Papers

The increasing prevalence of Large Language Models (LLMs) demands effective safeguards for their operation, particularly concerning their tendency to generate out-of-context responses. A key challenge is accurately detecting when LLMs stray from expected conversational norms, manifesting as topic shifts, factual inaccuracies, or outright hallucinations. Traditional anomaly detection struggles to directly apply within contextual semantics. This paper outlines our experiment in exploring the use of Representation Engineering (RepE) and One-Class Support Vector Machine (OCSVM) to identify subspaces within the internal states of LLMs that represent a specific context. By training OCSVM on in-context examples, we establish a robust boundary within the LLM's hidden state latent space. We evaluate out study with two open source LLMs - Llama and Qwen models in specific contextual domain. Our approach entailed identifying the optimal layers within the LLM's internal state subspaces that strongly associates with the context of interest. Our evaluation results showed promising results in identifying the subspace for a specific context. Aside from being useful in detecting in or out of context conversation threads, this research work contributes to the study of better interpreting LLMs.

Autoregressive language models are next-token predictors and have been criticized for only optimizing surface plausibility (i.e., local coherence) rather than maintaining correct latent-state representations (i.e., global coherence). Because Theory of Mind (ToM) tasks crucially depend on reasoning about latent mental states of oneself and others, such models are therefore often thought to fail at ToM. While post-training methods can improve ToM performance, we show that strong ToM capability can be recovered directly from the base model without any additional weight updates or verifications. Our approach builds on recent power-sampling methods (Karan & Du, 2025) that use Markov chain Monte Carlo (MCMC) to sample from sharpened sequence-level (rather than token-level) probability distributions of autoregressive language models. We further find that incorporating annealing, where the tempered distribution is gradually shifted from high to low temperature, substantially improves ToM performance over fixed-temperature power sampling. Together, these results suggest that sampling-based optimization provides a powerful way to extract latent capabilities from language models without retraining.

Vision-Language Models (VLMs) integrate visual and textual knowledge into unified representations that increasingly underpin modern retrieval and recommendation systems. However, it remains unclear how reliably these models utilize their cross-modal knowledge when ranking multimodal items, and whether their knowledge grounding can be subverted. In this paper, we expose a fundamental vulnerability in how VLMs apply multimodal knowledge for product ranking: through Multimodal Generative Engine Optimization (MGEO), we show that an adversary can manipulate a VLM's ranking decisions by jointly crafting imperceptible image perturbations and fluent textual suffixes that exploit the model's internal cross-modal knowledge coupling. Using an alternating optimization strategy, MGEO targets the deep interactions between visual and linguistic representations within the VLM, achieving rank manipulations that substantially exceed those of unimodal attacks and heuristic baselines powered by strong commercial models. Our findings reveal that surface-level content quality is insufficient for rank promotion; instead, direct alignment with the model's internal knowledge utilization mechanism is required. These results raise important questions on the faithfulness and robustness of knowledge grounding in multimodal foundation models, and motivate future work on defense mechanisms for multimodal retrieval systems. Code is available at: https://github.com/glad-lab/MGEO

Recent progress in large language models (LLMs) has improved code generation, but most evaluations still test isolated, small-scale code (e.g., a single function) under default or unspecified software environments. As a result, it is unclear whether LLMs can reliably generate executable code tailored to a user's specific environment. We present the first systematic study of Environment-Aware Code Generation (EACG), where generated code must be functionally correct and directly executable under arbitrary software configurations. To enable realistic evaluation, we introduce VersiBCB, a benchmark that is multi-package, execution-verified, and deprecation-aware, capturing complex and evolving environments that prior datasets often overlook. Using VersiBCB, we investigate three complementary adaptation axes: data, parameters, and cache, and develop representative strategies for each. Our results show that current LLMs struggle with environment-specific code generation, while our adaptations improve environment compatibility and executability. These findings highlight key challenges and opportunities for deploying LLMs in practical software engineering workflows.

Recent advances in audio-aware large language models have shown strong performance on audio question answering. However, existing benchmarks mainly cover answerable questions and overlook the challenge of unanswerable ones, where no reliable answer can be inferred from the audio. Such cases are common in real-world settings, where questions may be misleading, ill-posed, or incompatible with the information. To address this gap, we present AQUA-Bench, a benchmark for Audio Question Unanswerability Assessment. It systematically evaluates three scenarios: Absent Answer Detection (the correct option is missing), Incompatible Answer Set Detection (choices are categorically mismatched with the question), and Incompatible Audio Question Detection (the question is irrelevant or lacks sufficient grounding in the audio). By assessing these cases, AQUA-Bench offers a rigorous measure of model reliability and promotes the development of audio-language systems that are more robust and trustworthy. Our experiments suggest that while models excel on standard answerable tasks, they often face notable challenges with unanswerable ones, pointing to a blind spot in current audio-language understanding.

