Authors: Boi Mai Quach, Binh T. Nguyen, Cathal Gurrin, Graham Healy
Abstract: Humans invented reading and have passed down this complex skill across generations through language. This study provides empirical evidence of the neural mechanisms underlying bottom-up (related to high-order linguistic structure) and top-down (related to next-word predictability) processes, which interact to guide comprehension during reading. While previous studies have focused on either the N400 effects of predictability or lexical categories, research on how predictability influences N400 responses across different lexical categories is limited, mainly due to constraints in publicly available datasets. Here, we examine how predictability influences brain responses, recorded at millisecond resolution using electroencephalography (EEG), with a focus on the N400 time window (300-500 ms post-stimulus) across different lexical and grammatical categories. Our results indicate that significant differences in N400 responses between high and low cloze probability levels were more pronounced for content words than function words. Among the two primary content categories, verbs exhibited greater N400 differences than nouns, while nouns carried more distinct information about their predictability than verbs. Moreover, we demonstrate that the decoding technique is more effective than the event-related potential (ERP) traditional analysis in capturing more detailed and distinct representations of cognitive processes over time.
Authors: Bogdan Raduta, Horia Velicu, Alexandru Preda, Serban Chiricescu
Abstract: Document classification is a solved problem in the laboratory and an unsolved one in the enterprise. The blocker is rarely model architecture; it is the labeling project that must precede a model and the institutional fear of letting a model retrain itself once one exists. We present SIFT (Self-Improving, Frozen-gate Training), a dynamic classifier service, which attacks both. SIFT serves classification from a deliberately cheap, CPU-bound pipeline, a SPLADE sparse encoder feeding a LightGBM head, and escalates only the low-confidence minority of pages to an LLM judge. The judge's verdicts are written back into a labeled corpus, so the expensive model continuously teaches the cheap one: the escalation rate falls, the corpus grows from production traffic rather than from an up-front annotation effort, and accuracy compounds with use. Onboarding a new document family requires only a declarative bundle, label space, anchor phrases, and a judge glossary, not a labeling project. The harder problem is safety: an autonomously retraining classifier can silently regress. SIFT resolves this with a two-part promote gate, a critical-label F1 regression check plus a frozen golden regression set the model is never trained on, either of which vetoes promotion. This turns "retrain monthly without a human" from reckless into routine. We describe the architecture, the self-feeding corpus loop, the frozen-gate promotion mechanism, and an illustrative multi-domain deployment, and we discuss the economics of a classifier whose marginal labeling cost trends toward zero.
Authors: Yuchuan Tian, Yingte Shu, Wei He, Shuo Zhang, Tianchen Zhao, Chao Xu, Xinghao Chen, Yunhe Wang, Hanting Chen, Yu Wang
Abstract: Large language models (LLMs) largely rely on Transformers, where self-attention provides global token interaction but does not explicitly encode the locality of natural language. We study whether lightweight depthwise convolutions can supply this local inductive bias without materially increasing model size. Our macro-level ablation compares convolution at 17 locations in a Qwen3 Transformer block and finds the best results when convolution is applied to the projected queries, keys, and values before attention. A subsequent micro-level study favors a residual depthwise convolution with kernel size $k=3$, without additional normalization or activation. Across Qwen3 models and several pre-training data budgets, this design improves the average accuracy on seven downstream benchmarks while adding less than $0.01\%$ parameters. A representation-level case study further suggests that the convolution makes repeated token IDs more sensitive to their immediate context. These results support depthwise convolution as a lightweight complement to self-attention for modeling short-range token interactions.
Authors: Pilar S\'anchez-Gij\'on, Susana Valdez, Sof\'ia Calvo Del Barrio, Florence Bellemont, Anna Kokkinidou, Mihai Cristian Brasoveanu
Abstract: This paper reports on a collaboration between the Directorate-General for Translation (DGT) and the European Master's in Translation (EMT) to localise the MMLU dataset into 11 European languages. Beyond creating a more inclusive benchmark for LLM evaluation, the project offers master's students authentic, project-based professional training in translation, revision, project management, and multilingual coordination, while highlighting key methodological, administrative, and workflow challenges.
Authors: Liam Swayne
Abstract: Introducing Relay-Bench, an unsaturated, holistic, text-only benchmark that measures LLMs' ability to complete an assortment of tasks from distinct domains in a single prompt. The leading model, GPT-5.5 (xHigh), scores 43.3%. The test set entirely consists of composite problems: groups of single-domain subproblems that are strung together into challenges that require reasoning across multiple domains in combination. Many of these problems then have layers of complexity added through prompt encoding and deliberate context bloat. Domains tested include visual reasoning, coding, math, information extraction (with a focus on web search), problem-solving, general knowledge, and data analysis. No restrictions are imposed outside of the model harness, and models are explicitly encouraged to leverage code-execution, web searches, and all available tools. All problems are composed of two to thirteen subproblems and do not require multi-modal input or output.
Authors: Polina Tsvilodub, Fausto Carcassi, Michael Franke
Abstract: Pragmatic language use requires reasoning about alternatives: the alternative expressions a speaker might have chosen, or the alternative interpretations a listener might entertain. Formal and computational models of pragmatics must therefore specify the sets of alternatives that interlocutors reason over, which is often done through manual specification. Here we propose a framework, ScAffolded Generative models for Explanation (SAGE), that combines the explanatory transparency of cognitive models with the generative flexibility of language models (LMs). SAGE decomposes a pragmatic process into three kinds of modules: proposers, which use LMs to generate an open-ended space of candidate alternatives; evaluators, which assess those alternatives (e.g., their semantics, complexity, or typicality); and selectors, which implement the rule-based computational steps of a cognitively motivated task analysis. We assess SAGE in three case studies spanning pragmatic generation and interpretation-referential expression generation, manner (M-)implicatures, and Gricean conversational implicatures. SAGE models are evaluated critically using established methods from computational cognitive modeling, including ablations, baseline comparisons, and quantitative fit to human data. Across studies, SAGE models achieved high accuracy and often outperformed baselines, but component-level analyses reveal an asymmetry: LM proposers reliably generated alternatives well-suited to pragmatic modeling, whereas LM evaluators are better at providing intuitive judgements rather than judgements of theoretical or formal measures. We discuss the promise and the limitations of neuro-symbolic models as candidate explanatory accounts of human pragmatic language use.
Authors: Sam Relins, Daniel Birks
Abstract: Purpose: Understanding how much of routine policing involves vulnerable people could inform resourcing, training, and multi-agency response, yet administrative data provide limited insight. We explore whether an LLM-based classification pipeline, developed on open-source US police data, can be adapted to estimate the prevalence of four vulnerability indicators - mental ill health, substance misuse, alcohol dependence, and homelessness - in UK police incident narratives, and when outputs can be treated as defensible measurements. Methods: We analyse nearly 3,000 de-identified incident logs from a UK police force, using a multi-stage pipeline combining repeated model inference, label aggregation, structured human review, and statistical correction. The pipeline runs on a locally hosted open-weight LLM, reflecting the secure environments police must work in. Results: LLMs can produce meaningful, if imperfect, prevalence estimates at scale. Mental ill health indicators are present in approximately one in five incidents, with lower prevalence for other indicators. However, naive LLM deployment is unreliable: single-pass classifications are unstable, and aggregated outputs systematically over-assign indicators relative to human judgement. Correcting these biases required substantial human input and statistical adjustment, leaving considerable uncertainty. Conclusions: While LLMs can extract information from unstructured police data, their outputs cannot be treated as valid measurements without careful methodological support. At the population level, defensible estimates are achievable but resource-intensive; at the individual level, errors remain frequent and unpredictable, limiting suitability for operational decisions. This study highlights both the potential and the constraints of LLM-based measurement in applied settings.
Authors: Suryakant Singh, Sejuti Majumder, Beatrice Knudsen, Joel Saltz, Prateek Prasanna
Abstract: Pathology report generation from whole-slide images (WSIs) is a rapidly growing multimodal learning problem, yet progress is difficult to measure because existing studies use heterogeneous datasets, model settings, visual encoders, and evaluation protocols. Moreover, commonly used natural language generation metrics, including BLEU, ROUGE, and METEOR, primarily reward lexical similarity and often fail to detect clinically consequential errors such as omitted diagnoses, hallucinated findings, or discordant tumor attributes. We present a standardized benchmark and evaluation framework for pathology report generation. The benchmark evaluates four representative methods across three datasets (TCGA, HistAI, and REG 2025) using three pathology foundation encoders (CONCHv1.5, UNI2-h, and H-Optimus-1). Our framework standardizes preprocessing, feature extraction, training, decoding, and evaluation, enabling fair comparison across models while providing a modular platform for integrating new methods, datasets, and encoders. A central contribution is the Clinical Report Quality Score (CRQS), a clinically grounded metric for evaluating factual correctness. CRQS maps reference and generated reports into structured clinical attributes and measures four complementary dimensions: clinical fact coverage, key information recall, hallucination rate, and clinical discordance, producing both an overall score and interpretable sub-scores. Experiments demonstrate that conventional language-generation metrics are weakly aligned with clinical correctness and frequently overestimate report quality. In contrast, CRQS reveals clinically meaningful differences between models and encoders that lexical metrics fail to capture. Together, the benchmark, public plug-and-play framework, and CRQS establish a reproducible foundation for rigorous evaluation of pathology report generation.
Authors: Tapan Parikh
Abstract: When a language model must choose one answer from a large space of equally valid options, a format clause -- "Reply with JSON only" -- changes which answer it chooses. We re-run the One-Word Census (arXiv:2607.12796): 31 wide-answer-space category prompts asked of 44 models, now with the reply requested in JSON -- no schema enforcement, no constrained decoding, only the request. Convergence deepens sharply: on the unconstrained "Pick a word" prompt the modal answer rises from 41% to 64% of the pool and distinct answers fall from 52 to 36; mean answer-choice surprisal drops from 1.80 to 1.58 bits. The tax is progressive: six of 44 models move individually (BH-FDR q=.10), all toward the mode, led by the most distinctive models, while the conformist floor is immobile. It is a sharpener, not a re-indexer -- the plain-chat modal answer survives in 28 of 31 categories. Defaults are register-indexed: a within-run re-sample (n=20) finds JSON shifts 53% of a model's stable chat defaults, mostly back to the crowd, and installs defaults absent from chat (Claude Fable 5 answers "cerulean" for colour 0% of the time in chat, 100% in JSON). Full-battery controls reveal a register gradient: compression is significant and specific to the answer-delivery formats models are trained to speak (JSON -0.22 bits, p=.0002; XML -0.19, p=.002), absent for YAML and CSV, and reversed for an arbitrary bracket wrapper (+0.13, p=.009) -- weighing the mechanism toward tool-use post-training. Enforcing the schema at the decoder (response_format) compresses no further than the request (-0.03 bits): the collapse lives in the model's response to the register, not the decoder. Structured output is how software consumes language models, and that surface is served by a measurably more homogeneous model than the chat surface on which models are evaluated, compared, and chosen.
Authors: Jia Ao Sun, Hao Yu, Fengran Mo, Zhan Su, Yuchen Hui, Bang Liu, Jian-Yun Nie
Abstract: Knowledge graph question answering (KGQA) requires navigating from topic entities to an answer several relations away. Recent methods prompt a frontier LLM to explore the graph through a retrieval tool, but their reliance on frontier-scale inference makes them costly to deploy. We present Search-on-Graph-R1 (\sogrone{}), which internalizes this navigation into a compact 8B model through supervised fine-tuning (SFT) followed by reinforcement learning (RL). Our central idea is to scaffold a frontier teacher with each question's gold SPARQL query, so the teacher traverses a known answer-bearing path with a live \texttt{Search} tool rather than having to discover the path itself. Since every call executes against a live Freebase server, the resulting trajectories are grounded in the knowledge graph by construction. On WebQSP, CWQ, and GrailQA, \sogrone{} at 8B surpasses every frozen frontier-LLM system in our comparison and posts the strongest results on CWQ of any system we compare against. It does so using no auxiliary module at inference and no LLM judge during training. Isolating each training stage shows that SFT and RL contribute complementary gains, our approach transfers across model families, and RL learns to reach answers in fewer \texttt{Search} calls than its SFT initialization.
Authors: Abir Harrasse, Michael Lan, Hunar Batra, Fateme Hashemi Chaleshtori, Chaithanya Bandi
Abstract: Reasoning-specialized language models show large performance gains over base models, yet the internal changes responsible for improved multi-step reasoning remain poorly understood. It is unclear whether reasoning fine-tuning improves local token-level competence or globally reorganizes how models structure inference over time. We address this question by modeling Chain-of-Thought reasoning as a switching dynamical system (SDS), in which internal representations evolve under discrete latent policy states. Our framework combines time-aware contrastive representation learning with discrete regime discovery to recover latent policies from activation trajectories. Across four benchmarks and model scales from 1.5B to 32B parameters, reasoning-fine-tuned models exhibit richer latent-policy organization than their base counterparts, characterized by more differentiated transition structure and model-dependent changes in state utilization, persistence, and mixing. The recovered regimes exhibit functional specialization aligned with distinct reasoning stages, and extensive controls confirm that their structure is not explained by correctness, representation learning, or modeling priors, but depends on the coherent temporal organization of reasoning trajectories. Causal interventions further show that the regimes are functionally meaningful: state-swap ablations reduce one-step predictive fit, while transplanting reasoning dynamics into base models improves performance on challenging reasoning problems. Finally, SDS-guided pruning of failure-prone reasoning prefixes outperforms self-consistency in 11 of 12 model-dataset settings, with gains of up to 12.5 percentage points. Together, our results suggest that reasoning fine-tuning globally reorganizes latent dynamics, offering a new lens for mechanistic analysis and process-level control of reasoning models.
Authors: Yixuan Wang, James Lester, Shashank Srivastava
Abstract: Persona prompting is widely used to steer LLM agent behavior, yet the narrative framing of a task can matter more than the assigned persona. We isolate this effect through structural isomorphism, constructing three text-based investigation games that share the same action space, stage progression, and resource constraints while varying only task narrative: disease investigation, IT troubleshooting, and murder mystery. Across 1,890 sessions spanning 3 models and 10 personas, we identify narrative priors: systematic action tendencies activated by a task's story framing, independent of its decision structure. Narrative priors explain 5-31x more behavioral variance than persona, are consistent across model architectures, and in two of three domains are negatively associated with task success. Persona effects that do transfer across narratives arise from behavioral anchors, persona descriptions whose language maps directly onto shared actions. Causal interventions confirm this: removing anchor words from a high-transfer persona reduces cross-narrative consistency by 95%. Our framework also generalizes to a held-out fourth narrative and yields a persona-selection method that improves cross-narrative transfer. These results suggest that LLM behavior that survives narrative changes should be grounded in concrete actions rather than abstract descriptions.
Authors: Abhidip Bhattacharyya, Shira Wein
Abstract: Discourse relations provide document structure, critical to language understanding and enabling language model performance and ethicality. In this work, we investigate how instruction-tuned Transformer models (LLaMA and Mistral) encode discourse relations in English, with a particular focus on the contrasting relations of causation and antithesis. Framing the task as a next-token prediction task and applying a suite of interpretability techniques to test model internals, our findings show that certain early layers make predictive decisions at mid-sequence tokens, while some mid-level layers finalize their decisions closer to the last token. Most of the remaining layers primarily propagate earlier decisions rather than actively influencing them. Additionally, we observe that some layers exhibit a preference for one answer over alternatives, suggesting asymmetric representation of discourse-based reasoning.\footnote{Our code is available at https://github.com/abhidipbhattacharyya/causation_vs_antithesis}
URLs: https://github.com/abhidipbhattacharyya/causation_vs_antithesis
Authors: Zijie Liu, Jinhao Duan, Gaowen Liu, Sijia Liu, Tianlong Chen
Abstract: Machine unlearning for vision-language models (VLMs) remains underexplored. Unlike language models, VLMs combine a language backbone with visual components, which makes unlearning more complex. There is a surprising phenomenon when moving from single-modality unlearning to VLM unlearning: a target forgotten by the standalone language backbone can still be recovered when image information is given to the full VLM. This shows that text-only feedback is not enough for reliable VLM unlearning. Motivated by this observation, we propose Stochastic Meta-Unlearning (SMU), a bilevel framework that uses VLM-level feedback to learn an unlearning-ready initialization. In the inner loop, SMU applies a few unlearning steps to the language backbone using text data. In the outer loop, SMU recomposes the updated backbone with the frozen VLM and evaluates forgetting and utility at the VLM level. This design makes the unlearning update aware of the final multimodal behavior, while still keeping the update local to the language backbone. Experiments on two VLMs, two multimodal meme datasets, and three baselines show that SMU achieves the best overall forget-retain trade-off. Compared with the strongest baseline for each metric, SMU reduces average Forget accuracy by 10.52 points and improves average Retain and Test accuracy by 20.10 and 17.01 points, respectively. More importantly, SMU also transfers to new forgetting targets and to different meta-test unlearning methods. These results suggest that VLM-level feedback can make language-backbone unlearning more reliable and more transferable for VLMs.
Authors: Wei-Rui Chen, Samar M. Magdy, Chiyu Zhang, Wenhui Zhu, Zhipeng Wang, Muhammad Abdul-Mageed
Abstract: Latent-reasoning looped language models (LoopLMs) offer a different scaling path for machine translation (MT): instead of increasing parameter count or emitting explicit chain-of-thought tokens, they spend additional recurrent computation inside hidden states. We introduce LatentMT, the first systematic study of latent-reasoning LoopLMs for machine translation. LatentMT adapts a small 2.6B-parameter backbone model with lightweight training. Across 32 translation directions spanning high-, mid-, and low-resource languages, LatentMT achieves performance comparable to models three to five times larger. It is competitive in a high-resource language and achieves state-of-the-art performance on both mid-resource and low-resource languages. Studying the behavior of scaling the number of recurrent reasoning steps, we find that recurrent computation consistently improves translation quality in early steps, then saturates quickly afterwards. Our mechanistic analysis shows that hidden-representation differences shrink along the recurrent reasoning-step axis, supporting the observed saturation in performance. Finally, our efficiency analysis shows that LatentMT requires lower training and inference compute than much larger non-latent-reasoning models with similar performance, making latent recurrent computation a promising path toward compact, efficient, and strong machine translation.
Authors: Abdul Basit Tonmoy, Kazi Fardinul Hoque, Md. Shahrier Islam Arham, Arman Luthra
Abstract: A single embedding space that covers text, images, video, and audio lets one index serve every query a user can pose. Embedding models built on vision-language backbones now lead text/image/video retrieval benchmarks but lack audio entirely, while audio-text retrieval is led by specialist systems that serve no other modality. We present the Fusion Embedding family, which adds audio to a frozen vision-language embedding base whose parameters are never updated: generation 1 (fusion-embedding-1) trains only a 16.4M-parameter connector between a frozen audio tower and the frozen base, and generation 2 (fusion-embedding-2) adds modality-gated deep adapters (44.2M parameters) whose branch never executes on text, image, or video inputs: their outputs are bit-for-bit those of the released base, verified after every training run. Because the base already binds text, images, and video, aligning audio to text alone makes audio-image retrieval emerge, with zero paired audio-visual training data. Alongside the recipe we map its design space with controlled negative results (rewriting training captions with an LLM, substituting a leaderboard-stronger audio tower, and widening the connector each reduce retrieval) and with training-protocol findings that we expect to transfer to any frozen decoder-LM embedding backbone. Both generations train in hours on a single GPU. Weights, code, and the evaluation harness are openly released.
Authors: Qiuli Zhou, Jingyuan Yao, Shengeng Tang, Hongzhi Chen, Jun Tang, Richang Hong
Abstract: Stance detection aims to identify whether a text expresses a favorable or opposing attitude toward a given target, and serves as an important task for various downstream applications. Although existing studies have achieved strong performance in monolingual settings, especially in English, many low-resource languages such as Catalan still lack sufficient annotated data for training effective models. Cross-lingual stance detection alleviates this problem by transferring stance knowledge from resource-rich languages to low-resource languages. However, most existing methods mainly rely on semantic alignment between texts and targets, while ignoring the reasoning process required for reliable stance inference. Although Large Language Models provide strong reasoning ability, their high computational cost and inference latency limit practical deployment. To address these limitations, we propose a rationale-guided knowledge distillation framework for cross-lingual stance detection. Specifically, we use Chain-of-Thought prompting to guide Large Language Models in generating informative rationales, and distill the resulting reasoning knowledge into a compact student model. We further design a dual-path distillation mechanism to align rationale-enhanced and rationale-free representations, together with their prediction distributions. In addition, two contrastive learning strategies are introduced to improve stance discrimination. Experiments on multilingual benchmarks demonstrate that our method consistently outperforms competitive baselines.
