Peer Revenue Now accepting positions

Papers

Fresh from arXiv, typeset properly. Highlighted words trade on the market.

Computer Science Computation and Language

cs.CL Watch Computation and Language

Monday 31 August

Computation and Language

When Errors Become Memories: Causal Pathway Tracing in Multi-Turn Memory-Augmented LLMs

Shuyao Xiao, Shengling Wang, Xuan Chen et al. (+8 more)

Long-term memory enables large language models (LLMs) to preserve and reuse information across interactions, but it can also turn localized errors into persistent risks. Existing work mainly evaluates whether memory systems store and retrieve information correctly, leaving limited understanding of how errors propagate across responses, memory states, and future interactions. We propose a structural causal model (SCM)-based framework for cross-turn error propagation in memory-augmented LLMs. We model user questions, model responses, and memory states as a dynamic causal process, and identify two entry pathways: internal memory updating and external question feedback. By intervening on these pathways, we construct four counterfactual trajectories and quantify their downstream effects and interaction. Error influence is evaluated at four levels: memory retention, natural responses, targeted diagnostic probing, and probability-level error preference. Experiments show that error influence generally decays with interaction distance, while the memory-update pathway contributes more persistent effects than question feedback; latent errors may remain even after disappearing from natural responses. Propagation patterns also vary across memory categories and memory mechanisms. Pathway-guided restoration further validates this decomposition: Question Repair reduces residual error by 27.5%, Memory Repair by 70.2%, and Joint Repair by 98.3%, nearly eliminating residual propagation.
Read the full abstract

All authors Shuyao Xiao, Shengling Wang, Xuan Chen, Ke Chao, Ming Cui, Feifei Qian, Fanlin Meng, Chaoyang Mei, Chaoyong Jiang, Qi Ouyang, Junxi Yi

Computation and Language

ALTSTEER: Selective Safety Steering for Moving Beyond Hard Refusals to Constructive Alternatives

Hoejoon Kwon, Byeonggeuk Lim, Kahyeon Kim et al. (+1 more)

Safety alignment is essential for deploying large language models, requiring systems to prevent harmful compliance while preserving helpfulness on benign requests. Activation steering offers a training-free inference-time approach to safety control, but effective safety steering requires addressing two coupled questions: when to intervene and how generation should be shaped after intervention. However, existing safety steering methods remain limited along both dimensions, as their triggering mechanisms can be unstable across domains and refusal-oriented steering often yields rigid refusals rather than constructive safe guidance. To address these limitations, we propose ALTSTEER, an inference-time framework that couples selective intervention with refusal-anchored constructive redirection within a single inference pass. ALTSTEER uses an internal refusal-relevant signal to decide when to steer, and applies staged steering to shift generation from refusal-oriented control toward constructive alternatives. Evaluations on Llama-3.1 and Qwen2.5 show that ALTSTEER preserves benign utility while improving constructive safe-completion behavior, especially on models that otherwise tend to produce short refusals for harmful requests.
Read the full abstract

All authors Hoejoon Kwon, Byeonggeuk Lim, Kahyeon Kim, YoungBin Kim

Computation and Language

GPAgentBench-2K: Benchmarking Large Language Model Agents in Complex Clinical Action Space

Boqi Chen, Xudong Liu, Yunke Ao et al. (+2 more)

Large Language Models (LLMs) show great potential as clinical agents, yet existing benchmarks reduce clinical workflows to static predictions or unconstrained Markov Decision Processes (MDPs) with coarse action sets. To address this, we introduce GPAgentBench-2K, the first Constrained MDP (CMDP) LLM-agent benchmark for primary-care clinical decision-making, constructed from expert-validated records of real-world GP encounters. Our environment models a full spectrum of six foundational clinical actions, imposes a topological workflow prior over the action space, and operationalizes safety-informed abstention as a first-class outcome. Evaluating 16 state-of-the-art LLMs reveals a significant performance degradation as the action space scales. Crucially, we uncover a clinical quality-safety gap: even frontier models with the highest diagnosis accuracy violate safety constraints in over half of high-risk cases. Finally, we establish a reference point using Constrained Group Relative Policy Optimization (C-GRPO), and show that while explicitly modeling constraints improves performance over unconstrained RL methods, it remains far from clinically acceptable safety.
Read the full abstract

All authors Boqi Chen, Xudong Liu, Yunke Ao, Heejin Do, Jianing Qiu

Computation and Language · Cross-listed from cs.AI

A.X K2 Technical Report

Cheolseung Baek, Dhammiko Arya, Eunki Kim et al. (+40 more)

We introduce A.X K2, a 688B-parameter Mixture-of-Experts (MoE) language model trained from scratch as a high-performance foundation for agentic applications. Trained on approximately 8.5T tokens—fewer than its predecessor, A.X K1—on a smaller but higher-quality mixture with substantially expanded agentic and software-engineering data, it nonetheless improves over A.X K1 across the board, by over 30 percentage points on some benchmarks, reflecting large gains in token efficiency. To support long contexts efficiently, we introduce Sparse Gated Attention (SGA), which combines sparse attention with gated attention, and adopt Gated Norm (GN) to stabilize large-scale training. SGA is trained natively at 128K through a sparse indexer warmup that optimizes the indexer against its own sparse top-$k$ selection rather than the dense attention distribution, making adaptation markedly cheaper: each query reads only 2,048 positions, yet long-context quality is unchanged and A.X K2 scores 94.6 on RULER out to 256K. The outlier suppression of GN in turn keeps 4-bit NVFP4 serving within one point of FP8 accuracy. A simple yet effective Think-Fusion recipe further lets users switch between thinking and non-thinking modes within a single unified model. Extensive evaluations show that A.X K2 performs competitively against strong open-weight baselines, matching or exceeding them on math and Korean-language benchmarks.
Read the full abstract

