Computation and Language
ACCORD: Action-Conditioned Contextual Grounding for Language Agents
Lai Jiang, Cheng Qian, Zhenhailong Wang et al. (+3 more)
User instructions are often underspecified because humans rely on implicit assumptions about the surrounding environment. For large
language model (LLM)
agents operating in information-rich digital and physical environments, these assumptions cannot be inferred from the instruction alone; they must be recovered from the current state of tools, data, interfaces, and observations. Effective execution therefore requires
agents to identify missing context, ground it in observed evidence, and carry it forward into subsequent actions. We show that current
agents often fail to do so. They act from assumed rather than observed specifics, overlook information they could have gathered, and fail to incorporate evidence that has already been returned. Building on this insight, we propose ACCORD (Action-Conditioned Contextual Grounding), a simple and effective
agent framework for adaptive grounding. Before each action, ACCORD actively probes the environment for missing information and integrates relevant context from the
agent's trajectory that would otherwise be overlooked. Requiring no additional training or task-success signals, ACCORD improves task-goal completion on AppWorld by up to +20.6 points with GPT-5-mini, from 42.0% to 62.6%, compared to strong baselines. These gains persist with a substantially stronger base model (+10.8 with Claude-4.5-sonnet), an open-weight model (+10.1 with Qwen3.5-27B-FP8), and on the
embodied AlfWorld
benchmark (+7.4 success rate with GPT-5-mini).
Read the full abstract ▾
All authors Lai Jiang, Cheng Qian, Zhenhailong Wang, Pan Lu, Heng Ji, Hao Peng
Computation and Language · Cross-listed from cs.LG
Taylor-Calibrate: Principled Initialization for Hybrid Linear Attention Distillation
Zhongzhu Zhou, Qingyang Wu, Junxiong Wang et al. (+4 more)
Hybrid linear
attention models offer an appealing path to faster long-context inference: they reduce the quadratic cost and KV-cache burden of full softmax
attention while retaining much of the quality of
Transformer models. A practical way to obtain such models is to convert a pretrained
Transformer instead of pretraining a new architecture from scratch, but this conversion is still brittle. Simply copying the teacher
attention projections into a Gated DeltaNet (GDN) student does not specify the new recurrent decay, write, and output-gating dynamics. As a result, the converted model often starts in a poor dynamical regime and must spend many
distillation tokens repairing initialization rather than learning the remaining teacher behavior. We propose Taylor-Calibrate, a lightweight initialization method for hybrid GDN students. The method uses Taylor-guided teacher
attention statistics to set the value projection,
memory timescale, write gates, and output gate, then applies a short per-layer
alignment step to match each converted layer to the teacher output. Across four teacher settings and three retained-layer policies, Taylor-Calibrate gives substantially stronger zero-shot students, with up to an 88x improvement in a representative ablation, and reaches matched recovery targets with 4.9x–9.2x fewer training
tokens than naive conversion.
Read the full abstract ▾
All authors Zhongzhu Zhou, Qingyang Wu, Junxiong Wang, Mayank Mishra, Shuaiwen Leon Song, Ben Athiwaratkun, Chenfeng Xu
Computation and Language
LectūraAgents: A Multi-Agent Framework for Adaptive Personalized AI-Assisted Learning and Embodied Teaching
Jaward Sesay, Yue Yu, Siwei Dong et al. (+1 more)
Effective personalized AI-assisted learning demands systems that can not only generate accurate learner-specific educational materials, but also dynamically adapt their instruction to diverse learners. However, existing educational
agents have primarily focused on lecture content automation and simulations, which often fall short of modelling
multimodal and
embodied instructional methods tailored for the individual learner. To this end, we propose LectūraAgents - a multi-
agent framework that enables personalized learning through end-to-end adaptive
embodied teaching. At its core, LectūraAgents mirrors a professor-student relationship, in which a ProfessorAgent leads a collaborative team of specialized subordinate
agents through research, planning, review, and
embodied delivery of lecture contents that adapt to a learner's needs. The framework offers three main contributions: (1) a hierarchical multi-
agent architecture for end-to-end personalized learning; (2) an adaptive
embodied teaching mechanism, wherein the ProfessorAgent executes visible and pedagogically motivated teaching actions (e.g., handwrite, highlight, underline, etc.) over contents in a teaching environment; and (3) a Teaching Action-
Speech Alignment (TASA)
algorithm that employs salience-based heuristics and temporal semantic
segmentation to generate coherent teaching action sequences aligned with learner profiles. We evaluate LectūraAgents on diverse courses at high school, undergraduate, and graduate levels using sample-specific rubric-based analysis; with generated lecture materials and teaching actions assessed and validated by expert educators. Experimental results show consistent gains in lecture content quality,
embodied teaching quality, assessment, and personalization over existing approaches, positioning LectūraAgents as a pedagogically well-grounded framework for personalized learning at scale.