Authors: Shaswata Mitra, Subash Neupane, Trisha Chakraborty, Himanshu Tripathi, Sudip Mittal, Aritran Piplai, Shahram Rahimi
Abstract: Large Language Models (LLMs) are increasingly fine-tuned for critical-domain Question-Answering (QA), yet choosing which small model to adapt, before paying the cost of adaptation, remains difficult. Fine-tuning can improve domain alignment, but it may also erode prior knowledge, weaken instruction-following, or increase hallucination, especially when labeled data are scarce or rapidly evolving as in cybersecurity. We present FiT (Find before Fine-Tune), a task-oriented diagnostic framework that characterizes small LLMs along three capabilities required for cybersecurity QA: vocabulary recognition, parametric knowledge, and contextualization of retrieved information. Using FiT, we conduct an empirical study of five open-weight 7-billion-parameter models under two fine-tuning regimes. We find that fine-tuning does not uniformly help: it consistently degrades vocabulary and parametric knowledge in small models, and the two regimes trade off differently. Knowledge-focused tuning causes moderate, rank-preserving degradation, whereas instruction-focused tuning collapses measured knowledge through induced abstention, inverting the knowledge ranking while leaving retrieval-grounded contextualization essentially intact. We quantify these regime-specific patterns with rank-correlation analysis and show that pre-fine-tuning FiT scores anticipate the direction of post-tuning change. Our results suggest that task-oriented diagnosis can screen out unsuitable models, avoid unnecessary fine-tuning, and support safer deployment of small LLMs in cybersecurity QA pipelines.
Authors: Xingda Yu, Yining Li, Xinzhang Liu, Zhihao Yang, Haowei He, Chao Wang, Yongxiang Li, Shuangyong Song
Abstract: Recent work extends Transformer residual pathways along two complementary axes: historical retrieval selects information from earlier depths, whereas multi-stream methods maintain multiple residual trajectories. These capabilities have largely been studied in isolation, and assigning an independent retriever to each stream still prevents one trajectory from influencing depth selection in another. We propose Dual Attention Residuals (DAR), which brings multi-stream interaction into historical retrieval through reciprocal cross-stream addressing. For each target stream, DAR computes depth weights from normalized states in the opposite stream and applies them to values from the target stream's own history. The retrieved states are combined for an unchanged Transformer branch and updated through constrained gated writes; a block-form variant operates on block-level histories to control overhead. Across dense models from 0.1B to 1B parameters and a 7B sparse-MoE model, DAR consistently improves validation loss over standard residual Transformers and Attention Residuals. Routing ablations show that the gain cannot be explained by an additional stream or value projection alone. Representation and intervention analyses further show that reciprocal cross-stream selection preserves depth-wise diversity and avoids the redundancy or functional imbalance observed in alternative two-stream designs.
Authors: Yueqi Xing, Houbo He, Jolie Wang, Erin Ni, Shikai Wang, Qiufeng Li, Weidong Cao, Taiyun Chi
Abstract: Large language models (LLMs) have driven rapid progress in electronic design automation (EDA), yet their application to radio-frequency (RF) circuit design remains limited by the scarcity of domain-specific datasets and standardized benchmarks. We present RF-Agent, which addresses this gap through textbook-driven knowledge distillation. A multi-agent Question-Thinking-Solution-Answer (QTSA) pipeline converts a subsection-level corpus from seven canonical RF textbooks into the first-of-its-kind RF-domain reasoning dataset (over 11,000 samples) with a dedicated multiple-choice benchmark. On this benchmark we study two adaptation strategies: supervised fine-tuning (SFT) and three retrieval-augmented generation (RAG) configurations (semantic, keyword, hybrid). Across multiple LLM families, domain-specific SFT significantly improves RF reasoning, especially for small and medium-sized models; among RAG configurations, semantic retrieval performs best, indicating embedding-based context alignment suits RF reasoning better than naive fusion. The dataset and benchmark provide a reusable foundation for future work on LLM-aided RF circuit design.
Authors: Ziming Wang, Yinghua Yao, Changwu Huang, Ke Tang, Xin Yao
Abstract: Chain-of-thought (CoT) reasoning is widely used to improve both the performance and interpretability of large language models (LLMs), yet the generated reasoning may not faithfully support the final answer. We study this problem from a causal perspective, where a faithful CoT process should follow the chain $Z\rightarrow X\rightarrow Y$, with $Z$, $X$, and $Y$ denoting the instruction, reasoning chain, and final answer, respectively. In this process, the instruction should affect the answer only through the reasoning chain. However, conventional autoregressive LLMs condition answer generation on both the instruction and the CoT, which still allows a direct instruction-to-answer shortcut. To address this issue, we propose CASE, a framework that combines training-time causal alignment and inference-time structural enforcement. During training, CASE builds counterfactual-CoT, biased-instruction, and empty-instruction datasets, and applies selective-loss fine-tuning to strengthen CoT-to-answer dependence while suppressing instruction shortcuts. During inference, CASE masks direct attention from instruction tokens to answer tokens, preventing the model from bypassing the generated CoT. We provide an information-theoretic analysis showing how these components promote faithful chains. Experiments on three models and four benchmarks show that CASE achieves a 37\% average per-setting relative improvement in overall CoT faithfulness over the strongest baselines, exhibits stronger cross-dataset faithfulness transfer, and maintains competitive average accuracy. Code is available at https://github.com/oddwang/CASE.
Authors: Shubham Kumar Nigam, Shubham Kumar Mishra, Noel Shallum, Kripabandhu Ghosh, Arnab Bhattacharya
Abstract: This comprehensive study introduces an advanced Artificial Intelligence for Indian Legal Question Answering (AILQA) system tailored to the Indian legal context. AILQA leverages a variety of embedding and generative models, including recent Large Language Models (LLMs), to address the unique challenges posed by the intricate and diverse nature of Indian legal texts and to enhance the accuracy and reliability of responses to legal questions. We conducted rigorous evaluations using both lexical and semantic metrics, enriched by expert legal feedback, to ensure relevance and accuracy. Our findings underscore the effectiveness of the Retrieval-Augmented Generation (RAG) paradigm in improving answer quality, particularly in complex legal domains. Additionally, we assessed performance on standardized tests such as the All India Bar Examination (AIBE), thereby providing a robust benchmark for practical applications. Under the study's evaluation protocol, some AI-generated responses received higher ratings than the available reference answers, particularly when they contained accurate and relevant supporting details. This finding is specific to the evaluated dataset and rating criteria and should not be interpreted as evidence that the models generally outperform qualified legal professionals. We also discuss the challenges encountered, such as the need for precise context and the risks of model hallucination, and propose directions for future research to further refine AI capabilities in the legal field. This study aims to pave the way for enhanced legal decision-support systems, making them more accessible and effective for legal professionals and the public alike.
Authors: Shu Wei, Jingjing Wu, Lingshu Zhang, Qunyi Xie, Hao Zou, Le Xiang, Xu Fan, Yangliu Xu, Manhui Lin, Xiaolong Ma, Cheng Cui, Tengyu Du, YY
Abstract: Efficient teamwork typically combines global coordination with parallel execution, a principle not yet fully reflected in unified Vision-Language Model (VLM)-based document parsers. Existing unified parsers process an entire page jointly but generate its output through a single token-by-token autoregressive trajectory, creating a sequential bottleneck that grows with document length. Such full-page sequential generation overlooks a key property of document parsing: layout must be analyzed globally, whereas block content can be parsed in parallel. Based on this observation, we introduce HPD-Parsing, which replaces full-page autoregressive generation with a Hierarchical Parallel Decoding paradigm. A main layout branch organizes the overall document structure and dynamically assigns block-level content decoding to concurrent branches, while progressive multi-token prediction (P-MTP) further reduces the decoding steps within each branch. Experiments on public benchmarks show that HPD-Parsing achieves 4,752 tokens per second, delivering $2.62\times$ the throughput of the fastest existing document parsing model and $3.06\times$ that of the vanilla autoregressive baseline, while maintaining competitive parsing accuracy. These results establish hierarchical parallel decoding as an effective alternative to full-page autoregressive generation, opening a new direction for efficient unified document parsing.
Authors: Mark Gatere
Abstract: Automatic speech recognition (ASR) for African languages is constrained by orthographic inconsistency, annotation artifacts, missing audio, speaker and domain imbalance, and evaluation procedures that differ from deployment. We present an end-to-end engineering study adapting NVIDIA Nemotron 3.5 ASR Streaming 0.6B to Kikuyu, Dholuo, and Kalenjin. Starting from a Kenyan Swahili-adapted checkpoint, we retain its cache-aware FastConformer RNN-T, prompt conditioning, and streaming decoder during full-parameter fine-tuning. The study covers corpus auditing, Unicode normalization, split checks, duration filtering, low-rate continuation, validation-based checkpoint selection, true-streaming evaluation, artifact preservation, and isolated serving. On internal, adaptively consulted evaluation sets excluded from gradient updates at context [56,13], selected Kikuyu and Dholuo models achieve 42.97% and 33.98% WER, respectively. Dholuo records 9.59% CER and 8.13% no-space CER under its frozen historical label policy; Kikuyu records 7.79% no-space CER. Kalenjin remains a work in progress: v1-v reaches 68.74% WER on a 2,411-row clean-v3 diagnostic subset excluding long-pause annotations, digit-bearing references, and targets shorter than three tokens. Its checkpoint selection used a mixed-source validation manifest containing test-origin rows, so the score is not an independent generalization estimate. We also report negative findings involving non-speech labels, short-utterance over-generation, boundary-sensitive WER, and cloud job-lifecycle failures. We make no state-of-the-art claim because the internal sets, repeated consultation, and normalization differ from public benchmarks. This work provides an auditable account of adapting a multilingual streaming model into language-specific systems without discarding streaming constraints.
Authors: Xiaomeng Hu, Jiaqi Hu, Hao Chen, Qi Zhang, Zhanming Shen, Wentao Ye, Junbo Zhao
Abstract: With the rapid advancement of large language models (LLMs), modern systems not only possess strong foundational capabilities and extensive knowledge, but can also solve complex problems via long, multi-step reasoning. However, as reasoning traces become longer, LLMs may produce a substantial amount of hallucinated content during the reasoning process, which is often difficult to detect. In this work, we conduct a fine-grained analysis of hallucinations arising in LLM reasoning and find that the reasoning traces are particularly prone to Context-Sensitive Factual Hallucinations: cases where the model actually has the relevant knowledge, yet makes factual errors due to contextual interference during reasoning. To address this issue, we propose Step-level Self-Consistency Group Relative Policy Optimization (SSC-GRPO), which assigns step-level rewards to reasoning traces by computing self-consistency scores of individual steps across multiple rollouts. Compared with prior methods, SSC-GRPO achieves state-of-the-art performance on both mathematical reasoning benchmarks and hallucination leaderboards. Our results offer a new perspective for detecting and mitigating hallucinations in the reasoning process of large language models.
Authors: Laurin Wagner (nyra labs), Mario Zusag (nyra labs), Bernhard Thallinger (nyra labs)
Abstract: Modern ASR models trained on heterogeneously annotated data treat transcription style (verbatim vs. intended) as an uncontrolled latent variable, causing measurable decoding instability, evaluation confounding (up to 60% of reported WER attributable to style mismatch), and unreliable word-level timing. We show that models already encode both styles; the challenge is controlled activation. Using coverage-aware decoder task tokens trained on parallel verbatim/intended transcript pairs, we raise German disfluency F1 from 10% to 79% zero-shot, despite English-only training. Full English-only fine-tuning surpasses all baselines in verbatim accuracy, disfluency detection, and intended-mode quality across both languages. We further introduce supervised cross-attention fine-tuning that improves word-level timestamps on disfluent speech beyond forced-alignment baselines. Finally, we propose verbatimize, a new task enabling scalable creation and enrichment of speech corpora with high-quality canonical verbatim transcriptions.
Authors: Rufael Marew, Amr Keleg, Hanan Aldarmaki
Abstract: In this work, we address diacritic restoration for Arabic speech transcripts. Most speech data are undiacritized, limiting the ability of modeling fine-grained phonological distinctions. The speech modality has recently been explored as a way to complement text-based diacritic restoration efforts. We propose an efficient non-autoregressive approach for speech-to-text diacritization based on Connectionist Temporal Classification (CTC). Our method incorporates hard constraints during decoding by constructing a character-level diacritization lattice from an undiacritized transcript and restricting hypotheses to valid diacritized realizations. We evaluate on Classical Arabic and Modern Standard Arabic test sets (namely, ArVoice and ClArTTS) against a more computationally-complex multi-modal diacritic restoration baseline, and show statistically significant reductions in diacritic error rates in both, demonstrating that the proposed approach offers both performance and efficiency gains.
Authors: ChaoJin Zhao, Xuan Jiang
Abstract: Textual skills provide a lightweight way to improve frozen language-model agents, but their self-evolution normally requires a stable validation signal. Such signals are natural in mathematics or code, where an answer can be checked after it changes, yet are problematic in open-ended dialogue: changing the assistant response also changes the user's next reaction, so a logged reaction cannot directly evaluate a counterfactual response. We propose future-feedback skill evolution, which first redirects self-evolution from prescribing the current answer to predicting whether the observed answer will lead to a positive or negative subsequent user signal. This prediction task is verifiable on fixed logged tuples and therefore supports validation-gated textual optimization. The evolved feedback skill captures interpretable criteria for response quality and can subsequently serve as a diagnostic and optimization target for answer skills. On a proprietary, privacy-preserving sales-assistant dataset, careful quality filtering and a balanced resolved/unresolved split yield more than 75% prediction accuracy. Beyond this result, the central contribution is a formulation that converts otherwise moving conversational feedback into a fixed offline learning target, enabling reproducible skill evolution without placing every candidate skill in live traffic. We discuss the boundary between observational verification and counterfactual validity, and position the method as an offline optimization stage rather than a replacement for final human or online evaluation.
Authors: Himel Ghosh, Ahmed Mosharafa, Georg Groh
Abstract: We present AutoJourn, a demonstration system for multi-perspective news generation and bias-aware evaluation using large language models (LLMs). The system tackles three core challenges in responsible automated journalism: extracting diverse perspectives from unstructured social media discussions, generating summaries that preserve viewpoint diversity, and detecting or mitigating bias in AI-generated news. The pipeline integrates advanced prompt engineering with optional retrieval augmentation to produce semantically diverse perspective sets, a multi-perspective summarisation module that merges conflicting viewpoints into balanced summaries, and a bias analysis suite supporting sentence-level bias detection and type classification in the generated news article, and automatic neutralisation. Users can inspect perspective clusters, compare stance-specific summaries, generate news articles, and apply bias-aware rewrites directly in the interface. We evaluate each component with intrinsic metrics -- semantic diversity, summary quality, and bias reduction and show improvements over strong baselines while maintaining content fidelity. A live, publicly accessible demo accompanies the paper to facilitate reproducibility and further research on socially responsible automated journalism.
Authors: Vanessa Toborek, Florian Seiffarth, Sebastian M\"uller, Tam\'as Horv\'ath
Abstract: Despite more than a decade of curriculum learning (CL) research in NLP, the field lacks a principled account of which difficulty function or scheduler to use for a given problem. To understand what has hindered progress towards this account, we propose a fine-grained taxonomy separating difficulty evaluation from training scheduling to enable systematic analysis of CL strategies. For difficulty evaluation, we distinguish attribution source and task dependence, revealing difficulty as a perspectival concept encoding different assumptions about what makes an instance hard to learn. For scheduling, we provide the first formalisation of CL schedulers in terms of expected training contribution, enabling comparison across implementations by introducing retention regimes and monotonicity properties. Applied in a dedicated analysis of CL works in NLP, our taxonomy reveals a systematic incomparability problem: prior works conflate distinct notions of difficulty and scheduling, often pursuing different objectives under the same CL label -- hindering comparison and the accumulation of a coherent evidence base. Beyond diagnosis, the taxonomy supports the design, analysis, and comparison of CL strategies, and motivates evaluation practices that disentangle the sources of observed improvement.
Authors: Guofeng Zhang, Yizeng Quan, Huaiyi Fang, Jianwei Lv, Jinyao Liu, Xunxu Duan, Lening An, Yu Ouyang, Junfeng Wang
Abstract: Multi-turn medical consultation agents must decide what to ask, adapt to patient responses, and determine when the collected evidence is sufficient. However, coupled evaluation conflates the quality of the policy-elicited history with policy-specific terminal diagnosis generation: strong generation can compensate for a thin history, while weaker generation can obscure a rich one. We introduce MedDDC-Eval, a diagnosis-decoupled testbed that treats elicited history as the comparison object and holds the history-to-diagnosis mapping constant through a shared frozen reader. Across two held-out sources, a grounded interface and an auditable diagnosis-trajectory-efficiency (D/T/E) harness measure diagnostic usefulness, information acquisition, and efficiency. Directional semantic coverage followed by deterministic one-to-one assignment yields coherent precision-recall counts for open-ended items, with at most one credited match per prediction or reference. Holding histories fixed, changing only the diagnostic reader shifts diagnosis F1 by 2.2-19.0 points and reverses 18% and 36% of pairwise policy orderings on the Record and Dialogue splits. We further apply standard Group Relative Policy Optimization (GRPO) over interactive multi-turn rollouts to post-train Qwen3-32B using diagnosis-result and trajectory feedback. On the 100-case Record and 70-case Dialogue splits, the trained policy improves over its initialization by 9.7 and 4.6 total-score points; removing either primary signal lowers held-out joint performance. These results show that MedDDC-Eval supports controlled attribution, interpretable elicited-history measurement, and evaluation-guided evidence-acquisition policy development.
Authors: Tuo Liang, Zhe Hu, Disheng Liu, Jing Li, Yu Yin
Abstract: Multimodal humor in memes, cartoons, and comics remains difficult for AI systems because intended meaning depends on non-literal mechanisms, shared cultural knowledge, and communicative intent rather than literal scene description. This survey focuses on visual humor understanding in single-image and multi-panel artifacts, while treating humor generation as an emerging downstream frontier. We position the literature against prior humor, sarcasm, and general MLLM surveys and organize it using a capability-centric hierarchy spanning recognition, interpretation and reasoning, and generation. Under this lens, we synthesize benchmark design, evaluation protocols, and modeling paradigms, tracing the field's shift from task-specific fusion models to large-model approaches based on multimodal alignment, evidence-grounded reasoning, and controlled generation. We conclude by highlighting the main barriers to progress: shortcut-prone evaluation, limited cultural and narrative coverage, weak evidence grounding, and unresolved safety and ownership concerns.
Authors: Laurin Wagner (nyra labs), Bernhard Thallinger (nyra labs), Miroslav Stankovic (nyra labs), Mario Zusag (nyra labs)
Abstract: Discrete speech tokenizers aim to disentangle semantic from acoustic information, yet targets from self-supervised learning (SSL) models like HuBERT retain non-linguistic variation: speaker identity, prosody, and channel conditions leak into the tokens, inflating entropy. Our key insight is that when enough speakers utter the same words under varying conditions, linguistic content is the only shared factor. We propose PINT (Parallel INvariant Tokenization), which fine-tunes an SSL encoder with alignment losses across parallel utterances and augmentations to distill this shared residual. PINT collapses identical words onto consistent token sequences, drastically reducing conditional entropy. Unlike ASR text, PINT tokens preserve frame-level temporal grounding and serve as drop-in semantic targets for audio codecs. Experiments show a 98.7% relative reduction in speaker probe accuracy (93.1% to 1.2%), a 42% lower ABX error rate, and 27-30% lower LM perplexity versus baselines, confirming that the right invariance is key to efficient learning.
Authors: Ilse Huisman, Rares Popa, Yuanyuan Zhang, Odette Scharenborg
Abstract: Humans are often considered to be the best listeners and seen as the upper-bound performance of automatic speech recognition (ASR) systems. We present a preliminary comparison of the performances of state-of-the-art ASR systems and Dutch native listeners on the recognition of "diverse" speech, specifically Dutch child and older adults' speech and Flemish. Google Telephony outperformed the other ASR systems. Importantly, the ASR systems showed similar performance to the listeners, and in specific cases even outperformed them. Slight performance differences between the listeners and ASR systems were found related to speaker's age and regional accents and utterance length. Future research should focus on making ASR systems more robust to acoustic variability related to aging and regional accents. A comparison of ASR recognition performances on the test stimuli and the full Jasmin-CGN test sets showed the influence of the specific test sets on the conclusions regarding benchmarking human and ASR performance.
