Papers

Large Language Models (LLMs) deliver state-of-the-art performance across many tasks but impose high computational and memory costs, limiting their deployment in resource-constrained or real-time settings. To address this, we propose HierRouter, a hierarchical routing approach that dynamically assembles inference pipelines from a pool of specialized, lightweight language models. Formulated as a finite-horizon Markov Decision Process (MDP), our approach trains a Proximal Policy Optimization (PPO)-based reinforcement learning agent to iteratively select which models to invoke at each stage of multi-hop inference. The agent conditions on the evolving context and accumulated cost to make context-aware routing decisions. Experiments with three open-source candidate LLMs across six benchmarks, including QA, code generation, and mathematical reasoning, show that HierRouter improves response quality by up to 2.4x compared to using individual models independently, while incurring only a minimal additional inference cost on average. These results highlight the promise of hierarchical routing for cost-efficient, high-performance LLM inference. All codes can be found here https://github.com/ Nikunj-Gupta/hierouter.

Training Large Language Models (LLMs) for chain-of-thought reasoning presents a significant challenge: supervised fine-tuning on a single "golden" rationale hurts generalization as it penalizes equally valid alternatives, whereas reinforcement learning with verifiable rewards struggles with credit assignment and prohibitive computational cost. To tackle these limitations, we introduce InTRO (In-Token Rationality Optimization), a new framework that enables both token-level exploration and self-feedback for accurate and concise reasoning. Instead of directly optimizing an intractable objective over all valid reasoning paths, InTRO leverages correction factors-token-wise importance weights estimated by the information discrepancy between the generative policy and its answer-conditioned counterpart, for informative next token selection. This approach allows the model to perform token-level exploration and receive self-generated feedback within a single forward pass, ultimately encouraging accurate and concise rationales. Across six math-reasoning benchmarks, InTRO consistently outperforms other baselines, raising solution accuracy by up to 20% relative to the base model. Its chains of thought are also notably more concise, exhibiting reduced verbosity. Beyond this, InTRO enables cross-domain transfer, successfully adapting to out-of-domain reasoning tasks that extend beyond the realm of mathematics, demonstrating robust generalization.

Large language models (LLMs) have demonstrated impressive performance in text generation tasks; however, their embedding spaces often suffer from the isotropy problem, resulting in poor discrimination of domain-specific terminology, particularly in legal and financial contexts. This weakness in terminology-level representation can severely hinder downstream tasks such as legal judgment prediction or financial risk analysis, where subtle semantic distinctions are critical. To address this problem, we propose TermGPT, a multi-level contrastive fine-tuning framework designed for terminology adaptation. We first construct a sentence graph to capture semantic and structural relations, and generate semantically consistent yet discriminative positive and negative samples based on contextual and topological cues. We then devise a multi-level contrastive learning approach at both the sentence and token levels, enhancing global contextual understanding and fine-grained terminology discrimination. To support robust evaluation, we construct the first financial terminology dataset derived from official regulatory documents. Experiments show that TermGPT outperforms existing baselines in term discrimination tasks within the finance and legal domains.

The course forums are increasingly significant and play vital role in facilitating student discussions and answering their questions related to the course. It provides a platform for students to post their questions related to the content and admin issues related to the course. However, there are several challenges due to the increase in the number of students enrolled in the course. The primary challenge is that students' queries cannot be responded immediately and the instructors have to face lots of repetitive questions. To mitigate these issues, we propose a question answering system based on large language model with retrieval augmented generation (RAG) method. This work focuses on designing a question answering system with open source Large Language Model (LLM) and fine-tuning it on the relevant course dataset. To further improve the performance, we use a local knowledge base and applied RAG method to retrieve relevant documents relevant to students' queries, where the local knowledge base contains all the course content. To mitigate the hallucination of LLMs, We also integrate it with multi chain-of-thought reasoning to overcome the challenge of hallucination in LLMs. In this work, we experiment fine-tuned LLM with RAG method on the HotpotQA dataset. The experimental results demonstrate that the fine-tuned LLM with RAG method has a strong performance on question answering task.

