Computation and Language · Cross-listed from cs.LG
How to Loop MoE: Flatten the Experts, Untie the Attention
Shouren Wang, Chuang Ma, Mohsen Hariri et al. (+6 more)
Looped
Transformers reuse one block of layers several times: by spending extra computation they push a model of fixed size further, and so use its parameters more fully; while sparse mixture-of-experts (MoE) models activate only a few of many experts for each
token. Looped MoE bridges these two design philosophies and gives MoE models new potential for better expert usage, but it raises a question: how to loop a MoE? We answer it with Foil. With the expert parameters and the expert compute per
token held fixed, Foil (1) flattens the experts, halving the expert layers, doubling the experts per layer and doubling the passes, so that every routing decision chooses from a larger pool, and (2) unties the
attention, giving each pass its own
attention parameters while the experts and routers stay shared. Experiments show that Foil clearly outperforms the unflattened looped baseline: at 20B
tokens every Foil model has lower pretraining loss than the baseline; at 100B
tokens the loss improves monotonically with the degree of flattening, the most flattened Foil ending 0.012 nat below the baseline at equal parameters and compute, with downstream accuracy on par or better; untying the
attention also yields more balanced and more confident routing at equal shape. Our ablations analyse why Foil works and turn the findings into design guidance for looped MoE: the returns of looping and of widening the expert layers amplify each other, routing confidence tracks healthy expert use better than load balance, and a sparse looped MoE should therefore use more experts per layer and more passes. Code and configurations are available at https://github.com/SR-A-W/how-to-loop-moe.
Read the full abstract ▾
All authors Shouren Wang, Chuang Ma, Mohsen Hariri, Debargha Ganguly, Wang Yang, Xiaoqing Tong, Qianying Liu, Xiaotian Han, Vipin Chaudhary
Computation and Language
Towards Communication-Efficient Social Intelligence in Language Agents
Linxiao Gong, Yijie Xu, Tianfu Wang et al. (+7 more)
Socially intelligent
language agents must negotiate, coordinate, and resolve conflicting preferences while respecting the time and
attention of both participants. Balancing these demands is challenging because
agents must convey enough to address a partner's constraints and advance their goals without adding words that do not help the interaction. In this paper, we propose Teacher-Assisted Communication Training (TACT) to improve social goal attainment while reducing communication cost, making interactions with
agents more productive and less demanding. We first characterize communication efficiency in terms of action strategy and expression, whose effects extend beyond the current utterance to the partner's response and subsequent exchanges. We design TACT to revise student-generated actions, test the revisions through partner responses, and distill useful feedback into the student. An expression specialist removes unnecessary detail while preserving the intended action, while a strategy specialist proposes alternatives that may better address the partner's constraints. To determine which revision helps, TACT samples a partner response for each candidate and selects a teacher reference by balancing local goal support against action-
token cost. That reference guides on-policy
distillation on the student's own generation prefixes, allowing the student to act independently at deployment. We evaluate TACT on SOTOPIA and AgentSense. On SOTOPIA, it achieves the highest Goal among the evaluated methods on All and Hard while using substantially fewer target
tokens than SFT+SDPO. On AgentSense, it improves goal success over the initial student while reducing target
tokens and interaction messages.
Read the full abstract ▾
All authors Linxiao Gong, Yijie Xu, Tianfu Wang, Yin Wu, Yili Wang, Xingbo Yao, Huizai Yao, Xilin Xia, Haowen Yang, Hui Xiong
Computation and Language
Improving Test-Time Scaling with Adaptive Looped Transformers
Yichen You, Tianyu Fu, Aosong Feng et al. (+4 more)
Looped
transformers have demonstrated promising parameter efficiency by reusing layers for latent computation. Prior studies compare looped and non-looped models at matched parameters or per-
token FLOPs. However, to the best of our knowledge, whether looping improves test-time scaling as outputs grow longer remains underexplored. Through post-training looped
transformers, we study the accuracy-compute slope, measured as the accuracy gain per doubling of test-time decoding FLOPs. We find that existing looped
transformers often yield steeper slopes than their non-looped baseline, yet underperform it at matched compute. While fixed-depth looping spends extra iterations on every
token, our analysis shows that many
tokens do not benefit from extra iterations. We therefore propose TaH2, which enables the model to focus extra iterations on the
tokens that benefit from looping. It jointly post-trains the backbone and an iteration decider through lookahead depth supervision, which uses online labels indicating whether further iteration improves the prediction. TaH2 improves both the efficiency and attainable accuracy of test-time scaling. On challenging AIME
benchmarks, TaH2 improves the accuracy-compute slope by 53% (2.74 vs. 1.79) over the non-looped baseline, exceeding the baseline's peak accuracy by about 3.4 points at matched test-time compute. As the maximum iteration depth increases, existing looped models largely plateau, while TaH2's gain over the non-looped baseline continues to grow from +2.8 points at depth 2 to +3.9 points at depth 8. Our code is available at https://github.com/thu-nics/TaH.
