Authors: Yiming Huang, Zhenbo Shi, Xin-Cheng Wen, Jichuan Zeng, Cuiyun Gao, Peiyi Han, Chuanyi Liu
Abstract: Unsupervised reinforcement learning (RL) has emerged as a promising paradigm for enabling self-improvement in large language models (LLMs). However, existing unsupervised RL-based methods often lack the capacity to adapt to the model's evolving reasoning capabilities during training. Therefore, these methods can misdirect policy optimization in the absence of ground-truth supervision. To address this issue, we introduce FREIA, a novel RL-based algorithm built on two key innovations: (1) Free Energy-Driven Reward (FER) adapts rewards to balance consensus and exploration based on the Free Energy Principle. (2) Adaptive Advantage Shaping (AAS) adaptively adjusts learning signals based on the statistical characteristics of sampled rewards. Empirical evaluations on nine datasets across three reasoning tasks showcase that FREIA outperforms other unsupervised RL-based baselines. Notably, in mathematical reasoning tasks, FREIA surpasses other methods by an average of 0.5 to 3.5 points in Pass@1 using the DeepSeek-R1-Distill-Qwen-1.5B model.
Authors: Yiming Huang, Zhenbo Shi, Shuzheng Gao, Cuiyun Gao, Peiyi Han, Chuanyi Liu
Abstract: Reinforcement Learning with Verifiable Rewards (RLVR) is an essential paradigm that enhances the reasoning capabilities of Large Language Models (LLMs). However, existing methods typically rely on static policy optimization schemes that misalign with the model's evolving reasoning capabilities. To address this issue, we propose Adaptive Power-Mean Policy Optimization (APMPO), which comprises two main innovations: Power-Mean Policy Optimization (PMPO) and Feedback-Adaptive Clipping (FAC). Specifically, PMPO introduces a generalized power-mean objective. This enables the model to adaptively transition from the signal-amplifying behavior of the arithmetic mean to the consistency-enforcing behavior of the geometric mean. FAC adaptively adjusts clipping bounds based on real-time reward statistics to overcome the limitations of static mechanisms. Capitalizing on these innovations, APMPO improves learning dynamics and reasoning performance. Extensive experiments on nine datasets across three reasoning tasks showcase the superiority of APMPO over state-of-the-art RLVR-based baselines. For instance, APMPO boosts the average Pass@1 score on mathematical reasoning benchmarks by 3.0 points compared to GRPO when using Qwen2.5-3B-Instruct.
Authors: Vageesh Kumar Saxena
Abstract: This research investigated how online criminal activities can be better understood and connected using data-driven machine learning methods. Many illegal activities, such as human trafficking and illicit trade, have moved to online platforms where offenders hide behind anonymous accounts and frequently change identities. This makes it difficult for authorities to understand how large these networks are and how different online profiles may be linked. The research shows that people tend to maintain consistent patterns in how they write advertisements and present images online, even when they try to stay anonymous. By analysing these patterns across large collections of online advertisements, the research demonstrates how to link related accounts and identify repeated behaviour across illegal online markets. In addition, the research also addresses how such methods should be used responsibly. It proposes clear guidelines to ensure that privacy, fairness, and transparency are respected when these tools are applied. Overall, the research provides practical ways to support law enforcement investigations while emphasising careful and ethical use.
Authors: Elitsa Yotkova, Violeta Kastreva, Dimitar Dimitrov, Ivan Koychev, Preslav Nakov
Abstract: SemEval-2026 Task 13 investigates machine-generated code detection across multiple programming languages and application scenarios, asking participating systems to generalize to unseen languages and domains. This paper describes our participation in Subtask A (binary classification) and explores both pretrained code encoders and lightweight feature-based methods. We design ratio-based features that are less sensitive to snippet length. To support the extraction of descriptiveness-related signals, we use parsing engines and a programming-language classifier. Additionally, we train a separate code-vs-text line classifier to identify raw natural language segments embedded within samples. We combine a shallow decision tree with heuristic rules derived from data analysis to produce the final predictions. Our approach is computationally efficient, requires only CPU resources for training, and achieves near-instant inference time, offering a lightweight alternative to large pretrained models.
Authors: Humam Khan, Md Tabrez Nafis, Shahab Saquib Sohail, Aqeel Khalique, Rehan Hasan Khan
Abstract: Large Language models (LLMs) show extraordinary abilities, but they are still prone to hallucinations, especially when we use them for generating Academic content. We have investigated four popular LLMs, ChatGPT, Grok, Gemini, and Copilot for hallucinations specifically for academic writing. We have designed 80 prompts across four categories, namely, reference generation, factual explanation, abstract generation, and writing improvement. We evaluated the model using a 0-5 rubric score, which checks factual accuracy, reference validity, coherence, style consistency, and academic tone. A novel weighted metric, Hallucination Index (HI), was introduced to measure hallucination in the responses generated by the models. Some of the most widely used evaluation metrics often fail to check errors which alter sentiment in machine-translated text. We found that Grok and Copilot perform better on reference generation tasks, but they often struggle with abstract or stylistic prompts, with HI values of 0.67 and 0.70, respectively. Whereas, Gemini and ChatGPT have done well with having stronger tone control, but they lack in writing factual tasks and higher hallucination risk with HI scores of 0.53 and 0.57, respectively. Our study found that hallucination behavior does not depend solely on model architecture but also on the type of task and the prompting conditions we are providing. We propose that our work opens new research dimensions for future researchers.
Authors: Hoffmann Muki, Olukunle Owolabi
Abstract: As LLMs enter conflict monitoring, understanding systematic distortions in their outputs is critical for humanitarian accountability. We evaluate four vanilla open-weight models Gemma 3 4B, Llama 3.2 3B, Mistral 7B, and OLMo 2 7B and two domain-adapted models, AfroConfliBERT and AfroConfliLLAMA, on Nigeria and Cameroon conflict-event classification against ACLED, a gold-standard dataset with multi-stage verification. We find a bifurcated divergence in normative directionality. Open-weight models exhibit statistically significant False Illegitimation bias: Gemma misclassifies to 18.29% of legitimate battles as civilian-targeted violence while making zero False Legitimation errors. By contrast, AfroConfliBERT and AfroConfliLLAMA achieve near-directional neutrality, with Legitimization Bias differences indistinguishable from zero. Yet domain adaptation does not eliminate actor-based selection bias. Both adapted models show statistically significant actor bias comparable to vanilla LLMs; in Nigeria, state actors are legitimized 36.5% more often than non-state actors in identical tactical contexts. Open-weight outputs are also fragile to geography-specific lexical framing: delegitimizing phrases produce flip rates up to 66.7% in Cameroon and 34.2% in Nigeria, while perturbations salient in one context may not matter in another. Error trace profiling shows models mask normative bias through unfaithful rationale confabulations. In contrast, AfroConfliBERT and AfroConfliLLAMA are largely robust, with near-zero flip rates across perturbation categories. Overall, current models are not ready for unsupervised deployment in conflict monitoring. We call for fairness-aware fine-tuning to reduce actor-based selection bias, mandatory adversarial robustness evaluation against lexical manipulation, and context-specific human-in-the-loop oversight calibrated to regional difficulty.
Authors: Tung Sum Thomas Kwok, Qian Qian, Xiaofeng Lin, Dongxu Zhang, Jun Han, Zhichao Yang, Davin Hill, Tamer Soliman, Sanjit Singh Batra, Robert Tillman, Guang Cheng
Abstract: Large Language Models exhibit strong reasoning and semantic understanding capabilities but often hallucinate in domains that require expert knowledge, among which fabrications, the generation of factually incorrect yet fluent statements, pose the greatest risk in medical contexts. Existing medical hallucination datasets inadequately capture fabrication phenomena due to limited fabrication coverage, stylistic disparities between human and LLM-authored texts, and distributional drift during hallucinated sample synthesis. To address this, we propose a data-centric pipeline to generate realistic and word-level fabrications that preserve syntactic and stylistic fidelity while introducing subtle factual deviations, resulting in MedFabric. Building upon this dataset, we introduce ETHER, a modular word-level fabrication detector integrating Text2Table Decomposition, Word Masking and Filling and Hybrid Sentence Pair Evaluation to enhance factual alignment. Empirical results demonstrate that MedFabric outperforms state-of-the-art detectors by over 15% on word-level fabrication benchmarks while maintaining consistent performance across structural similarities, offering a comprehensive framework for reliable and domain-specific factuality detection.
Authors: Oona Itkonen, J\"org Tiedemann
Abstract: Knowledge transfer, especially across related languages, has been found beneficial for multilingual neural machine translation (MNMT), but some aspects are still under-explored and deserve further investigation. A joint vocabulary is most often applied to form a uniform word embedding space, but since the impact of a disjoint vocabulary on model performance is far less studied, there is no consensus on how much knowledge transfer is mainly due to vocabulary overlap. In this paper, we present systematic experiments with joint and disjoint vocabularies, and auxiliary languages related and unrelated to the source language. We design this experiment in an out-of-domain setup in order to emphasize transfer and the impact of the auxiliary language. As expected, we yield better results with more extensive vocabulary overlaps typical for related languages, but our experiments also show that domain-match and language relatedness are more important than a joint vocabulary.
Authors: Stephen E. Moore, Mich-Seth Owusu, Akwasi Asare, Lawrence Adu Gyamfi, Paul Azunre, Joel Budu, Jonathan Asiamah, Elias Dzobo, Kelvin Newman, Edmund O. Benefo, Gerhardt Datsomor, Onesimus Addo Appiah, Ama Branoa Banful, Lucas Woedem Kpatah, Saani Mustapha Deishini, John Ayernor
Abstract: Large language models (LLMs) have demonstrated impressive multilingual capabilities for well-resourced languages, yet their performance on low-resource African languages remains poorly understood and largely unevaluated. This paper presents Nsanku, a systematic benchmark that evaluates the zero-shot machine translation performance of 19 open-weight and proprietary LLMs across 43 Ghanaian languages paired with English. Evaluation sentences were sourced from the YouVersion Bible platform, providing 300 sentence pairs per language. Two complementary automatic metrics are employed: Bilingual Evaluation Understudy (BLEU) and Character n-gram F-Score (chrF), alongside an average accuracy score and a cross-language consistency dimension. Nsanku represents the most comprehensive LLM translation evaluation for Ghanaian languages conducted to date. Results show that gemini-2.5-flash achieves the highest overall average score of 26.88 (BLEU: 24.60, chrF: 29.16), followed by claude-sonnet-4-5 at 24.87 (BLEU: 22.46, chrF: 27.28) and gpt-4.1 at 23.20 (BLEU: 21.15, chrF: 25.24). Among open-weight models, kimi-k2-instruct-0905 leads at an average score of 20.87. A critical finding from the consistency analysis is that no model and no language reached the Leaders quadrant of high performance and high consistency simultaneously, indicating that current LLMs are not yet reliably usable for Ghanaian language translation at scale. Siwu achieved the highest per-language average score at 25.73 while Nkonya scored lowest at 11.65. Nsanku establishes a publicly available, community-extensible evaluation infrastructure for African language NLP research.
Authors: Yao-Shun Chuang, Tushti Mody, Uday Pratap Singh, Shirindokht Shiraz, Chun-Teh Lee, Ryan Brandon, Muhammad F Walji, Xiaoqian Jiang, Bunmi Tokede
Abstract: Clinical named entity recognition from dental progress notes is challenging because documentation is highly unstructured, domain-specific, and often privacy-sensitive. We developed a locally deployable framework that enables small language models to self-generate, verify, refine, and evaluate entity-specific prompts for extracting multiple clinical entities from dental notes. Using 1,200 annotated notes, we evaluated candidate open-weight models with multi-prompt ensemble inference and further adapted selected models using QLoRA-based supervised fine-tuning and direct preference optimization. Model performance varied substantially, highlighting the need for task-specific evaluation rather than reliance on generic benchmarks. Qwen2.5-14B-Instruct achieved the strongest baseline performance. After DPO, Qwen2.5-14B-Instruct and Llama-3.1-8B-Instruct achieved micro/macro F1 scores of 0.864/0.837 and 0.806/0.797, respectively. These findings suggest that automated prompt optimization combined with lightweight preference-based post-training can support scalable clinical information extraction using locally deployed small language models.
Authors: Achuth Chandrasekhar, Omid Barati Farimani, Radheesh Sharma Meda, Amir Barati Farimani
Abstract: Materials science workflows rely on structured and unstructured data from the vast body of available scientific literature. However, most of the experimental details remain buried in text, tables, graphs and figures. Thus, constructing databases that incorporate this data is a manual, time-consuming, and hard-to-scale process. Multimodal large language models have made it feasible to extract information from text and scientific figures with high speed and accuracy. This opens the possibility of an AI system that can create production-scale material databases. Material Database Agent (MDA) is a modular, multi-agent system architecture for converting research literature into structured databases. MDA accepts article PDFs as input, which are subsequently processed in parallel into markdown files and figures. Multiple sub-agents read these markdown files and figures in parallel to assemble sub-databases for each paper. These sub-databases are then compiled into a single tabular database by an agent. As opposed to using either a rule-based approach or a single-pass pipeline for extracting information, MDA is a specialized architecture for transforming the literature into a database in the field of materials science. More generally, this study provides a basis for positioning multimodal agentic information extraction as a viable means for constructing next-generation scientific databases from the primary literature.
Authors: Stefano Bann\`o, Kate Knill, Mark Gales
Abstract: Automated essay scoring (AES) research often relies on rank-based correlation metrics to validate analytic assessment. However, such metrics obscure both intrinsic intercorrelations among analytic dimensions that arise from the structure of writing proficiency itself and halo effects, whereby holistic impressions bleed into fine-grained component scores. As a result, high correlations may mask a system's true diagnostic behaviour. In this study, we propose a novel self-referential assessment evaluation framework that focuses on identifying intra-learner strengths and weaknesses rather than assessing inter-learner rankings. We conduct experiments on the publicly available ICNALE GRA, a uniquely dense second-language writing dataset annotated holistically and analytically by up to 80 trained raters. To obtain reliable reference scores, we apply two-facet Rasch modelling to calibrate rater severity and derive fair average scores across ten analytic aspects and holistic proficiency. We compare the analytic scoring performance of human operational raters and three large language models (LLMs) in a zero-shot setting. Our results show that LLMs tend to outperform single human raters in identifying relative weaknesses (negative feedback) across several proficiency aspects, while human raters remain stronger at identifying relative strengths (positive feedback). Overall, our findings highlight the limitations of rank-based evaluation for analytic assessment and demonstrate the value of intra-learner, profile-based methods for assessing and deploying LLMs in AES.
Authors: Ziping Ye, Gourab Dey, Christos Christodoulopoulos, Charith Peris, Anil Ramakrishna, Weitong Ruan, Aram Galstyan, Kai-Wei Chang, Rahul Gupta, Ninareh Mehrabi
Abstract: We introduce SWAN (Semantic Watermarking with Abstract Meaning Representation), a novel framework that embeds watermark signatures into the semantic structure of a sentence using Abstract Meaning Representation (AMR). In contrast to existing watermarking methods, which typically encode signatures by adjusting token selection preferences during text generation, SWAN embeds the signature directly in the sentence's semantic representation. As the signature is encoded at the semantic structure level, any paraphrase that preserves meaning automatically preserves the signature. SWAN is training-free: watermark injection is achieved by prompting an LLM to generate sentences guided by a selected AMR template while maintaining contextual coherence, and detection uses an off-the-shelf AMR parser followed by a simple one-proportion z-test. Empirical evaluation on the RealNews benchmark shows SWAN matches state-of-the-art detection performance on unaltered watermarked text, while significantly improving robustness against paraphrasing, increasing detection AUC by up to 13.9 percentage points compared to prior methods. These results demonstrate that SWAN's approach of anchoring watermarks in AMR semantic structures provides a simple, effective, and prompt-based method for robust text provenance verification under paraphrasing, opening new avenues for semantic-level watermarking research.
Authors: Zhi Xu, Yun Fu
Abstract: Causal reasoning in natural language requires identifying relevant variables, understanding their interactions, and reasoning about effects and interventions, often under noisy or ambiguous conditions. While large language models (LLMs) exhibit strong general reasoning abilities, they struggle to disentangle correlation from causation, particularly when observations are partially incorrect or irrelevant information is present. In this work, we introduce NoisyCausal, a new benchmark designed to evaluate causal reasoning under structured noise. Each instance is generated from a ground-truth causal graph and contextualized with a natural language scenario by injecting controllable forms of noise, such as irrelevant distractors, value perturbations, confounding, and partial observability. Moreover, we propose a modular reasoning framework that combines LLMs with explicit causal structure to address these challenges. Our method prompts the LLM to extract variables, construct a causal graph from context, and then reformulates the reasoning task as a structured prompt grounded in this graph. Rather than relying on statistical patterns alone, the LLM is guided by symbolic structure, enabling more interpretable and robust inference. Experimental results show that our method significantly outperforms standard prompting and reasoning baselines on NoisyCausal. Furthermore, it generalizes well to external benchmarks such as Cladder without task-specific tuning. Our findings highlight the importance of combining causal abstractions with language-driven reasoning to achieve faithful and robust causal understanding in LLMs.
Authors: Mikhail L. Arbuzov, Sisong Bei, Ziwei Dong, Dmitri Kalaev, Alexey A. Shvets
Abstract: We introduce Telegraph English (TE), a prompt-compression protocol that rewrites natural language into a symbol-rich, formally-structured dialect. Where token-deletion methods such as LLMLingua-2 train a classifier to delete low-importance tokens at a fixed ratio, TE performs a full semantic rewrite: it decomposes the input into atomic fact lines, substitutes verbose phrases with $\sim$40 logical and relational symbols, and lets the compression ratio adapt to each document's information density. A consequence of the line-structure rule is that compression and semantic chunking become the same operation -- each output line is an independently addressable fact, so the compressed representation is simultaneously a semantic index. We evaluate TE on 4{,}081 question-answer pairs from LongBench-v2 across five OpenAI models and two difficulty levels. At roughly 50\% token reduction, TE preserves 99.1\% accuracy on key facts with GPT-4.1 and outperforms LLMLingua-2 at matched compression ratios on every model and task tested. The gap widens on smaller models -- up to 11 percentage points on fine-detail tasks -- suggesting that explicit relational structure compensates for limited model capacity. We release the grammar specification, compression prompt, benchmark data, and reference implementation.
Authors: Ziqi Zhu, Adithya Suresh, Tomal Deb, Iman Abbasnejad
Abstract: Dialogue State Tracking (DST) requires precise extraction of structured information from multi-domain conversations, a task where Large Language Models (LLMs) struggle despite their impressive general capabilities. We present GEM (Graph-Enhanced Mixture-of-Experts), a novel framework that combines language models and graph-structured dialogue understanding with ReAct agent-based reasoning for superior DST performance. Our approach dynamically routes between specialized experts: a Graph Neural Network that captures dialogue structure and turn-level dependencies, and a finetuned T5-Small encoder-decoder for sequence modeling, coordinated by an intelligent router. For complex value generation tasks, we integrate ReAct agents that perform structured reasoning over dialogue context. On MultiWOZ 2.2, GEM achieves 65.19% Joint Goal Accuracy, substantially outperforming end-to-end LLM approaches (best: 38.43%) and surpassing state-of-the-art (SOTA) methods including TOATOD (63.79%), D3ST (58.70%), and Diable (56.48%). Our graph-enhanced mixture-of-experts architecture with ReAct integration demonstrates that combining structured dialogue representation with dynamic expert routing and agent-based reasoning provides a powerful paradigm for dialogue state tracking, achieving superior accuracy while maintaining computational efficiency through selective expert activation.
