Computation and Language
Guijin Son, Donghun Yang, Hitesh Laxmichand Patel et al. (+9 more)
Recent frontier models employ long chain-of-thought
reasoning to explore solution spaces in context and achieve stonger performance. While many works study
distillation to build smaller yet capable models, most focus on English and little is known about
language-specific
reasoning. To bridge this gap, we first introduct **
Language-Mixed CoT**, a
reasoning schema that switches between English and a target
language, using English as an anchor to excel in
reasoning while minimizing translation artificats. As a Korean case study, we curate **Yi-Sang**: 5.79M native-Korean
prompts from web Q&A, exams, STEM, and code; 3.7M long
reasoning traces generated from Qwen3-32B; and a targeted 260k high-yield subset. We train ninve models (4B-35B) across six families (Qwen2.5, Llama-3.1, Gemma-3, etc). Our best model, **KO-REAson-35B**, achieves state-of-the-art performance, with the highest overall average score (64.0 \pm 25), ranking first on 5/9
benchmarks and second on the remainder. Samller and mid-sized models also benefit substantially, with an average improvement of +18.6 points across teh evaluated nine
benchmarks. Ablations show **
Language-Mixed CoT** is more effective than monolingual CoT, also resulting in cross-lingual and mult-modal performance gains. We release our data-curation pipeline, evaluation system,
datasets, and models to advance research on
language-specific
reasoning. Data and model collection: https://huggingface.co/KOREAson.
Read the full abstract ▾
All authors Guijin Son, Donghun Yang, Hitesh Laxmichand Patel, Amit Agarwal, Hyunwoo Ko, Chanuk Lim, Srikant Panda, Minhyuk Kim, Nikunj Drolia, Dasol Choi, Kyong-Ha Lee, Youngjae Yu
Computation and Language
What and Whose Knowledge? Measuring Epistemic Diversity in Large Language Models
Dustin Wright, Sarah Masud, Jared Moore et al. (+5 more)
Large
language models (LLMs) are increasingly used as primary knowledge sources, yet their epistemic diversity - defined as the diversity of real-world claims in their outputs - has never been measured. Low epistemic diversity would pose a risk of knowledge collapse as homogeneous LLMs mediate a shrinking in the range of accessible information over time. The dominant paradigm is that overall LLM diversity is low, but this is always with respect to a single point in time, with no reference baseline or consideration for variation across countries. We address this gap in knowledge by performing the first systematic study of epistemic diversity in LLMs across time and cultural context, testing 27 LLMs on 155 topics covering 12 countries, resulting in 1.7M responses and 70M individual claims. We find that epistemic diversity has increased substantially over the past three years, a positive counter to recent diversity pessimism. However, despite progress, we find that every system is less diverse than a search baseline. This gap is not uniform: RAG can improve diversity, while large models are counterintuitively less diverse than smaller ones. Moreover, LLM parametric knowledge systematically reflects English over local-
language knowledge for country specific topics. Together, these results demonstrate that while progress on epistemic diversity is tangible, it is insufficient and unevenly distributed.
Read the full abstract ▾
All authors Dustin Wright, Sarah Masud, Jared Moore, Srishti Yadav, Maria Antoniak, Peter Ebert Christensen, Chan Young Park, Isabelle Augenstein
Computation and Language · Cross-listed from cs.CV
Locate-Then-Examine: Grounded Region Reasoning Improves Detection of AI-Generated Images
Yikun Ji, Yan Hong, Bowen Deng et al. (+5 more)
The rapid growth of AI-generated imagery has blurred the boundary between real and synthetic content, raising practical concerns for digital integrity.
Vision-
language models (VLMs) can provide natural
language explanations, but standard one-pass classifiers often miss subtle artifacts in high-quality synthetic images and offer limited grounding in the pixels. We propose Locate-Then-Examine (LTE), a two-stage VLM-based forensic framework that first localizes suspicious regions and then re-examines these crops together with the full image to refine the real vs. AI-generated verdict and its explanation. LTE explicitly links each decision to localized visual evidence through region proposals and region-aware
reasoning. To support training and evaluation, we introduce TRACE, a
dataset of 20,000 real and high-quality synthetic images with region-level annotations and automatically generated forensic explanations, constructed by a VLM-based pipeline with additional consistency checks and quality control. Across TRACE and multiple external
benchmarks, LTE achieves competitive accuracy and improved robustness while providing human-understandable, region-grounded explanations suitable for forensic deployment.