Read the full abstract ▾
All authors Jaward Sesay, Yue Yu, Siwei Dong, Börje F. Karlsson
Computation and Language
PathRouter: Aligning Rewards with Retrieval Quality in Agentic Graph Retrieval-Augmented Generation
Bo Wang, Heyan Huang, Yaolin Li et al. (+6 more)
Agentic GraphRAG trains
language-model
agents to iteratively retrieve and reason over graph-structured evidence, enabling more accurate and context-aware decision-making by efficiently navigating complex information networks. However, outcome-only reinforcement learning suffers from \textit{answer-path reward aliasing}, where correct answers may come from shortcuts rather than useful evidence paths. It also exhibits \textit{search-update ambiguity}, as scalar trajectory-level feedback does not indicate which
retrieval actions to adjust. To mitigate these shortcomings, we present PathRouter, a path-aware training framework for agentic GraphRAG. PathRouter jointly evaluates each trajectory along answer correctness and evidence-path overlap, yielding four trajectory categories with differentiated GRPO advantage scaling that suppresses shortcut reinforcement while preserving evidence-seeking behavior. For evidence-poor trajectories, a frozen gold-evidence teacher provides
token-level KL guidance on
reasoning and search-query
tokens, excluding answer
tokens to avoid direct response imitation. Experiments on six QA
benchmarks across three model sizes show that PathRouter consistently improves answer F1 and evidence-path overlap, achieving average F1 gains of 3.1 on 3B and 4.9 on 7B models compared to a strong baseline.
Read the full abstract ▾
All authors Bo Wang, Heyan Huang, Yaolin Li, Wei Tang, Yuan Zhang, Wenbo Li, Mingze Gao, Ge Shi, Chong Feng
Computation and Language
A Mechanistic Understanding of Pronoun Fidelity in LLMs
Katharina Trinley, Jesujoba O. Alabi, Dietrich Klakow et al. (+1 more)
Faithful and robust pronoun use is important for fair and coherent generations, yet large
language models largely fail when multiple referents use different pronouns. To study the interplay of
reasoning, repetition, and bias in this task, prior work relies exclusively on behavioural approaches, which may not reflect a model's internal workings. Therefore, we provide a mechanistic, model-internal perspective on pronoun fidelity, testing whether three mechanisms – group entity binding (G), recency bias (R), and stereotypical bias (S) – are causally implemented across several SOTA
language models. Using Boundless Distributed
Alignment Search, we find all three coexist as causal subspaces distributed across network depth. No single mechanism fully explains model behaviour, but a combination of the three consistently accounts for 91-99.5%. An
attention head analysis further reveals two competing copying routes; group binding and stereotype share a localized concept-level route that retrieves a bound occupation-pronoun unit, while recency uses a distributed
token-level route that repeats surface forms. In sum, pronoun fidelity arises from competition between simultaneously active causal subspaces.
Read the full abstract ▾
All authors Katharina Trinley, Jesujoba O. Alabi, Dietrich Klakow, Vagrant Gautam
Computation and Language
Surpassing Scale by Efficiency: A Compact 135M Parameter Foundational LLM Natively Adapted for the Bangla Language
Rabindra Nath Nandi
While the NLP landscape is dominated by multi-billion parameter architectures, their deployment in low-resource, non-Latin scripts remains computationally prohibitive for edge configurations, mobile systems, and decentralized local hardware. This paper presents bangla-smollm-135m, a highly compact 135-million parameter decoder-only foundational model engineered explicitly for high-efficiency
language modeling in the Bangla script. By leveraging a deterministic intersect-and-append
token merging strategy between TituLLMs and SmolLM2-135M, the model overcomes subword script fragmentation without destabilizing early pretrained parameter states. In zero-shot multi-task
benchmark evaluations (PIQA_bn, OpenBookQA_bn, CommonsenseQA_bn, and Bangla_MMLU), bangla-smollm-135m matches or outperforms models twice its size (Gemma-3-270m) and achieves parity with models in the 1B parameter tier. The model is available at rnnandi/bangla-smollm-135m
Read the full abstract ▾
Computation and Language
Evaluating LLM Personalization via Semantic Constraint Verification
Xuran Li, Guanqin Zhang, Imran Razzak et al. (+4 more)
Current evaluation paradigms for Large
Language Model (LLM) personalization rely heavily on brittle surface-matching metrics or computationally expensive LLM-as-a-judge protocols, both of which lack interpretability. To address these limitations, we introduce Natural
Language Inference Constraint Verification (NLICV), a scalable, semantically invariant framework that maps sentence meanings to truth-condition sets to verify personalization constraints via a Natural
Language Inference (NLI) model. Moving beyond binary scoring, NLICV categorizes LLM behaviors into four distinct modes: personalization, generalization, sycophancy, and failure. Extensive experiments demonstrate that NLICV aligns closely with human annotations while drastically reducing the latency and
token costs associated with LLM judges (up to 2100 inference speedup). Finally, through an ablation-based procedure, NLICV pinpoints the exact sentences driving the constraint verification, yielding faithful, understandable evidence for its evaluations.
Read the full abstract ▾
All authors Xuran Li, Guanqin Zhang, Imran Razzak, Hakim Hacid, Eleanna Kafeza, Hao Xue, Flora D. Salim
Computation and Language
Tyler: Typed Latent Reasoning for Language Models – When to Think, What to Compute, and How Much to Allocate
Hanyu Lin, Min Cai, Jiawei Wen et al. (+1 more)
Chain-of-thought (CoT) prompting improves
reasoning in large
language models (LLMs) by externalizing intermediate computation as discrete text
tokens, but this textual interface also introduces redundancy and inference overhead. Latent
reasoning offers a promising alternative by carrying part of the computation in continuous representations. However, existing methods typically predefine when latent computation is invoked and how it is allocated during decoding, leaving a key problem unresolved: when to invoke latent computation, what type of computation to perform, and how much budget to allocate. We propose Typed Latent
Reasoning (Tyler), a typed and budget-aware framework for latent
reasoning during autoregressive decoding. Tyler learns a policy that, at each decoding step, chooses between emitting a text
token and switching to a latent computation module specialized for a particular
reasoning function. Once invoked, an operator maps the current
reasoning state into latent
tokens that support global planning, local state updates, or reusable procedural abstraction. Across extensive experiments on three backbone LLMs, Tyler improves accuracy by up to 14.49 points over CoT and by up to 4.30 points over the strongest competing baseline. It further generalizes across diverse
reasoning domains and achieves the best final-stage performance with the lowest forgetting.
