Computation and Language
The Geometry of Knowledge Accessibility in Large Language Models
Lihu Chen
Large
language models (LLMs) contain broad knowledge, but they cannot access all of it reliably. We study this problem through knowledge accessibility, which describes whether the knowledge needed for a query can be recalled from the model. We find that knowledge accessibility has a simple
geometric structure in the model's representation of the query alone, before any generation. More accessible queries are closer to a center in the representation space, while less accessible queries are farther away. This geometry reveals a knowledge boundary that separates more accessible queries from less accessible ones. Accessibility consistently decreases with distance from the center, and this distance-based ordering transfers across
datasets even when the centers differ. Controlled experiments further show that the centered geometry is more closely related to knowledge accessibility than to
reasoning difficulty. The geometry also reveals when different interventions are useful. Query rewriting helps more for accessible queries, chain-of-thought
reasoning helps more near the boundary, and
retrieval gives larger gains beyond the boundary. These findings not only provide a new
geometric view of how knowledge is organized in
language models, but also suggest a useful pre-generation signal for adaptive inference.
Read the full abstract ▾
Computation and Language
HyperThink: Text-to-Parameter Hypernetworks for Efficient Reasoning
Donggyun Kim, Jack Lu, Chanwoo Kim et al. (+2 more)
Long-form thinking traces can substantially improve the multi-step
reasoning performance of large
language models (LLMs), but they introduce high inference-time overhead, with latency dominated by sequential decoding. We propose HyperThink, a text-to-parameter approach that amortizes this
reasoning computation into a single query-conditioned parameter update: a lightweight hypernetwork reads the question and predicts updates to a small subset of the base LLM's parameters, while a vector-quantized decoder constrains them to a finite set of reusable patterns to improve robustness and transfer. Trained end-to-end on outputs from the base model itself, HyperThink eliminates long thinking traces at test time: after one hypernetwork forward pass, the adapted model generates a concise step-by-step solution and final answer without an intermediate trace, using far fewer
tokens while retaining strong
reasoning performance. Empirically, HyperThink improves the low-latency region of the accuracy-latency trade-off on mathematical and general
reasoning tasks, with its strongest gains in the near-non-thinking regime.
Read the full abstract ▾
All authors Donggyun Kim, Jack Lu, Chanwoo Kim, Mengye Ren, Seunghoon Hong
Computation and Language · Cross-listed from cs.LG
Adaptive Second-Order Solvers for Fast Stochastic Diffusion Sampling
Ella Kemperman and Luca Ambrogioni
Diffusion models rely on numerical solvers requiring time-discretization, which has a large influence on the tradeoff between sampling cost and quality. However, the computational difficulty of the reverse process varies along the sampling trajectory and across data distributions, making the choice of discretization important. We adapt proportional-integral (PI) step-size control to
diffusion, using our
diffusion noise-normalised error
estimator. Unlike existing adaptive methods in
diffusion that respond only to the current error, the PI solver also incorporates the previous error, yielding smoother step adaptation. We further show that these per-sample trajectories exhibit shared structure and can be aggregated into a fixed schedule that retains much of the benefit of adaptive sampling. We evaluate both approaches on natural-image and
language datasets, in terms of quality, measured by FID at a matched number of neural network evaluations (NFE), comparing them with widely used
stochastic solvers and schedules. For images, our fixed discretization outperforms the commonly used EDM schedule in terms of sample quality when used with the
stochastic Heun sampler, and with the EDM-churn sampler at low NFE. Additionally, our PI adaptive solver obtains better FID than most
stochastic and adaptive baselines, although it does not beat the EDM-churn sampler at low NFE. Moreover, we find our solver outperforms both the EDM and the
entropy schedule on
language diffusion at low-to-medium NFE in terms of perplexity, with the drawback of lower
token entropy. Lastly, we find that the benefit of per-sample adaptivity is problem-dependent. It is highly beneficial in 1D toy examples, while only marginal for image and
language data, where the average schedule sometimes even outperforms the PI-adaptive solver. Code is available at https://github.com/ellakemperman/adaptive-second-order-
diffusion-solvers
Read the full abstract ▾
Computation and Language · Cross-listed from cs.LG
Tailoring the Quantization Space for 1-Bit KV Cache Compression
Minsoo Cheong, Donghyun Son and Sungjoo Yoo
The key-value (KV) cache becomes a major
memory bottleneck in long-context LLM inference, placing substantial pressure on
memory capacity and bandwidth. To mitigate this bottleneck, vector quantization (VQ) has emerged as a promising approach for aggressive KV cache compression. However, existing VQ methods degrade substantially in the 1-bit regime. At such extreme compression, each codebook must represent a larger group of channels with a limited set of centroids, making effective use of its capacity increasingly challenging. To address this, we introduce $\textbf{TaSQ}$, which tailors the VQ target space by combining query-guided channel weighting, cross-head normalization, and covariance-aware channel grouping to better reflect the error sensitivity and statistical structure of cached activations. Since these transforms are RoPE-compatible and can be easily merged into projection weights and codebooks, TaSQ preserves the conventional VQ lookup structure and adds negligible serving overhead. Across general, long-chain-of-thought
reasoning, and long-context
retrieval benchmarks, TaSQ consistently outperforms existing low-bit KV cache VQ baselines while preserving
reasoning stability. On a single RTX 6000 Ada GPU, its SGLang implementation supports up to $14\times$ larger batch sizes and achieves $1.87\times$ higher peak throughput compared to the BF16 baseline.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.AI
Verifiable, Articulable, and Tacit Components of Preference
Alexander Spangher, Sheldon S. Huang, Andreas Haupt et al. (+4 more)
What makes a short story gripping; a news article newsworthy; or a math proof elegant? These constructs resist articulation or verification; their meaning is at least partially tacit. However, modern AI models are improved primarily via articulated constitutions, rubrics and verifiers (i.e. in RLAIF and RLVR); tacit components of preferences are typically understudied. We introduce a large, labeled preference
dataset CreativePreferences, containing 2.8M texts labeled by 317M human preference judgments across 7 creative domains, with 42
benchmark tasks. We model these labels with executable programs, rubric banks and densely trained models (V, A and VAT, respectively). We observe robust articulability gaps, VAT-VA; and verifiability gaps, VAT-V; we estimate upper and lower bounds for each gap with a novel measurement approach that discovers articulable and verifiable metrics, identifies spurious variables and estimates the value of undiscovered metrics using capture-recapture. These gaps occur across all domains, even in domains traditionally treated as fully verifiable: correctness-centered domains (i.e. mathematics and software engineering) and claim- and novelty-centric domains (i.e. news, patents, peer review). The size of the gap varies based on domain (e.g. peer review and creative writing have the largest articulability gaps) and widens as more people take part in the judgment, consistent with Collins' collective tacit knowledge. We show two consequences: (1) on human generations, the full model more closely matches human preferences, often in disagreement with articulated criteria, and (2) in an analogy to Goodhart's law, articulating preference shifts it away from the tacit dimension. Articulability and verifiability gaps are consequential; we give recommendations on when tasks can be prompted; how learning mechanisms might improve; and when to leave judgments with humans.