Evaluating complex texts across domains requires converting user defined criteria into quantitative, explainable indicators, which is a persistent challenge in search and recommendation systems. Single prompt LLM evaluations suffer from complexity and latency issues, while criterion specific decomposition approaches rely on naive averaging or opaque black-box aggregation methods. We present an interpretable aggregation framework combining LLM scoring with the Analytic Hierarchy Process. Our method generates criterion specific scores via LLM as judge, measures discriminative power using Jensen Shannon distance, and derives statistically grounded weights through AHP pairwise comparison matrices. Experiments on Amazon review quality assessment and depression related text scoring demonstrate that our approach achieves high explainability and operational efficiency while maintaining comparable predictive power, making it suitable for real time latency sensitive web services.

Text classification with hierarchical taxonomies is a fundamental requirement in IT Service Management (ITSM) systems, where support tickets must be categorized into tree-structured taxonomies. We present a dual-embedding centroid-based classification framework that maintains separate semantic and lexical centroid representations per category, combining them through reciprocal rank fusion at inference time. The framework achieves performance competitive with Support Vector Machines (hierarchical F1: 0.731 vs 0.727) while providing interpretability through centroid representations. Evaluated on 8,968 ITSM tickets across 123 categories, this method achieves 5.9 times faster training and up to 152 times faster incremental updates. With 8.6-8.8 times speedup across batch sizes (100-1000 samples) when excluding embedding computation. These results make the method suitable for production ITSM environments prioritizing interpretability and operational efficiency.

Recent advances in text-to-speech (TTS) have enabled models to clone arbitrary unseen speakers and synthesize high-quality, natural-sounding speech. However, evaluation methods lag behind: typical mean opinion score (MOS) estimators perform regression over entire utterances, while failures usually occur in a few problematic words. We observe that encoder-decoder ASR models (e.g., Whisper) surface word-level mismatches between speech and text via cross-attention, providing a fine-grained reward signal. Building on this, we introduce Word-level TTS Alignment by ASR-driven Attentive Reward (W3AR). Without explicit reward annotations, W3AR uses attention from a pre-trained ASR model to drive finer-grained alignment and optimization of sequences predicted by a TTS model. Experiments show that W3AR improves the quality of existing TTS systems and strengthens zero-shot robustness on unseen speakers. More broadly, our results suggest a simple recipe for generative modeling: understanding models can act as evaluators, delivering informative, fine-grained feedback for optimization.

Large Language Models (LLMs) have been positioned as having the potential to expand access to health information in the Global South, yet their evaluation remains heavily dependent on benchmarks designed around Western norms. We present insights from a preliminary benchmarking exercise with a chatbot for sexual and reproductive health (SRH) for an underserved community in India. We evaluated using HealthBench, a benchmark for conversational health models by OpenAI. We extracted 637 SRH queries from the dataset and evaluated on the 330 single-turn conversations. Responses were evaluated using HealthBench's rubric-based automated grader, which rated responses consistently low. However, qualitative analysis by trained annotators and public health experts revealed that many responses were actually culturally appropriate and medically accurate. We highlight recurring issues, particularly a Western bias, such as for legal framing and norms (e.g., breastfeeding in public), diet assumptions (e.g., fish safe to eat during pregnancy), and costs (e.g., insurance models). Our findings demonstrate the limitations of current benchmarks in capturing the effectiveness of systems built for different cultural and healthcare contexts. We argue for the development of culturally adaptive evaluation frameworks that meet quality standards while recognizing needs of diverse populations.

Analyzing spoken discourse is a valid means of quantifying language ability in persons with aphasia. There are many ways to quantify discourse, one common way being to evaluate the informativeness of the discourse. That is, given the total number of words produced, how many of those are context-relevant and accurate. This type of analysis is called Correct Information Unit (CIU) analysis and is one of the most prevalent discourse analyses used by speech-language pathologists (SLPs). Despite this, CIU analysis in the clinic remains limited due to the manual labor needed by SLPs to code and analyze collected speech. Recent advances in machine learning (ML) seek to augment such labor by automating modeling of propositional, macrostructural, pragmatic, and multimodal dimensions of discourse. To that end, this study evaluated five ML models for reliable identification of Correct Information Units (CIUs, Nicholas & Brookshire, 1993), during a picture description task. The five supervised ML models were trained using randomly selected human-coded transcripts and accompanying words and CIUs from persons with aphasia. The baseline model training produced a high accuracy across transcripts for word vs non-word, with all models achieving near perfect performance (0.995) with high AUC range (0.914 min, 0.995 max). In contrast, CIU vs non-CIU showed a greater variability, with the k-nearest neighbor (k-NN) model the highest accuracy (0.824) and second highest AUC (0.787). These findings indicate that while the supervised ML models can distinguish word from not word, identifying CIUs is challenging.