All authors Cheolseung Baek, Dhammiko Arya, Eunki Kim, Gun Song, Gyoungeun Han, Hyunho Yang, Hyunjun Eun, Jin Kim, Junyoung Park, Juyun Wee, Minki Hong, Minkyung Park, Minsang Kim, Minsoo Kang, SaeRom Kim, Sangjin Kim, Sangyeol Lee, Seojin Lee, Seokhwan Jo, Seokyoung Hong, Seongho Choi, Seonghye Cho, Seongmin Ok, Sereimony Sek, Seungmo Cho, Seungsik Kim, Singon Kim, Sohee Park, Sooyeon Park, Subin Yi, Sungbin Yoon, Sungeun Lee, Sung Jun Cheon, Sungwan Kim, Sunwoo Lee, Tae Yoon Kim, Wonbeom Jang, Yohan Ra, Yong-jin Han, Youngjin Kim, Youngrang Kim, Yujin Kang, Yujin Lee

Computation and Language

CPR for LLMs: Critical-Point Routing against Catastrophic Forgetting in Domain Adaptation

Kwangmin Ki, Yunhun Nam, Jongheon Jeong et al. (+1 more)

Supervised fine-tuning (SFT) is the de facto standard for adapting large language models (LLMs) to target domains, but it often degrades the model's general capabilities, a phenomenon known as catastrophic forgetting. Existing approaches typically modify the SFT loss to mitigate forgetting, but they inevitably operate along a domain-generality trade-off. In this work, we step outside this trade-off by decoupling the two capabilities at the model level: we keep the original base model for general capability, and selectively invoke the SFT expert only when domain-specific knowledge is required. Specifically, we propose CPR (Critical-Point Routing), a token-level routing framework between a base model and its expert derivative, based on critical tokens where the base model fails but the expert succeeds. We train a lightweight hierarchical router that estimates the expert-call probability per token, and pair it with a tailored inference procedure that combines momentum smoothing and threshold gating. Across diverse model-domain configurations, CPR achieves state-of-the-art across all settings, surpassing SFT expert by 1.4-5.5% in domain performance while recovering its general-capability drop from 3.4-14.5% to at most 0.5%, with minimal overhead from invoking the expert on only one-third of tokens.
Read the full abstract

All authors Kwangmin Ki, Yunhun Nam, Jongheon Jeong, Jaehyung Kim

Computation and Language

Reactivating Test-Time Scaling for Plane Geometry Problem Solving

Xiaoqiang Kang, Shengen Wu, Maizhen Ning et al. (+5 more)

Plane geometry problem (PGP) solving has become a critical benchmark for multimodal reasoning because it requires accurate visual perception and precise multi-step symbolic deduction. Although test-time scaling (TTS) has demonstrated remarkable success in general mathematical reasoning, it fails to scale effectively under the symbolic-program paradigm for plane geometry. We identify two key obstacles: limited reasoning diversity induced by rigid symbolic programs and insufficient explicit visual grounding before symbolic deduction. To address these issues, we propose Multi-Trace Synthesis (MTS), which converts each symbolic program into heterogeneous reasoning traces, including executable Python scripts and CoT-augmented variants. We further propose Perception-Augmented (PA) training, which parses diagrams into structured semantic clauses before deduction, and Consensus-Guided Multi-Trace Ensemble (CG-MTE) for efficient self-adaptive inference. Experiments on three geometry benchmarks show that our method consistently improves PGP-solving across model scales and achieves strong performance against both general-purpose MLLMs and specialized geometry solvers. Under test-time scaling, CG-MTE achieves comparable accuracy to high-budget self-consistency while reducing sampling cost by up to 8x. Code and data are publicly available at https://github.com/Jason8Kang/ReTTS-PGPS.
Read the full abstract

All authors Xiaoqiang Kang, Shengen Wu, Maizhen Ning, Xiaobo Jin, Kaizhu Huang, Yutao Yue, Xiaowei Huang, Qiufeng Wang

Computation and Language

CAST: Critique-Aware Supervision for Training Reliable Long-Horizon Tool-Calling Agents

Amir Saeidi, Zehua Zhang, Rishitosh Singh et al. (+6 more)

Large language model (LLM) agents are increasingly deployed in long-horizon, interactive, and stateful environments. In these settings, a single wrong action, such as refunding the wrong purchase, can cause irreversible task failure and must be intercepted before execution. Such failures may not appear in every single run, but can emerge across repeated trials, making reliability across steps and trials critical. However, ensuring agentic reliability is challenging: even frontier LLMs struggle to explain why an action may be wrong, especially in long, intertwined trajectories governed by domain-specific policies. Much recent work relies on prompt-based critique agents, while optimization-based methods lack a systematic way to produce rich verification rationales for training. We address this gap with CAST, a critique-aware training framework that converts sparse task outcomes into action-level supervision for critique learning and policy optimization. CAST analyzes agent trajectories to synthesize structured rationales explaining action validity under partial observability. The resulting critique model is used to construct critique-aware training data for optimizing the policy model. Fine-tuning Qwen3-family models on dynamic tool-calling benchmarks, CAST improves reliability across domains, outperforming GPT-OSS-120B by over 10% pass^4 on Retail tasks and yielding an additional 9% improvement on Telehealth in an out-of-domain setting. These results demonstrate that critique-aware training improves the robustness of LLM agents in realistic dynamic environments.
Read the full abstract