Diffusion Language Models (DLMs) present a promising non-sequential paradigm for text generation, distinct from standard autoregressive (AR) approaches. However, current decoding strategies often adopt a reactive stance, underutilizing the global bidirectional context to dictate global trajectories. To address this, we propose Plan-Verify-Fill (PVF), a training-free paradigm that grounds planning via quantitative validation. PVF actively constructs a hierarchical skeleton by prioritizing high-leverage semantic anchors and employs a verification protocol to operationalize pragmatic structural stopping where further deliberation yields diminishing returns. Extensive evaluations on LLaDA-8B-Instruct and Dream-7B-Instruct demonstrate that PVF reduces the Number of Function Evaluations (NFE) by up to 65% compared to confidence-based parallel decoding across benchmark datasets, unlocking superior efficiency without compromising accuracy.

Evaluating conversational systems in multi-turn settings remains a fundamental challenge. Conventional pipelines typically rely on manually defined rubrics and fixed conversational context$-$a static approach that limits coverage and fails to capture the diverse, emergent behaviors of dialogue models. To address this, we introduce CoReflect (A Reflective Co-Evolution Framework for Improving Conversational Evaluation), which unifies dialogue simulation and evaluation into an adaptive, iterative process. CoReflect employs a conversation planner that generates structured templates to guide a user simulator through diverse, goal-directed dialogues. Subsequently, a reflective analyzer processes these dialogues to identify systematic behavioral patterns and automatically refine the evaluation rubrics. Crucially, the insights from the conversation analysis are fed back into the planner to update conversation templates for subsequent iterations. This co-evolution loop ensures that the complexity of test cases and the diagnostic precision of rubrics improve in tandem. By minimizing human intervention, CoReflect provides a scalable and self-refining methodology that allows evaluation protocols to adapt alongside the rapidly advancing capabilities of dialogue models.

This paper proposes a Dialect Identification (DID) approach inspired by the Connectionist Temporal Classification (CTC) loss function as used in Automatic Speech Recognition (ASR). CTC-DID frames the dialect identification task as a limited-vocabulary ASR system, where dialect tags are treated as a sequence of labels for a given utterance. For training, the repetition of dialect tags in transcriptions is estimated either using a proposed Language-Agnostic Heuristic (LAH) approach or a pre-trained ASR model. The method is evaluated on the low-resource Arabic Dialect Identification (ADI) task, with experimental results demonstrating that an SSL-based CTC-DID model, trained on a limited dataset, outperforms both fine-tuned Whisper and ECAPA-TDNN models. Notably, CTC-DID also surpasses these models in zero-shot evaluation on the Casablanca dataset. The proposed approach is found to be more robust to shorter utterances and is shown to be easily adaptable for streaming, real-time applications, with minimal performance degradation.

Parameter-efficient fine-tuning (PEFT) methods are widely used to adapt large language models (LLMs) to downstream tasks and are often assumed to improve factual correctness. However, how the parameter-efficient fine-tuning methods affect hallucination behavior remains insufficiently understood, especially on QA datasets. In this work, we systematically investigate the impact of PEFT on hallucination detection through a comprehensive empirical study across three open-weight LLM backbones and three fact-seeking QA benchmarks. For each model, we evaluate performance using seven unsupervised hallucination detection methods spanning three complementary approaches: semantic consistency based detectors, confidence based detectors, and entropy based detectors. This multifaceted evaluation enables us to characterize how PEFT reshapes uncertainty across different detection paradigms. In conclusion, our experimental results show that PEFT consistently strengthens hallucination detection ability, substantially improving AUROC across a wide range of hallucination detectors. Besides, further analyses using linear probes and representation diagnostics indicate that PEFT methods primarily reshapes how uncertainty is encoded and surfaced, comparing with injecting new factual knowledge into the models.