Read the full abstract ▾
All authors Alexander Spangher, Sheldon S. Huang, Andreas Haupt, Noah D. Goodman, Diyi Yang, Daniel E. Ho, Sanmi Koyejo
Computation and Language · Cross-listed from cs.CV
ReSCUE: Re-translation with Sentence Commitment for Unsegmented Long-Form Simultaneous Sign Language Translation
Sihan Ren, Gaozheng Li, Yuanshang Quan et al. (+5 more)
Simultaneous Sign
Language Translation (SLT) is critical for real-time communication, yet existing methods remain largely confined to sentence-level, offline settings that assume pre-segmented inputs. These assumptions hinder deployment in realistic scenarios involving continuous, unsegmented
video streams. We present ReSCUE, a unified framework for simultaneous SLT on unsegmented long-form sign
language videos that aligns training and inference with realistic streaming conditions. ReSCUE combines inference-aware training to handle partial inputs, non-signing pauses, and multi-sentence contexts, stabilized re-translation to enable low-latency yet revisable predictions with reduced output flicker, and a sentence commitment mechanism for online segmentation and
memory management. Experiments on standard sentence-level
benchmarks show that ReSCUE achieves lower latency and the best translation quality under low-latency settings. On long-form unsegmented
datasets, ReSCUE approaches the translation quality of oracle offline systems that use ground-truth sentence boundaries, while operating at substantially lower latency, demonstrating its practicality for real-world streaming scenarios.
Read the full abstract ▾
All authors Sihan Ren, Gaozheng Li, Yuanshang Quan, Yiming Qin, Fuyi Yang, Chang Liu, Lan Xu, Minye Wu
Computation and Language · Cross-listed from cs.AI
Personalized Automatic Speech Recognition for a Dysarthric and Tracheostomic Speaker using Artificial Conversations
David Nadrchal, Monorama Swain, Florian Schmid et al. (+2 more)
This work presents an automatic
speech recognition (ASR) system personalized for a Czech speaker with a permanent tracheal stoma and severe dysarthria rendering their
speech unintelligible to untrained listeners. We release a public
dataset containing 33 annotated hours of the speaker's
speech, collected using a novel "artificial conversation" protocol designed for high engagement and dialogue realism. We propose a multi-stage training pipeline based on Whisper Base: fine-tuning on standard Czech
speech, acoustically simulated tracheostomic
speech, and the speaker's data. We evaluate the system across three near real-time scenarios: scripted conversations, question answering, and spontaneous dialogue, achieving a 50% relative reduction in Character Error Rate compared to Whisper Base baseline and surpassing the average recognition accuracy of their assistants in acoustic recognition of isolated utterances. We demonstrate that even for severely impeded
speech, a helpful ASR is achievable, as evidenced by the quantitative results and the feedback from the speaker.
Read the full abstract ▾
All authors David Nadrchal, Monorama Swain, Florian Schmid, Gerhard Widmer, Paul Primus
Computation and Language
Recursive Self-Improvement in Unified Multimodal Models
Huijuan Wang, Chufan Shi, Cheng Yang et al. (+3 more)
Unified
multimodal models (UMMs) understand and generate both text and images, which lets a model produce its own training data. Existing self-improvement in UMMs keeps supervision on the visual side, where image understanding judges image generation. We propose recursive cross-capability self-improvement (RSI), a training loop in which the text and visual abilities of a UMM supply training data for one another. In each round, the model generates images and reads them to find where it falls short. It then writes programs aimed at these shortcomings, and execution verifies every result against its specification. Verified renders train image generation, while labeled renders and the model's own correct programs train visual understanding and program writing. Program execution thus acts as a source of truth outside the model, so errors do not accumulate across rounds. We study RSI on charts and build BasicChartBench to evaluate open models early in training. On requests worded differently from training, four rounds of RSI raise the score from 45.7% to 60.2%, while continued training stays at 46.3%. Verified construction carries most of the gain, and targeting the model's failures adds 3.5%. Along the way, the share of verified programs rises from 48.9% to 95.2%, and the reader's accuracy on edited renders rises from 55.6% to 87.4%.