Authors: Bryan Li, William Walden, Yu Hou, Gabrielle Kaili-May Liu, Dawn Lawrie, Jame Mayfield, Eugene Yang, Chris Callison-Burch, Laura Dietz
Abstract: Evaluation of long-form, citation-backed reports has lately received significant attention due to the wide-scale adoption of retrieval-augmented generation (RAG) systems. Core to many evaluation frameworks is the use of atomic facts, or nuggets, to assess a report's coverage of query-relevant information attested in the underlying collection. While nuggets have traditionally been represented as short statements, recent work has used question-answer (QA) representations, enabling fine-grained evaluations that decouple the information need (i.e. the question) from the potentially diverse content that satisfies it (i.e. its answers). A persistent challenge for nugget-based evaluation is the need to manually curate sets of nuggets for each topic in a test collection -- a laborious process that scales poorly to novel information needs. This challenge is acute in cross-lingual settings, where information is found in multilingual source documents. Accordingly, we introduce DoGMaTiQ, a pipeline for generating high-quality QA-based nugget sets in three stages: (1) document-grounded nugget generation, (2) paraphrase clustering, and (3) nugget subselection based on principled quality criteria. We integrate DoGMaTiQ nuggets with AutoArgue -- a recent nugget-based evaluation framework -- to enable fully automatic evaluation of generated reports. We conduct extensive experiments on two cross-lingual TREC shared tasks, NeuCLIR and RAGTIME, showing strong rank correlations with both human-in-the-loop and fully manual judgments. Finally, detailed analysis of our pipeline reveals that a strong LLM nugget generator is key, and that the system rankings induced by DoGMaTiQ are robust to outlier systems. We facilitate future research in report evaluation by publicly releasing our code and artifacts at https://github.com/manestay/dogmatiq.
Authors: Zhipeng Song, Yizhi Zhou, Xiangyu Kong, Jiulong Jiao, Xuezhou Ye, Chunqi Gao, Xueqing Shi, Yuhang Zhou, Heng Qi
Abstract: Retrieval-Augmented Generation (RAG) depends on document ranking to provide useful evidence for generation, but conventional reranking methods mainly optimize query-document relevance rather than generation usefulness. A relevant document may still introduce noise, while a lower-ranked document may better reduce the generator's uncertainty. We propose CAR (Confidence-Aware Reranking), a query-guided, training-free, and plug-and-play reranking framework that uses generator confidence change as a document usefulness signal. CAR estimates confidence through the semantic consistency of multiple sampled answers under query-only and query-document conditions. Documents that significantly increase confidence are promoted, those that decrease confidence are demoted, and uncertain cases preserve the baseline order, while a query-level gate avoids unnecessary intervention on already confident queries. Experiments on four BEIR datasets show that CAR consistently improves NDCG@5 across sparse and dense retrievers, LLM-based and supervised rerankers, and four LLM backbones. Notably, CAR improves the YesNo reranker by 25.4 percent on average under Contriever retrieval, and its ranking gains strongly correlate with downstream generation F1 improvements, achieving Spearman rho = 0.964.
Authors: Zhenliang Zhang, Wenqing Wang, Yong Hu, Yaming Yang, Jiaheng Gao, Chen Shen, Xiaojun Wan
Abstract: Long-Text Understanding (LTU) at million-token scale requires balancing reasoning fidelity with computational efficiency. Frontier long-context LLMs can process millions of token contexts end-to-end, but they suffer from high token consumption and attention dilution. In parallel, specialized LTU agents often sacrifice fidelity through task-agnostic abstractions like graph construction or indexing. We identify a key insight for LTU: query-relevant information is typically sparse relative to the full document, so effective reasoning should rely on a query-sufficient subset rather than the entire context. To address this, we propose SCOUT, a new paradigm for LTU that shifts from passive processing to active information foraging. It treats the document as an explorable environment and answers from a compact, provenance-grounded epistemic state. Guided by state-level gap diagnosis, SCOUT adaptively alternates between coarse-to-fine exploration and anchored state updates that progressively contract its epistemic state toward query sufficiency. Experiments show that SCOUT matches state-of-the-art proprietary models while reducing token consumption by up to 8x. Moreover, SCOUT remains stable as context length scales, substantially alleviating the practical cost-performance trade-off.
Authors: Jinju Kim, Haeji Jung, Youjeong Roh, Jong Hwan Ko, David R. Mortensen
Abstract: Low-resource language varieties used by specific groups remain neglected in the development of Multilingual Language Models. A great deal of cross-lingual research focuses on inter-lingual language transfer which strives to align allied varieties and minimize differences between them. However, for low-resource varieties, linguistic dissimilarity is also an important cue allowing generalization to unseen varieties. Unlike prior approaches, we propose a two-stage Language Generalization framework that focuses on capturing variety-specific cues while also exploiting rich overlap offered by high-resource source variety. First, we propose TOPPing, a source-selection method specifically designed for low-resource varieties. Second, we suggest a lightweight VACAI-Bowl architecture that learns variety-specific attributes with one branch while a parallel branch captures variety-invariant attributes using adversarial training. We evaluate our framework on structural prediction tasks, which are among the few tasks available, as proxy for performance on other downstream tasks. Using VACAI-Bowl with TOPPing yields an average 54.62% improvement in the dependency parsing task, which serves as a proxy for performance on other downstream tasks across 10 low-resource varieties.
Authors: Zongqi Cui, Baihan Lin
Abstract: Negotiation agents must infer what their counterpart values, update those beliefs over dialogue turns, and choose actions under uncertainty. End-to-end large language models (LLMs) can imitate negotiation dialogue, but their opponent beliefs are usually implicit and difficult to inspect. We propose BOND (Bayesian Opponent-belief Negotiation Distillation), a framework for auditable negotiation. BOND consists of an LLM-based Bayesian teacher that scores dialogue contexts against the six possible opponent priority orderings, updates a posterior over those orderings, and uses the posterior for menu-based decision making, as well as a smaller 8B student language model that emits both negotiation actions and normalized posterior beliefs as tagged text. In the CaSiNo negotiation dataset, BOND outperforms the state-of-the-art and achieves mean Brier score 0.085 over opponent-priority posteriors. The distilled student preserves much of this belief signal, achieving Brier 0.114, below the uniform six-ordering reference of 5/36, approximately 0.139. Compared with a 70B structured-CoT baseline, the significantly smaller 8B student model yields substantially better elicited posterior calibration. We further showcase auditability through posterior trajectories, belief-versus-policy error decomposition, and posterior-prefix interventions. These diagnostics reveal that distillation preserves a scoreable belief report more strongly than causal belief-conditioned control, making weak belief-action coupling visible, not hidden.
Authors: Ivan Bondarenko, Roman Derunets, Oleg Sedukhin, Mikhail Komarov, Ivan Chernov, Mikhail Kulakov
Abstract: We present our winning system for Task~B (generation with reference passages) in SemEval-2026 Task~8: MTRAGEval. Our method is a heterogeneous ensemble of seven LLMs with two prompting variants, where a GPT-4o-mini judge selects the best candidate per instance. We ranked 1st out of 26 teams, achieving a conditioned harmonic mean of 0.7827 and outperforming the strongest baseline (gpt-oss-120b, 0.6390). Ablations show that diversity in model families, scales, and prompting strategies is essential, with the ensemble consistently beating any single model. We also introduce Meno-Lite-0.1, a 7B domain-adapted model with a strong cost--performance trade-off, and analyse MTRAGEval, highlighting annotation limitations and directions for improvement. Our code is publicly available: https://github.com/RaguTeam/ragu_mtrag_semeval
Authors: Qiming Bao, Juho Leinonen, Paul Denny, Michael J. Witbrock
Abstract: Direct Preference Optimization (DPO), the efficient alternative to PPO-based RLHF, falls short on knowledge-intensive generation: standard preference signals from human annotators or LLM judges exhibit a systematic verbosity bias that rewards fluency over logical correctness. This blindspot leaves a logical alignment gap -- SFT models reach NLI entailment of only 0.05-0.22 despite producing fluent text. We propose RLearner-LLM with Hybrid-DPO: an automated preference pipeline that fuses a DeBERTa-v3 NLI signal with a verifier LLM score, removing human annotation while overcoming the "alignment tax" of single-signal optimization. Evaluated across five academic domains (Biology, Medicine, Law) with three base architectures (LLaMA-2-13B, Qwen3-8B, Gemma 4 E4B-it), RLearner-LLM yields up to 6x NLI improvement over SFT, with NLI gains in 11 of 15 cells and consistent answer-coverage gains. On Gemma 4 E4B-it (4.5B effective params), Hybrid-DPO lifts NLI in four of five domains (+11.9% to +2.4x) with faster inference across all five, scaling down to compact base models without losing the alignment-tax mitigation. Our Qwen3-8B RLearner-LLM wins 95% of pairwise comparisons against its own SFT baseline; GPT-4o-mini in turn wins 95% against our concise output -- alongside the 69% win the same judge gives a verbose SFT over our DPO model, this replicates verbosity bias on a frontier comparator and motivates logic-aware metrics (NLI, ACR) over LLM-as-a-judge for knowledge-intensive generation.
Authors: Yepeng Weng, Qiao Hu, Takehisa Yairi
Abstract: Speculative decoding accelerates Large Language Models via draft-then-verify, where verification can be framed as an Optimal Transport (OT) problem. Existing approaches typically handle multi-draft and multi-step aspects in isolation, applying either flat OT to single-step drafts or per-token rejection sampling to tree-structured candidates. This separation leaves the joint regime (where multi-step dependencies meet multi-draft branching) poorly optimized, as local verification rules fail to exploit the coupling between horizontal and vertical dimensions of candidate trees. In this paper, we propose a unified perspective that casts tree-based verification as a conditional OT problem. Our key insight is that vertical dependencies can be abstracted through prefix acceptance probabilities, which act as dynamic scaling factors to actively guide horizontal draft selection. Based on this principle, we introduce UniVer, a verification algorithm that jointly optimizes across tree levels by composing local optimal transport plans under prefix constraints. We prove that UniVer remains lossless and achieves the optimal acceptance rate under the proposed conditional framework. Extensive experiments across different tasks and models demonstrate that UniVer improves acceptance length by 4.2% to 8.5% over standard recursive rejection sampling without replacement, while maintaining exact distributional alignment with the target model.
Authors: Brielen Madureira, Andreas Niekler, Marc Keuschnigg, Mariana Madruga de Brito
Abstract: Media coverage influences disaster response, yet the drivers of international media attention to local events remain unevenly understood. Brazil offers a compelling case: some of its natural and technological disasters occasionally hit the international headlines. However, systematic analyses of what makes these events be discussed abroad are still missing. Addressing this gap requires representative, validated and country-specific news datasets. This paper presents a peak analysis of 2k news about Brazilian fires and landslides in German newspapers from 2000 to 2024. Using time series segmentation to detect news event peaks, we examine the extent to which they can be temporally aligned with observations in national and global disaster databases.
Authors: Mullosharaf K. Arabov
Abstract: This paper presents the first benchmark for the task of automatic part-of-speech (POS) tagging for the Tajik language. Despite the existence of multilingual language models demonstrating high effectiveness for many of the world's languages, their capacity for grammatical analysis of Tajik has remained unexplored until now. The aim of this study is to fill this gap through a systematic comparison of classical neural network architectures and modern multilingual transformers. Experiments were conducted on the TajPersParallel corpus, a parallel lexical resource comprising approximately 44,000 dictionary entries. Due to the absence of full-fledged example sentences in the current version of the corpus, the task was performed at the level of isolated lexical units, representing a challenging case of context-independent classification. The study compares the following architectures: a recurrent BiLSTM-CRF model, as well as multilingual models XLM-RoBERTa (large), mBERT, ParsBERT (Persian), and ruBERT (Russian), adapted using the parameter-efficient fine-tuning method LoRA. The testing results showed that the best performance is achieved by the mBERT + LoRA model (macro F1-score = 0.11, weighted F1-score = 0.62). It was established that in the absence of syntactic context, all models experience significant difficulty in resolving morphological ambiguity, successfully classifying primarily high-frequency classes ("noun," "adjective") while demonstrating zero effectiveness for rare function words. Zero-shot evaluation revealed the greatest typological proximity of Tajik to Persian (ParsBERT) and Russian (ruBERT). The obtained results form a foundation for further research and development in the field of automatic processing of the Tajik language.
Authors: Mullosharaf K. Arabov
Abstract: The Tajik language, written in Cyrillic script, remains severely under-resourced in terms of publicly available natural language processing (NLP) toolkits, hindering both linguistic research and applied development. This paper introduces TajikNLP, an open-source Python library that provides the first comprehensive pipeline for processing authentic Tajik text while preserving the original Cyrillic orthography. The library implements a modular architecture centered around a unified Doc object, enabling sequential application of components for cleaning, normalization, tokenization (including subword BPE), morphemic segmentation, part-of-speech tagging, stemming, lemmatization, and sentence splitting. A novel unified morphology engine is introduced, offering controlled and deep analysis modes that significantly improve handling of Tajik's agglutinative nominal and verbal inflections. The release further incorporates a lexicon-based sentiment analyser and pre-trained Word2Vec/FastText embeddings loaded directly from the Hugging Face Hub. To ensure reproducibility and facilitate future research, four accompanying linguistic datasets -- a POS-tagged corpus (52.5k entries), a sentiment lexicon (3.5k entries), a toponym gazetteer (5.6k entries), and a personal names dataset (3.8k entries) -- have been openly published under permissive licenses. The library's reliability is validated by an extensive test suite of 616 automated tests achieving 93% source code coverage. TajikNLP thus establishes a foundational technological infrastructure for Tajik language processing, lowering the barrier to entry for both academic and industrial applications in low-resource Cyrillic-script environments.
Authors: Mingda Li, Rundong Lv, Xinyu Li, Weinan Zhang, Ting Liu
Abstract: Uncertainty quantification (UQ) is an important technique for ensuring the trustworthiness of LLMs, given their tendency to hallucinate. Existing state-of-the-art UQ approaches for free-form generation rely heavily on sampling, which incurs high computational cost and variance. In this work, we propose the first gradient-based UQ method for free-form generation, SemGrad, which is sampling-free and computationally efficient. Unlike prior gradient-based methods developed for classification tasks that operates in parameter space, we propose to consider gradients in semantic space. Our method builds on the key intuition that a confident LLM should maintain stable output distributions under semantically equivalent input perturbations. We interpret the stability as the gradients in semantic space and introduce a Semantic Preservation Score (SPS) to identify embeddings that best capture semantics, with respect to which gradients are computed. We further propose HybridGrad, which combines the strengths of SemGrad and parameter gradients. Experiments demonstrate that both of our methods provide efficient and effective uncertainty estimates, achieving superior performance than state-of-the-art methods, particularly in settings with multiple valid responses.
Authors: Yifei Yuan, Luis Salamanca, Sophia Schlosser, Laurence Brandenberger
Abstract: Approximating the ideological position of Members of Parliament (MPs) is a fundamental task in political science, helping researchers understand legislative behavior, party alignment, and policy preferences. While Large Language Models (LLMs) have shown promising results in estimating MPs' ideological stances, there are more actors and elements in the parliamentary system, and relations between them, that could provide a wider and more informative picture. However, due to the complexity of integrating them in the prediction task, these additional elements are generally ignored. In this work, we propose an LLM framework, PG-RAG, that implements a retrieval-augmented generation pipeline: it first queries a political knowledge graph (KG) and then integrates the resulting graph-structured information into the context. This allows for capturing both textual semantics and inter-MP relationships, another relevant information source in any parliamentary system. We evaluate the approach on the task of ideology prediction, using data from a Swiss parliamentary dataset. When comparing graph-augmented models against several state-of-the-art baselines, the results demonstrate that incorporating this enriched information, which encodes information about different entities and relations, improves prediction performance. These results help to highlight the value of domain-specific relational information in modeling political behavior.
Authors: Shuai-long Lei, Xiaobin Zhu, Jiarui Liang, Guoxi Sun, Zhiyu Fang, Xu-Cheng Yin
Abstract: Temporal knowledge graph (TKG) reasoning aims to predict future events from historical facts. A key challenge lies in jointly capturing two sources of predictive information in TKGs: historical evidence and evolutionary dynamics. However, existing methods typically focus on only one of these sources, which limits the ability to fully exploit the complementary predictive signals in TKGs. To address this, we propose CHE-TKG, a novel collaborative dual-view learning framework for TKG reasoning. CHE-TKG explicitly separates and jointly models historical evidence and evolutionary dynamics, aiming to learn and exploit their complementary predictive signals. Specifically, CHE-TKG constructs a historical evidence graph to capture long-term structural regularities and stable relational constraints, alongside an evolutionary dynamics graph to model temporal transitions and recent changes, with dedicated encoders for each view. We further employ relation decomposition and a contrastive alignment objective to better capture the predictive signals across the two views. Extensive experiments demonstrate that CHE-TKG achieves state-of-the-art performance on multiple benchmarks.
Authors: Aofan Liu, Jingxiang Meng
Abstract: When the substantive content of a request is rewritten, do large language models still answer in the format the original task asked for? We find that they often do not, even at temperature zero. On a 150-query evaluation over five compact 2025-era LLMs and four task types, we observe a systematic failure mode we call prompt-variant output-mode collapse: when a closed-form prompt asks for a bare label or a single choice token, content-preserving prompt variants can push the model into conversational prose, the requested format dissolves, and exact-match evaluation pipelines silently misjudge the result. To make this measurable, we release PARACONSIST, a 900-prompt benchmark of 150 base queries with five lexical, syntactic, and semantic-expansion prompt variants each, and a Semantic Consistency Score that decomposes prompt-variant robustness into answer consistency, sentence-BERT semantic similarity, and length stability. Under a whole-word answer-set match, only ~22% of closed-form variant responses preserve the ground-truth label inside their output, while ~78% drift away from the answer space entirely. In our pool, the dominant predictor of collapse is task structure rather than model identity, with model differentiation jointly carried by answer consistency and length stability. Robustness audits should therefore track response-mode preservation as a first-class reliability target alongside answer accuracy.
Authors: Yaxun Dai, Baolin Sun, Junying Wang, Pengfei Wang, Yingqi Gao, Xuemei Dong, Mengdie Chu, Xiang Qi, Pingfu Chao
Abstract: Tool-integrated Text-to-SQL parsing has emerged as a promising paradigm, framing SQL generation as a sequential decision-making process interleaved with tool execution. However, existing reinforcement learning approaches mainly rely on coarse-grained outcome supervision, resulting in a fundamental credit assignment problem: models receive the same reward for any trajectory that yields the correct answer, even when intermediate steps are redundant, inefficient, or erroneous. Consequently, models are encouraged to explore suboptimal reasoning spaces, limiting both efficiency and generalization. To address this problem, we propose FineStep, a novel framework for step-level credit assignment in tool-augmented Text-to-SQL. First, we introduce a reward design with independent process rewards to alleviate the signal sparsity of outcome supervision. Next, we present a step-level credit assignment mechanism to precisely quantify the value of each reasoning step. Finally, we develop a policy optimization method based on step-level advantages for efficient updates. Extensive experiments on BIRD benchmarks show that FineStep achieves state-of-the-art performance and reduces redundant tool interactions, with a 3.25% average EX gain over GRPO at the 4B scale.
Authors: Venkat Srinivasan, Vishaal Jatav, Anushka Chandrababu, Geetika Sharma
Abstract: Transformer based pre-trained large language models have become ubiquitous. There is increasing evidence to suggest that even with large scale pre-training, these models do not capture complete compositional context and certainly not, the full human analogous context. Besides, by the very nature of the architecture, these models hallucinate, are difficult to maintain, are not easily interpretable and require enormous compute resources for training and inference. Here, we describe Gyan, an explainable language model based on a novel non-transformer architecture, without any of these limitations. Gyan achieves SOTA performance on 3 widely cited data sets and superior performance on two proprietary data sets. The novel architecture decouples the language model from knowledge acquisition and representation. The model draws on rhetorical structure theory, semantic role theory and knowledge-based computational linguistics. Gyan's meaning representation structure captures the complete compositional context and attempts to mimic humans by expanding the context to a 'world model'. AI model adoption critically depends on trust and transparency especially in mission critical use cases. Collectively, our results demonstrate that it is possible to create models which are trustable and reliable for mission critical tasks. We believe our work has tremendous potential for guiding the development of transparent and trusted architectures for language models.
Authors: Ge Lei, Samuel J. Cooper
Abstract: Large language models are increasingly used as surrogate models for low-data optimization, but their optimizer-facing prediction and its uncertainty remain poorly understood. We study the surrogate belief elicited from an LLM under sparse observations, showing that it depends strongly on prompt text and query protocol. We introduce an uncertainty-alignment criterion that measures whether model uncertainty tracks residual ambiguity among sample-consistent functions. Across controlled inference tasks and Bayesian optimization studies, we find that structural prompts act as effective priors, POINTWISE and JOINT querying induce different beliefs, and sequential evidence leads to non-monotonic, order-sensitive confidence updates. These effects change downstream acquisition decisions and regret, showing that elicitation protocol is part of the LLM surrogate specification, not a formatting detail.