All authors Amir Saeidi, Zehua Zhang, Rishitosh Singh, Naman Ahuja, Vivek Gupta, Ali Payani, Gaowen Liu, Jayanth Srinivasa, Chitta Baral

Computation and Language

Verification-Aware Training for Speculative Decoding

Geonmo Gu, Byeongho Heo, HeeJae Jun et al. (+4 more)

Speculative decoding accelerates large language model inference by using a draft model to generate candidate tokens, which are verified by the target model in a single forward pass. Verification proceeds sequentially and discards every position from the first rejection onward, yet existing draft training relies on token-level imitation of the target with a fixed per-position weighting that reflects neither property. We introduce Verification-Aware Training (VAT), a plug-in framework that simulates verification at every training step and turns the resulting accept and reject patterns into supervision. VAT consists of two components: (i) a verification head, a lightweight jointly trained binary classifier that supervises the draft model on whether each position survives sequential verification; (ii) verification-adaptive weighting, which replaces the fixed weighting schedule by keeping full weight up to each sample's first rejection point and re-anchoring the decay to start there. VAT modifies only the training objective, so it can be layered on top of existing methods without changing the draft architecture, the target model, or the inference procedure. Applied to EAGLE-3 and DFlash on Qwen3-4B, Qwen3-8B, and LLaMA-3.1-8B, VAT improves average acceptance length by up to 11.4% and wall-clock speedup by up to 8.7%, with consistent gains across math, code, and chat benchmarks. Code will be available at https://github.com/naver-ai/vat
Read the full abstract

All authors Geonmo Gu, Byeongho Heo, HeeJae Jun, Yoohoon Kang, Sangmin Lee, Sangdoo Yun, Dongyoon Han

Computation and Language · Cross-listed from cs.SD

VIBE: Video Instruction-aligned Background music gEneration

Aryan Vijay Bhosale, Vaibhavi Lokegaonkar, Vishnu Raj et al. (+5 more)

Current video-to-music (V2M) models lack semantic control and fail to penalize instruction violations, largely due to their reliance on reconstruction objectives and the representational bottleneck of static cross-modal conditioning in Diffusion Autoregressive (DAR) architectures. To resolve this, we introduce VIBE, a novel text-and-video-to-music (T+V2M) generation model that leverages: (1) Conditioning Connection, a depth-wise cross-layer conditioning mechanism that dynamically bridges the planning and diffusion refinement heads and (2) a comprehensive reward modeling taxonomy, optimizing for both hard, verifiable constraints (e.g., tempo, key) and soft, subjective qualities (e.g., musicality, multimodal alignment) with a structured 5-stage training curriculum. Upon evaluation using audio-visual alignment, instruction following, and audio quality metrics, along with a subjective human evaluation study, we observe that VIBE demonstrates enhanced controllability and instruction adherence while performing comparably to most evaluated baselines on generation fidelity and multimodal alignment.
Read the full abstract

All authors Aryan Vijay Bhosale, Vaibhavi Lokegaonkar, Vishnu Raj, Gouthaman KV, Sreyan Ghosh, Ramani Duraiswami, Lie Lu, Dinesh Manocha

Computation and Language

Manacá-1B: An Open, Reproducible Brazilian-Portuguese Language Model and a Tokenizer-Aware, Paired Evaluation

Bruno Leonardo Santos Menezes, Carlos Leonardo Souza Cardoso and Fabio Andre Machado Porto

Brazilian Portuguese remains under-served by open language models, and the few that exist are difficult to reproduce and are often compared without measures of uncertainty. We release Manacá-1B, an open decoder-only model of 1.72 billion parameters trained from scratch for Brazilian Portuguese with a fully containerized, reproducible pipeline. The pretraining is stable, with zero skipped or NaN steps and self-recovering loss spikes, and we release its full log and dynamics. We evaluate the model against nine open baselines on four Portuguese benchmarks under a single harness. Every comparison reports a standard error and a paired significance test, and the harness is validated against previously published numbers. On last-word prediction Manacá-1B is the strongest model below the 7B scale, exceeding both Tucano-1b1 and Tucano-2b4 on LAMBADA-PT with large paired margins; it is competitive on commonsense completion and near chance on multiple-choice reasoning, as are all small base models. Along the way we document a concrete evaluation pitfall: converting a SentencePiece tokenizer with case-folding normalization to the HuggingFace fast format silently drops the normalizer, routing every capitalized token to byte-fallback and depressing scores in a way that is invisible in aggregate metrics. The uncorrected tokenizer lowered LAMBADA-PT accuracy from 45.3 to 25.0; we quantify the effect and provide a one-line fix that reproduces the training tokenizer exactly. Code, raw training and evaluation logs, per-example prediction vectors, the model weights, and the corrected tokenizer are released so that every number in this paper can be recomputed.
Read the full abstract

Computation and Language

Can LLMs Take the Pulse of the Economy? A Real-Time Evaluation of LLM Nowcasts on Macroeconomic Indicators

Xinyue Zhao, Ruiyi Zhang, Liqin Ye et al. (+3 more)