Read the full abstract ▾
All authors Huijuan Wang, Chufan Shi, Cheng Yang, Yaokang Wu, Taylor Berg-Kirkpatrick, Xuezhe Ma
Computation and Language
OmniConfess: Eliciting Token Confessions to Mitigate Omni-Modal Hallucination
Huiqiang Rong, Haoran Luo, Hui Feng et al. (+4 more)
Omni-modal large
language models (OmniLLMs) unify text, images, audio, and
video, yet hallucinate when generation relies on the wrong evidence. Existing inference-time methods can reduce
hallucinations, but rarely reveal which evidence sustains a generated commitment. We introduce OmniConfess, a training-free method for mitigating omni-modal
hallucinations. It fixes a candidate response and re-scores it at
token resolution under controlled channel-wise evidence interventions, producing a structured
token-by-channel confession that reveals the response's evidential dependence. OmniConfess uses this confession to preserve grounded content and correct commitments driven by irrelevant or contradictory evidence. To evaluate OmniConfess, we construct OmniHalluBench, a 3,540-example
benchmark built from six
datasets spanning text, image, audio, and
video settings and both judgment and free-form generation. Experiments show that OmniConfess mitigates
hallucinations across heterogeneous modality and task settings. Our code and
benchmark are publicly available at https://github.com/RongHuiQiang/OmniConfess.
Read the full abstract ▾
All authors Huiqiang Rong, Haoran Luo, Hui Feng, Zhonghong Ou, Kaiwen Xue, Guoxin Zhang, Yifan Zhu
Computation and Language · Cross-listed from cs.LG
Sentry: Learning to Recover from LLM Agent Failures at Test Time
Changxiu Ji, Amy Lu, Qizheng Zhang et al. (+1 more)
LLM
agents often fail mid-task due to invalid tool calls, repeated actions, or poorly grounded
reasoning, and learning from these failures is a path to reliability. We find that how failure knowledge reaches the
agent matters as much as what it contains. Failure lessons are conditional: kept in the
agent's context, they misfire when their failure is absent, and removing them from an evolving playbook improves performance. Runtime interventions, in contrast, act only when a failure occurs but do not learn from their repairs. We argue that failure knowledge is conditional knowledge and should be conditionally exposed, and instantiate this principle in Sentry, a failure-management layer that runs alongside the
agent. When Sentry detects a failure, it retrieves matching lessons from an external playbook to guide recovery, verifies without access to task rewards whether the
agent recovered, and stores a new lesson only if it did; the full playbook never enters the
agent's context. Across multiple agentic
benchmarks, Sentry outperforms the strongest runtime-intervention baseline on every
benchmark, by 37% on average, and the strongest context-
evolution baseline by 39% on the two
benchmarks where both are evaluated; combining Sentry with context
evolution yields further gains. Learned lessons transfer to held-out tasks, and controlled experiments show that exposing the full playbook to the
agent lowers performance even when relevant lessons remain available on demand.
Read the full abstract ▾
All authors Changxiu Ji, Amy Lu, Qizheng Zhang, Kunle Olukotun
Computation and Language
OLMo-Detect: A Multi-Stage, Confounder-Controlled Benchmark for Membership Inference on Large Language Models
Tao Shi, Chaoyi Xiang, Qiongkai Xu et al. (+1 more)
Membership inference on large
language models (LLMs) aims to determine whether a given text sample was included in an LLM's training data, without access to its training corpus. Despite recent progress, existing
benchmarks suffer from three limitations: limited coverage of training stages, insufficient distributional
alignment between members and non-members, and lack of rigorous filtering of non-members against the training corpus. To address these limitations, we propose OLMo-Detect, a multi-stage, confounder-controlled
benchmark built upon the fully open OLMo 2 pipeline. OLMo-Detect spans pre-training, mid-training, and post-training, explicitly aligns members and non-members on three key axes, and rigorously filters non-members via infini-gram. To assess robustness to distribution shifts, we further introduce OLMo-Detect (Shifted), a variant where members are misaligned with non-members. We evaluate 15 unsupervised and 3 supervised membership inference attacks (MIAs) across the OLMo 2 family, finding that: (i) overall performance is limited: the best unsupervised and supervised MIAs both reach an AUC of only 0.68, and supervised MIAs degrade under cross-domain evaluation; (ii) MIA performance peaks at mid-training and is lower at pre-training and post-training, a pattern driven by data type rather than a stage effect: curated math data is far more detectable than other types; (iii) overall scores improve from 1B to 13B but plateau at 32B; and (iv) no unsupervised MIA is robust to distribution shifts, with AUCs shifting by up to 0.42. Finally, we find that our findings on OLMo 2 generalize to OLMo 3 and non-OLMo models.
Read the full abstract ▾
All authors Tao Shi, Chaoyi Xiang, Qiongkai Xu, Jey Han Lau
Computation and Language
A Guideline-Augmented Multi-Agent Framework for Schema-as-Code Biomedical Named Entity Recognition
Songtao Li, Yijia Zhang, Shidi Zhang et al. (+3 more)
Large
language models (LLMs) have shown promising potential for biomedical named entity recognition (BioNER) through instruction following and in-context learning. However, existing LLM-based BioNER methods still face two key limitations. First, retrieved demonstrations and external biomedical knowledge provide limited support for
dataset-specific annotation semantics, leaving entity boundaries, type scopes, and annotation conventions ambiguous. Second, free-form generation lacks sufficient structural control, often leading to invalid formats, hallucinated mentions, duplicated entities, and boundary errors. To address these limitations, we propose GAMA, a guideline-augmented multi-
agent framework for schema-as-code BioNER. GAMA first induces candidate annotation rules from labeled training instances and verifies them against annotated data to construct reliable
dataset-specific guideline
memory. Guided by these verified rules, a planning component generates ranked span-type hypotheses with rationales, and a coding component converts them into schema-constrained entity objects. A verification module then checks span grounding, type validity, and structural compliance, and performs dual-loop refinement to correct invalid or low-confidence predictions. Experiments on five widely used BioNER
datasets with multiple LLM backbones show that GAMA consistently outperforms strong LLM-based baselines. Ablation and parameter analyses further verify the effectiveness of the proposed components.