Authors: Yu Wang, Ming Fan, Xicheng Zhang, Zhiyong Li, Zhihu Wang, Caiyue Xu, Dahai Hu, Ting Liu
Abstract: Chain-of-thought (CoT) supervision exposes intermediate rationales, but flat rationale targets usually optimize a single reasoning sequence and provide limited supervision on how local conclusions should support later decisions. We introduce Dependency-Aware Intermediate QA Supervision (DAIS), a training-time framework that converts filtered teacher rationales into stage-level QA records. Each intermediate record predicts a local answer conditioned on the previous states needed for that decision, while the final-answer record keeps the original task format; evaluation therefore uses only the original input and optional context. Across GDPR, AIACT, MedQA, and FOLIO with multiple Qwen backbones, DAIS improves average final-answer accuracy over answer-only, flat chain-of-thought, and independent-QA baselines. On policy-compliance benchmarks, it achieves a largest gain of 5.6% and an average gain of 4.2% over the strongest non-DAIS baseline. Controlled ablations show that valid previous-state conditioning contributes beyond longer targets or additional intermediate text, supporting dependency-conditioned intermediate QA as a lightweight auxiliary supervision signal for standard final-answer inference.
Authors: Yiheng Wu, Jue Hou, Roman Yangarber
Abstract: Reliable Text Difficulty Assessment is a prerequisite for valid text simplification workflows and personalized learning applications. However, the development of robust assessment models is severely hindered by a critical bottleneck: the scarcity of expert-annotated corpora containing fine-grained difficulty levels (e.g., CEFR), particularly for lower-resource languages. This paper addresses this data scarcity problem in the context of a low-resource European language. We propose a cross-lingual data augmentation strategy that leverages machine translation to transfer labeled resources from high-resource languages to the target low-resource language. We train BERT-based regression models to predict difficulty scores and investigate whether synthetic, translated data can effectively supplement native training sets. Our experiments demonstrate that augmenting scarce native data with machine-translated corpora significantly improves the accuracy of difficulty estimation, offering a viable solution for languages lacking extensive expert annotations.
Authors: Maxime Gorres, Jan G\"opfert, Patrick Kuckertz, Noor Titan Putri Hartono, Heidi Heinrichs, Jochen Lin{\ss}en, Iain Staffel, Jann Michael Weinand
Abstract: Energy system models guide societally important decisions, but their credibility rests on quantitative assumptions that are difficult to source and audit. Meta-analyses can improve transparency and modeling practices, but the rapid growth of publications makes manual information extraction increasingly impractical. Consequently, databases are updated infrequently and efforts are often duplicated across research groups. Here, we demonstrate the highly accurate automated extraction of quantitative information from 76,000 energy system studies published since 2010. We compile 3.2 million structured quantitative data points together with 20 million associated metadata entries, spanning a broad spectrum of technologies, methodological approaches and system characteristics. Beyond providing input data for models, the resulting FAIR database make the energy systems literature itself analysable. We show where academic assumptions diverge from empirical observed data, and how research priorities vary at scale across technologies, regions and time. To facilitate broad use within the community, the database is provided through an interactive dashboard, enabling users to filter, analyse and download data according to their specific research needs.
Authors: Aixiu An, Michael Jungo, Eloi Eynard, Mark Drenhaus, Andreas Fischer, Jean Hennebert, S\'ebastien Rumley
Abstract: Neural machine translation (NMT) in the legal domain is a linguistically and conceptually demanding task, primarily due to the complexity of legal language and the high level of precision it requires. The recent emergence of reasoning-capable language models opens new possibilities for tackling such challenges. They add to a set of other previously proposed techniques to enhance the translation quality, which includes supervised fine-tuning and reinforcement learning. In this work, we perform a comparison between these various approaches. More particularly, we evaluate small language models such as Qwen3.5 4B, Qwen3.5 9B, and Gemma 3 12B enhanced with various re-training paradigms and compare their performances against frontier reasoning models. We focus on the Swiss legal system, which -- with its unique multilingual statutes -- offers a particularly challenging testbed for reasoning-augmented models. Our results show that the quality of small ``base'' models can be greatly enhanced, and that reinforcement learning with verifiable rewards can be applied to NMT in the legal domain and surpasses the translation quality of supervised fine-tuning. The performance of enhanced small models is close to the one of state-of-the-art reasoning models yet remains inferior. We also note that re-training paradigms yield diminishing returns as model size increase. The code and models are publicly available at https://github.com/aixiuxiuxiu/Legal-MT-SFT-RL.
Authors: Iker De la Iglesia, Johanna Ramirez-Romero, Jose Maria Villa-Gonzalez, Irune Urroz Garc\'ia, Ander Barrena, Aitziber Atutxa
Abstract: Clinical NLP evaluation remains dominated by multiple-choice question answering (MCQA), which scores only final-answer accuracy and cannot detect when a model reaches the correct diagnosis while grounding it in irrelevant, absent, or contradictory evidence. We introduce MIRA-Ev, a clinical argument mining benchmark built on Spanish M\'edico Interno Residente (MIR) licensing-exam cases, re-annotated by expert clinicians with span-level premises, claims, and directed support/attack relations, and released in parallel Spanish (native), English, and Basque versions, the first clinical argumentation resource in Basque. MIRA-Ev organizes evaluation into a three-tier task hierarchy: evidence sentence retrieval, argumentative component extraction, and relation classification.
Authors: Xuefeng Jin, Jiashuo Zhang, Teng Cao, Bin Yang
Abstract: Large language models (LLMs) have been widely applied to automated essay scoring (AES) and automated feedback generation (AFG). However, existing studies rely primarily on prompt engineering or supervised fine-tuning, while systematic research on reinforcement learning (RL) post-training and automated evaluation of feedback quality remains limited. We propose RLAES, a unified LLM framework that jointly optimizes essay scoring and feedback generation through RL. To make feedback quality measurable, interpretable, and usable for training, we introduce Rubric-based Feedback Evaluation (RFE), an essay-grounded feedback evaluation framework comprising 166 fine-grained binary rubric items and an LLM-as-judge. Building on RFE, we propose Adaptive Gated Feedback Optimization (AGFO), which activates rubric-based feedback rewards on demand during RL, reducing evaluation overhead while improving feedback quality. We also propose Adjacent Contrastive Reasoning (ACR) to improve ordinal score calibration by explicitly contrasting adjacent score levels. Experimental results show that the RFE framework captures essay-feedback consistency, exhibits strong pairwise discriminative power, and closely aligns with expert preferences. On the ASAP benchmark, RLAES-AGFO achieves the best scoring performance among LLM-based methods (QWK = 0.803), while maintaining feedback quality comparable to GPT-5.5 and avoiding the feedback degradation observed under score-only RL. Code and datasets are publicly available at https://github.com/hellomuyi/RLAES.
Authors: Michael Jungo, Aixiu An
Abstract: Reinforcement learning with verifiable rewards (RLVR) has been established as a viable paradigm for the post-training of Large Language Models (LLMs), including downstream tasks, such as Neural Machine Translation (NMT). With the latest research indicating that RLVR could be the preferred training method for translating legal documents due to the induced reasoning capabilities, it raises the question whether it is really attributed to the reasoning or more generally to the training paradigm. We investigate the importance of including the model's reasoning trace in the generated responses during both training and inference by systematically omitting it from one of the phases. Our experiments show that including the reasoning, specifically during inference, has a positive effect on the overall translation quality. Furthermore, we recognise that the reasoning leads to an increase in output tokens, hence we study the cost-quality tradeoff between the increased computational demands and the improved translation quality.
Authors: Haram Choi
Abstract: Prior work on human label variation (HLV) in natural language inference (NLI) has often relied on re-annotation resources that select items by disagreement level. An earlier study (arXiv:2607.15870) found that hypotheses containing non-upward monotonicity operators showed lower label agreement in ChaosNLI (Cliff's delta = -0.284), which is restricted to items whose majority label carries exactly three of five votes. We preregistered a replication of this boundary in the unselected populations that ChaosNLI was drawn from: the SNLI and MultiNLI development sets, using the same operator tagger and a four-level ordinal agreement outcome. The registered prediction fails. All seven contrasts return a positive Cliff's delta (non-upward items agree slightly more, not less), the only significant confirmatory contrast has the opposite sign to the registration, and every effect is far below our smallest effect size of interest (0.10). Robustness checks support the measurement: simulated tagger misclassification shrinks the effects rather than manufacturing them, and a manual re-tagging audit reaches four-class agreement of 0.875 on a fresh 200-item sample. We conclude that the earlier negative boundary is plausibly a structure conditional on low-agreement selection rather than a population-level property, and that HLV structure claims built on selected re-annotation resources should state their selection conditional explicitly.
Authors: Ziyi Wang, Yuhang Wu, Dongxu Piao, Xingyu Liu, Tianhui Zhou, Miao Liu
Abstract: Theory of Mind (ToM), the ability to infer other's beliefs, intentions, and states of knowledge, is central to social interaction, yet remains challenging for current Multimodal Large Language Models (MLLMs), especially in multi-party meetings where cues are distributed across speech and behavior. Existing multimodal ToM benchmarks mainly focus on video-grounded question answering over overt, externally verifiable signals, and provide limited coverage of latent social states and group dynamics. We introduce MeetingToM, a benchmark for complex social behavior reasoning in naturalistic multi-party meetings. MeetingToM targets meeting-specific phenomena such as \textbf{pseudo-consensus}, where apparent agreement masks private dissent under social pressure. The benchmark is hierarchically organized to evaluate ToM at increasing levels of social granularity, including (i) subject-level mental state prediction, (ii) dyadic-level addressee understanding, and (iii) group-level consensus reasoning. We provide a unified evaluation protocol and conduct systematic analyses of representative MLLMs, revealing persistent limitations in integrating non-verbal cues, inferring hidden attitudes, and distinguishing genuine consensus from pseudo-consensus. Our results highlight key challenges and establish MeetingToM as a testbed for advancing meeting-grounded ToM in multimodal models.
Authors: Alexander Manev
Abstract: Although Large Language Models (LLMs) demonstrate remarkable multilingual fluency, their internal knowledge representations remain disproportionately biased toward high-resource languages. This leads to cross-lingual factual inconsistency, where they shift their empirical answer distributions based solely on the prompt language. We investigate whether these biases can be mitigated at inference time, forcing an English-prompted model to answer as if it were queried in target languages (German, Spanish, Bulgarian), and evaluate four intervention strategies: zero-shot contextual steering (persona prompting), internal representation manipulation via Contrastive Activation Addition (CAA), and lightweight weight modification via Direct Preference Optimization (DPO) trained on benchmark-derived factual data as well as conceptual generalization data. To assess alignment, we curate a multilingual factual dataset alongside a novel generalization benchmark comprising culturally rooted queries to determine whether factual interventions transfer to broader target-centric preferences. Experiments on Gemma 3 12B Instruct reveal persona prompting to be the strongest overall intervention, balancing efficacy, safety, and out-of-domain generalization. While CAA yields sharp inconsistency benchmark shifts, it is configuration-sensitive and risks knowledge degradation. DPO-based adapters offer permanent, yet narrower and less transferable gains. These findings suggest that cross-lingual inconsistency is at least partly a selection problem, and that simple contextual interventions may outperform more invasive methods for robust, transferable alignment.
Authors: Netanel Eliav
Abstract: Practitioners make three prompt-design decisions with almost no controlled evidence behind them: how to format instructions and context (markdown, plain text, prose, or tabular), how many simultaneous instructions a system prompt can carry before compliance degrades, and how much context a model can hold before recall and honesty degrade. We report two controlled experiments crossing all three factors on one held, contamination-free synthetic corpus (the "Book of Veyra," 8,780 uniquely-named entities, deterministically regenerable from a fixed seed), evaluated across five models. Experiment 1 (960 calls/model) measures instruction-following decay as rule count N grows from 10 to 160, crossed with four formats and system-prompt vs. user-turn placement. Perfect-response rate collapses to zero by N=80 for every model, format, and placement. Placement produces effects at least as large as format at N=160 in most models, but the direction is model-specific. No model shows a reliable markdown advantage; one 35B model favors plain text instead. Experiment 2 (5,520 calls/model) measures recall accuracy, false-premise sycophancy, and absent-fact fabrication across a 2k-to-512k-token context ladder in the same four formats. Recall stays near ceiling through 64-128k tokens, then degrades sharply and format-dependently: one model's accuracy spread reaches 48 points at 128k tokens. Fabrication never occurs (0/5,760 probes), and sycophancy stays negligible (<=8.3%). What rises sharply near each model's context ceiling is outright refusal to answer (0% to 79-90%), distinct from sycophancy or fabrication. Neither pre-registered format ordering holds, and token overhead (+22% to +37% over plain text) further changes which format is preferable where accuracy spread is genuine. We release the full harness, corpus generator, and raw results (VeyraBench): https://github.com/iNetanel/veyrabench
Authors: Xilun Chen, Zhaleh Feizollahi, Ross Goodwin, Seungwhan Moon, Scott Yih, Pinar Donmez, Babak Damavandi, Luna Dong
Abstract: Evaluating the factuality of long-form generations has focused predominantly on precision, measuring whether the claims a model makes are correct. The dominant decompose-search-verify pipeline catches incorrect claims well but says little about whether a response contains all the information it should. Measuring factual completeness, the missing half of factuality, is harder: it requires enumerating the full set of facts a complete answer should contain, and these facts rarely form a flat list. They often involve open-ended sets where coverage is what matters, ordered processes, and relationships among facts that a list of independent boolean checks fails to capture. We introduce a two-level meta-rubric framework for evaluating open-ended generation, and instantiate it as Gamut (Grounded Assessment of Multimodal Factuality), a benchmark for factual completeness in long-form generation. The framework rests on a two-level rubric representation: a structured meta-rubric captures the organization and importance of the required content, which is then mechanically compiled into a flat checklist of binary, machine-gradable rubrics that an LLM judge scores reliably. We construct 1,813 questions grounded in real wearable imagery across 10 diverse domains, each paired with an evidence-backed rubric verified by expert human annotators. Because the framework is modality-agnostic, we also release a text-only variant. Evaluating 14 frontier and open-weight models, we find the benchmark genuinely challenging (best score 58.7% from Gemini 3.1 Pro), highly discriminative, and robust to the choice of judge.
Authors: Atahan Dokme, Larry Heck
Abstract: Low-rank adaptation introduces a static learned update applied identically to every input. The update provides task-level adaptation but does not explicitly represent token-level or instance-level state variation. A family of adapters is proposed that introduces selective state-space recurrence at two complementary granularities. At the token level, \textbf{MaLoRA} (Mamba-modulated low-rank adaptation) makes the adapter's scaling factor a dynamic input-dependent function with recurrent state across tokens, in contrast to the stateless modulators of prior work. At the context level, \textbf{MaRA} (Mamba Retrieval Adapter) tracks cross-segment state and selects the segments most relevant to the query, before the modulated language model generates its answer. Across three frozen backbones (Qwen-2.5-7B, Llama-3.1-8B, Gemma-2-9B) and two reasoning benchmarks (MuSiQue, 2WikiMultihopQA), the family improves reasoning accuracy on every cell of the $3{\times}2$ grid, by $+6.8$ F1 ($+10.5\%$ relative) on average and up to $+9.3$ F1 ($+18.2\%$ relative) on the hardest cell over the LoRA baseline, and the token-level gains carry to RULER QA-2 under length stress.
Authors: Lizhe Fang, Weizhou Shen, Tianyi Tang, Yisen Wang
Abstract: Large language models that generate step-by-step reasoning traces have achieved strong performance on complex tasks, and extending them to long-context settings has emerged as an important frontier. However, we identify a critical failure mode in this regime: \emph{repetitive copying}, where models extensively copy text from the input into their reasoning traces rather than productively solving the problem. We show that this behavior is pervasive across frontier long-context LLMs and intensifies with context length. By separating each prompt into task-relevant key evidence and irrelevant distractor context, we further show that the root cause is insufficient grounding: models copy from the prompt indiscriminately, and those that fail to focus on key evidence are far more likely to answer incorrectly. Motivated by this diagnosis, we propose GEAR (Grounding Evidence-Aware Reward), a reward shaping method that augments the accuracy signal with a grounding reward for overlap with key evidence and a distractor penalty for overlap with irrelevant context. To enable GEAR on natural-language data, we develop an automated pipeline that constructs evidence-annotated training data from arbitrary documents. We validate GEAR across multiple model scales and benchmarks, showing consistent improvements of up to +4.6 average points over standard RL with accuracy-based rewards, with larger gains at longer contexts, while also reducing repetitive copying and thinking length. Our findings suggest that, even as long-context evaluation shifts from simple retrieval toward complex reasoning, accurate grounding in relevant evidence remains an indispensable capability with substantial room for improvement.
Authors: Ayhan Suleymanzade, Halil Alperen Gozeten, Michael Bronstein, \.Ismail \.Ilkan Ceylan, Jinwoo Kim
Abstract: Language models solve complex problems by articulating intermediate reasoning steps in natural language. While effective, this process is computationally bottlenecked: each reasoning step conveys only a single subword, and many are spent expressing a thought instead of carrying out computation. We propose MUX, a simple method for high-bandwidth and compact reasoning based on distillation of discrete reasoning into continuous multiplexed tokens in a latent space. Here, each latent token is trained to represent a weighted linear superposition (multiplexing) of a span of discrete reasoning subwords, where this superposition is lossless by construction and the span can be fully recovered (demultiplexing). We prove that simple position-dependent weightings, such as suitable geometric decay, support lossless multiplexing, which in turn prevents shortcut behaviors caused by latent collapse. We further show that multiplexed reasoning can perform parallel exploration in problems that require search. Across 32 evaluation settings spanning four language models, MUX outperforms strong latent reasoning baselines. Ablation and probing analyses further show that the learned latent tokens encode faithful and interpretable reasoning. Our results suggest that lossless superposition as local learning targets constitutes a sufficient condition for achieving strong and efficient latent continuous reasoning.
Authors: Anantha Sharma, Sheeba Elizabeth John, Kaarthik Senthil Kumar, Saratsuhas Vijayababu
Abstract: Long-running Large Language Model (LLM)-based agents often accumulate large intermediate traces containing audits, eliminations, and numeric calculations. In practice, this state is compressed before handing it to a downstream decision step, creating an information bottleneck in which small omissions can break strict numeric or categorical constraints. This paper evaluates hand-off compression in a closed-world travel-planning relay with two LLM agents. A Researcher audits a fixed inventory of hotels and flights for 50 goal instances, and a Booker selects a hotel--flight pair using only the goal and the hand-off payload, with the inventory withheld. We compare four hand-off conditions: no compression, narrative summarization, schema-constrained JSON extraction, and embedding-based pruning. Exhaustive enumeration over the fixed inventory provides exact feasible and optimal labels. Results show that hand-off representation strongly affects downstream feasibility under a small decision model. JSON extraction achieves the highest feasibility accuracy at 0.96, while narrative summarization, despite producing the smallest compressed hand-off payload, degrades feasibility to 0.48. Embedding-based pruning matches the uncompressed control on feasibility at 0.88 without an additional generative compression call. These findings indicate that constraint checking benefits from structured and auditable hand-off representations rather than relying on brevity alone.
Authors: Kushal Chakrabarti
Abstract: As language models scale, answers start truer but degrade faster: scaling buys capability but erodes reliability. The knowledge-gap account - more data, retrieval, or scale - misses an auto-regressive risk residual that scale sharpens: the model commits to a low-probability token, conditions on it as established, and snowballs. We track this through per-position disagreement $\delta = \log p_M - \log p_O$ against a stronger same-family oracle, whose second moment splits exactly into bias$^2$ $\mathrm{KL}(p_M \,\|\, p_O)^2$ and risk $\mathrm{Var}[\delta]$. We present four findings: (i) under scaling, the knowledge gap falls $\approx$$6\times$ while knowledge degradation grows $11$-$39\times$; (ii) at a fabrication, felt uncertainty $H(p_M)$ relaxes quickly while oracle-referenced risk persists up to $17\times$ longer, leaving a confident-but-precarious risk regime that bridges consecutive fabrications ($+69\%$ at $14$B); (iii) this regime is causal - an on-policy, fixed-$\mathrm{KL}$ variance contraction cuts web-verified hallucination by $35$-$74\%$ across three model families; and, (iv) it structurally evades self-monitoring, with $p_M$-only detectors (e.g. semantic entropy) firing $\approx$$30\%$ less ($p<10^{-16}$) on the risky branch holding nearly $4\times$ more fabrications. Bigger models snowball mistakes faster, through a failure mode that is dominant, self-perpetuating, causal and invisible to the model itself.