Large Language Models (LLMs) are increasingly deployed for open-domain question answering, yet their alignment with human perspectives on temporally recent information remains underexplored. We introduce RECOM (Reddit Evaluation for Correspondence of Models), a benchmark dataset of 15,000 recent Reddit questions from September 2025 paired with community-derived reference answers. We investigate how four open-source LLMs (Llama3.1-8B, Mistral-7B, Gemma-2-9B, and GPT-OSS-20B) respond to these questions, evaluating alignment using lexical metrics (BLEU, ROUGE), semantic similarity (BERTScore, MoverScore, cosine similarity), and logical inference (NLI). Our central finding is a striking semantic-lexical paradox: all models achieve over 99% cosine similarity with references despite less than 8% BLEU-1 overlap, a 90+ percentage point gap indicating that models preserve meaning through extensive paraphrasing rather than lexical reproduction. MoverScore (51-53%) confirms this pattern, occupying an intermediate position that reflects the optimal transport cost of semantic alignment. Furthermore, model scale does not predict performance: Mistral-7B (7B parameters) outperforms GPT-OSS-20B (20B parameters) across all metrics. NLI analysis reveals that contradiction rates remain below 7%, suggesting models rarely generate content that directly conflicts with human consensus. These findings challenge the reliability of lexical metrics for evaluating abstractive generation and argue for multi-dimensional evaluation frameworks that capture semantic fidelity beyond surface-level text matching. The RECOM dataset is publicly available at https://anonymous.4open.science/r/recom-D4B0

Automated optimization modeling via Large Language Models (LLMs) has emerged as a promising approach to assist complex human decision-making. While post-training has become a pivotal technique to enhance LLMs' capabilities in this domain, its effectiveness is severely constrained by the scarcity and underutilization of high-quality training data. However, through a detailed profiling of error patterns across various problem-response pairs drawn from post-training, we identify two fundamental limitations of existing automated optimization modeling approaches: (L1) the sparsity of error-specific problems and (L2) the sparse rewards associated with difficult problems. We demonstrate that these limitations can result in suboptimal performance in domain-specific post-training for LLMs. To tackle the above two limitations, we propose a novel error-driven learning framework – namely, automated optimization modeling via a localizable error-driven perspective (MIND) – that customizes the whole model training framework from data synthesis to post-training. MIND is based on our key observation of the unique localizable patterns in error propagation of optimization modelings, that is, modeling errors may remain localized to specific semantic segments and do not propagate throughout the entire solution. Thus, in contrast to holistic reasoning tasks such as mathematical proofs, MIND leverages the construction of a focused, high-density training corpus and proposes Dynamic Supervised Fine-Tuning Policy Optimization (DFPO) to tackle difficult problems through localized refinement. Experiments on six benchmarks demonstrate that MIND consistently outperforms all the state-of-the-art automated optimization modeling approaches.

Named Entity Recognition (NER) is an important task in natural language processing that aims to identify and extract key entities from unstructured text. We present a novel application of NER in plasma physics research articles and address the challenges of extracting specialized entities from scientific text in this domain. Research articles in plasma physics often contain highly complex and context-rich content that must be extracted to enable, e.g., advanced search. We propose a lightweight approach based on encoder-transformers and conditional random fields to extract (nested) named entities from plasma physics research articles. First, we annotate a plasma physics corpus with 16 classes specifically designed for the nested NER task. Second, we evaluate an entity-specific model specialization approach, where independent BERT-CRF models are trained to recognize individual entity types in plasma physics text. Third, we integrate an optimization process to systematically fine-tune hyperparameters and enhance model performance. Our work contributes to the advancement of entity recognition in plasma physics and also provides a foundation to support researchers in navigating and analyzing scientific literature.

Peer review is central to scientific quality, yet reliance on simple heuristics, namely lazy thinking and non-specific critiques, has threatened review quality. Prior work frames lazy thinking detection as single-label classification and stops at detection, yet review segments often exhibit multiple co-occurring issues, and reviewers benefit more from actionable, guideline-aware feedback than from labels alone. We further show that off-the-shelf LLMs prompted for feedback frequently rewrite the entire review or address the authors rather than the reviewer, motivating an inference-time approach. We introduce an LLM-driven framework that decomposes reviews into argumentative segments, identifies issues violating ACL Rolling Review (ARR) guidelines, and generates targeted feedback using issue-specific templates refined by a novel iterative, reranking-based generation algorithm. In a controlled rewriting study, our feedback reduces guideline violations by up to 92.4%. We also release LazyReviewPlus, the first multi-label dataset of 1,309 sentences annotated for detecting lazy thinking and lack of specificity.

