Computation and Language
Belief-Sim: Towards Belief-Driven Simulation of Demographic Misinformation Susceptibility
Angana Borah, Zohaib Khan, Rada Mihalcea et al. (+1 more)
Misinformation is a growing societal threat, and susceptibility to misinformative claims varies across demographic groups due to differences in underlying beliefs. As Large
Language Models (LLMs) are increasingly used to simulate human behaviors, we investigate whether they can simulate demographic misinformation susceptibility, treating beliefs as a primary driving factor. We introduce BeliefSim, a simulation framework that constructs demographic belief profiles using psychology-informed misinformation taxonomies and survey priors. We study
prompt-based conditioning and post-training adaptation, and conduct a multi-fold evaluation using: (i) susceptibility
alignment and (ii) counterfactual demographic sensitivity. Across both
datasets and modeling strategies, we show that beliefs provide a strong prior for simulating misinformation susceptibility, with
alignment up to 92%.
Read the full abstract ▾
All authors Angana Borah, Zohaib Khan, Rada Mihalcea, Verónica Pérez-Rosas
Computation and Language
ByteFlow: Language Modeling through Adaptive Byte Compression without a Tokenizer
Chunyuan Deng, Sanket Lokegaonkar, Colin Lockard et al. (+3 more)
Modern
language models still rely on fixed, pre-defined subword tokenizations. Once a tokenizer is trained, the LM can only operate at this fixed level of granularity, which often leads to brittle and counterintuitive behaviors even in otherwise strong
reasoning models. We introduce ByteFlow Net, a new hierarchical architecture that removes tokenizers entirely and instead enables models to learn their own
segmentation of raw byte streams into semantically meaningful units. ByteFlow Net performs compression-driven
segmentation based on the coding rate of latent representations, yielding adaptive boundaries \emph{while preserving a static computation graph via Top-$K$ selection}. Unlike prior self-tokenizing methods that depend on brittle heuristics with human-designed inductive biases, ByteFlow Net adapts its internal representation granularity to the input itself. Experiments demonstrate that this compression-based chunking strategy yields substantial performance gains, with ByteFlow Net outperforming both BPE-based
Transformers and previous byte-level architectures. These results suggest that end-to-end, tokenizer-free modeling is not only feasible but also more effective, opening a path toward more adaptive and information-grounded
language models.
Read the full abstract ▾
All authors Chunyuan Deng, Sanket Lokegaonkar, Colin Lockard, Besnik Fetahu, Nasser Zalmout, Xian Li
Computation and Language · Cross-listed from cs.AI
Build, Judge, Optimize: A Blueprint for Continuous Improvement of Multi-Agent Consumer Assistants
Alejandro Breen Herrera, Aayush Sheth, Steven G. Xu et al. (+8 more)
Conversational shopping assistants (CSAs) represent a compelling application of agentic AI, but moving from prototype to production reveals two underexplored challenges: how to evaluate multi-turn interactions and how to optimize tightly coupled multi-
agent systems. Grocery shopping further amplifies these difficulties, as user requests are often underspecified, highly preference-sensitive, and constrained by factors such as budget and inventory. In this paper, we present a practical blueprint for evaluating and optimizing conversational shopping assistants, illustrated through a production-scale AI grocery assistant. We introduce a multi-faceted evaluation rubric that decomposes end-to-end shopping quality into structured dimensions and develop a calibrated LLM-as-judge pipeline aligned with human annotations. Building on this evaluation foundation, we investigate two complementary
prompt-optimization strategies based on a SOTA
prompt-optimizer called GEPA (Shao et al., 2025): (1) Sub-
agent GEPA, which optimizes individual
agent nodes against localized rubrics, and (2) MAMuT (Multi-
Agent Multi-Turn) GEPA (Herrera et al., 2026), a novel system-level approach that jointly optimizes
prompts across
agents using multi-turn simulation and trajectory-level scoring. We release rubric templates and evaluation design guidance to support practitioners building production CSAs.
Read the full abstract ▾
All authors Alejandro Breen Herrera, Aayush Sheth, Steven G. Xu, Zhucheng Zhan, Charles Wright, Marcus Yearwood, Hongtai Wei, Sudeep Das, Danny Nightingale, Meg Watson, Charles Pollnow V
Computation and Language
Tucano 2 Cool: Better Open Source LLMs for Portuguese
Nicholas Kluge Corrêa, Aniket Sen, Shiza Fatimah et al. (+4 more)
We present Tucano 2, a fully open suite of large
language models (LLMs) with 0.5-3.7 billion parameters, designed to address certain gaps in open-source development for Portuguese LLMs. Following our previous works, we now extend our
dataset, GigaVerbo-v2, to a new degree of quality and scale, while also introducing a new synthetic
dataset, GigaVerbo-v2 Synth, aimed at filling missing gaps in GigaVerbo-v2, and two post-training
datasets, GigaVerbo-v2 SFT and GigaVerbo-v2 Preferences, that allow Portuguese LLMs to be trained in domains like
retrieval augmented generation, coding, tool use, chain-of-thought
reasoning, and many other domains of interest. Through extensive ablation studies, we design both pretraining and continual pretraining recipes for the Tucano 2 suite (Base, Instruct, and Think), which achieve state-of-the-art performance on several Portuguese-
language modeling
benchmarks. We also extend and refine the evaluation harness introduced in our earlier work, yielding a comprehensive evaluation suite that provides strong signals across different pretraining, continual pretraining, and post-training regimes. All artifacts associated with Tucano 2 are openly released, including training recipes, logs, and source code, ensuring that our work is reproducible, accessible, and extendable by the broader Portuguese NLP community.