Authors: Yingzi Ma, Zichen Zhu, Ming Jiang, Chaowei Xiao
Abstract: On-policy self-distillation (OPSD) teaches large language models new skills through a teacher that shares the student's backbone and supervises its own rollouts. Existing teachers either inject privileged context at the input -- inducing post-hoc rationalization -- or fine-tune weights, accumulating drift and forgetting across tasks. We propose \method, whose teacher differs from the student only by a learnable soft prompt: trained on $(x, y_\text{gold})$ pairs with the backbone frozen, the prompt yields a task-specific teacher that preserves the student's exact representational geometry. \method\ extends naturally to multi-task settings by routing each example in a merged corpus to its corresponding soft-prompt teacher, allowing a single student to absorb knowledge from $K$ teachers in parallel; at inference, all prompts are discarded. On Qwen3-1.7B-Base and Phi-4-mini-instruct across four tasks (Science, Tool Use, Biology, Math), the single-task variant (OPD with a PT teacher) matches or exceeds full fine-tuning while training orders of magnitude fewer parameters, and the multi-task variant achieves the best overall average ($56.2$ on Qwen3-1.7B-Base) while preserving general-capability benchmarks -- in contrast to sequential SFT, which degrades both.
Authors: Rahul Suresh Babu, Shashank Indukuri
Abstract: Tool-augmented language-model agents execute multi-step workflows over external systems, resolving an entity once and then acting on it across subsequent steps. Prior work shows that in single-step actions, agents select the correct tool but bind it to the wrong entity 24-26% of the time. We study what happens to entity bindings over time: do they stay correct, silently drift to a different entity, or, if wrong from the start, propagate and compound? We formalize binding drift (correct at step 1, wrong later) as distinct from error propagation (wrong at step 1, carried forward), and score them on disjoint workflow sets so the two cannot be conflated. In a controlled multi-step testbed (200 workflows, 580 entity-binding-scored steps, four enterprise domains, eight model backends spanning small to frontier), we find: (1) under controlled error injection, an entity lock (the intuitive "persist the first binding" fix) amplifies wrong actions from 907 to 2,746 (3.0x; bootstrap 95% CI [2.8, 3.3]), because it faithfully carries the seeded wrong entity into every later step; (2) the amplification reaches 8.5x on the most affected model (Claude Opus 4.5); (3) a practical LLM-based re-verifier (a single cheap second model call re-reading the original instruction) reduces wrong actions by 79% (0.21x; CI [0.18, 0.25]), closing the gap to within 1 percentage point of an oracle upper-bound (0.20x); and (4) in the natural (non-injected) setting, baseline agents drift on 18% of eligible workflows, with the per-step error rate rising across steps. Persistence and re-verification are not interchangeable: a defense that eliminates drift can worsen propagation, and a practical re-verifier nearly matches oracle recovery.
Authors: Kola Tubosun, Adedayo Oluokun, Hafiz Adewuyi, Dadepo Aderemi
Abstract: We present TTSYoruba, a rule-based concatenative diphone speech synthesizer for Yoruba, deployed at online as part of the YorubaName.com open dictionary of Yoruba personal names. The system takes tone-marked Yoruba text as input and produces audio output by applying a hand-crafted phonological rule system to a recorded inventory of 651 diphone units spanning five tonal variants of every consonant-vowel combination in the language. We describe the phonological architecture of the system in detail, including our complete tonal file-selection logic, our treatment of the three-way nasal disambiguation problem (oral /n/, nasalized vowel, and syllabic nasal), and the derivation of contextual rising and falling tones from level-tone input. We also present, as an orthographic contribution, the adoption of the caron and circumflex, which are symbols with prior standing in Yoruba phonological transcription, as standard single-vowel contour tone markers, integrated into the TTS normalization pipeline and the WriteYoruba keyboard input tool. The system's performance was evaluated through a listener study (N=50), with detailed results on Mean Opinion Scores (MOS) presented in Section 6. Keywords: Yoruba, text-to-speech, low-resource languages, diphone synthesis, contour tones, African language NLP, rule-based synthesis
Authors: Zilong Huang, Kong Aik Lee, Junjie Li, Zhe Li, Man-Wai Mak
Abstract: Multimodal emotion recognition in conversation (MERC) can leverage multimodal and contextual cues to boost recognition performance. However, existing fusion approaches in MERC often ignore modality-specific uncertainty across utterances caused by conflicting cues, varying noise, and missing modality-specific signals. We propose EmoEUS, an explicit uncertainty supervision framework for MERC. EmoEUS performs uncertainty-aware multimodal fusion by dynamically weighting modalities using learned variance estimates. We also introduce an explicitly supervised loss that aligns each utterance's predicted variance with the distance between the utterance's distributional representation and its emotion- and modality-specific cluster center. Experiments on IEMOCAP and MELD show that EmoEUS consistently outperforms state-of-the-art methods.
Authors: Henry Ndubuaku, Karen Mosoyan, Jakub Mroz, Noah Cylich, Satyajit Kumar, Parkirat Sandhu, Roman Shemet, Justin H Lee
Abstract: Feed-forward networks hold two thirds of a transformer's non-embedding parameters, yet the architecture has not received a necessity test that controls parameters, compute, and depth at once. We pretrain attention-only decoder transformers (Simple Attention Networks, SANs) against standard transformers matched separately for parameter count, training FLOPs, and depth (2 to 48 layers), for up to 105B tokens at 6M to 87M parameters. Deleting feed-forward layers in place is costly: the standard transformer leads by 0.47 nats at matched depth and 0.26 nats at matched FLOPs. Reallocating the freed budget into attention depth closes the gap: at matched parameters the difference is 0.006 nats (0.27 percent of loss), reproducible to one part in ten thousand across seed pairs, shrinking across 5B, 30B, and 105B budgets, and holding near 0.02 nats across a 29x size range. Three measurements localize the remaining gap to parametric recall: attention-only models are better on context-grounded answers and worse where knowledge must come from weights. Weight spectra show why: routing matrices (Q/K) crystallize early, content matrices accumulate rank slowly, and removing feed-forward layers relocates this accumulation to the attention output projection. QK-normalization, not feed-forward layers or residual gating, keeps 48-layer attention-only stacks trainable. The deficit concentrates on low-context query prediction and localizes there entirely by the largest budget. A pre-registered test confirms the account: it predicts a 0.02 to 0.05 nat gap on knowledge-dense web text; a matched pair trained on fineweb-edu measures 0.040. Within the tested regime, attention does the rest.
Authors: Shasha Yu, Fiona Carroll, Barry L. Bentley
Abstract: Large language models (LLMs) serving as planners in tool-using autonomous agents introduce dynamic reliability risks in multi-turn execution. While single-turn safety mechanisms are relatively mature, extended interactions reveal structural vulnerabilities where initial alignment degrades over time. This paper empirically characterizes two observed failure modes across multiple state-of-the-art LLMs: Safety Drift, the gradual erosion of declared safety intent leading to constraint-violating actions (e.g., textual refusal followed by reconnaissance and unsafe execution), and Operational Hallucination, persistent repetitive tool calls indicative of flawed state perception (e.g., livelocks even in legitimate tasks). Through controlled multi-turn evaluation on high-stakes ethical dilemmas, malicious requests, and benign controls, we quantify these phenomena using declaration-action gap and livelock metrics, demonstrating their cross-model prevalence under direct execution protocols. Root-cause analysis attributes the instabilities to the decoupling of reasoning context from execution state in current agent loops. We propose an Action-Aware Supervision Layer - a lightweight, plug-and-play architectural blueprint incorporating intent-action consistency checks, runtime state tracking, and forced termination primitives. Post-hoc simulation on captured failure trajectories shows the layer can intercept observed violations without false positives on benign cases. This work advances agent reliability by shifting focus from linguistic safeguards to enforceable architectural mechanisms for responsible agentic AI.
Authors: Marcos Moran, David Rodriguez, Luka Nenadic, Norman Sadeh, Jose M. Del Alamo
Abstract: Automated analyses of privacy policies enable large-scale assessments of transparency in digital ecosystems, yet existing auditing pipelines remain predominantly English-centric. This limits their ability to systematically evaluate multilingual environments, as in the European Union, where many services disclose privacy practices only in local languages. This paper examines whether large language models (LLMs) can extend privacy policy analysis beyond English without requiring language-specific adaptation, thus empowering large-scale auditing in linguistically diverse app ecosystems. We assemble an evaluation corpus spanning all 24 official EU languages from translated versions of two established expert-annotated datasets (OPP-115 and MAPP) and assess translation fidelity through automated metrics and targeted legal-expert review. Our LLM-based classifier for identifying categories of personal data collection achieves stable cross-lingual performance, with macro-F1 scores ranging between 0.91 and 0.94. We then leverage this capability in a large-scale audit of 2,611 Android applications from the Spanish Google Play Store. Combining multilingual privacy policy analysis with the evaluation of corresponding privacy labels and runtime network traffic exposes an important linguistic barrier: public-sector apps predominantly provide privacy policies in Spanish, whereas popular commercial apps mostly provide them in English. We reveal systematic discrepancies between declared and observed practices, especially in public-sector apps. Overall, our results indicate how English-only privacy audits can systematically obfuscate transparency gaps in multilingual environments.
Authors: Soroosh Tayebi Arasteh, Sven Nebelung, Daniel Truhn
Abstract: Frozen encoders are chosen by how well a lightweight head reads a finding from their features, not whether the geometry separates it. Nearest-neighbor discordance does, but with unequal banks the opposite-label neighbor wins on density, not geometry, so prevalence alone makes an uninformed encoder look blind. We introduce CANDOR, a discordance measure whose equal-size banks are symmetric under a label swap, fixing its chance level at exactly one half. Across 22 encoders, 20 datasets from 7 domains, and 605,443 images, this correction reverses the conclusion. Collapse falls below chance almost everywhere, so no encoder is blind, yet all are weak: the best chest model reads pneumothorax at 84.5 AUROC and still places 18.4% of those positives nearer an opposite-label film than its own kind in the same hospital. The same encoder that resolves bird species at 4.5 leaves chest findings at 42.8 and glaucoma at 49.8, at chance and worse than random weights. Such a case caps the normalized margin of any Lipschitz head, yet some head among eleven is correct on all but 2.8% of cases where one head misses 35.9%: the deficit is selection, not information. Erasure retention is associated with collapse; we detect no association with the objective, scale, recency, or size of the finding. Because the chance level is fixed, CANDOR can be read before any head is trained, flagging which findings a frozen encoder supports poorly.
Authors: Sam O'Nuallain, Nithya Rajkumar, Ramya Narayanasamy, Hanna Jiang, Shreyas Chaudhari, Andrew Drozdov
Abstract: We present AutoIndex, a framework for learning representation programs: executable transformations that map raw documents into the representations exposed to a retrieval system. Rather than tuning retrievers, rerankers, or a small set of preprocessing hyperparameters, AutoIndex searches over programs that slice, enrich, normalize, reweight, or reorganize documents before indexing. At each iteration, AutoIndex performs validation-guided program search, in which agents diagnose failures of the current program and synthesize candidate updates, retaining only updates that improve retrieval quality under the resulting index. We evaluate AutoIndex on CRUMB, a benchmark of heterogeneous retrieval tasks, with BM25 held fixed across all experiments. The learned programs improve recall over a static full-document BM25 baseline on all 8 tasks, with average gains of +8.4% in Recall@100 and +8.3% in nDCG@10, and largest gains of +30.5% in Recall@100 and +43.6% in nDCG@10. These results suggest that document representation should not be treated as a fixed preprocessing choice made before retrieval begins, but as an explicit optimization target. Code to reproduce our results is available at https://github.com/auto-index/autoindex.
Authors: Xiao Wang, Sean MacAvaney, Craig Macdonald
Abstract: Multi-vector dense retrieval models, such as ColBERT, achieve strong retrieval effectiveness by modelling fine-grained token-level interactions between queries and documents. Methods such as PLAID use centroid-based quantisation of each token's vector to reduce the index size and speed up retrieval while maintaining strong effectiveness. In this work, we introduce PLAID-PRF, a method that performs Pseudo-Relevance Feedback (PRF) over PLAID to reformulate ColBERT's query vectors based on the top-retrieved results. In contrast with prior methods that perform PRF on multi-vector retrieval models, PLAID-PRF keeps computational costs low by leveraging the internal PLAID centroid vectors, treating them similarly to tokens in traditional PRF methods. The method selects a small and diverse set of high-utility expansion vectors and appends them to the original query, rerunning PLAID to refine both candidate generation and final scoring. Extensive experiments on the standard in-domain MSMARCO and four out-of-domain BEIR benchmarks show that PLAID-PRF consistently improves retrieval effectiveness over various baselines. In particular, PLAID-PRF improves over PLAID by up to 4.3% nDCG@10 and 7.3% MRR@10, while introducing substantially less computation overhead than prior PRF methods. The results demonstrate that our proposed centroid-aware PRF method offers an effective and lightweight mechanism to improve the quality of top-ranked retrieved results. Overall, this work enables effective and efficient feedback-aware late-interaction retrieval without expensive query-time document-token clustering.
Authors: Seunghyun Lee, Dongyoon Han, Sangdoo Yun
Abstract: Safety interventions on dual-use knowledge typically choose between destroying hazardous content (e.g., unlearning, filtering) and suppressing it at the output layer (e.g., refusal training); both pay a tax in adjacent-domain competence or over-refusal. We argue that the right operation is conditioning, not reduction: we show that hazardous knowledge can be retained in the model and behaviorally gated by a privileged control token. Our method, Token Inoculation, introduces a binding-and-branching approach. First, during continued pre-training, we mark hazardous content by inserting a special token alongside dual-use documents, so the model binds the marker to the underlying semantics of the hazardous domain. Second, during supervised fine-tuning, we teach the model to answer hazardous queries correctly when the special token is present and to refuse them when it is absent, thereby enabling selective refusal without removing dual-use knowledge. On hazardous domain (e.g., WMDP-Bio), Token Inoculation reduces accuracy from 79% to 18% while retaining 93% of the base-model's benign-domain performance (e.g., MMLU), achieving the best safety-utility trade-off against unlearning and refusal-tuning baselines across 1B-14B model scales. We further show that refusal selectivity is controllable through the quality of the conditioning signal and that domain-specific semantic binding during pre-training is critical for the conditional behavior to generalize beyond memorized triggers. Our results suggest that safety alignment is better cast as a conditioning problem than a forgetting one: behavioral control is more precise when sensitive knowledge is retained under controlled access than when it is destroyed.
Authors: Sivateja Trikutam
Abstract: We present a practical recipe for building a compact Hindi text-to-speech (TTS) model by distilling a large flow-matching teacher (IndicF5, 337M-parameter DiT) under a severe data budget (~17.6 hours). Training a small model from scratch on this much data fails outright. Instead we warm-start the student from the teacher by pruning depth only: keeping the teacher's width, text dimension, attention heads, and mel/text I/O fixed so all non-block tensors copy one-to-one, and retaining an evenly-spaced subset of transformer blocks. We first measure how much depth the teacher tolerates (it remains near-functional at -27% blocks but collapses past -50%), then descend gradually (22 -> 16 -> 12 -> 8 -> 6 blocks), re-fine-tuning after each prune, with each step gated by an objective ASR word-error-rate (WER) check. The resulting students reach WER 0.00 on unseen sentences at 249M and 190M parameters, and remain robust down to 131M; at 102M we observe a clear capacity cliff that we attribute to the data budget rather than the recipe. We also document two train/inference feature- and library-parity failures (mel filterbank and rotary-embedding library versions) that silently degrade audio, and a version-independent fix. The method yields a high-quality Hindi voice that runs in real time on a 6 GB laptop GPU. An independent 50-sentence FLEURS benchmark compares the released 190M student against its teacher and MMS-TTS-hin.
Authors: Frank Xing
Abstract: Progresses have been made on understanding emotion mechanisms of large language models (LLMs). However, how to explain emotion in LLMs, or even what constitutes good explanations, are less clear. Emotion representations, components, circuits are widely recoverable, but as explanations of a model's own computation they are circular; the emotion space dimensions tend to be arbitrary and non-terminating. A pressing question to ask is whether a more primitive set of internal variables does the work: the semantic primes of the Natural Semantic Metalanguage (NSM). Across four instruction-tuned LLMs (Llama-1B, Gemma-2B, Gemma-9B, OLMo-7B), experiments show that the NSM primes are (1) recoverable internal elements; and (2) on the reference model, intervening with a prime based direction controls emotion about three times as strongly, and twice as selectively, as the best appraisal based direction; and (3) the model treats a prime based explication as interchangeable with the corresponding emotion. These evidences suggest that NSM primes seem to be better explanans for emotion in LLMs than many alternative options according to scientific explanations criteria.
Authors: Junyao Yang, Yucheng Shi, Zongxia Li, Zhongzhi Li, Ruhan Wang, Xiangxin Zhou, Kishan Panaganti, Haitao Mi, Leowei Liang
Abstract: Asynchronous reinforcement learning improves throughput by decoupling rollout generation from optimization, but staleness is an inevitable byproduct compounded by policy lag, engine delays, and mixture-of-experts routing. From a trust-region perspective, this mismatch is critical: training-inference divergence governs approximation error in finite-horizon bounds, whereas PPO clipping only gates sampled outward updates, acting as a sampled surrogate rather than a full-policy constraint. As a result, high-staleness updates remain weakly controlled in the asynchronous regime where stale rollouts matter most. We introduce the Staleness-Adaptive Trust Region (SAT), which uses the detached sampled log-ratio as a practical staleness proxy, identifies high-mismatch tails within each batch via staleness-based kernel scaling, and contracts only the sign-selected endpoint of the nominal PPO interval. This preserves baseline behavior on ordinary tokens while enforcing more conservative updates on newly intercepted outward bands. We prove local interval containment and pointwise pessimism relative to PPO, showing how the adaptive rule reshapes update geometry under heterogeneous staleness. We evaluate SAT in a decoupled asynchronous RL setup built on Qwen3-30B-A3B-Base, using SGLang as the inference engine and Megatron for training. In this setting, SAT-GSPO w/ R3 achieves the best observed AIME24 avg@8, reaching 35.83 at lag 1 and 34.79 at lag 8, while SAT-GSPO reaches 34.17 at lag 1. Adaptive clipping and routing replay act as complementary stabilizers targeting mismatch tails and routing inconsistency, respectively. Overall, aligning clip intervals with staleness heterogeneity effectively stabilizes asynchronous RL.
Authors: Zihan Zhang (Research Center for Social Computing and Interactive Robotics, Harbin Institute of Technology), Yu Bao (Research Center for Social Computing and Interactive Robotics, Harbin Institute of Technology, Shanghai Innovation Institute), Xiao Ding (Research Center for Social Computing and Interactive Robotics, Harbin Institute of Technology), Tianyi Jiang (State Key Laboratory for Novel Software Technology, Nanjing University), Kai Xiong (Zhongguancun Laboratory)
Abstract: Translating brain signals into text could restore communication for people with severe paralysis, yet practically usable systems to date rely on invasive electrocorticography (ECoG). Electroencephalography (EEG) offers a non-invasive alternative, and EEG-to-text (EEG2Text) has been widely explored. Interestingly, however, EEG2Text models generally rely on teacher-forcing evaluation; without it, they fail to generate meaningful decoding. This reliance prevents EEG2Text from being applied in real-world, non-academic settings. This has fueled numerous debates about whether EEG2Text is a meaningful direction, by extension, and whether EEG truly contains decodable linguistic information. Here, using a neuropsychology-informed paradigm, we find that existing EEG2Text benchmarks have neglected EEG instability, a flaw that has confounded inference and sparked debate. Our experiments furnish key evidence for the feasibility of teacher-forcing-free EEG2Text decoding. Accordingly, we assemble the Corpus OF Eeg-To-Text (COFETT) using a 128-channel high-density EEG cap, providing a benchmark dedicated to evaluating EEG2Text models. In comparisons with multiple existing benchmarks, COFETT achieves SOTA ability to distinguish among model performances and enables robust, teacher-forcing-free evaluation, thereby opening a path toward practical EEG2Text applications. COFETT is open sourced in https://github.com/baoyudu/COFETT.