Read the full abstract ▾
All authors Hanyu Lin, Min Cai, Jiawei Wen, Haodi Zhang
Computation and Language
TMASC: Transmasculine Attitude and Speech Corpus
Sidney Wong
We introduce the Transmasculine Attitudes and
Speech Corpus (TMASC), a
multimodal corpus of 196 transmasculine individuals, including questionnaire responses and 66 audio recordings. The questionnaire includes items exploring the vocal health of transmasculine individuals. The audio recordings include cough and throat-clearing samples, a reading passage, and additional session-specific questions. This paper outlines the development of this corpus and the data collection procedures. To illustrate the utility of this corpus, we present three case studies demonstrating how this crowd-sourced
multimodal corpus can be used to support transmasculine individuals. These include the integration of perceptual and acoustic data, the identification of group-level characteristics, and the calibration of acoustic measurements.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.AI
Whose hotel does the AI recommend? An algorithm audit of reputation signals in LLM-assisted hotel selection
Mirza Samad Ahmed Baig, Syeda Anshrah Gillani and Asher Ali
Travelers increasingly ask large
language model (LLM) assistants which hotel to book, making these systems gatekeepers of property visibility – yet what moves their recommendations is undocumented. We conduct a pre-specified
algorithm audit using a randomized choice-based conjoint: across personas,
prompt templates, and twelve open-weight and proprietary models, assistants choose among five hotels whose guest rating, review volume and recency, management response, chain affiliation, price, eco-certification, and list position are independently randomized. We estimate the average marginal component effect of each signal on the probability of recommendation. Guest rating and price dominate (a top rating raises selection by 31.6 percentage points; a high price lowers it by 30.0), reproducing human valence-and-price primacy but over-weighting eco-certification and ignoring management response. List position – a content-free artifact – shifts recommendations causally, worth about \$12 per night. Stated reasons track revealed weights imperfectly. The findings ground generative engine optimization and the accountability of AI infomediaries in causal evidence.
Read the full abstract ▾
Computation and Language
PaperJury: Due-Process Review for Bounded LaTeX Revision
Yiran Wang, Ruixuan An, Biao Wu et al. (+1 more)
Pre-submission hardening of human-authored LaTeX computer science papers differs from drafting assistance because it requires
adversarial whole-paper review, explicit no-fix outcomes, and bounded artifact-safe revision. Existing writing assistants, critique generators, and judge-centered loops lack durable issue identity across rounds, deterministic routing from critique to adjudication, and manuscript control that can reject invalid concerns or defer author-dependent ones. We present PaperJury, a closed-loop review-verdict-revise-verify system built on a deterministic-versus-semantic split: deterministic orchestration manages decomposition, a frozen claim spine, a durable ledger, routing, stopping, and exact-once patch application, while semantic
agents are limited to bounded review, judgment, and repair. PaperJury combines bounded holistic review, contestability-based routing, a due-process trial, and risk-proportional guard chains for anchor-bounded edits, yielding terminal outcomes of invalid-drop, valid-fixable, and author-required. In a two-arm expert-review evaluation on held-out
Vision, natural
language processing, and machine learning papers against four baselines, we assess issue quality, verdict and routing quality, edit safety, convergence behavior, and cost, supporting the thesis that load-bearing safety and completion logic should reside in deterministic orchestration rather than model discretion. PaperJury is available at https://github.com/u7079256/paperjury.
Read the full abstract ▾
All authors Yiran Wang, Ruixuan An, Biao Wu, Wenhao Wang
Computation and Language · Cross-listed from cs.LG
QK-Normed MLA: QK normalization without full key caching
Yizhou Han, Yao Zhao, Jun Zhou et al. (+2 more)
Query-key (QK) normalization stabilizes
attention by controlling the scale of queries and keys before the dot product, but is not immediately compatible with Multi-head Latent
Attention (MLA). MLA achieves efficient decoding by caching low-dimensional latent states instead of full keys, whereas post-projection QK RMSNorm appears to require the fully projected key for every cached
token. We show this apparent incompatibility is an implementation artifact, not an architectural constraint. RMSNorm decomposes into a static affine weight and a dynamic scalar RMS statistic. The static key-side weight can be absorbed into the MLA query-side projection; the dynamic key statistic reduces to one inverse-RMS scalar per
token and KV group. The resulting formulation is exactly equivalent to explicit post-projection QK RMSNorm in exact arithmetic and preserves MLA's latent decode path. In our 400M runs trained for up to 100B
tokens, QK-Normed MLA achieves lower training loss and better downstream accuracy than QK clipping, while H800 decode
benchmarks show less than 2% latency overhead up to 256k context. These results make QK normalization a practical stabilization option for MLA models without requiring full-key caching.