Read the full abstract ▾
All authors Songtao Li, Yijia Zhang, Shidi Zhang, Jianyuan Yuan, Fengyu Zhang, Hongfei Lin
Computation and Language · Cross-listed from cs.LG
Understanding Trajectory Heterogeneity in Federated World Model Learning
Yipan Wei, Zhaokun Yan, Ziming Hong et al. (+2 more)
World models learn state
evolution from trajectories, making access to temporal context a central training requirement.
Federated learning can use distributed records, while ownership boundaries within a trajectory restrict the examples each client can construct. Our study
benchmarks this cross-time setting through hourly action-conditioned clinical prediction on eight MIMIC-IV disease cohorts, comprising 40.87 million transition memberships. We specify severity-based client ownership, patient-separated construction, local history and future-window rules, and paired rollout evaluation from one to 32 hours. A
matrix of ten
federated algorithms covers 32 disease–partition configurations under five rounds of ten-percent participation. Three findings emerge from existing results and training logs. First, client ownership and participation jointly restrict long-window coverage: only 7.55%–21.36% of pooled-available 32-step windows have a locally complete anchor visited during training, averaged across diseases. Second, finer severity partitions accompany higher FedAvg error in 15 of 16 paired comparisons, while
algorithm gains are small and horizon-dependent: FedProx reduces mean error by 0.56%, with no consistent improvement at 32 steps. Third,
algorithm labels conceal distinct update behavior, including inactive extrapolation and orders-of-magnitude differences in update scale. Cached-update performance also varies strongly across trajectory partitions under the same
benchmark protocol. These results establish temporal access, participation coverage, optimization behavior, and horizon-resolved prediction as complementary dimensions for evaluating
federated clinical world models.
Read the full abstract ▾
All authors Yipan Wei, Zhaokun Yan, Ziming Hong, Jiaqi Wu, Lixu Wang
Computation and Language
Enhancing Biomedical Named Entity Recognition via Multiple Programming Languages Instruction Tuning and Ensemble Method
Songtao Li, Yijia Zhang, Jianyuan Yuan et al. (+3 more)
Instruction tuning has become a common paradigm for applying large
language models (LLMs) to biomedical named entity recognition (BioNER). However, existing instruction-tuning approaches still face two key challenges. First, conventional natural-
language instructions typically serialize BioNER annotations as flat textual outputs, providing limited structural constraints for typed entity extraction. Second, high-quality biomedical annotations are limited, and learning from a single serialized output form may restrict structural diversity and reduce model robustness. Although external biomedical knowledge can be introduced to alleviate data scarcity, it often requires costly resource construction. To address these challenges, we propose MITE, a Multiple Programming
Languages Instruction Tuning and Ensemble method for BioNER. MITE reformulates BioNER as a structure-to-structure generation task by representing both instructions and entity outputs in code-formatted representations. Specifically, each training instance is transformed into multiple programming-
language formats, including Python, C++, and Java, while preserving the same underlying entity semantics. These
language-specific representations provide structurally diverse supervision without requiring external biomedical knowledge or additional annotations. During inference, MITE aggregates predictions from different code formats through an entity-level voting strategy, reducing
language-specific prediction variance and improving robustness. Experiments on six widely used BioNER
datasets demonstrate that MITE consistently outperforms representative BERT-based and LLM-based baselines and exhibits strong cross-
dataset generalization. Ablation and parameter analyses further verify the effectiveness and robustness of the proposed components.
Read the full abstract ▾
All authors Songtao Li, Yijia Zhang, Jianyuan Yuan, Shidi Zhang, Fengyu Zhang, Hongfei Lin
Computation and Language · Cross-listed from cs.AI
Continual Graph Memory for Mathematical Research Agents
Junyi Zhang, Jinxi Yu, Eric Hanchen Jiang et al. (+14 more)
Using frontier
agent harnesses to tackle mathematical research problems has emerged as an effective means of advancing mathematics. However, solving frontier problems in mathematics may require a massive number of
agents working in parallel for extended periods to construct proofs, thereby generating an enormous volume of intermediate proof results. Organizing these intermediate results throughout a long-horizon proof-search process and reusing knowledge gained from prior explorations remain major challenges. We present Ansatz, a mathematical research
agent built around Continual Graph
Memory, a graph-based, evolvable, cross-problem mathematical research
memory system that explicitly organizes the entire proof search process and reuses information from exploration trajectories of previous problems. Specifically, we develop a unified graph
memory that represents all intermediate exploration results, including facts, plans, and counterexamples, together with edges that explicitly represent the relationships among them; dependency-aware
retrieval supplies precisely targeted local context; an evidence-sensitive curator updates the research frontier and distills lessons from prior attempts; and scoped recall surfaces earlier statements and negative findings for local re-proving rather than uncritical reuse. Experiments cover runs across all ten First Proof Second Batch problems, together with four component studies. Ansatz reports closure on all ten research tasks, demonstrating its ability to sustain and resume long-horizon mathematical search. Beyond these problems, Ansatz also produces solutions to the Jamison caterpillar
conjecture and Erdős Problems 289, 348, and 488 without human intervention, and makes partial progress on several open problems, illustrating its strong ability to solve open mathematical research problems.