Authors: Haotian Xia, Hao Peng, Yunjia Qi, Xiaozhi Wang, Bin Xu, Lei Hou, Juanzi Li
Abstract: Story generation aims to automatically produce coherent, structured, and engaging narratives. Although large language models (LLMs) have significantly advanced text generation, stories generated by LLMs still diverge from human-authored works regarding complex narrative structure and human-aligned preferences. A key reason is the absence of effective modeling of human story preferences, which are inherently subjective and under-explored. In this work, we systematically evaluate the modeling of human story preferences and introduce StoryRMB, the first benchmark for assessing reward models on story preferences. StoryRMB contains $1,133$ high-quality, human-verified instances, each consisting of a prompt, one chosen story, and three rejected stories. We find existing reward models struggle to select human-preferred stories, with the best model achieving only $66.3\%$ accuracy. To address this limitation, we construct roughly $100,000$ high-quality story preference pairs across diverse domains and develop StoryReward, an advanced reward model for story preference trained on this dataset. StoryReward achieves state-of-the-art (SoTA) performance on StoryRMB, outperforming much larger models. We also adopt StoryReward in downstream test-time scaling applications for best-of-n (BoN) story selection and find that it generally chooses stories better aligned with human preferences. We will release our dataset, model, and code to facilitate future research. Related code and data are available at https://github.com/THU-KEG/StoryReward.
Authors: Eduardo Santos, Juliana Carvalho, C\'esar Renn\'o-Costa
Abstract: This paper presents the development and validation of an eye-tracking dataset designed to investigate how second-language (L2) learners process idiomatic expressions. While native speakers often rely on direct retrieval of figurative meanings, L2 speakers frequently adopt a literal-first approach, which incurs measurable cognitive costs. This resource captures these costs through ocular metrics recorded from Portuguese L1 speakers of English across all CEFR proficiency levels (A1-C2). Although the study uses entry-level 60 Hz hardware (Tobii Pro Spark), we demonstrate that this sampling rate provides sufficient data density to detect macro-cognitive events such as fixations and regressions in reading. Preliminary analysis validates the dataset by revealing a strong inverse correlation between language proficiency and regressive eye movements. Integrated into the MIA (Modeling Idiomaticity in Human and Artificial Language Processing) initiative, this dataset serves as a cognitively grounded benchmark for evaluating both human processing models and the alignment of large language models with human-like figurative understanding.
Authors: Maria Luongo, Davide Marocco, Nicola Milano
Abstract: Recent advances in natural language processing have enabled increasingly accurate estimation of psychological traits from language. However, most existing approaches rely on supervised models trained to predict questionnaire scores, limiting interpretability and generalizability across contexts. The present study introduces a theory-driven and fully unsupervised framework for measuring psychological states directly from natural language using semantic projection. Psychological constructs were operationalized as interpretable semantic axes derived from lexical anchors and items from validated clinical scales assessing depression, anxiety, and worry. Participants textual responses were embedded using Sentence-BERT and projected onto these axes to generate continuous psychological scores across multiple response formats, including selected words, generated words, phrases, and free-text responses. Projection scores were evaluated through correlations with standardized clinical measures , split-half reliability analyses, attenuation corrections, distributional similarity using Wasserstein distance, and comparisons with lexicon-based sentiment analysis (VADER). Results showed strong associations between projection scores and clinical measures, particularly for structured formats such as selected words, written words, and phrases. Free-text responses produced weaker results when analyzed as whole texts, but performance improved substantially when sentence-level aggregation strategies were applied. These findings support semantic projection as an interpretable and scalable alternative to supervised language models for psychological assessment and highlight the importance of response format and text-processing strategies in language-based mental health measurement.
Authors: Enrico Maria Fenoaltea, Filippo Santoro, Giordano De Marzo, Segun Taofeek Aroyehun, Andrea Tacchella
Abstract: Forecasting innovation, intended as the emergence of new technological combinations, is a fundamental challenge for science and policy. We show that forthcoming combinations leave an early trace in the collective language of patents, with predictive signals detectable even decades in advance. We show that signal is not attributable to any single inventor, but emerges as a collective shift in how technologies are described across thousands of patents. To this end, we introduce TechToken, a transformer-based model that treats technologies, classified by International Patent Classification codes, as words in its vocabulary, learning the language of technologies by embedding these codes during fine-tuning. We define context similarity between code embeddings as a measure of linguistic convergence and show that it accurately predicts first technological combinations. TechToken also improves general representation quality, outperforming state-of-the-art models across different patent-related tasks.
Authors: Tanty Widiyastuti, Mayada, Adisty Syawalda Ariyanto, Luluk Muthoharoh, Ardika Satria, Martin Clinton Tosima Manullang
Abstract: This paper compares a PyCaret AutoML branch and a CNN-BiLSTM branch for binary hate speech detection on Indonesian Twitter using the HS label from the corpus of Ibrohim and Budi. Both branches share the same preprocessing pipeline so that the comparison reflects modelling differences rather than inconsistent data preparation. The conventional branch uses TF-IDF with a lexicon-based abusive-word count, whereas the neural branch learns dense token representations and captures both local phrase patterns and bidirectional context. The benchmark is built from the released 13,130-row annotation table, whose HS label yields a 58:42 class ratio. On the held-out split, CNN-BiLSTM achieves the best result with 83.8% accuracy, 79.8% precision, 82.7% recall, and 81.2% F1-score. Within the PyCaret branch, Random Forest is the strongest conventional model with 77.2% accuracy and 77.0% F1-score. The neural branch therefore improves accuracy by 6.6 points and F1-score by 4.2 points. Exploratory corpus analysis, learning curves, and confusion matrices show that the dataset is short-text, moderately imbalanced, and still difficult because many decisions depend on local lexical cues plus short contextual composition. The study concludes that PyCaret AutoML is an effective conventional benchmarking framework, whereas CNN-BiLSTM is the stronger end model for the reported benchmark setting.
Authors: Arnault Chatelain, \'Etienne Ollion, Qianwen Guan, Diandra Fabre, Lorraine Goeuriot, Emile Chapuis, Abdelkrim Beloued, Marie Candito, Nicolas Herv\'e, Didier Schwab
Abstract: This position paper argues that the under-representation of social science tasks in contemporary LLM benchmarks limits advances in both LLM evaluation and social scientific inquiry. Benchmarks -- standardized tools for assessing computational systems -- are pivotal in the development of artificial intelligence (AI), including large language models (LLMs). Benchmarks do more than measure progress -- they actively structure it, shaping reputations, research agendas, and commercial outcomes. Despite this central role, the social sciences are largely absent from mainstream evaluation frameworks, even though scholars in these fields generate dozens of rigorously annotated, context-sensitive datasets each year. Integrating this work into benchmark design could significantly improve the generalization and robustness of AI models. In turn, models trained on social scientific tasks would likely yield better performance on classic and contemporary tasks in disciplines as diverse as history, sociology, political science or economics. This is all the more pressing as these disciplines are quickly turning to LLMs for assistance. To address this gap, we introduce BenCSSmark, a benchmark composed of datasets annotated by computational social scientists. By integrating social scientific perspectives into benchmarking, BenCSSmark seeks to promote more robust, transparent, and socially relevant AI systems and to foster efficient collaboration.
Authors: Ridho Benedictus Togi Manik, Muhammad Aqil Ramadhan, Ihsan Maulana Yusuf, Luluk Muthoharoh, Ardika Satria, Martin Clinton Tosima Manullang
Abstract: The exponential expansion of digital commerce in Indonesia has significantly shifted consumer interactions toward video-centric social networks, particularly YouTube. Consequently, the sheer volume of unstructured, multi-contextual comments poses a tremendous challenge for manual sentiment tracking. This study investigates and constructs a predictive model for customer satisfaction leveraging the Extreme Gradient Boosting (XGBoost) architecture coupled with Term Frequency-Inverse Document Frequency (TF-IDF) vectorization. By utilizing a secondary dataset of YouTube comments retrieved from e-commerce review videos, the raw text underwent rigorous preprocessing to generate normalized numerical features. The experimental results demonstrate that the PyCaret-optimized machine learning framework delivers superior classification resilience. Beyond standard performance metrics, lexical evaluations and feature-importance mapping uncover a notable phenomenon: e-commerce discourse is heavily infiltrated by socio-political terminologies, which ultimately influence the polarity of audience satisfaction.
Authors: Vita Anggraini, Cintya Bella, Bastian, Luluk Muthoharoh, Ardika Satria, Martin C. T. Manullang
Abstract: The exponential growth of social media has created an urgent need for automated systems to analyze unstructured public sentiment in real time. This study compares a traditional Logistic Regression model using TF-IDF features with a deep learning Bidirectional Long Short-Term Memory (BiLSTM) architecture on a 10,000-tweet subset of the Sentiment140 dataset. Experimental results show that Logistic Regression outperformed BiLSTM, achieving an accuracy of 73.5% compared with 69.17%, while the deep learning model exhibited mild overfitting. These findings suggest that for medium-scale informal text data, classical machine learning with robust feature extraction can outperform more complex deep learning approaches. Finally, the trained models were integrated into an interactive web application using Streamlit and deployed on Hugging Face Spaces for public access.
Authors: Joshua Adler, Guy Zehavi
Abstract: Extraction at ingestion is the wrong primitive for agent memory: content discarded before the query is known cannot be recovered at retrieval time. We propose True Memory, a six-layer architecture that shifts the center of the system from a storage schema to a multi-stage retrieval pipeline operating over events preserved verbatim. The full system runs as a single SQLite file on commodity CPU with no external database, vector index, graph store, or GPU. On LoCoMo (1,540 questions across 10 multi-session conversations), True Memory Pro reaches 93.0% accuracy (3-run mean) against 61.4% for Mem0, 65.4% for Supermemory, approximately 71% for Zep, and 94.5% for EverMemOS under a matched gpt-4.1-mini answer model. On LongMemEval (500 questions), True Memory Pro reaches 87.8% (3-run mean). On BEAM-1M (700 questions at the 1-million-token scale), True Memory Pro reaches 76.6% (3-run mean), above the prior published result of 73.9% for Hindsight. A 56-configuration ablation shows a 1.3-percentage-point spread within the top-performing configuration family.
Authors: Hengyu Shi, Tianyang Han, Peizhe Wang, Zhiling Wang, Xu Yang, Junhao Su
Abstract: LLM post-training typically propagates task gradients through the full depth of the model. Although this end-to-end structure is simple and general, it couples task adaptation to full-depth activation storage, long-range backward dependencies and direct task-gradient access to pretrained representations. We argue that this full-depth backward coupling can be unnecessarily expensive and intrusive, particularly when post-training supervision is much narrower than pre-training. To this end, we propose \textbf{LoPT}: Local-Learning Post-Training, a simple post-training strategy that makes gradient reach an explicit design choice. LoPT places a single gradient boundary at the transformer midpoint: the second-half block learns from the task objective, while the first-half block is updated by a lightweight feature-reconstruction objective to preserve useful representations and maintain interface compatibility. LoPT shortens the task-induced backward path while limiting direct interference from narrow task gradients on early-layer representations. Extensive experiments demonstrate that LoPT achieves competitive performance with lower memory cost, higher training efficiency and better retention of pretrained capabilities. Our code is available at: https://github.com/HumyuShi/LoPT
Authors: Bingkun Zhao, Chenwei Zhang, Hao Peng
Abstract: Promotional language has been increasingly used to aid the communication of innovative ideas in science. Yet, less is known about its role in the context of technological innovation. Here, we use a validated and domain-diagnosed lexicon of 135 promotional words to study the association between promotional language and patent evaluation outcomes among 2.7 million USPTO patent applications. Our large-scale study reveals three unexpected findings. First, in contrast to scientific evaluation, we find that a higher frequency of promotional words is negatively associated with the probability of an application being (i) granted a patent, (ii) transferred ownership, and (iii) successfully appealed. This promotional penalty holds even after accounting for a range of confounding factors and is largely robust across different technological areas. Among matched samples, the difference in the success rate between the lowest and highest promotional density quintile is 5.5, 5.9, and 5.3 percentage points for patentability, transferability, and rejection reversal. Second, contrary to institutional skepticism, we show that promotional language is not a mask of weak technology, but objectively reflects the degree of combinatorial novelty and future citation impact. Third, digging into the mechanisms, we find that the tolerance to promotional framing is strongly moderated by human factors, with men and experienced examiners showing a higher acceptance of promotional narratives than women and novice examiners. By revealing an emerging paradox in the patent system, our study offers theoretical and practical implications for improving patent evaluation through more objective scrutiny of linguistic patterns in patent filings.
Authors: Ivan Kart\'a\v{c}, Krist\'yna Onderkov\'a, Jan Bronec, Zden\v{e}k Kasner, Mateusz Lango, Ond\v{r}ej Du\v{s}ek
Abstract: This paper describes our system submitted to SemEval-2026 Task 11: Disentangling Content and Formal Reasoning in Large Language Models. We present an efficient modular neuro-symbolic approach, combining a symbolic prover with small reasoning LLMs (4B parameters). The system consists of an LLM-based parser that translates natural language syllogisms to a first-order logic (FOL) representation, an automated theorem prover, and two optional modules: machine translation for multilingual inputs and a symbolic retrieval component for the identification of relevant premises. The system achieves competitive accuracy and relatively low content effect on most subtasks. Our ablations show that this approach outperforms LLM-based zero-shot baselines in this parameter size range, but also reveal limited multilingual capabilities of small LLMs. Finally, we include a discussion of the task's main ranking metric and analyze its limitations.
Authors: Mullosharaf K. Arabov, Svetlana S. Khaybullina
Abstract: This paper presents a comparative study of parameter-efficient fine-tuning (PEFT) methods, including LoRA and QLoRA, applied to the task of adapting large language models to the Bashkir language, a low-resource agglutinative language of the Turkic family. Experimental evaluation is conducted on a Bashkir text corpus of 71k documents (46.9M tokens) using models of various architectures: DistilGPT2, GPT-2 (base, medium), Phi-2, Qwen2.5-7B, DeepSeek-7B, and Mistral-7B. To improve the reliability of results, each configuration was trained with three different random seeds. The lowest perplexity on the test set was obtained for GPT-2 medium with full fine-tuning (3.34). Meanwhile, QLoRA applied to Mistral-7B (3.79) and Phi-2 (3.81) achieved comparable quality with over 40 times fewer trainable parameters. However, we also observed cases of significant quality degradation when using PEFT for certain architectures (e.g., DeepSeek-7B with rank 8, perplexity = 129.55), indicating that the outcome depends critically on the choice of the base model and its tokenizer. Additionally, a qualitative analysis of generated texts based on Bashkir prompts revealed that models with the best perplexity do not necessarily produce the most coherent outputs: QLoRA-tuned models generated monolingual Bashkir continuations, whereas the fully fine-tuned model with the lowest perplexity frequently switched to English. The results suggest that QLoRA on 7B-scale models offers an effective compromise between quality and computational cost for Bashkir. To ensure reproducibility, open data, code, and trained adapters will be released upon acceptance.
Authors: Minjie Qiang, Mingming Zhang, Xiaoyi Bao, Xing Fu, Yu Cheng, Weiqiang Wang, Zhongqing Wang, Ningtao Wang
Abstract: Foundation models have established unified representations for natural language processing, yet this paradigm remains largely unexplored for tabular data. Existing methods face fundamental limitations: LLM-based approaches lack retrieval-compatible vector outputs, whereas text embedding models often fail to capture tabular structure and numerical semantics. To bridge this gap, we first introduce the Tabular Embedding Benchmark (TabBench), a comprehensive suite designed to evaluate the tabular understanding capability of embedding models. We then propose TabEmbed, the first generalist embedding model that unifies tabular classification and retrieval within a shared embedding space. By reformulating diverse tabular tasks as semantic matching problems, TabEmbed leverages large-scale contrastive learning with positive-aware hard negative mining to discern fine-grained structural and numerical nuances. Experimental results on TabBench demonstrate that TabEmbed significantly outperforms state-of-the-art text embedding models, establishing a new baseline for universal tabular representation learning. Code and datasets are publicly available at https://github.com/qiangminjie27/TabEmbed and https://huggingface.co/datasets/qiangminjie27/TabBench.
URLs: https://github.com/qiangminjie27/TabEmbed, https://huggingface.co/datasets/qiangminjie27/TabBench.
Authors: Tzu-Mi Lin, Wataru Hirota, Tatsuya Ishigaki, Lung-Hao Lee, Chung-Chi Chen
Abstract: Aligning large language models with expert judgment is especially difficult in subjective evaluation tasks, where experts may disagree, rely on tacit criteria, and change their judgments over time. In this paper, we study expert alignment as a way to understand this difficulty. Using expert evaluations and follow-up questionnaires, we examine how different forms of expert information affect alignment and what this reveals about subjective judgment. Our findings show four consistent patterns. First, alignment difficulty varies substantially across experts, suggesting that expert evaluation styles differ widely in their distance from a model's prior behavior. Second, explicit criteria and reasoning do not always improve alignment, indicating that expert judgment is not fully captured by verbalized rules. Third, editing is sensitive to both the number and the identity of examples, with small numbers of edits providing useful but unstable gains. Fourth, alignment difficulty differs across evaluation dimensions: dimensions grounded more directly in proposal content are easier to align, while dimensions requiring external knowledge or value-based judgment remain harder. Taken together, these results suggest that expert alignment is difficult not only because of model limitations, but also because subjective evaluation is inherently heterogeneous, partly tacit, dimension-dependent, and temporally unstable.
Authors: Gayane Ghazaryan, Esra D\"onmez
Abstract: Reward models are a key component of large language model alignment, serving as proxies for human preferences during training. However, existing evaluations focus primarily on broad instruction-following benchmarks, providing limited insight into whether these models capture socially desirable preferences. As a result, important failures in social alignment can remain hidden. We extend reward-model benchmarking to four socially consequential domains: bias, safety, morality, and ethical reasoning. We introduce a framework that converts social evaluation datasets into pairwise preference data, leveraging gold labels where available and directional bias indicators otherwise. This enables us to test whether reward models prefer socially undesirable responses, and whether their preferences produce systematically biased distributions over selected outputs. Across five publicly available reward models and two instruction-tuned models used as reward proxies, we find substantial variation across domains, with no single model performing best overall. The models fall well short of strong social intelligence: they often prefer socially undesirable options, and their preferences produce systematically biased distributions. Moreover, stronger bias avoidance can reduce sensitivity to context, revealing a key alignment trade-off between avoiding biased outcomes and preserving contextual faithfulness. These findings show that standard reward benchmarks are insufficient for assessing social alignment and highlight the need for evaluations that directly measure the social preferences encoded in reward models.
Authors: Gijs van Dijk
Abstract: We propose a lightweight and single-pass uncertainty quantification method for detecting hallucinations in Large Language Models. The method uses attention matrices to estimate uncertainty without requiring repeated sampling or external models. Specifically, we measure the Kullback-Leibler divergence between each attention head's distribution and a uniform reference distribution, and use these features in a logistic regression probe. Across multiple datasets, task types, and model families, attention divergence is highly predictive of answer correctness and performs competitively with existing uncertainty estimation methods. We find that this signal is concentrated in middle layers and on factual tokens such as named entities and numbers, suggesting that attention dynamics provides an efficient and interpretable white-box signal of model uncertainty.
Authors: Yan Zhou
Abstract: We identify and prove a fundamental trade-off governing long-sequence models: no model can simultaneously achieve (i) per-step computation independent of sequence length (Efficiency), (ii) state size independent of sequence length (Compactness), and (iii) the ability to recall a number of historical facts proportional to sequence length (Recall). We formalize this trade-off within an Online Sequence Processor abstraction that unifies Transformers, state space models, linear recurrent networks, and their hybrids. Using the Data Processing Inequality and Fano's Inequality, we prove that any model satisfying Efficiency and Compactness can recall at most O(poly(d)/log V) key-value pairs from a sequence of arbitrary length, where d is the model dimension and V is the vocabulary size. We classify 52 architectures published before March 2026 into the triangle, showing that each achieves at most two of the three properties and that hybrid architectures trace continuous trajectories in the interior. Experiments on synthetic associative recall tasks with five representative architectures validate the theoretical bound: empirical recall capacity lies strictly below the information-theoretic limit, and no architecture escapes the triangle.