Read the full abstract ▾
All authors Yizhou Han, Yao Zhao, Jun Zhou, Longfei Li, Ruoyu Sun
Computation and Language · Cross-listed from cs.AI
State-Grounded Multi-Agent Synthetic Data Generation for Tool-Augmented LLMs
Rahul Khedar, Eshita, Sneha Teja Sree Reddy Thondapu et al. (+10 more)
Training tool-augmented LLM
agents requires large corpora of multi-turn, tool-grounded conversational data that is expensive to annotate,
privacy-constrained in production settings, and largely absent from public
datasets. We present StateGen, a synthetic data generation platform that produces scored,
reasoning-trace-rich training conversations by orchestrating a four-role LLM loop: a persona-conditioned user simulator, an
agent under test, a state-grounded tool simulator, and a multi-axis LLM judge. The key architectural contribution is an authoritative state manager that maintains a structured world-state object across turns, enforcing a backend-is-truth invariant that eliminates the dominant class of tool-call
hallucinations by construction. StateGen extends naturally to hierarchical multi-
agent settings by declaring sub-
agents as tools, all sharing a single state object. We report results on 64,698 evaluated conversations across three production corpora: tool-call
hallucination scores reach 9.66/10, the system supports persona-driven variation via a 23-dimensional trait vector, and a cleanly separated train and golden evaluation set split confirms the data is not memorization bait (per-criterion gap analysis). Comparison with eight external systems shows that no single publicly available platform combines multi-turn generation, state-grounded tool simulation, hierarchical multi-
agent support, and built-in judge scoring.
Read the full abstract ▾
All authors Rahul Khedar, Eshita, Sneha Teja Sree Reddy Thondapu, Mayank Malhotra, Arup Das, Jitesh Chandra, Yun-Shiuan Chuang, Chaitanya Kulkarni, Arun Menon, Linsey Pang, Avinash Karn, Mouli V, Prakhar Mehrotra
Computation and Language · Cross-listed from cs.CV
VisualClaw: A Real-Time, Personalized Agent for the Physical World
Haoqin Tu, Jianwen Chen, Zijun Wang et al. (+14 more)
Vision language models are serving as general-purpose interfaces for complex
multimodal tasks. However, deployment still faces three gaps: VLMs typically incur high latency and cost when processing dense
video frames and long
prompts, the
agent scaffold remains static after deployment, and standard
video-QA
benchmarks do not test whether
agents can use visual evidence inside tool-using workspaces. We present VisualClaw, a self-evolving
multimodal agent built around two principles. First, hybrid encoding reduces deployment cost by filtering less informative streaming frames with a cascaded gate and compressing the text skill bank through hot/cold top-k injection. Second, skill
evolution lets the
agent learn from failures: retrieved
memories condition an evolver as direct concatenated context or as guided evidence, producing skill-bank updates that help future questions. Across 4
video-QA
benchmarks with 2 VLMs, VisualClaw cuts per-question API cost by an average -98% versus full-frame upload and by -25.9% over the offline uniform 8 frame baseline, while boosting accuracy in most settings, e.g., an average +3.85% and a peak +15.80% on EgoSchema with Gemini 3 Flash. To address the gap, we curate VisualClawArena, a 200-scenario
multimodal agentic
benchmark built through a strict five-stage pipeline; models must use
video evidence, documents, dynamic updates, and executable checks inside a workspace. On VisualClawArena, the same framework with computer-use
agent backends improves macro accuracy by +2.9% for Codex (GPT-5.5) and +3.2% for Claude Code (Sonnet 4.6) over no-
evolution baselines, with a -9.5% cost reduction compared to the uniform-sampled baseline. These properties make VisualClaw a natural fit for edge applications, where the cascade reduces a 1-hour streaming session from ~3,600 API uploads down to only 5-20 calls and the self-
evolution makes it a perfect personalized assistant.
Read the full abstract ▾
All authors Haoqin Tu, Jianwen Chen, Zijun Wang, Siwei Han, Juncheng Wu, Hardy Chen, Haonian Ji, Kaiwen Xiong, Jiaqi Liu, Peng Xia, Jieru Mei, Hongliang Fei, Jason Eshraghian, Zeyu Zheng, Yuyin Zhou, Huaxiu Yao, Cihang Xie
Computation and Language
HiMPO: Hindsight-Informed Memory Policy Optimization for Less-Entangled Credit in Long-Horizon Agents
Jiangze Yan, Yi Shen, Wenjing Zhang et al. (+6 more)
Long-horizon
agents compress interaction histories into
memory, but final outcomes entangle
memory-writing credit with tool failures and
reasoning errors. We propose HiMPO, a Hindsight-Informed
Memory Policy Optimization framework that supplements trajectory-level reinforcement learning with
memory-specific credit. For each update, HiMPO compares the target answer's log-likelihood under the previous and updated
memories while holding the remaining pre-write context fixed. A soft gate based on group-relative hindsight relevance then weights the local utility advantage. Trajectory-level credit applies to all generated
tokens;
memory tokens additionally receive the auxiliary advantage. Across judge-based open-domain tasks and objective compressive-
memory QA, HiMPO improves task performance while preserving compressed-context efficiency. On Local Wiki Search, it improves average F1 and exact match over MemPO by 2.6 and 2.2 percentage points, respectively. Controlled interventions and live
memory-reversion studies provide evidence of reduced blame leakage and
alignment between
memory credit and behavioral effects. With the policy fixed, the credit
estimator remains robust to the tested forms of target noise.