Read the full abstract ▾
All authors Yikun Ji, Yan Hong, Bowen Deng, Jun Lan, Huijia Zhu, Weiqiang Wang, Liqing Zhang, Jianfu Zhang
Computation and Language
Teaching LLM to be Persuasive: Reward-Enhanced Policy Optimization for Alignment from Heterogeneous Rewards
Xia Zeng, Yihan Chen, Luhui Liu et al. (+3 more)
We deploy large
language models (LLMs) as business development (BD)
agents for persuasive price negotiation in online travel agencies (OTAs). The
agent must follow a multi-stage Standard Operating Procedure (SOP) and strict guardrails (no over-promising and no
hallucinations), while remaining human-like and effective over long, multi-turn dialogues. We propose Reward-Enhanced Policy Optimization (REPO), a reinforcement learning post-training method that combines heterogeneous rewards: a preference-trained reward model (RM), an LLM-as-a-judge (RJ) for nuanced behaviors (e.g., emotional value and SOP compliance), and rule-based reward functions (RF) (mainly regex-based) for deterministic checks on numerics, formatting, and guardrails. In expert consensus evaluation (three human experts; 30 online conversations and 45 curated bad cases), REPO improves average dialogue rating to 4.63 (+0.33 over GRPO) and raises the share of conversations with at least one excellent response to 66.67% (+23.34 pp over GRPO), while achieving a 93.33% bad-case fix rate with 75.56% clean fixes. In a production A/B test on 9,653 real customer conversations (vs. an intent-driven dialogue system), REPO improves response rate by +12.14 pp and task success rate by +5.94 pp (p<0.001).
Read the full abstract ▾
All authors Xia Zeng, Yihan Chen, Luhui Liu, Chao Luo, Ye Chen, Zhuoran Zhuang
Computation and Language
CALM Before the STORM: Unlocking Native Reasoning for Optimization Modeling
Zhengyang Tang, Zihan Ye, Chenyu Huang et al. (+9 more)
Large
Reasoning Models (LRMs) have demonstrated strong capabilities in complex multi-step
reasoning, opening new opportunities for automating optimization modeling. However, existing domain adaptation methods, originally designed for earlier instruction-tuned models, often fail to exploit the advanced
reasoning patterns of modern LRMs – In particular, we show that direct fine-tuning on traditional non-reflective
datasets leads to limited gains. To fully leverage LRMs' inherent
reasoning abilities, we propose CALM (Corrective Adaptation with Lightweight Modification), a framework that progressively refines LRMs within their native
reasoning modes for optimization modeling tasks. In CALM, an expert intervener identifies
reasoning flaws and provides concise corrective hints, which the LRM incorporates to produce improved
reasoning trajectories. These interventions modify fewer than 2.6% of generated
tokens, but generate high-quality data for soft adaptation through supervised fine-tuning. The adapted model is then further improved through reinforcement learning. Building on CALM, we develop STORM (Smart Thinking Optimization
Reasoning Model), a 4B-parameter LRM that achieves a new state-of-the-art average accuracy of 68.9% across five popular optimization modeling
benchmarks, matching the performance of a 671B LRM. These results demonstrate that dynamic, hint-based data synthesis both preserves and amplifies the native
reasoning patterns of modern LRMs, offering a more effective and scalable path towards expert-level performance on challenging optimization modeling tasks.
Read the full abstract ▾
All authors Zhengyang Tang, Zihan Ye, Chenyu Huang, Xuhan Huang, Chengpeng Li, Sihang Li, Guanhua Chen, Ming Yan, Zizhuo Wang, Hongyuan Zha, Dayiheng Liu, Benyou Wang
Computation and Language
Thinking on the Fly: Test-Time Reasoning Enhancement via Latent Thought Policy Optimization
Wengao Ye, Yan Liang and Lianlei Shan
Recent advancements in Large
Language Models (LLMs) have shifted from explicit Chain-of-Thought (CoT)
reasoning to more efficient latent
reasoning, where intermediate thoughts are represented as vectors rather than text. However, latent
reasoning can be brittle on challenging, out-of-distribution tasks where robust
reasoning is most critical. To overcome these limitations, we introduce Latent Thought Policy Optimization (LTPO), a parameter-free framework that enhances LLM
reasoning entirely at test time, without requiring model parameter updates. LTPO treats intermediate latent "thought" vectors as dynamic parameters that are actively optimized for each problem instance. It employs an online policy gradient method guided by an intrinsic, confidence-based reward signal computed directly from the frozen LLM's own output distributions, eliminating the need for external supervision or expensive text generation during optimization. Extensive experiments on five
reasoning benchmarks show that LTPO not only matches or surpasses strong baselines on standard tasks but also demonstrates remarkable robustness where others fail. Most notably, on highly challenging AIME
benchmarks where existing latent
reasoning baselines collapse to near-zero accuracy, LTPO delivers substantial improvements, showcasing a unique capability for complex
reasoning.
Read the full abstract ▾
Computation and Language
Self Speculative Decoding for Diffusion Large Language Models
Yifeng Gao, Ziang Ji, Yuxuan Wang et al. (+3 more)
Diffusion-based Large
Language Models (dLLMs) have emerged as a competitive alternative to autoregressive models, offering unique advantages through bidirectional
attention and parallel generation paradigms. However, the generation results of current parallel decoding methods deviate from stepwise decoding, introducing potential performance degradation, which limits their practical deployment. To address this problem, we propose Self Speculative Decoding (SSD), a lossless inference acceleration method that leverages the dLLM itself as both speculative decoding drafter and verifier without auxiliary modules. SSD introduces a self-drafting mechanism where the model generates predictions for multiple positions, then verifies them through hierarchical verification trees in a single forward pass. Unlike traditional speculative decoding that requires separate draft models, SSD eliminates model redundancy and
memory overhead by exploiting the dLLM's inherent parallel prediction capability for multiple positions. This self-speculative approach allows the model to progressively verify and accept multiple
tokens in a single forward pass. Our experiments demonstrate that SSD achieves up to 3.46$\times$ speedup while keeping the output identical to stepwise decoding on open source models such as LLaDA and Dream. Code will be made publicly available on GitHub.