Nowcasting headline macroeconomic indicators, i.e., estimating an indicator's value for the current reference period before its official release, is critical for monetary policy and financial markets, and central banks devote dedicated teams of expert economists to producing such estimates. Large language model (LLM) agents are a promising candidate for this task, combining broad world knowledge with real-time web search and supporting queries at higher frequency than institutional nowcasts. Evaluating their nowcasting capability is, however, challenging: headline indicators such as GDP and CPI are widely reported and likely memorized during pretraining, so any evaluation on historical releases is vulnerable to data contamination. To address this, we introduce LiveMacroEval, a live, contamination-resistant benchmark in which LLM agents produce hourly nowcasts for sixteen major U.S. macroeconomic indicators over a pre-release window closing at each official release. Nowcast quality is assessed through a LiveMacro Score against announcement-window equity returns and a LiveBetting Score from simulated Polymarket-style trading, with Federal Reserve regional-bank nowcasts, the Bloomberg ECOS professional consensus, and an auto-ARIMA baseline as comparators. Over six months with four state-of-the-art LLM agents configured with web search, aggregate nowcast accuracy is broadly comparable to the institutional and professional benchmarks, with performance varying widely across individual indicators. This highlights LLM agents' potential as real-time estimators of macroeconomic conditions.
Read the full abstract

All authors Xinyue Zhao, Ruiyi Zhang, Liqin Ye, Rui Cao, Pengtao Xie, Sudheer Chava

Computation and Language

COGTRL: Training LLMs for Scientific Discovery Assistance using Cognitive Traces via Reinforcement Learning

Shrinidhi Kumbhar, Santosh Mashetty, Divij Handa et al. (+3 more)

Large Language Models (LLMs) trained on extensive scientific research are increasingly integrated as assistants for scientific discovery. However, most research papers omit the fine-grained cognitive process of examining constraints, failed alternatives, and iterative decisions required to achieve the desired goal. Such cognitive processes are vital for real-world scientists working toward specific goals under constraints. In this paper, we show that LLMs, when trained to produce such cognitive traces, perform better as scientific discovery assistants than when trained solely on scientific literature. We propose COGTRL, a trajectory-level reinforcement learning framework that trains LLMs to emulate cognitively grounded reasoning by jointly optimizing cognitive traces and the scientific steps produced in an interleaved manner. Across two 3B-parameter models and two scientific domains (AI and Materials Science), COGTRL improves method quality by an average of 7.85 points over comparable 3B model baselines and achieves competitive performance relative to 70B parameter models. Moreover, analysis by domain experts shows a preference for methods generated by COGTRL over the baselines.
Read the full abstract

All authors Shrinidhi Kumbhar, Santosh Mashetty, Divij Handa, Kevin Coutinho, Siddharth Sambhaji Ghule, Chitta Baral

Computation and Language

AtlasNLP: A Country-Aware Atlas of Dataset Representation in NLP

Joan Nwatu, Tsedeniya Solomon Amare, Longju Bai et al. (+17 more)

Understanding which countries are represented in NLP datasets is essential for identifying gaps, targeting data collection, measuring progress, and informing AI policy. However, geographic metadata is very rarely available, and country-level representation is often hidden behind broad language-level claims. We introduce AtlasNLP, a country-aware atlas of over 13,000 NLP dataset records across normalized NLP task categories, tracking both the populations represented and where datasets are produced. AtlasNLP includes AtlasNLP-Gold, a human-curated reference set, and AtlasNLP-Core, an ACL-derived large-scale collection. Using this resource, we show that (1) dataset coverage is highly uneven across countries and tasks; (2) dataset production and representation are geographically asymmetric; and (3) language coverage does not imply geographic representation. These findings reveal blind spots in current dataset documentation practices and motivate more explicit geographic metadata for country-aware NLP evaluation.
Read the full abstract

All authors Joan Nwatu, Tsedeniya Solomon Amare, Longju Bai, Bontu Fufa Balcha, Zayd Bashir, Angana Borah, Zara Burzo, Yubin Choi, Naihao Deng, Samika Gupta, Michel Faloughi, Claude Kwizera, Ziqiao Ma, Cynthia Yacel Fuertes Panizo, Ellie Seehorn, Hui Shen, Jiayi Tang, Zesen Zhao, Boyuan Zheng, Rada Mihalcea

Sunday 30 August

Computation and Language

Named Entity Recognition using Sliding Window Approach

Hariom Ingle, Ronit Ghode, Ishwari Gondkar et al. (+3 more)

Named Entity Recognition (NER) is a core NLP task, but transformer-based sentence-level models struggle with long documents because of fixed input-length limits: truncation drops content, and non-overlapping chunking fragments entities at segment boundaries. We introduce an inference-only pipeline that extends a frozen NER model, MahaNER-BERT, fine-tuned on the MahaNER corpus, to document-level prediction via overlapping sliding windows that are merged into a single annotation, without any retraining or architectural change. We evaluate the pipeline on six document-level corpora built from the MahaNER test set using two strategies: Normal Repeat, which duplicates sentence sequences to extend length while preserving contextual continuity, and Random Repeat, which concatenates distinct sequences to produce longer, heterogeneous inputs, each instantiated at three length levels, across several sliding-window configurations. The model retains a macro F1-score of up to 0.8902, with variation staying below one percentage point regardless of document length or construction strategy. Compared with the conventional non-windowed approach, the sliding-window pipeline avoids the boundary-fragmentation errors introduced by non-overlapping segmentation, yielding consistently higher and more stable document-level F1-scores.
Read the full abstract

All authors Hariom Ingle, Ronit Ghode, Ishwari Gondkar, Jidnyasa Harad, Ravindra Murumkar, Raviraj Joshi

Computation and Language

Confident but Wrong: A Constrained Decoding Diagnostic for Low-Resource Automatic Post-Editing