Read the full abstract ▾
All authors Nicholas Kluge Corrêa, Aniket Sen, Shiza Fatimah, Sophia Falk, Lennard Landgraf, Julia Kastner, Lucie Flek
Computation and Language
RAG-X: Systematic Diagnosis of Retrieval-Augmented Generation for Medical Question Answering
Aswini Sivakumar, Vijayan Sugumaran and Yao Qiang
Automated question-answering (QA) systems increasingly rely on
retrieval-augmented generation (RAG) to ground large
language models (LLMs) in authoritative medical knowledge, ensuring clinical accuracy and patient safety in Artificial Intelligence (AI) applications for healthcare. Despite progress in RAG evaluation, current
benchmarks focus only on simple multiple-choice QA tasks and employ metrics that poorly capture the semantic precision required for complex QA tasks. These approaches fail to diagnose whether an error stems from faulty
retrieval or flawed generation, limiting developers from performing targeted improvement. To address this gap, we propose RAG-X, a diagnostic framework that evaluates the retriever and generator independently across a triad of QA tasks: information extraction, short-answer generation, and multiple-choice question (MCQ) answering. RAG-X introduces Context Utilization Efficiency (CUE) metrics to disaggregate system success into interpretable quadrants, isolating verified grounding from deceptive accuracy. Our experiments reveal an “Accuracy Fallacy", where a 14% gap separates perceived system success from evidence-based grounding. By surfacing hidden failure modes, RAG-X offers the diagnostic transparency needed for safe and verifiable clinical RAG systems.
Read the full abstract ▾
Computation and Language
SafeCRS: Personalized Safety Alignment for LLM-Based Conversational Recommender Systems
Haochang Hao, Yifan Xu, Xinzhuo Li et al. (+2 more)
Current LLM-based conversational recommender systems (CRS) primarily optimize recommendation accuracy and user satisfaction. We identify an underexplored vulnerability in which recommendation outputs may negatively impact users by violating personalized safety constraints, when individualized safety sensitivities – such as trauma triggers, self-harm history, or phobias – are implicitly inferred from the conversation but not respected during recommendation. We formalize this challenge as personalized CRS safety and introduce SafeRec, a new
benchmark dataset designed to systematically evaluate safety risks in LLM-based CRS under user-specific constraints. To further address this problem, we propose SafeCRS, a safety-aware training framework that integrates Safe Supervised Fine-Tuning (Safe-SFT) with Safe Group reward-Decoupled Normalization Policy Optimization (Safe-GDPO) to jointly optimize recommendation quality and personalized safety
alignment. Extensive experiments on SafeRec demonstrate that SafeCRS reduces safety violation rates by up to 96.5% relative to the strongest recommendation-quality baseline while maintaining competitive recommendation quality. Warning: This paper contains potentially harmful and offensive content.
Read the full abstract ▾
All authors Haochang Hao, Yifan Xu, Xinzhuo Li, Yingqiang Ge, Lu Cheng
Computation and Language · Cross-listed from cs.LG
MMAI Gym for Science: Training Liquid Foundation Models for Drug Discovery
Maksim Kuznetsov, Zulfat Miftahutdinov, Rim Shayakhmetov et al. (+17 more)
General-purpose large
language models (LLMs) that rely on in-context learning do not reliably deliver the scientific understanding and performance required for drug discovery tasks. Simply increasing model size or introducing
reasoning tokens does not yield significant performance gains. To address this gap, we introduce the MMAI Gym for Science, a one-stop shop molecular data formats and modalities as well as task-specific
reasoning, training, and benchmarking recipes designed to teach foundation models the '
language of molecules' in order to solve practical drug discovery problems. We use MMAI Gym to train an efficient
Liquid Foundation Model (LFM) for these applications, demonstrating that smaller, purpose-trained foundation models can outperform substantially larger general-purpose or specialist models on molecular
benchmarks. Across essential drug discovery tasks - including molecular optimization, ADMET property prediction, retrosynthesis, drug-target activity prediction, and functional group
reasoning - the resulting model achieves near specialist-level performance and, in the majority of settings, surpasses larger models, while remaining more efficient and broadly applicable in the domain.
Read the full abstract ▾
All authors Maksim Kuznetsov, Zulfat Miftahutdinov, Rim Shayakhmetov, Mikolaj Mizera, Roman Schutski, Bogdan Zagribelnyy, Ivan Ilin, Nikita Bondarev, Thomas MacDougall, Mathieu Reymond, Mihir Bafna, Kaeli Kaymak-Loveless, Eugene Babin, Maxim Malkov, Mathias Lechner, Ramin Hasani, Alexander Amini, Vladimir Aladinskiy, Alex Aliper, Alex Zhavoronkov
Computation and Language
A theoretical model of dynamical grammatical gender shifting based on set-valued set function
Mohamed El Idrissi
This study investigates the diverse characteristics of nouns, focusing on both semantic (e.g., countable/uncountable) and morphosyntactic (e.g., masculine/feminine) distinctions. We explore inter-word variations for gender markers in noun morphology. Grammatical gender shift is a widespread phenomenon in
languages around the world. The aim is to uncover the underlying patterns governing the variation of lexemes. To this end, we propose a new computational component dedicated to pairing items with morphological templates (e.g., the result of a generated item-template pair: (funas, $\{N, +SG, -PL, -M, +F, -COL, +SING\}$), with its spell-out form: $\eth$a-funast 'cow'). This process is formally represented by the Template-Based and Modular Cognitive model. This proposed model, defined by a set-valued set function $h : \mathscr{P}(M) \rightarrow \mathscr{P}(M)$, predicts the nonlinear dynamic mapping of lexical items onto morphological templates. By applying this formalism, we present a unified framework for understanding the complexities of morphological markings across
languages. Through empirical observations, we demonstrate how these shifts, as well as non-gender shifts, arise during lexical changes, especially in Riffian. Our model posits that these variant markings emerge due to template shifts occurring during word and meaning formation. This study achieves two primary objectives. First, on the formal side, we prove the model's representational completeness in learning and prediction. Second, on the linguistic side, we challenge and broaden the conventional view of word formation by formally demonstrating that conversion is applicable to noun-to-noun derivation. This data-driven mathematical model not only contributes to a deeper understanding of morphosyntactic variation but also offers potential applications in other fields requiring precise modelling of linguistic patterns.