Large Language Models (LLMs) are the engines driving today's AI agents. The better these models understand human languages, the more natural and user-friendly the interaction with AI becomes, from everyday devices like computers and smartwatches to any tool that can act intelligently. Yet, the ability of industrial-scale LLMs to comprehend low-resource languages, such as Tunisian Arabic (Tunizi), is often overlooked. This neglect risks excluding millions of Tunisians from fully interacting with AI in their own language, pushing them toward French or English. Such a shift not only threatens the preservation of the Tunisian dialect but may also create challenges for literacy and influence younger generations to favor foreign languages. In this study, we introduce a novel dataset containing parallel Tunizi, standard Tunisian Arabic, and English translations, along with sentiment labels. We benchmark several popular LLMs on three tasks: transliteration, translation, and sentiment analysis. Our results reveal significant differences between models, highlighting both their strengths and limitations in understanding and processing Tunisian dialects. By quantifying these gaps, this work underscores the importance of including low-resource languages in the next generation of AI systems, ensuring technology remains accessible, inclusive, and culturally grounded.

Running LLMs locally has become increasingly common, but users face a complex design space across models, quantization levels, inference engines, and serving scenarios. Existing inference benchmarks are fragmented and focus on isolated goals, offering little guidance for practical deployments. We present Bench360, a framework for evaluating local LLM inference across tasks, usage patterns, and system metrics in one place. Bench360 supports custom tasks, integrates multiple inference engines and quantization formats, and reports both task quality and system behavior (latency, throughput, energy, startup time). We demonstrate it on four NLP tasks across three GPUs and four engines, showing how design choices shape efficiency and output quality. Results confirm that tradeoffs are substantial and configuration choices depend on specific workloads and constraints. There is no universal best option, underscoring the need for comprehensive, deployment-oriented benchmarks.

Vector databases (VecDBs) are increasingly deployed in retrieval-augmented generation (RAG) pipelines where query processing and document ingestion occur concurrently. The index layer needs to provide low-latency search while incorporating new vectors without frequent global rebuilding. Existing VecDB pipelines typically operate within a uniform representation regime, despite substantial variation in the semantic granularity required across queries. This motivates an index design that supports incremental updates while adapting retrieval depth to query distribution and complexity. We propose Semantic Pyramid Indexing (SPI), a VecDB-layer indexing framework that organizes embeddings into $L$ semantically aligned resolution levels and selects retrieval depth per query via a lightweight uncertainty-aware controller. SPI supports progressive coarse-to-fine ANN search, level-wise streaming insertion without global rebuilds, and distributed execution through LSH partitioning with asynchronous gRPC coordination. Unlike hierarchical ANN structures with fixed traversal rules (e.g., SPANN), SPI adapts resolution at query time while remaining compatible with FAISS and Qdrant backends. On MS MARCO and Natural Questions, SPI achieves competitive Recall@10 with lower latency under the same dense encoder family, yielding a \textbf{1.4–2.3$\times$} average retrieval latency reduction under fixed Recall@10 targets relative to comparable approximate-ANN baselines. A prototype scaling study up to 8 nodes shows $6.2\times$ throughput scaling (${\approx}73\%$ efficiency); the 16-node configuration is included for completeness but shows diminishing efficiency. We provide a top-$K$ stability guarantee: queries with sufficient retrieval margin return an identical top-$K$ set at a shallower level. Code and configurations are available at https://github.com/FastLM/SPI_VecDB.

Despite rapid advances in multilingual natural language processing (NLP), the Bantu language Shona remains under-served in terms of morphological analysis and language-aware tools. This paper presents Shona spaCy, an open-source, rule-based morphological pipeline for Shona built on the spaCy framework. The system combines a curated JSON lexicon with linguistically grounded rules to model noun-class prefixes (Mupanda 1-18), verbal subject concords, tense-aspect markers, ideophones, and clitics, integrating these into token-level annotations for lemma, part-of-speech, and morphological features. The toolkit is available via pip install shona-spacy, with source code at https://github.com/HappymoreMasoka/shona-spacy and a PyPI release at https://pypi.org/project/shona-spacy/0.1.4/. Evaluation on formal and informal Shona corpora yields 90% POS-tagging accuracy and 88% morphological-feature accuracy, while maintaining transparency in its linguistic decisions. By bridging descriptive grammar and computational implementation, Shona spaCy advances NLP accessibility and digital inclusion for Shona speakers and provides a template for morphological analysis tools for other under-resourced Bantu languages.