Read the full abstract ▾
All authors Jiangze Yan, Yi Shen, Wenjing Zhang, Jieyun Huang, Ping Chen, Zhaoxiang Liu, Ning Wang, Kai Wang, Shiguo Lian
Computation and Language
Who Should Lead Decoding Now? Tracking Reliable Trajectories for Ensembling Masked Diffusion Language Models
Heecheol Yun, Joonhyung Park, Joowon Kim et al. (+1 more)
Masked
Diffusion Language Models (MDLMs) have emerged as a distinct paradigm for sequence generation. As MDLMs become diverse in capabilities and knowledge coverage, an important question is how to combine their knowledge. Toward this, we first investigate the unique decoding dynamics of MDLMs. We find that successful generations exhibit stable confidence dynamics over answer-relevant positions, while unreliable trajectories can often be corrected by injecting promising intermediate states from other models. Guided by this observation, we propose $\textbf{TIE}$ ($\textbf{T}$rajectory-based $\textbf{I}$terative $\textbf{E}$nsembling), a knowledge fusion framework in which MDLMs iteratively identify reliable decoding trajectories and relay them across models. TIE tracks confidence dynamics over answer-relevant positions to determine which model currently follows a more reliable trajectory and selectively transfers partially denoised sequences across models. As the model on the more promising trajectory often changes across denoising steps, TIE allows different models to contribute complementary strengths at different stages of generation. Strong performance across diverse
reasoning tasks, along with our analyses, suggests that TIE offers a practical approach to the underexplored problem of MDLM ensembling.
Read the full abstract ▾
All authors Heecheol Yun, Joonhyung Park, Joowon Kim, Eunho Yang
Computation and Language · Cross-listed from cs.AI
UXBench: Measuring the Actionability of LLM-Generated UX Critiques
Wenjie Wang, Yue Huang, Zipeng Ling et al. (+11 more)
Large
language models (LLMs) are increasingly deployed as UX judges that inspect interfaces, diagnose usability problems, and propose repairs. Yet no controlled
benchmark measures whether the resulting critiques are reliable and actionable across heterogeneous product surfaces. We introduce UXBench, a
benchmark for evaluating LLMs as interaction-grounded UX judges. UXBench comprises local-first runnable web fixtures spanning ten product-surface families, paired with coverage-gated browser exploration that forces models to collect interaction evidence before reporting. Each judge model produces a structured UX report over seven rubric dimensions; report quality is measured by whether a fixed downstream repair
agent can improve the interface based on the critique. We evaluate eight frontier models under both an automated repair-lift protocol and a blind human validation study. Results show that UX judging is neither saturated nor one dimensional: models differ meaningfully in report actionability, exhibit distinct rubric-level repair signatures, vary in fixture-level reliability, and trade leadership across surface categories
Read the full abstract ▾
All authors Wenjie Wang, Yue Huang, Zipeng Ling, Han Bao, Hang hua, Xiaonan Luo, Yu Jiang, Shiyi Du, Yuexing Hao, Xiaomin Li, Yuchen Ma, Dianzhuo Wang, Yanfang Ye, Xiangliang Zhang
Computation and Language · Cross-listed from cs.LG
Demystifying Training-Time Augmentation for Data-Constrained Language Model Pretraining
Michael K. Chen, Xikun Zhang, Fan Bai et al. (+2 more)
As AI labs approach a data ceiling where compute capacity outpaces the rate of new high-quality text generation,
language model pretraining is shifting toward a data-constrained, compute-abundant regime that demands productive multi-epoch training on fixed corpora. Standard autoregressive (AR) pretraining overfits severely in this setting, reaching its optimum early and then continuously deteriorating. We investigate training-time data augmentation as a regularizer to mitigate this overfitting and enable productive training for hundreds of epochs on the same data. We introduce three orthogonal categories of augmentation for AR pretraining:
token-level noise (masking, random replacement), sequence permutations (right-to-left prediction, Fill-in-the-Middle), and target offset prediction ($x_{t+i}$ for $i > 1$). Through systematic ablations, we find that individual augmentations delay overfitting and lower validation loss relative to the baseline, with random
token replacement achieving the best minimum loss among individual methods. Combining augmentation categories further lowers the minimum validation loss. Our experiments demonstrate that data augmentations mitigate AR pretraining's data inefficiency and offer a promising solution to the data-constrained regime~\footnote{All code and data are available at https://github.com/ michaelchen-lab/ data-augmentations-for-pretraining.
Read the full abstract ▾
All authors Michael K. Chen, Xikun Zhang, Fan Bai, Zhengding Hu, Zhen Wang
Computation and Language · Cross-listed from cs.LG
LiFT: Local Search via Linear Programming for Overfitting-Controlled Transformers
Abhishek Shukla, Anikeit Khanna, Ankur Sinha et al. (+1 more)
This paper proposes a Linear Programming (LP)-based local search framework for fine-tuning pretrained
transformer models with explicit control against overfitting. The approach formulates
transformer fine-tuning as a bilevel optimization-based regularization problem, in which model parameters and regularization hyperparameters are jointly updated. Information collected during initial warm-up iterations, including validation gradients and training Hessian information, is used to construct a local descent direction by solving an LP that minimizes a scaled directional derivative while preserving training optimality. This validation-aware descent direction enables focused local updates of both parameters and regularization hyperparameters, reducing overfitting without requiring repeated full retraining cycles. The resulting method, termed Linear Programming-based Fine-Tuning (LiFT) for
transformers, differs from conventional fine-tuning by systematically identifying task-specific updates rather than relying on heuristic or grid-based hyperparameter selection. Experiments on GPT-2 Small fine-tuned on WikiText-2 demonstrate that LiFT enables effective adaptation through selective tuning of
transformer blocks and regularization parameters, yielding consistent improvements in test perplexity across multiple layer configurations and regularization settings, with particularly pronounced gains in overfitting-prone scenarios. Beyond empirical performance, LiFT establishes a principled connection between
transformer fine-tuning, bilevel optimization, local search, and regularization theory.