Authors: Hubert Plisiecki, Sabina Siudaj, Kacper Dudzic, Anna Sterna, Maciej Gorski, Karolina Drozdz, Marcin Moskalewicz
Abstract: We administer 45 validated psychometric questionnaires to 50 large language models (LLMs) to identify the dimensions along which LLMs differ psychometrically. Using Supervised Semantic Differential (SSD), we find that the primary axis of between-model variance separates items describing phenomenally rich experience, including embodied sensation, felt affect, inner speech, imagery, and empathy, from items describing stimulus-driven behavioral reactivity ($R^2_{adj}=.037$, $p<.0001$). To test this hypothesis at the item level, we introduce the Pinocchio score ($\pi_i$), the ratio of inter-model response variance under neutral prompting to that under a human-simulation prompt, as an annotation-free measure of each item's experiential demand. $\pi_i$ predicts condition-induced shifts in primary factor loading magnitudes ($\rho=-.215$, $p<.0001$, $n=1292$--$1310$ items), confirming that between-model divergence on experiential items is structured rather than noisy. Applying PCA to per-model EFA scores across all questionnaires reveals one dominant dimension, the Pinocchio Axis ($\Pi$): the degree to which a model presents itself as a locus of phenomenal experience rather than a system of behavioral responses. This axis captures 47.1% of cross-questionnaire between-model variance in primary factor scores and converges with item-level Pinocchio scores ($r=.864$). Marked within-provider divergence across closely related model variants is consistent with post-training fine-tuning as a key contributor, supporting the interpretation that $\Pi$ reflects a training-shaped self-representational tendency governing how a model treats experiential language as self-applicable. The dominant axis of between-model psychometric variation is therefore not a conventional personality trait but a self-representational stance toward one's own nature as an experiencer.
Authors: Quintin Pope, Ajay Hayagreeve Balaji, Jacques Thibodeau, Xiaoli Fern
Abstract: We present an automated, contrastive evaluation pipeline for auditing the behavioral impact of interventions on large language models. Given a base model $M_1$ and an intervention model $M_2$, our method compares their free-form, multi-token generations across aligned prompt contexts and produces human-readable, statistically validated natural-language hypotheses describing how the models differ, along with recurring themes that summarize patterns across validated hypotheses. We evaluate the approach in synthetic setting by injecting known behavioral changes and showing that the pipeline reliably recovers them. We then apply it to three real-world interventions, reasoning distillation, knowledge editing and unlearning, demonstrating that the method surfaces both intended and unexpected behavioral shifts, distinguishes large from subtle interventions, and does not hallucinate differences when effects are absent or misaligned with the prompt bank. Overall, the pipeline provides a statistically grounded and interpretable tool for post-hoc auditing of intervention-induced changes in model behavior.
Authors: Nicholas S. Kersting, Vittorio Castelli, Chieh Ting Yeh, Xinzhu Wang, Saad Taame
Abstract: We introduce the **Concept Field** of a text corpus: a local drift field with pointwise uncertainty, estimated in sentence-embedding space from the deltas between consecutive sentences. Given a candidate sentence transition, we score its agreement with the field by $\zeta$, the mean absolute z-distance between the observed delta and the field's local Gaussian estimate. The score is black-box (no model internals), corpus-attributable (every score traces to nearby corpus sentences), and admits a direct probabilistic reading. We support the computation with the introduction of a **Vector Sequence Database (VSDB)** that stores embeddings together with sequence-position and next-delta metadata. We evaluate this approach on two large-scale settings: hallucination-style groundedness detection over the U.S. Code of Federal Regulations, and novelty detection over Project Gutenberg. Using controlled LLM-generated rewrites, Concept Fields achieve strong selective classification performance under a grounded / ungrounded / unsure triage policy, which unlike retrieval-centric baselines have similar coverage-risk behavior across both domains, supporting a probability-based interpretation that transfers across domains. We also sketch how divergence and curl of the Concept Field, computed on dense clusters, surface qualitatively meaningful semantic patterns (logic sources, sinks, and implicit topics), which we offer as hypothesis-generating rather than as a quantitative result. Concept Fields provide a fast, lightweight, and interpretable signal for groundedness and novelty, complementary to LLM-as-judge and white-box detectors.
Authors: Yucheng Ruan, Ling Huang, Qika Lin, Kai He, Mengling Feng
Abstract: Automated mental health prediction using textual data has shown promising results with deep learning and large language models. However, deploying these models in high-stakes real-world settings remains challenging, as existing approaches largely rely on semantic representations and often produce overconfident predictions under ambiguous, noisy, or shifted data. Moreover, most methods lack reliable uncertainty estimation, undermining trust in risk-sensitive mental health applications. To address these limitations, we formulate the task as a multi-view learning problem that integrates semantic information from encoder-only models with higher-level reasoning information from decoder-only models, where reasoning-aware representations and uncertainty modeling are obtained in a trustworthy manner. To ensure reliable fusion, we adopt an evidential learning framework based on Subjective Logic to explicitly model uncertainty and introduce an evidential fusion strategy that balances complementary views while discounting unreliable evidence. Benchmarking on three real-world datasets, Dreaddit, SDCNL, and DepSeverity, reports accuracies of 0.835, 0.731, and 0.751, respectively, demonstrating its potential for reliable mental health prediction. Additional experiments on robustness to noise and case studies for interpretability confirm that our proposed framework not only improves predictive performance but also provides trustworthy uncertainty estimates and human-understandable reasoning signals, making it suitable for risk-sensitive applications in mental health assessment.
Authors: Srikar Kashyap Pulipaka
Abstract: We present our system for SemEval-2026 Task 9: Multilingual Polarization Detection, a binary classification task spanning 22 languages. Our approach fine-tunes separate Gemma~3 models (12B and 27B parameters) per language using Low-Rank Adaptation (LoRA), augmented with synthetic data generated by a large language model (LLM). We employ three synthetic data strategies (direct generation, paraphrasing, and contrastive pair creation) using GPT-4o-mini, with a multi-stage quality filtering pipeline including embedding-based deduplication. We find that per-language threshold tuning on the development set yields 2 to 4\% F1 improvements without retraining. We also use weighted ensembles of 12B and 27B model predictions with per-language strategy selection. Our final system achieves a mean macro-F1 of 0.811 across all 22 languages, ranking 2nd overall of the participating teams, with 1st place finishes in 3 languages and top-3 in 8 languages. We also find that alternative architectures (XLM-RoBERTa, Qwen3) that showed strong development set performance suffered 30 to 50\% F1 drops on the test set, highlighting the importance of generalization.
Authors: Mina Gabriel
Abstract: Self-consistency detects hallucinations by generating multiple sampled answers to a question and measuring agreement, but this requires repeated decoding and can be sensitive to lexical variation. Semantic self-consistency improves this by clustering sampled answers by meaning using natural language inference, but it adds both sampling cost and external inference overhead. We show that first-token confidence, phi_first, computed from the normalized entropy of the top-K logits at the first content-bearing answer token of a single greedy decode, matches or modestly exceeds semantic self-consistency on closed-book short-answer factual question answering. Across three 7-8B instruction-tuned models and two benchmarks, phi_first achieves a mean AUROC of 0.820, compared with 0.793 for semantic agreement and 0.791 for standard surface-form self-consistency. A subsumption test shows that phi_first is moderately to strongly correlated with semantic agreement, and combining the two signals yields only a small AUROC improvement over phi_first alone. These results suggest that much of the uncertainty information captured by multi-sample agreement is already available in the model's initial token distribution. We argue that phi_first should be reported as a default low-cost baseline before invoking sampling-based uncertainty estimation.
Authors: Yingshan Susan Wang, Linlu Qiu, Zhaofeng Wu, Roger P. Levy, Yoon Kim
Abstract: Grammaticality and likelihood are distinct notions in human language. Pretrained language models (LMs), which are probabilistic models of language fitted to maximize corpus likelihood, generate grammatically well-formed text and discriminate well between grammatical and ungrammatical sentences in tightly controlled minimal pairs. However, their string probabilities do not sharply discriminate between grammatical and ungrammatical sentences overall. But do LMs implicitly acquire a grammaticality distinction distinct from string probability? We explore this question through studying internal representations of LMs, by training a linear probe on a dataset of grammatical and (synthetic) ungrammatical sentences obtained by applying perturbations to a naturalistic text corpus. We find that this simple grammaticality probe generalizes to human-curated grammaticality judgment benchmarks and outperforms LM probability-based grammaticality judgments. When applied to semantic plausibility benchmarks, in which both members of a minimal pair are grammatical and differ in only plausibility, the probe however performs worse than string probability. The English-trained probe also exhibits nontrivial cross-lingual generalization, outperforming string probabilities on grammaticality benchmarks in numerous other languages. Additionally, probe scores correlate only weakly with string probabilities. These results collectively suggest that LMs acquire to some extent an implicit grammaticality distinction within their hidden layers.
Authors: Bryan Cheng, Jasper Zhang
Abstract: Understanding how large language models encode task identity from few-shot demonstrations is a central open problem in mechanistic interpretability. Prior work uses linear probing to localize task representations, reporting high classification accuracy at specific layers. We reveal a striking dissociation: probing accuracy completely fails to predict causal importance. Single-position activation intervention achieves 0% task transfer across all 28 layers of Llama-3.2-3B-despite 100% probing accuracy at those same positions. This null result is itself a key finding, demonstrating that task encoding is fundamentally distributed. Multi-position intervention-replacing activations at all demonstration output tokens simultaneously-achieves up to 96% transfer (N=50, 95% CI: [87%, 99%]) at layer 8, pinpointing for the first time the causal locus of ICL task identity. We establish the generality of these findings across four models spanning three architecture families (LLaMA, Qwen, Gemma), discovering a universal intervention window at ~30% network depth. Causal tracing uncovers an asymmetric architecture: the query position is strictly necessary (53-100% disruption) while no individual demonstration position is necessary (0% disruption)-resolving a key ambiguity in prior accounts. Crucially, transfer depends on internal representation compatibility, not surface similarity (r=-0.05 vs r=0.31), ruling out trivial explanations. These results establish the distributed template hypothesis: ICL task identity is encoded as output format templates distributed across demonstration tokens, fundamentally reshaping our understanding of how in-context learning operates.
Authors: Chris Sainsbury, Feng Dong, Andreas Karwath
Abstract: Sparse autoencoders (SAEs) have been applied to large language models and protein language models, but not systematically to electronic health record (EHR) foundation models. We train TopK SAEs on FlatASCEND, a 14.5-million-parameter autoregressive clinical sequence model, at all 10 residual stream extraction points on INSPECT (outpatient) and MIMIC-IV (ICU). SAE decomposition reveals progressive abstraction across transformer depth: layer-0 features are near-perfect token detectors (45.7% singleton), while layer-6 features span approximately 30 token types across multiple clinical categories (0.5% singleton). Under full-sequence simple linear probes, SAE features outperform dense representations for discrete event prediction (mortality) while dense representations outperform for continuous magnitude prediction (length of stay) - a probe-level representational phenomenon that does not extend to clinically relevant leakage-safe windows, where dense representations match or exceed SAE features across all tested settings (eICU-CRD 48-hour AUC: SAE 0.871 versus dense 0.880; base model zero-shot, SAE dictionaries trained on eICU activations; MIMIC-IV: 0.836 versus 0.914; INSPECT 1-year/3-year: 0.697 versus 0.800). A delta-mode intervention method reduces SAE perturbation noise by 86x, enabling cleaner feature-level experiments, though the resulting perturbation effects are larger than random controls in 3 of 4 conditions but not formally significant. Feature reproducibility across random seeds is 21%, and individual features should be interpreted as illustrative rather than stable.
Authors: Sihao Liu, YuFan Xiong, Zhonghua Jiang, Zhaode Wang, chengfei lv Shengyu Zhang
Abstract: Multimodal Large Language Models face severe challenges in computational efficiency and memory consumption due to the substantial expansion of the visual KV cache when processing long visual contexts. Existing KV cache compression methods typically rely on the "persistence of importance" hypothesis to prune tokens. However, this approach proves fragile in multimodal settings due to two key issues: 1) Visual tokens display "deferred importance," initially exhibiting low salience but becoming pivotal during later decoding, which can lead to premature eviction. 2) Discrete pruning disrupts the inherent spatial continuity of visual cues. To address these challenges, we propose RetentiveKV, an entropy-driven KV cache optimization method that reformulates KV eviction from "discrete context truncation" to "continuous memory evolution" based on State Space Models. Our method leverages information entropy to quantify the information potential of low-attention tokens and integrates tokens scheduled for eviction into a continuous state space through entropy-guided state transitions, enabling their dynamic reactivation when semantic relevance arises during subsequent decoding. Extensive experiments on multimodal benchmarks demonstrate that RetentiveKV achieves 5.0 times KV cache compression and 1.5 times decoding acceleration.
Authors: Zhiyuan Zeng, Jiameng Huang, Zhangyue Yin, Jiashuo Liu, Ziniu Li, Bingrui Li, Yuhao Wu, Yining Zheng, Ge Zhang, Wenhao Huang, Xipeng Qiu
Abstract: Reinforcement learning with verifiable rewards (RLVR) has become a central paradigm for improving reasoning and code generation in large language models, and GRPO-style training is widely adopted for its simplicity and effectiveness. However, an important design choice remains underexplored: how token-level policy gradient terms are aggregated within each sampled group. Standard GRPO uses sequence aggregation, while recent work has advocated token aggregation as a better alternative. We show that these two rules induce different optimization biases: token aggregation introduces sign-length coupling, while sequence aggregation implicitly downweights longer responses through sequence-level equal weighting. To address this tension, we propose \textbf{Balanced Aggregation (BA)}, a simple drop-in replacement that computes token-level means separately within the positive and negative subsets and then combines them with sequence-count-based weights. Experiments with Qwen2.5-Math-7B and Qwen3-1.7B on DAPO-17k and Polaris, evaluated on six reasoning and coding benchmarks, show that BA consistently improves training stability and final performance over standard token and sequence aggregation. Our analysis further shows that the relative effectiveness of token and sequence aggregation is largely governed by response-length variation and the positive-negative length gap, highlighting aggregation as a critical design dimension in GRPO-style RLVR.
Authors: Lydie Bednarczyk, Jamil Zaghir, Julien Ehrsam, Maria Tcherepanova, Christian Skalafouris, Karim Gariani, Catherine Geslin, Claire-B\'en\'edicte Rivara, Pascal Bonnabry, Laetitia Gosetto, Richard Dubos, Mina Bjelogrlic, Christophe Gaudet-Blavignac, Christian Lovis
Abstract: Objectives: Large language models (LLMs) are increasingly used for clinical text summarization, yet structured methods to assess associated patient safety risks remain limited. Failure Mode, Effects, and Criticality Analysis (FMECA) provides a proactive framework for systematic risk identification but has not been adapted to LLM-generated clinical content. This study aimed to develop and validate a novel FMECA framework for the prospective assessment of patient safety risks in LLM-generated clinical summaries. Materials and Methods: An interdisciplinary expert panel (n = 8) developed a taxonomy of failure modes through literature review and brainstorming. Standard FMECA dimensions (occurrence, severity, detectability) were adapted into 5-point ordinal scales. The framework was applied to 36 discharge summaries from four patients, generated by an open LLM (GPT-OSS 120B) using real-world clinical data from the Geneva University Hospitals. Reviewers independently annotated the summaries across two rounds. Inter-rater reliability was assessed at failure mode, severity and detectability score levels. Usability and content validity were evaluated using an adapted System Usability Scale and structured feedback. Results: The final framework comprised 14 failure modes organized into categories. Inter-rater agreement improved between rounds, reaching moderate-to-substantial agreement for failure mode identification and good agreement for severity and detectability scoring. Usability was rated as good (mean SUS: 79.2/100), with high evaluator confidence. Discussion and Conclusion: This study presents the first FMECA-based framework for systematic patient safety risk assessment of LLM-generated clinical summaries. The framework provides a structured and reproducible method for identifying clinically relevant risks caused by these summaries.
Authors: Matteo Gambella, Fabrizio Pittorino, Manuel Roveri
Abstract: Neural Architecture Search (NAS) has emerged as a powerful framework for automatically discovering neural architectures that balance accuracy and efficiency. However, as AI transitions from static benchmarks to real-world deployment, the traditional focus on hardware-aware efficiency is no longer sufficient. We observe that modern NAS methods, especially those that target edge AI, are evolving to address a triple objective: Efficiency, Robustness, and Continual Learning. While efficiency ensures feasibility in resource-constrained environments, robustness guarantees reliability under environmental variabilities, and continual learning enables adaptation to sequential tasks without catastrophic forgetting. We propose a taxonomy of NAS approaches through this triple lens, distinguishing between methods targeting resource optimization, environmental resilience, and architectural plasticity. This unified perspective reveals that these axes, though often studied in isolation, are mutually reinforcing. Building on this taxonomy, we map the current landscape of these NAS methods into a new framework called Hardware-Efficient, Robust, and ContinUal LEarning Search (HERCULES). We define the desiderata, the twelve labours of HERCULES, addressing the non-trivial challenge of balancing an adequate search-space exploration with the immense computational costs of a multi-objective NAS, accounting for these crucial objectives of current AI systems. By identifying critical gaps in existing research, this survey outlines a roadmap toward integrated algorithmic, architectural, and hardware-software co-design for truly deployable, lifelong-learning AI systems.
Authors: Furkan Sakizli
Abstract: Production agent frameworks (OpenAI Function Calling, Anthropic Tool Use, MCP) transmit tool schemas as JSON, a format designed for machine parsing, not for interpretation by language models. For small models (4B-14B), this protocol mismatch accounts for the majority of tool-use failure at production catalog sizes. We present TSCG, a deterministic tool-schema compiler that resolves this mismatch at the API boundary, converting JSON schemas into token-efficient structured text without model access, fine-tuning, or runtime search. TSCG combines eight composable operators with a formal compression bound (>=51% on well-formed schemas). On TSCG-Agentic-Bench (about 19,000 calls, 12 models, 5 scenarios), TSCG restores Phi-4 14B from 0% to 84.4% accuracy at 20 tools (90.3% at 50 tools) and achieves 108-181% accuracy-retained ratio across three models on BFCL. Format-versus-compression decomposition (R^2=0.88 -> 0.03) establishes representation change as the dominant mechanism. Per-operator isolation across three frontier models reveals three distinct operator-response profiles: operator-hungry (Opus 4.7), operator-sensitive (GPT-5.2), and operator-robust (Sonnet 4), providing per-model deployment guidance. Scaling experiments show accuracy advantages persisting on heavy production MCP schemas (+5.0 pp at about 10,500 input tokens) despite saturation on light synthetic catalogs, with 52-57% token savings throughout. The synthetic benchmark generalizes to real MCP schemas within 0.1 accuracy points. TSCG ships as a 1,200-line zero-dependency TypeScript package.
Authors: Devon Jarvis, Richard Klein, Benjamin Rosman, Steven James, Stefano Sarao Mannelli
Abstract: Model collapse, the degradation in performance that arises when generative models are trained on the outputs of prior models, is an increasing concern as artificially generated content proliferates. Related critiques of large language models have highlighted their tendency to reproduce frequent patterns in training data, their reliance on vast datasets, and their substantial environmental cost. Together, these factors contribute to data degradation, the reinforcement of cultural biases, and inefficient resource use. In this position paper we aim to combine these views and argue that model collapse threatens current efforts to democratize AI. By reducing training efficiency and skewing data distributions away from the tails of their support, model collapse disproportionately impacts low-resource and marginalized communities. We examine both the environmental and cultural implications of this phenomenon, situate our position within recent position papers on model collapse, and conclude with a call to action. Finally, we outline initial directions for mitigating these effects.
Authors: Lin Song, Wenbo Li, Guoqing Ma, Wei Tang, Bo Wang, Yuan Zhang, Yijun Yang, Yicheng Xiao, Jianhui Liu, Yanbing Zhang, Guohui Zhang, Wenhu Zhang, Hang Xu, Nan Jiang, Xin Han, Haoze Sun, Maoquan Zhang, Haoyang Huang, Nan Duan
Abstract: We present JoyAI-Image, a unified multimodal foundation model for visual understanding, text-to-image generation, and instruction-guided image editing. JoyAI-Image couples a spatially enhanced Multimodal Large Language Model (MLLM) with a Multimodal Diffusion Transformer (MMDiT), allowing perception and generation to interact through a shared multimodal interface. Around this architecture, we build a scalable training recipe that combines unified instruction tuning, long-text rendering supervision, spatially grounded data, and both general and spatial editing signals. This design gives the model broad multimodal capability while strengthening geometry-aware reasoning and controllable visual synthesis. Experiments across understanding, generation, long-text rendering, and editing benchmarks show that JoyAI-Image achieves state-of-the-art or highly competitive performance. More importantly, the bidirectional loop between enhanced understanding, controllable spatial editing, and novel-view-assisted reasoning enables the model to move beyond general visual competence toward stronger spatial intelligence. These results suggest a promising path for unified visual models in downstream applications such as vision-language-action systems and world models.