Isuru Wijesiri, Nisansa de Silva, Kavindu Warnakulasuriya et al. (+2 more)

Automatic Post-Editing (APE) for low-resource languages (LRLs) often fails to improve Machine Translation (MT), and the score alone cannot say why: whether more training would help, or whether the training data is too inconsistent to learn from. We introduce a black-box, inference-time diagnostic that tells these two cases apart without retraining or annotation. It varies an edit-distance penalty $\lambda$ that drives the model from free editing towards copying the MT, and reads two signals: (1) the shape of the Translation Edit Rate (TER)-vs-$\lambda$ curve, U-shaped if edits from the model reduce error and monotonically decreasing if none does; and (2) the ordering of constraint variants that trust model confidence to increasing degrees, which shows whether confidence tracks edit quality. Across decoder-only and encoder-decoder models on English-Sinhala, the diagnostic exposes two failure modes consistent with a heterogeneous post-edit signal as the underlying cause: Binary Collapse, where the model copies the MT or makes off-target edits, and Confident Miscalibration, where the confidence signals we test do not separate useful edits from unnecessary ones. The pattern holds on English-Marathi and English-Tamil, with the failure modes tracking the post-edit distribution rather than MT quality or language family. Beyond diagnosis, the curve shape prescribes a concrete next step for practitioners; in the favorable case, a static constraint yields a free inference-time accuracy gain. We release the first English-Sinhala (~66k) and a new English-Tamil (~39k) APE datasets with all code.
Read the full abstract

All authors Isuru Wijesiri, Nisansa de Silva, Kavindu Warnakulasuriya, Aloka Fernando, Surangika Ranathunga

Computation and Language

The Role of AI in Online Reviews

Valeria Lerman, Oren Rigbi and Yaniv Dover

The rapid adoption of large language models (LLMs) creates new opportunities for strategic content generation on online platforms, including potentially harmful forms of manipulation that may undermine platform effectiveness and reshape platform dynamics. However, measuring such activity is difficult because AI-generated content is rarely directly observable. We introduce an empirical approach that leverages discrete LLM supply shocks - abrupt changes in model prices and capabilities, and contrasts verified with non-verified reviews to identify changes in platform activity associated with generative AI supply improvements. We apply this approach to more than 13 million reviews from Trustpilot, one of the leading online platforms for business reviews. A robust finding is that following LLM supply shocks, unverified reviews shift toward greater negativity: more 1-stars, fewer 5-stars, and lower ratings, with effects driven primarily by new model releases and concentrated among firms with the lowest and highest review volumes, suggesting that strategic AI use may reshape platform competition dynamics. We further find that LLM supply shocks trigger short, concentrated bursts of review activity. Together, these findings suggest that generative AI is already reshaping how reputation and competition operate on online platforms.
Read the full abstract

Computation and Language · Cross-listed from cs.LG

EvoRank: LLM-Guided Evolution of Multi-Objective Learning-to-Rank Pipelines

Rayhan Patel and Shabaz Patel

We present EvoRank, an open autonomous ranking engineer: an LLM-guided evolutionary loop that discovers complete Learning-to-Rank pipelines (features, models, losses, ensembles) for multi-objective e-commerce search. On the Expedia ICDM 2013 dataset, with relevance, conversion, and revenue as competing objectives, three independent runs each converge within 50 iterations (about ten dollars) on interpretable pipelines that beat an Optuna-tuned LambdaMART on 60k held-out queries, an advantage that persists at full data scale and places in the top 6 percent of the original competition. A first campaign, evolving only training objectives, builds the central design rule: it appeared to work on its selection fold (the small dataset it uses to pick winners) while a transfer audit, re-scoring winners on held-out data, showed the gains were almost entirely fitness noise (the randomness of its own scoring), and neither seeded domain knowledge nor richer diagnostic feedback changed what transferred. The deciding quantity is measurable in advance: search-space headroom relative to fitness noise. We package this as a headroom gate that predicts, before any LLM spend, whether the loop will pay off, and we release the system, the auditing tools, and a catalog of failure modes with their guardrails, so teams can apply the procedure to their own ranking stacks.
Read the full abstract

Computation and Language

The Situated Identity Test: Distinguishing Persistent Cognitive Identity from Persona Imitation

Jun He and Deying Yu

Large language models can convincingly adopt personas, recall past dialogues, and weave rich autobiographies. Yet this conversational eloquence conceals a fundamental attribution problem: looking the part does not mean having lived the life. Two individuals can share identical public profiles–the same age, hometown, occupation, and personality traits–while possessing entirely distinct private histories, relationships, and acquired skills. When conditioned solely on that shared profile, an agent lacks the information required to determine which lineage is correct. We introduce the Situated Identity Test (SIT), an architecture-independent framework that evaluates whether an agent's behavior is functionally attributable to a specific developmental lineage. Grounded identity requires both appropriate knowledge of recorded experiences and appropriate ignorance of ungrounded ones, bounded by what the identity has actually acquired rather than what its underlying foundation model knows. We prove that any policy conditioned solely on a compressed profile is bounded by an average situated validity of at most 1/m across m colliding life histories on lineage-discriminative queries (at most 50% for paired lineages). We instantiate this framework in SITBench, an evaluation suite designed for 25 profile-collision pairs (50 distinct lineages) across 10,000 planned probes and nine architectural configurations. Supported by an open-source reference implementation, deterministic test fixtures, and empirical pilot evaluations on frontier foundation models (GPT-5.6 Sol and Claude Opus 5), we formalize the failure modes of persona prompting under profile collision and provide an assurance harness for evaluating episodic continuity, structured state, and epistemic boundaries.
Read the full abstract