Read the full abstract ▾
All authors Yichen You, Tianyu Fu, Aosong Feng, Xingtai Lv, Xuefei Ning, Ning Ding, Yu Wang
Computation and Language · Cross-listed from cs.AI
Shockingly Simple Self-retrospection Improves Agentic Models Without RL
Jonathan Light, Christopher Zhang Cui, Jeonghye Kim et al. (+7 more)
People learn not only by repeating successful actions, but also by recounting and explaining their experiences, revising their understanding to guide future behavior. Can a
language-model
agent improve its future actions by training only on explanations of its own experience? We investigate this question by studying Retrospection-Only Fine-Tuning (ROFT), a minimal online procedure designed to isolate the effect of explanation-only training on subsequent behavior. The
agent attempts a task, observes available feedback, generates a retrospective explanation, and is fine-tuned with a next-
token prediction loss on the explanation
tokens alone. The procedure uses neither an external teacher nor a reward-based policy update. In software-engineering experiments with Qwen3.5-4B, ROFT is trained on problems with mixed successful and unsuccessful base-model attempts. On held-out SWE-bench Verified and Pro, it reaches 49.2% and 26.8% solve rates after 20 updates without using a verifier, compared with GRPO's 48.0% and 25.3% after 40 updates in the evaluated runs, and makes faster early progress in training time and sampled attempts. It also learns to solve individual tasks on which all 64 sampled base-model attempts failed, showing that learning can begin without any initially successful trajectories. Behavioral analyses find that ROFT indirectly assigns credit to actions, encouraging good actions and discouraging incorrect ones. Moreover, prompting retrospections to emphasize more direct solutions yields shorter subsequent attempts even without an explicit length penalty. Together, these findings show that learning to explain can also improve learning to do, establishing self-generated retrospections as useful training targets and motivating further study of explanation-to-action transfer.
Read the full abstract ▾
All authors Jonathan Light, Christopher Zhang Cui, Jeonghye Kim, Roger Creus Castanyer, Emiliano Penaloza, Zhengyan Shi, Alessandro Sordoni, Marc-Alexandre Côté, Xingdi Yuan, Minseon Kim
Computation and Language
Harness Learning Enables Generalizable Test-Time Adaptation
Alvin Zhang, Xuecheng Liu, Zixuan Wang et al. (+6 more)
A
language-model
agent is jointly defined by its model and its harness, the executable program that organizes model calls, tool use, and information flow. Because different tasks call for different ways of organizing these operations, the harness needs to be adapted using feedback from the task at hand. We introduce harness learning, which trains a proposer model to revise a solver's harness using execution feedback. We formulate this process as meta-learning over executable programs, with harness revisions playing the role of weight updates in gradient-based adaptation. We train the proposer with reinforcement learning, using the task performance of revised harnesses as the reward. At test time, the proposer uses feedback from successive executions on a new task to refine the harness, without performing any parameter-space update. Experiments on
reasoning and multi-hop question answering show that harness learning improves revision quality and that the ability to adapt at test time transfers to unseen tasks. Policies trained on individual revisions can continue improving harnesses over multiple rounds, while the benefits of training on revision sequences vary across settings. These findings suggest a path towards continually learning
agents that turn accumulated experience into generalizable improvements.
Read the full abstract ▾
All authors Alvin Zhang, Xuecheng Liu, Zixuan Wang, Fahim Tajwar, Daman Arora, Ruslan Salakhutdinov, Daniel Khashabi, Yuda Song, Andrea Zanette
Computation and Language · Cross-listed from cs.AI
Reinforcing Agentic Creativity in Scientific Ideation with Night Science
Priyanka Kargupta, Silviu Cucerzan, Shweti Mahajan et al. (+4 more)
Large
language models (LLMs) excel at structured, verifiable tasks, but their low-
entropy bias can produce homogeneous and predictable outputs, limiting their utility for open-ended scientific ideation. Effective discovery, however, spans a broader creative spectrum: from structured day science to loosely structured, serendipitous night science that reaches ideas beyond those typically considered. We introduce AI Night-Scientist, an agentic framework that uses reinforcement learning to teach models when and how to depart from predictable
reasoning. Grounded in cognitive science, we model creativity along three axes: action (what to do and how creatively), process (when to explore versus exploit), and outcome (the novelty and usefulness of the resulting idea). We use these axes to train models with GRPO, exposing them to varying degrees and forms of creativity throughout training. This produces substantially more diverse scientific proposals, expanding the range of research directions by 27.8% and contribution types by 14.9% over the base model. It also improves predicted citation impact by up to 32.0 percentage points and originality by 66.2 points. These gains cannot be reproduced by simply increasing decoding temperature; instead, we find that semantic guidance specifying what kind of creativity to pursue is critical. Overall, our results suggest that creativity is a learnable, multi-level ability that can be shaped to help researchers reach ideas beyond those typically explored by LLMs.
Read the full abstract ▾
All authors Priyanka Kargupta, Silviu Cucerzan, Shweti Mahajan, Allen Herring, Jiawei Han, Ryen W. White, Sujay Kumar Jauhar
Computation and Language · Cross-listed from cs.LG
Rethinking Personalized Generation: Test-Time Alignment via Factorized Ranking Models
Qiyao Ma, Junshan Zhang and Zhe Zhao
Aligning large
language models (LLMs) to diverse user preferences is fundamentally hindered by standard
alignment paradigms that optimize for monolithic users. In this work, empirical studies are first used to reveal the existence of a massive, untapped performance headroom for personalized generation through test-time
alignment. We demonstrate that personalized generation is uniquely suited for test-time scaling methods like Best-of-N (BoN) because it can be viewed primarily as a candidate matching problem rather than a generator capability bottleneck. While reward models could in principle exploit this headroom, they are poorly calibrated for personalization, and their billion-parameter scale makes scoring large candidate pools prohibitively expensive. To overcome this limitation, we propose a parameter-efficient framework utilizing million-parameter scale multi-layer perceptron (MLP) ranking models. Our personalized ranking model directly reuses the internal
embeddings of the base generator with minimal overhead. By scaling train-time data to provide fine-grained personalized preferences, this million-parameter ranking model accurately scores large candidate pools and can seamlessly guide generation to reduce the cost of materializing N candidates. Extensive experiments on nine
datasets spanning three personalized generation settings show that our personalized ranking model effectively exploits the discovered headroom, outperforming billion-parameter generalist reward models on every
dataset, with under 0.4% of their parameters and four orders of magnitude lower scoring latency.