Authors: David Gringras, Misha Salahshoor
Abstract: Readers of applied-domain LLM capability evaluations want to know what AI systems can currently do. That literature answers a related, but consequentially different, question: what older, cheaper, less-elicited models could do months or years earlier (a 2026 paper evaluating GPT-4o-mini zero-shot, say, against a frontier of reasoning-capable, tool-using systems like GPT-5.5 Pro and Claude Opus 4.7), often reported with sparse configuration details and abstracted upward into claims about "AI" that propagate through citations, media, and policy. We measure the 'publication elicitation gap' (the gap between these answers) in a pre-registered audit of 112,303 LLM-keyword-matched candidate records (2022-01 to 2026-04; 18,574 admissible, 4,766 full-paper texts retrievable), comparing tested models to the contemporaneous frontier on the Epoch AI Capabilities Index (ECI), reproduced under Arena Elo and Artificial Analysis. The median paper evaluates a model +10.85 ECI (~1.4x the distance between Claude Sonnet 3.7 and Claude Opus 4.5) behind the contemporaneous frontier at evaluation time (H1); an exploratory rational-lag baseline (H8) decomposes this into ~25% peer-review latency, ~75% excess lag. The gap is widening at +5.53 ECI/year (H2; 95% CI [+5.03, +5.83]). Meanwhile, only 3.2% of abstracts (21.2% of full-texts) disclose reasoning-mode status on reasoning-capable models (H4) and 52.5% (95% CI [48.2, 56.9]) state conclusions at the level of "AI" rather than the evaluated model(s), rising at OR = 1.23/year. Proposed remedies include API-access subsidies and editorial enforcement of reporting frameworks mandating configuration-surface disclosure (model snapshot, reasoning mode/effort, tool access, scaffolding, prompting, etc.); VERSIO-AI is a 13-item checklist (Core 3 desk-reject) extending existing frameworks at the elicitation surface, with per-DOI analysis at frontierlag.org.
Authors: Yaobo Zhang
Abstract: Relative positional encodings determine which functions of query-key lag can enter the primitive attention logit. RoPE supplies a rotary phase, while ALiBi supplies an additive distance bias. Motivated by group-theoretic views of linear translation-invariant positional encodings, we study a non-semisimple case in which a complex rotary eigenvalue and a nilpotent response live in the same defective Jordan block. The resulting relative operator generates oscillatory-polynomial features such as $e^{-\gamma d}\cos(\omega d)$, $e^{-\gamma d}\sin(\omega d)$, $d e^{-\gamma d}\cos(\omega d)$, and $d e^{-\gamma d}\sin(\omega d)$, for causal lag $d=i-j\geq 0$. Thus the construction realizes a distance-modulated phase basis $d e^{i\omega d}$, rather than merely adding a separate distance channel to RoPE. We formulate Exact Jordan-RoPE as a non-semisimple one-parameter representation, give its real block form, and specify the contragredient query action required by non-orthogonal positional maps. We also distinguish this exact representation from stabilized variants whose bounded shear improves numerical behavior but breaks the exact group law. Kernel-level diagnostics and a Jordan-friendly synthetic language-model task show that the coupled Jordan basis is useful when the target contains distance-modulated phase interactions. On a small WikiText-103 byte language model, a scaled-exact variant improves over RoPE and direct-sum baselines within the Jordan family, while RoPE+ALiBi remains strongest overall. The evidence is structural rather than a broad performance claim.
Authors: Qihua Dong, Ruozhen He, Junwen Chen, Yizhou Wang, Xu Ma, Songyao Jiang, Yun Fu
Abstract: Advanced chart question answering requires both precise perception of small visual elements and multi-step reasoning across several subplots. While existing MLLMs are strong at understanding single plots, they often struggle with multi-step reasoning across multiple subplots. We propose HierVA, a hierarchical visual agent framework for chart reasoning that iteratively constructs and updates a working context in a joint image--text space. A high-level manager generates plans and maintains a compact context containing only key information, while specialized workers perform reasoning, gather evidence, and return results. In particular, the agent maintains separate visual and textual contexts, using a zoom-in tool to restrict the visual context. Experiments on the CharXiv reasoning subset demonstrate consistent improvements over strong multimodal baselines, and ablation studies verify that hierarchical architecture, scoped visual context, and distilled context contribute complementary gains.
Authors: Mohammed Sabry, Anya Belz
Abstract: We study distillation for large language models under explicit compute constraints, with the goal of producing student models that are not only cheaper to train, but structurally efficient at inference time. While prior approaches to parameter-efficient distillation, such as LoRA, reduce adaptation cost, they leave the dense backbone unchanged and therefore fail to deliver meaningful inference savings. We propose Budgeted LoRA, a distillation framework that treats model compression as a structured compute allocation problem. Instead of using a fixed student architecture, we introduce a global compute budget that sets the final target fraction of dense computation retained. Under this constraint, the model redistributes capacity across dense and low-rank pathways via (i) module-level dense retention coefficients, (ii) adaptive low-rank allocation, and (iii) post-training compression that selectively removes, approximates, or preserves dense components. This formulation yields a family of students controlled by a single budget dial. Empirically, Budgeted LoRA matches standard LoRA perplexity at a moderate budget with a 1.74x compressed-module speedup; at an aggressive budget it achieves a 4.05x speedup with moderate perplexity degradation, and it preserves higher accuracy on function-style in-context learning probes. These results suggest that, under compute-constrained distillation, retaining behavior is less about matching perplexity or removing more parameters than it is about controlling how dense computation is transferred to low-rank pathways.
Authors: Yixuan Mei, Zikun Li, Zixuan Chen, Shiqi Pan, Mengdi Wu, Xupeng Miao, Zhihao Jia, K. V. Rashmi
Abstract: The usage of large language models (LLMs) has grown increasingly fragmented, with no single model dominating. Meanwhile, cloud providers offer a wide range of mid-tier and older-generation GPUs that enjoy better availability and deliver comparable performance per dollar to top-tier hardware. To efficiently harness these heterogeneous resources for serving multiple LLMs concurrently, we introduce Coral, an adaptive heterogeneity-aware multi-LLM serving system. The key idea behind Coral is to jointly optimize resource allocation and the serving strategy of each model replica across all models. To keep pace with shifting throughput demand and resource availability, Coral applies a lossless two-stage decomposition that preserves joint optimality while cutting online solve time from hours to tens of seconds. Our evaluation across 6 models and 20 GPU configurations shows that Coral reduces serving cost by up to 2.79$\times$ over the best baseline, and delivers up to 2.39$\times$ higher goodput under scarce resource availability.
Authors: Xinyu Wang, Changzhi Sun, Yuanbin Wu, Xiaoling Wang
Abstract: Post-training large language models (LLMs) often suffers from catastrophic forgetting, where improvements on a target objective degrade previously acquired capabilities. Recent evidence suggests that this phenomenon is primarily driven by excessive distributional drift during optimization. Motivated by this perspective, we propose Anchored Learning, a simple framework that explicitly controls distributional updates during offline fine-tuning via a dynamically evolving moving anchor. Instead of matching a fixed reference distribution, the anchor interpolates between the current model and a frozen reference to construct an intermediate target that the model distills toward, transforming global fine-tuning into a sequence of local trust-region updates in distribution space. Theoretically, we prove this anchor-based update admits a linear KL-divergence upper bound per iteration, ensuring a stable transition between model distributions. Extensive experiments on iGSM, MedCalc, and IFEval show that Anchored Learning consistently lies on the Pareto frontier of gain-stability trade-offs, achieving near-optimal performance improvements while substantially reducing degradation compared to strong baselines. For example, while standard SFT suffers from over 53% performance degradation on iGSM and MedCalc, Anchored Learning slashes this drop to under 5% while maintaining near-optimal gains (e.g., 75.2% on iGSM).
Authors: Do Minh Duc, Quan Xuan Truong, Viet Tran Hong, Le Hoang Anh, Mac Thi Minh Tra, Nguyen Van Thuy, Le Hai Ha, Vinh Nguyen Van
Abstract: Named Entity Recognition (NER) is a critical component of Natural Language Processing with diverse applications in information extraction and conversational AI. However, NER in specific domains for low-resource languages faces challenges such as limited annotated data and heterogeneous label sets. This study addresses these issues by proposing a hybrid neurosymbolic framework that integrates rule-based processing with deep learning models for Vietnamese NER. The core idea involves a two-stage pipeline: first, a rule-based component reduces label complexity by grouping relational and special categories; second, pre-trained language models are fine-tuned for high-precision extraction. A post-processing module is then utilized to restore fine-grained labels, preserving expressiveness for application-level usability. To mitigate data scarcity, a scalable data augmentation strategy leveraging Large Language Models (LLMs) is introduced to expand the label set without full re-annotation, which is a significant novelty of this work. The effectiveness of this method was evaluated across five specific-domain datasets, including logistics, wildlife, and healthcare. Experimental results demonstrate substantial improvements over strong RoBERTa-based baselines. Specifically, the proposed system achieved F1 scores of 90 percent in Customer Service, up from 83 percent; 84 percent in GAM, up from 73 percent; 83 percent in AI Fluent, up from 80 percent; 94 percent in PhoNER_Covid19, up from 91 percent; and 60 percent in Rare Wildlife, up from 36 percent. These findings confirm that the hybrid approach effectively captures the linguistic complexity of Vietnamese and contextual nuances in specialized domains, offering a robust contribution to low-resource NER research.
Authors: Jingtao Zhou, Xirui Kang, Feiyang Huang, Lai-Man Po
Abstract: Existing prompt learning for VLMs exhibits a modality asymmetry, predominantly optimizing text tokens while still relying on frozen visual encoder as holistic extractor and neglecting the spectral granularity essential for fine-grained discrimination. To bridge this, we introduce Disentangling Spectral Granularity for Prompt Learning (SpecPL), which approaches prompt learning from a novel spectral perspective via Counterfactual Granule Supervision. Specifically, we leverage a frozen VAE to decompose visual signals into semantic low-frequency bands and granular high-frequency details. A frozen Visual Semantic Bank anchors text representations to universal low-frequency invariants, mitigating overfitting. Crucially, fine-grained discrimination is driven by counterfactual granule training: by permuting high-frequency signals, we compel the model to explicitly distinguish visual granularity from semantic invariance. Uniquely, SpecPL serves as a universal plug-and-play booster, revitalizing text-oriented baselines like CoOp and MaPLe via visual-side guidance. Experiments on 11 benchmarks demonstrate competitive state-of-the-art performance, achieving a new performance ceiling of 81.51\% harmonic-mean accuracy. These results validate that spectral disentanglement with counterfactual supervision effectively bridges the gap in the stability-generalization trade-off. Code is released at https://github.com/Mlrac1e/SpecPL-Prompt-Learning.
Authors: Wei Liu, Jiaxin Lin, Rui Chen
Abstract: Recent studies have shown that large generative models can solve vision tasks they were not explicitly trained for. However, existing evidence relies on closed-source models~(Veo~3, Nano Banana Pro) or requires task-specific instruction tuning, leaving open whether publicly available image-editing models possess zero-shot vision abilities out of the box. We conduct a systematic evaluation of three open-source image-editing models -- Qwen-Image-Edit, FireRed-Image-Edit, and LongCat-Image-Edit -- on dense visual prediction tasks \emph{without any fine-tuning}. We benchmark monocular depth estimation on NYUv2 and DIODE, surface normal estimation on NYUv2, and semantic segmentation on Cityscapes, covering both geometric and semantic scene understanding. Results show that open-source image-editing models exhibit non-trivial zero-shot visual understanding. On NYUv2 surface normals, FireRed-Image-Edit achieves a mean angular error of $17.69^\circ$, surpassing the fine-tuned Marigold ($20.86^\circ$) and matching the instruction-tuned Vision Banana ($17.78^\circ$) without any task-specific training. On NYUv2 depth estimation, LongCat-Image-Edit obtains $\delta_1{=}0.822$ with affine alignment, and Qwen-Image-Edit leads on DIODE Indoor ($\delta_1{=}0.868$). On Cityscapes semantic segmentation, Qwen-Image-Edit reaches 25.7 mIoU at the 19-class level and 49.5 mIoU at a coarser 7-category level. By comparing three independently trained editors, we test whether zero-shot vision ability is an emergent property of image-editing pretraining rather than a model-specific artifact. Code, evaluation scripts, and all results are publicly released to serve as a reproducible baseline for future work.
Authors: Guangsheng Bao, Hongbo Zhang, Han Cui, Yanbin Zhao, Yue Zhang
Abstract: Adapting pretrained models typically involves a trade-off between the high training costs of backpropagation and the heavy inference overhead of memory-based or in-context learning. We propose FAAST, a forward-only associative adaptation method that analytically compiles labeled examples into fast weights in a single pass. By eliminating memory or context dependence, FAAST achieves constant-time inference and decouples task adaptation from pretrained representation. Across image classification and language modeling benchmarks, FAAST matches or exceeds backprop-based adaptation while reducing adaptation time by over 90\% and is competitive to memory/context-based adaptation while saving memory usage by up to 95\%. These results demonstrate FAAST as a highly efficient, scalable solution for supervised task adaptation, particularly for resource-constrained models. We release the code and models at https://github.com/baoguangsheng/faast.
Authors: Zheng Fang, Xiaosen Wang, Shenyi Zhang, Shaokang Wang, Zhijin Ge
Abstract: Jailbreak attacks on audio language models (ALMs) optimize audio perturbations to elicit unsafe generations, and they typically update the entire waveform densely throughout optimization. In this work, we investigate the necessity of such dense optimization by analyzing the structure of token-aligned gradients in ALMs. We find that gradient energy is highly non-uniform across audio tokens, indicating that only a small subset of token-aligned audio regions dominates the optimization signal. Motivated by this observation, we propose Token-Aware Gradient Optimization (TAGO), which enables sparse jailbreak optimization by retaining only waveform gradients aligned with audio tokens that have high gradient energy, while masking the remaining gradients at each iteration. Across three ALMs, TAGO outperforms baselines, and substantial sparsification preserves strong attack success rates (e.g. on Qwen3-Omni, $\mathrm{ASR}_{l}$ remains at 86% with a token retention ratio of 0.25, compared to 87% with full token retention). These results demonstrate that dense waveform updates are largely redundant, and we advocate that future audio jailbreak and safety alignment research should further leverage this heterogeneous token-level gradient structure.
Authors: Huatian Zhang, Zhendong Mao, Lei Zhang, Yongdong Zhang
Abstract: Direct Preference Optimization (DPO) has proven to be an effective solution for mitigating hallucination in Multimodal Large Language Models (MLLMs) by learning from preference pairs. One of its key challenges lies in how to transfer the sequence-level preference into fine-grained supervision on visual fidelity. To safeguard vision-related tokens that are prone to hallucination, existing methods typically allocate training emphasis according to the model's self-assessed visual sensitivity signals. However, such sensitivity, estimated by a model still under training, introduces self-referential bias: reinforcing already well-learned visual cues while neglecting hard-to-perceive but critical details, thereby limiting deeper alignment. In this work, we propose an Uncertainty-aware Exploratory Direct Preference Optimization (UE-DPO) method for MLLMs, which enables the model to uncover its cognitive deficiencies and actively explore for self-correction, guided by token-level epistemic uncertainty. Specifically, we first quantify the uncertainty from the model's failure to ground token predictions in the given image. Then, based on an uncertainty-aware exploration intensity, we encourage more learning pressure on visually deficient tokens in preferred samples, and alleviate the over-penalization of beneficial knowledge in dispreferred samples. Further, we provide a theoretical justification for our method, and extensive experiments demonstrate its effectiveness and robustness.
Authors: Dominik Dahlem, Diego Maniloff, Mac Misiura
Abstract: Large language models hallucinate in predictable ways: attention routing fails by over-concentrating on a narrow set of positions, or by spreading so diffusely that relevance is diluted, and the shape of the failure carries diagnostic signal. A widely used family of spectral methods analyzes the symmetric component of the degree-normalized attention operator, which governs transport capacity; we prove that every transpose-invariant spectral diagnostic of this operator is structurally orientation-blind (it cannot distinguish an operator from its transpose, and therefore cannot detect information-flow direction), with a quantitative converse establishing the asymmetry coefficient $G$ as the unique control parameter for direction. Pairing this with a closed-form bipartite-Cheeger landscape for canonical causal architectures, we show that uniform causal attention satisfies an $n$-independent floor $\phi \ge 1/5$ with worst cut at $t^\ast/n \approx 0.32$, while window attention pierces the floor as $O(w/n)$; failure modes are shape-different, not just value-different. The resulting two-axis diagnostic ($\phi$ for capacity, $G$ for direction) yields a falsifiable polarity prediction: bottleneck- and diffuse-dominated benchmarks should exhibit opposite polarity. Under length-controlled evaluation, transport features retain interpretable signal (LC-AUROC from 0.62 to 0.84) on tested models up to 8B parameters, with polarity reversing as predicted between HaluEval and MedHallu.
Authors: Xiyan Fu, Wei Liu
Abstract: Compositional generalization refers to correctly interpret novel combinations of known primitives, which remains a major challenge. Existing approaches often rely on supervised fine-tuning, which encourages models to imitate target outputs. This token-level training paradigm fails to capture the global compositional structure required for generalizing to unseen combinations. In this work, we investigate whether compositional generalization can instead be improved through outcome-level reinforcement learning. We adopt Group Relative Policy Optimization to optimize models based on feedback on their final outputs. Within this framework, we explore both a simple binary outcome reward and a composite reward that provides additional composition feedback. Experiments on multiple compositional benchmarks show that reinforcement learning improves compositional generalization compared to supervised fine-tuning. Further analysis reveals that supervised models tend to overfit frequent training compositions, whereas reinforcement learning improves compositional generalization by reshaping the output distribution, particularly for more complex composition types.
Authors: Kyungwon Jeong, Won-Gi Paeng, Honggyo Suh
Abstract: Weight matrices in deep networks exhibit geometric continuity -- principal singular vectors of adjacent layers point in similar directions. While this property has been widely observed, its origin remains unexplained. Through experiments on toy MLPs and small transformers, we identify two mechanisms: residual connections create cross-layer gradient coherence that aligns weight updates across layers, and symmetry-breaking nonlinearities constrain all layers to a shared coordinate frame, preventing the rotation drift that would otherwise destabilize weight structure. Crucially, a nonlinear but rotation-preserving activation fails to retain continuity, isolating symmetry breaking -- not nonlinearity itself -- as the active ingredient. Activation and normalization play distinct roles: activation concentrates continuity in the leading singular direction, while normalization distributes it across multiple directions. In transformers, continuity is projection-specific: Q, K, Gate, and Up (which read from the residual stream) develop input-space ($\mathbf{v}_1$) continuity; O and Down (which write to it) develop output-space ($\mathbf{u}_1$) continuity; V alone, lacking an adjacent nonlinearity, develops only low continuity.
Authors: Ilias Triantafyllopoulos, Young-Min Cho, Ren Tao, Miranda Muqing Miao, Sunny Rai, Lyle Ungar, Sharath Chandra Guntuku, Neville Ryant, Jo\~ao Sedoc
Abstract: Activation-based steering provides control of LLM behavior at inference time, but the dominant paradigm reduces each concept to a single direction whose geometry is left largely unexamined. Rather than selecting a single steering direction, we use conceptors: soft projection matrices estimated from activations pooled across both poles of a bipolar concept, which preserve the concept's full multidimensional subspace. A geometric analysis shows the bipolar subspace strictly subsumes the single-vector baseline. We further show that the conceptor quota provides a parameter-free layer-selection diagnostic, predicting concept separability with Pearson correlations up to r=0.96 across three instruction-tuned models and three semantic dimensions. Beyond selection, conceptors admit a closed-form Boolean algebra (AND, OR, NOT): we evaluate conceptor compositionality on thematically related sub-concepts. Across a systematic five-axis design-space evaluation, conceptors match or outperform additive baselines at layers where concept subspaces are multi-dimensional while producing substantially fewer degenerate outputs. Conceptor steering is a geometrically principled, compositional, and practically safer alternative to single-direction steering from a limited number of contrastive pairs.