Large Language Models (LLMs) are increasingly using external web content. However, much of this content is not easily digestible by LLMs due to LLM-unfriendly formats and limitations of context length. To address this issue, we propose a method for generating general-purpose, information-dense summaries that act as plain-text repositories of web content. Inspired by Hegel's dialectical method, our approach, denoted as Chain of Summaries (CoS), iteratively refines an initial summary (thesis) by identifying its limitations through questioning (antithesis), leading to a general-purpose summary (synthesis) that can satisfy current and anticipate future information needs. Experiments on the TriviaQA, TruthfulQA, and SQUAD datasets demonstrate that CoS outperforms zero-shot LLM baselines by up to 66% and specialized summarization methods such as Chain of Density, BRIO and PEGASUS by up to 27%. CoS-generated summaries yield higher Q&A performance compared to the source content, while requiring substantially fewer tokens and being agnostic to the specific downstream LLM. CoS thus resembles an appealing option for website maintainers to make their content more accessible for LLMs, while retaining possibilities for human oversight.

Standardized patients (SPs) are indispensable for clinical skills training but remain expensive and difficult to scale. Although large language model (LLM)-based virtual standardized patients (VSPs) have been proposed as an alternative, their behavior remains unstable and lacks rigorous comparison with human standardized patients. We propose EasyMED, a multi-agent VSP framework that separates case-grounded information disclosure from response generation to support stable, inquiry-conditioned patient behavior. We also introduce SPBench, a human-grounded benchmark with eight expert-defined criteria for interaction-level evaluation. Experiments show that EasyMED more closely matches human SP behavior than existing VSPs, particularly in case consistency and controlled disclosure. A four-week controlled study further demonstrates learning outcomes comparable to human SP training, with stronger early gains for novice learners and improved flexibility, psychological safety, and cost efficiency.

What is reasoning? This question has driven centuries of philosophical inquiry, from Aristotle's syllogisms to modern computational complexity theory. In the age of large language models achieving superhuman performance on benchmarks like GSM8K (95% accuracy) and HumanEval (90% pass@1), we must ask: have these systems learned to reason, or have they learned to pattern-match over reasoning traces? This paper argues for a specific answer: reasoning is iterative operator application in state spaces, converging to fixed points. This definition is not merely philosophical – it has concrete architectural implications that explain both the failures of current systems and the path to genuine reasoning capabilities. Our investigation begins with a puzzle (OpenXOR), progresses through theory (OpenOperator), and culminates in a working solution (OpenLM) that achieves 76% accuracy where state-of-the-art LLMs achieve 0%. This is not about criticizing existing systems, but about understanding what reasoning requires and building architectures that provide it.

Generative AI chatbots built for mental health could extend access to care, but evidence from real-world use is limited. We report a single-arm, naturalistic pilot of a foundation model trained for mental health, among 299 US adults with at least moderate depressive or anxiety symptoms who were followed for up to 12 months. Depression and anxiety symptoms fell by 10 weeks (Cohen's d 0.93 and 0.79), with loneliness, behavioral activation, and social interaction improving. Clinicians confirmed that automated safeguards were escalated appropriately. Participants fell into non-responding (57.2%), improving (37.1%) and rapidly improving (5.7%) trajectories. Early working alliance and greater engagement were associated with better outcomes. The AI deployed ten identified intervention families whose delivery varied with baseline anxiety and depression, with the overall ratio of clinical to non-clinical content increasing according to severity without survey information access. These findings support the feasibility of purpose-built AI for mental health.

A growing literature documents that generative artificial intelligence (GenAI) is changing scientific writing, yet most studies focus on absolute changes in vocabulary or readability. An important question remains unanswered: Does GenAI use lead to systematic convergence, or a narrowing of stylistic gaps relative to the dominant form of scientific English? Unlike absolute changes, convergence signals whether language-related publication barriers are declining and suggests broader implications for participation and competition in global science. This study directly addresses this question using 5.65 million English-language scientific articles published from 2021 to 2024 and indexed in Scopus. We measure linguistic similarity to a U.S. benchmark corpus using SciBERT text embeddings, and estimate dynamic changes using an event-study difference-in-differences design with repeated cross-sections centered on the late-2022 release of ChatGPT. We find that GenAI-assisted publications from non-English-speaking countries exhibit statistically significant and increasing convergence toward U.S. scientific English, relative to non-GenAI-assisted publications from these countries. This effect is strongest for domestic author teams from countries more linguistically distant from English and for articles published in lower-impact journals – precisely the contexts where language barriers have historically been most consequential. The results suggest that GenAI tools are reducing language-related barriers in scientific publications. Whether this represents genuine inclusion or a deepening dependence on a single linguistic standard remains an open question.