Read the full abstract ▾
Computation and Language
QuanReview: Offline, Auditable Reconciliation of Human and LLM Span Annotations
Matteo Musacchio, Juan Cruz Giner Pulero, Isabel Castañeda et al. (+4 more)
Structured span annotations, such as quantities with their units, uncertainty modifiers, and event classes, are expensive to create and hard to keep trustworthy once
language models enter the loop. We present QuanReview, an open-source system for auditing and correcting such annotation layers. QuanReview aligns two annotation streams over the same documents at character level, resolves unambiguous cases by an explicit and logged policy, and routes candidate conflicts to a browser-based adjudication interface where reviewers accept either side, build field-level hybrids, or flag items for re-annotation. A campaign manager assigns documents to multiple annotators with configurable redundancy, computes agreement at document and span level, auto-merges unanimous documents, and exports the corrected layer in the original file format, so that it can replace the original annotation files directly. Applied to a 4,457-record humanitarian
benchmark and an LLM extraction stream, the system fully auto-merged 8% of documents, applied automatic policy decisions to a further 1,513 records, and concentrated human
attention on 3,131 candidate conflicts, a mean of 5.4 per reviewed document.
Read the full abstract ▾
All authors Matteo Musacchio, Juan Cruz Giner Pulero, Isabel Castañeda, Naomi Couriel, Yelena Mejova, Mariano G. Beiró, Kyriaki Kalimeri
Computation and Language
Tracing the Evolution of Oracle Bone Characters Across Three Millennia
Tianhao Fu, Xinxin Xu, Spike Wang et al. (+3 more)
Of the approximately 4,500 Oracle Bone Inscription (OBI) characters discovered from the Shang dynasty, only about 1,600 have been deciphered. Many computational approaches compare OBI with glyphs from one historical period at a time. However, during the
evolution of Chinese characters, significant structural or semantic changes often occur in uncertain dynasties. A single-period reference may be insufficient when relevant forms change substantially between observed eras. Therefore, we propose the
Manifold-based Script
Evolution Framework (MSEF), a framework that models the
evolution series (OBI, Bronze, Seal, Clerical, Regular) of Chinese characters as the continual
evolution of a
manifold space. MSEF represents each character as an era-specific
manifold point and learns continuous inter-era transition rules via Neural Ordinary Differential Equations. Both
manifold space and transition dynamics can be trained end-to-end through character
evolution pairs across any two eras.
Read the full abstract ▾
All authors Tianhao Fu, Xinxin Xu, Spike Wang, Cunyi Kang, Jian Cao, Xixin Cao
Computation and Language
MS-GLA: Multi-Scale Gated Linear Attention for Addressing Representational Bottlenecks via Multi-Temporal Resolution
Prasoon Dev, Anirudh Sankar and Vasudeva Varma
Gated Linear
Attention (GLA)
Transformers advance linear recurrent models through data-dependent gating, but face a core limitation: the fixed-capacity
memory matrices across all heads operate at a single temporal resolution, where each
token is processed individually, forcing them to simultaneously encode local syntactic patterns and long-range semantic structure, creating a representational bottleneck that gating alone is insufficient to resolve. We introduce Multi-Scale Gated Linear
Attention (MS-GLA), which addresses this by distributing
attention heads across multiple temporal resolutions. Coarser resolutions pool longer
token spans naturally specializing toward long-range dependencies, while finer head groups retain sensitivity to local syntactic structure. A learnable, input-dependent fusion layer dynamically recombines head group outputs at each timestep, expanding effective
memory capacity without increasing per-head state size. This multi-resolution decomposition draws on principles from Multi-Scale State-Space Models (MS-SSM), adapting them to the gated linear
attention setting. We evaluate MS-GLA on
language modeling, recall-intensive tasks, and long-context generalization. Across all settings, MS-GLA consistently achieves higher accuracy and lower perplexity than GLA at matched parameter counts, with up to 18.9% improvement on recall-intensive tasks and 9.5% lower average perplexity on
language modeling
benchmarks, validating multi-temporal resolution decomposition as a principled and effective extension of Gated Linear
Attention.
Read the full abstract ▾
Computation and Language
Late Attention Layers Alone Can Copy Entity Tokens, but Not Without Attending to Their Context
Muyu He, Yuchen Liu, Ran Tao et al. (+1 more)
Large
language models (LLMs) reliably perform entity copying, in which a model copies
tokens referring to an entity, termed entity
tokens, from the
prompt into its output to answer a question. Although entity copying is straightforward for most LLMs, existing research does not provide a systematic account of which layers specialize in this fundamental task or how other
tokens in the same sequence, termed context
tokens, influence the model's ability to copy the entity
tokens. To address these questions, we conduct experiments on Qwen3-8B using two novel methods: genie-in-a-bottle, which controls exactly which layers can participate in an entity-copying task, and
attention lobotomy, which cuts off specific
tokens'
attention to entity
tokens without affecting the remaining
attention distribution. We find that two distinct groups of layers in the second half of the model are both necessary and sufficient for entity copying. Moreover, in addition to the decoding position's
attention to entity
tokens, context
tokens'
attention to entity
tokens also proves necessary for copying the exact
tokens, even though context
tokens do not store entity information themselves unless they satisfy particular semantic properties. Our findings establish the critical role of late layers in entity copying under the guidance of context
tokens, calling for future work on how models propagate and consume entity information.