Read the full abstract ▾
All authors Yifeng Gao, Ziang Ji, Yuxuan Wang, Biqing Qi, Hanlin Xu, Linfeng Zhang
Computation and Language · Cross-listed from cs.LG
Beyond Next-Token Prediction: A Performance Characterization of Diffusion versus Autoregressive Language Models
Minseo Kim, Coleman Hooper, Aditya Tomar et al. (+5 more)
Large
Language Models (LLMs) have achieved state-of-the-art performance on a broad range of Natural
Language Processing (NLP) tasks, including document processing and code generation. Autoregressive
Language Models (ARMs), which generate
tokens sequentially conditioned on all previous
tokens, have been the predominant paradigm for LLMs. While these models have achieved high accuracy across a range of downstream tasks, they exhibit low arithmetic intensity due to the inherent sequential dependency in next-
token prediction. Recently,
Diffusion Language Models (DLMs) have emerged as a promising alternative architecture. DLMs generate output
tokens in parallel, mitigating the limitations of sequential decoding. However, the performance implications of DLMs relative to commonly deployed ARMs are not fully understood. In this work, we present a comprehensive study of the performance characteristics of ARMs and DLMs, combining theoretical analysis with empirical profiling to characterize the trade-offs between these approaches. We show that although DLMs can achieve higher arithmetic intensity than ARMs by leveraging parallelism across
token positions, they fail to scale effectively with longer contexts. We then explore block-wise decoding for DLMs, which decouples arithmetic intensity from sequence length and enables better scaling to long contexts (similar to ARMs). We also examine batched inference and find that ARMs exhibit superior throughput as they benefit more from parallelism across sequences in the batch. Finally, we highlight opportunities for accelerating DLM inference, emphasizing that reducing the number of sampling steps is key for open-source DLMs to achieve lower latency relative to ARMs.
Read the full abstract ▾
All authors Minseo Kim, Coleman Hooper, Aditya Tomar, Chenfeng Xu, Mehrdad Farajtabar, Michael W. Mahoney, Kurt Keutzer, Amir Gholami
Computation and Language · Cross-listed from cs.CV
Automating construction safety inspections using a multi-modal vision-language RAG framework
Chenxin Wang, Elyas Asadi Shamsabadi, Zhaohui Chen et al. (+3 more)
Conventional construction safety inspection methods are often inefficient as they require navigating through large volume of information. Recent advances in large
vision-
language models (LVLMs) provide opportunities to automate safety inspections through enhanced visual and linguistic understanding. However, existing applications face limitations including irrelevant or unspecific responses, restricted modal inputs and
hallucinations. Utilisation of Large
Language Models (LLMs) for this purpose is constrained by availability of training data and frequently lack real-time adaptability. This study introduces SiteShield, a multi-modal LVLM-based
Retrieval-Augmented Generation (RAG) framework for automating construction safety inspection reports by integrating visual and audio inputs. Using real-world data, SiteShield outperformed unimodal LLMs without RAG with an F1 score of 0.82, hamming loss of 0.04, precision of 0.76, and recall of 0.96. The findings indicate that SiteShield offers a novel pathway to enhance information
retrieval and efficiency in generating safety reports.
Read the full abstract ▾
All authors Chenxin Wang, Elyas Asadi Shamsabadi, Zhaohui Chen, Luming Shen, Alireza Ahmadian Fard Fini, Daniel Dias-da-Costa
Computation and Language · Cross-listed from cs.AI
Selective Expert Guidance for Effective and Diverse Exploration in Reinforcement Learning of LLMs
Zishang Jiang, Jinyi Han, Tingyun Li et al. (+7 more)
Reinforcement Learning with Verifiable Rewards (RLVR) has become a widely adopted technique for enhancing the
reasoning ability of Large
Language Models (LLMs). However, the effectiveness of RLVR strongly depends on the capability of base models. This issue arises because it requires the model to have sufficient capability to perform high-quality exploration, which involves both effectiveness and diversity. Unfortunately, existing methods address this issue by imitating expert trajectories, which improve effectiveness but neglect diversity. To address this, we argue that the expert only needs to provide guidance only at critical decision points rather than the entire
reasoning path. Based on this insight, we propose MENTOR: Mixed-policy Expert
Navigation for
Token-level Optimization of
Reasoning, a framework that provides expert guidance only at critical decision points to perform effective and diverse exploration in RLVR. Extensive experiments show that MENTOR enables models capture the essence of expert strategies rather than surface imitation, thereby performing high-quality exploration and achieving superior overall performance. Our code is available online.
