Food assistance referral requires conversational agents to translate underspecified, often noisy help-seeking dialogues into locally valid resource recommendations. We present Food4All, an agentic food-resource referral framework and benchmark grounded in 686 structured Indiana food resources. Food4All couples a food-specific search tool with 300 multi-turn evaluation tasks spanning single food needs, composite cases with access or document constraints, and five non-ideal user interaction traits: unreasonable demands, rambling responses, impatience, incomplete answers, and inconsistent information. We evaluate six Large Language Models (LLMs) on requirement grounding, resource retrieval, final referral correctness, and interaction efficiency. Although the strongest model achieves 96.33% referral accuracy, our diagnostics reveal persistent failures in grounding schedule, eligibility, intake, and document constraints, as well as failures to preserve valid retrieved resources in the final recommendation. Trait-level analysis further shows that different non-ideal behaviors stress different parts of the referral pipeline. Food4All provides a controlled testbed for studying tool-calling agents in constraint-sensitive food assistance referral under realistic user interaction challenges.

Language models are increasingly capable and are being rapidly deployed on a population-level scale. As a result, the safety of these models is increasingly high-stakes. Fortunately, advances in alignment have significantly reduced the likelihood of harmful model outputs. However, when models are queried billions of times in a day, even rare worst-case behaviors will occur. Current safety evaluations focus on capturing the distribution of inputs that yield harmful outputs. These evaluations disregard the probabilistic nature of models and their tail output behavior. To measure this tail risk, we propose a method to efficiently estimate the probability of harmful outputs for any input query. Instead of naive brute-force sampling from the target model, where harmful outputs could be rare, we operationalize importance sampling by creating unsafe versions of the target model. These unsafe versions enable sample-efficient estimation by making harmful outputs more probable. On benchmarks measuring misuse and misalignment, these estimates match brute-force Monte Carlo estimates using 10-20x fewer samples. For example, we can estimate probability of harmful outputs on the order of 10^-4 with just 500 samples. Additionally, we find that these harmfulness estimates can reveal the sensitivity of models to perturbations in model input and predict deployment risks. Our work demonstrates that accurate rare-event estimation is both critical and feasible for safety evaluations. Code is available at https://github.com/rangell/LMTailRisk

Effective contact-rich manipulation requires robots to synergistically leverage vision, force, and proprioception. However, Reinforcement Learning agents struggle to learn in such multisensory settings, especially amidst sensory noise and dynamic changes. We propose MultiSensory Dynamic Pretraining (MSDP), a novel framework for learning expressive multisensory representations tailored for task-oriented policy learning. MSDP is based on masked autoencoding and trains a transformer-based encoder by reconstructing multisensory observations from only a subset of sensor embeddings, leading to cross-modal prediction and sensor fusion. For downstream policy learning, we introduce a novel asymmetric architecture, where a cross-attention mechanism allows the critic to extract dynamic, task-specific features from the frozen embeddings, while the actor receives a stable pooled representation to guide its actions. Our method demonstrates accelerated learning and robust performance under diverse perturbations, including sensor noise, and changes in object dynamics. Evaluations in multiple challenging, contact-rich robot manipulation tasks in simulation and the real world showcase the effectiveness of MSDP. Our approach exhibits strong robustness to perturbations and achieves high success rates on the real robot with as few as 6,000 online interactions, offering a simple yet powerful solution for complex multisensory robotic control. Website: https://msdp-pearl.github.io/

Language Models (LLMs) have emerged as powerful reasoning engines for embodied control. In particular, In-Context Learning (ICL) enables off-the-shelf, text-only LLMs to predict robot actions without any task-specific training while preserving their generalization capabilities. Applying ICL to bimanual manipulation remains challenging as the high-dimensional joint action space and tight inter-arm coordination constraints rapidly overwhelm standard context windows. To address this, we introduce BiCICLe (Bimanual Coordinated In-Context Learning), the first framework that enables standard LLMs to perform few-shot bimanual manipulation without fine-tuning. BiCICLe frames bimanual control as a multi-agent leader-follower problem, decoupling the action space into sequential, conditioned single-arm predictions. Evaluated on 13 tasks from the TWIN benchmark, BiCICLe achieves 70.5% average success rate, outperforming the best training-free baseline by 6.1 percentage points and surpassing most supervised methods. We also demonstrate superior real-world performance on 3 tasks without hardware-specific retraining.

