradient descent on in-context examples, achieving up to 95% steering effectiveness with 50 times fewer samples than existing methods.
The paper introduces Beyond Rows to Reasoning (BRTR), a multimodal agentic framework that overcomes the limitations of single-pass retrieval and context compression in enterprise spreadsheet analysis by employing an iterative tool-calling loop, achieving state-of-the-art performance on multiple benchmarks while maintaining full auditability.
This paper presents a multilingual context-aware ASR framework that integrates a frozen speech encoder with a decoder-only language model via a lightweight projection module and contrastive learning to align speech and contextual representations, achieving over 5% performance gains across 11 languages and 5 English dialects.
The KCLarity team's SemEval-2026 paper details their comparative evaluation of encoder-based and zero-shot decoder-only models for detecting political evasion, finding that while RoBERTa-large excelled on the public test set, GPT-5.2 demonstrated superior generalization on the hidden evaluation set.
This paper introduces EHRSQL, a practical text-to-SQL benchmark for electronic health records built from real hospital staff queries, which challenges models to handle diverse clinical needs, complex time expressions, and unanswerable questions to bridge the gap between research and healthcare deployment.
This paper introduces Contrastive Prompting, a simple zero-shot method that significantly enhances large language models' reasoning capabilities across arithmetic, commonsense, and symbolic tasks by instructing them to generate both correct and incorrect answers, often outperforming state-of-the-art prompting techniques without requiring hand-crafted examples.
This paper proposes INMS, an asynchronous multi-agent framework that establishes a shared conversational memory pool through real-time filtering and retrieval, enabling dynamic knowledge exchange and collective self-enhancement to significantly improve LLM-based agent performance in open-ended scenarios.