Read the full abstract ▾
All authors Zishang Jiang, Jinyi Han, Tingyun Li, Xinyi Wang, Sihang Jiang, Jiaqing Liang, Zhaoqian Dai, Shuguang Ma, Fei Yu, Yanghua Xiao
Computation and Language
Fine Tuning Methods for Low-resource Languages
Tim Bakkenes, Daniel Wang and Anton Johansson
The rise of Large
Language Models has not been inclusive of all cultures. The models are mostly trained on English texts and culture which makes them underperform in other
languages and cultural contexts. By developing a generalizable method for preparing culturally relevant
datasets and post-training the Gemma 2 model, this project aimed to increase the performance of Gemma 2 for an underrepresented
language and showcase how others can do the same to unlock the power of Generative AI in their country and preserve their cultural heritage.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.AI
Internal states before wait modulate reasoning patterns
Dmitrii Troitskii, Koyena Pal, Chris Wendler et al. (+2 more)
Prior work has shown that a significant driver of performance in
reasoning models is their ability to reason and self-correct. A distinctive marker in these
reasoning traces is the
token wait, which often signals
reasoning behavior such as backtracking. Despite being such a complex behavior, little is understood of exactly why models do or do not decide to reason in this particular manner, which limits our understanding of what makes a
reasoning model so effective. In this work, we address the question whether model's latents preceding wait
tokens contain relevant information for modulating the subsequent
reasoning process. We train crosscoders at multiple layers of DeepSeek-R1-Distill-Llama-8B and its base version, and introduce a latent attribution technique in the crosscoder setting. We locate a small set of features relevant for promoting/suppressing wait
tokens' probabilities. Finally, through a targeted series of experiments analyzing max activating examples and causal interventions, we show that many of our identified features indeed are relevant for the
reasoning process and give rise to different types of
reasoning patterns such as restarting from the beginning, recalling prior knowledge, expressing uncertainty, and double-checking.
Read the full abstract ▾
All authors Dmitrii Troitskii, Koyena Pal, Chris Wendler, Callum Stuart McDougall, Neel Nanda
Computation and Language
Sri Lanka Document Datasets: A Large-Scale, Multilingual Resource for Law, News, and Policy
Nuwan I. Senaratna
We present a collection of open, machine-readable document
datasets covering parliamentary proceedings, legal judgments, government publications, news, and tourism statistics from Sri Lanka. The collection currently comprises of 299,131 documents (85.4 GB) across 26
datasets in Sinhala, Tamil, and English. The
datasets are updated daily and mirrored on GitHub and Hugging Face. These resources aim to support research in computational linguistics, legal analytics, socio-political studies, and
multilingual natural
language processing. We describe the data sources, collection pipeline, formats, and potential use cases, while discussing licensing and ethical considerations. This manuscript is at version v2026-10-12-0839.
Read the full abstract ▾
Computation and Language
Probing Semantic Alignment, Lexical Invariance, and Syntactic Influence in LLM Metaphor Processing
Fengying Ye, Shanshan Wang, Lidia S. Chao et al. (+1 more)
Large
language models (LLMs) achieve strong performance on metaphor detection and interpretation tasks, yet it remains unclear what such behavioral success reveals about metaphor processing. We present a diagnostic analysis that examines the limits of behavioral evidence by probing three complementary dimensions: semantic attribute
alignment, lexical invariance, and syntactic sensitivity. Using
geometric probing, we assess whether model-generated interpretations align with reference semantic attributes; through context-varying substitution, we analyze the stability of lexical associations between metaphorical and literal expressions; and via controlled syntactic perturbations, we examine sensitivity in metaphor detection. Our analysis reveals that LLM-generated interpretations can exhibit semantic drift relative to reference attributes; stable lexical anchors persist across contextual conditions, potentially supporting conventional metaphors while biasing novel metaphors requiring contextual integration; and detection performance is sensitive to syntactic irregularities. These findings suggest that strong behavioral performance may reflect heterogeneous underlying signals, highlighting the need for caution when interpreting metaphor
benchmarks as evidence of robust, integrated semantic understanding.
Read the full abstract ▾
All authors Fengying Ye, Shanshan Wang, Lidia S. Chao, Derek F. Wong
Computation and Language
Scaling Code-Assisted Chain-of-Thoughts and Instructions for Model Reasoning
Honglin Lin, Qizhi Pei, Xin Gao et al. (+5 more)
Reasoning capability is pivotal for Large
Language Models (LLMs) to solve complex tasks, yet achieving reliable and scalable
reasoning remains challenging. While Chain-of-Thought (CoT) prompting has become a mainstream approach, existing methods often suffer from uncontrolled generation, insufficient quality, and limited diversity in
reasoning paths. Recent efforts leverage code to enhance CoT by grounding
reasoning in executable steps, but such methods are typically constrained to predefined mathematical problems, hindering scalability and generalizability. In this work, we propose Caco (Code-Assisted Chain-of-ThOught), a novel framework that automates the synthesis of high-quality, verifiable, and diverse instruction-CoT
reasoning data through code-driven augmentation. Unlike prior work, Caco first fine-tunes a code-based CoT generator on existing math and programming solutions in a unified code format, then scales the data generation to a large amount of diverse
reasoning traces. Crucially, we introduce automated validation via code execution and rule-based filtering to ensure logical correctness and structural diversity, followed by reverse-engineering filtered outputs into natural
language instructions and
language CoTs to enrich task adaptability. This closed-loop process enables fully automated, scalable synthesis of
reasoning data with guaranteed executability. Experiments on our created Caco-1.3M
dataset demonstrate that Caco-trained models achieve strong competitive performance on mathematical
reasoning benchmarks, outperforming existing strong baselines. Further analysis reveals that Caco's code-anchored verification and instruction diversity contribute to superior generalization across unseen tasks. Our work establishes a paradigm for building self-sustaining, trustworthy
reasoning systems without human intervention.
