Post-training methods have improved the performance and enhanced the reasoning capability for mainstream large language models (LLMs), but the same is challenging for Japanese LLMs to achieve due to the amount of resources required. Inspired by task vectors that extract the change of weights before and after training, specifically for a certain task, we obtain reasoning vectors from reasoning LLMs and apply them to Japanese LLMs to boost their performance. While the resources available present a challenge to improve Japanese LLMs, we present a simple and effective way to obtain high improvement and hope to inspire for other languages.

Sensorial language -- the language connected to our senses including vision, sound, touch, taste, smell, and interoception, plays a fundamental role in how we communicate experiences and perceptions. We explore the relationship between sensorial language and traditional stylistic features, like those measured by LIWC, using a novel Reduced-Rank Ridge Regression (R4) approach. We demonstrate that low-dimensional latent representations of LIWC features r = 24 effectively capture stylistic information for sensorial language prediction compared to the full feature set (r = 74). We introduce Stylometrically Lean Interpretable Models (SLIM-LLMs), which model non-linear relationships between these style dimensions. Evaluated across five genres, SLIM-LLMs with low-rank LIWC features match the performance of full-scale language models while reducing parameters by up to 80%.

There is a broad consensus that the inability to form long-term plans is one of the key limitations of current foundational models and agents. However, the existing planning benchmarks remain woefully inadequate to truly measure their planning capabilities. Most existing benchmarks either focus on loosely defined tasks like travel planning or end up leveraging existing domains and problems from international planning competitions. While the former tasks are hard to formalize and verify, the latter were specifically designed to test and challenge the weaknesses of existing automated planners. To address these shortcomings, we propose a procedure for creating a planning benchmark centered around the game called Countdown, where a player is expected to form a target number from a list of input numbers through arithmetic operations. We discuss how this problem meets many of the desiderata associated with an ideal benchmark for planning capabilities evaluation. Specifically, the domain allows for an intuitive, natural language description for each problem instance, it is computationally challenging (NP-complete), and the instance space is rich enough that we do not have to worry about memorization. We perform an extensive theoretical analysis, establishing the computational complexity result and demonstrate the advantage of our instance generation procedure over public benchmarks. We evaluate a variety of existing LLM-assisted planning methods on instances generated using our procedure. Our results show that, unlike other domains like 24 Game (a special case of Countdown), our proposed dynamic benchmark remains extremely challenging for existing LLM-based approaches.

This paper introduces VisuCraft, a novel framework designed to significantly enhance the capabilities of Large Vision-Language Models (LVLMs) in complex visual-guided creative content generation. Existing LVLMs often exhibit limitations in maintaining high visual fidelity, genuine creativity, and precise adherence to nuanced user instructions when generating long-form texts. VisuCraft addresses these challenges by integrating a multimodal structured information extractor (E) and a dynamic prompt generation module (G). The extractor distills fine-grained visual attributes from input images into a rich, structured representation, which the dynamic prompt module then combines with user instructions to create highly optimized prompts for underlying LVLMs (e.g., LLaVA, InstructBLIP). Evaluated on the self-constructed ImageStoryGen-500K dataset using VisuGen Metrics (Visual Grounding, Creativity, and Instruction Adherence), VisuCraft consistently outperforms baseline LVLMs across tasks like story generation and poetry composition. Our results demonstrate remarkable improvements, particularly in creativity and instruction adherence, validating VisuCraft's effectiveness in producing imaginative, visually grounded, and user-aligned long-form creative text. This work unlocks new potential for LVLMs in sophisticated creative AI applications.

--Despite significant advancements, current large language models (LLMs) and vision-language models (L VLMs) continue to struggle with complex, multi-step, cross-modal common sense reasoning tasks, often exhibiting a lack of "deliberative thinking." They tend to rely on superficial associations rather than deep, chained inference, particularly when integrating visual information with abstract concepts. T o address this, we propose the Coherent Multimodal Reasoning Framework (CMRF), a novel approach that enhances L VLMs' common sense reasoning capabilities through an iterative, self-evaluating inference mechanism. CMRF mimics human problem-solving by decomposing complex queries, generating step-by-step inferences, and self-correcting errors. Coupled with an Adaptive Iterative Refinement strategy, CMRF systematically refines its reasoning paths. Built upon LLaV A-1.6-34B and trained on a novel Multimodal Daily Activity Reasoning (MDAR) dataset, CMRF achieves state-of-the-art performance among open-source L VLMs on challenging benchmarks like VCR, A-OKVQA, and DailyLife-MRC. Extensive ablation studies and human evaluations confirm the critical contributions of each module and the effectiveness of iterative refinement in fostering more coherent and accurate reasoning. The remarkable advancements in large language models (LLMs) [1], [2] and vision-language models (L VLMs) have revolutionized various aspects of artificial intelligence, demonstrating unprecedented capabilities in understanding, generating, and processing information across modalities [3]. These models excel in tasks ranging from complex question answering to creative content generation, largely due to their extensive pre-training on vast amounts of data.