Read the full abstract ▾
All authors Junyi Zhang, Jinxi Yu, Eric Hanchen Jiang, Jiachen Lu, Zhi Zhang, Xinjie He, Hyunsik Chae, Ethan Ji, Alexander K Taylor, Vigyan Sahai, Yiwen Kou, Kai-Wei Chang, Raghu Meka, Nanyun Peng, Amit Sahai, Terence Tao, Wei Wang
Computation and Language
Output Language Confusion under Multilingual Prompt Contamination
Riju Marwah, Ritvik Garimella, Khusham Bansal et al. (+2 more)
Standard factual
benchmarks assume clean monolingual
prompts and exact-match scoring, two assumptions that break simultaneously in real-world multilingual deployment, from
retrieval-augmented generation pipelines returning mixed-
language passages to users pasting multilingual web content. We introduce Multilingual Distractor Interference (MDI), a lightweight and fully replicable evaluation protocol requiring no new data or annotation, in which factual questions are preceded by a semantically irrelevant foreign-
language sentence, and evaluate five instruction-tuned LLMs across TruthfulQA and TriviaQA under eight distractor conditions (40,000 evaluations). Our central finding is a metric confound: for Llama-3.1-8B under a Hindi distractor, 58% of responses switch to Devanagari script, yielding a raw
hallucination proxy of 0.710, but manual review reveals that 120 of 148 script-switched responses that were correct under clean conditions remain semantically correct despite being written in the wrong script, reducing the adjusted semantic
hallucination rate to 0.470. All other models respond through abstention escalation with no
hallucination increase. A paragraph-length English distractor triggers near-universal abstention (0.806-0.998) across all models, consistent with reading-comprehension confusion, a failure mode with direct consequences for multilingual RAG pipelines. TruthfulQA multiple-choice accuracy is unaffected under all single-sentence conditions. These results show that exact-match
hallucination rates in mixed-
language settings should be decomposed into script-switching and semantic error components before drawing conclusions about model reliability.
Read the full abstract ▾
All authors Riju Marwah, Ritvik Garimella, Khusham Bansal, Atishay Jain, Amit Sheth
Computation and Language · Cross-listed from cs.LG
Probe the Harness: Setup Checks for Stale-Data RL Comparisons in Language Models
Taiheng Pan
Methods for training
language models on stale samples are judged by comparisons against importance-corrected baselines. We show that details of the experimental harness can reverse the observed ranking of methods, and we introduce PTH (Probe The Harness), a set of checks that makes the harness visible. Our case is a comparison between SAN, a behaviour-free method, and truncated importance sampling (TIS) on verl and in a single-GPU trainer, in which SAN first finished ahead in both stacks. Four details of the harness changed this comparison: the PPO ratio was taken against the learner's own recomputed probabilities, the data seed did not reach the TIS arm, the replay queue reused its first batch for 33 updates, and two loss normalisers differed from their description. In each case the logged quantity looked consistent with a working setup, while the quantity that defines the comparison went unchecked. With the harness checked, TIS matches SAN on verl, and in the trainer TIS learns steadily while SAN keeps a margin. We contribute the signature of each detail and its effect on the comparison, reference results for TIS and uncorrected GRPO under sampler lag, and the PTH checklist.
Read the full abstract ▾
Computation and Language
Misinformation Without Triggers: From Factual Answers to Downstream Decisions
Lin Tian and Marian-Andrei Rizoiu
Language models learn from web documents, some of them false, and false content can reach a model's answer to a factual question and the summaries and decisions that use it. Most data-poisoning studies add a trigger to the training data and activate it in the
prompt. False documents can also change factual responses without any trigger, but we do not know whether the direct answer predicts the decision. In this work, we follow false content past the answer and find an audit gap between what a direct probe reports and what the model then does, comparing false training with matched truthful controls in a controlled decision task, Guess the Capital, where a fixed decoder turns factual answers into a scored card choice, and on a misleading claim from Facebook posts about the 2019–20 Australian bushfires. Across eight models at dose 1,000, direct injected-choice rates reach 95.8–100%, while injected game choices increase by 1.7–14.4 percentage points over matched truthful training. The gap runs the other way too. Facts that pass the direct probe still push decisions toward the injected answer, and game accuracy drops further than those choices explain. Truthful correction brings the fact back but not the decisions built on it. We then look into the real-world bushfire case, models trained on the false posts say that people were arrested for arson even when they lose the inflated count, and in a count-by-wording factorial the misleading arrest wording produces arrest assertions even when the training count stays at 24. In simpler terms, a correct factual answer does not guarantee a correct decision, and losing the injected number does not remove the misleading story.
Read the full abstract ▾
Computation and Language
Evaluating VQA in Vision Language Models using Cooperative Principles
Monika Shah, Sudarshan Balaji, Somdeb Sarkhel et al. (+2 more)
We evaluate the performance of
Vision Language Models in Visual Question Answering (VQA) when questions violate Grice's maxims. To do this, we use VLMs to generate question modifiers that add non-essential, ambiguous or false information and show that in the presence of such violations, the VLMs that we evaluate (ChatGPT, Claude, Gemini and Llava) show diminished performance. Further, we empirically show the difference between how humans reason pragmatically compared to VLMs, and the difference in VLM
reasoning when it resolves violations that are human-induced compared to those that are AI-generated. Finally, we show that human cognitive effort (measured through time-on-task in an experiment) is lower for resolving VLM-induced violations, but VLMs themselves perform less accurately in such cases.