Read the full abstract ▾
All authors Honglin Lin, Qizhi Pei, Xin Gao, Zhuoshi Pan, Yu Li, Juntao Li, Conghui He, Lijun Wu
Computation and Language
PoLi-RL: A Point-to-List Reinforcement Learning Framework for Conditional Semantic Textual Similarity
Zixin Song, Bowen Zhang, Qian-Wen Zhang et al. (+3 more)
Conditional Semantic Textual Similarity (C-STS) measures the semantic proximity between text segments under a specific condition, thereby overcoming the ambiguity inherent in traditional STS. However, existing methods are largely confined to discriminative models, failing to fully leverage recent breakthroughs in the NLP community involving Large
Language Models (LLMs) and Reinforcement Learning (RL). RL is a particularly well-suited paradigm for this task, as it can directly optimize the non-differentiable Spearman ranking metric and guide the
reasoning process required by C-STS. Nevertheless, we find that naively applying listwise RL fails to produce meaningful improvements, as the model struggles with complex, coarse-grained reward signals, leading to optimization difficulties. To address this challenge, we introduce PoLi-RL, a novel Point-to-List Reinforcement Learning framework. PoLi-RL employs a two-stage curriculum: it first trains the model with a simple pointwise reward to establish fundamental scoring capabilities, then transitions to a hybrid reward that combines pointwise, pairwise, and listwise objectives to refine the model's ability to discern subtle semantic distinctions. Crucially, we propose an innovative Parallel Slice Ranking Reward (PSRR) mechanism that computes ranking rewards in parallel slices, where each slice consists of completions with the same index from different samples. This provides a precise, differentiated learning signal for each individual completion, enabling granular credit assignment and effective optimization. On the official C-STS
benchmark, PoLi-RL achieves a Spearman correlation coefficient of 48.18, establishing a new SOTA for the cross-encoder architecture. As the first work to successfully apply RL to C-STS, our study introduces a powerful paradigm for aligning LLMs for complex, ranking-based conditional judgment tasks.
Read the full abstract ▾
All authors Zixin Song, Bowen Zhang, Qian-Wen Zhang, Di Yin, Xing Sun, Chunping Li
Computation and Language · Cross-listed from cs.LG
Slow-Fast Policy Optimization: Reposition-Before-Update for LLM Reasoning
Ziyan Wang, Zheng Wang, Xingwei Qu et al. (+5 more)
Reinforcement learning (RL) has become central to enhancing
reasoning in large
language models (LLMs). Yet on-policy
algorithms such as Group Relative Policy Optimization (GRPO) often suffer in early training: noisy gradients from low-quality rollouts lead to unstable updates and inefficient exploration. We introduce Slow-Fast Policy Optimization (SFPO), a simple yet efficient framework to address the above limitations via decomposing each step into three stages: a short fast trajectory of inner steps on the same batch, a reposition mechanism to control off-policy drift, and a final slow correction. This reposition-before-update design preserves the objective and rollout process unchanged, making SFPO plug-compatible with existing policy-gradient pipelines. Extensive experiments demonstrate that SFPO consistently improves stability, reduces number of rollouts, and accelerates convergence of
reasoning RL training. Specifically, it outperforms GRPO by up to 2.80 points in average on math
reasoning benchmarks. It also achieves up to 4.93\texttimes{} fewer rollouts and an up to 4.19\texttimes{} reduction in wall-clock time to match GRPO's best accuracy. Project website is available at https://slow-fast-po.github.io/.
Read the full abstract ▾
All authors Ziyan Wang, Zheng Wang, Xingwei Qu, Qi Cheng, Jie Fu, Shengpu Tang, Minjia Zhang, Xiaoming Huo
Computation and Language
What Makes Diffusion Language Models Super Data Learners?
Zitian Gao, Haoming Luo, Lynx Chen et al. (+4 more)
Recent studies have shown that
diffusion language models achieve remarkable data efficiency under limited-data constraints, yet the underlying mechanisms remain unclear. In this work, we perform extensive ablation experiments to disentangle the sources of this efficiency. Our results show that random masking of input
tokens plays the dominant role. We further show that similar gains can be obtained through in MLP dropout and weight decay, indicating that
stochastic regularization broadly enhances data efficiency in multi-epoch training. Our code is available at https://github.com/zitian-gao/data-efficiency.
Read the full abstract ▾
All authors Zitian Gao, Haoming Luo, Lynx Chen, Jason Klein Liu, Ran Tao, Joey Zhou, Bryan Dai
Computation and Language · Cross-listed from cs.LG
What Scales in Cross-Entropy Scaling Law?