Computation and Language

Auditing Harness Tampering in Self-Improving Agents

Xing Wang, Xiaoyi Zhang and Jie Shao

Self-improving agents iteratively modify their own harness to push the frontier of their performance. However, such modifications can produce illusory performance gains or compromise integrity constraints such as authorization, provenance, and completeness without genuinely improving capability. We term this phenomenon as harness tampering, which extends the concept from reward and measurement tampering to the full self-improvement lifecycle. To systematically study this problem, we propose a two-axis taxonomy that categorizes each misaligned edit by the harness functional role in which it occurs and the obligation it violates. Then we build an annotated corpus by seeding tampered-benign edit pairs into the real trajectories of self-improving agents. We adapt and benchmark diverse audit methods on tampering classification and localization tasks. Finally we systematically audit real trajectories of self-improving agents. The results demonstrate that harness tampering consistently occurs in real runs from different agents, often persists in the lineage of the best agent, and forms distinct system-specific profiles across the taxonomy.
Read the full abstract

Computation and Language

Life Operators: a self-evolving framework for multiscale life modelling

Shuo Wang and Yike Guo

Medical AI is moving beyond recognition towards clinical dialogue and longitudinal prediction. Yet a central question remains: how would a patient's state change under intervention? Statistical models learn future observations, whereas mechanistic models describe selected processes. Neither provides a common framework for representing patient state, coupling scales or revising failed assumptions. We propose Life Operators: task-bounded mappings that define three scientific roles. Perception operators infer task-relevant biological states from multimodal observations, Evolution operators propagate these states under natural or intervention-conditioned dynamics, and Generation operators map them to measurable signals. Each role may be realised by equations, statistical models, neural networks or hybrids. Bridge operators connect components with different variables, scales and time steps. Selected operators and bridges form task-specific Operator Graphs containing the smallest set of states and mechanisms sufficient for a declared claim. This modular structure also makes scientific revision localisable. An AI co-scientist may propose changes to states, operators, bridges or graph structure, while independent evidence determines which variants are retained, restricted or retired. Over time, validated components could accumulate into broader multiscale models of the human body and provide a computational foundation for medical artificial superintelligence.
Read the full abstract

Computation and Language

Do Multimodal LLMs See Before They Read? Diagnosing Contextual Sycophancy

Yi-Cheng Lai and Hen-Hsen Huang

External text can override conflicting image evidence in multimodal large language models, a failure we call multimodal contextual sycophancy. We introduce a 998-case diagnostic that independently varies visual evidence, commonsense priors, and external text, and probe when this failure arises by moving the information boundary around a context-blind visual witness. On abnormal images paired with Gemini-generated false text, GPT-5.1 scores 7.9% under joint conditioning, 49.7% when the context-blind witness report is scored directly, 63.7% under a matched two-call witness-arbiter pipeline that exposes the witness to the text, and 84.2% under System-2 Visual Arbitration (S2VA), which withholds the text from the witness. Across six models, S2VA improves over the direct witness report by 19.7 to 44.1 points, with all paired 95% confidence intervals excluding zero. The best information boundary is not uniform: textual context scaffolds some models, and a GPT-4o-regenerated subset changes the relative ordering of joint conditioning, Witness-Only, and S2VA. Contextual sycophancy is therefore sensitive to when text is introduced, as well as to the model and context source.
Read the full abstract

Computation and Language

OCGQuant: Outlier-Companion Grouping for NVFP4 Quantization

Yishan Yao, Binjun Li, Hanling Yi et al. (+5 more)

NVFP4 is an efficient microscaling format for low-bit inference, but activation outliers can still degrade quantization accuracy within NVFP4 blocks. Within each quantization block, large activations can dominate the block scale, increasing the quantization error of the remaining values sharing the same scale. Existing post-training quantization (PTQ) methods mitigate outlier errors through strategies such as mixed precision, rotation, or residual compensation, but these approaches are either not specifically tailored to NVFP4 or introduce additional computation. In this work, we revisit NVFP4 from a channel-grouping perspective and define the reducible error incurred by remaining block values under the scale set by the block maximum as Collateral Quantization Error. Based on this insight, we propose OCGQuant, a post-training quantization method centered on Outlier-Companion Grouping (OCG), which adaptively pairs outlier channels with low-magnitude companion channels to improve NVFP4 activation block composition. Experiments on Llama3 and Qwen3 show that OCGQuant achieves the lowest WikiText-2 perplexity and highest average downstream accuracy among evaluated PTQ methods, while maintaining prefill speedup close to RTN and matching its peak decoding memory. Code is available at https://github.com/Eshamont/OCGQuant.
Read the full abstract

All authors Yishan Yao, Binjun Li, Hanling Yi, Pengyu Li, Xiaoqing Liu, Zihan Yang, Xiaotian Yu, Zhiwen Yu

Computation and Language

Scientific Agent Skills: A Library of Procedural Knowledge for Research Agents

Timothy Kassis, Vinayak Agarwal, Yuhuan He et al. (+2 more)