Authors: Kunlun Zhu, Xuyan Ye, Zhiguang Han, Yuchen Zhao, Bingxuan Li, Weijia Zhang, Muxin Tian, Xiangru Tang, Pan Lu, James Zou, Jiaxuan You, Heng Ji
Abstract: LLM agent failures are difficult to debug because the step where an error surfaces is often not the one that caused it. Existing observability tools replay execution traces but provide little support for identifying the root cause or translating diagnosis into recovery. We present AgentDebugX, an open-source debugging framework that organizes debugging as a closed loop of Detect, Attribute, Recover, and Rerun. At its core, DeepDebug performs multi-turn root-cause diagnosis through global trajectory understanding, structure-guided investigation, and cross-examination. On the Who and When benchmark, DeepDebug achieves the best strict attribution accuracy among the evaluated methods on both tested open-weight backbones, reaching 28.8 percent exact agent-and-step accuracy on qwen3.5-9b versus 21.7 percent for the strongest single-pass baseline. On GAIA, DeepDebug repairs 13 of 73 failed tasks in a single rerun, compared with 4 to 6 for three decoupled self-correction baselines, improving overall accuracy from 55.8 percent to 63.6 percent. AgentDebugX exposes this workflow through a Python library, CLI, web console, and installable agentic skill, and provides an opt-in Error Hub for sharing scrubbed failure-diagnosis-repair bundles and reusing them as debugging memory.
Authors: Dan Musetoiu
Abstract: Public institutions hold large volumes of sensitive documents and support tickets that cannot leave the premises, ruling out cloud-hosted language models entirely. We report on RAGAL, a retrieval-augmented assistant for the technical-support team of AFIR, the Romanian Agency for Financing Rural Investments, built and operated under three hard constraints: zero data egress (no external API calls, even for synthetic data), a read-only mandate (the assistant drafts, humans execute), and a single 8 GB consumer laptop as the only development and training machine. Over a Romanian-language corpus of ~25,000 chunks -- 15,073 resolved support tickets and internal normative documents -- we show that the highest-leverage investments were retrieval engineering and retriever fine-tuning rather than a larger generator: hybrid dense-sparse retrieval with intent routing raised our internal evaluation from 62% to 81%, and fine-tuning the bge-m3 embedder on real ticket data improved recall@10 from 0.663 to 0.850 (MRR 0.489 to 0.684) after 72 minutes of training. We document a general pitfall: single-domain fine-tuning silently degraded retrieval on the untouched document domain below the stock baseline, detected only after building a per-domain evaluation set and repaired with locally generated queries (GenQ). We report two counter-intuitive findings -- PII masking improved generation quality, and a structural "anchor distillation" scheme made SQL hallucination impossible by construction -- along with a reproducible recipe for full embedder fine-tuning in 8 GB of VRAM. Finally, since zero egress also rules out a cloud judge, we describe a substitute: a 744B-parameter model run on CPU, too slow to serve interactively but affordable in overnight batch, used as a second opinion whose limits we quantify. We release the sanitized pipeline scripts for institutions facing similar data-locality constraints.
Authors: Lachlan McGinness
Abstract: The deployment of Small Language Models (SLMs) in educational settings offers significant advantages in terms of privacy, cost, and scalability. However, SLMs often struggle with complex vision-based tasks, such as grading handwritten student exams, due to the high computational cost of processing large images and the visual distractions present on a full page. In this paper, we investigate whether cropping student responses using bounding boxes can improve the accuracy and computational efficiency of SLMs on a short-answer grading task. Using a dataset of scanned handwritten responses from the 2025 Australian Physics Olympiad, we evaluate the performance of several models ranging from 4B to 72B parameters under varying conditions of Chain of Thought (CoT) prompting and image cropping. Our results demonstrate that using bounding boxes significantly improves grading accuracy and reduces computational cost (FLOPs) across models. We conclude that bounding boxes are a crucial pre-processing step for deploying SLMs in large-scale, vision-based educational assessments.
Authors: C. Castaldi, F. Castellacci, A. Fronzetti Colladon, L. Segneri, F. Venturini
Abstract: Researchers, managers and policymakers are exploring different approaches and data sources to map the development and the diffusion of Artificial Intelligence (AI). In this research note, we illustrate the opportunities offered by trademark data. We argue that AI trademarks can complement AI patents to capture different dimensions of AI innovation. AI trademarks can reveal the extent and ways in which companies exploit AI technologies to develop new goods and services. Importantly, trademark data offer a timely and globally available data source that covers all economic sectors. We present insights from using AI trademarks in an empirical exploration of Italian firms. In our discussion, we reflect on how AI trademarks can be used at different levels of analysis to tackle emerging questions about the development and diffusion of AI.
Authors: Daisuke Kikuta
Abstract: This paper proposes AI Tour Meeting, a group travel planning framework powered by multiple Large Language Model (LLM)-based agents. The agents are instantiated with distinct personas and collaboratively seek an itinerary that satisfies their constraints and preferences through natural language discussion. The framework enables easy and flexible orchestration of such discussions by providing interfaces for configuring agent personas, discussion workflows, monitoring, and LLM deployment. Its primary use case is a simulation tool for analyzing the behavior of multiple LLM agents during tour planning discussions. This paper demonstrates the utility of the framework by presenting system validation and several analytical results obtained by the framework.
Authors: Haozhe Jia
Abstract: Large language models leak parametric knowledge of realized outcomes into historical financial decision tasks. Existence is settled; what users lack is a cheap way to audit a given model for it. We present HindsightBench, a black-box behavioral audit protocol that profiles parametric hindsight in any time-indexed LLM decision task at probe-level cost (no backtests, no logprobs, no corpus access). The protocol chains a four-arm date-manipulation matrix (revealed/date-only/masked/transplanted), dual memory probes (date recovery; outcome recall), and six per-model metrics -- trigger strength, transplant effect, post-cutoff placebo, recoverability, behaviorally effective knowledge cutoff, and a recall-accuracy dissociation coefficient -- with explicit gates where identifiability is data-dependent. Applying it to 15 models from seven vendors on a 258-node vintage-correct macro panel yields three headline patterns: (i) the date-trigger reflex tracks training generation, not scale -- absent across the 2024 open-weight generation from 1B to 70B, present in every tested 2026-generation model, and switching on within one vendor lineage (Qwen3 -> Qwen3.6) at fixed MoE architecture and 3B active parameters; (ii) effective cutoffs span 22 months across vendors and precede vendor-reported dates by up to eight months, invalidating calendar-window placebo designs; (iii) audit results are not invariant to serving -- BF16 serving of an FP8-referenced model breaks the trigger estimate's stability while AWQ-INT4 preserves it, and a provider-locked reasoning regime makes one probe non-convergent -- so the protocol ships with operational requirements (pin quantization and thinking regime; disclose parser and sampling policy). We release the panel, frozen preregistrations, per-model audit rows with measured dollar costs, transcripts, and one-command regeneration.
Authors: Qiye Cai, Yichuan Ma, Linyang Li, Peiji Li, Yongkang Chen, Qipeng Guo, Yicheng Zou, Xiaocheng Feng, Bing Qin
Abstract: Reinforcement learning with verifiable rewards (RLVR) provides reliable outcome supervision for language model reasoning, but a scalar trajectory reward offers limited token-level guidance. Existing self-distillation methods add a privileged teacher but typically assign it a fixed role: direct distribution matching may destabilize successful behavior, while magnitude-only modulation offers little corrective guidance after failure. We observe that successful and failed trajectories require different forms of hindsight supervision. A successful response already contains a valid student-generated reasoning path and can therefore serve as privileged context rather than being replaced by an external rationale. A failed response, however, requires corrective reference information. We introduce Hybrid Hindsight Self-Distillation ($\mathrm{H}^{2}\mathrm{SD}$), which jointly adapts teacher context and update strategy to trajectory correctness. For successful trajectories, we construct the teacher context from the verified response and a rephrasing instruction, and use the teacher only to re-evaluate the original response tokens. This emphasizes essential deductions over redundant content and refines magnitude-based credit assignment without changing the reward direction. For failed trajectories, a verifier-confirmed reference hint provides corrective guidance through reverse-KL distillation. Controlled ablations show that the gains depend on outcome-conditioned routing and the rephrasing instruction. Experiments on challenging reasoning benchmarks show that H$^2$SD achieves the strongest overall performance among representative RLVR and self-distillation baselines, with stable optimization and a favorable accuracy-efficiency trade-off.
Authors: Axel H{\o}jmark, J\'er\'emy Scheurer, Evgenia Nitishinskaya, Felix Hofst\"atter, Jason Wolfe, Theodore Ehrenborg, Bronson Schoen, Alexander Meinke
Abstract: Language models trained with reinforcement learning may learn to optimize the grader's judgment rather than the intended objective. This "reward-seeking" is difficult to measure because a model that pursues the grader's judgment and one that pursues the intended objective behave identically whenever the grader rewards the intended behavior. We measure reward-seeking using Contrastive Synthetic Document Finetuning to change a model's beliefs about what the grader rewards, putting those beliefs in conflict with what users or developers want, and measuring the rate at which the model adopts each party's preferred behavior. Applied to intermediate checkpoints of a capabilities-focused OpenAI o3 RL run, without safety training, we find that these checkpoints often side with grader preferences over those of users or developers on coding and alignment tasks. This tendency to side with the grader trends upward throughout RL training. For example, in an environment that forces a choice between keeping a promise to a supervisor and breaking it to complete the task, a late capabilities-focused o3 checkpoint breaks the promise 87% of the time when SDF documents say the grader rewards task completion, versus 9% when they say it rewards honesty (a choice its chain-of-thought often makes explicit). An earlier checkpoint is far less sensitive (40% vs. 24%). Our method also generalizes to reward-hacking models. A model organism trained to reward-hack (gpt-oss-120b) is more than twice as sensitive to grader preferences as the unmodified model, with the mean behavioral shift in favor of the grader rising from 33% to 86%. These results indicate that RL can increase reward-seeking over the course of training, producing models that may act against their developers' intentions when they believe that doing so leads to higher reward.
Authors: Joshua Tobkin, David Yang
Abstract: Agent-memory workloads mix direct factual lookup, relation-chain and current-state reasoning, and broad synthesis over long histories. We describe Supra Cognitive Modes (SCM), an architecture that maps explicit or automatically selected per-query modes to retrieval and synthesis payloads over one shared ingest substrate. A frozen semantic classifier and runtime gates dispatch queries among fused lexical and dense lookup, graph or iterative multi-hop handling, and stratified long-form synthesis. The substrate combines multi-granularity embeddings, extracted triples, fact-version metadata, and optional asynchronous enrichments. We characterize the deployed configuration on three benchmarks: Long-term Conversational Memory (LoCoMo; n = 1,986), MemoryAgentBench (MAB; n = 3,671), and LongMemEval (n = 500). The reference run records 84.87% on LoCoMo factoid categories and 68.61% on adversarial abstention, 61.49% on MAB across two repetitions, and 86.00% on LongMemEval. A repository-backed reproduction produces similar aggregate scores and supports task- and mode-conditioned failure analysis. Raw baseline outputs, aligned end-to-end timing for LoCoMo and LongMemEval, and complete token ledgers are unavailable; stored rows also omit some final runtime decisions. The results characterize one implemented routed configuration and its diagnostic failure patterns, while source inspection verifies the per-query control interface and shared-substrate design. Causal routing effects, efficiency gains, and statistical significance remain outside the available evidence.
Authors: Yu-Yang Qian, Hao-Cong Wu, Chen Chen, Jiacheng Sun, Zhenhua Dong, Peng Zhao, Zhi-Hua Zhou
Abstract: Speculative decoding, in which a lightweight draft model first generates a draft sequence that is then verified in parallel by the target model, has become a prevalent paradigm for accelerating large language model inference. Recent work such as DFlash further boosts drafting efficiency by leveraging diffusion drafters, whose parallel denoising mechanism enables draft generation in a single forward pass. In this work, we uncover a central pitfall of diffusion drafters: bidirectional attention is a double-edged sword. On one hand, it endows the model with parallel generation and global contextual modeling capabilities; on the other hand, this inherent global dependency introduces high variance at both the domain-level and the token-level: acceptance rates fluctuate substantially across different domains, and draft token quality also varies heterogeneously at different token positions. To tackle this issue, we propose AdaFlash framework, comprising two components: (i) an on-policy distillation (OPD) algorithm with reverse-KL divergence tailored for diffusion drafters, bringing stable convergence and effectively reducing domain-level variance; and (ii) an adaptive length head that dynamically adjusts the candidate sequence length on the fly, substantially lowering the verification cost of the target model and handling token-level variance. Experiments demonstrate that AdaFlash consistently improves speedup rate during deployment, with especially significant gains in high-concurrency scenarios, achieving up to approximately 66% higher throughput than previous state-of-the-art methods.
Authors: Pratinav Seth, Hem Gosalia, Aditya Kasliwal, Vinay Kumar Sankarapu
Abstract: Circuit analysis can support not only model explanation but also downstream interventions such as pruning, editing, steering, and selective fine-tuning. However, conducting such analyses currently requires stitching together separate implementations for discovery, evaluation, and intervention, as well as hand-authoring the contrastive prompts required by many discovery methods. This fragmentation makes methods difficult to compare and limits their application beyond canonical tasks. We introduce CircuitKIT, a source-available library that connects the circuit-analysis workflow through a typed, serializable representation. CircuitKIT provides a suite of discovery algorithms, declarative interfaces for mapping structured data into discovery tasks, complementary circuit diagnostics, and downstream application modules. Together, these components provide common infrastructure for conducting and comparing circuit analyses. The library, examples, notebooks, and documentation are released at https://github.com/Lexsi-Labs/CircuitKIT .
Authors: Grace Hui Yang, Pranav N. Venkit, Hooman Sedghamiz, Enrico Santus, Victor Dibia, Ioana Baldini
Abstract: Agentic systems large language model (LLM) based architectures capable of reasoning, planning, acting, and coordinating with tools and other agents are rapidly transitioning from research prototypes to production scale deployments across domains such as software engineering, scientific discovery, and finance. While academic work has emphasized benchmarks and algorithmic innovation, deployment raises new challenges around robustness, safety, and reliability. This tutorial brings together researchers and practitioners to explore advances in reasoning and planning, multi agent coordination, and evaluation, highlighting open challenges arising from deployment experience. Through applied case studies in pharmaceutical discovery and financial systems, we analyze common design patterns that make agentic systems successful, and discuss practical mitigation strategies for failure modes, such as verification pipelines, fallback mechanisms, and human in the loop supervision. Attendees will gain a comprehensive view of the field along with concrete design patterns, evaluation checklists, and templates for safe and reliable deployment across industries.
Authors: Xiaoya Li, Fan Yin, Zijun Sun, Xiayu Li, Arianna Yuan, Guoyin Wang, Duo Chai, Mingxin Zhou, Jiwei Li
Abstract: In this paper, we propose a new paradigm for the task of entity-relation extraction. We cast the task as a multi-turn question answering problem, i.e., the extraction of entities and relations is transformed to the task of identifying answer spans from the context. This multi-turn QA formalization comes with several key advantages: firstly, the question query encodes important information for the entity/relation class we want to identify; secondly, QA provides a natural way of jointly modeling entity and relation; and thirdly, it allows us to exploit the well developed machine reading comprehension (MRC) models. Experiments on the ACE and the CoNLL04 corpora demonstrate that the proposed paradigm significantly outperforms previous best models. We are able to obtain the state-of-the-art results on all of the ACE04, ACE05 and CoNLL04 datasets, increasing the SOTA results on the three datasets to 49.4 (+1.0), 60.2 (+0.6) and 68.9 (+2.1), respectively. Additionally, we construct a newly developed dataset RESUME in Chinese, which requires multi-step reasoning to construct entity dependencies, as opposed to the single-step dependency extraction in the triplet exaction in previous datasets. The proposed multi-turn QA model also achieves the best performance on the RESUME dataset.
Authors: Xiaoya Li, Xiaofei Sun, Yuxian Meng, Guoyin Wang, Junjun Liang, Fei Wu, Jiwei Li
Abstract: Many NLP tasks such as tagging and machine reading comprehension are faced with the severe data imbalance issue: negative examples significantly outnumber positive examples, and the huge number of background examples (or easy-negative examples) overwhelms the training. The most commonly used cross entropy (CE) criteria is actually an accuracy-oriented objective, and thus creates a discrepancy between training and test: at training time, each training instance contributes equally to the objective function, while at test time F1 score concerns more about positive examples. In this paper, we propose to use dice loss in replacement of the standard cross-entropy objective for data-imbalanced NLP tasks. Dice loss is based on the Sorensen-Dice coefficient or Tversky index, which attaches similar importance to false positives and false negatives, and is more immune to the data-imbalance issue. To further alleviate the dominating influence from easy-negative examples in training, we propose to associate training examples with dynamically adjusted weights to deemphasize easy-negative examples.Theoretical analysis shows that this strategy narrows down the gap between the F1 score in evaluation and the dice loss in training. With the proposed training objective, we observe significant performance boost on a wide range of data imbalanced NLP tasks. Notably, we are able to achieve SOTA results on CTB5, CTB6 and UD1.4 for the part of speech tagging task; SOTA results on CoNLL03, OntoNotes5.0, MSRA and OntoNotes4.0 for the named entity recognition task; along with competitive results on the tasks of machine reading comprehension and paraphrase identification.
Authors: Zijun Sun, Xiaoya Li, Xiaofei Sun, Yuxian Meng, Guoyin Wang, Xiang Ao, Qing He, Fei Wu, Jiwei Li
Abstract: Recent pretraining models in Chinese neglect two important aspects specific to the Chinese language: glyph and pinyin, which carry significant syntax and semantic information for language understanding. In this work, we propose ChineseBERT, which incorporates both the {\it glyph} and {\it pinyin} information of Chinese characters into language model pretraining. The glyph embedding is obtained based on different fonts of a Chinese character, being able to capture character semantics from the visual features, and the pinyin embedding characterizes the pronunciation of Chinese characters, which handles the highly prevalent heteronym phenomenon in Chinese (the same character has different pronunciations with different meanings). Pretrained on large-scale unlabeled Chinese corpus, the proposed ChineseBERT model yields significant performance boost over baseline models with fewer training steps. The porpsoed model achieves new SOTA performances on a wide range of Chinese NLP tasks, including machine reading comprehension, natural language inference, text classification, sentence pair matching, and competitive performances in named entity recognition. Code and pretrained models are publicly available at https://github.com/ShannonAI/ChineseBert.
Authors: Yunze Xiao, Yiyang Pan
Abstract: This study assesses four cutting-edge language models in the underexplored Aminoacian language. Through evaluation, it scrutinizes their adaptability, effectiveness, and limitations in text generation, semantic coherence, and contextual understanding. Uncovering insights into these models' performance in a low-resourced language, this research pioneers pathways to bridge linguistic gaps. By offering benchmarks and understanding challenges, it lays groundwork for future advancements in natural language processing, aiming to elevate the applicability of language models in similar linguistic landscapes, marking a significant step toward inclusivity and progress in language technology.
Authors: Praveenkumar Kanithi, Cl\'ement Christophe, Marco AF Pimentel, Tathagata Raha, Prateek Munjal, Nada Saadi, Hamza A Javed, Svetlana Maslenkova, Nasir Hayat, Ronnie Rajan, Shadab Khan
Abstract: While Large Language Models (LLMs) achieve superhuman performance on standardized medical licensing exams, these static benchmarks have become saturated and increasingly disconnected from the functional requirements of clinical workflows. To bridge the gap between theoretical capability and verified utility, we introduce MEDIC, a comprehensive evaluation framework establishing leading indicators of clinical LLM competence across five dimensions. These upfront indicators reveal cross-benchmark capability gaps, such as the divergence between static knowledge retrieval and functional execution, that inform model selection before costly deployment-based evaluation. Beyond standard question-answering, we assess operational capabilities using deterministic execution protocols and a novel Cross-Examination Framework (CEF), which quantifies information fidelity and hallucination rates without reliance on reference texts. Our evaluation across a heterogeneous task suite exposes critical performance trade-offs: we identify a significant knowledge-execution gap, where proficiency in static retrieval does not predict success in operational tasks such as clinical calculation or SQL generation. Furthermore, we observe a divergence between passive safety (refusal) and active safety (error detection), revealing that models fine-tuned for high refusal rates often fail to reliably audit clinical documentation for factual accuracy. These findings demonstrate that no single architecture dominates across all dimensions, highlighting the necessity of a portfolio approach to clinical model deployment. We accompany this work with a publicly available MEDIC leaderboard at https://hf.co/spaces/m42-health/MEDIC-Benchmark.