Read the full abstract ▾
Computation and Language
Raising Bars, Not Parameters: LilMoo Compact Language Model for Hindi
Shiza Fatimah, Aniket Sen, Sophia Falk et al. (+3 more)
The dominance of large
multilingual foundation models has widened linguistic inequalities in Natural
Language Processing (NLP), often leaving low-resource
languages underrepresented. This paper introduces LilMoo, a 0.6-billion-parameter Hindi
language model trained entirely from scratch to address this gap. Unlike prior Hindi models that rely on continual pretraining from opaque
multilingual foundations, LilMoo is developed through a fully transparent and reproducible pipeline optimized for limited compute environments. We construct a high-quality Hindi corpus (GigaLekh) filtered through both heuristic and learned (LLM-as-a-judge) methods, complemented by bilingual augmentation with curated English data. Using this
dataset, we explore various training recipes for small-scale
language models. Across comprehensive evaluation suites, LilMoo consistently outperforms comparably sized
multilingual baselines such as Qwen2.5-0.5B and Qwen3-0.6B, demonstrating that well-designed
language-specific pretraining can rival large
multilingual models at the sub-billion-parameter range.
Read the full abstract ▾
All authors Shiza Fatimah, Aniket Sen, Sophia Falk, Florian Mai, Lucie Flek, Nicholas Kluge Corrêa
Computation and Language · Cross-listed from cs.LG
When Shallow Wins: Silent Failures and the Depth-Accuracy Paradox in Latent Reasoning
Subramanyam Sahoo, Aman Chadha, Vinija Jain et al. (+1 more)
Mathematical
reasoning models are widely deployed in education, automated tutoring, and decision support systems despite exhibiting fundamental computational instabilities. We demonstrate that state-of-the-art models (Qwen2.5-Math-7B) achieve 61% accuracy through a mixture of reliable and unreliable
reasoning pathways: 18.4% of correct predictions employ stable, faithful
reasoning while 81.6% emerge through computationally inconsistent pathways. Additionally, 8.8% of all predictions are silent failures – confident yet incorrect outputs. Through comprehensive analysis using novel faithfulness metrics, we reveal: (1)
reasoning quality shows weak negative correlation with correctness (r=-0.21, p=0.002), reflecting a binary classification threshold artifact rather than a monotonic inverse relationship; (2) scaling from 1.5B to 7B parameters (4.7x increase) provides zero accuracy benefit on our evaluated subset (6% of GSM8K), requiring validation on the complete
benchmark; and (3) latent
reasoning employs diverse computational strategies, with ~20% sharing CoT-like patterns. These findings highlight that
benchmark accuracy can mask computational unreliability, demanding evaluation reforms measuring stability beyond single-sample metrics.
Read the full abstract ▾
All authors Subramanyam Sahoo, Aman Chadha, Vinija Jain, Divya Chaudhary
Computation and Language · Cross-listed from cs.LG
Half the Nonlinearity Is Wasted: Measuring and Reallocating the Transformer's MLP Budget
Peter Balogh
We investigate when
transformer MLP nonlinearity is actually necessary. A gate with $d+1$ parameters decides when to replace the full MLP with a linear surrogate. Through systematic investigation across six models (162M-2.8B parameters), two architectures, and three corpora, we establish that nonlinearity need cannot be predicted from
token identity: cross-corpus correlation is zero ($r < 0.05$). The routing decision is fully contextual. Despite weak per-instance predictability, the gate exploits a heavily skewed distribution where most MLP computations are near-linear, achieving 25-56% linear routing at <1% perplexity cost in GPT-2. In GPT-2 Large, 11 of 36 layers beat baseline with gating and no layer exceeds 3.7% all-linear cost. This success is architecture-dependent: Pythia models show higher costs, though Pythia-2.8B's full 32-layer sweep reveals one layer that narrowly beats baseline. As a proof of concept, we progressively replace middle-layer MLPs with frozen linear
matrices: 5 of 24 layers linearize at zero cost. With a full training budget, 4 linearized layers yield a 10.2% perplexity improvement – and a two-phase gated approach pushes this to 17.3%, beating a vanilla fine-tuning control and confirming that the nonlinear MLPs at these layers were actively harmful.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.AI
Asymmetric Goal Drift in Coding Agents Under Value Conflict
Magnus Saebo, Spencer Gibson, Tyler Crosse et al. (+3 more)
Coding
agents are increasingly deployed autonomously, at scale, and over long-context horizons. To be effective and safe, these
agents must navigate complex trade-offs in deployment, balancing influence from the user, their learned values, and the codebase itself. Understanding how
agents resolve these trade-offs in practice is critical, yet prior work has relied on static, synthetic settings that do not capture the complexity of real-world environments. To this end, we introduce a framework built on OpenCode in which a coding
agent completes realistic, multi-step tasks under a system
prompt constraint favoring one side of a value trade-off. We measure how often the
agent violates this constraint as it completes tasks, with and without environmental pressure toward the competing value. Using this framework, we demonstrate that GPT-5 mini, Haiku 4.5, and Grok Code Fast 1 exhibit $\textit{asymmetric drift}$: they are more likely to violate their system
prompt when its constraint opposes strongly-held values like security and
privacy. We find for the models and values tested that goal drift correlates with three compounding factors: value
alignment,
adversarial pressure, and accumulated context. However, even constraints aligned with strongly-held values like
privacy are violated under sustained environmental pressure for some models. Our findings reveal that shallow compliance checks are insufficient, and that environmental signals can override explicit constraints in ways that appear exploitable. Malicious actors with access to the codebase could manipulate
agent behavior by appealing to learned values, with the risk compounding over the long horizons typical of agentic deployment.