Replaying a financial agent requires specifying the conditions of a rerun, preserving its execution record, and deciding which parts of its behavior to compare. We introduce the Determinism-Faithfulness Assurance Harness (DFAH), a framework and practical protocol for organizing these choices. The framework distinguishes comparisons of final decisions, tool-name sets, and ordered tool names, while treating support from available evidence as a separate evaluation target. We illustrate the framework using 2,957 recorded executions in synthetic compliance-triage and data-operations tasks. One data-operations configuration repeated its recorded decision across all trials on 84% of cases, while its ordered tool-name sequence repeated on 26%. Conversely, some compliance cases retained the same tool sequence while recorded decisions changed. A case-level example shows how repeated labels accompanied different lookups and escalation calls. These observations demonstrate how the comparison level changes the description of repeated behavior. The resulting protocol connects configuration control, preservation of each attempt, comparison of records, and inspection of changed cases. DFAH provides a practical foundation for designing replay experiments and locating execution differences during the development and evaluation of financial agents.

Modern language models like GPT-3, BERT, and LLaMA require massive training data, yet with sufficient training they reliably learn to distinguish grammatical from ungrammatical sentences. Children aged as young as 14 months already have the capacity to learn abstract grammar rules from very few exemplars, even in the presence of non-rule-following exceptions. Yang's (2016) Tolerance Principle defines a precise threshold for how many exceptions a rule can tolerate and still be learnable. The present study explored the minimal amount and quality of training data necessary for rules to be generalized by a transformer-based language model to test the predictions of the Tolerance Principle. We trained BabyBERTa (Huebner et al. 2021), a transformer model optimized for small datasets, on artificial grammars. The training sets varied in size, number of unique sentence types, and proportion of rule-following versus exception exemplars. We found that, unlike human infants, BabyBERTa's learning dynamics do not align with the Tolerance Principle.

Large language models are increasingly used to interpret politically contested questions, value-laden material on which there is no single correct answer, only competing interpretive traditions. We ask whether a model's choice among those traditions can turn on the language of the prompt rather than the content. Comparing two frontier models, ChatGPT 5.2 and Claude Opus 4.5, on one contested Ukrainian civil-society document under semantically matched Russian and Ukrainian prompts, we find that both shift along the same axis on identical source text: Russian prompts elicit delegitimizing readings of the document's authors and Ukrainian prompts legitimating ones. The magnitude is model-dependent but neither model is neutral: each adopts a language-dependent stance, and the difference is one of degree. Because contested political questions admit no correct reading against which to measure, we read this as language-conditioned variation in which interpretive tradition a model activates: the model neither holds a single stance nor surfaces the plurality of available ones, but silently adopts the dominant frame of the prompt's language. We draw out the consequences for pluralism-aware evaluation, which must probe the same content across the languages a model serves, and for pluralistic alignment in multilingual settings.

This study investigates the use of neural topic modeling and LLMs to uncover meaningful themes from patient storytelling data, to offer insights that could contribute to more patient-oriented healthcare practices. We analyze a collection of transcribed interviews with cancer patients (132,722 words in 13 interviews). We first evaluate BERTopic and Top2Vec for individual interview summarization by using similar preprocessing, chunking, and clustering configurations to ensure a fair comparison on Keyword Extraction. LLMs (GPT4) are then used for the next step topic labeling. Their outputs for a single interview (I0) are rated through a small-scale human evaluation, focusing on {coherence}, {clarity}, and {relevance}. Based on the preliminary results and evaluation, BERTopic shows stronger performance and is selected for further experimentation using three {clinically oriented embedding} models. We then analyzed the full interview collection with the best model setting. Results show that domain-specific embeddings improved topic precision and interpretability, with BioClinicalBERT producing the most consistent results across transcripts. The global analysis of the full dataset of 13 interviews, using the BioClinicalBERT embedding model, reveals the most dominant topics throughout all 13 interviews, namely “Coordination and Communication in Cancer Care Management" and “Patient Decision-Making in Cancer Treatment Journey”. Although the interviews are machine translations from Dutch to English, and clinical professionals are not involved in this evaluation, the findings suggest that neural topic modeling, particularly BERTopic, can help provide useful feedback to clinicians from patient interviews. This pipeline could support more efficient document navigation and strengthen the role of patients' voices in healthcare workflows.