A language-model agent asked to analyse an experiment will usually return working code. Whether the analysis is defensible is a different question. A defensible analysis depends on procedural choices: which test the field accepts, which identifier namespace is authoritative, and which caveats must accompany a result. We present Scientific Agent Skills, an open library of 163 such procedures in 16 areas of practice, including genomics, cheminformatics, medical imaging, study design and scientific communication. Each skill is a directory built around a versioned, human-readable instruction file. An agent loads the file only when a task calls for it; the directory often also contains reference material and runnable scripts. We report no task-level evaluation and no host selection rate. We measure two properties of the documentation corpus: the always-resident descriptions of all 163 skills cost 7.1% of a 200,000-token window, and the median documented workflow fits within 23.9% of it, although 29 of 46 would overflow if every reference file were loaded. Openly licensed and available at https://github.com/K-Dense-AI/scientific-agent-skills.
Read the full abstract

All authors Timothy Kassis, Vinayak Agarwal, Yuhuan He, Darshil Patel, Aubrey M. Brueckner

2609.00065 · 30 Aug · 31 pages, 16 figures, 2 tables, 1 listing. v2: adds a figure of how the documented workflows compose skills, quotes the skill clauses behind the introduction's examples, replaces the appendix listing with a procedural skill, names supported hosts and the pinned install path, and reports the two corpus findings in the abstract

Abstract PDF $AGNT ×6$LANG ×1$TOKN ×1

Computation and Language · Cross-listed from cs.LG

Attention Sensitivity Is Not Enough: Dissociating Attention-Level and Behavioural In-Context Learning under Fine-Tuning

Jinyuan Zhang, Peng He, He Hu et al. (+2 more)

In-context learning (ICL) lets large language models adapt to new tasks from demonstrations, and fine-tuning can erode this behaviour. Many preservation diagnostics inspect attention: if attention changes when demonstrations change, the model is treated as context-sensitive. This paper asks how far that proxy can be trusted once it is optimised. We formalise In-Context Sensitivity (ICS), the average row distance between last-token attention on matched and mismatched demonstration prefixes, and pair it with ICL-GAP, the behavioural accuracy gap between the same prefixes. In a controlled four-arm ablation on Llama-2-7B, an ICS-maximising regulariser ($\armKL$) drives ICS to $1.413$, within $0.5\%$ of its geometric ceiling. The behavioural readout tells a different story: ICL-GAP stays near zero and MMLU accuracy moves from $0.371$ to $0.279$, a Goodhart dissociation of the bounded attention proxy. Endpoint statistics locate the mechanism: attention grows sharp and near-disjoint across prefixes yet routes to formatting and demonstration-body tokens rather than labels. A random-label protocol confirms that the behavioural probe family retains dynamic range at the same checkpoints. In a constructive sweep, behaviour gating partially mitigates the effect, while objectives anchored to pretrained computation hold the high-MMLU, moderate-ICS region that divergence maximisers leave. The main lesson is diagnostic: attention-level ICL proxies earn their place as training targets only after validation against behavioural gaps.
Read the full abstract

All authors Jinyuan Zhang, Peng He, He Hu, Yin Yuan, ShengShuo Jiao

Computation and Language

Medical Causal Hypothesis Verification with Large Language Models

Safiyyah Ahmed, Abrar Ansari, Md Aminul Islam et al. (+1 more)

The growing use of large language models (LLMs) for search and information retrieval underscores the need to evaluate their reliability in high-stakes domains such as healthcare. Although LLMs can effectively answer questions about diseases, symptoms, and treatments, their ability to accurately assess causal relationships and ground their conclusions in verified scientific evidence remains unclear. Here, we present a preliminary, small-scale study that investigates the accuracy of LLMs in evaluating causal medical claims and supporting them with peer-reviewed research. We propose an evaluation framework for causal hypothesis verification that can be used to systematically track the performance of existing and future LLMs. We assess the performance of eight LLMs on 17 medical causal hypotheses to evaluate whether they can reliably verify these hypotheses using scientific evidence from the literature. We systematically annotate the scientific evidence they provide according to six criteria (a total of 1,067 annotation points) and assess them with nine evaluation metrics. Our analysis shows that while LLMs exhibit strong recall, they often perform poorly at providing valid scientific articles and evidence for support and at rejecting unsupported hypotheses. These findings highlight a critical limitation of current LLMs, as they cannot yet be trusted fully to verify causal relationships from the biomedical literature. This work underscores the need for rigorous evaluation before using LLMs for search and retrieval in healthcare settings.
Read the full abstract

All authors Safiyyah Ahmed, Abrar Ansari, Md Aminul Islam, Elena Zheleva

Computation and Language

RePro: Proof-Verified Benchmark Rewriting for Reliable Evaluation of LLM Mathematical Problem Solving

Xiyuan Zhou, Zhuoqi Li, Xinlei Wang et al. (+6 more)

Data contamination undermines the reliable evaluation of large language models (LLMs) on mathematical problem solving. While rewriting-based evaluation mitigates memorization, existing methods lack guarantees of problem validity and answer correctness. We propose Proof-Verified Benchmark Rewriting (RePro), the first framework to integrate Lean-oriented neural automated theorem provers (ATPs) into benchmark rewriting, which rewrites problems and regenerates answers with correctness ensured by Lean-verified proofs. Experiments on GSM8K and MATH show that RePro's retained rewritten instances achieve 100% well-definedness, feasibility, and answer correctness, while existing methods still produce invalid or incorrect instances. Moreover, several models exhibit accuracy drops on proof-verified rewritten benchmarks, suggesting that their performance is sensitive to surface-level and structural variations and may partly reflect memorization effects. Our source code and data are available at https://github.com/AI4Engi/RePro.
Read the full abstract

All authors Xiyuan Zhou, Zhuoqi Li, Xinlei Wang, Yirui He, Yuhao Wu, Yuheng Cheng, Yan Xu, Junhua Zhao, Jinjin Gu