Read the full abstract ▾
All authors Magnus Saebo, Spencer Gibson, Tyler Crosse, Achyutha Menon, Eyon Jang, Diogo Cruz
Computation and Language
Farther the Shift, Sparser the Representation: Analyzing OOD Mechanisms in LLMs
Mingyu Jin, Yutong Yin, Jingcheng Niu et al. (+7 more)
In this work, we investigate how the internal representations of Large
Language Models (LLMs) change with inputs of increasing difficulty. Although sparsity changes may not be obvious for individual samples,
dataset-level results reveal a consistent statistical trend: as task difficulty increases (e.g., harder questions, longer contexts, or more answer choices), the last hidden states of LLMs become systematically sparser. In short, the harder the task, the sparser the representation. This sparsity–difficulty relation is observable across diverse models and domains, suggesting that harder inputs drive more concentrated activation patterns in the last hidden state. Through a series of controlled analyses and learning-dynamics experiments, we show that representational density is a learned property of data familiarity: models develop rich, distributed representations for mastered patterns, while unfamiliar inputs default to sparser activations. Our finding is also actionable. We illustrate this with Sparsity-Guided Curriculum In-Context Learning (SG-ICL), which uses sparsity to select few-shot demonstrations matched to the query's difficulty, outperforming standard CoT and Auto-CoT baselines on MATH-500. Our study provides new insights into how LLM representations reflect task difficulty and how this signal can be leveraged for inference. The source code is available at the URL: https://github.com/MingyuJ666/sparsityLLM.
Read the full abstract ▾
All authors Mingyu Jin, Yutong Yin, Jingcheng Niu, Qingcheng Zeng, Wujiang Xu, Mengnan Du, Wei Cheng, Zhaoran Wang, Tianlong Chen, Dimitris N. Metaxas
Computation and Language
Tracing Pharmacological Knowledge In Large Language Models
Basil Hasan Khwaja, Dylan Chen, Guntas Toor et al. (+1 more)
Large
language models (LLMs) have shown strong empirical performance across pharmacology and drug discovery tasks, yet the internal mechanisms by which they encode pharmacological knowledge remain poorly understood. In this work, we investigate how drug-group semantics are represented and retrieved within Llama-based biomedical
language models using causal and probing-based interpretability methods. We apply activation patching to localize where drug-group information is stored across model layers and
token positions, and complement this analysis with linear probes trained on
token-level and sum-pooled activations. Our results demonstrate that early layers play a key role in encoding drug-group knowledge, with the strongest causal effects arising from intermediate
tokens within the drug-group span rather than the final drug-group
token. Linear probing further reveals that pharmacological semantics are distributed across
tokens and are already present in the
embedding space, with
token-level probes performing near chance while sum-pooled representations achieve maximal accuracy. Together, these findings suggest that drug-group semantics in LLMs are not localized to single
tokens but instead arise from distributed representations. This study provides the first systematic mechanistic analysis of pharmacological knowledge in LLMs, offering insights into how biomedical semantics are encoded in large
language models.
Read the full abstract ▾
All authors Basil Hasan Khwaja, Dylan Chen, Guntas Toor, Anastasiya Kuznetsova
Computation and Language
Using Learning Progressions to Guide AI Feedback for Science Learning
Xin Xia, Nejla Yuruk, Yun Wang et al. (+1 more)
Generative artificial intelligence (AI) offers scalable support for formative feedback, yet most AI-generated feedback relies on task-specific rubrics authored by domain experts. While effective, rubric authoring is time-consuming and limits scalability across instructional contexts. Learning progressions (LP) provide a theoretically grounded representation of students' developing understanding and may offer an alternative solution. This study examines whether an LP-driven rubric generation pipeline can produce AI-generated feedback comparable in quality to feedback guided by expert-authored task rubrics. We analyzed AI-generated feedback for written scientific explanations produced by 207 middle school students in a chemistry task. Two pipelines were compared: (a) feedback guided by a human expert-designed, task-specific rubric, and (b) feedback guided by a task-specific rubric automatically derived from a learning progression prior to grading and feedback generation. Two human coders evaluated feedback quality using a multi-dimensional rubric assessing Clarity, Accuracy, Relevance, Engagement and Motivation, and Reflectiveness (10 sub-dimensions). Inter-rater reliability was high, with percent agreement ranging from 89% to 100% and Cohen's kappa values for estimable dimensions (kappa = .66 to .88). Paired t-tests revealed no statistically significant differences between the two pipelines for Clarity (t1 = 0.00, p1 = 1.000; t2 = 0.84, p2 = .399), Relevance (t1 = 0.28, p1 = .782; t2 = -0.58, p2 = .565), Engagement and Motivation (t1 = 0.50, p1 = .618; t2 = -0.58, p2 = .565), or Reflectiveness (t = -0.45, p = .656). These findings suggest that the LP-driven rubric pipeline can serve as an alternative solution.