Read the full abstract ▾
All authors Abhishek Shukla, Anikeit Khanna, Ankur Sinha, Faiz Hamid
Computation and Language · Cross-listed from cs.LG
Rapid Poison: Practical Poisoning Attacks Against the Rapid Response Framework
David Huang, Jaewon Chang, Avidan Shah et al. (+2 more)
The Rapid Response (RR) framework, deployed in production systems, including Anthropic's ASL-3 safeguards, continuously improves jailbreak-detection classifiers. When new jailbreaks emerge that bypass these classifiers, Rapid Response generates synthetic variants for training, helping the model generalize from the new attacks and quickly adapt. We reveal that
prompt injection can infiltrate this pipeline to deliver poisoned samples into the classifier's training set, enabling two attack objectives: (I) targeted poisoning attacks that create false positives on harmless samples by categorizing them as a jailbreak, with a specific desired feature (e.g., certain formatting, subject, or keyword), (II) concept-based
backdoor attacks that induce false negatives on jailbreak inputs, generalizing even to jailbreaks from attack strategies the defender explicitly trained against, when the
backdoor trigger is present. Importantly, our threat model restricts adversaries to modifying only jailbreak samples (not benign data or labels), a constraint unexplored by prior work that makes the second objective particularly challenging. We address this with Omission Attack, which exploits a new phenomenon: when training on concept-absent unsafe samples, the classifier misassociates that concept's presence with the safe label. Both attacks cause substantial and in some cases near-complete label flipping at only a 1% poisoning rate, achieving up to 100% false positive rates and up to 96% false negative rates.
Read the full abstract ▾
All authors David Huang, Jaewon Chang, Avidan Shah, Prateek Mittal, Chawin Sitawarin
Computation and Language
Creative Collision: Directorial Persona Steering and Competition in Large Language Models
Subramanyam Sahoo and Justin Shenk
Activation steering has emerged as a powerful tool for shaping the behaviour of large
language models at inference time, yet most prior work injects a single semantic direction into the residual stream. We study the richer setting in which two semantically opposing steering vectors are superimposed – a regime we call Creative Collision. Concretely, we construct directorial persona vectors for Steven Spielberg (optimistic, redemptive moral valence) and Martin Scorsese (dark, morally ambiguous) via mean-difference activation contrast on curated screenplay-derived corpora, then interpolate between them with a scalar mixing parameter $\alpha \in [0,1]$ and a steering coefficient $\lambda$. Across five evaluation axes – moral valence, generation coherence, surface style, directional dominance, and vector geometry – three principal findings emerge: (i)~Spielberg's representational signature exhibits robust directional dominance, suppressing Scorsese's moral influence across almost the entire interpolation range; (ii)~intermediate collision points paradoxically improve generation coherence relative to pure single-director steering at high $\lambda$; and (iii)~both personas localise maximally to layer~28 of a 40-layer decoder-only
transformer, revealing a shared moral-tone substrate. These results illuminate the geometry of competing semantic directions in
transformer residual streams and have direct implications for controllable creative generation and value-aligned narrative synthesis.
Read the full abstract ▾
Computation and Language
PACT: Privileged Trace Co-Training for Multi-Turn Tool-Use Agents
Zhenbang Du, Jun Luo, Zhiwei Zheng et al. (+8 more)
Multi-turn tool-use
agents must reason, call tools, and adapt to observations across several interaction turns. Post-training such
agents is challenging, as reinforcement learning often suffers from sparse rewards and weak credit assignment despite matching the
prompt-only inference setting, while supervised fine-tuning on expert traces provides dense process supervision but can over-constrain the model to fixed trajectories. To tackle this, we propose PACT, a Privileged trAce Co-Training framework for multi-turn tool-use
agents. The key idea is to use expert traces only as training-time optimization signals rather than rollout-time hints. PACT keeps rollout generation
prompt-only, then uses expert traces to guide optimization through two complementary signals: a trace-conditioned RL surrogate that evaluates
prompt-only rollouts under expert-trace context, and a component-aware SFT loss that provides annealed process supervision over
reasoning and tool-call components. To reduce over-reliance on the training-only trace context, PACT incorporates
prompt-only anchoring with standard
prompt-only RL updates. Experiments on FTRL, BFCL, and ToolHop show that PACT consistently outperforms strong SFT- and RL-based baselines. Further analyses support the complementarity of the two training objectives and the benefits of component-aware supervision and
prompt-only anchoring. These results highlight the effectiveness of privileged trace co-training for multi-turn tool-use learning. The project page is available at https://zhenbangdu.github.io/pact-project-page.