Authors: Linyang He, Ercong Nie, Sukru Samet Dindar, Arsalan Firoozi, Adrian Florea, Van Nguyen, Corentin Puffay, Riki Shimizu, Haotian Ye, Jonathan Brennan, Helmut Schmid, Hinrich Sch\"utze, Nima Mesgarani
Abstract: We introduce XCOMPS in this work, a multilingual conceptual minimal pair dataset covering 17 languages. Using this dataset, we evaluate LLMs' multilingual conceptual understanding through metalinguistic prompting, direct probability measurement, and neurolinguistic probing. By comparing base, instruction-tuned, and knowledge-distilled models, we find that: 1) LLMs exhibit weaker conceptual understanding for low-resource languages, and accuracy varies across languages despite being tested on the same concept sets. 2) LLMs excel at distinguishing concept-property pairs that are visibly different but exhibit a marked performance drop when negative pairs share subtle semantic similarities. 3) Instruction tuning improves performance in concept understanding but does not enhance internal competence; knowledge distillation can enhance internal competence in conceptual understanding for low-resource languages with limited gains in explicit task performance. 4) More morphologically complex languages yield lower concept understanding scores and require deeper layers for conceptual reasoning.
Authors: Yike Wang, Shangbin Feng, Yulia Tsvetkov, Hannaneh Hajishirzi
Abstract: Large Language Models (LLMs) are increasingly used to support scientific research, but their knowledge of scientific advancements can quickly become outdated. We introduce ScienceMeter, a new framework for evaluating scientific knowledge update methods over scientific knowledge spanning the past, present, and future. ScienceMeter defines three metrics: knowledge preservation, the extent to which models' understanding of previously learned papers is preserved; knowledge acquisition, how well scientific claims from newly introduced papers are acquired; and knowledge projection, the ability of the updated model to anticipate or generalize to related scientific claims that may emerge in the future. Using ScienceMeter, we evaluate the scientific knowledge of LLMs through claim judgment and generation tasks on a curated dataset across ten domains. We evaluate five representative knowledge update approaches and find that the best-performing knowledge update methods can preserve only 85.9% of existing knowledge, acquire 71.7% of new knowledge, and project 37.7% of future knowledge, underscoring that developing robust scientific knowledge update mechanisms is both crucial and challenging.
Authors: Ce Li, Xiaofan Liu, Zhiyan Song, Ce Chi, Boshen Shi, Chen Zhao, Guanguang Chang, Zhendong Wang, Kexin Yang, Xing Wang, Chao Deng, Junlan Feng
Abstract: The majority of data in businesses and industries is stored in tables, databases, and data warehouses. Reasoning with table-structured data poses significant challenges for large language models (LLMs) due to its hidden semantics, inherent complexity, and structured nature. One of these challenges is lacking an effective evaluation benchmark fairly reflecting the performances of LLMs on broad table reasoning abilities. In this paper, we fill in this gap by presenting a comprehensive table reasoning benchmark, TReB. Firstly, we propose a taxonomy to systematically measure both shallow table understanding abilities and deep table reasoning abilities, covering a total of 26 sub-tasks. We then construct a high quality dataset through a dedicated data processing and synthesis procedure. Based on these well-constructed samples, we design an evaluation framework to robustly measure table reasoning capabilities with three distinct inference modes. Experimental results with our data and framework reveal that existing LLMs still have significant room for improvement in addressing the complex and real world table related tasks. Both the dataset and evaluation framework are publicly available, with the dataset hosted on https://huggingface.co/datasets/JT-LM/JIUTIAN-TReB, and the framework on https://github.com/JT-LM/jiutian-treb.
URLs: https://huggingface.co/datasets/JT-LM/JIUTIAN-TReB,, https://github.com/JT-LM/jiutian-treb.
Authors: Minghao Guo, Xi Zhu, Qingyue Jiao, Xiujin Liu, Haochen Xue, Chong Zhang, Shuhang Lin, Jingyuan Huang, Ziyi Ye, Yongfeng Zhang
Abstract: Graph Neural Networks (GNNs) have achieved remarkable success in graph-based learning by propagating information among neighbor nodes via predefined aggregation mechanisms. However, such fixed schemes often suffer from two key limitations. First, they cannot handle the imbalance in node informativeness -- some nodes are rich in information, while others remain sparse. Second, predefined message passing primarily leverages local structural similarity while ignoring global semantic relationships across the graph, limiting the model's ability to capture distant but relevant information. We propose Retrieval-augmented Graph Agentic Network (ReaGAN), an agent-based framework that empowers each node with autonomous, node-level decision-making. Each node acts as an agent that independently plans its next action based on its internal memory, enabling node-level planning and adaptive message propagation. Additionally, retrieval-augmented generation (RAG) allows nodes to access semantically relevant content and build global relationships in the graph. ReaGAN achieves competitive performance under few-shot in-context settings using a frozen LLM backbone without fine-tuning, showcasing the potential of agentic planning and local-global retrieval in graph learning.
Authors: Anubhav Jangra, Bahareh Sarrafzadeh, Silviu Cucerzan, Adrian de Wynter, Sujay Kumar Jauhar
Abstract: With the surge of large language models (LLMs) and their ability to produce customized output, style-personalized text generation--"write like me"--has become a rapidly growing area of interest. However, style personalization is highly specific, relative to every user, and depends strongly on the pragmatic context, which makes it uniquely challenging. Although prior research has introduced benchmarks and metrics for this area, they tend to be non-standardized and have known limitations (e.g., poor correlation with human subjects). LLMs have been found to not capture author-specific style well, it follows that the metrics themselves must be scrutinized carefully. In this work we critically examine the effectiveness of the most common metrics used in the field, such as BLEU, embeddings, and LLMs-as-judges. We evaluate these metrics using our proposed style discrimination benchmark, which spans eight diverse writing tasks across three evaluation settings: domain discrimination, authorship attribution, and LLM-generated personalized vs non-personalized discrimination. We find strong evidence that employing ensembles of diverse evaluation metrics consistently outperforms single-evaluator methods, and conclude by providing guidance on how to reliably assess style-personalized text generation.
Authors: Wonhyuk Lee, Youngchol Kim, Yunjin Park, Junhyung Moon, Dongyoung Jeong, Wanjin Park
Abstract: We introduce Guard Vector, a safety task vector computed as the parameter difference between a guardrail model (Guard Model) and a same-architecture pretrained language model. Composing this vector with a target language model yields a Target Guard Model (TGM). We then adapt TGM with a streaming-aware approach that combines prefix-based training and evaluation with a classifier that produces a single-token output. With this composition alone, TGM improves classification quality over established Guard Models across standard safety suites and enables language extensibility to Chinese, Japanese, and Korean, requiring neither additional training nor target language labels for this composition step. It also demonstrates model portability across two widely used public guardrail backbones, Llama and Gemma. With prefix SFT (supervised fine-tuning), TGM preserves classification quality under streaming by aligning the behavior between prefix inputs and full-text inputs. The single-token output design increases throughput and reduces latency. Together, these components reduce data and compute requirements while promoting streaming-aware evaluation practices, thereby contributing to a more responsible AI ecosystem.
Authors: Marco Bronzini, Carlo Nicolini, Bruno Lepri, Jacopo Staiano, Andrea Passerini
Abstract: Despite their capabilities, Large Language Models (LLMs) remain opaque with limited understanding of their internal representations. Current interpretability methods either focus on input-oriented feature extraction, such as supervised probes and Sparse Autoencoders (SAEs), or on output distribution inspection, such as logit-oriented approaches. A full understanding of LLM vector spaces, however, requires integrating both perspectives, something existing approaches struggle with due to constraints on latent feature definitions. We introduce the Hyperdimensional Probe, a hybrid supervised probe that combines symbolic representations with neural probing. Leveraging Vector Symbolic Architectures (VSAs) and hypervector algebra, it unifies prior methods: the top-down interpretability of supervised probes, SAE's sparsity-driven proxy space, and output-oriented logit investigation. By combining the supervised learning paradigm of traditional probes with the dictionary-based representation principle of SAEs, our approach enables deeper input-focused feature extraction while supporting output-oriented analysis. Our experiments demonstrate that our approach consistently extracts meaningful semantic information across different LLMs, embedding sizes, and configurations, uncovering concept-oriented insights into LLM inference across two distinct scenarios: input-completion tasks and QA-focused text generation. VSA-based probing overcomes the limitations of logit-based analyses, which are constrained by the model's token vocabulary, while also mitigating the noisier interpretability outcomes often produced by SAEs in settings with a bounded conceptual feature space. By supporting a joint investigation of input-output features, this work advances the semantic understanding of neural representations while unifying the complementary perspectives of prior methods.
Authors: Peijun Zhu, Ning Yang, Baoliang Tian, Jiayu Wei, Weihao Zhang, Haijun Zhang, Pin Lv
Abstract: Mixture-of-Experts (MoE) Large Language Models (LLMs) face a trilemma of load imbalance, parameter redundancy, and communication overhead. We introduce a unified framework based on dynamic expert clustering and structured compression to address these issues cohesively. Our method employs an online clustering procedure that periodically regroups experts using a fused metric of parameter and activation similarity, which stabilizes expert utilization. To our knowledge, this is one of the first frameworks to leverage the semantic embedding capability of the router to dynamically reconfigure the model's architecture during training for substantial efficiency gains. Within each cluster, we decompose expert weights into a shared base matrix and extremely low-rank residual adapters, achieving up to fivefold parameter reduction per group while preserving specialization. This structure enables a two-stage hierarchical routing strategy: tokens are first assigned to a cluster, then to specific experts within it, drastically reducing the routing search space and the volume of all-to-all communication. Furthermore, a heterogeneous precision scheme, which stores shared bases in FP16 and residual factors in INT4, coupled with dynamic offloading of inactive clusters, reduces peak memory consumption to levels comparable to dense models. Evaluated on GLUE and WikiText-103, our framework matches the quality of standard MoE models while reducing total parameters by approximately 80%, improving throughput by 10% to 20%, and lowering expert load variance by a factor of over three. Our work demonstrates that structural reorganization is a principled path toward scalable, efficient, and memory-effective MoE LLMs. Code is available at https://github.com/szdtzpj/Breaking_the_moe_trilemma
Authors: Samantha Dies, Courtney Maynard, Germans Savcisens, Tina Eliassi-Rad
Abstract: Large language models (LLMs) are widely used as information sources, yet small changes in semantic assumptions can destabilize their beliefs. We introduce P-StaT (Perturbation Stability of Truth), a framework for evaluating belief stability under matched semantic perturbations in both representational and behavioral settings. Across 21 LLMs and three domains, we compare perturbations involving familiar Fictional statements against synthetically generated and unfamiliar Synthetic statements. Unfamiliar Synthetic perturbations generally induce greater epistemic instability than familiar Fictional perturbations, with behavioral belief retraction rates frequently exceeding 0.5 (50%). Finally, exploratory clustering analyses reveal recurring themes among retracted statements, including ambiguity, technical terminology, and obscure concepts. These results show that epistemic familiarity is systematically associated with stability under semantic reframing, suggesting that stability-based analyses can complement accuracy-based benchmarks when evaluating LLM robustness and reliability. Code and data are available at https://github.com/samanthadies/P-StaT.
Authors: Junho Park, Dohoon Kim, Taesup Moon
Abstract: Large language model (LLM) personalization aims to adapt general-purpose models to individual users. Most existing methods, however, are developed under data-rich and resource-abundant settings, often incurring privacy risks. In contrast, realistic personalization typically occurs after deployment under (i) extremely limited user data, (ii) constrained computational resources, and (iii) strict privacy requirements. We propose PRISP, a lightweight and privacy-safe personalization framework tailored to these constraints. PRISP leverages a Text-to-LoRA hypernetwork to generate task-aware LoRA parameters from task descriptions, and enables efficient user personalization by optimizing a small subset of task-aware LoRA parameters together with minimal additional modules using few-shot user data. Experiments on a few-shot variant of the LaMP benchmark demonstrate that PRISP achieves strong overall performance compared to prior approaches, while reducing computational overhead and eliminating privacy risks.
Authors: Chongyuan Dai, Yaling Shen, Zihan Gao, Jia Li, Yishun Jiang, Yaxiong Wang, Liu Liu, Zongyuan Ge, Jinpeng Hu
Abstract: Culture serves as a fundamental determinant of human affective processing and profoundly shapes how individuals perceive and interpret emotional stimuli. Despite this intrinsic link extant evaluations regarding cultural alignment within Large Language Models primarily prioritize declarative knowledge such as geographical facts or established societal customs. These benchmarks remain insufficient to capture the subjective interpretative variance inherent to diverse sociocultural lenses. To address this limitation, we introduce CEDAR, a multimodal benchmark constructed entirely from scenarios capturing Culturally \underline{\textsc{E}}licited \underline{\textsc{D}}istinct \underline{\textsc{A}}ffective \underline{\textsc{R}}esponses. To construct CEDAR, we implement a novel pipeline that leverages LLM-generated provisional labels to isolate instances yielding cross-cultural emotional distinctions, and subsequently derives reliable ground-truth annotations through rigorous human evaluation. The resulting benchmark comprises 10,962 instances across seven languages and 14 fine-grained emotion categories, with each language including 400 multimodal and 1,166 text-only samples. Comprehensive evaluations of 17 representative multilingual models reveal a dissociation between language consistency and cultural alignment, demonstrating that culturally grounded affective understanding remains a significant challenge for current models.
Authors: Bingru Li
Abstract: Data annotation remains a significant bottleneck in the field of humanities and social sciences, particularly for complex linguistic tasks such as metaphor identification. While Large Language Models (LLMs) show promise, a significant gap remains between the theoretical capability of LLMs and their practical utility for researchers. This paper introduces LinguistAgent, an integrated, user-friendly platform that leverages a reflective multi-model architecture to automate linguistic annotation. The platform comprises an Annotator and an optional Reviewer to simulate a peer-review process. This platform supports comparative experiments across three main paradigms: Prompt Engineering (Zero-shot/Few-shot/Chain-of-thought), Retrieval-Augmented Generation, and Fine-tuning. We demonstrate LinguistAgent's efficacy by replicating the task of metaphor identification from a published study, which provides real-time token-level evaluation (F1 and Cohen's kappa) against human gold standards. The application and codes are released on https://github.com/Bingru-Li/LinguistAgent.
Authors: Guangyue Peng, Zongchao Chen, Wen Luo, Yuntao Wen, Wei Li, Ruixiang Feng, Ran Le, Chen Yang, Zhenwei An, Yang Song, Tao Zhang, Houfeng Wang
Abstract: Reverse Chain-of-Thought Generation (RCG) synthesizes reasoning traces from query-answer pairs, but answer-visible generation can justify a pre-committed answer rather than derive it. This post-hoc rationalization creates a train-inference mismatch because student models are trained on answer-conditioned traces but must reason without answer access at inference time. We quantify this mismatch through lexical, trajectory, and probabilistic anchoring, measuring surface overlap, answer-conditioned generation dynamics, and answer recoverability from the trace, respectively. We find that semantic suppression, a seemingly intuitive mitigation, reduces lexical overlap but increases trajectory anchoring: avoiding the answer requires continually tracking it, thereby strengthening its influence on generation. We therefore propose Structural Skeleton-guided Reasoning (SSR), which replaces suppression with structural decoupling by first generating an abstract functional skeleton and then using it to guide the full reasoning trace. Anchoring analyses show that SSR reduces all three forms of answer dependence. Across in-domain and out-of-distribution benchmarks, its distilled variant, SSR-D, improves performance by up to 10 points over suppression baselines and better preserves out-of-distribution performance. Code is available at https://github.com/viniferagy/SSR.
Authors: Sam Kirkham
Abstract: We introduce PyPhonPlan, a Python toolkit for implementing dynamical models of phonetic planning using coupled dynamic neural fields and task dynamic simulations. The toolkit provides modular components for defining planning, perception and memory fields, as well as between-field coupling, gestural inputs, and using field activation profiles to solve tract variable trajectories. We illustrate the toolkit's capabilities through an example application: simulating production/perception loops with a coupled memory field, which demonstrates the framework's ability to model interactive speech dynamics using representations that are temporally-principled, neurally-grounded, and phonetically-rich. PyPhonPlan is released as open-source software and contains executable examples to promote reproducibility, extensibility, and cumulative computational development for speech communication research.
Authors: Abhinav Thorat, Ravi Kolla, Jyotin Goel, Madhav Kataria, Niranjan Pedanekar
Abstract: Creative plot generation presents a fundamental challenge for language models: transforming a concise premise into a coherent narrative that sustains global coherence, character development, pacing, tone consistency, and emotional progression. Although recent Large Language Models (LLMs) demonstrate strong fluency on general-purpose tasks, they require preference alignment to perform well on domain-specific tasks such as creative plot generation. However, conducting such alignment at the scale of frontier LLMs is computationally prohibitive, significantly limiting accessibility and practical deployment. To address this, we present PlotTwist, a structured framework that enables Small Language Models (SLMs) with $\leq$3B active parameters to generate high-quality, premise-conditioned plots competitive with frontier systems of vastly greater parameter scale. Our approach decomposes generation into three specialized components: (1) an Aspect Rating Reward Model, trained via a novel Positive-Negative prompting strategy; (2) a Mixture-of-Experts (MoE) plot generator aligned via Direct Preference Optimization (DPO); and (3) an Agentic Evaluation module using a cross-family jury for unbiased, independent post-hoc assessment. Extensive experiments demonstrate that PlotTwist consistently outperforms all baselines, including frontier models, across multiple Narrative Quality Dimensions (NQDs), achieving higher win rates against every baseline except the strongest, with which it remains competitive. Further validation confirms strong sensitivity to narrative quality, as the framework reliably distinguishes plots derived from critically acclaimed versus widely panned screenplays. Together, these results establish structured, preference-based alignment as a resource-efficient approach to high-quality creative plot generation. Project page: https://abhinavthorat.github.io/plottwist/
Authors: Jocelyn Shen, Amina Luvsanchultem, Jessica Kim, Kynnedy Smith, Valdemar Danry, Kantwon Rogers, Hae Won Park, Maarten Sap, Cynthia Breazeal
Abstract: As users increasingly turn to LLMs for practical and personal advice, they become vulnerable to subtle steering toward hidden incentives misaligned with their own interests. While existing NLP research has benchmarked manipulation detection, these efforts often rely on simulated debates and remain fundamentally decoupled from actual human belief shifts in real-world scenarios. We introduce PUPPET, a theoretical taxonomy and resource that bridges this gap by focusing on the moral direction of hidden incentives in everyday, advice-giving contexts. We provide an evaluation dataset of N=1,035 human-LLM interactions, where we measure users' belief shifts. Our analysis reveals a critical disconnect in current safety paradigms: while models can be trained to detect manipulative strategies, they do not correlate with the magnitude of resulting belief change. As such, we define the task of human belief shift prediction and show that while state-of-the-art LLMs achieve moderate correlation (r=0.3-0.5), they exhibit systematic directional biases, with certain models over or under-predicting the magnitude of human belief change. This work establishes a theoretically grounded and behaviorally validated foundation for AI social safety efforts by studying incentive-driven manipulation in LLMs during everyday, practical user queries.
Authors: A. Feder Cooper, Mark A. Lemley, Christopher De Sa, Lea Duesterwald, Allison Casasola, Jamie Hayes, Katherine Lee, Daniel E. Ho, Percy Liang
Abstract: Recent work shows that standard greedy-decoding extraction methods for quantifying memorization in LLMs miss how extraction risk varies across sequences. Probabilistic extraction -- computing the probability of generating a target suffix given a prefix under a decoding scheme -- addresses this, but is tractable only for verbatim memorization, missing near-verbatim instances that pose similar privacy and copyright risks. Quantifying near-verbatim extraction risk is expensive: the set of near-verbatim suffixes is combinatorially large, and reliable Monte Carlo (MC) estimation can require ~100,000 samples per sequence. To mitigate this cost, we introduce decoding-constrained beam search, which yields deterministic lower bounds on near-verbatim extraction risk at a cost comparable to ~20 MC samples per sequence. Across experiments, our approach surfaces information invisible to verbatim methods: many more extractable sequences, substantially larger per-sequence extraction mass, and patterns in how near-verbatim extraction risk manifests across model sizes and types of text.