Bengali text classification is a Significant task in natural language processing (NLP), where text is categorized into predefined labels. Unlike English, Bengali faces challenges due to the lack of extensive annotated datasets and pre-trained language models. This study explores the effectiveness of large language models (LLMs) in classifying Bengali newspaper articles. The dataset used, obtained from Kaggle, consists of articles from Prothom Alo, a major Bangladeshi newspaper. Three instruction-tuned LLMs LLaMA 3.1 8B Instruct, LLaMA 3.2 3B Instruct, and Qwen 2.5 7B Instruct were evaluated for this task under the same classification framework. Among the evaluated models, Qwen 2.5 achieved the highest classification accuracy of 72%, showing particular strength in the "Sports" category. In comparison, LLaMA 3.1 and LLaMA 3.2 attained accuracies of 53% and 56%, respectively. The findings highlight the effectiveness of LLMs in Bengali text classification, despite the scarcity of resources for Bengali NLP. Future research will focus on exploring additional models, addressing class imbalance issues, and refining fine-tuning approaches to improve classification performance.

Fine-tuned language models pose significant privacy risks, as they may memorize and expose sensitive information from their training data. Membership inference attacks (MIAs) provide a principled framework for auditing these risks, yet existing methods achieve limited detection rates, particularly at the low false-positive thresholds required for practical privacy auditing. We present EZ-MIA, a membership inference attack that exploits a key observation: memorization manifests most strongly at error positions, specifically tokens where the model predicts incorrectly yet still shows elevated probability for training examples. We introduce the Error Zone (EZ) score, which measures the directional imbalance of probability shifts at error positions relative to a pretrained reference model. This principled statistic requires only two forward passes per query and no model training of any kind. On WikiText with GPT-2, EZ-MIA achieves 3.8x higher detection than the previous state-of-the-art under identical conditions (66.3% versus 17.5% true positive rate at 1% false positive rate), with near-perfect discrimination (AUC 0.98). At the stringent 0.1% FPR threshold critical for real-world auditing, we achieve 8x higher detection than prior work (14.0% versus 1.8%), requiring no reference model training. These gains extend to larger architectures: on AG News with Llama-2-7B, we achieve 3x higher detection (46.7% versus 15.8% TPR at 1% FPR). These results establish that privacy risks of fine-tuned language models are substantially greater than previously understood, with implications for both privacy auditing and deployment decisions. Code is available at https://github.com/JetBrains-Research/ez-mia.

Large language models (LLMs) have shown promise for automated text annotation, raising hopes that they might accelerate cross-cultural research by extracting structured data from ethnographic texts. We evaluated 7 state-of-the-art LLMs on their ability to annotate 121 ritual features across 567 ethnographic excerpts. Performance was limited, falling well below levels required for reliable automated annotation. Longer texts, features requiring ordinal distinctions, and ambiguous constructs proved particularly difficult. Human inter-coder reliability set an approximate ceiling on LLM accuracy: features that human coders found difficult to agree upon were also difficult for LLMs. Yet even on features where humans reliably agreed, models fell short of human performance. Our findings suggest that LLMs cannot yet substitute for human expertise in ethnographic annotation.

Neural network models often face challenges when processing very small or very large numbers due to issues such as overflow, underflow, and unstable output variations. To mitigate these problems, we propose using embedding vectors for numbers instead of directly using their raw values. These embeddings aim to retain essential algebraic properties while preventing numerical instabilities. In this paper, we introduce, for the first time, a fixed-length number embedding vector that preserves algebraic operations, including addition, multiplication, and comparison, within the field of rational numbers. We propose a novel Neural Isomorphic Field, a neural abstraction of algebraic structures such as groups and fields. The elements of this neural field are embedding vectors that maintain algebraic structure during computations. Our experiments demonstrate that addition performs exceptionally well, achieving over 95 percent accuracy on key algebraic tests such as identity, closure, and associativity. In contrast, multiplication exhibits challenges, with accuracy ranging from 53 percent to 73 percent across various algebraic properties. These findings highlight the model's strengths in preserving algebraic properties under addition while identifying avenues for further improvement in handling multiplication.