Speech translation for low-resource languages remains fundamentally limited by the scarcity of high-quality, diverse parallel speech data, a challenge that is especially pronounced in African linguistic contexts. To address this, we introduce NaijaS2ST, a parallel speech translation dataset spanning Igbo, Hausa, Yorùbá, and Nigerian Pidgin paired with English. The dataset comprises approximately 50 hours of speech per language and captures substantial variation in speakers and accents, reflecting realistic multilingual and multi-accent conditions. With NaijaS2ST, we conduct a comprehensive benchmark of cascaded, end-to-end (E2E), and AudioLLM-based approaches across bidirectional translation settings. Our results show that audio LLMs with few-shot examples are more effective for speech-to-text translation than cascaded and end-to-end methods trained on fine-tuned data. However, for speech-to-speech translation, the cascaded and audio LLM paradigms yield comparable performance, indicating that there is still considerable room for improvement in developing targeted, task-specific models for this setting. By providing both a high-quality dataset and a systematic benchmark, we hope that NaijaS2ST will serve as a strong foundation for advancing research in low-resource, multilingual speech translation.

Reinforcement learning (RL) enables high-frequency, closed-loop control for robotic manipulation, but scaling to long-horizon tasks with sparse or imperfect rewards remains difficult due to inefficient exploration and poor credit assignment. Vision-Language-Action (VLA) models leverage large-scale multimodal pretraining to provide generalist, task-level reasoning, but current limitations hinder their direct use in fast and precise manipulation. In this paper, we propose Vision-Language-Action Jump-Starting (VLAJS), a method that bridges sparse VLA guidance with on-policy RL to improve exploration and learning efficiency. VLAJS treats VLAs as transient sources of high-level action suggestions that bias early exploration and improve credit assignment, while preserving the high-frequency, state-based control of RL. Our approach augments Proximal Policy Optimization (PPO) with a directional action-consistency regularization that softly aligns the RL agent's actions with VLA guidance during early training, without enforcing strict imitation, requiring demonstrations, or relying on continuous teacher queries. VLA guidance is applied sparsely and annealed over time, allowing the agent to adapt online and ultimately surpass the guiding policy. We evaluate VLAJS on six challenging manipulation tasks: lifting, pick-and-place, peg reorientation, peg insertion, poking, and pushing in simulation, and validate a subset on a real Franka Panda robot. VLAJS consistently outperforms PPO and distillation-style baselines in sample efficiency, reducing required environment interactions by over 50% in several tasks. Real-world experiments demonstrate zero-shot sim-to-real transfer and robust execution under clutter, object variation, and external perturbations.

The robustness of machine learning models can be compromised by spurious correlations between non-causal features in the input data and target labels. A common way to test for such correlations is to train on data where the label is strongly tied to some non-causal cue, then evaluate on examples where that tie no longer holds. This idea is well established for classification tasks, but for semantic segmentation the specific failure modes are not well understood. We show that a model may achieve reasonable overlap while assigning the wrong semantic label, swapping one plausible foreground class for another, even when object boundaries are largely correct. We focus on this semantic label-flip behaviour and quantify it with a simple diagnostic (Flip) that counts how often ground truth foreground pixels are assigned the wrong foreground identity while remaining predicted as foreground. In a setting where category and scene are correlated during training, increasing the correlation consistently widens the gap between common and rare test conditions and increases these within-object label swaps on counterfactual groups. Overall, our results motivate assessing segmentation robustness under distribution shift beyond overlap by decomposing foreground errors into correct pixels, flipped-identity pixels, and missed-to-background pixels. We also propose an entropy-based, ground truth label-free `flip-risk' score, which is computed from foreground identity uncertainty, and show that it can flag flip-prone cases at inference time. Code is available at https://github.com/acharaakshit/label-flips.

Despite recent progress in text-prompt-based medical image segmentation, existing LISA-like MLLM-based methods typically generate masks regardless of whether the target specified in the query is present, leading to hallucinated segmentation. In this work, we propose MedVeriSeg, a training-free query verification framework that enables LISA-like medical segmentation models to reject false segmentation queries. MedVeriSeg first quantifies the response quality between the [SEG] token and image features through a Similarity Response Quality Scoring Module. To further improve robustness, it employs a Lightweight Routed Multi-Agent Verification Module, which fuses quantitative score evidence with qualitative agent evidence to comprehensively verify the validity of the query. To support systematic evaluation, we construct MedVeriSeg-Bench, a benchmark designed for query verification in medical image segmentation. Experimental results demonstrate that MedVeriSeg effectively identifies false segmentation queries and reduces hallucinated segmentation, while maintaining a high acceptance rate for valid queries, thereby largely preserving the segmentation utility of LISA-like medical segmentation models.