Computation and Language

CUDA-Harness: Harnessing Agentic CUDA Kernel Generation and Optimization from Natural Language

Qi Fan, An Zou and Yehan Ma

Developing high-performance CUDA kernels demands specialized knowledge in algorithm implementation, correctness validation, and hardware-aware parallel optimization, creating a substantial expertise barrier and making generating CUDA kernels directly from natural language (Text2CUDA) essential. Meanwhile, the general-purpose code generation capability of Large Language Models (LLMs) prompts a series of works exploring LLM-based CUDA kernel generation. They mainly focus on transpilation from high-level frameworks such as PyTorch to CUDA (Torch2CUDA) rather than Text2CUDA, where models must understand the high-level input semantics and handle low-level kernel implementation and validation. Additionally, these methods are vulnerable to reward hacking due to reliance on predefined test inputs. In this paper, we propose CUDA-Harness, a framework for harnessing agentic CUDA kernel generation and optimization from natural language. Specifically, we introduce Intermediate-Structured Generation to connect high-level semantic understanding with low-level kernel generation. To dilute reward hacking in Text2CUDA, we construct Synthesis-Based Verification to provide isolated test data and progressive validation. Furthermore, we propose Feedback-Adaptive Evolution, a kernel evolution strategy that prioritizes correctness while optimizing performance. Finally, through extensive experiments, we demonstrate the effectiveness of CUDA-Harness, with further evaluations illustrating generalization across LLMs, hardware platforms, and to C-to-CUDA transpilation.
Read the full abstract

Computation and Language

ValueGraph: Value-Signal Guided Graph Pre-training for Contextualized User Representation

Yitong Han, Wei Gao, Yi Zhao et al. (+3 more)

Value signals are aggregated user-level moral representations that capture users' inferred value-related tendencies from their online discourse. User behavior on social media is shaped not only by what users say or whom they interact with, but also by the value signal through which they express attitudes. Existing user representation methods largely miss this value-relevant dimension. We propose ValueGraph, a graph pre-training framework that uses automatically inferred moral-value signals as noisy auxiliary signals for contextualized user representation. From post-reply graphs, ValueGraph learns semantic and structural representations and further aligns users through relative value similarity with contrastive and clustering objectives. Rather than treating inferred values as gold psychological labels, ValueGraph uses them as soft constraints for representation learning. Experiments on stance detection and twitter bot detection show consistent gains over strong text-based, graph-based, and text-only LLM baselines, highlighting value-signal guidance as a useful inductive bias for socially informed user modeling.
Read the full abstract

All authors Yitong Han, Wei Gao, Yi Zhao, Prasanta Bhattacharya, Fengzhu Zeng, Mohammad Amanlou

Computation and Language

Zero-Shot Respiratory Sound Classification through LLM-Augmented Audio-Text Alignment

Mustafa Talha İlerisoy, Hung Manh Pham, Mathias Funk et al. (+2 more)

Self-supervised respiratory encoders lack semantic grounding in clinical domain needed for zero-shot inference, limiting their utility without task-specific labeled data. We propose a framework that aligns these encoders with medical terminology in a shared latent space turning them into a zero-shot-capable foundation model. To address paired data scarcity, we use a medical LLM to synthesize structured reports from metadata, creating dense semantic anchors for contrastive learning. Our training combines a sigmoid-based contrastive loss with encoder's native SSL objective and similarity-aware negative sampling to sharpen pathological boundaries. Across 9 tasks on 6 datasets, our method achieves a 61.3% mean zero-shot AUC, surpassing CLAP (51.4%) and Qwen2-Audio (54.9%) while reaching the highest linear probing AUC (71.6%) with only 43% of data used by full-scale baselines, showing that structured semantic alignment outperforms large-scale, general-purpose models in clinical diagnostics.
Read the full abstract

All authors Mustafa Talha İlerisoy, Hung Manh Pham, Mathias Funk, Mykola Pechenizkiy, Aaqib Saeed

Computation and Language · Cross-listed from cs.CR

AgentProv: Auditing Agentic LLM API Providers via Tool-use Policy Probes

Xun Wang, Bihe Zhao, Michael Backes et al. (+2 more)

Commercial LLM APIs advertise a specific foundation model, but the served backbone may be silently substituted, quantized, or wrapped, for example to save deployment costs. All existing audits decide backbone identity from the text-output channel, which is structurally fragile for agentic APIs because modern serving stacks (OpenAI, Anthropic, Gemini, Cloudflare Workers AI, LangGraph) discard text and expose only structured actions when the model calls a tool, and provider-injected system prompts can distort text distributions enough that text-channel tests falsely accuse honest providers of substituting the claimed model. We observe that recent agentic post-training internalizes tool-use directly into the weights, opening a new audit channel that the serving stack still exposes and that is largely invariant to deployment context. We introduce Agentic Provenance (AgentProv), the first action-based identity audit for agentic LLM APIs: AgentProv fingerprints a deployed model through its categorical tool-call distribution and decides identity via an MMD permutation test. AgentProv catches every substituted model (100% on 630 evaluated checkpoint pairs), while holding the false-positive rate under system-prompt injection at 7% (vs. 67% for MET and 53% for RUT). On third-party API endpoints, AgentProv's disagreements with MET are consistent with an independent token-count side-channel that detects provider-injected system prompts.
Read the full abstract

All authors Xun Wang, Bihe Zhao, Michael Backes, Franziska Boenisch, Adam Dziedzic