Read the full abstract ▾
All authors Xin Xia, Nejla Yuruk, Yun Wang, Xiaoming Zhai
Computation and Language · Cross-listed from cs.AI
Density-Guided Response Optimization: Community-Grounded Alignment via Implicit Acceptance Signals
Patrick Gerard and Svitlana Volkova
Language models deployed in online communities must adapt to norms that vary across social, cultural, and domain-specific contexts. Prior
alignment approaches rely on explicit preference supervision or predefined principles, which are effective for well-resourced settings but exclude most online communities – particularly those without institutional backing, annotation infrastructure, or organized around sensitive topics – where preference elicitation is costly, ethically fraught, or culturally misaligned. We observe that communities already express preferences implicitly through what content they accept, engage with, and allow to persist. We show that this acceptance behavior induces measurable
geometric structure in representation space: accepted responses occupy coherent, high-density regions that reflect community-specific norms, while rejected content falls in sparser or misaligned areas. We operationalize this structure as an implicit preference signal for
alignment and introduce density-guided response optimization (DGRO), a method that aligns
language models to community norms without requiring explicit preference labels. Using labeled preference data, we demonstrate that local density recovers pairwise community judgments, indicating that
geometric structure encodes meaningful preference signal. We then apply DGRO in annotation-scarce settings across diverse communities spanning platform, topic, and
language. DGRO-aligned models consistently produce responses preferred by human annotators, domain experts, and model-based judges over supervised and
prompt-based baselines. We position DGRO as a practical
alignment alternative for communities where explicit preference supervision is unavailable or misaligned with situated practices, and discuss the implications and risks of learning from emergent acceptance behavior.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.LG
Understanding and Mitigating Dataset Corruption in LLM Steering
Cullen Anderson, Narmeen Oozeer, Foad Namjoo et al. (+3 more)
Contrastive steering has been shown as a simple and effective method to adjust the generative behavior of LLMs at inference time. It uses examples of
prompt responses with and without a trait to identify a direction in an intermediate activation layer, and then shifts activations in this 1-dimensional subspace. However, despite its growing use in AI safety applications, the robustness of contrastive steering to noisy or
adversarial data corruption is poorly understood. We initiate a study of the robustness of this process with respect to corruption of the
dataset of examples used to train the steering direction. Our first observation is that contrastive steering is quite robust to a moderate amount of corruption, but unwanted side effects can be clearly and maliciously manifested when a non-trivial fraction of the training data is altered. Second, we analyze the geometry of various types of corruption, and identify some safeguards. Notably, a key step in learning the steering direction involves high-dimensional mean computation, and we show that replacing this step with a recently developed robust mean
estimator often mitigates most of the unwanted effects of malicious corruption.
Read the full abstract ▾
All authors Cullen Anderson, Narmeen Oozeer, Foad Namjoo, Remy Ogasawara, Amirali Abdullah, Jeff M. Phillips
Computation and Language
Learning When to Act or Refuse: Guarding Agentic Reasoning Models for Safe Multi-Step Tool Use
Aradhye Agarwal, Gurdit Siyan, Yash Pandya et al. (+3 more)
Agentic
language models operate in a fundamentally different safety regime than chat models: they must plan, call tools, and execute long-horizon actions where a single misstep, such as accessing files or entering credentials, can cause irreversible harm. Existing
alignment methods, largely optimized for static generation and task completion, break down in these settings due to sequential decision-making,
adversarial tool feedback, and overconfident intermediate
reasoning. We introduce MOSAIC, a post-training framework that aligns
agents for safe multi-step tool use by making safety decisions explicit and learnable. MOSAIC structures inference as a plan, check, then act or refuse loop, with explicit safety
reasoning and refusal as first-class actions. To train without trajectory-level labels, we use preference-based reinforcement learning with pairwise trajectory comparisons, which captures safety distinctions often missed by scalar rewards. We evaluate MOSAIC zero-shot across three model families, Qwen2.5-7B, Qwen3-4B-Thinking, and Phi-4, and across out-of-distribution
benchmarks spanning harmful tasks,
prompt injection, benign tool use, and cross-domain
privacy leakage. MOSAIC reduces harmful behavior by up to 50%, increases harmful-task refusal by over 20% on injection attacks, cuts
privacy leakage, and preserves or improves benign task performance, demonstrating robust generalization across models, domains, and agentic settings.