Authors: Senkang Hu, Yong Dai, Xudong Han, Zhengru Fang, Yuzhi Zhao, Sam Tak Wu Kwong, Yuguang Fang
Abstract: Long-horizon LLM agents depend on intermediate information-gathering turns, yet training feedback is usually observed only at the final answer, because process-level rewards require high-quality human annotation. Existing turn-level shaping methods reward turns that increase the likelihood of a gold answer, but they require answer supervision or stable task-specific verifiers. Conversely, label-free RL methods extract self-signals from output distributions, but mainly at the answer or trajectory level and therefore cannot assign credit to intermediate turns. We propose Self-Induced Outcome Potential (SIOP), which treats semantic clusters of final answers as latent future outcome states for potential-based turn-level credit assignment. For each query, SIOP samples multiple rollouts, clusters final answers into semantic outcome modes, and builds a reliability-aware target distribution over these states. It then rewards turns for increasing posterior support for reliable future states using a tractable cluster-level approximation. The objective generalizes information-potential shaping from gold-answer supervision to settings without task-specific gold verifiers while avoiding the broadcasted rollout-level advantages used by standard GRPO. We formalize the framework, characterize its supervised gold-answer limit, and show that SIOP improves average performance over verifier-free outcome-level baselines on seven search-augmented agentic reasoning benchmarks while approaching a gold-supervised outcome baseline. Code is available at https://github.com/dl-m9/SIOP.git.
Authors: Philip Wootaek Shin, Ajay Narayanan Sridhar, Sivani Devarapalli, Rui Zhang, Jack Sampson, Vijaykrishnan Narayanan
Abstract: Vision-language models (VLMs) achieve strong multimodal performance but remain prone to relation hallucination, which requires accurate reasoning over inter-object interactions. We study the impact of visual perturbations, specifically rotation and noise, and show that even mild distortions significantly degrade relational reasoning across models and datasets. We further evaluate prompt-based augmentation and preprocessing strategies (orientation correction and denoising), finding that while they offer partial improvements, they do not fully resolve hallucinations. Our results reveal a gap between perceptual robustness and relational understanding, highlighting the need for more robust, geometry-aware VLMs.
Authors: Andreas Pattichis, Constantine Dovrolis
Abstract: LLMs are trained once, then deployed into a world that never stops changing. External memory compensates for this, but most systems manage it explicitly rather than letting it adapt on its own. Biological memory works differently: coupled multi-timescale dynamics make new associations immediately usable, strengthen what repetition confirms, and let the rest fade. We argue that external memory should follow a similar principle. In Memini, this view takes the form of an associative memory that organizes knowledge as a directed graph. Each edge carries two coupled internal variables, one fast and one slow, following the Benna-Fusi model of synaptic consolidation. From this coupling, episodic sensitivity, gradual consolidation, and selective forgetting emerge as facets of a single mechanism, reframing external memory as a learning substrate that reorganizes through its own dynamics.
Authors: Perry E. Radau
Abstract: Background: Existing MRI LLM benchmarks rely mainly on review-book multiple-choice questions, where top proprietary models already score highly, limiting discrimination. No systematic benchmark has evaluated vendor-specific scanner operational knowledge central to research MRI practice. Purpose: We developed MRI-Eval, a tiered benchmark for relative model comparison on MRI physics and GE scanner operations knowledge using primary multiple-choice questions (MCQ), with stem-only and primed diagnostic conditions as complementary analyses. Methods: MRI-Eval includes 1365 scored items across nine categories and three difficulty tiers from textbooks, GE scanner manuals, programming course materials, and expert-generated questions. Five model families were evaluated (GPT-5.4, Claude Opus 4.6, Claude Sonnet 4.6, Gemini 2.5 Pro, Llama 3.3 70B). MCQ was primary; stem-only removed options and used an independent LLM judge; primed stem-only tested responses to incorrect user claims. Results: Overall MCQ accuracy was 93.2% to 97.1%. GE scanner operations was the lowest category for every model (88.2% to 94.6%). In stem-only, frontier-model accuracy fell to 58.4% to 61.1%, and Llama 3.3 70B fell to 37.1%; GE scanner operations stem-only accuracy was 13.8% to 29.8%. Conclusion: High MCQ performance can mask weak free-text recall, especially for vendor-specific operational knowledge. MRI-Eval is most informative as a relative comparison benchmark rather than an absolute competency measure and supports caution in using raw LLM outputs for GE-specific protocol guidance.
Authors: Sruly Rosenblat, Tim O'Reilly, Ilan Strauss
Abstract: Using a legally obtained dataset of 34 copyrighted O'Reilly Media books, we apply the DE-COP membership inference attack method to investigate whether OpenAI's large language models show recognition of copyrighted content. Our results based on this small sample suggest that GPT-4o, OpenAI's more recent and capable model, exhibits patterns consistent with recognition of pay-walled book content, with an AUROC score of 0.82 (95% bootstrapped CI: 0.60-0.96), though this wide confidence interval reflects substantial uncertainty due to the limited number of books tested. GPT-4o Mini, as a much smaller model, shows little recognition of any O'Reilly Media content with an AUROC score of 0.56 (0.28-0.83) for non-public data. Testing multiple models, with the same cutoff date, provides a partial control for potential language shifts over time that might bias our findings, though differences in model size, architecture, and potentially training data composition limit the strength of this control. These preliminary results underscore the importance of increased corporate transparency regarding pre-training data sources and the development of formal licensing frameworks for AI content training. Our principal contribution is our examination of public and non public data separately.
Authors: Henry Peng Zou, Wei-Chieh Huang, Yaozu Wu, Jizhou Guo, Yankai Chen, Chunyu Miao, Hoang Nguyen, Yue Zhou, Weizhi Zhang, Liancheng Fang, Hanrong Zhang, Fangxin Wang, Pengfei Zhang, Huacan Wang, Langzhou He, Yangning Li, Dongyuan Li, Renhe Jiang, Xue Liu, Philip S. Yu
Abstract: Recent advances in large language models (LLMs) have sparked growing interest in building fully autonomous agents. However, fully autonomous LLM-based agents still face significant challenges, including limited reliability due to hallucinations, difficulty in handling complex tasks, and substantial safety and ethical risks, all of which limit their feasibility and trustworthiness in real-world applications. To overcome these limitations, LLM-based human-agent systems (LLM-HAS) incorporate human-provided information, feedback, or control into the agent system to enhance system performance, reliability, and safety. These human-agent collaboration systems enable humans and LLM-based agents to collaborate effectively by leveraging their complementary strengths. This paper provides the first comprehensive and structured survey of LLM-HAS. It clarifies fundamental concepts, systematically presents core components shaping these systems, including environment and profiling, human feedback, interaction types, orchestration, and communication, explores emerging applications, and discusses unique challenges and opportunities arising from human-AI collaboration. By consolidating current knowledge and offering a structured overview, we aim to foster further research and innovation in this rapidly evolving interdisciplinary field. Paper lists and resources are available at https://github.com/HenryPengZou/Awesome-Human-Agent-Collaboration-Interaction-Systems.
URLs: https://github.com/HenryPengZou/Awesome-Human-Agent-Collaboration-Interaction-Systems.
Authors: Yukin Zhang, Qi Dong, Kemu Xu
Abstract: Why do language models from different architecture families respond so differently to the same perturbation? We argue that the answer is not scale, but \emph{how architecture shapes information compression}. Analyzing eight Transformer models (7B--70B parameters) from the Llama and Qwen families, we show that every model spontaneously develops discrete functional boundaries dividing its layers into Local, Intermediate, and Global processing segments -- yet boundary locations and per-segment brittleness are determined overwhelmingly by architecture family rather than model size or training configuration. We formalize this regularity as the \textbf{Multi-Scale Probabilistic Generation Theory} (MSPGT), which models an autoregressive Transformer as a Hierarchical Variational Information Bottleneck system and derives a tiered set of falsifiable predictions. Three predictions are strongly confirmed: all eight models exhibit two prominent phase-transition boundaries (P1.1); Llama boundary positions are stable across a $10{\times}$ parameter range ($\mathrm{CV}{=}0.067$--$0.095$) while Qwen positions vary widely ($\mathrm{CV}{=}0.465$--$0.726$), precisely matching our strong- and weak-dominance conditions; and cross-architecture local-segment brittleness spans \textbf{three orders of magnitude} ($493{\times}$ ratio) -- a gap that architecture family alone predicts and that dwarfs any within-family or scale-driven variation.
Authors: Mamnuya Rinki, Chahat Raj, Anjishnu Mukherjee, Ziwei Zhu
Abstract: Evaluations of Large Language Models (LLMs) often overlook intersectional and culturally specific biases, particularly in underrepresented multilingual regions like South Asia. This work addresses these gaps by conducting a multilingual and intersectional analysis of LLM outputs across 10 Indo-Aryan and Dravidian languages, identifying how cultural stigmas influenced by purdah and patriarchy are reinforced in generative tasks. We construct a culturally grounded bias lexicon capturing previously unexplored intersectional dimensions including gender, religion, marital status, and number of children. We use our lexicon to quantify intersectional bias and the effectiveness of self-debiasing in open-ended generations (e.g., storytelling, hobbies, and to-do lists), where bias manifests subtly and remains largely unexamined in multilingual contexts. Finally, we evaluate two self-debiasing strategies (simple and complex prompts) to measure their effectiveness in reducing culturally specific bias in Indo-Aryan and Dravidian languages. Our approach offers a nuanced lens into cultural bias by introducing a novel bias lexicon and evaluation framework that extends beyond Eurocentric or small-scale multilingual settings.
Authors: Mikhail L. Arbuzov, Sisong Bei, Ziwei Dong, Dmitri Kalaev, Alexey A. Shvets
Abstract: The prevailing assumption of an exponential decay in large language model (LLM) reliability with sequence length, predicated on independent per-token error probabilities, posits an inherent limitation for long autoregressive outputs. Our research fundamentally challenges this view by synthesizing emerging evidence that LLM errors are not uniformly distributed but are concentrated at sparse "key tokens" ($5-10\%$ of total tokens) representing critical decision junctions. By distinguishing these high-impact tokens from the increasingly predictable majority, we introduce a new reliability formula explaining the sustained coherence of modern LLMs over thousands of tokens. Converging research streams reveal that long-context performance primarily depends on accurately navigating a few crucial semantic decision points rather than on uniform token-level accuracy, enabling targeted strategies that significantly outperform brute-force approaches. We thus propose a framework for next-generation systems centered on selective preservation of semantically vital tokens, dynamic computational allocation at uncertain decision boundaries, multi-path exploration at ambiguities, and architectures aligned with natural semantic domains. This marks a fundamental shift from raw scaling to strategic reasoning, promising breakthrough performance without proportionate computational scaling and offering a more nuanced understanding that supersedes the exponential decay hypothesis, thereby opening pathways toward substantially more powerful and efficient language systems.
Authors: Yuanhao Shen, Daniel Xavier de Sousa, Ricardo Mar\c{c}al, Hongyu Guo, Xiaodan Zhu
Abstract: Innovation is a key driving force of human civilization. As the body of knowledge has grown considerably, bridging knowledge across different disciplines, where significant innovation often emerges, has become increasingly challenging. The recent advancements in machine learning models, particularly Large Language Models (LLMs), have provided effective access to extensive knowledge sources and shown impressive abilities in reasoning, rendering significant opportunities for interdisciplinary discovery. Our research aims to understand the capabilities of state-of-the-art LLMs in integrating knowledge from different fields for interdisciplinary research (IDR). To address this fundamental problem, we introduce IDRBench, a pioneering framework that includes both datasets and evaluation tasks: (1) IDR Paper Identification, (2) IDR Idea Integration, and (3) IDR Idea Recommendation. Our study on ten mainstream LLMs provides a comprehensive analysis of their behavior and establishes benchmarks and baselines for future research. To the best of our knowledge, IDRBench is the first to provide a comprehensive investigation of LLMs' IDR capability.
Authors: Yuquan Wang, Mi Zhang, Yining Wang, Geng Hong, Mi Wen, Xiaoyu You, Min Yang
Abstract: Large Reasoning Models (LRMs) have demonstrated impressive performance in reasoning-intensive tasks, but they remain vulnerable to harmful content generation, particularly in the mid-to-late steps of their reasoning processes. Current defense methods, however, depend on costly fine-tuning and additional expert knowledge, which limits their scalability. In this work, we propose ReasoningGuard, an inference-time safeguard for LRMs. It injects timely safety aha moments during the reasoning process to guide the model towards harmless yet helpful reasoning. Our approach leverages the internal attention mechanisms of the LRM to accurately identify key points in the reasoning path, triggering safety-oriented reflections. To safeguard both the subsequent reasoning steps and the final answers, we implement a scaling sampling strategy during decoding to select the optimal reasoning path. With minimal additional inference cost, ReasoningGuard effectively mitigates four types of jailbreak attacks, including recent ones targeting the reasoning process of LRMs. Our approach outperforms nine existing safeguards, providing state-of-the-art defenses while avoiding common exaggerated safety issues.
Authors: Xu Pan, Ely Hahami, Jingxuan Fan, Ziqian Xie, Haim Sompolinsky
Abstract: Large language models (LLMs) are often used in environments where facts evolve, yet factual knowledge updates via fine-tuning on unstructured text often suffer from 1) reliance on compute-heavy paraphrasing augmentation and 2) the reversal curse. Recent studies show diffusion large language models (dLLMs) require fewer training samples to achieve lower loss in pre-training and are more resistant to the reversal curse, suggesting dLLMs may learn new knowledge more easily than autoregressive LLMs (arLLMs). We test this hypothesis in controlled knowledge fine-tuning experiments and find that while arLLMs rely on paraphrase augmentation to generalize knowledge text into question-answering (QA) capability, dLLMs do not require paraphrases to achieve high QA accuracy. To further investigate whether the demasking objective alone can induce such a knowledge injection advantage in dLLMs regardless of their diffusion denoising paradigm, we propose masked fine-tuning for arLLMs, which prompts an arLLM to reconstruct the original text given a masked version in context. The masked fine-tuning for arLLMs substantially improves the efficacy of knowledge injection, i.e. no paraphrase needed and resistant to the reversal curse, closing the gap between arLLMs and dLLMs. We also demonstrate broader applicability: on a large-scale knowledge-intensive dataset (1.2M samples), masked SFT achieves the best downstream accuracy on GPQA-diamond among all fine-tuning variants. The demasking objective also improves SFT on math tasks, suggesting broad utility beyond factual knowledge injection.
Authors: Ziyi Liu, Bahar Sarrafzadeh, Pei Zhou, Longqi Yang, Jieyu Zhao, Ashish Sharma
Abstract: While Large Language Models (LLMs) are increasingly used in agentic frameworks to assist individual users, there is a growing need for agents that can proactively manage complex, multi-party collaboration. Systematic evaluation methods for such proactive agents remain scarce, limiting progress in developing AI that can effectively support multiple people together. Negotiation offers a demanding testbed for this challenge, requiring socio-cognitive intelligence to navigate conflicting interests between multiple participants and multiple topics and build consensus. Here, we present ProMediate, the first framework for evaluating proactive AI mediator agents in complex, multi-topic, multi-party negotiations. ProMediate consists of two core components: (i) a simulation testbed based on realistic negotiation cases and theory-driven difficulty levels (ProMediate-Easy, ProMediate-Medium, and ProMediate-Hard), with a plug-and-play proactive AI mediator grounded in socio-cognitive mediation theories, capable of flexibly deciding when and how to intervene; and (ii) a socio-cognitive evaluation framework with a new suite of metrics to measure consensus changes, intervention latency, mediator effectiveness, and intelligence. Together, these components establish a systematic framework for assessing the socio-cognitive intelligence of proactive AI agents in multi-party settings. Our results show that a socially intelligent mediator agent outperforms a generic baseline, via faster, better-targeted interventions. In the ProMediate-Hard setting, our social mediator increases consensus change by 3.6 percentage points compared to the generic baseline (10.65\% vs 7.01\%) while being 77\% faster in response (15.98s vs. 3.71s). In conclusion, ProMediate provides a rigorous, theory-grounded testbed to advance the development of proactive, socially intelligent agents.
Authors: P. Bilha Githinji, Aikaterini Melliou, Zeming Liang, Lian Zhang, Peiwu Qin
Abstract: The growing public demand for accessible biomedical information calls for scalable text simplification. While large language models (LLMs) offer solutions, they too struggle with balancing improved readability against preservation of meaning. This report empirically compares how two LLMs - instruction-tuned Mistral-Small 3 24B and the reasoning-augmented QWen2.5 32B- navigate this trade-off in biomedical text simplification, benchmarked against human performance. Our analysis highlights how each model applies distinct operational strategies when simplifying biomedical text. Mistral exhibits a tempered lexical simplification approach that consistently enhances readability across multiple metrics while preserving discourse fidelity (BERTScore: 0.91, statistically comparable to that of humans). In comparison, QWen also attains enhanced readability performance and a reasonable BERTScore of 0.89, but presents a disconnect in balancing between readability and accuracy. Additionally, a comprehensive correlation analysis of a suite of 21 metrics confirms strong functional redundancies in metrics and informs adaptation requirements.
Authors: Buu Phan, Ashish Khisti, Karen Ullrich
Abstract: Computing next-token likelihood ratios between two language models (LMs) is a standard task in training paradigms such as knowledge distillation. Since this requires both models to share the same probability space, it becomes challenging when the teacher and student LMs use different tokenizers, for instance, when edge-device deployment necessitates a smaller vocabulary size to lower memory overhead. This work addresses this vocabulary misalignment problem by uncovering an implicit recursive structure in the commonly deployed Byte-Pair Encoding (BPE) algorithm and utilizing it to create a probabilistic framework for cross-tokenizer likelihood scoring. Our method enables sequence likelihood evaluation for vocabularies different from the teacher model native tokenizer, addressing two specific scenarios: when the student vocabulary is a subset of the teacher vocabulary, and the general case where it is arbitrary. In the subset regime, our framework computes exact likelihoods and provides next-token probabilities for sequential sampling with only ${O}(1)$ model evaluations per token. When used for distillation, this yields up to a $12\%$ reduction in memory footprint for the Qwen2.5-1.5B model while also improving baseline performance up to $4\%$ on the evaluated tasks. For the general case, we introduce a rigorous lossless procedure that leverages BPE recursive structure, complemented by a fast approximation that keeps large-vocabulary settings practical. Applied to GSM8K mathematical reasoning distillation, our method improves accuracy by over $2\%$ the current state of the art. Code: github.com/truongbuu/cross-tokenizer-scoring
Authors: Weiqin Wang, Yile Wang, Kehao Chen, Hui Huang
Abstract: Test-time reinforcement learning mitigates the reliance on annotated data by using majority voting results as pseudo-labels, emerging as a complementary direction to reinforcement learning with verifiable rewards (RLVR) for improving reasoning ability. However, this voting strategy often induces confirmation bias and suffers from sparse rewards, limiting the overall performance. In this work, we propose subgroup-specific step-wise confidence-weighted pseudo-label estimation (SCOPE), a framework integrating model confidence and dynamic subgroup partitioning to address these issues. Specifically, SCOPE integrates the proposed step-wise confidence into pseudo label estimation, prioritizing high-quality reasoning paths over simple frequency count. Furthermore, it dynamically partitions the candidate outputs pool into independent subgroups by balancing reasoning quality against exploration diversity. By deriving local consensus via repeat sampling for each sub group, SCOPE provides diverse supervision targets to encourage broader exploration. We conduct experiments across various models and benchmarks, experimental results show that SCOPE consistently outperforms recent baselines. Notably, SCOPE achieving relative improvements of 13.1% on challenging AIME 2025 and 8.1% on AMC. The code is released at https://github.com/szu-tera/SCOPE.
Authors: Ziyi Zhu, Olivier Tieleman, Caitlin A. Stamatis, Luka Smyth, Thomas D. Hull, Daniel R. Cahn, Jinghong Chen, Matteo Malgaroli
Abstract: Realistic user simulation is crucial for training and evaluating multi-turn dialogue systems, yet creating simulators that accurately replicate human behavior remains a significant challenge. An effective simulator must expose the failure modes of the systems under evaluation. This work introduces Direct Iterative Adversarial Learning (DIAL), an adversarial framework that iteratively enhances user simulator realism through a competitive dynamic between a generator (user simulator) and a discriminator. When applied to mental health support, a domain characterized by diverse failure types and a critical dependence on realistic user behavior for failure detection, DIAL restores lexical diversity diminished by supervised fine-tuning and drastically reduces discriminator accuracy. The resulting simulator exhibits a strong correlation between simulated and real failure occurrence rates while maintaining low distributional divergence of failure modes. These findings indicate that DIAL is a promising method for developing realistic user simulators in multi-turn dialogue, facilitating reliable and cost-effective system evaluation prior to deployment.