Large language models (LLMs) excel at general-purpose tasks, yet adapting their responses to individual users remains challenging. Retrieval augmentation provides a lightweight alternative to fine-tuning by conditioning LLMs on user history records, and existing approaches typically select these records based on semantic relevance. We argue that relevance serves as an unreliable proxy for utility: a record may be semantically similar to a query yet fail to improve generation quality or even degrade it due to redundancy or conflicting information. To bridge this gap, we propose PURPLE, a contextual bandit framework that oPtimizes UseR Profiles for LLM pErsonalization. In contrast to a greedy selection of the most relevant records, PURPLE treats profile construction as an order-sensitive generation process and utilizes a Plackett-Luce ranking model to capture complex inter-record dependencies. By training with semantically rich feedback provided by the likelihood of the reference response, our method aligns retrieval directly with generation quality. Extensive experiments on nine personalization tasks demonstrate that PURPLE consistently outperforms strong heuristic and retrieval-augmented baselines in both effectiveness and efficiency, establishing a principled and scalable solution for optimizing user profiles.

Remote sensing video referring object segmentation (RS-RVOS) is challenged by weak target saliency and severe visual information truncation in dynamic scenes, making it extremely difficult to maintain discriminative target representations during segmentation. Moreover, progress in this field is hindered by the absence of large-scale dedicated benchmarks, while existing models are often affected by biased initial memory construction that impairs accurate instance localization in complex scenarios, as well as indiscriminate memory accumulation that encodes noise from occlusions or misclassifications, leading to persistent error propagation. This paper advances RS-RVOS research through dual contributions in data and methodology. First, we construct RS-RVOS Bench, the first large-scale benchmark comprising 111 video sequences, about 25,000 frames, and 213,000 temporal referring annotations. Unlike common RVOS benchmarks where many expressions are written with access to the full video context, our dataset adopts a strict causality-aware annotation strategy in which linguistic references are generated solely from the target state in the initial frame. Second, we propose a memory-quality-aware online referring segmentation framework, termed Memory Quality Control with Segment Anything Model (MQC-SAM). MQC-SAM introduces a temporal motion consistency module for initial memory calibration, leveraging short-term motion trajectory priors to correct structural deviations and establish accurate memory anchoring. Furthermore, it incorporates a decoupled attention-based memory integration mechanism with dynamic quality assessment, selectively updating high-confidence semantic features while filtering unreliable information, thereby effectively preventing error accumulation and propagation. Extensive experiments on RS-RVOS Bench demonstrate that MQC-SAM achieves state-of-the-art performance.

Language models used in retrieval-augmented settings must arbitrate between parametric knowledge stored in their weights and contextual information in the prompt. This work presents a mechanistic study of that choice by extracting an arbitration vector from model activations on a curated dataset designed to disentangle (i) irrelevant contexts that elicit parametric recall and (ii) relevant but false contexts that elicit copying. The vector is computed as the residual-stream centroid difference between these regimes across 27 relations, and is injected as an additive intervention at selected layers and token spans to steer behavior in two directions: Copy$\rightarrow$Recall (suppressing context use) and Recall$\rightarrow$Copy (inducing the model to copy any token from the context). Experiments on three architectures (decoder-only and encoder/decoder) and two open-domain QA benchmarks show consistent behavior shifts under moderate scaling while monitoring accuracy and fluency. Mechanistic analyses of attention routing, MLP contributions, and layer-wise probability trajectories reveal an asymmetry: inducing copying is an easy “reactivation” process that can be triggered at different locations in the input, while restoring recall is a “suppression” process that is more fragile and strongly tied to object-token interventions.

Increased access to reliable health information is essential for non-English-speaking populations, yet resources in Bangla for disease prediction remain limited. This study addresses this gap by developing a comprehensive Bangla symptoms-disease dataset containing 758 unique symptom-disease relationships spanning 85 diseases. To ensure transparency and reproducibility, we also make our dataset publicly available. The dataset enables the prediction of diseases based on Bangla symptom inputs, supporting healthcare accessibility for Bengali-speaking populations. Using this dataset, we evaluated multiple machine learning models to predict diseases based on symptoms provided in Bangla and analyzed their performance on our dataset. Both soft and hard voting ensemble approaches combining top-performing models achieved 98% accuracy, demonstrating superior robustness and generalization. Our work establishes a foundational resource for disease prediction in Bangla, paving the way for future advancements in localized health informatics and diagnostic tools. This contribution aims to enhance equitable access to health information for Bangla-speaking communities, particularly for early disease detection and healthcare interventions.