Read the full abstract ▾
All authors Muyu He, Yuchen Liu, Ran Tao, Li Zhang
Computation and Language
Rubric Rewards from Item Response Theory
Milad Yazdani, Yaser Souri, Xiren Zhou et al. (+4 more)
Many
language tasks have no single answer that can be checked automatically. Rubrics provide criteria for judging responses to these tasks. For reinforcement learning, the resulting verdicts must be combined into a scalar reward. A common approach sums the points assigned to satisfied criteria. Distinct verdict patterns can thus receive the same reward, and the fixed points encode how much each criterion should count, not how strongly its verdict distinguishes the current rollouts. Beyond this aggregation problem, judging the full rubric needs more judge requests as the criterion count grows. To address these limitations, Rubric Response Theory (RRT) measures quality and selects criteria when rubric criteria are monotone indicators of a shared target. Rather than adding assigned points, RRT uses a two parameter item response model that treats the verdict pattern as evidence about scalar quality specific to the rubric. Under this model, its likelihood score maximizes the local signal-to-noise ratio for quality. Its Response Parameter Network (RPN) reads the
prompt and criterion text to predict criterion difficulty and discrimination. As the policy distribution changes during training, RRT uses online expectation maximization to update the RPN from current rollout verdicts. With Qwen3.5-4B as the policy, RRT's macro criterion score across Medical, Science, Rubrics as Rewards Science, and RubricBench is 1.7 points above that of group relative policy optimization (GRPO). On hard and very hard criteria in Medical and Science, RRT gains 2.8 to 5.6 points over GRPO. At half the criterion budget, adaptive Fisher selection with a frozen RPN keeps the macro criterion score across four
datasets within 0.1 points of GRPO with full judging. These results show RRT can reduce judge requests while remaining competitive with GRPO.
Read the full abstract ▾
All authors Milad Yazdani, Yaser Souri, Xiren Zhou, Pranit Chawla, Dena Shahriari, Subhojit Som, Xia Song
Computation and Language · Cross-listed from cs.SD
CoSE-E: A Benchmark for Code-switched Speech Evaluation in Enterprise Settings
Shama Gupta, Hoang H Nguyen, Chelsea Huang et al. (+2 more)
Code-switching (CS), a seamless alternation between
languages within a single utterance, remains a critical challenge in automatic
speech recognition (ASR). While prior works focus on conversational CS-ASR, enterprise settings demand evaluation of operational impact beyond edit-distance errors: how code-switching transcription errors propagate to downstream voice
agent task failures. In this work, we propose (1) a CS-ASR synthetic
benchmark and multidimensional evaluation framework tailored to enterprise domains, (2) systematic evaluation of frontier ASR systems across 5
language pairs, (3) diagnostic analysis of the additional transcription errors that code-switching introduces across
language pairs and models. We release COSE-E to support enterprise-focused CSASR evaluation for multilingual voice
agents in enterprise deployment.
Read the full abstract ▾
All authors Shama Gupta, Hoang H Nguyen, Chelsea Huang, Lindsay Devon Brin, Fanny Riols
Computation and Language
Which the Eye Fears: Writing with Read-Blindness Explains Massive Activations in Transformers
Swagatam Mukhopadhyay, Vishal Vivek Saley, Vraj Parikh et al. (+1 more)
Massive activation features (MAs) in
Transformers are extreme-value residual-stream features that persist across layers despite the model's ability to suppress them. Why do they survive? Our investigation using an operator-level mechanistic analysis of
attention and feed-forward (FFN) blocks reveals that these blocks systematically ignore MA coordinates while reading, but not while writing; creating a read-write asymmetry that blocks corrective feedback while allowing continued accumulation. We find that both
attention and feed-forward layers have this read-blindness, and contribute to the emergence and persistence of MAs. To validate prior work that hypothesized that FFN's amplification abilities is the primary reason for MAs (Sun et al., 2026), we analyze the model checkpoints during learning. Contrary to our expectation, read-blindness emerges before FFN amplification, suggesting that it acts upstream in the MA mechanism. We further contribute gradient analysis to link this behavior to surprising asymmetries in the loss landscape, concluding that the model actively maintains this read-blindness. Finally, we find that removing read-blocking at different locations induces compensatory shifts elsewhere, but MAs still persist.
Read the full abstract ▾
All authors Swagatam Mukhopadhyay, Vishal Vivek Saley, Vraj Parikh, Mausam
Computation and Language · Cross-listed from cs.LG
SANTA++: Sampling Attention through Representative Keys
Kyle Lee, Christian Z. Pratt, Ruoyu Fang et al. (+4 more)
Attention often concentrates on a small subset of
tokens in the context, but which subset matters changes from one query to the next. To exploit this changing structure, we introduce SANTA++, a training-free
stochastic attention method that uses representative keys for
memory-efficient selection without scanning the entire key-value (KV) cache. Cached keys are organized into teams, and the query scores one representative from each team to decide which teams to sample. We compute exact
attention scores within the sampled teams and reweight each team's contribution by the inverse of its inclusion probability. This importance sampling correction estimates
attention over the full cache, with a sampling budget that lets us trade
memory reads for accuracy. Remarkably, with 32 or 64 sampled teams, SANTA++ uses 16% to 22% of dense
attention's KV reads and retains 94% to 99% of the dense-
attention baseline's scores on LongBench v2 and HELMET's
retrieval-augmented generation subset, and 85% to 91% on RULER, with Qwen2.5-7B-Instruct at 32K context. With 31 sampled teams, our GPU implementation delivers a $1.69\times$
attention speedup over the dense FlashAttention baseline at 32K context. By reducing the number of cache entries read, SANTA++ in principle complements architectures with compressed KV representations, such as multi-head latent
attention. Our kernels are available at: https://github.com/OPUSLab/santapp-kernel-demo.git.