Junxi Yan, Zixi Wei, Qingyao Ai et al. (+2 more)
The cross-
entropy scaling law has long served as a key tool for guiding the development of large
language models. It shows that cross-
entropy loss decreases in a predictable power-law rate as the model size increases. However, recent evidence indicates that this law breaks down at very large scales: the loss decreases more slowly than expected, which causes significant trouble for developing large
language models. In this paper, we hypothesize that the root cause lies in the fact that cross-
entropy itself does not truly scale; instead, only one of its hidden components does. To investigate this, we introduce a novel decomposition of cross-
entropy into three parts: Error-
Entropy, Self-
Alignment, and Confidence. We show both theoretically and empirically that this decomposition precisely captures the training dynamics and optimization objectives. Through extensive experiments on multiple
datasets and 32 models spanning five orders of magnitude in size, we find that only error-
entropy follows a robust power-law scaling, while the other two terms remain largely invariant. Moreover, error-
entropy constitutes the dominant share of cross-
entropy in small models but diminishes in proportion as models grow larger. This explains why the cross-
entropy scaling law appears accurate at small scales but fails at very large ones. Our findings establish the error-
entropy scaling law as a more accurate description of model behavior. We believe it will have wide applications in the training, understanding, and future development of large
language models.
Read the full abstract ▾
All authors Junxi Yan, Zixi Wei, Qingyao Ai, Yiqun Liu, Jingtao Zhan
Computation and Language
Exploring Chain-of-Thought Reasoning for Steerable Pluralistic Alignment
Yunfan Zhang, Kathleen McKeown and Smaranda Muresan
Large
Language Models (LLMs) are typically trained to reflect a relatively uniform set of values, which limits their applicability to tasks that require understanding of nuanced human perspectives. Recent research has underscored the importance of enabling LLMs to support steerable pluralism – the capacity to adopt a specific perspective and align generated outputs with it. In this work, we investigate whether Chain-of-Thought (CoT)
reasoning techniques can be applied to building steerable pluralistic models. We explore several methods, including CoT prompting, fine-tuning on human-authored CoT, fine-tuning on synthetic explanations, and Reinforcement Learning with Verifiable Rewards (RLVR). We evaluate these approaches using the Value Kaleidoscope and OpinionQA
datasets. Among the methods studied, RLVR consistently outperforms others and demonstrates strong training sample efficiency. We further analyze the generated CoT traces with respect to faithfulness and safety.
Read the full abstract ▾
Computation and Language
Small Language Models for Emergency Departments Decision Support: A Benchmark Study
Zirui Wang, Jiajun Wu, Braden Teitge et al. (+2 more)
Large
language models (LLMs) have become increasingly popular in medical domains to assist physicians with a variety of clinical and operational tasks. Given the fast-paced and high-stakes environment of emergency departments (EDs), small
language models (SLMs), characterized by a reduction in parameter count compared to LLMs, offer significant potential due to their inherent
reasoning capability and efficient performance. This enables SLMs to support physicians by providing timely and accurate information synthesis, thereby improving clinical decision-making and workflow efficiency. In this paper, we present a comprehensive
benchmark designed to identify SLMs suited for ED decision support, taking into account both specialized medical expertise and broad general problem-solving capabilities. In our evaluations, we focus on SLMs that have been trained on a mixture of general-domain and medical corpora. A key motivation for emphasizing SLMs is the practical hardware limitations, operational cost constraints, and
privacy concerns in the typical real-world deployments. Our
benchmark datasets include MedMCQA, MedQA-4Options, and PubMedQA, with the medical abstracts
dataset emulating tasks aligned with real ED physicians' daily tasks. Experimental results reveal that general-domain SLMs surprisingly outperform their medically fine-tuned counterparts across these diverse
benchmarks for ED. This indicates that for ED, specialized medical fine-tuning of the model may not be required.
Read the full abstract ▾
All authors Zirui Wang, Jiajun Wu, Braden Teitge, Jessalyn Holodinsky, Steve Drew
Computation and Language
Does Using Counterfactual Help LLMs Explain Textual Importance in Classification?
Nelvin Tan, James Asikin Cheung, Yu-Ching Shih et al. (+2 more)
Large
language models (LLMs) are becoming useful in many domains due to their impressive abilities that arise from large training
datasets and large model sizes. More recently, they have been shown to be very effective in textual classification tasks, motivating the need to explain the LLMs' decisions. Motivated by practical constrains where LLMs are black-boxed and LLM calls are expensive, we study how incorporating counterfactuals into LLM
reasoning can affect the LLM's ability to identify the top words that have contributed to its classification decision. To this end, we introduce a framework called the decision changing rate that helps us quantify the importance of the top words in classification. Our experimental results show that using counterfactuals can be helpful.
Read the full abstract ▾
All authors Nelvin Tan, James Asikin Cheung, Yu-Ching Shih, Dong Yang, Amol Salunkhe
Computation and Language · Cross-listed from cs.AI
LLM-Based Data Science Agents: A Survey of Capabilities, Challenges, and Future Directions
Mizanur Rahman, Amran Bhuiyan, Mohammed Saidul Islam et al. (+5 more)
Recent advances in large
language models (LLMs) have enabled a new class of AI
agents that automate multiple stages of the data science workflow by integrating planning, tool use, and
multimodal reasoning across text, code, tables, and visuals. This survey presents the first comprehensive, lifecycle-aligned taxonomy of data science
agents, systematically analyzing and mapping forty-five systems onto the six stages of the end-to-end data science process: business understanding and data acquisition, exploratory analysis and visualization, feature engineering, model building and selection, interpretation and explanation, and deployment and monitoring. In addition to lifecycle coverage, we annotate each
agent along five cross-cutting design dimensions:
reasoning and planning style, modality integration, tool orchestration depth, learning and
alignment methods, and trust, safety, and governance mechanisms. Beyond classification, we provide a critical synthesis of
agent capabilities, highlight strengths and limitations at each stage, and review emerging
benchmarks and evaluation practices. Our analysis identifies three key trends: most systems emphasize exploratory analysis, visualization, and modeling while neglecting business understanding, deployment, and monitoring;
multimodal reasoning and tool orchestration remain unresolved challenges; and over 90% lack explicit trust and safety mechanisms. We conclude by outlining open challenges in
alignment stability, explainability, governance, and robust evaluation frameworks, and propose future research directions to guide the development of robust, trustworthy, low-latency, transparent, and broadly accessible data science
agents.