Privacy-preserving machine learning (PPML) is critical to ensure data privacy in AI. Over the past few years, the community has proposed a wide range of provably secure PPML schemes that rely on various cryptography primitives. However, when it comes to large language models (LLMs) with billions of parameters, the efficiency of PPML is everything but acceptable. For instance, the state-of-the-art solution for confidential LLM inference represents at least 10,000-fold slower performance compared to plaintext inference. The performance gap is even larger when the context length increases. In this position paper, we propose a novel framework named Agentic-PPML to make PPML in LLMs practical. Our key insight is to employ a general-purpose LLM for intent understanding and delegate cryptographically secure inference to specialized models trained on vertical domains. By modularly separating language intent parsing - which typically involves little or no sensitive information - from privacy-critical computation, Agentic-PPML completely eliminates the need for the LLMs to process the encrypted prompts, enabling practical deployment of privacy-preserving LLM-centric services.

--Retrieval-Augmented Generation (RAG) has emerged as a powerful approach to boost the capabilities of large language models (LLMs) by incorporating external, up-to-date knowledge sources. However, this introduces a potential vulnerability to knowledge poisoning attacks, where attackers can compromise the knowledge source to mislead the generation model. One such attack is the PoisonedRAG in which the injected adversarial texts steer the model to generate an attacker chosen response for a target question. In this work, we propose novel defense methods, FilterRAG and ML-FilterRAG, to mitigate the PoisonedRAG attack. First, we propose a new property to uncover distinct properties to differentiate between adversarial and clean texts in the knowledge data source. Next, we employ this property to filter out adversarial texts from clean ones in the design of our proposed approaches. Evaluation of these methods using benchmark datasets demonstrate their effectiveness, with performances close to those of the original RAG systems. KEY challenge associated with large language models (LLMs) [1]-[3] is their tendency of becoming outdated and struggling to integrate the most recent knowledge [4], [5]. This fundamental short-coming is addressed by the recent emergency of retrieval-augmented generation (RAG) [6]-[9].

Automation in software engineering increasingly relies on large language models (LLMs) to generate, review, and assess code artifacts. However, establishing LLMs as reliable evaluators remains an open challenge: human evaluations are costly, subjective and non scalable, while existing automated methods fail to discern fine grained variations in artifact quality. We introduce REFINE (Ranking Evaluators for FIne grained Nuanced Evaluation), an automated framework for benchmarking LLM based evaluators across software engineering tasks. REFINE comprises of two modules: Hierarchy Dataset Builder applies novel generation techniques to automatically synthesize artifacts with progressively reduced quality, and Evaluator Tester quantifies each candidate evaluator configuration by measuring how closely its rankings align with expected ordering. A key feature of REFINE is controllability: users can tune the granularity of degradation to progressively refine evaluator configurations, from coarse filtering to stress testing on subtle quality gaps. While the methodology is general, we focus on coding tasks reflecting the practical demands in our production setting. REFINE was integrated into IBM's internal development workflows and applied to code generation, translation, and summarization for COBOL, an enterprise critical programming language, using industrial data. It was used to identify LLM as a Judge configurations that lifted alignment scores from below $0.7$ to above $0.9$ in some coding tasks. These nuance sensitive evaluators are now actively used by model training teams to support model release decisions.

Conversational LLMs have been widely adopted by domain users with limited programming experience to solve domain problems. However, these users often face misalignment between their intent and generated code, resulting in frustration and rounds of clarification. This work first investigates the cause of this misalignment, which dues to bidirectional ambiguity: both user intents and coding tasks are inherently nonlinear, yet must be expressed and interpreted through linear prompts and code sequences. To address this, we propose direct intent-task matching, a new human-LLM interaction paradigm that externalizes and enables direct manipulation of the LLM understanding, i.e., the coding tasks and their relationships inferred by the LLM prior to code generation. As a proof-of-concept, this paradigm is then implemented in NeuroSync, which employs a knowledge distillation pipeline to extract LLM understanding, user intents, and their mappings, and enhances the alignment by allowing users to intuitively inspect and edit them via visualizations. We evaluate the algorithmic components of NeuroSync via technical experiments, and assess its overall usability and effectiveness via a user study (N=12). The results show that it enhances intent-task alignment, lowers cognitive effort, and improves coding efficiency.

Radiological imaging is central to diagnosis, treatment planning, and clinical decision-making. Vision-language foundation models have spurred interest in automated radiology report generation (RRG), but safe deployment requires reliable clinical evaluation of generated reports. Existing metrics often rely on surface-level similarity or behave as black boxes, lacking interpretability. We introduce ICARE (Interpretable and Clinically-grounded Agent-based Report Evaluation), an interpretable evaluation framework leveraging large language model agents and dynamic multiple-choice question answering (MCQA). Two agents, each with either the ground-truth or generated report, generate clinically meaningful questions and quiz each other. Agreement on answers captures preservation and consistency of findings, serving as interpretable proxies for clinical precision and recall. By linking scores to question-answer pairs, ICARE enables transparent, and interpretable assessment. Clinician studies show ICARE aligns significantly more with expert judgment than prior metrics. Perturbation analyses confirm sensitivity to clinical content and reproducibility, while model comparisons reveal interpretable error patterns.