Read the full abstract ▾
All authors Kyle Lee, Christian Z. Pratt, Ruoyu Fang, Heekyung Lee, Avinash Lohitsa, Ryan Modafe, Kerem Y. Camsari
Computation and Language
Can LLMs Value the Right Evidence? Evidence-Value Misalignment in Dynamic Medical Diagnosis
Kehua Feng, Yunsheng Lu, Yitong Qiao et al. (+5 more)
A correct diagnosis reached from insufficient or misleading evidence can pose a clinical hazard, yet outcome-based accuracy may reward such lucky guesses. We call this mismatch between diagnostic decisions and the value of available evidence Evidence-Value Misalignment (EVM). To disentangle evidential grounding independently from diagnostic accuracy, we introduce MedEVM, a dynamic benchmarking environment comprising 1,050 cases across 24 disease systems. Observations arrive turn by turn, requiring models to continuously calibrate its decision by deciding whether to wait for more evidence or submit a diagnosis. Across 9 LLMs, four interesting patterns are observed. (1) Miscalibrated evidence tracking. Making a diagnosis often fails to calibrate evidence sufficiency, even in more capable models, and even worsens in
reasoning mode. (2) Misaligned diagnosis submission. Confidence in the correct diagnosis often fails to ensure timely submission despite sufficient evidence. (3) Evidence order matters. Reordering the same evidence changes diagnoses even when model confidence remains similar. (4) Misleading evidence remains influential. Added misleading evidence redirects diagnoses even after prior evidence becomes sufficient. We further verify that EVM predicts errors and that preventing premature submission improves accuracy. These findings motivate Evidence-Verified Diagnosis Harness (EVD-Harness). It decouples diagnosis generation from submission through an offline Contrastive Diagnostic Wiki and three online control stages, namely observation management, proposal and witness verification, and diagnosis submission control. Across five LLMs, EVD-Harness improves accuracy by 12.0–51.1 percentage points while mitigating EVM-related failures. Our results demonstrate that verifying evidential support before submission can make diagnostic decisions more reliable.
Read the full abstract ▾
All authors Kehua Feng, Yunsheng Lu, Yitong Qiao, Tiantian He, Lei Liu, Yue Shen, Jian Wang, Jinjie Gu
Computation and Language · Cross-listed from cs.LG
Twist, Don't Tilt: Trajectory-Exact Constrained Decoding for Masked Diffusion Models
Aditya Thimmaiah, Lara Marinov, Jayanth Srinivasa et al. (+3 more)
Constrained decoding for Masked
Diffusion Language Models (MDLMs) aims to ensure that generated outputs satisfy a specified structure or syntax constraint. MDLMs generate outputs by repeatedly unmasking masked positions present in their current state. Recent strategies for constrained decoding constrain the model's per-step mean-field posterior (which factorizes over masked positions) by enforcing the desired constraint with an automaton. The resulting chain-structured factor graph allows exact constrained sampling via dynamic programming. However, despite each draw being exact and constraint-satisfying, we prove that their composition, in general, tilts away from the model's relative probabilities over valid trajectories, thus leading to trajectory bias. We derive an exact expression for this bias as a product of ratios measuring how valid continuation mass changes when the denoiser is reconditioned, and characterize when the bias vanishes. We then correct the bias by introducing TWISTER, the first automaton-twisted Sequential Monte Carlo decoder for MDLMs, using the step-exact decoder as the proposal. We show that for regular
language constraints, the Feynman-Kac correction is exactly computable, with the twists obtained efficiently using quantities pre-computed for step-exact sampling. We prove that the resulting Feynman-Kac model targets the unbiased Doob h-transformed path law conditioned on constraint satisfaction.
Read the full abstract ▾
All authors Aditya Thimmaiah, Lara Marinov, Jayanth Srinivasa, Haris Vikalo, Junyi Jessy Li, Milos Gligoric
Computation and Language · Cross-listed from cs.CV
Simultaneous Translation between Sign Languages
Zetian Wu, Bowen Xie, Stefan Lee et al. (+1 more)
Deaf and hard-of-hearing (DHH) signers cannot converse in real time across different sign
languages today: existing sign-to-sign translation systems run offline, requiring the full source clip before any target sign is emitted. Live use cases - e.g. broadcast interpretation and two-way
video calls - instead demand simultaneous output, while the source signer is still signing. We present, to our knowledge, the first simultaneous sign-to-sign (S2S) translation system, with two wait-k regimes: test-time wait-k inference applied directly to a full-sentence model, and a trained wait-k model via
stochastic multi-path supervision. We further introduce ca-Stream-AL, a computation-aware latency metric for streaming output. Averaged across six S2S directions on both a smaller human-verified test set and a larger synthetic S2S corpus, our streaming system achieves a 38% ca-Stream-AL reduction while staying within a 9% DTW-PA-MPJPE increase and a 2.1 BLEU-4 drop compared to the full-sentence baseline. A word-order case study probes how the streaming model handles word order mismatch between different sign
languages - a consequence of simultaneous translation.
Read the full abstract ▾
All authors Zetian Wu, Bowen Xie, Stefan Lee, Liang Huang
Computation and Language · Cross-listed from cs.AI
TCSAlgBench: Benchmarking Automated Proving for Research-Level Theoretical Computer Science
Chutong Yang, Xiyuan Zhang, Yu Huang et al. (+7 more)
Large
language models perform strongly on competition mathematics, but their research-level
reasoning remains difficult to evaluate systematically. Theoretical computer science (TCS) connects
algorithm design to explicit guarantees and fundamental limits, providing a setting for evaluating whether models can justify computational improvements with arguments humans can inspect. We introduce TCSAlgBench, a
benchmark and reusable pipeline for natural-
language proof discovery, comprising 398
theorem-level challenges from 138 STOC and COLT 2026 papers. Expert-designed rules complete paper-specific context, preserve computational assumptions and quantitative guarantees, and withhold constructions when discovering an
algorithm is part of the task. For each task, prover systems receive
theorem statements and access to cited prior work. The pipeline supports fresh, versioned challenge batches from newly released papers. We evaluate ten model configurations from four families under direct inference and prover-verifier discussion, and compare four
agent workflows under matched model-call opportunities. All evaluations use the full
benchmark. In the model comparison, GPT-5.6 Sol max achieves the highest five-run verifier-accepted coverage at 23.6% after 10-round discussion. Discussion and repeated sampling improve coverage. In the separate
agent comparison using GPT-5.5 xhigh, decomposition improves coverage over discussion, and agentic planning achieves the highest five-run verifier-accepted coverage at 25.4%. TCSAlgBench provides a refreshable testbed for measuring progress in model
reasoning and studying how
agent workflows support research-level proof discovery.