Read the full abstract ▾
All authors Monika Shah, Sudarshan Balaji, Somdeb Sarkhel, Sanorita Dey, Deepak Venugopal
Computation and Language
Evaluating LLM-as-a-Judge Beyond Score Alignment: A Psychometric Analysis of Residual Judging Difficulty
Longwei Cong, Sonja Hahn, Sebastian Gombert et al. (+4 more)
Large
language models (LLMs) are widely used as automatic judges, with validity typically assessed via
alignment with human scores. However, aggregate agreement fails to reveal whether humans and LLMs find the same evaluation cases difficult. In this paper, we study this problem in summarization evaluation from a psychometric perspective. We fit Many-Facet Rasch Models separately to human and LLM ratings to decompose scores into latent summary quality, rater severity, dimension severity, and rating-scale thresholds. Building on this decomposition, we define residual hardness as a model-adjusted measure of judging difficulty and compare whether human and LLM judges share the same hardness structure. Across 17 open-weight LLM judges on SummEval, we find that moderate
alignment in latent summary quality does not imply
alignment in residual hardness. Human and LLM judges differ in which summary–dimension units remain difficult, and this mismatch is strongly dimension-dependent. Consistency shows a pronounced LLM-hard shift, whereas coherence shows a human-hard shift. We further show that human-easy but LLM-hard cases are partially predictable from observable source–summary properties. These findings suggest that aggregate human
alignment reflects only part of LLM-as-a-judge reliability, while psychometric residual diagnostics support more informative judge evaluation and more targeted human–LLM collaboration.
Read the full abstract ▾
All authors Longwei Cong, Sonja Hahn, Sebastian Gombert, Leon Camus, Fabian Zehner, Hendrik Drachsler, Ulf Kroehne
Computation and Language
Query-aware routing for Cross-lingual performance gains in Encoders
Akshay Jain and Edward Kim
Multilingual encoders can exhibit reduced
retrieval effectiveness when queries and relevant documents differ in
language, despite strong same-
language performance. We investigate whether Finnish and Swedish cross-lingual
retrieval can improve while preserving an encoder's existing same-
language performance and document index. We combine a query-only low-rank adapter, trained against frozen document
embeddings, with deterministic routing based on query and index
languages. Cross-
language queries use the adapter, while same-
language queries use the original encoder. SampoTron, our fine-tuned low-rank (LoRA) adapter alongwith the Nemotron-3-Embed-1B model, improves average
retrieval quality across six English, Finnish, and Swedish directions from 0.241 to 0.291 in normalized discounted cumulative gain (nDCG) at rank ten, a 20.9% relative gain on a sampled financial
benchmark. All six cross-lingual directions improve, and routing preserves the original same-
language performance, including two full-corpus Finnish evaluations. The approach enables selective cross-
language specialization with reusable document
embedding vectors.
Read the full abstract ▾
Computation and Language
ConvoDrift: A Multi-Turn Conversational Dataset for Modeling Stylistic Tone Evolution
Vihindi Kotalawala, Pamoda Dilranga, Gayani Thoradeniya et al. (+1 more)
The
evolution of linguistic style in conversations is an underexplored issue in NLP. Most style-control
datasets focus on sentences or assume a static style throughout, missing the dynamic shifts that occur as user preferences change during interactions. We introduce ConvoDrift, a
dataset designed to model progressive stylistic conversational tone drift under fixed semantic intent. It is built on 15,727 shared multi-turn conversational structures for adaptation and persona-conditioned
alignment methods. It consists of six
prompt-response pairs per conversation, each with the annotation of style drift and style direction labels. These pairs cover a range of communication genres. We further derive a complementary pairwise
dataset by pairing semantically equivalent but stylistically distinct responses and annotating persona-conditioned preferences using five distinct style communication personas, enabling the controlled study of personalisation and pluralistic
alignment in
language tone. In addition to
dataset construction, we conduct a comprehensive evaluation involving human validation, LLM-as-judge assessment, and automatic lexical and semantic evaluations. Across seven Likert criteria annotated by three human annotators, the average Krippendorff's alpha is 0.88, and our lexical and semantic analyses show that drift events induce lexical changes while preserving semantic similarity.
Read the full abstract ▾
All authors Vihindi Kotalawala, Pamoda Dilranga, Gayani Thoradeniya, Prasan Yapa
Computation and Language
Adaptive Mutual Distillation for Balanced Multi-Task Post-Training of Large Language Models
Baohang Li, Xiaocheng Feng, Yichong Huang et al. (+6 more)
Multi-task post-training of large
language models (LLMs) aims to improve performance across tasks with unequal amounts of training data. Existing methods focus primarily on balancing task contributions during single-model training. Different task-balancing strategies can produce models with complementary strengths, creating opportunities for mutual
distillation. However, the usefulness of cross-model supervision can vary across tasks, transfer directions, and stages of training. We propose Adaptive Mutual
Distillation (AMD), a collaborative post-training framework that jointly trains two models with different task-balancing strategies. AMD evaluates candidate adjustments to
distillation weights through short training probes shared across tasks, then uses task-wise validation scores to select an adjustment for each task and transfer direction. Across six
benchmarks and three LLM backbones, both AMD models achieve higher average
benchmark scores than supervised fine-tuning (SFT) baselines trained with the same sampling strategies. They also outperform the task-balancing methods evaluated in our experiments. Merging the two trained models can further improve their average
benchmark score while yielding a single model for inference. The merged models outperform multi-task SFT by an average of 2.91 points across the three backbones.
Read the full abstract ▾
All authors Baohang Li, Xiaocheng Feng, Yichong Huang, Chengpeng Fu, Wenshuai Huo, Zekun Zhou, Zekun Yuan, Tingjia Zhang, Bing Qin
Computation and Language
How Robust Is Multimodal Claim Verification to LLM Rewriting?