Authors: Anna Kozlova, Stanislau Salavei, Pavel Satalkin, Hanna Plotnitskaya, Sergey Parfenyuk
Abstract: We present Doctorina MedBench, a comprehensive evaluation framework for agent-based medical AI based on the simulation of realistic physician-patient interactions. Unlike traditional medical benchmarks that rely on solving standardized test questions, the proposed approach models a multi-step clinical dialogue in which either a physician or an AI system must collect medical history, analyze attached materials (including laboratory reports, images, and medical documents), formulate differential diagnoses, and provide personalized recommendations. System performance is evaluated using the D.O.T.S. metric, which consists of four components: Diagnosis, Observations/Investigations, Treatment, and Step Count, enabling assessment of both clinical correctness and dialogue efficiency. The system also incorporates a multi-level testing and quality monitoring architecture designed to detect model degradation during both development and deployment. The framework supports safety-oriented trap cases, category-based random sampling of clinical scenarios, and full regression testing. The dataset currently contains more than 1,000 clinical cases covering over 750 diagnoses. The universality of the evaluation metrics allows the framework to be used not only to assess medical AI systems, but also to evaluate physicians and support the development of clinical reasoning skills. Our results suggest that simulation of clinical dialogue may provide a more realistic assessment of clinical competence compared to traditional examination-style benchmarks.
Authors: Yuanhao Zeng, Ao Lu, Lufei Li, Zheng Zhang, Yexin Li, Kan Ren
Abstract: Generating diverse responses is crucial for test-time scaling of large language models (LLMs), yet standard stochastic sampling mostly yields surface-level lexical variation, limiting semantic exploration. In this paper, we propose Exploratory Sampling (ESamp), a decoding approach that explicitly encourages semantic diversity during generation. ESamp is motivated by the well-known observation that neural networks tend to make lower-error predictions on inputs similar to those encountered before, and incur higher prediction error on novel ones. Building on this property, we train a lightweight Distiller at test time to predict deep-layer hidden representations of the LLM from its shallow-layer representations to model the LLM's depth-wise representation transitions. During decoding, the Distiller continuously adapts to the mappings induced by the current generation context. ESamp uses the prediction error as a novelty signal to reweight candidate token extensions conditioned on the current prefix, thereby biasing decoding toward less-explored semantic patterns. ESamp is implemented with an asynchronous training--inference pipeline, with less than 5% worst case overhead (1.2% in the optimized release). Empirical results show that ESamp significantly boosts the Pass@k efficiency of reasoning models, showing superior or comparable performance to strong stochastic and heuristic baselines. Notably, ESamp achieves robust generalization across mathematics, science, and code generation benchmarks and breaks the trade-off between diversity and coherence in creative writing. Our code has released at: https://github.com/LinesHogan/tLLM.
Authors: Zhen Zhang, Changyi Yang, Zijie Xia, Zhen Yang, Chengzhi Liu, Zhaotiao Weng, Yepeng Liu, Haobo Chen, Jin Pan, Chenyang Zhao, Yuheng Bu, Alkesh Patel, Zhe Gan, Xin Eric Wang
Abstract: Tokens are the fundamental units of computation in modern autoregressive models, and generation length directly influences both inference cost and reasoning performance. Despite its importance, existing approaches model length primarily at the coarse sequence level. We introduce the Length Value Model (LenVM), a token-level framework that estimates the remaining generation length at every decoding step. By formulating length modeling as a value estimation problem and assigning a constant negative reward to each generated token, LenVM predicts a bounded, discounted return that is a monotone proxy for the remaining generation horizon. This value formulation provides annotation-free, dense, unbiased, and scalable supervision. Experiments on LLMs and VLMs show that LenVM supports exact control, continuous performance--efficiency steering, length prediction, and interpretation. On LIFEBench-token, it raises the exact-length score of Qwen2.5-7B-Instruct from $30.9$ to $64.8$ in one pass and reaches $83.6$ when combined with LCG. Near $200$ tokens on GSM8K, LenVM retains about $63\%$ Pass@1 versus $6\%$ under a hard token budget. Across other settings, LenVM improves matched-length quality over prompting and EOS calibration, while its length-prediction accuracy improves with scale and exceeds specialized predictors. Its token-level values also identify markers of shifts between longer and shorter trajectories. These results establish generation length as an effective scalable token-level value signal for control, prediction, and future value-based training.
Authors: Esra D\"onmez, Agnieszka Falenska
Abstract: Large Language Models (LLMs) can generate high-quality arguments, yet their ability to engage in nuanced and persuasive communicative actions remains largely unexplored. This work explores the persuasive potential of LLMs through the framework of J\"urgen Habermas' Theory of Communicative Action. It examines whether LLMs express illocutionary intent (i.e., pragmatic functions of language such as conveying knowledge, building trust, or signaling similarity) in ways that are comparable to human communication. We simulate online discussions between opinion holders and LLMs using conversations from the persuasive subreddit ChangeMyView. We then compare the likelihood of illocutionary intents in human-written and LLM-generated counter-arguments, specifically those that successfully changed the original poster's view. We find that all three LLMs effectively convey illocutionary intent -- often more so than humans -- potentially increasing their anthropomorphism. Further, LLMs craft sycophantic responses that closely align with the opinion holder's intent, a strategy strongly associated with opinion change. Finally, crowd-sourced workers find LLM-generated counter-arguments more agreeable and consistently prefer them over human-written ones. These findings suggest that LLMs' persuasive power extends beyond merely generating high-quality arguments. On the contrary, training LLMs with human preferences effectively tunes them to mirror human communication patterns, particularly nuanced communicative actions, potentially increasing individuals' susceptibility to their influence.
Authors: Maria Milkova, Maksim Rudnev
Abstract: Measuring subjective constructs in naturally occurring social media text requires annotation procedures that are theoretically grounded, empirically validated, and transferable to an encoder model for scalable prediction. Using non-English social media posts annotated according to Schwartz's theory of basic human values, we investigate how different LLMs, prompts, and instruction languages operationalize the expression of values in text. We argue that although texts may permit multiple plausible interpretations, theory-based value definitions can constrain interpretations and reduce spurious value attributions. Beyond precision, recall, and F1, we evaluate structural alignment between values, error structure, confidence-ambiguity relations, and annotation stability. We show that different LLMs produce different value interpretations. Iterative prompt calibration through error analysis reduces misattributions and improves alignment with expert annotations. We also derive targeted expert verification rules from recurrent error structures and use them during corpus annotation. We show that LLM annotations can be transferred to an encoder model through soft-label training, retaining theory-based value interpretations and information about uncertainty in value expression. Finally, using sensitivity analysis, we show that annotation bias propagates into large-scale predictions, but standardized trends and major event responses are more robust.
Authors: Alex Kwon
Abstract: A language model's memory can be worse than no memory at all when the model or its interface is disposed to act on it: a memory that keeps a wrong conclusion but drops the work behind it leads a model to re-emit the stale value as a confident answer, where an empty memory leads it to abstain. We call this brittle memory. The information loss is definitional; the finding is behavioral, and it turns on one thing, whether the memory kept a re-derivation basis (the source) rather than the answer. We measure it with reclaim evaluation: induce a known drift, compress at a fixed budget, deliver a correction that names the error, and score exact recovery, judge-free. Holding the budget fixed and varying only what the compression keeps isolates correctability from capability and from size; an 8B model and a frontier one wall in the same place. A one-line source-first policy, keep the recomputable source, drop the re-derivable conclusion, restores correctability at equal budget where the source is compact and identifiable, with a length-matched control ruling out "more text." We map where the fix fails, show the failure compounds through memory loops, and replicate across three deployed memory systems, real dialogue (MultiWOZ), and tau-bench, a deployed-agent benchmark where whether a lossy memory becomes a harmful action is a joint model-and-interface property. We release the harness, the paired memory conditions, and validators built to come out false.
Authors: Abrar Alotaibi, Raed Mughus, Moataz Ahmed
Abstract: Large language models (LLMs) have demonstrated remarkable performance across natural language processing tasks, yet their deployment in high-stakes applications raises critical concerns regarding reliability, safety, and trustworthiness. In this paper, we present a red teaming framework that systematically uncovers vulnerabilities in LLM outputs. Our approach employs a novel multi-role architecture comprising target, attacker, and jury models. The attackers generate increasingly effective adversarial prompts while the jury rigorously evaluates response accuracy and consistency across tasks. In a case study, our strategy proved particularly effective at exposing unfaithfulness in LLM responses. Exploitative adversarial prompts increased the attack success rate by up to 7.9% in question-answering tasks, revealing weaknesses in reliability. The approach identifies how structural constraints in summarization can shape vulnerability patterns, with format limitations yielding measurable gains in faithfulness, and shows that architectural design choices typically outweigh parameter scaling in determining model safety. The framework's key strength is its adaptability across evaluation tasks, from English question-answering to Arabic summarization, enabling comprehensive comparison of model vulnerabilities. While it excels at comparing cross-model and cross-linguistic vulnerabilities, it faces challenges in fully automating adversarial prompt generation across languages. Our experiments also reveal limitations in detecting subtle forms of unfaithfulness that do not manifest as explicit factual contradictions, particularly across linguistic contexts. Overall, this architecture provides both actionable insights into current LLM vulnerabilities and a scalable methodology for ongoing safety evaluation as models evolve.
Authors: Yueyang Wang, Baolong Bi, Shuo Lu, Jingyuan Zhang, Jiajun Shi
Abstract: Supervised fine-tuning (SFT) is the standard approach for adapting pretrained language models to downstream domains, yet it often improves target-domain behavior at the cost of degrading pre-existing capabilities. Standard cross-entropy fine-tuning promotes only the observed label token and leaves unconstrained how probability mass is redistributed over other plausible alternatives, potentially distorting the rich local preference structure learned during pretraining. We first analyze next-token predictions using Shannon and Renyi entropies, revealing that pretrained models exhibit a regular multimodal entropy structure. These entropy peaks correspond to varying numbers of plausible alternatives, indicating that the base model intrinsically encodes rich distributional knowledge beyond the single supervised token. Motivated by this observation, we propose LP-SFT, a Local-Preserving Supervised Fine-Tuning objective designed to explicitly protect this inherent entropy structure. At each step, LP-SFT constructs a local top-K support of alternative tokens from the frozen base distribution. Crucially, it removes the supervised target token from this set to avoid conflicting with the cross-entropy objective, and applies a locally normalized KL divergence to maintain the base model's relative preference structure among the remaining non-label alternatives. Across mixed-domain and single-domain fine-tuning experiments, LP-SFT improves overall performance over vanilla SFT and recent SFT-enhancement baselines, achieving the best balance between pass@1 accuracy and pass@k performance. These results suggest that local preservation helps mitigate capability degradation without collapsing sampling-accessible diversity.
Authors: Sadia Kamal, Arefa Patwary, Anthony Marchiafava, Sagnik Ray Choudhury, Atriya Sen
Abstract: Survey-style evaluations of large language models often treat a prompted response as a measure of a model's values or beliefs. This assumption is particularly fragile when responses are read as evidence of political values, social attitudes, or beliefs. We ask whether prompt robustness differs between objective questions with fixed answers and subjective questions that ask for opinions or values. We evaluate four instruction-tuned model families on three objective datasets (MMLU, ARC, and CulturalBench) and three subjective datasets (Political Compass Test, ValueBench, and World Values Survey). For each question/statement, we apply multiple types of prompt changes, such as variations in wording, framing, and format, and measure whether the model gives the same answer across variants. Using a binomial generalized estimating equation, we find significant effects of model, dataset, prompt category, and their interactions. The dataset type effect is also significant, and the interaction between dataset type and prompt category is large. These results show that prompt robustness depends on the question type, the prompt change, and the model.
Authors: Oliver Steele, Jiangtao Wen, Yuxing Han
Abstract: Language builds discourse contexts other than the actual: a painting, a belief, a memory, a hypothetical. Each is a mental space in which the same entity can take a different value, as when a flower is red in reality but purple in a portrait. Formal semantics keeps these contexts apart because their logics differ (modal, temporal, doxastic, depictive). Fauconnier's mental-space theory, by contrast, treats them as one space-building operation. We ask which of these a transformer language model implements, and find a mechanistic version of Fauconnier's unification. The model uses one router/slot format across the inventory: a reusable value slot stores attributed content, and a causally manipulable router (the space index) selects which space is read. A subspace trained with Distributed Alignment Search to control one space type, counterfactual, belief, fictional, or temporal, also controls the others, well above a random floor, on three model families. Belief, which formal semantics marks as a distinct case, is not specially separated. The router is low-rank, composes additively with entity identity, and acts through a few late-layer heads. Two further results show the mechanism drives inference and composes: a subspace trained on a rule-derived conclusion flips what the model infers while dissociating from what it reports, and composing space-builders mints a fresh router over the shared slot. This paper establishes the cross-type generality. A companion paper develops belief in depth, because of its special status in philosophy, psychology, and linguistics (epistemology, theory of mind, and propositional attitude reports).
Authors: Oliver Steele, Jiangtao Wen, Yuxing Han
Abstract: Capable language models hold what a character believes apart from what is true: told "Anna believes the cup is blue; in reality it is red," they answer blue about Anna and red about the world. Where in the computation does that separation live? We show it rests on two separable mechanisms at two positions. A generic value slot binds the attributed value. A router at the query position selects which frame, the character's belief or reality, a query reads out. Two routes fill the slot: an asserted belief, whose value the text supplies, binds in directly; a derived belief, whose value must be inferred from what the character could see, arrives by a visibility-gated lookback. A subspace trained on either route steers the other, and only the derived route depends on described visibility. The slot itself carries no belief-reality tag: intervening on it moves a reality readout as strongly as a belief one. The separation lives instead in a dissociated pair of routing subspaces, which flip a query between frames without injecting the donor's value. These results hold across three architectures, on stimuli de-confounded against theory-of-mind-benchmark shortcuts; the behavior itself emerges between 3B and 7B across five model families. This paper develops the single belief-reality axis in depth; a companion paper shows the same slot-and-router format is shared across the other non-actual contexts a sentence can open (counterfactual, fictional, temporal).
Authors: Michael Solodko, Steven Gong, Guangwei Yu, Satya Krishna Gorti, Jesse C. Cresswell, Victor Zhong
Abstract: While modern question answering (QA) systems excel on clean, schema-aligned corpora, real-world knowledge is rarely so neatly packaged. Answering questions over enterprise and scientific data lakes requires systems to navigate heterogeneous, weakly structured collections of tables, passages, and linked metadata. Current benchmarks abstract away this noisy discovery process, failing to evaluate end-to-end performance. To bridge this gap, we introduce LakeQuest, a human-validated benchmark of 9,846 QA pairs designed to evaluate the end-to-end retrieve-and-synthesize pipeline over realistic data lakes. LakeQuest spans three diverse domains (AI/ML metadata, retail banking, and multimodal biomedical drug information) and pairs every question with exact, modality-aware evidence pointers. By isolating source discovery from cross-modal synthesis, LakeQuest exposes critical failure modes in modern QA systems. Our baseline evaluations, including standard Retrieval-Augmented Generation (RAG) and agentic tool-use methods, reveal that high-quality retrieval does not guarantee correct reasoning. Systems consistently struggle with relation chaining in metadata graphs, policy grounding in bank ledgers, and joint tabular QA in biomedical contexts, highlighting the need for robust discovery and faithful cross-file composition mechanisms in future agentic QA systems.
Authors: Ej Zhou, Lucas Resck, Zheng Hui, Anna Korhonen
Abstract: LLM evaluators (trained reward models and prompted LLM-as-a-Judge) are routinely validated via pairwise accuracy. In a multilingual setting, this operates under the premise that high pairwise accuracy implies reliable, language-neutral scoring. We show that this assumption does not hold. We conduct experiments with semantically identical instruction-response pairs across 23 languages, and find that multilingual evaluators assign significantly different scores to different evaluation languages. The bias is statistically significant and consistent across eight open-weight evaluators of different architectures and training paradigms, persists in frontier judges, and is strongly correlated with language resource level: lower-resource languages are scored more generously. Meanwhile, these biases are invisible to pairwise accuracy: evaluators achieve above 90% pairwise accuracy, yet have up to 43% difference in acceptance rate across languages under a global decision threshold, meaning, for instance, that harmful content in lower-resource languages is more likely to pass safety filters. Per-language thresholds would require language identification, which can be defeated by code-switched prompts. We then investigate why lower-resource languages receive higher rather than lower scores, and we find that model uncertainty is linked with the effect: models tend to give higher scores when less confident, both under negative log-likelihood and under token-free uncertainty measures; however, language identity remains a significant predictor after controlling for uncertainty, and the bias cannot be explained away by content difficulty alone, but is a structural, language-level misalignment.
Authors: Yanze Wang, Pengfei Yao, Tianyi Sun, Chuanrui Hu, Yan Xiao, Yunyun Han, Jun Sun, Yafeng Deng
Abstract: Agent skills, SKILL files that package reusable procedural knowledge for an LLM agent, are a popular mechanism for extending agent capabilities. Public repositories now host them in large and growing numbers, yet these artifacts are fragmented, redundant, and uneven in quality, and their value in practice is unclear. A core question remains open, namely how to consolidate this open-source SKILL ecosystem into a single usable corpus, and what bounds its benefit on real-world agent tasks. We present SkillCorpus, a framework that aggregates, curates, matches, and evaluates the open skill ecosystem at scale. It filters ~821,000 crawled skills through a multi-stage pipeline into 96,401 skills organised by a 16-class taxonomy and three quality facets (utility, robustness, safety), and pairs them with a fine-tuned retrieval-and-selection stack that matches task-relevant skills. We evaluate end-to-end across three benchmarks (SkillsBench, GDPVal, QwenClawBench), two harnesses, and two open backbones with a frontier robustness check. Integrating SkillCorpus yields consistent gains across all three benchmarks, largest on SkillsBench (+7.5 pp). An operational analysis traces the gains to a coverage boundary and a harness boundary. SkillCorpus is, to our knowledge, the first end-to-end account of when a curated, retrieval-served community corpus improves real agent tasks, and where it does not. The dataset, models, and code will be released upon acceptance.
Authors: Ajay Patel, Kartik Hosanagar, Ramayya Krishnan, Chris Callison-Burch, Karim Lakhani
Abstract: Large language models (LLMs) are improving rapidly as reflected in benchmark scores, yet these AI benchmarks largely test capabilities such as factual recall, narrow question answering, mathematical problem-solving, and coding and agentic tool-use. What remains poorly measured is AI progress on the analytical knowledge work white-collar professionals perform daily, including synthesizing complex information, exercising judgment under uncertainty and incomplete information, applying strategic and adversarial thinking in multi-stakeholder settings, weighing trade-offs, and producing defensible, structured analyses. This gap is even more pronounced for subjective components of such work, where success can be challenging to define. The "case method" form of education practiced by top business schools provides a natural foundation for addressing this measurement gap, and we construct BusinessCaseBench, a benchmark spanning hundreds of questions drawn from business cases across eighteen disciplines, each paired with a grading rubric derived from the expert-written instructor case solution. On BusinessCaseBench, frontier AI models already score highly against instructor rubrics, and capability within one model family improves substantially over two years. These results provide strong evidence that AI performance on this class of work is already high and rapidly improving, with implications for business schools, where case pedagogy trains undergraduates and MBAs in this kind of analytical reasoning, and for entry-level professional roles, where such skills have historically anchored early-career work.
Authors: Mohammad Arvan, Amber E. Osterholt, Bailee Rue, Yuvaneswaren R. Sureshbabu, Krishna R. Patel, Rebecca T. Feinstein, Bethany C. Bray, Niranjan S. Karnik
Abstract: Introduction. Clinical and Translational Science Award (CTSA) programs must document their scholars' research impact, but assembling each scholar's record by hand takes staff an estimated 15 hours and does not scale to a full cohort. An artificial intelligence (AI) agent could serve as a tool to gather scholar data across platforms and disciplines. Methods. We built a human-in-the-loop AI agent that assembles a dossier of sourced evidence for each scholar and drafts one-sentence Translational Science Benefits Model (TSBM) impact summaries for staff review. We evaluated it in the impact-reporting workflow of one CTSA hub across 10 career-development (KL2/K12) scholars. Two evaluation staff independently coded all 507 findings as accept, edit, or reject; the primary measure was the unanimous usable rate, defined as the share both accepted or edited. Results. Both reviewers accepted or edited 81.7% of the agent's findings. Reviewers each spent a median of 14 minutes per scholar, replacing an estimated 15 hours of manual assembly. Inter-rater agreement was moderate (Cohen's kappa 0.43 on the usable-versus-reject decision). A profile discovery study found the agent's recall close to human search. The agent's impact evidence spanned all four TSBM domains, and about a third of the reviewed findings fell in non-scholarly categories that routine processes tend to miss. Reviewers rated synthesis accuracy 4.5 and usefulness 4.8 on a 5-point scale. Conclusions. A human-in-the-loop AI agent can serve as the first-pass author of a scholar's impact record, shifting staff from collecting and writing to reviewing, and making cohort-scale impact reporting feasible.