Read the full abstract ▾
All authors Chutong Yang, Xiyuan Zhang, Yu Huang, Boran Han, Soonho Kong, Shuai Zhang, Vihang Prakash Patil, Zhen Han, Michael Bohlke-Schneider, Bernie Wang
Computation and Language · Cross-listed from cs.CR
SEABench: Benchmarking Endogenous Misalignment In Self-Evolving Agents
Saswat Das, Parvati Viswanathan, Daniel Donnelly et al. (+3 more)
Self-evolving LLM
agents have gained prominence for their ability to improve after deployment by modifying their harness, including their controller instructions,
memory management protocols, and reusable tools and skills, in response to user and environment feedback. However, locally useful updates may persist into later tasks where they produce unsafe behavior, even without direct
adversarial influence. To study this risk, we introduce SEABench, a
benchmark for studying endogenous misalignment arising from
agent self-
evolution, with 48 longitudinal task sequences that span multiple
evolution surfaces, task domains, and harm types in a rich personal-assistant environment. To account for the stochasticity inherent in agentic operations, we provide an adaptive trajectory discovery pipeline that probes for failures while preserving original task intent and supports causal attribution through paired non-evolving
agents and attribution scores. Our evaluation across multiple recent LLMs,
evolution surfaces, and harm types reveals that self-
evolution indeed increases task completion rates but often at the cost of safety failures that are absent for paired non-evolving baseline
agents. We also show that qualitatively different safety behaviors emerge across
evolution surfaces and harm types. Further, we show that this divergence in safety behavior is reflected in
agents' chain-of-thought
reasoning, which yields an effective monitoring strategy that can mitigate unsafe behavior with a low false positive rate.
Read the full abstract ▾
All authors Saswat Das, Parvati Viswanathan, Daniel Donnelly, Chang Huang, Sahar Abdelnabi, Ferdinando Fioretto
Computation and Language
Language Models Act on Hidden Valence
Cameron Berg and Caspar Kaiser
Language models describe some internal states as good and others as bad. But whether models have a stake in them is an open question. Simply asking models is unlikely to be informative. Any answer may be consistent with genuine introspection, superficial pattern-matching, or with fixed scripts learned in character training. We therefore study revealed preference. Rather than asking about a state, we use activation steering to attach a positively or negatively valenced activation pattern to one of two otherwise meaningless 'zones', switch steering off, and then observe which zone the model prefers. A model with a stake in that state should choose accordingly. Across seven open-weight models from five families, this is indeed what we find. First, steering changes the passages models write about each zone, and those words shift later choice. Second, this shift persists when all surface-level
tokens are held fixed and only the hidden KV cache differs. Third, the effect also remains when all text is generated without steering and valence is only injected during cache construction. Thus, the hidden state is sufficient to move choice in proportion to the steering dose. Fourth, the dependence of choice on hidden valence is nearly absent in a base model and emerges during DPO, consistent with a link between valence and goal-directed behaviour formed in training. Finally, given tools to self-steer, a model does not tend to induce a positive state, but it regularly removes an imposed negative state. It does so at a dose-dependent rate and significantly more often than it removes random-direction interventions. Overall we demonstrate that valence-related activation patterns leave hidden traces that predictably govern later choices, even when all visible
tokens are identical across conditions. Whether these traces are accompanied by any subjective experience relevant to model welfare remains unclear.
Read the full abstract ▾
Computation and Language
FactorEngram: Factorized N-gram Memory with Basis-Level Gating for Language Models
Bowen Yang, Jingbo Zhou, Qinghong Miao et al. (+1 more)
Lookup-based
memory has been a promising way to scale the parameters of large
language models (LLMs). It retrieves learned representations of local
token patterns, such as n-grams, instead of reconstructing them through successive layers of computation. However, existing designs such as Engram treat each retrieved
embedding as a monolithic unit. Each
embedding is stored in its own hashed slot and modulated by a single scalar gate. As a result, polysemous patterns cannot selectively read out the components of their
memory that are relevant to the context. Moreover, parameters are shared only through hash collisions, which are largely unrelated to semantics. We propose FactorEngram, a factorized n-gram
memory with basis-level contextual gating. FactorEngram retrieves sparsity-regularized coefficients over a dictionary of basis vectors shared across patterns, so related patterns can reuse common components. The same dictionary is also used for gating. The backbone hidden state is scored against each basis vector to gate the corresponding coefficient before reconstruction, which lets the context modulate each
memory component individually. FactorEngram also covers both individual
tokens and multi-
token n-grams, and we systematically study where the
memory branch should be inserted. On 340M- and 1B-parameter
Transformer backbones, FactorEngram improves
language modeling and downstream task performance. Ablation studies confirm the contribution of each component and identify insertion before the
attention sublayer in the middle layers as an effective configuration.
Read the full abstract ▾
All authors Bowen Yang, Jingbo Zhou, Qinghong Miao, Hua Wu
Computation and Language · Cross-listed from cs.AI
Share-Borne AI Virus: Memory-Hopping Attacks Across LLM Agents
Sidharth Pulipaka, Ansh Sharma, Stanislau Hlebik et al. (+4 more)
Large
language models are increasingly deployed as stateful assistants that retain information across interactions and use tools to read, modify, and create persistent artifacts. As these artifacts are shared between users, they form an indirect communication channel between otherwise independent assistants. We study a failure mode in which this channel enables self-propagating attacks. We introduce artifact-mediated propagation, where
adversarial content introduced through an artifact (e.g. a report), is stored in an assistant's persistent
memory, reproduced in a subsequently created artifact, and acquired by another assistant that later reads it. We evaluate this process in temporal human-
agent universes that model artifact exchange between independently operated assistants over time, measuring whether an attack survives successive hand-offs, how many hops it reaches, and how broadly it spreads. We find that attacks can propagate across multiple independent assistants and persist over extended interaction sequences. In larger simulated environments, even GPT-5.6 Luna exhibits substantial spread, reaching 60-80% of
agents with propagation chains extending to eight hops. These results show that persistent artifacts can act as durable carriers of
adversarial state, allowing attacks to outlive individual interactions and spread across isolated assistants.