Yun-Ang Wu, Xanh Ho, Andre Greiner-Petter et al. (+4 more)
LLMs are known to introduce stylistic changes into generated text, yet how these stylistic shifts affect model decisions on scientific tasks remains underexplored. In this paper, we focus on
multimodal claim verification, where the goal is to determine whether a textual claim is grounded in a given piece of evidence. We apply two rewriting strategies: natural rewriting, which simulates how researchers routinely use LLMs to polish academic text, and controlled injection, which inserts a single LLM-associated word to isolate the effect of vocabulary choice. We evaluate 11 open-weight models spanning five VLM families and ranging from 2B to 38B parameters. We find that models are robust to these modifications: most show no significant drop in accuracy, and compared to prior work on review-score manipulation, verification appears far more stable. However, consistent probability shifts do occur. Hedging-oriented conditions produce significant shifts across nearly all models, while boosting conditions show a weaker effect and general polishing conditions (e.g., grammar correction, fluency improvement) have little effect.
Read the full abstract ▾
All authors Yun-Ang Wu, Xanh Ho, Andre Greiner-Petter, Sunisth Kumar, Tian Cheng Xia, Florian Boudin, Akiko Aizawa
Computation and Language · Cross-listed from cs.LG
To Explore The Strange New World Beyond Data Distribution: System Behavior, Causality Tax, and Non-causal Base Model
Xianzhi Zeng, Jiangneng Li and Gao Cong
We show that the causality of
language models (LMs) may not be necessary nor optimal. This is the case when system behavior (denoted as $S$) is incorporated as a first-principle
Bayesian feature. Here, $S$ refers to extra dominant factors beyond the data space, and they involve coupled effects. Despite being the de facto foundation of modern architecture, recent studies indicate persistent mismatches and contradictions with causality. These issues largely stem from system behavior rather than the data distribution. We therefore propose the SBD framework, which incorporates $S$ as an irreducible component of the evidence lower bound (ELBO). SBD theoretically reveals a counter-intuitive Causality Tax phenomenon, where causality emerges as a suboptimal approximation with an additional structural error, due to the obliviousness to $S$. To address the challenge of latent variable analysis, we validate the SBD-predicted impact of $S$ via implicit measurements, theoretical-bound-guided controls, and Neural Tangent Kernel (NTK) evaluations. In particular, we construct Green Shell (GSH) to show the possibility of reducing Causality Tax. GSH is a non-causal variational family, and it replaces the sequential dependency chain of $S$ components with a divide-and-conquer partition. NTK spectra in the lazy-training regime confirm that GSH always achieves significantly tighter error bounds than causality, with $7dB+$ improvement in signal-to-noise ratio. In the relatively later stage of lazy-training, GSH further leads to superior generalization (up to $20\%$ richer multi-scale fitting capabilities). Taken together, SBD establishes system behavior as a complementary theoretical abstraction besides causality and distribution fitting, opening new research avenues such as designing and optimizing LM base models.
Read the full abstract ▾
Computation and Language
Clinical Concept Centers in LLMs
Aishik Nagar, Abhishek Vaidyanathan, Arun-Kumar Kaliya-Perumal et al. (+2 more)
Large
language models are increasingly used in clinical settings. However, research into the reliability and performance of these models has focused almost entirely on the
language substrate, scoring what the model says. Mechanistic interpretability has found that the latent space carries a higher fidelity of representation than the text: internal representations not only encode substantially more than the output verbalizes, but the stated
reasoning also systematically omits features that causally drive the answer. An evaluation of model behavior in terms of mechanistic interpretability has not been explored in clinical decision support. In this work, we extend behavioral evaluation into the latent space and ask whether clinical concepts exist as locatable, causally used representations inside open-weight LLMs. We find dedicated clinical concept centers in the latent space of all eleven open models we test. These concept centers are interpretable, firing only on their aligned clinical narratives, and meaningfully and causally drive model behavior in both constrained and open-ended settings. They are not just analytical representations, but
circuits that can be utilized in clinical practice, and we explore their use from the perspective of both evaluation and performance. From the evaluation standpoint, models stay internally coherent and keep using the relevant concept centers even under
adversarial role-based priming, while aligned priming improves downstream clinical performance. From a performance perspective, we simulate realistic deployment settings and find that steering models along these centers leads to meaningful downstream improvements. Finally, we conduct a blinded clinician validation and find the activation and usage of these concept centers predicts clinicians preferences.
Read the full abstract ▾
All authors Aishik Nagar, Abhishek Vaidyanathan, Arun-Kumar Kaliya-Perumal, Elijah Tzen Hsuen Boey, Stefan Winkler
Computation and Language · Cross-listed from cs.AI
FSPO: Policy-Consistent Risk and Pareto-Feasible Control for Budgeted LLM RL Post-Training
Miaobo Hu, Shuhao Hu, Xiaobo Guo et al. (+4 more)
Adaptive LLM reinforcement-learning post-training changes multiple training actuators online, including rollout temperature, group size, clipping, KL regularization, verifier allocation, and update budget. Three coupled issues remain unresolved. A future-risk model trained from behavior trajectories need not estimate the risk induced by the controller that will be deployed; a score calibrated on logged state-action pairs can become miscalibrated after selective action choice; and independent per-resource minimum costs do not in general certify a feasible multi-resource continuation. We introduce FSPO, a feedback-state controller for budgeted LLM RL post-training that addresses these issues jointly. FSPO learns a policy-consistent risk-to-go model whose Bellman target follows the same frozen controller used for future decisions, together with a long-horizon utility model. Decision-conditioned trajectory calibration (DCTC) calibrates risk on cross-fitted trajectories generated by actions selected by provisional controllers. A Pareto resource continuation certificate (PRCC) admits an action only when a non-dominated cumulative reservation remains feasible over the residual horizon. Under a matched GRPO resource envelope, FSPO reaches 66.11% held-out and 59.43% OOD accuracy, compared with 64.47% and 57.03% for PB2, the strongest evaluated adaptive baseline. Three paired training seeds give gains of +2.42 and +3.19 percentage points over the contextual bandit on held-out and OOD evaluation. Under high behavior-deployment mismatch, policy-consistent risk lowers selected-decision ECE from 0.108 to 0.053; DCTC lowers it from 0.039 to 0.022 at matched acceptance; PRCC removes false-feasible admissions on an 18-action catalog ($0.197\rightarrow0.000$); and enabling all three components reduces trajectory failure from 0.181 to 0.083 in a factorial ablation.