Executing a multi-agent plan can be challenging when an agent is delayed, because this typically creates conflicts with other agents. So, we need to quickly find a new safe plan. Replanning only the delayed agent often does not yield an efficient plan, and sometimes cannot even yield a feasible one. On the other hand, replanning other agents may lead to a cascade of changes and delays, and it is computationally expensive. We show how to efficiently replan a single delayed agent by tracking and using the temporal flexibility of other agents while avoiding cascading delays. This flexibility is the maximum delay that the agent can take without changing the order with agents other than the initially delayed agent, or further delaying other agents. Our algorithm, FlexSIPP, precomputes all possible plans for the delayed agent and returns the changes to the other agents within the given scenario. We demonstrate our method in a real-world case study of replanning trains in the densely-used Dutch railway network and in the MovingAI MAPF benchmark set. Our experiments show that FlexSIPP provides effective solutions relevant to real-world adjustments, and within a reasonable timeframe.

Fairness in human-robot interaction critically depends on the reliability of the perceptual models that enable robots to interpret human behavior. While demographic biases have been widely studied in high-level facial analysis tasks, their presence in facial landmark detection remains unexplored. In this paper, we conduct a systematic audit of demographic bias in this task, analyzing the age, gender, and race biases. To this end, we introduce a controlled statistical methodology to disentangle demographic effects from confounding visual factors. Our analysis demonstrates that visual confounders, particularly head pose and face resolution, heavily outweigh the impact of demographic attributes. Notably, after accounting for these confounders, performance disparities across gender and race vanish. However, we identify a statistically significant age-related bias, with higher localization errors for older individuals. This shows that fairness issues can emerge even in low-level vision components and can propagate through the HRI pipeline. We argue that auditing and correcting such biases is a necessary step toward trustworthy and equitable robot perception systems.

Computational models have emerged as powerful tools for multi-scale energy modeling research at the building and urban scale, supporting data-driven analysis across building and urban energy systems. However, these models require large amounts of building parameter data that is often inaccessible, expensive to collect, or subject to privacy constraints. We introduce a modular, multimodal generative Artificial Intelligence (AI) framework that integrates image, tabular, and simulation-based components and produces synthetic residential building datasets from publicly available county records and images, and present an end-to-end pipeline instantiating this framework. To reduce typical Large Language Model (LLM) challenges, we evaluate our model's components using occlusion-based visual focus analysis. Our analysis demonstrates that our selected vision-language model achieves greater visual focus than a GPT-based alternative for building image processing. We also assess realism of our results against a national reference dataset, finding that our synthetic data overlaps more than 95% for three of the four selected variables. This work reduces dependence on costly or restricted data sources, lowering barriers to building-scale energy research and Machine Learning (ML)-driven urban energy modeling, and therefore enabling scalable downstream tasks such as energy modeling, retrofit analysis, and urban-scale simulation under data scarcity.

Speech language models (SLMs) are systems of systems: independent components that unite to achieve a common goal. Despite their heterogeneous nature, SLMs are often studied end-to-end; how information flows through the pipeline remains obscure. We investigate this question through the lens of backdoor attacks. We first establish that backdoors can propagate through the SLM, leaving all tasks highly vulnerable. From this, we design a component analysis to discover the role each component takes in backdoor learning. We find that backdoor persistence or erasure is highly dependent on the targeted component. Beyond propagation, we examine how backdoors are encoded in shared multitask embeddings, showing that poisoned samples are not directly separable from benign ones, challenging a common separability assumption used in filtering defenses. Our findings emphasize the need to treat multimodal pipelines as intricate systems with unique vulnerabilities, not solely extensions of unimodal ones.