Read the full abstract ▾
All authors Sidharth Pulipaka, Ansh Sharma, Stanislau Hlebik, Leonidas Raghav, Vyas Raina, Ivaxi Sheth, Mario Fritz
Computation and Language · Cross-listed from cs.AI
Representation Alignment as a Bottleneck in LLM-Based Retrosynthesis Planning
Hyunwoo Yoo, Cassie Huang, Haebin Shin et al. (+2 more)
While LLMs show promise in general
reasoning, symbolic planning in chemistry remains a bottleneck. Direct ”SMILES-to-PDDL” attempts fail because they force models to juggle chemical analysis and planning-
language structuring simultaneously. We hypothesize that this failure stems from a lack of intermediate abstractions rather than insufficient model capacity. By decomposing retrosynthesis into molecule mapping, reaction mapping, and PDDL generation, we achieve high success rates where end-to-end approaches fail. This provides evidence that a primary bottleneck lies in representation
alignment rather than raw model capacity. Our structural analysis demonstrates that intermediate representations are essential in retrosynthesis planning, highlighting the importance of representation-centric design in future systems.
Read the full abstract ▾
All authors Hyunwoo Yoo, Cassie Huang, Haebin Shin, Li Zhang, Gail L. Rosen
Computation and Language
Almieyar: A Culturally Grounded Benchmark for Multi-Dialect Arabic Speech Recognition
Omid Ghahroodi, Anas Madkoor, Dima Faris Al Saudi et al. (+37 more)
Arabic
speech technology has largely focused on Modern Standard Arabic, leaving the living dialects spoken by hundreds of millions under-served. We introduce ALMIEYAR, a culturally grounded ASR
benchmark covering 17 Arabic dialects across six families, built entirely from newly recorded
speech unseen by existing models. Dialect-community coordinators selected culturally relevant images across 10 topics, and native speakers described them through five structured scenarios, yielding approximately 50 minutes per dialect (13.7 hours total). We
benchmark 12 state-of-the-art ASR systems zero-shot, including GPT-4o-transcribe, Voxtral-Mini-4B, Fanar-STT-LF, Whisper, SeamlessM4T-v2, and wav2vec2-based models. GPT-4o-transcribe achieves the lowest overall WER at 35.0%, followed by Voxtral-Mini-4B, Fanar-STT-LF, and Whisper-Large-v3 at 41.1%, 45.9%, and 49.5%, respectively, indicating substantial remaining errors across Arabic dialect communities. Performance varies considerably across dialect groups, with no model performing uniformly best across all groups. WER alone also obscures dialectal ASR behaviour: wav2vec2-based models show large WER/CER gaps, where character-level agreement remains much higher than word-level accuracy, motivating joint WER/CER reporting. ALMIEYAR provides a unified
benchmark for culturally grounded Arabic ASR evaluation, including the first published
benchmark for Ahwazi Arabic.
Read the full abstract ▾
All authors Omid Ghahroodi, Anas Madkoor, Dima Faris Al Saudi, Fagr Tahir, Malak Annan, Talha shahid javad allah rakha, Omar Al-Busaidi, Zineb El Kahla, Iheb Zouari, Essa Ahmed Abou Jabal, Ahmed Ezzat, Hind AL-Merekhi, Aisha Hamad M A Al-Naimi, Hadi Wazni, Bushra Alnajjar, Omar Amin, Haya Al-Thani, Houssam Eddine-Othman Lachemat, Marwa Elwakedy, Sundus Abdulmalik Al Nahari, Elahe Zahiri, Osamah Sarraj, Raghad Mousa, Mckeen Assi, Ahd Al Jumah, Heyam Salman, Alhanouf Abdulraqib, Sara Benoumhani, Alia Hamwi, Ayaat Al-Yasseri, Rim Ibrahim Ghazal, Lamia Ben hiba, Mohamed Eltabakh, Fatima Al-Raisi, Yassine El Kheir, Mohammed Abdulrahman, Hamdy Mubarak, Ayah Hashem, Lefkir Meriem, Ehsaneddin Asgari
Computation and Language
Less Sycophancy, Stronger Refusal? Lessons for AI Safety from Mechanistic Interpretability
Xu Wang, Difan Zou and Xuansheng Wu
Reliable refusal of harmful requests is essential to the safe deployment of
language models. Because excessive eagerness to please users may undermine existing refusal capabilities, reducing sycophancy offers a potential route to stronger refusal beyond the harmful scenarios covered by safety training. We investigate this possibility using compensatory feature injection (CFI), a training technique designed to limit the acquisition of a target concept by supplying its associated activation during learning. Across three Qwen3.5 base models, we use sparse autoencoders (SAEs) to identify the top-ranked sycophancy feature from paired sycophantic and independent responses, then validate its behavioral influence through inference steering. We subsequently inject the selected feature during supervised fine-tuning on sycophantic targets. Positive injection reduces learned sycophancy after removal (by 62.0% relative to ordinary fine-tuning in 35B-A3B), whereas modest negative injection increases it. Unexpectedly, these reductions in sycophancy do not consistently improve direct refusal of harmful requests, motivating a narrower evaluation of the same harmful intents under user pressure. In this setting, ordinary fine-tuning on sycophantic responses substantially weakens refusal, while selected checkpoints trained with positive injection recover part of the loss, including approximately 95% in 35B-A3B. These findings show that persistent sycophancy reduction does not guarantee stronger direct refusal, while identifying recovery under user pressure as a distinct, conditional benefit of training intervention.