Read the full abstract ▾
All authors Miaobo Hu, Shuhao Hu, Xiaobo Guo, Xin Wang, Bokun Wang, Daren Zha, Jun Xiao
Computation and Language
Text-Centric Post-Training for Omni-Modal Reasoning
Ziyang Cheng, Yuhao Wang, Hongcheng Liu et al. (+5 more)
Improving joint audio-visual
reasoning in Omni Large
Language Models typically incurs substantial data construction and training costs. Our diagnostics reveal multi-hop
reasoning difficulties despite correct answers to all corresponding single-hop questions and suggest partial decoupling in the local optimization of perception and
reasoning objectives. This motivates post-training with different emphases on these capabilities. Text-only
reasoning training yields gains across data sources, model scales, and families. With the best-performing text-only configuration, supervised fine-tuning followed by reinforcement learning (RL) raises Qwen2.5-Omni-7B's
geometric mean of nine
reasoning scores by 25.83% over the base model, outperforming the complete native audio-visual route with 56.6% fewer GPU-hours. Training on data synthesized entirely by a text-only LLM raises this
geometric mean by 21.01% without audio-visual data in construction or training. However, text-only training degrades perception. We therefore propose a text-centric post-training paradigm: text-only training provides the main
reasoning optimization, and reduced-data native audio-visual RL then refines perception. Refinement uses about 90% fewer input
tokens than full-data audio-visual RL, restores perception above the base level, and retains 93.5% of the best-performing text-only pipeline's
reasoning gain.
Read the full abstract ▾
All authors Ziyang Cheng, Yuhao Wang, Hongcheng Liu, Qimin Wu, Jingru Fan, Chen Qian, Yanfeng Wang, Yu Wang
Computation and Language · Cross-listed from cs.CR
RMCW: A Deletion-Robust Watermark Based on Reed–Muller Codes for Language Models
Yi Wang, Baicheng Chen, Yu Wang et al. (+3 more)
Large
Language Model (LLM) watermarking provides a lightweight mechanism for identifying text generated by a specific model, but its robustness remains fragile under post-processing attacks. Deletion attacks are particularly challenging because they shift
token positions and break the
alignment between observed
tokens and their original watermark positions. We propose Reed–Muller Code Watermarking (RMCW), an LLM watermarking method based on Reed–Muller codes. In contrast to global codeword recovery, RMCW searches for surviving local algebraic structure, leveraging the Reed–Solomon consistency induced by affine-line restrictions of Reed–Muller codewords. During generation, RMCW injects a Reed–Muller structure into the sequence via a secret-keyed vocabulary partition. During detection, it maps the given text to keyed vocabulary bins and tests local subsequences for low-degree Reed–Solomon consistency using Berlekamp–Welch tests. Experiments on C4 and ELI5
datasets with OPT-1.3B and Llama-3.1-8B-Instruct show that RMCW preserves strong clean-text detectability and outperforms or matches the baseline methods under several deletion and rewriting attacks. Our code is available at https://github.com/BaichengDanny/RMCW.
Read the full abstract ▾
All authors Yi Wang, Baicheng Chen, Yu Wang, Jian Zhao, Yilei Chen, Tianxing He
Computation and Language · Cross-listed from cs.AI
ROUTEAUDIT: Interaction-Aware Identification for Budgeted Multi-Verifier Routing
Miaobo Hu, Shuhao Hu, Xiaobo Guo et al. (+5 more)
Adaptive multi-verifier systems are commonly compared through endpoint quality-cost gaps, even when the verifier catalog, availability, accounting, information filtration, or scorer changes with the policy. We formulate verifier routing as a contract-conditioned identification problem. The contract records request support, verifier catalog, realized availability, resource accounting, online filtration, and post-trace scoring; a matched route contrast changes only the policy coordinate. ROUTEAUDIT adds three measurable objects to this contract. A contract
lattice averages coordinate increments over every admissible bridge order and reports the resulting attribution together with its path sensitivity. A policy-independent response tape identifies paired sequential contrasts when adaptive policies reveal different observations. For incomplete matching, request-level bounds use whichever potential outcome remains observed and give a sharp finite-population interval. The protocol commits paid observations and ledger events before the oracle join and returns an attribution certificate for each comparison. On two held-out raw-tail caches, matched static SF+SA equals the cascade, assigning the apparent gains of 0.1797 and 0.1250 over full static to the verifier-set edge. On 1,319 held-out task requests, the learned and RLVR studies report quality 0.9522 and 0.9553 versus 0.9484 for matched static; the RLVR-static paired difference is +0.0068 with a request-paired interval $[0.0015,0.0122]$ and a training-seed-by-request hierarchical interval $[0.0006,0.0131]$. Controlled attribution recovery yields route mean absolute error 0.0011 and endpoint reconstruction error 0.0004. Factorial, bridge-order, and
stochastic-provider studies evaluate the certificate interface; RLVR supplies a learned-policy stress test under the same identification contract.
Read the full abstract ▾
All authors Miaobo Hu, Shuhao Hu, Xiaobo Guo, Xin Wang, Bokun Wang, Tianshu Fu, Daren Zha, Jun Xiao