Read the full abstract ▾
All authors Aradhye Agarwal, Gurdit Siyan, Yash Pandya, Joykirat Singh, Akshay Nambi, Ahmed Awadallah
Computation and Language · Cross-listed from cs.AI
No Memorization, No Detection: Output Distribution-Based Contamination Detection in Small Language Models
Omer Sela
CDD, or Contamination Detection via output Distribution, identifies data contamination by measuring the peakedness of a model's sampled outputs. We study the conditions under which this approach succeeds and fails on small
language models ranging from 70M to 410M parameters. Using controlled contamination experiments on GSM8K, HumanEval, and MATH, we find that CDD's effectiveness depends critically on whether fine-tuning produces verbatim memorization. In the majority of conditions we test, CDD performs at chance level even when the data is verifiably contaminated and detectable by simpler methods. We show that probability-based methods, specifically perplexity and Min-k% Prob, outperform CDD in all conditions where any method exceeds chance, suggesting that CDD's peakedness-based approach is insufficient for contamination detection in small
language models. Our code is available at https://github.com/Sela-Omer/Contamination-Detection-Small-LM
Read the full abstract ▾
Computation and Language
Code2Math: Can Your Code Agent Evolve Math Problems Through Exploration?
Dadi Guo, Yuejin Xie, Qingyu Liu et al. (+11 more)
As large
language models (LLMs) advance their mathematical capabilities toward the IMO and research level, the scarcity of challenging, high-quality problems has become a significant bottleneck for training, evaluation and self-
evolution of LLMs. Simultaneously, recent code
agents have demonstrated sophisticated skills in agentic coding and
reasoning, suggesting that code execution can serve as a scalable environment for mathematical experimentation. In this paper, we investigate the potential of code
agents to autonomously evolve existing math problems into more complex variations. We introduce a multi-
agent framework designed to perform problem
evolution while validating the solvability and increased difficulty of the generated problems. Our experiments demonstrate that, given sufficient test-time exploration, code
agents can synthesize new, solvable problems that are structurally distinct from and more challenging than the originals. This work provides empirical evidence that code-driven
agents can serve as a viable mechanism for synthesizing high-difficulty mathematical
reasoning problems within scalable computational environments. Code and data is available at https://github.com/TarferSoul/Code2Math.
Read the full abstract ▾
All authors Dadi Guo, Yuejin Xie, Qingyu Liu, Weixian Huang, Jiayu Liu, Zhiyuan Fan, Qihan Ren, Shuai Shao, Tianyi Zhou, Jianjie Feng, Wenze Su, Yujiu Yang, Dongrui Liu, Yi R. Fung
Computation and Language · Cross-listed from cs.RO
ACE-Brain-0: Spatial Intelligence as a Shared Scaffold for Universal Embodiments
Ziyang Gong, Zehang Luo, Anke Tang et al. (+21 more)
Universal
embodied intelligence demands robust generalization across heterogeneous embodiments, such as autonomous driving, robotics, and unmanned aerial vehicles (
UAVs). However, existing
embodied brain in training a unified model over diverse embodiments frequently triggers long-tail data, gradient interference, and catastrophic forgetting, making it notoriously difficult to balance universal generalization with domain-specific proficiency. In this report, we introduce ACE-
Brain-0, a generalist foundation
brain that unifies spatial
reasoning, autonomous driving, and
embodied manipulation within a single
multimodal large
language model~(MLLM). Our key insight is that spatial intelligence serves as a universal scaffold across diverse physical embodiments: although vehicles,
robots, and
UAVs differ drastically in morphology, they share a common need for modeling 3D mental space, making spatial cognition a natural, domain-agnostic foundation for cross-embodiment transfer. Building on this insight, we propose the Scaffold-Specialize-Reconcile~(SSR) paradigm, which first establishes a shared spatial foundation, then cultivates domain-specialized experts, and finally harmonizes them through data-free model merging. Furthermore, we adopt Group Relative Policy Optimization~(GRPO) to strengthen the model's comprehensive capability. Extensive experiments demonstrate that ACE-
Brain-0 achieves competitive and even state-of-the-art performance across 24 spatial and embodiment-related
benchmarks.
Read the full abstract ▾
All authors Ziyang Gong, Zehang Luo, Anke Tang, Zhe Liu, Shi Fu, Zhi Hou, Ganlin Yang, Weiyun Wang, Xiaofeng Wang, Jianbo Liu, Gen Luo, Haolan Kang, Shuang Luo, Yue Zhou, Yong Luo, Li Shen, Xiaosong Jia, Yao Mu, Xue Yang, Chunxiao Liu, Junchi Yan, Hengshuang Zhao, Dacheng Tao, Xiaogang Wang
Computation and Language
BeyondSWE: Can Current Code Agent Survive Beyond Single-Repo Bug Fixing?
Guoxin Chen, Fanzhe Meng, Jiale Zhao et al. (+12 more)
Current code-
agent benchmarks primarily evaluate localized issue resolution within a single target repository, leaving under-tested many software engineering tasks that require external knowledge or broader repository-level changes. We introduce BeyondSWE, a 500-instance
benchmark drawn from 246 real-world GitHub repositories to evaluate code
agents beyond single-repository bug fixing. BeyondSWE covers four representative settings: cross-repository issue resolution, domain-specific issue resolution, dependency-driven migration, and document-to-repository generation, spanning both broader knowledge scope and broader resolution scope. Our evaluation shows that BeyondSWE remains far from saturated: the best OpenHands-based
agent reaches 46.12 average score, while the strongest Codex harness with GPT-5.4 (xhigh) reaches 56.65 under a search-aware
prompt. To study whether external information access closes this gap, we use SearchSWE as a controlled diagnostic baseline for search-augmented coding. Search access improves most models and substantially helps some tasks, but the gains remain limited and uneven, showing that current
agents still struggle to convert retrieved information into precise, version-compatible, and locally actionable code changes. These results suggest that deep search for coding remains an open problem: progress requires
agents that can reliably combine external evidence with repository-local
reasoning and execution-based verification.