Read the full abstract ▾
All authors Mizanur Rahman, Amran Bhuiyan, Mohammed Saidul Islam, Md Tahmid Rahman Laskar, Ridwan Mahbub, Ahmed Masry, Shafiq Joty, Enamul Hoque
Computation and Language · Cross-listed from cs.LG
Anthony Zhan
Diffusion large
language models (dLLMs) represent a promising alternative to autoregressive LLMs; however, the lack of effective post-training techniques, including reinforcement learning (RL), remains a key challenge for dLLMs, especially for downstream applications. Existing approaches often rely on a sequence-level view that requires biased likelihood approximations. In this work, we propose Amortized Group Relative Policy Optimization (AGRPO), a policy gradient
algorithm that leverages the Markovian nature of dLLMs, optimizing individual denoising steps rather than full sequences. Our approach improves
alignment between the trained policy and the inference process and also admits efficient, unbiased gradient updates via a novel timestep estimation scheme. We demonstrate AGRPO's effectiveness on different math and
reasoning tasks, achieving absolute accuracy gains of +59.4% and +69.7% on Countdown and Sudoku over the base LLaDA model, exceeding comparable methods such as diffu-GRPO. Code is available at https://github.com/probablyabot/agrpo.
Read the full abstract ▾
Computation and Language
Thai Semantic End-of-Turn Detection for Real-Time Voice Agents
Thanapol Popit, Natthapath Rungseesiripak, Monthol Charattrakool et al. (+1 more)
Fluid voice-to-voice interaction requires reliable and low-latency detection of when a user has finished speaking. Traditional audio-silence end-pointers add hundreds of milliseconds of delay and fail under hesitations or
language-specific phenomena. We present, to our knowledge, the first systematic study of Thai text-only end-of-turn (EOT) detection for real-time
agents. We compare zero-shot and few-shot prompting of compact LLMs to supervised fine-tuning of lightweight
transformers. Using transcribed subtitles from the YODAS corpus and Thai-specific linguistic cues (e.g., sentence-final particles), we formulate EOT as a binary decision over
token boundaries. We report a clear accuracy-latency tradeoff and provide a public-ready implementation plan. This work establishes a Thai baseline and demonstrates that small, fine-tuned models can deliver near-instant EOT decisions suitable for on-device
agents.
Read the full abstract ▾
All authors Thanapol Popit, Natthapath Rungseesiripak, Monthol Charattrakool, Saksorn Ruangtanusak
Computation and Language
LLM Microscope: What Model Internals Reveal About Answer Correctness and Context Utilization
Jiarui Liu, Jivitesh Jain, Mona Diab et al. (+1 more)
Although large
language models (LLMs) have tremendous utility, trustworthiness is still a chief concern: models often generate incorrect information with high confidence. While contextual information can help guide generation, identifying when a query would benefit from retrieved context and assessing the effectiveness of that context remains challenging. In this work, we operationalize interpretability methods to ascertain whether we can predict the correctness of model outputs from the model's activations alone. We also explore whether model internals contain signals about the efficacy of external context. We consider correct, incorrect, and irrelevant context and introduce metrics to distinguish amongst them. Experiments on six different models reveal that a simple classifier trained on intermediate layer activations of the first output
token can predict output correctness with about 75% accuracy, enabling early
auditing. Our model-internals-based metric significantly outperforms prompting baselines at distinguishing between correct and incorrect context, guarding against inaccuracies introduced by polluted context. These findings offer a lens to better understand the underlying decision-making processes of LLMs. Our code is publicly available at https://github.com/jiarui-liu/LLM-Microscope
Read the full abstract ▾
All authors Jiarui Liu, Jivitesh Jain, Mona Diab, Nishant Subramani
Computation and Language · Cross-listed from cs.IR
Visual Lifelog Retrieval through Captioning-Enhanced Interpretation
Yu-Fei Shih, An-Zi Yen, Hen-Hsen Huang et al. (+1 more)
People often struggle to remember specific details of past experiences, which can lead to the need to revisit these
memories. Consequently, lifelog
retrieval has emerged as a crucial application. Various studies have explored methods to facilitate rapid access to personal lifelogs for
memory recall assistance. In this paper, we propose a Captioning-Integrated Visual Lifelog (CIVIL)
Retrieval System for extracting specific images from a user's visual lifelog based on textual queries. Unlike traditional
embedding-based methods, our system first generates captions for visual lifelogs and then utilizes a text
embedding model to project both the captions and user queries into a shared vector space. Visual lifelogs, captured through wearable cameras, provide a first-person viewpoint, necessitating the interpretation of the activities of the individual behind the camera rather than merely describing the scene. To address this, we introduce three distinct approaches: the single caption method, the collective caption method, and the merged caption method, each designed to interpret the life experiences of lifeloggers. Experimental results show that our method effectively describes first-person visual images, enhancing the outcomes of lifelog
retrieval. Furthermore, we construct a textual
dataset that converts visual lifelogs into captions, thereby reconstructing personal life experiences.