Read the full abstract ▾
All authors Zhenbang Du, Jun Luo, Zhiwei Zheng, Xiangchi Yuan, Kejing Xia, Dachuan Shi, Qirui Jin, Qijia He, Shaofeng Zou, Yingbin Liang, Wenke Lee
Computation and Language
Weaving Multi-Source Evidence for Biomedical Reasoning: The BioMedHop Benchmark and BioWeave Framework
Xingyu Tan, Shiyuan Liu, Xiaoyang Wang et al. (+5 more)
Biomedical question answering (QA) increasingly requires
reasoning over interacting entities, where supporting evidence is scattered across biomedical knowledge graphs, literature documents, and web-accessible resources. However, existing biomedical QA
benchmarks mainly focus on exam-style knowledge, literature comprehension, or short-range multi-hop inference, leaving source-conditioned graph
reasoning and evidence
topology construction underexplored. To fill this gap, we introduce BioMedHop, a multi-source graph-grounded
benchmark for evaluating biomedical
reasoning over structured evidence
topologies. BioMedHop contains 10,045 instances across KG, document, web, and hybrid evidence settings, covering shared-neighbor matching, intersection
reasoning, path-based
reasoning, and counting, with option-based, open-ended, and numeric count renderings. To support this
benchmark, we further propose BioWeave, a source-aware
reasoning framework that retrieves biomedical KG paths, gathers supporting clues from documents and web sources, assembles them into a unified evidence graph, and verifies answers through entity-level evidence support. Comprehensive experiments show that BioWeave achieves the best overall performance among compared methods on BioMedHop, outperforming the strong hybrid baseline ToG-2 by 10.5% in the overall average. Moreover, BioWeave consistently improves different LLM backbones and enables smaller models, such as Qwen3-4B, to achieve
reasoning performance comparable to GPT-4-Turbo.
Read the full abstract ▾
All authors Xingyu Tan, Shiyuan Liu, Xiaoyang Wang, Qing Liu, Xiwei Xu, Xin Yuan, Liming Zhu, Wenjie Zhang
Computation and Language · Cross-listed from cs.AI
Beyond Helpfulness: A Teaching-over-Solving Diagnostic for Measuring Educational Impact in LLM Tutors
Junyi Yao, Zihao Zheng and Baichuan Li
Large
language models are increasingly proposed as educational tutors, yet stronger task-solving ability does not necessarily imply stronger learning support. Motivated by recent calls to measure the social impact of NLP systems in practice, we study whether public LLM tutoring
benchmarks distinguish learning-supportive behavior from mere answer production. We propose a lightweight diagnostic based on the gap between solving-oriented and pedagogy-oriented
benchmark performance. Using public MathTutorBench leaderboard results, we show that these dimensions are only partially aligned: across eight publicly reported models, the correlation between solving and pedagogy composites is 0.421, and several models shift meaningfully in rank when evaluation moves from solving to pedagogy. We then analyze the public TutorBench sample and show that agency-relevant behaviors are explicitly encoded in
benchmark rubrics, especially in active-learning settings that reward guiding questions, calibrated hints, and non-disclosive scaffolding. Together, these findings suggest that educational-impact evaluation should not treat task success as a sufficient proxy for learning support. We argue that public tutoring
benchmarks can better support positive-impact evaluation by reporting solving-oriented and pedagogy-oriented scores separately and by making disclosure-sensitive, student-agency-preserving criteria more explicit.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.LG
LLM-Powered Virtual Population for Demand Simulation and Pricing
Chengpiao Huang and Kaizheng Wang
We develop an LLM-powered virtual population model that simulates demand for pricing decisions, in settings where products are described by rich unstructured information, such as text descriptions and images, and where decision makers need not only mean-demand predictions but also uncertainty estimates for counterfactual prices. Our model represents exposed customers as draws from a finite mixture of customer personas. For each persona, product, and candidate price, an LLM elicits a persona-level purchase probability using both structured persona information and unstructured product information. These probabilities are aggregated through calibrated mixture weights to form a predictive distribution of aggregate demand. The resulting simulator can evaluate counterfactual prices under various pricing objectives, including expected revenue and risk-aware criteria such as conditional value at risk. We test the framework on an online H&M fashion
dataset with product descriptions and images. The calibrated LLM-based simulator achieves the best overall predictive performance among the models considered, and supports sample-efficient pricing decisions. Our framework provides a practical way to use LLMs as demand simulators for products with limited historical demand data but rich product information. By producing a full predictive demand distribution rather than only a point forecast, it enables managers to compare candidate prices, quantify demand uncertainty, and choose prices that target either average-case revenue or risk-aware objectives.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.CV
Focus When Necessary: Adaptive Routing and Collaborative Grounding for Training-Free Visual Grounding
Yifan Wang, Peiming Li, Shiyu Li et al. (+5 more)
While
Multimodal Large
Language Models (MLLMs) excel in cross-modal
reasoning, they often struggle to perceive fine-grained details in complex high-resolution images. Recent training-free methods address this through image scaling and localized cropping. However, applying these manipulations indiscriminately introduces computational redundancy for simple queries and can degrade accuracy by truncating essential global context or introducing irrelevant background noise. To this end, we propose LazyMCoT, a dynamic and training-free framework that adaptively allocates visual grounding efforts based on sample difficulty. The framework features an Adaptive Routing mechanism that evaluates predictive uncertainty using first-
token statistics from a single forward pass. This efficiently bypasses confident cases while ensuring the recall of difficult samples via conformal calibration. For these challenging cases, a Collaborative Grounding module integrates the inherent cross-modal
attention of the model with an external visual expert through a two-stage refinement process. This refinement process generates a precise localized display to recover small or occluded targets. Extensive experiments across diverse
benchmarks demonstrate that LazyMCoT rivals training-based approaches by simultaneously improving
reasoning accuracy and reducing average inference latency. Our code is availble at https://github.com/TencentBAC/LazyMCoT.