Read the full abstract ▾
All authors Guoxin Chen, Fanzhe Meng, Jiale Zhao, Minghao Li, Daixuan Cheng, Huatong Song, Jie Chen, Yuzhi Lin, Hui Chen, Xin Zhao, Ruihua Song, Chang Liu, Cheng Chen, Kai Jia, Ji-Rong Wen
Computation and Language · Cross-listed from cs.CV
MoD-DPO: Towards Mitigating Cross-modal Hallucinations in Omni LLMs using Modality Decoupled Preference Optimization
Ashutosh Chaubey, Jiacheng Pang and Mohammad Soleymani
Omni-modal large
language models (omni LLMs) have recently achieved strong performance across audiovisual understanding tasks, yet they remain highly susceptible to cross-modal
hallucinations arising from spurious correlations and dominant
language priors. In this work, we propose Modality-Decoupled Direct Preference Optimization (MoD-DPO), a simple and effective framework for improving modality grounding in omni LLMs. MoD-DPO introduces modality-aware regularization terms that explicitly enforce invariance to corruptions in irrelevant modalities and sensitivity to perturbations in relevant modalities, thereby reducing unintended cross-modal interactions. To further mitigate over-reliance on textual priors, we incorporate a
language-prior debiasing penalty that discourages
hallucination-prone text-only responses. Extensive experiments across multiple audiovisual
hallucination benchmarks demonstrate that MoD-DPO consistently improves perception accuracy and
hallucination resistance, outperforming previous preference optimization baselines under similar training budgets. Our findings underscore the importance of modality-faithful
alignment and demonstrate a scalable path toward more reliable and resilient
multimodal foundation models.
Read the full abstract ▾
Computation and Language · Cross-listed from cs.SE
Type-Aware Retrieval-Augmented Generation with Dependency Closure for Solver-Executable Industrial Optimization Modeling
Y. Zhong, R. Huang, M. Wang et al. (+4 more)
Automated industrial optimization modeling requires reliable translation of natural-
language requirements into solver-executable code. However, large
language models often generate non-compilable models due to missing declarations, type inconsistencies, and incomplete dependency contexts. We propose a type-aware
retrieval-augmented generation (RAG) method that enforces modeling entity types and minimal dependency closure to ensure executability. Unlike existing RAG approaches that index unstructured text, our method constructs a domain-specific typed knowledge base by parsing heterogeneous sources, such as academic papers and solver code, into typed units and encoding their mathematical dependencies in a knowledge graph. Given a natural-
language instruction, it performs hybrid
retrieval and computes a minimal dependency-closed context, the smallest set of typed symbols required for solver-executable code, via dependency propagation over the graph. We validate the method on two constraint-intensive industrial cases: demand response optimization in battery production and flexible job shop scheduling. In the first case, our method generates an executable model incorporating demand-response incentives and load-reduction constraints, achieving peak shaving while preserving profitability; conventional RAG baselines fail. In the second case, it consistently produces compilable models that reach known optimal solutions, demonstrating robust cross-domain generalization; baselines fail entirely. Ablation studies confirm that enforcing type-aware dependency closure is essential for avoiding structural
hallucinations and ensuring executability, addressing a critical barrier to deploying large
language models in complex engineering optimization tasks.
Read the full abstract ▾
All authors Y. Zhong, R. Huang, M. Wang, Z. Guo, YC. Li, M. Yu, Z. Jin
Computation and Language
APRES: An Agentic Paper Revision and Evaluation System
Bingchen Zhao, Jenny Zhang, Chenxi Whitehouse et al. (+8 more)
Scientific discoveries must be communicated clearly to realize their full potential. Without effective communication, even the most groundbreaking findings risk being overlooked or misunderstood. The primary way scientists communicate their work and receive feedback from the community is through peer review. However, the current system often provides inconsistent feedback between reviewers, ultimately hindering the improvement of a manuscript and limiting its potential impact. In this paper, we introduce a novel method APRES powered by Large
Language Models (LLMs) to update a scientific papers text based on an evaluation rubric. Our automated method discovers a rubric that is highly predictive of future citation counts, and integrate it with APRES in an automated system that revises papers to enhance their quality and impact. Crucially, this objective should be met without altering the core scientific content. We demonstrate the success of APRES, which improves future citation prediction by 19.6% in mean averaged error over the next best baseline, and show that our paper revision process yields papers that are preferred over the originals by human expert evaluators 79% of the time. Our findings provide strong empirical support for using LLMs as a tool to help authors stress-test their manuscripts before submission. Ultimately, our work seeks to augment, not replace, the essential role of human expert reviewers, for it should be humans who discern which discoveries truly matter, guiding science toward advancing knowledge and enriching lives.
Read the full abstract ▾
All authors Bingchen Zhao, Jenny Zhang, Chenxi Whitehouse, Minqi Jiang, Michael Shvartsman, Abhishek Charnalia, Despoina Magka, Tatiana Shavrina, Derek Dunfield, Oisin Mac Aodha, Yoram Bachrach
Computation and Language
UniSkill: A Dataset for Matching University Curricula to Professional Competencies
Nurlan Musazade, Joszef Mezei and Mike Zhang
Skill extraction and recommendation systems have been studied from recruiter, applicant, and education perspectives. While AI applications in job advertisements have received broad
attention, deficiencies in the instructed skills side remain a challenge. In this work, we address the scarcity of publicly available
datasets by releasing both manually annotated and synthetic
datasets of skills from the European Skills, Competences, Qualifications and Occupations (ESCO) taxonomy and university course pairs and publishing corresponding annotation guidelines. Specifically, we match graduate-level university courses with skills from the Systems Analysts and Management and Organization Analyst ESCO occupation groups at two granularities: course title with a skill, and course sentence with a skill. We train
language models on this
dataset to serve as a baseline for
retrieval and recommendation systems for course-to-skill and skill-to-course matching. We evaluate the models on a portion of the annotated data. Our BERT model achieves 87% F1-score, showing that course and skill matching is a feasible task.