Authors: Pere Martra
Abstract: Structured width pruning of GLU-MLP layers, guided by the Maximum Absolute Weight (MAW) criterion, reveals a systematic dichotomy in how reducing the expansion ratio affects different model capabilities. While performance on tasks relying on parametric knowledge (e.g., MMLU, GSM8K) and perplexity metrics degrades predictably, instruction-following capabilities improve substantially (+46% to +75% in IFEval for Llama-3.2-1B and 3B models), and multi-step reasoning remains robust (MUSR). This pattern challenges the prevailing assumption that pruning induces uniform degradation. We evaluated seven expansion ratio configurations using comprehensive benchmarks assessing factual knowledge, mathematical reasoning, language comprehension, instruction-following, and truthfulness. Our analysis identifies the expansion ratio as a critical architectural parameter that selectively modulates cognitive capabilities, rather than merely serving as a compression metric. We provide the first systematic characterization of this selective preservation phenomenon. Notably, we document a robust inverse correlation (r = -0.864, p = 0.012 in Llama-3B) between factual knowledge capacity (MMLU) and truthfulness metrics (TruthfulQA-MC2): as knowledge degrades, the model's ability to discriminate misconceptions improves consistently. This connects two previously distinct research areas, demonstrating that MAW-guided width pruning acts as a selective filter, reducing parametric knowledge while preserving or enhancing behavioral alignment. Additionally, we quantify context-dependent efficiency trade-offs: pruned configurations achieve up to 23% reduction in energy consumption (J/token) but incur penalties in single-request latency, whereas batch processing workloads benefit uniformly.
Authors: Tao Liu, Taiqiang Wu, Runming Yang, Shaoning Sun, Junjie Wang, Yujiu Yang
Abstract: Supervised fine-tuning (SFT) is a fundamental post-training strategy to align Large Language Models (LLMs) with human intent. However, traditional SFT often ignores the one-to-many nature of language by forcing alignment with a single reference answer, leading to the model overfitting to non-core expressions. Although our empirical analysis suggests that introducing multiple reference answers can mitigate this issue, the prohibitive data and computational costs necessitate a strategic shift: prioritizing the mitigation of single-reference overfitting over the costly pursuit of answer diversity. To achieve this, we reveal the intrinsic connection between token probability and semantic importance: high-probability tokens carry the core logical framework, while low-probability tokens are mostly replaceable expressions. Based on this insight, we propose ProFit, which selectively masks low-probability tokens to prevent surface-level overfitting. Extensive experiments confirm that ProFit consistently outperforms traditional SFT baselines on general reasoning and mathematical benchmarks.
Authors: Rapha\"el Sarfati, Eric Bigelow, Daniel Wurgaft, Siddharth Boppana, Jack Merullo, Atticus Geiger, Owen Lewis, Tom McGrath, Ekdeep Singh Lubana
Abstract: Large language models (LLMs) form implicit beliefs (posteriors over latent variables) from prompts, but we lack a mechanistic account of how these beliefs are encoded in representation space, how they update with new evidence, and how interventions reshape them. We study a controlled setting in which Llama-3.2 infers the parameters of a normal distribution from in-context samples. We show that parameter posteriors are encoded as curved manifolds in representation space, and trace how they evolve along the prompt. Standard linear steering moves representations off-manifold, inducing unintended, coupled changes, whereas geometry-aware methods preserve the target belief family. Our work demonstrates an example of linear field probing (LFP) as a principled approach to tile the data manifold and make interventions that respect the underlying geometry. Our results suggest that LLM beliefs are inherently geometric objects, and that globally linear representations are often inadequate abstractions.
Authors: Hao Fang, Tianyi Zhang, Tianqu Zhuang, Jiawei Kong, Kuofeng Gao, Bin Chen, Leqi Zheng, Shu-Tao Xia, Ke Xu
Abstract: Proprietary large language models (LLMs) embody substantial economic value and are generally exposed only as black-box APIs, yet adversaries can still exploit their outputs to extract knowledge via distillation. Existing defenses focus exclusively on text-based distillation, leaving the important logit-based distillation largely unexplored. In this work, we analyze this problem and present an effective solution from an information-theoretic perspective. We characterize distillation-relevant information in teacher outputs using the conditional mutual information (CMI) between teacher logits and input queries conditioned on ground-truth labels. This quantity captures contextual information beneficial for model extraction, motivating us to defend distillation via CMI minimization. Guided by our theoretical analysis, we propose learning a transformation matrix that purifies the original outputs to enhance distillation resistance. We further derive a CMI-inspired anti-distillation objective to optimize this transformation, which effectively removes distillation-relevant information while preserving output utility. Extensive experiments across multiple LLMs and strong distillation algorithms demonstrate that the proposed method significantly degrades distillation performance while preserving task accuracy, effectively protecting models' intellectual property.
Authors: Yining Lu, Meng Jiang
Abstract: We study a persistent failure mode in multi-objective alignment for large language models (LLMs): training improves performance on only a subset of objectives while causing others to degrade. We formalize this phenomenon as cross-objective interference and conduct the first systematic study across scalarization algorithms, showing that interference is pervasive and exhibits strong model dependence. To explain this phenomenon, we derive a local covariance law showing that an objective improves when its reward exhibits positive covariance with the scalarized score. We extend this analysis to clipped surrogate objectives used in modern alignment, demonstrating that the covariance law remains valid under mild conditions despite clipping. Building on this analysis, we propose Covariance Targeted Weight Adaptation (CTWA), a plug-and-play method that maintains positive covariance between objective rewards and the training signal to effectively mitigate cross-objective interference. Finally, we complement these local improvement conditions with a global convergence analysis under the Polyak--\L{}ojasiewicz condition, establishing when non-convex scalarized optimization achieves global convergence and how cross-objective interference depends on specific model geometric properties.
Authors: Sasha Boguraev, Qing Yao, Kyle Mahowald
Abstract: Sentences like "She will go to France or Spain, or perhaps to Germany or France." appear formally redundant, yet become acceptable in contexts such as "Mary will go to a philosophy program in France or Spain, or a mathematics program in Germany or France." While this phenomenon has typically been analyzed using symbolic formal representations, we aim to provide an account grounded in artificial neural mechanisms. We first present new behavioral evidence from humans and large language models demonstrating the robustness of this apparent non-redundancy across contexts. We then show that, in language models, redundancy avoidance arises from two interacting mechanisms: models learn to bind contextually relevant information to repeated lexical items, and Transformer induction heads selectively attend to these context-licensed representations. We argue that this neural explanation sheds light on the mechanisms underlying context-sensitive semantic interpretation, and that it complements existing symbolic analyses.
Authors: Yingtao Shen, An Zou
Abstract: Large Language Models (LLMs) have demonstrated exceptional performance across diverse domains but are increasingly constrained by high inference latency. Early Exit has emerged as a promising solution to accelerate inference by dynamically bypassing redundant layers. However, in decoder-only architectures, the efficiency of Early Exit is severely bottlenecked by the KV Cache Absence problem, where skipped layers fail to provide the necessary historical states for subsequent tokens. Existing solutions, such as recomputation or masking, either introduce significant latency overhead or incur severe precision loss, failing to bridge the gap between theoretical layer reduction and practical wall-clock speedup. In this paper, we propose River-LLM, a training-free framework that enables seamless token-level Early Exit. River-LLM introduces a lightweight KV-Shared Exit River that allows the backbone's missing KV cache to be naturally generated and preserved during the exit process, eliminating the need for costly recovery operations. Furthermore, we utilize state transition similarity within decoder blocks to predict cumulative KV errors and guide precise exit decisions. Extensive experiments on mathematical reasoning and code generation tasks demonstrate that River-LLM achieves 1.71 to 2.16 times of practical speedup while maintaining high generation quality.
Authors: Eoghan Cunningham, Derek Greene, James Cross, Antonio Rago
Abstract: Understanding how policy is debated and justified in parliament is a fundamental aspect of the democratic process. However, the volume and complexity of such debates mean that outside audiences struggle to engage. Meanwhile, Large Language Models (LLMs) have been shown to enable automated summarisation at scale. While summaries of debates can make parliamentary procedures more accessible, evaluating whether these summaries faithfully communicate argumentative content remains challenging. Existing automated summarisation metrics have been shown to correlate poorly with human judgements of consistency (i.e., faithfulness or alignment between summary and source). In this work, we propose a formal framework for evaluating parliamentary debate summaries that grounds argument structures in the contested proposals up for debate. Our novel approach, driven by computational argumentation, focuses the evaluation on formal properties concerning the faithful preservation of the reasoning presented to justify or oppose policy outcomes. We demonstrate our methods using a case-study of debates from the European Parliament and associated LLM-driven summaries.
Authors: Hagai Astrin, Ayal Swaid, Avi Segal, Kobi Gal
Abstract: Mental health challenges are increasing worldwide, straining emotional support services and leading to counselor overload. This can result in delayed responses during critical situations, such as suicidal ideation, where timely intervention is essential. While large language models (LLMs) have shown strong generative capabilities, their application in low-resource languages, especially in sensitive domains like mental health, remains underexplored. Furthermore, existing LLM-based agents often struggle to replicate the supportive language and intervention strategies used by professionals due to a lack of training on large-scale, real-world datasets. To address this, we propose CARE (Counselor-Aligned Response Engine), a GenAI framework that assists counselors by generating real-time, psychologically aligned response recommendations. CARE fine-tunes open-source LLMs separately for Hebrew and Arabic using curated subsets of real-world crisis conversations. The training data consists of sessions rated as highly effective by professional counselors, enabling the models to capture interaction patterns associated with successful de-escalation. By training on complete conversation histories, CARE maintains the evolving emotional context and dynamic structure of counselor-help-seeker dialogue. In experimental settings, CARE demonstrates stronger semantic and strategic alignment with gold-standard counselor responses compared to non-specialized LLMs. These findings suggest that domain-specific fine-tuning on expert-validated data can significantly support counselor workflows and improve care quality in low-resource language contexts.
Authors: Shouren Wang, Wang Yang, Chuang Ma, Debargha Ganguly, Vikash Singh, Chaoda Song, Xinpeng Li, Xianxuan Long, Vipin Chaudhary, Xiaotian Han
Abstract: Hybrid-thinking language models expose explicit think and no-think modes, but current designs do not separate them cleanly. Even in no-think mode, models often emit long and self-reflective responses, causing reasoning leakage. Existing work reduces this issue through better data curation and multi-stage training, yet leakage remains because both modes are still encoded in the same feed-forward parameters. We propose Path-Lock Expert (PLE), an architecture-level solution that replaces the single MLP in each decoder layer with two semantically locked experts, one for think and one for no-think, while keeping attention, embeddings, normalization, and the language-model head shared. A deterministic control-token router selects exactly one expert path for the entire sequence, so inference preserves the dense model's per-token computation pattern and each expert receives mode-pure updates during supervised fine-tuning. Across math and science reasoning benchmarks, PLE maintains strong think performance while producing a substantially stronger no-think mode that is more accurate, more concise, and far less prone to reasoning leakage. On Qwen3-4B, for example, PLE reduces no-think reflective tokens on AIME24 from 2.54 to 0.39 and improves no-think accuracy from 20.67% to 40.00%, all while preserving think-mode performance. These results suggest that controllable hybrid thinking is fundamentally an architectural problem, and separating mode-specific feed-forward pathways is a simple and effective solution.
Authors: Jiawei Wu, Doudou Zhou
Abstract: Machine unlearning has emerged as a critical capability for addressing privacy, safety, and regulatory concerns in large language models (LLMs). Existing methods operate at the sequence level, applying uniform updates across all tokens despite only a subset encoding the knowledge targeted for removal. This introduces gradient noise, degrades utility, and leads to suboptimal forgetting. We propose TokenUnlearn, a token-level attribution framework that identifies and selectively targets critical tokens. Our approach combines knowledge-aware signals via masking, and entropy-aware signals to yield importance scores for precise token selection. We develop two complementary strategies: hard selection, applying unlearning only to high-importance tokens, and soft weighting, modulating gradient contributions based on importance scores. Both extend existing methods to token-level variants. Theoretical analysis shows token-level selection improves gradient signal-to-noise ratio. Experiments on TOFU and WMDP benchmarks across three model architectures demonstrate consistent improvements over sequence-level baselines in both forgetting effectiveness and utility preservation.
Authors: Jiaqi Wei, Xuehang Guo, Pengfei Yu, Xiang Zhang, Wanli Ouyang, Siqi Sun, Qingyun Wang, Chenyu You
Abstract: In single-stream autoregressive interfaces, the same tokens both update the model state and constitute an irreversible public commitment. This coupling creates a silence tax: additional deliberation postpones the first task-relevant content, while naive early streaming risks premature commitments that bias subsequent generations. We introduce Side-by-Side (SxS) Interleaved Reasoning, which makes disclosure timing a controllable decision within standard autoregressive generation. SxS interleaves partial disclosures with continued private reasoning in the same context, but releases content only when it is supported by the reasoning so far. To learn such pacing without incentivizing filler, we construct entailment-aligned interleaved trajectories by matching answer prefixes to supporting reasoning prefixes, then train with SFT to acquire the dual-action semantics and RL to recover reasoning performance under the new format. Across two Qwen3 architectures/scales (MoE Qwen3-30B-A3B, dense Qwen3-4B) and both in-domain (AIME25) and out-of-domain (GPQA-Diamond) benchmarks, SxS improves accuracy--content-latency Pareto trade-offs under token-level proxies such as inter-update waiting.
Authors: Robert Dilworth
Abstract: In this study, we more rigorously evaluated our attack script $\textit{TraceTarnish}$, which leverages adversarial stylometry principles to anonymize the authorship of text-based messages. To ensure the efficacy and utility of our attack, we sourced, processed, and analyzed Reddit comments -- comments that were later alchemized into $\textit{TraceTarnish}$ data -- to gain valuable insights. The transformed $\textit{TraceTarnish}$ data was then further augmented by $\textit{StyloMetrix}$ to manufacture stylometric features -- features that were culled using the Information Gain criterion, leaving only the most informative, predictive, and discriminative ones. Our results found that function words and function word types ($L\_FUNC\_A$ $\&$ $L\_FUNC\_T$); content words and content word types ($L\_CONT\_A$ $\&$ $L\_CONT\_T$); and the Type-Token Ratio ($ST\_TYPE\_TOKEN\_RATIO\_LEMMAS$) yielded significant Information-Gain readings. The identified stylometric cues -- function-word frequencies, content-word distributions, and the Type-Token Ratio -- serve as reliable indicators of compromise (IoCs), revealing when a text has been deliberately altered to mask its true author. Similarly, these features could function as forensic beacons, alerting defenders to the presence of an adversarial stylometry attack; granted, in the absence of the original message, this signal may go largely unnoticed, as it appears to depend on a pre- and post-transformation comparison. "In trying to erase a trace, you often imprint a larger one." Armed with this understanding, we framed $\textit{TraceTarnish}$'s operations and outputs around these five isolated features, using them to conceptualize and implement enhancements that further strengthen the attack.
Authors: Qiming Bao, Xiaoxuan Fu, Michael Witbrock
Abstract: Large language models (LLMs) achieve high accuracy on many reasoning benchmarks but remain brittle under structural perturbations of rule-based systems. We introduce a diagnostic framework with four stress tests -- redundant vs. essential rule deletion, contradictory-rule injection, logic-preserving rewrites, and multi-law stacking -- and use it to expose Logic Inertia: the tendency of generative LLMs (Qwen2/3, TinyLlama, GPT-4o, Gemma-3-4B-IT) and the encoder-only BERT baseline to persist along learned deductive trajectories under inconsistent premises. The collapse is sharp: untreated baselines fall from accuracy 1.00 on the base task to 0.00 on contradiction injection (instance-level exact match), and GPT-4o resolves only 56.0% of contradiction cases. We propose Conflict-Aware Fusion, a four-stage training pipeline that enforces verification-before-deduction as a learned structural prior: (i) SFT establishes the verification preamble; (ii) DPO sharpens the halt-on-contradiction decision boundary; (iii) Logical Invariance REgularisation (LIRE) penalises divergence between logically equivalent rule formulations via symmetric KL; (iv) Reinforcement Learning from Verification Feedback (RLVF) uses a symbolic forward-chaining engine as a deterministic oracle reward, jointly optimising invariance and sensitivity. The pipeline saturates all four primary stress tests for both 1.5B and 8B backbones. We further validate a Phase 2 extension that replaces the propositional oracle with a Lean 4 kernel, attaining 99.0% kernel agreement on the 105 classically-derivable (T) questions within a stratified 187-question Lean-translated sample (overall 71.7% across both polarities), providing a sound upgrade path to formally verified RL training. Code and benchmark: https://github.com/14H034160212/lemo
Authors: Dingwei Zhu, Zhiheng Xi, Shihan Dou, Jiahan Li, Chenhao Huang, Junjie Ye, Sixian Li, Mingxu Chai, Yuhui Wang, Yajie Yang, Ming Zhang, Jiazheng Zhang, Shichun Liu, Caishuang Huang, Yunke Zhang, Yuran Wang, Tao Gui, Xipeng Qiu, Qi Zhang, Xuanjing Huang
Abstract: Training reinforcement learning (RL) systems in real-world environments remains challenging due to noisy supervision and poor out-of-domain (OOD) generalization, especially in LLM post-training. Recent distributional RL methods improve robustness by modeling values with multiple quantile points, but they still learn each quantile independently as a scalar. This results in rough-grained value representations that lack fine-grained conditioning on state information, struggling under complex and OOD conditions. We propose DFPO (Distributional Value Flow Policy Optimization with Conditional Risk and Consistency Control), a robust distributional RL framework that models values as continuous flows across time steps. By scaling value modeling through learning of a value flow field instead of isolated quantile predictions, DFPO captures richer state information for more accurate advantage estimation. To stabilize training under noisy feedback, DFPO further integrates conditional risk control and consistency constraints along value flow trajectories. Experiments on dialogue, math reasoning, and scientific tasks show that DFPO outperforms PPO, FlowRL, and other robust baselines under noisy supervision, achieving improved training stability and generalization.
Authors: Nikolai Ludwig, Wasi Uddin Ahmad, Somshubra Majumdar, Boris Ginsburg
Abstract: We introduce SWE-ZERO to SWE-HERO, a two-stage SFT recipe that achieves state-of-the-art results on SWE-bench by distilling open-weight frontier LLMs. Our pipeline replaces resource-heavy dependencies with an evolutionary refinement strategy: (1) SWE-ZERO utilizes large-scale, execution-free trajectories to master code semantics and repository-level reasoning, and (2) SWE-HERO applies targeted, execution-backed refinement to transition these semantic intuitions into rigorous engineering workflows. Our empirical results set a new benchmark for open-source models of comparable size. We release a dataset of 300k SWE-ZERO and 13k SWE-HERO trajectories distilled from Qwen3-Coder-480B, alongside a suite of agents based on the Qwen2.5-Coder series. Notably, SWE-HERO-32B achieves a 62.2% resolution rate on SWE-bench Verified. Furthermore, despite being trained exclusively on Python, our agents demonstrate robust zero-shot transferability on SWE-bench Multilingual, reaching 44.1% and confirming the paradigm's generalizability across diverse languages.
Authors: Matthew Penaroza
Abstract: Language model agents reason from scratch on every query, discarding their chain of thought after each run. The result is lower accuracy and high run-to-run variance. We introduce reasoning graphs, which persist the per-evidence chain of thought as structured edges. Unlike prior memory that retrieves distilled strategies by query similarity, reasoning graphs enable evidence-centric feedback: for every candidate item, the system traverses all incoming evaluation edges across prior runs to surface how that specific item has been judged before. We further introduce retrieval graphs, which feed a planner that prunes consistently-rejected candidates over successive runs. Together they form a ROZA graph: a self-improving feedback loop in which accuracy gains scale with gold-passage reuse (reasoning graph) and efficiency gains scale with candidate-pool overlap (retrieval graph). The base model remains frozen; all gains come from context engineering via graph traversal. We evaluate on MuSiQue and HotpotQA, plus a high-reuse deployment subset. Four findings stand out. (1) Dose-response: accuracy improves monotonically with evidence-profile coverage, reaching +10.6pp over Vanilla RAG at 50%+ coverage on the same questions (47% error reduction, $p<0.0001$; per-question Spearman $\rho=+0.144$, $p<10^{-6}$, $n=1{,}100$). (2) Multi-hop scaling: 4-hop accuracy improves by +11.0pp ($p=0.0001$). (3) Cross-cluster prediction: the cluster-level gain is predicted by gold-passage reuse density ($r=0.604$, $p=0.001$, $n=26$ clusters). (4) High-reuse Pareto dominance: highest or tied-for-highest accuracy alongside 46% lower cost and 46% lower latency. Per-passage decision consistency across repeated runs ($N=73$ paired probes, $K=10$ runs each, two model families, three temperatures) rises by +8 to +13pp on a fixed 20-passage context and by +12 to +21pp when the retrieval graph also prunes (all $p<0.005$).