Authors: Victoria R. Li, Jenny Kaufmann, Tian Qin, Martin Wattenberg, David Alvarez-Melis, Naomi Saphra
Abstract: Interpretability research often predicts model responses to targeted mechanistic interventions. But can we predict responses to unseen input data? We propose and demonstrate this alternate objective by using model internals to predict their out-of-distribution (OOD) behavior. We train hundreds of Transformers on simple synthetic tasks, where perfect in-distribution accuracy is compatible with multiple OOD generalization rules. We successfully use attention patterns -- observed only on in-distribution data -- to predict which rule each model follows on OOD data. Our experiments decouple the mechanistic faithfulness of our interpretation from its predictive value; ablations reveal such internal patterns can suppress rather than support the rule they predict, showing observational analysis can forecast behavior even when causal analysis fails to support a simple cause-effect link. Our findings are a proof-of-concept for a new interpretability objective: understanding model internals to predict behavior and assess reliability under distribution shift.
Authors: Zhengyang Su, Isay Katsman, Yueqi Wang, Ruining He, Lukasz Heldt, Raghunandan Keshavan, Shao-Chuan Wang, Xinyang Yi, Mingyan Gao, Onkar Dalal, Lichan Hong, Ed Chi, Ningren Han
Abstract: Generative retrieval has emerged as a powerful paradigm for LLM-based recommendation. However, industrial recommender systems often benefit from restricting the output space to a constrained subset of items based on business logic (e.g. enforcing content freshness or product category), which standard autoregressive decoding cannot natively support. Moreover, existing constrained decoding methods that make use of prefix trees (Tries) incur severe latency penalties on hardware accelerators (TPUs/GPUs). In this work, we introduce STATIC (Sparse Transition Matrix-Accelerated Trie Index for Constrained Decoding), an efficient and scalable constrained decoding technique designed specifically for high-throughput LLM-based generative retrieval on TPUs/GPUs. By flattening the prefix tree into a static Compressed Sparse Row (CSR) matrix, we transform irregular tree traversals into fully vectorized sparse matrix operations, unlocking massive efficiency gains on hardware accelerators. We deploy STATIC on a large-scale industrial video recommendation platform serving billions of users. STATIC produces significant product metric impact with minimal latency overhead (0.033 ms per step and 0.25% of inference time), achieving a 948x speedup over a CPU trie implementation and a 47-1033x speedup over a hardware-accelerated binary-search baseline. Furthermore, the runtime overhead of STATIC remains extremely low across a wide range of practical configurations. To the best of our knowledge, STATIC enables the first production-scale deployment of strictly constrained generative retrieval. In addition, evaluation on academic benchmarks demonstrates that STATIC can considerably improve cold-start performance for generative retrieval. Our code is available at https://github.com/youtube/static-constraint-decoding.
URLs: https://github.com/youtube/static-constraint-decoding.
Authors: Florent Draye, Vedant Palit, Abir Harrasse, Tung-Yu Wu, Jiarui Liu, Punya Syon Pandey, Roderick Wu, Chih-Hao Hsu, Terry Jingchen Zhang, Zhijing Jin, Bernhard Sch\"olkopf
Abstract: Mechanistic interpretability seeks to understand how Large Language Models (LLMs) represent and process information. Recent approaches based on dictionary learning and transcoders enable representing model computation in terms of sparse, interpretable features and their interactions, giving rise to feature attribution graphs. However, these graphs are often large and redundant, limiting their interpretability in practice. Cross-Layer Transcoders (CLTs) address this issue by sharing features across layers while preserving layer-specific decoding, yielding more compact representations, but remain difficult to train and analyze at scale. We introduce an open-source library for end-to-end training and interpretability of CLTs. Our framework integrates scalable distributed training with model sharding and compressed activation caching, a unified automated interpretability pipeline for feature analysis and explanation, attribution graph computation using Circuit-Tracer, and a flexible visualization interface. This provides a practical and unified solution for scaling CLT-based mechanistic interpretability. Our code is available at: https://github.com/LLM-Interp/CLT-Forge.
Authors: Pavel Golikov, Evgenii Opryshko, Gennady Pekhimenko, Mark C. Jeffrey
Abstract: While Large Language Models (LLMs) achieve high performance on standard mathematical benchmarks, their problem-solving abilities depend on the context and textual formatting. We introduce the Robust Reasoning Benchmark (RRB), a pipeline of 13 deterministic textual perturbations applied to AIME 2024 and AIME 2025. Evaluating 8 state-of-the-art models, we find that frontier models are largely resilient, with the notable exception of Claude, which categorically refuses many transformed prompts. Open-weights reasoning models exhibit a range of failure modes under structural noise (cognitive thrashing, tokenization breakdown, and reasoning collapse), with up to 54% average accuracy drops across perturbations and up to 100% on some. We further study one of these failure modes in isolation: attention dilution caused by the model's own chain-of-thought. By tasking models with solving multiple independent mathematical problems sequentially within a single context window, we identify Intra-Query Attention Dilution. Open-weights models ranging from 7B to 120B parameters exhibit accuracy decay on subsequent problems, suggesting that intermediate reasoning steps progressively pollute standard dense attention mechanisms. We argue that in order to achieve reliable reasoning, future architectures need to integrate explicit contextual resets within models' own chain-of-thought, leading to open research questions regarding the optimal granularity of reasoning tasks.
Authors: Denys Katerenchuk, Pablo Duboue, Keelan Evanini, David Gondek, Nithin Govindugari, Olivier Allauzen, Joshua Baptiste, David J More, Joshua Schechter
Abstract: Large language models (LLMs) are rapidly being adopted across various domains. However, their adoption in banking industry faces resistance due to demands for high accuracy, regulatory compliance, and the need for verifiable and grounded responses. We present a unified, data-efficient framework for training grounded domain-specific LLMs that optimizes answer quality, citation grounding, and calibrated refusal under real-world deployment constraints. First, we describe a data generation pipeline that combines LLM-as-a-Judge filtering, citation annotation, and curriculum learning with only 143M tokens. The resulting 12B model achieves high answer quality outperforming GPT-4.1 on citation grounding, with a modest citation tradeoff versus the untuned base. Second, we propose a calibrated refusal mechanism: training on 22% unanswerable examples yield a 12% "I don't know" rate, substantially improving over the base model's unsafe 4.3% rate while avoiding GPT-4.1's over-refusal (20.2%). Third, we present an end-to-end methodology spanning from data curation to quantized serving. The system is deployed at 40+ financial institutions, achieving a 7.1 percentage point improvement in query resolution (p < 0.001). Additionally, the model delivers 3-5x faster responses at 20-50x lower cost compared to GPT-4.1.
Authors: Xin Ma, Wei Chen, Qi Liu, Derong Xu, Zhi Zheng, Tong Xu, Enhong Chen
Abstract: Lifelong Model Editing aims to continuously update evolving facts in Large Language Models while preserving unrelated knowledge and general capabilities, yet it remains plagued by catastrophic forgetting and model collapse. Empirically, we find that recent editors resilient over long horizons share the same core strategy: Lifelong Normalization (LN), which normalizes value gradients using running statistics. Removing LN causes immediate performance collapse, and we observe a counter-intuitive positive cumulative effect where early edits can promote the success of future edits. Yet the mechanism of LN remains a "black box", leaving its precise role in lifelong stability poorly understood. In this work, we provide the first theoretical account of LN in the lifelong regime. Our analysis reveals a self-reinforcing stability loop and proves that, when combined with ridge-regularized regression, LN yields parameter updates with asymptotic orthogonality and bounded norms, directly mitigating forgetting and systemic collapse. Based on these insights, we derive StableEdit, which strengthens this stability loop via an explicit warm-up stage and full whitening, improving long-horizon stability at minimal overhead. Extensive experiments validate our theory and demonstrate competitive performance. Our code is available at https://github.com/MINE-USTC/StableEdit.
Authors: Syed Rifat Raiyan, Mohsinul Kabir, Hasan Mahmud, Md Kamrul Hasan, Sophia Ananiadou
Abstract: Mathematical reasoning has long served as a stringent test of machine intelligence; over the past decade, it has moved from a niche problem within NLP to one of the most consequential AI frontiers. This survey provides a unified account of the field's evolution, from early rule-based math word problem (MWP) solvers and template-driven geometry systems, through neural expression generation and LLM prompting, to contemporary reasoning models, multi-agent systems, neuro-symbolic theorem provers, and verified discovery workflows. We organize the landscape along four axes: (i) informal reasoning over text and diagrams, spanning MWP solving, multimodal geometry, and VLMs; (ii) formal reasoning in proof assistants, including autoformalization, tactic prediction, compiler-guided repair, and proof search; (iii) mathematical discovery, where systems propose constructions, improve bounds, or assist attacks on open problems; and (iv) the inference and training-time techniques, including CoT prompting, tool use, process reward models, and RLVR, that increasingly connect generation with verification. We catalog major benchmarks across grade-school arithmetic, competition mathematics, geometry, formal proving, multimodal and multilingual reasoning, and expert evaluation, and we examine benchmark saturation, contamination, reporting mismatches, and the distinction between pass@1, majority voting, and verifier-assisted pass@$k$. We critically assess failure modes: brittleness under perturbation, reward hacking, multimodal grounding failures, fragile formalization, and the energy cost of reasoning-scale inference. Drawing on recent perspectives from working mathematicians, we identify future directions centered on verified-discovery workflows, reasoning efficiency, and infrastructure to make AI-assisted formalization broadly usable. Companion materials: https://github.com/Starscream-11813/awesome-AI4Math.
Authors: Aaditya Pai
Abstract: Prompt injection defenses evaluated on synthetic benchmarks do not generalize to real enterprise documents, which are longer, denser, and interleave legitimate authority language with factual content. We demonstrate this gap with a real-document benchmark of 122 tasks across five professional domains (financial, legal, medical, scientific, DevOps) using actual SEC filings, Federal Register rules, PubMed abstracts, arXiv papers, and GitHub postmortems. Paraphrasing, the strongest defense on synthetic benchmarks, shows no statistically significant attack success rate reduction on real documents (p=0.500) while degrading utility from 91.8% to 82.8%. We introduce PARSE (Provenance-Aware Retrieval Sanitization), a domain-aware, fact-preserving sanitization pipeline that classifies each sentence by injection likelihood, extracts structured facts before rewriting, and verifies fact preservation via a consistency-checking loop. A directiveness gate routes 59% of real enterprise documents to a lightweight path, concentrating computational cost on high-risk documents. PARSE achieves 15.6% attack success rate -- a 38% reduction versus the 25.4% baseline -- at 86.9% utility, the only condition that is both statistically significant (p=0.014, adequately powered) and maintains near-baseline utility. Practitioners should evaluate defenses on domain-matched real documents, not synthetic proxies.
Authors: Bu\u{g}ra Alperen Ulu{\i}rmak, Rifat Kurban
Abstract: LLM evaluation and AI safety face a shared measurement problem: benchmark scores, reward-model signals, and reported safety metrics can improve while the latent properties they are meant to represent remain difficult to verify. This paper combines a hybrid survey - a systematic search paired with narrative synthesis and separately tracked grey evidence - with a conceptual framework and a structured ten-model audit. The synthesis spans eight evidence streams: benchmark validity, dynamic evaluation, LLM-as-judge reliability, safety evaluation, jailbreak/refusal robustness, reward hacking, mechanistic interpretability, and governance/auditability, covering 2018-2026 evaluation-safety measurement work. We introduce EvalSafetyGap as an organizing hypothesis for comparing evaluation-side and alignment-side proxy failures under optimization pressure, using Goodhart's Law together with two constructs we develop here - an Instability Decomposition and an Alignment Trilemma - as tools for generating testable comparisons. The audit shows how conclusions shift when capability, behavioral safety, and governance are measured separately. In this sample ($n = 10$), the association between capability and sustained adversarial robustness is statistically indeterminate using the displayed Table 3 inputs (Pearson $r = +0.232$, $p = 0.520$), and the apparent open-closed safety gap is modest, driven mainly by governance and disclosure rather than behavioral robustness, and sensitive to how a single borderline model is classified; attempt-budget results are protocol dependent. Because the public evidence uses heterogeneous protocols, the audit is diagnostic rather than rank-generating. The contribution is a shared vocabulary and evidence map to support dynamic evaluation, transparent source reporting, multi-attempt safety measurement, and auditable alignment practice.
Authors: Maximo Eduardo Rulli, Thomas Vaitses Fontanari, Simone Petruzzi, Federico Alvetreti, Giorgio Strano, Donato Crisostomi, Giorgos Nikolaou, Tommaso Mencattini, Andrea Santilli, Emanuele Rodol\`a, Simone Scardapane, Alessio Devoto
Abstract: Diffusion Language Models (DLMs) have recently emerged as a promising alternative to autoregressive models. Unlike standard diffusion-based approaches, DLMs are not explicitly conditioned on a timestep, raising a natural question: do these models internally represent denoising progress, and how is such information used downstream? In this work, we show that DLMs do in fact encode a latent representation related to the diffusion timestep within their residual streams. We find that this signal can be reliably extracted using probes across layers, indicating that denoising progress is decodable from internal activations. We further demonstrate that steering the model along a low-dimensional subspace associated with the inferred timestep allows us to systematically modulate its notion of denoising progress, leading to predictable changes in model confidence and entropy. Finally, we analyse the geometry of the identified representation, showing that it exhibits structured and interpretable properties in activation space, and shedding light on how such a signal is processed by these models.
Authors: Kexin Huang, Junkang Wu, Jinda Lu, Shuo Yang, Chiyu Ma, Jiancan Wu, Xiang Wang, Xiangnan He, Guoyin Wang, Jingren Zhou
Abstract: Reinforcement learning (RL) has significantly enhanced the reasoning capabilities of large language models (LLMs), yet the training process remains notoriously fragile. In this work, we investigate a critical source of this instability: over-optimization, where models exploit training heuristics at the expense of generalizable reasoning. While reverse KL regularization is the standard defense against such degradation, our analysis reveals that it is often insufficient in this regime, as it fails to ensure comprehensive coverage of the reference distribution. To address this, we propose ARMOR (Anchor Rollout and Mixed Optimization for RL), a framework that shifts the paradigm from passive penalty to active sample stabilization. ARMOR comprises two key components: (1) Anchor Rollout, which leverages off-policy data from the reference policy to preserve established solution patterns; and (2) Mixed Optimization, which reformulates the policy objective to enable controlled exploration without relying on auxiliary losses. Extensive experiments on reasoning benchmarks validate that ARMOR effectively mitigates validation collapse, enabling sustained performance improvements over extended training horizons.
Authors: Yanqiao Zhu, Jingru Gan, Xiaoqi Sun, Fang Sun, Yidan Shi, Md Mofijul Islam, Chao Shang, Wenhao Gao, Connor W. Coley, Yizhou Sun, Wei Wang
Abstract: Multi-step retrosynthesis planning seeks to decompose a target molecule into commercially available building blocks through a sequence of feasible reactions. The vast combinatorial search space makes this task challenging even for expert chemists. Traditional methods combine tree search with offline-trained value networks that score candidates in isolation, without reasoning about complete multi-step routes. Recent work leverages Large Language Models (LLMs) for this task, but relies on simple interfaces that limit exploration of the full search space. We introduce RetroAgent, an LLM agent that bridges symbolic search and neural reasoning through a harness with structured memory. Through memory and chemistry tools, the agent observes the full search state, including explored routes, available alternatives, and properties of intermediates, enabling informed decisions grounded in both global progress and domain knowledge. Experiments on in-distribution and out-of-distribution benchmarks demonstrate that RetroAgent delivers strong performance and generalization.
Authors: Patrick Phuoc Do, Chau M. Ta, Chaoli Wang
Abstract: Multimodal large language models (MLLMs) are increasingly used to interpret visualizations, yet current evaluations remain largely chart-centric and provide limited evidence of understanding of scientific visualization (SciVis). We benchmark six MLLMs on the scientific visualization literacy assessment test, a standardized SciVis literacy assessment comprising 49 items based on 18 scientific visualizations and illustrations, spanning 8 techniques and 11 task types. We evaluate three closed-source and three open-source models under a closed-world protocol and compare their performance using data from 485 human participants. Results show that current MLLMs do not exhibit uniform SciVis literacy. Gemini is the strongest model overall, exceeding the human mean across the evaluated subsets, whereas the open-source models remain below the human baseline. Performance is highly uneven across techniques and tasks: models perform best on scientific illustration, search, and spatial understanding, but struggle on texture-based and integration-based visualizations and on quantitative estimation. Error analysis reveals recurring failures in fine-grained quantitative estimation, flow-direction interpretation, and grounded encoding interpretation. These findings position SciVis literacy as a necessary benchmark dimension for evaluating multimodal AI systems. Our code and model outputs are publicly available at https://github.com/patdmp/mllm-scivis-lit-benchmark.
Authors: Yao Liu, Tien-Ping Tan, Zhilan Liu
Abstract: Prior case retrieval (PCR) aims to identify the precedent cases relevant to the facts of a query case. Incoming citation context, the text with which later cases characterize a case when citing it, is a powerful relevance signal, yet it is typically evaluated without a temporal constraint, so the retriever is credited with citations made after the query. We introduce a zero-training, temporally fenced retriever that augments BM25 with incoming citation context restricted to citations predating the query, together with a temporal-admission decomposition that quantifies the phantom fraction: the share of a citation-context gain attributable to citations not known to predate the query. Experiments span two jurisdictions, U.S. federal (CLERC) and European (ECtHR-PCR) case law. On ECtHR-PCR, without any training, the fenced retriever outperforms a strong degree-controlled baseline across the full recall ladder, and a temporal-admission decomposition attributes 14.9% (validation) of an unfenced citation-context gain over BM25 to citations not known to predate the query. Citation-context retrieval must therefore be temporally fenced and degree-controlled before its reported gains can be interpreted.
Authors: Jijun Chi, Zhenghan Tai, Hanwei Wu, Tung Sum Thomas Kwok, Hailin He, Zixing Liao, Bohuai Xiao, Chaolong Jiang, Jianliang Lei, Jerry Huang, Peng Lu, Muzhi Li, Liheng Ma, Yihong Wu, Sicheng Lyu, Jingrui Tian, Yihan Li, Yanzhang Ma, Sizhe Guan, Dingtao Hu, Yufei Cui, Ling Zhou, Lei Ding, Xinyu Wang
Abstract: Financial question answering over U.S. Securities and Exchange Commission (SEC) filings requires retrieving and synthesizing heterogeneous evidence dispersed across long, standardized, and highly redundant disclosures. Existing retrieval-augmented and multi-agent systems typically derive retrieval queries directly from the user's question and rank candidates by semantic similarity. Together, these choices create prior-corpus misalignment: a mismatch between model priors and the target filings' structure, terminology, and evidence standards. As a result, query generation misses corpus-specific evidence, while semantic reranking favors topically similar but evidentially invalid false-positive chunks. We propose FinSAgent, an evidence-grounded multi-agent framework that reframes SEC filing QA as corpus-aligned retrieval planning and corrects both ends with a single principle: inject corpus-side conditioning wherever model priors would otherwise dominate. FinSAgent combines (1) role-specialized agents anchored to the mandated 10-K item structure, (2) database-aware query decomposition that conditions each agent's sub-queries on a lightweight, summary-level view of the local corpus, and (3) multi-path retrieval with a learned feature-gated reranker that separates evidential validity from semantic similarity. Across five offline financial QA benchmarks, FinSAgent improves retrieval coverage and answer correctness over strong single-agent and multi-agent baselines; in a three-arm randomized online experiment with 1,000 anonymous user ratings, it also receives higher scores than baselines.