The surge in intelligent applications driven by large language models (LLMs) has made it increasingly difficult for bandwidth-limited cloud servers to process extensive LLM workloads in real time without compromising user data privacy. To solve these problems, recent research has focused on constructing cloud-edge consortia that integrate server-based LLM with small language models (SLMs) on mobile edge devices. Furthermore, designing collaborative training mechanisms within such consortia to enhance inference performance has emerged as a promising research direction. However, the cross-domain deployment of SLMs, coupled with structural heterogeneity in SLMs architectures, poses significant challenges to enhancing model performance. To this end, we propose Co-PLMs, a novel co-tuning framework for collaborative training of large and small language models, which integrates the process of structure-agnostic mutual learning to realize knowledge exchange between the heterogeneous language models. This framework employs distilled proxy models (DPMs) as bridges to enable collaborative training between the heterogeneous server-based LLM and on-device SLMs, while preserving the domain-specific insights of each device. The experimental results show that Co-PLMs outperform state-of-the-art methods, achieving average increases of 5.38% in Rouge-L and 4.88% in EM.

Sparse attention is crucial in long-context Transformers, which restricts each token to a limited neighborhood and thereby reduces the quadratic cost of full self-attention. Local windows capture nearby context effectively, yet they induce a receptive-field bottleneck for dependencies beyond the window, limiting long-range modeling under moderate depth. In this paper, we propose $\pi$-Attention, an online efficient sparse attention operator: as tokens arrive, each step maintains a streaming working set of local neighbors plus a $\pi$-indexed long-range fetch, fused by an adaptive prior under a shared softmax. Rather than materializing a global sparse mask in advance, $\pi$-Attention computes attention on the live working set with hierarchy-aware IO. We analyze causal reachability and minimum depth under this online rule, and show per-step cost remains $\mathcal{O}(k)$. Experiments on language modeling, Long Range Arena, and efficiency profiling—across 4K–32K context lengths—show consistent gains over local-window and other sparse baselines, approaching dense attention quality at linear cost.

This paper systematically examines nation-level biases exhibited by Large Language Models (LLMs) within the domain of International Relations (IR). Leveraging historical records from the United Nations Security Council (UNSC), we developed a bias evaluation framework comprising three distinct tests to explore nation-level bias in various LLMs, with a particular focus on the five permanent members of the UNSC. Experimental results show that, even with the general bias patterns across models (e.g., favorable biases toward the western nations, and unfavorable biases toward Russia), these still vary based on the LLM. Notably, even within the same LLM, the direction and magnitude of bias for a nation change depending on the evaluation context. This observation suggests that LLM biases are fundamentally multidimensional, varying across models and tasks. We also observe that models with stronger reasoning abilities show reduced bias and better performance. Building on this finding, we introduce a debiasing framework that improves LLMs' factual reasoning combining Retrieval-Augmented Generation with Reflexion-based self-reflection techniques. Experiments show it effectively reduces nation-level bias, and improves performance, particularly in GPT-4o-mini and LLama-3.3-70B. Our findings emphasize the need to assess nation-level bias alongside performance when applying LLMs in the IR domain.

Following an instruction involves distinct sub-processes, such as reading content, reading the instruction, executing it, and producing an answer. We ask where, along the layer stack, instruction following begins, the point where reading gives way to doing. We introduce three simple datasets (Key-Value, Quote Attribution, Letter Selection) and two hop compositions of these tasks. Using activation patching on minimal-contrast prompt pairs, we measure a layer-wise flip rate that indicates when substituting selected residual activations changes the predicted answer. Across models in the Llama family, we observe an inflection point, which we term onset, where interventions that change predictions before this point become largely ineffective afterward. Multi-hop compositions show a similar onset location. These results provide a simple, replicable way to locate where instruction following begins and to compare this location across tasks and model sizes.