Authors: Jon M Laurent, Albert Bou, Michael Pieler, Conor Igoe, Alex Andonian, Siddharth Narayanan, James Braza, Alexandros Sanchez Vassopoulos, Jacob L Steenwyk, Blake Lash, Andrew D White, Samuel G Rodriques
Abstract: Optimism for accelerating scientific discovery with AI continues to grow. Current applications of AI in scientific research range from training dedicated foundation models on scientific data to agentic autonomous hypothesis generation systems to AI-driven autonomous labs. The need to measure progress of AI systems in scientific domains correspondingly must not only accelerate, but increasingly shift focus to more real-world capabilities. Beyond rote knowledge and even just reasoning to actually measuring the ability to perform meaningful work. Prior work introduced the Language Agent Biology Benchmark LAB-Bench as an initial attempt at measuring these abilities. Here we introduce an evolution of that benchmark, LABBench2, for measuring real-world capabilities of AI systems performing useful scientific tasks. LABBench2 comprises nearly 1,900 tasks and is, for the most part, a continuation of LAB-Bench, measuring similar capabilities but in more realistic contexts. We evaluate performance of current frontier models, and show that while abilities measured by LAB-Bench and LABBench2 have improved substantially, LABBench2 provides a meaningful jump in difficulty (model-specific accuracy differences range from -26% to -46% across subtasks) and underscores continued room for performance improvement. LABBench2 continues the legacy of LAB-Bench as a de facto benchmark for AI scientific research capabilities and we hope that it continues to help advance development of AI tools for these core research functions. To facilitate community use and development, we provide the task dataset at https://huggingface.co/datasets/futurehouse/labbench2 and a public eval harness at https://github.com/EdisonScientific/labbench2.
URLs: https://huggingface.co/datasets/futurehouse/labbench2, https://github.com/EdisonScientific/labbench2.
Authors: Wei Jiang, Wei Wang
Abstract: Sub-token routing provides a finer compression axis for transformer efficiency than the coarse units used in most prior work, such as tokens, pages, heads, or layers. In this paper, we study routing within a token representation itself in LoRA-adapted transformers. We consider two settings. In the query-independent setting, we combine routed subspace LoRA with value-group routing on the KV path for compression-aware language modeling. In the query-aware setting, we use a predictor-based selector to allocate a global retention budget over context-token/value-group pairs using query-conditioned relevance. Experiments show that the query-independent design improves language-model quality under reduced KV budgets, while the query-aware design preserves downstream behavior well under KV compression. We further show that sub-token routing is most effective as a complementary compression axis to token-level query-aware selection: token-level methods decide which tokens survive globally, while sub-token routing determines how the surviving tokens are compressed internally. Their combination enables deeper KV compression at nearly unchanged task accuracy.
Authors: Samer Attrah
Abstract: We present Code Broker, a multi agent system built on Google s Agent Development Kit ADK that analyses Python source code from individual files, local directory trees, or remote GitHub repositories and generates structured, actionable quality assessment reports. The system realises a hierarchical five agent architecture in which a root orchestrator coordinates a sequential pipeline agent that, in turn, dispatches three specialised agents concurrently a Correctness Assessor, a Style Assessor, and a Description Generator before synthesising their findings through an Improvement Recommender. Reports quantify four quality dimensions correctness, security, style, and maintainability on a normalised scale and are rendered in both Markdown and HTML for integration into diverse developer workflows. Code Broker fuses LLM based semantic reasoning with deterministic static analysis signals from Pylint, employs asynchronous execution with exponential backoff retry logic to improve robustness under transient API failures, and explores lightweight session memory for retaining and querying prior assessment context across runs. We frame this paper as a technical report on system design, prompt engineering, and tool orchestration, and present a preliminary qualitative evaluation on representative Python codebases of varying scale. The results indicate that parallel specialised agents produce readable, developer oriented feedback that complements traditional linting, while also foregrounding current limitations in evaluation depth, security tooling, large repository handling, and the exclusive reliance on in memory persistence. All code and reproducibility materials are publicly available: https://github.com/Samir-atra/agents_intensive_dev.
Authors: Audrey Cherilyn, Houman Safaai
Abstract: We study the organization of channel-level importance in transformer feed-forward networks (FFNs). Using a Fisher-style loss proxy (LP) based on activation-gradient second moments, we show that loss sensitivity is concentrated in a small set of channels within each layer. In Llama-3.1-8B, the top 1% of channels per layer accounts for a median of 58.7% of LP mass, with a range of 33.0% to 86.1%. We call these loss-critical channels supernodes. Although FFN layers also contain strong activation outliers, LP-defined supernodes overlap only weakly with activation-defined outliers and are not explained by activation power or weight norms alone. Around this core, we find a weaker but consistent halo structure: some non-supernode channels share the supernodes' write support and show stronger redundancy with the protected core. We use one-shot structured FFN pruning as a diagnostic test of this organization. At 50% FFN sparsity, baselines that prune many supernodes degrade sharply, whereas our SCAR variants explicitly protect the supernode core; the strongest variant, SCAR-Prot, reaches perplexity 54.8 compared with 989.2 for Wanda-channel. The LP-concentration pattern appears across Mistral-7B, Llama-2-7B, and Qwen2-7B, remains visible in targeted Llama-3.1-70B experiments, and increases during OLMo-2-7B pretraining. These results suggest that LLM FFNs develop a small learned core of loss-critical channels, and that preserving this core is important for reliable structured pruning.
Authors: William Oliveira
Abstract: On-device Small Language Models (SLMs) promise fully offline, private AI experiences for mobile users (no cloud dependency, no data leaving the device). But is this promise achievable in practice? This paper presents a longitudinal practitioner case study documenting the engineering challenges of integrating SLMs (Gemma 4 E2B, 2.6B parameters; Qwen3 0.6B, 600M parameters) into Palabrita, a production Android word-guessing game. Over a 5-day development sprint comprising 204 commits (~90 directly AI-related), the system underwent a radical transformation: from an ambitious design where the LLM generated complete structured puzzles (word, category, difficulty, and five hints as JSON) to a pragmatic architecture where curated word lists provide the words and the LLM generates only three short hints, with a deterministic fallback if it fails. We identify five categories of failures specific to on-device SLM integration: output format violations, constraint violations, context quality degradation, latency incompatibility, and model selection instability. For each failure category, we document the observed symptoms, root causes, and the prompt engineering and architectural strategies that effectively mitigated them, including multi-layer defensive parsing, contextual retry with failure feedback, session rotation, progressive prompt hardening, and systematic responsibility reduction. Our findings demonstrate that on-device SLMs are viable for production mobile applications, but only when the developer accepts a fundamental constraint: the most reliable on-device LLM feature is one where the LLM does the least. We distill our experience into eight actionable design heuristics for practitioners integrating SLMs into mobile apps.
Authors: Yida Xue, Ningyu Zhang, Tingwei Wu, Zhe Ma, Daxiong Ji, Zhao Wang, Guozhou Zheng, Huajun Chen
Abstract: The vast and underexplored ocean plays a critical role in regulating global climate and supporting marine biodiversity, yet artificial intelligence has so far delivered limited impact in this domain due to a fundamental data bottleneck. Specifically, ocean data are highly fragmented across disparate sources and inherently exhibit multi-modal, high-noise, and weakly labeled characteristics, lacking unified schemas and semantic alignment. Although Multimodal Large Language Models (MLLMs) have achieved remarkable success in general domains, their application to ocean science remains severely constrained by the absence of large-scale, well-aligned multimodal datasets tailored to marine environments. To bridge this gap, we introduce OceanPile, a large-scale multimodal corpus designed for ocean foundation models. It comprises three key components: OceanCorpus, a unified collection integrating sonar data, underwater imagery, marine science visuals, and scientific text from diverse authoritative sources; OceanInstruction, a high-quality instruction dataset synthesized via a novel pipeline guided by a hierarchical Ocean Concept Knowledge Graph; and OceanBenchmark, a manually curated evaluation benchmark for rigorous assessment. We establish a multi-stage quality control process to ensure scientific validity and alignment across modalities. Experimental validation demonstrates significant performance improvements for models trained on our data. All datasets are publicly released to advance the field of marine artificial intelligence and empower domain-specific MLLMs.
Authors: Sen Fang, Hongbin Zhong, Yanxin Zhang, Dimitris N. Metaxas
Abstract: Existing large-scale sign language resources typically provide supervision only at the level of raw video-text alignment and are often produced in laboratory settings. While such resources are important for semantic understanding, they do not directly provide a unified interface for open-world recognition and translation, or for modern pose-driven sign language video generation frameworks: 1. RGB-based pretrained recognition models depend heavily on fixed backgrounds or clothing conditions during recording, and are less robust in open-world settings than style-agnostic pose-processing models. 2. Recent pose-guided image/video generation models mostly use a unified keypoint representation such as DWPose as their control interface. At present, the sign language field still lacks a data resource that can directly interface with this modern pose-native paradigm while also targeting real-world open scenarios. We present SignVerse-2M, a large-scale multilingual pose-native dataset for sign language pose modeling and evaluation. Built from publicly available multilingual sign language video resources, it applies DWPose in a unified preprocessing pipeline to convert raw videos into 2D pose sequences that can be used directly for modeling, resulting in a consolidated corpus of about two million clips covering more than 55 sign languages. Unlike many laboratory datasets, this resource preserves the recording conditions and speaker diversity of real-world videos while reducing appearance variation through a unified pose representation. Toward this goal, we further provide the data construction pipeline, task definitions, and a simple SignDW Transformer baseline, demonstrating the feasibility of this resource for multilingual pose-space modeling and its compatibility with modern pose-driven pipelines, while discussing the evaluation claims it can support as well as its current limitations.
Authors: David Wilmot
Abstract: Stories hold a reader's attention because they have causes, secrets, and consequences. Shadow-Loom is an experimental open-source framework that turns a narrative into a versioned graphical world model and lets two engines act on it: a causal physics grounded in Pearl's ladder of causation and a recently proposed counterfactual calculus over Ancestral Multi-World Networks; and a narrative physics that scores the same graph against four structural reader-states -- mystery, dramatic irony, suspense, and surprise -- in the tradition of Sternberg's curiosity/suspense/surprise triad, with suspense formalised in the structural-affect line of work on story comprehension and computational suspense. Large language models are used only at the boundary: extraction, rendering, and audit; identification, intervention, and counterfactual reasoning are carried out in typed code over the graph. The system is offered as a research artefact rather than as a benchmarked NLP model; code, fixtures, and pipeline are released open source.
Authors: Jinliang Xu
Abstract: As the ecosystem of Large Language Model (LLM)-based agents expands rapidly, efficient and accurate Agent Discovery becomes a critical bottleneck for large-scale multi-agent collaboration. Existing approaches typically face a dichotomy: either relying on heavy-weight LLMs for intent parsing, leading to prohibitive latency (often exceeding 30 seconds), or using monolithic vector retrieval that sacrifices semantic precision for speed. To bridge this gap, we propose \textbf{GRAIL} (Granular Resonance-based Agent/AI Link), a novel framework achieving sub-400ms discovery latency without compromising accuracy. GRAIL introduces three key innovations: (1) \textbf{SLM-Enhanced Prediction}, replacing the generalized LLM parser with a specialized, fine-tuned Small Language Model (SLM) for millisecond-level capability tag prediction; (2) \textbf{Pseudo-Document Expansion}, augmenting agent descriptions with synthetic queries to enhance semantic density for robust dense retrieval; and (3) \textbf{MaxSim Resonance}, a fine-grained matching mechanism computing maximum similarity between user queries and discrete agent usage examples, effectively mitigating semantic dilution. Validated on \textbf{AgentTaxo-9K}, our new large-scale dataset of 9,240 agents, GRAIL reduces end-to-end discovery latency by over \textbf{79$\times$} compared to LLM-parsing baselines, while significantly outperforming traditional vector search in Recall@10. This framework offers a scalable, industrial-grade solution for the real-time ``Internet of Agents."
Authors: Fan Ma, Yuntian Liu, Xiang Lan, Weipeng Zhou, Jun Ni, Mauro Giuffr\`e, Lingfei Qian, Xueqing Peng, Yujia Zhou, Ruey-Ling Weng, Huan He, Lu Li, Huiyuan Wang, Qingyu Chen, Andrew Loza, Laila Rasmy, Degui Zhi, Yuan Lu, Chenjie Zeng, Joshua C Denny, Lee Schwamm, Daniella Meeker, Lucila Ohno-Machado, Yong Chen, Hua Xu
Abstract: Evidence derived from large-scale real-world data (RWD) is increasingly informing regulatory evaluation and healthcare decision-making. Administrative claims provide population-scale, longitudinal records of healthcare utilization, expenditure, and detailed coding of diagnoses, procedures, and medications, yet their potential as a substrate for healthcare foundation models remains largely unexplored. Here we present ReClaim, a generative transformer trained from scratch on 43.8 billion medical events from more than 200 million enrollees in the MarketScan claims data spanning 2008-2022. ReClaim models longitudinal trajectories across diagnoses, procedures, medications, and expenditure, and was scaled to 140 million, 700 million, and 1.7 billion parameters. Across over 1,000 disease-onset prediction tasks, ReClaim achieved a mean AUC of 75.6%, substantially outperforming disease-specific LightGBM (66.3%) and the transformer-based Delphi model (69.4%), with the largest gains for rare diseases. These advantages held across retrospective and prospective evaluations and in external validation on two independent datasets. Performance improved monotonically with scale, and post-training added 13.8 percentage points over pre-training alone. Beyond disease prediction, ReClaim captured financial outcomes and improved real-world evidence (RWE) analyses: for healthcare expenditure forecasting it increased explained variance from 0.28 to 0.37 relative to LightGBM, and in a target trial emulation it reduced systematic bias by 72% on average relative to Delphi. Together, these results establish administrative claims as a scalable substrate for healthcare foundation models and show that learned representations generalize across time periods and data sources, supporting disease surveillance, expenditure forecasting, and RWE generation.
Authors: Cheng Qian, Hyeonjeong Ha, Jiayu Liu, Jeonghwan Kim, Jiateng Liu, Bingxuan Li, Aditi Tiwari, Dwip Dalal, Zhenhailong Wang, Xiusi Chen, Mahdi Namazifar, Yunzhu Li, Heng Ji
Abstract: Recent advances in large language models have led to strong performance on reasoning and environment-interaction tasks, yet their ability for creative problem-solving remains underexplored. We study this capability through the lens of creative tool use, where a model repurposes available objects by reasoning about their affordances and attributes rather than relying on canonical usage. As a first step, we introduce CreativityBench, a benchmark for evaluating affordance-based creativity in LLMs. To this end, we build a large-scale affordance knowledge base (KB) with 4K entities and 150K+ affordance annotations, explicitly linking objects, parts, attributes, and actionable uses. Building on this KB, we generate 14K grounded tasks that require identifying non-obvious yet physically plausible solutions under constraints. Evaluations across 10 state-of-the-art LLMs, including closed and open-source models, show that models can often select a plausible object, but fail to identify the correct parts, their affordances, and the underlying physical mechanism needed to solve the task, leading to a significant drop in performance. Furthermore, improvements from model scaling quickly saturate, strong general reasoning does not reliably translate to creative affordance discovery, and common inference-time strategies such as Chain-of-Thought yield limited gains. These results suggest that creative tool use remains a major challenge for current models, and that CreativityBench provides a useful testbed for studying this missing dimension of intelligence, with potential implications for planning and reasoning modules in future agents.
Authors: Alexandria K. Vail, Marcelo Cicconet, Katie Aafjes-van Doorn, Ryan Maroney, Marc Aafjes
Abstract: Modeling latent clinical constructs from unconstrained clinical interactions is a unique challenge in affective computing. We present ADAPTS (Agentic Decomposition for Automated Protocol-agnostic Tracking of Symptoms), a framework for automated rating of depression and anxiety severity using a mixture-of-agents LLM architecture. This approach decomposes long-form clinical interviews into symptom-specific reasoning tasks, producing auditable justifications while preserving temporal and speaker alignment. Generalization was evaluated across two independent datasets ($N=204$) with distinct interview structures. On high-discrepancy interviews, automated ratings approximated expert benchmarks ($\text{absolute error}=22$) more closely than original human ratings ($\text{absolute error}=26$). Implementing an ``extended'' protocol that incorporates qualitative clinical conventions significantly stabilized ratings, with absolute agreement reaching $\text{ICC(2,1)} = 0.877$. These findings suggest that the ADAPTS framework enables promising evaluations of psychiatric severity. While the current implementation is purely text-based, the underlying architecture is readily extensible to multimodal inputs, including acoustic and visual features. By approximating expert-level precision in a protocol-agnostic manner, this framework provides a foundation for objective and scalable psychiatric assessment, especially in resource-limited settings.
Authors: Tianyang Han, Hengyu Shi, Junjie Hu, Xu Yang, Zhiling Wang, Junhao Su
Abstract: Reinforcement learning with verifiable rewards has become a common way to improve explicit reasoning in large language models, but final-answer correctness alone does not reveal whether the reasoning trace is faithful, reliable, or useful to the model that consumes it. This outcome-only signal can reinforce traces that are right for the wrong reasons, overstate reasoning gains by rewarding shortcuts, and propagate flawed intermediate states in multi-step systems. To this end, we propose TraceLift, a planner-executor training framework that treats reasoning as a consumable intermediate artifact. During planner training, the planner emits tagged reasoning. A frozen executor turns this reasoning into the final artifact for verifier feedback, while an executor-grounded reward shapes the intermediate trace. This reward multiplies a rubric-based Reasoning Reward Model (RM) score by measured uplift on the same frozen executor, crediting traces that are both high-quality and useful. To make reasoning quality directly learnable, we introduce TRACELIFT-GROUPS, a rubric-annotated reason-only dataset built from math and code seed problems. Each example is a same-problem group containing a high-quality reference trace and multiple plausible flawed traces with localized perturbations that reduce reasoning quality or solution support while preserving task relevance. Extensive experiments on code and math benchmarks show that this executor-grounded reasoning reward improves the two-stage planner-executor system over execution-only training, suggesting that reasoning supervision should evaluate not only whether a trace looks good, but also whether it helps the model that consumes it. Our code is available at: https://github.com/MasaiahHan/TraceLift
Authors: Skye Gunasekaran, T\'ea Wright, Rui-Jie Zhu, Jason Eshraghian
Abstract: Several recent Transformer architectures expose later layers to representations computed in the earliest layers, motivated by the observation that low-level features can become harder to recover as the residual stream is repeatedly transformed through depth. The cheapest among these methods add static value residuals: learned mixing coefficients that expose the first-layer value projection V_1 uniformly across tokens and heads. More expressive dense or dynamic alternatives recover finer-grained access, but at higher memory cost and lower throughput. The usefulness of V_1 is unlikely to be constant across tokens, heads, and contexts; different positions plausibly require different amounts of access to early lexical or semantic information. We therefore treat early-representation reuse as a retrieval problem rather than a connectivity problem, and introduce Selective Access Transformer (SATFormer), which preserves the first-layer value pathway while controlling access with a context-dependent gate. Across models from 130M to 1.3B parameters, SATFormer consistently improves validation loss and zero-shot accuracy over the static value-residual and Transformer baselines. Its strongest gains appear on retrieval-intensive benchmarks, where it improves over static value residuals by approximately 1.5 average points, while maintaining throughput and memory usage close to the baseline Transformer. Gate analyses suggest sparse, depth-dependent, head-specific, and category-sensitive access patterns, supporting the interpretation that SATFormer learns selective reuse of early representations rather than uniform residual copying. Our code is available at https://github.com/SkyeGunasekaran/SATFormer.