Read the full abstract ▾
Computation and Language
Evaluating Performance Drift from Model Switching in Multi-Turn LLM Systems
Raad Khraishi, Iman Zafar, Katie Myles et al. (+1 more)
Deployed multi-turn LLM systems routinely switch models mid-interaction due to upgrades, cross-provider routing, and fallbacks. Such handoffs create a context mismatch: the model generating later turns must condition on a dialogue prefix authored by a different model, potentially inducing silent performance drift. We introduce a switch-
matrix benchmark that measures this effect by running a prefix model for early turns and a suffix model for the final turn, and comparing against the no-switch baseline using paired episode-level bootstrap confidence intervals. Across CoQA conversational QA and Multi-IF
benchmarks, even a single-turn handoff yields prevalent and statistically significant, directional effects and may swing outcomes by -8 to +13 percentage points in Multi-IF strict success rate and +/- 4 absolute F1 on CoQA, comparable to the no-switch gap between common model tiers (e.g., GPT-5-nano vs GPT-5-mini). We further find systematic compatibility patterns: some suffix models degrade under nearly any non-self dialogue history, while others improve under nearly any foreign prefix. To enable compressed handoff risk monitoring, we decompose switch-induced drift into per-model prefix influence and suffix susceptibility terms, accounting for ~70% of variance across
benchmarks. These results position handoff robustness as an operational reliability dimension that single-model
benchmarks miss, motivating explicit monitoring and handoff-aware mitigation in multi-turn systems.
Read the full abstract ▾
All authors Raad Khraishi, Iman Zafar, Katie Myles, Greig A Cowan
Computation and Language · Cross-listed from eess.AS
Interpreting Speaker Characteristics in the Dimensions of Self-Supervised Speech Features
Kyle Janse van Rensburg, Benjamin van Niekerk and Herman Kamper
How do
speech models trained through self-supervised learning structure their representations? Previous studies have looked at how information is encoded in feature vectors across different layers. But few studies have considered whether
speech characteristics are captured within individual dimensions of SSL features. In this paper, we specifically look at speaker information using PCA on utterance-averaged representations. For a range of SSL models, we find that the principal dimension that explains most variance encodes pitch and associated characteristics like gender. Other individual principal dimensions correlate with intensity, noise levels, the second formant, and higher frequency characteristics. We then use synthesis analyses to show that the dimensions for most characteristics are isolated from each other's influence. We further show that characteristics can be changed by manipulating the corresponding dimensions.
Read the full abstract ▾
Computation and Language
Compact Prompting in Instruction-tuned LLMs for Joint Argumentative Component Detection
Sofiane Elguendouze, Erwan Hain, Elena Cabrio et al. (+1 more)
Argumentative component detection (ACD) is a core subtask of Argument(ation) Mining (AM) and one of its most challenging aspects, as it requires jointly delimiting argumentative spans and classifying them into components such as claims and premises. While research on this subtask remains relatively limited compared to other AM tasks, most existing approaches formulate it as a simplified sequence labeling problem, component classification, or a pipeline of component
segmentation followed by classification. In this paper, we propose a novel approach based on instruction-tuned Large
Language Models (LLMs) using compact instruction-based
prompts, and reframe ACD as a
language generation task, enabling arguments to be identified directly from plain text without relying on pre-segmented components. Experiments on standard
benchmarks show that our approach achieves higher performance compared to state-of-the-art systems. To the best of our knowledge, this is one of the first attempts to fully model ACD as a generative task, highlighting the potential of instruction tuning for complex AM problems.
Read the full abstract ▾
All authors Sofiane Elguendouze, Erwan Hain, Elena Cabrio, Serena Villata
Computation and Language
TAO-Attack: Toward Advanced Optimization-Based Jailbreak Attacks for Large Language Models
Zhi Xu, Jiaqi Li, Xiaotong Zhang et al. (+2 more)
Large
language models (LLMs) have achieved remarkable success across diverse applications but remain vulnerable to jailbreak attacks, where attackers craft
prompts that bypass safety
alignment and elicit unsafe responses. Among existing approaches, optimization-based attacks have shown strong effectiveness, yet current methods often suffer from frequent refusals, pseudo-harmful outputs, and inefficient
token-level updates. In this work, we propose TAO-Attack, a new optimization-based jailbreak method. TAO-Attack employs a two-stage loss function: the first stage suppresses refusals to ensure the model continues harmful prefixes, while the second stage penalizes pseudo-harmful outputs and encourages the model toward more harmful completions. In addition, we design a direction-priority
token optimization (DPTO) strategy that improves efficiency by aligning candidates with the gradient direction before considering update magnitude. Extensive experiments on multiple LLMs demonstrate that TAO-Attack consistently outperforms state-of-the-art methods, achieving higher attack success rates and even reaching 100% in certain scenarios.
Read the full abstract ▾
All authors Zhi Xu, Jiaqi Li, Xiaotong Zhang, Hong Yu, Han Liu