Voice-based artificial intelligence is increasingly expected to adhere to human social conventions, but can it learn implicit cues that are not explicitly programmed? This study investigates whether state-of-the-art text-to-speech systems have internalized the human tendency to reduce speech rate to convey politeness - a non-obvious prosodic marker. We prompted 22 synthetic voices from two leading AI platforms (AI Studio and OpenAI) to read a fixed script under both "polite and formal" and "casual and informal" conditions and measured the resulting speech duration. Across both AI platforms, the polite prompt produced slower speech than the casual prompt with very large effect sizes, an effect that was statistically significant for all of AI Studio's voices and for a large majority of OpenAI's voices. These results demonstrate that AI can implicitly learn and replicate psychological nuances of human communication, highlighting its emerging role as a social actor capable of reinforcing human social norms.

The potential data contamination issue in contemporary large language models (LLMs) benchmarks presents a fundamental challenge to establishing trustworthy evaluation frameworks. Meanwhile, they predominantly assume benign, resource-rich settings, leaving the behavior of LLMs under pressure unexplored. In this paper, we introduce Squid Game, a dynamic and adversarial evaluation environment with resource-constrained and asymmetric information settings elaborated to evaluate LLMs through interactive gameplay against other LLM opponents. Squid Game consists of six elimination-style levels, focusing on multi-faceted abilities, including instruction-following, code, reasoning, planning, and safety alignment. We evaluate over 50 LLMs on Squid Game, presenting the largest behavioral evaluation study of general LLMs on dynamic adversarial scenarios. We observe a clear generational phase transition in performance in the same model lineage and find evidence that some models resort to speculative shortcuts to win the game, indicating the possibility of higher-level evaluation paradigm contamination in static benchmarks. We also compare prominent LLM benchmarks and Squid Game, highlighting that dynamic evaluation can serve as a complementary part for static evaluations. Project page: https://github.com/zijianchen98/LLM_Squid_Game.

Recent closed-source multimodal systems have made great advances, but their hidden language for understanding the world remains opaque because of their black-box architectures. In this paper, we use the systems' preference bias to study their hidden language: During the process of compressing the input images (typically containing multiple concepts) into texts and then reconstructing them into images, the systems' inherent preference bias introduces specific shifts in the outputs, disrupting the original input concept co-occurrence. We employ the multi-round "telephone game" to strategically leverage this bias. By observing the co-occurrence frequencies of concepts in telephone games, we quantitatively investigate the concept connection strength in the understanding of multimodal systems, i.e., "hidden language." We also contribute Telescope, a dataset of 10,000+ concept pairs, as the database of our telephone game framework. Our telephone game is test-time scalable: By iteratively running telephone games, we can construct a global map of concept connections in multimodal systems' understanding. Here we can identify preference bias inherited from training, assess generalization capability advancement, and discover more stable pathways for fragile concept connections. Furthermore, we use Reasoning-LLMs to uncover unexpected concept relationships that transcend textual and visual similarities, inferring how multimodal systems understand and simulate the world. This study offers a new perspective on the hidden language of multimodal systems and lays the foundation for future research on the interpretability and controllability of multimodal systems.

Recursive prompting with large language models enables scalable synthetic dataset generation but introduces the risk of bias amplification. We investigate gender bias dynamics across three generations of recursive text generation using three complementary evaluation frameworks: rule-based pattern matching, embedding-based semantic similarity, and downstream task performance. Experiments with three initial bias levels (0.1, 0.3, 0.6) and four mitigation strategies reveal equilibrium dynamics rather than monotonic amplification. The low initial bias amplifies toward the model's inherent bias level (+36%), whereas the high initial bias decays toward it (-26%). Among mitigation methods, contrastive augmentation, which introduces gender-swapped variants, achieves significant downstream bias reduction (98.8% for low initial bias and 91% on average) despite producing higher embedding-based bias scores. This paradox demonstrates that semantic similarity metrics may diverge from behavioral fairness outcomes, highlighting the need for multidimensional evaluation in responsible synthetic data generation.