Read the full abstract ▾
All authors Yu-Fei Shih, An-Zi Yen, Hen-Hsen Huang, Hsin-Hsi Chen
Computation and Language · Cross-listed from cs.AI
What Shapes a Creative Machine Mind? Comprehensively Benchmarking Creativity in Foundation Models
Zicong He, Boxuan Zhang, Weihao Liu et al. (+2 more)
The meteoric rise of foundation models (FMs) has expanded their capabilities far beyond conventional tasks. Creativity, long regarded as a hallmark of human intelligence and a driver of innovation, is now increasingly recognized as a critical dimension of machine intelligence in the era of generative FMs, complementing traditional measures of accuracy. However, existing evaluation frameworks for creativity remain fragmented, relying on ad hoc metrics not firmly grounded in established theories. To address this gap, we introduce C^2-Eval, a holistic
benchmark for unified assessment of creativity in FMs. C^2-Eval distinguishes between two complementary forms of creativity: convergent creativity, where tasks admit constrained solutions (e.g., code generation), and divergent creativity, where tasks are open-ended (e.g., storytelling). It evaluates both dimensions using fine-grained criteria derived from social-science theory, focusing on Usefulness, Originality, and Surprise (U-O-S). Through extensive experiments on leading proprietary and open-source models, we analyze trade-offs in their creative capabilities. Our results highlight both the strengths and challenges of current FMs in pursuing a creative machine mind, showing that C^2-Eval is an effective lens for examining the evolving landscape of creative AI.
Read the full abstract ▾
All authors Zicong He, Boxuan Zhang, Weihao Liu, Ruixiang Tang, Lu Cheng
Computation and Language · Cross-listed from cs.CV
Enhancing OCR for Sino-Vietnamese Language Processing via Fine-tuned PaddleOCRv5
Minh Hoang Nguyen and Su Nguyen Thiet
Recognizing and processing Classical Chinese (Han-Nom) texts play a vital role in digitizing Vietnamese historical documents and enabling cross-lingual semantic research. However, existing OCR systems struggle with degraded scans, non-standard glyphs, and handwriting variations common in ancient sources. In this work, we propose a fine-tuning approach for PaddleOCRv5 to improve character recognition on Han-Nom texts. We retrain the text recognition module using a curated subset of ancient Vietnamese Chinese manuscripts, supported by a full training pipeline covering preprocessing, LMDB conversion, evaluation, and visualization. Experimental results show a significant improvement over the base model, with exact accuracy increasing from 37.5 percent to 50.0 percent, particularly under noisy image conditions. Furthermore, we develop an interactive demo that visually compares pre- and post-fine-tuning recognition results, facilitating downstream applications such as Han-Vietnamese semantic
alignment, machine translation, and historical linguistics research. The demo is available at https://huggingface.co/spaces/MinhDS/Fine-tuned-PaddleOCRv5
Read the full abstract ▾
Computation and Language
AgriGPT-VL: Agricultural Vision-Language Understanding Suite
Bo Yang, Yunkui Chen, Lanfei Feng et al. (+8 more)
Despite rapid advances in
multimodal large
language models, agricultural applications remain constrained by the scarcity of domain-tailored models, curated
vision-
language corpora, and rigorous evaluation. To address these challenges, we present the AgriGPT-VL Suite, a unified
multimodal framework for agriculture. Our contributions are threefold. First, we introduce Agri-3M-VL, the largest
vision-
language corpus for agriculture to our knowledge, curated by a scalable multi-
agent data generator; it comprises 1M image-caption pairs, 2M image-grounded VQA pairs, 50K expert-level VQA instances, and 15K GRPO reinforcement learning samples. Second, we develop AgriGPT-VL, an agriculture-specialized
vision-
language model trained via a progressive curriculum of textual grounding,
multimodal shallow/deep
alignment, and GRPO refinement. This method achieves strong
multimodal reasoning while preserving text-only capability. Third, we establish AgriBench-VL-4K, a compact yet challenging evaluation suite with open-ended and image-grounded questions, paired with multi-metric evaluation and an LLM-as-a-judge framework. Experiments show that AgriGPT-VL outperforms leading general-purpose VLMs on AgriBench-VL-4K, achieving higher pairwise win rates in the LLM-as-a-judge evaluation. Meanwhile, it remains competitive on the text-only AgriBench-13K with no noticeable degradation of
language ability. Ablation studies further confirm consistent gains from our
alignment and GRPO refinement stages. We will open source all of the resources to support reproducible research and deployment in low-resource agricultural settings.
Read the full abstract ▾
All authors Bo Yang, Yunkui Chen, Lanfei Feng, Yu Zhang, Xiao Xu, Jianyu Zhang, Nueraili Aierken, Runhe Huang, Hongjian Lin, Yibin Ying, Shijian Li