Read the full abstract ▾
All authors Yifan Wang, Peiming Li, Shiyu Li, Zhiyuan Hu, Xiaochen Yang, Wenming Yang, Yang Tang, Zheng Wei
Computation and Language
GRACE: Step-Level Benchmark for Faithful Reasoning over Context
Hoang Pham, Dong Le and Anh Tuan Luu
Many
reasoning tasks require models to reason over input context, from document-grounded question answering to rule-based deduction. Chain-of-Thought (CoT) prompting produces traces that appear transparent, yet individual steps can silently deviate from the source evidence, even when the final answer is correct. Existing methods detect
hallucinations at the response level but fail to identify where in the chain a failure occurs or what type it is. We introduce GRACE, the first human-annotated step-level faithfulness
benchmark with a data-driven error taxonomy for context-grounded textual
reasoning. GRACE covers CoT traces from 10 models across 4 source
datasets, with each step annotated for faithfulness, error category, and natural
language explanation. A data-driven taxonomy, discovered bottom-up via unsupervised clustering, organizes failures into two tracks: GRACE-Inference (deductive errors) and GRACE-Grounding (factual grounding errors), with four categories each. The evaluation set is human-annotated and challenging by design. Our experiments reveal substantial headroom for current models. In addition, integrating step-level faithfulness signals into reinforcement learning pipelines improves both downstream accuracy and
reasoning reliability.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.AI
VibeThinker-3B: Exploring the Frontier of Verifiable Reasoning in Small Language Models
Sen Xu, Shixi Liu, Wei Wang et al. (+6 more)
This technical report introduces VibeThinker-3B, a compact dense model with 3B parameters developed to investigate how far verifiable
reasoning can be pushed within a strictly small-model regime. Building upon the Spectrum-to-Signal post-training paradigm, we systematically enhance the model through an optimized pipeline that includes curriculum-based supervised fine-tuning, multi-domain reinforcement learning, and offline self-
distillation. Experimental evaluations demonstrate that VibeThinker-3B achieves frontier-level performance on highly demanding verifiable tasks. Specifically, it attains a score of 94.3 on AIME26 (improving to 97.1 with claim-level test-time scaling), an 80.2 Pass@1 on LiveCodeBench v6, and exhibits strong out-of-distribution generalization with a 96.1% acceptance rate on recent unseen LeetCode contests. This effectively places it in the performance band of first-tier
reasoning systems, matching or exceeding flagship models that are orders of magnitude larger, such as DeepSeek V3.2, GLM-5, and Gemini 3 Pro. Furthermore, a score of 93.4 on IFEval confirms that this extreme
reasoning enhancement does not compromise strict instruction controllability. Extending our previous 1.5B work, these findings motivate the Parametric Compression-Coverage Hypothesis, which views verifiable
reasoning as compressible into compact
reasoning cores, while open-domain knowledge and general-purpose competence require broad parameter coverage over facts, concepts, and long-tail scenarios. This perspective suggests that compact models are not merely deployment-efficient substitutes, but a complementary path toward frontier-level performance in parameter-dense capability regimes.
Read the full abstract ▾
All authors Sen Xu, Shixi Liu, Wei Wang, Jixin Min, Yingwei Dai, Zhibin Yin, Yirong Chen, Xin Zhou, Junlin Zhang
Computation and Language
XAI-Grounded Explanation Generation for Speech Deepfake Detection with Training-Free Multimodal Large Language Models
Yupei Li, Qiyang Sun, Xiaoliang Wu et al. (+3 more)
Speech deepfake detection (SDD) systems require trustworthy explanations for reliable decision-making. Existing explanation ways mainly fall into two categories. Traditional explainable AI (XAI), such as gradient-based attribution, produces low-level attribution signals tightly coupled with model decisions, and harder to be understood by human than natural
language explanations. Meanwhile, large
language model (LLM)-based explanation generation often produces generic and ungrounded descriptions due to the lack of heuristic evidence and task-specific supervision, stemming from limited grounded explanation
datasets for SDD. We therefore propose a training-free explanation framework that integrates XAI evidence with
multimodal LLMs to generate grounded and specific explanations. Using the PartialSpoof
dataset, we construct a grounded explanation
dataset and show that methods with XAI increase inside accuracy by over 45%, verified through human evaluation and faithfulness checks.
Read the full abstract ▾
All authors Yupei Li, Qiyang Sun, Xiaoliang Wu, Chenxi Wang, Berrak Sisman, Björn W. Schuller
Computation and Language
Andreas Einwiller, Max Klabunde and Florian Lemmerich
The worldwide rise of authoritarianism and the growing role of Large
Language Models (LLMs) in users' everyday lives raise the question of whether specific models exhibit or promote authoritarian attitudes. We introduce AuAu, a comprehensive
benchmark for assessing the risk of authoritarian tendencies in LLM responses. AuAu combines three evaluation approaches: (i) psychometric questions from 15 human-validated instruments, (ii) vignettes probing intended behavior in concrete situations, and (iii) responses to realistic user
prompts. Unlike prior work, AuAu measures not only overall authoritarian
alignment but also its established sub-concepts: Authoritarian Aggression, Authoritarian Submission, and Conventionalism. Evaluating 17 models from China, the EU, Russia, and the USA, we find substantial authoritarian response rates on psychometric instruments across all models, though rates drop significantly on more realistic downstream tasks. Moreover, a simple authoritarian system
prompt manipulates 15 of 17 models into promoting increased authoritarianism. Our results underscore the need for continued, systematic
auditing of LLM-based AI systems to detect and mitigate authoritarian tendencies in their outputs.
Read the full abstract ▾