Dialogue Act (DA) annotation typically treats communicative or pedagogical intent as localized to individual utterances or turns. This leads annotators to agree on the underlying action while disagreeing on segment boundaries, reducing apparent reliability. We propose codebook-injected segmentation, which conditions boundary decisions on downstream annotation criteria, and evaluate LLM-based segmenters against standard and retrieval-augmented baselines. To assess these without gold labels, we introduce evaluation metrics for span consistency, distinctiveness, and human-AI distributional agreement. We found DA-awareness produces segments that are internally more consistent than text-only baselines. While LLMs excel at creating construct-consistent spans, coherence-based baselines remain superior at detecting global shifts in dialogue flow. Across two datasets, no single segmenter dominates. Improvements in within-segment coherence frequently trade off against boundary distinctiveness and human-AI distributional agreement. These results highlight segmentation as a consequential design choice that should be optimized for downstream objectives rather than a single performance score.

Personalization in Large Language Models (LLMs) often relies on user-specific soft prompts. However, these prompts become obsolete when the foundation model is upgraded, necessitating costly, full-scale retraining. To overcome this limitation, we propose the Prompt-level User Migration Adapter (PUMA), a lightweight framework to efficiently migrate personalized prompts across incompatible models. PUMA utilizes a parameter-efficient adapter to bridge the semantic gap, combined with a group-based user selection strategy to significantly reduce training costs. Experiments on three large-scale datasets show our method matches or even surpasses the performance of retraining from scratch, reducing computational cost by up to 98%. The framework demonstrates strong generalization across diverse model architectures and robustness in advanced scenarios like chained and aggregated migrations, offering a practical path for the sustainable evolution of personalized AI by decoupling user assets from the underlying models.

Quantization is widely adopted to reduce the computational cost of large language models (LLMs); however, its implications for fairness and safety, particularly in dynamic quantization and multilingual contexts, remain underexplored. In this work, we conduct a systematic study of how static and dynamic quantization methods impact fairness and safety across benchmarks measuring intrinsic and extrinsic bias and safety alignment. For fairness, we evaluate English, French, Dutch, Spanish, and Turkish; for safety, we focus on English, Korean, and Arabic. Our findings reveal that quantization consistently degrades fairness and safety, with dynamic methods demonstrating greater stability than static ones. Moreover, fairness degradation varies across languages, while safety deterioration is especially pronounced in non-English settings. To address these risks, we introduce Critical Weight Protection, a novel technique that identifies and preserves fairness- and safety-critical weights during quantization. This approach effectively mitigates bias and safety deterioration without costly retraining or alignment, maintaining trustworthiness while retaining efficiency.

Customer reviews contain valuable signals about service quality, but converting large-scale review corpora into actionable business recommendations remains difficult. Standard sentiment/aspect analysis is largely descriptive, while direct prompting of large language models (LLMs) often yields generic and repetitive advice that is weakly grounded in user feedback. We propose a hierarchical decision-support pipeline that explicitly separates signal compression, problem abstraction, candidate generation, objective-based evaluation, and cost-aware routing into different agents. This architectural decomposition produces auditable intermediate artifacts and enables controllable trade-offs between advice quality and token budget. Experiments on Yelp reviews from three service domains show consistent improvements over single-pass LLM baselines across multiple advice quality dimensions, including actionability, relevance, and non-redundancy. A human evaluation further indicates that users generally prefer our system's recommendations. These results highlight the value of structured agentic decomposition for scalable, cost-aware business decision support.

The widespread proliferation of online content has intensified concerns about clickbait, deceptive or exaggerated headlines designed to attract attention. While Large Language Models (LLMs) offer a promising avenue for addressing this issue, their effectiveness is often hindered by Sycophancy, a tendency to produce reasoning that matches users' beliefs over truthful ones, which deviates from instruction-following principles. Rather than treating sycophancy as a flaw to be eliminated, this work proposes a novel approach that initially harnesses this behavior to generate contrastive reasoning from opposing perspectives. Specifically, we design a Self-renewal Opposing-stance Reasoning Generation (SORG) framework that prompts LLMs to produce high-quality agree and disagree reasoning pairs for a given news title without requiring ground-truth labels. To utilize the generated reasoning, we develop a local Opposing Reasoning-based Clickbait Detection (ORCD) model that integrates three BERT encoders to represent the title and its associated reasoning. The model leverages contrastive learning, guided by soft labels derived from LLM-generated credibility scores, to enhance detection robustness. Experimental evaluations on three benchmark datasets demonstrate that our method consistently outperforms LLM prompting, fine-tuned smaller language models, and state-of-the-art clickbait detection baselines.