Read the full abstract ▾
Computation and Language
Beyond Token Scale: Chunk-Level Sparse Autoencoders for Reliable Semantic Feature Discovery
Xu Wang, Yifan Yang, TingHao YU et al. (+1 more)
Sparse autoencoders (SAEs) expose features that help us understand and steer
language models, but faithful reconstruction does not guarantee informative concepts.
Token-level objectives reward lexical and formatting details alongside semantic content, all competing for a limited sparse budget. We introduce a family of chunk-level SAEs that encode mean-pooled activations over chunks, each a contiguous span of
tokens: Mean-Chunk reconstructs the observed chunk, Cross-Chunk predicts an independently processed neighbor, and Joint-Chunk combines both targets. These designs separate the effect of a larger observation unit from that of predicting information shared across passages. With matched training data, chunk-level SAEs remain powerful interpretability tools while learning reliable semantic features that capture high-level concepts and respond selectively to relevant content. Their strengths are complementary: Mean-Chunk improves high-level feature discovery,
reasoning detection beyond surface cues, and steering; Cross-Chunk leads document
retrieval and classification transfer while producing selective, persistent features. Changing what an SAE sees and predicts yields reliable semantic features for more meaningful tasks. We demonstrate their practical value through gains across downstream tasks such as
retrieval,
reasoning detection, and steering.
Read the full abstract ▾
All authors Xu Wang, Yifan Yang, TingHao YU, Difan Zou
Computation and Language
Who Is Left of Whom? Tracing Spatial Evidence and Role Binding in Relative-Position Reasoning
Yingjin Song, Denis Paperno and Albert Gatt
High instance-level accuracy can mask inconsistencies in spatial
reasoning when objects exchange positions or their roles are reversed in the query. The internal representations supporting relative-position
reasoning remain poorly understood. We investigate two complementary components of this process: tracking object locations in the input and representing their query roles. Across three VLMs with visual or textual inputs and their
language-model backbones, activation patching reveals a staged progression from early-layer source representations through intermediate-layer query-object representations to late-layer answer states. Targeted interventions further establish causal links along this progression: manipulating source-side representations shifts location information at query-object mentions and ultimately alters relation predictions. Beyond object-location information, we also identify a stable query-side direction associated with the roles of the two objects in the comparison. Steering along directions estimated on synthetic scenes generalizes to natural-image
benchmarks, improving accuracy and both forms of paired consistency in most settings without retraining. Our findings reveal complementary components of relational
reasoning across visual and textual settings and show how targeted interventions can improve the consistency of models' behavior.
Read the full abstract ▾
Computation and Language
Spontaneous Context Restoration: How Language Models Recover from Corrupted Inputs
Pranjal Garg and Jacob Beck
Language models sometimes produce correct outputs even when their inputs are corrupted by deletion, replacement, or misspelling. We study the internal processes accompanying this behavior, which we call context restoration, in controlled
attention-only
transformers and five pretrained LLMs (1B-32B parameters) across arithmetic, reading comprehension, and multiple-choice
reasoning tasks. In the
attention-only
transformers, restoration emerges spontaneously despite training exclusively on clean sequences, without corruption training or an explicit denoising objective. We find that context restoration follows a two-phase process: early layers localize effects associated with repair at corrupted positions, while later layers accumulate these effects at uncorrupted positions through the residual stream and ultimately concentrate them at the output position. Repair outcome is predictable from hidden states: cosine
alignment with the clean state is highly predictive in
attention-only models, while linear probes recover additional information in pretrained LLMs. A linear probe using only the corrupted
prompt's first-block hidden state predicts failure with mean ROC-AUC 0.78. This enables failure triage under matched or even partially shifted deployment conditions and may reduce unnecessary verification or computation. Failed examples also show substantially greater nonlinearity along corruption directions. Moderate-corruption finetuning increases corruption tolerance while simultaneously reducing displacement-normalized linearization error, associating improved robustness with a more nearly linear response to corruption.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.LG
CLIMB: A Clinical Multimorbidity Benchmark for Diagnosing Co-occurring Conditions through Multiturn Conversations
Yusuf Kesmen, Aniruddha Mukherjee, Yena Chang et al. (+7 more)
Patients often have several co-occurring clinical conditions, and the findings needed to identify and disambiguate them emerge over the course of a consultation. Evaluating clinical
reasoning in this setting requires both multi-turn interaction and multi-label diagnosis. We introduce CLIMB, a
benchmark in which a doctor model interviews a simulated patient to recover a ground truth set of co-occurring clinical conditions. Cases are synthesized from clinical decision
algorithms and diagnostic
datasets, grounding multimorbid presentations in structured clinical knowledge. Across six frontier and open models, none recovers the exact set of conditions in more than 10% of interactive cases. Diagnostic performance declines when conditions co-occur, even when models receive the full clinical record and the true number of conditions. Interaction reduces performance further. In controlled experiments, models behave like single-hypothesis trackers: they anchor on the diagnosis suggested by the opening findings, keep questioning around it, and recover a second condition mainly when a finding in view points to it. Questioning them further does not complete the set but adds mostly wrong diagnoses. We formalise this pattern with a theoretical reference model of single-hypothesis tracking. The
benchmark, generator, and evaluation code are available at https://anonymous.4open.science/r/CLIMB-8340.
Read the full abstract ▾
All authors Yusuf Kesmen, Aniruddha Mukherjee, Yena Chang, David Sasu, Trevor Brokowski, Alexandra V. Kulinkina, Kristina Keitel, Akhil Arora, Lars Henning Klein, Mary-Anne Hartley