As large language models (LLMs) are adopted in an increasingly wide range of applications, user-model interactions have grown in both frequency and scale. Consequently, research has focused on evaluating the robustness of LLMs, an essential quality for real-world tasks. In this paper, we employ simple multi-turn follow-up prompts to evaluate models' answer changes, model accuracy dynamics across turns with Markov chains, and examine whether linear probes can predict these changes. Our results show significant vulnerabilities in LLM robustness: a simple "Think again" prompt led to an approximate 10% accuracy drop for Gemini 1.5 Flash over nine turns, while combining this prompt with a semantically equivalent reworded question caused a 7.5% drop for Claude 3.5 Haiku. Additionally, we find that model accuracy across turns can be effectively modeled using Markov chains, enabling the prediction of accuracy probabilities over time. This allows for estimation of the model's stationary (long-run) accuracy, which we find to be on average approximately 8% lower than its first-turn accuracy for Gemini 1.5 Flash. Our results from a model's hidden states also reveal evidence that linear probes can help predict future answer changes. Together, these results establish stationary accuracy as a principled robustness metric for interactive settings and expose the fragility of models under repeated questioning. Addressing this instability will be essential for deploying LLMs in high-stakes and interactive settings where consistent reasoning is as important as initial accuracy.

This paper aims to address the challenge of selecting relevant courses for students by proposing the design and development of a course recommendation system. The course recommendation system utilises a combination of data analytics techniques and machine learning algorithms to recommend courses that align with current industry trends and requirements. In order to provide customised suggestions, the study entails the design and implementation of an extensive algorithmic framework that combines machine learning methods, user preferences, and academic criteria. The system employs data mining and collaborative filtering techniques to examine past courses and individual career goals in order to provide course recommendations. Moreover, to improve the accessibility and usefulness of the recommendation system, special attention is given to the development of an easy-to-use front-end interface. The front-end design prioritises visual clarity, interaction, and simplicity through iterative prototyping and user input revisions, guaranteeing a smooth and captivating user experience. We refined and optimised the proposed system by incorporating user feedback, ensuring that it effectively meets the needs and preferences of its target users. The proposed course recommendation system could be a useful tool for students, instructors, and career advisers to use in promoting lifelong learning and professional progression as it fills the gap between university learning and industry expectations. We hope that the proposed course recommendation system will help university students in making data-drive and industry-informed course decisions, in turn, improving graduate outcomes for the university sector.

Compared to English and other high-resource languages, spellchecking for Khmer remains an unresolved problem due to several challenges. First, there are misalignments between words in the lexicon and the word segmentation model. Second, a Khmer word can be written in different forms. Third, Khmer compound words are often loosely and easily formed, and these compound words are not always found in the lexicon. Fourth, some proper nouns may be flagged as misspellings due to the absence of a Khmer named-entity recognition (NER) model. Unfortunately, existing solutions do not adequately address these challenges. This paper proposes a holistic approach to the Khmer spellchecking problem by integrating Khmer subword segmentation, Khmer NER, Khmer grapheme-to-phoneme (G2P) conversion, and a Khmer language model to tackle these challenges, identify potential correction candidates, and rank the most suitable candidate. Experimental results show that the proposed approach achieves a state-of-the-art Khmer spellchecking accuracy of up to 94.4%, compared to existing solutions. The benchmark datasets for Khmer spellchecking and NER tasks in this study will be made publicly available.

Retrieval-Augmented Generation (RAG) improves factuality but retrieving for every query often hurts quality while inflating tokens and latency. We propose Training-free Adaptive Retrieval Gating (TARG), a single-shot policy that decides when to retrieve using only a short, no-context draft from the base model. From the draft's prefix logits, TARG computes lightweight uncertainty scores-mean token entropy, a margin signal derived from the top-1/top-2 logit gap via a monotone link, or small-N variance across a handful of stochastic prefixes-and triggers retrieval only when the score exceeds a threshold. The gate is model-agnostic, adds only tens to hundreds of draft tokens, and requires no additional training or auxiliary heads. On five QA benchmarks spanning short-answer (NQ-Open, TriviaQA, PopQA), multi-hop (MuSiQue), and long-form (ASQA) tasks, TARG consistently pushes the accuracy-efficiency frontier: compared with Alway-RAG, TARG matches or improves EM/F1 while reducing retrieval by 70-90% and cutting end-to-end latency, and it remains close to Never-RAG in overhead. A central empirical finding is that under modern instruction-tuned LLMs the margin signal is a robust default (entropy compresses as backbones sharpen), with small-N variance offering a conservative, budget-first alternative. We provide ablations over gate type and prefix length and use a $\Delta$-latency view to make budget trade-offs explicit.