Many strategic decision-making problems, such as environment design for warehouse robots, can be naturally formulated as bi-level reinforcement learning (RL), where a leader agent optimizes its objective while a follower solves a Markov decision process (MDP) conditioned on the leader's decisions. In many situations, a fundamental challenge arises when the leader cannot intervene in the follower's optimization process; it can only observe the optimization outcome. We address this decentralized setting by deriving the hypergradient of the leader's objective, i.e., the gradient of the leader's strategy that accounts for changes in the follower's optimal policy. Unlike prior hypergradient-based methods that require extensive data for repeated state visits or rely on gradient estimators whose complexity can increase substantially with the high-dimensional leader's decision space, we leverage the Boltzmann covariance trick to derive an alternative hypergradient formulation. This enables efficient hypergradient estimation solely from interaction samples, even when the leader's decision space is high-dimensional. Additionally, to our knowledge, this is the first method that enables hypergradient-based optimization for 2-player Markov games in decentralized settings. Experiments highlight the impact of hypergradient updates and demonstrate our method's effectiveness in both discrete and continuous state tasks.

Retrieval-Augmented Generation (RAG) improves large language model applications by grounding generation in retrieved evidence, but also introduces corpus poisoning as a new attack surface. In this setting, an adversary injects or edits passages so that they enter the Top-$K$ results for target queries and influence downstream generation. Existing defences often rely on content filtering, auxiliary models, or generator-side reasoning, which complicates deployment. We propose ProGRank, a post hoc, training-free retriever-side defence for dense-retriever RAG. ProGRank stress-tests each query--passage pair under mild randomized perturbations, extracts probe gradients from a small fixed parameter subset, and derives two instability signals: representational consistency and dispersion risk. It then combines these signals with a score gate for reranking. ProGRank preserves the original passage content, requires no retraining, and supports a surrogate-based variant when the deployed retriever is unavailable. Experiments across datasets, retrievers, attacks, and retrieval-stage and end-to-end settings show that ProGRank improves robustness and maintains a favorable robustness--utility trade-off, including under adaptive evasive attacks.

Benchmarks like MMLU suggest flagship language models approach factuality saturation above 90\%. \emph{LLMpedia} shows this picture is incomplete. We materialize ${\sim}$1.3M encyclopedia articles entirely from parametric memory across three model families, then audit every claim against Wikipedia and curated web evidence. For \texttt{gpt-5-mini}, the verifiable true rate is 68.4\% on Wikipedia-covered subjects - more than 21\,pp below MMLU - and the gap is driven by \emph{unverifiability} (30.5\%), not refutation (1.2\%). Beyond Wikipedia, frontier articles audited against curated web evidence reach 57.6\%; Wikipedia covers only 56.7\% of model-surfaced subjects, and three model families overlap in just 7.3\% of subject choices. In a retrieval-trap benchmark inspired by prior analysis of Grokipedia, LLMpedia is more factual at roughly half the textual similarity to Wikipedia. Every prompt, article, and verdict is released. Data, code, interface: https://llmpedia.net.

Recent advances in diffusion models have significantly elevated the visual fidelity of Virtual Try-On (VTON) systems, yet reliable evaluation remains a persistent bottleneck. Traditional metrics struggle to quantify fine-grained texture details and semantic consistency, while existing datasets fail to meet commercial standards in scale and diversity. We present OpenVTON-Bench, a large-scale benchmark comprising approximately 100K high-resolution image pairs (up to $1536 \times 1536$). The dataset is constructed using DINOv3-based hierarchical clustering for semantically balanced sampling and Gemini-powered dense captioning, ensuring a uniform distribution across 20 fine-grained garment categories. To support reliable evaluation, we propose a multi-modal protocol that measures VTON quality along five interpretable dimensions: background consistency, identity fidelity, texture fidelity, shape plausibility, and overall realism. The protocol integrates VLM-based semantic reasoning with a novel Multi-Scale Representation Metric based on SAM3 segmentation and morphological erosion, enabling the separation of boundary alignment errors from internal texture artifacts. Experimental results show strong agreement with human judgments (Kendall's $τ$ of 0.833 vs. 0.611 for SSIM), establishing a robust benchmark for VTON evaluation.

Composed Video Retrieval (CoVR) aims to find a target video given a reference video and a textual modification. Prior work assumes the modification text fully specifies the visual changes, overlooking after-effects and implicit consequences (e.g., motion, state transitions, viewpoint or duration cues) that emerge from the edit. We argue that successful CoVR requires reasoning about these after-effects. We introduce a reasoning-first, zero-shot approach that leverages large multimodal models to (i) infer causal and temporal consequences implied by the edit, and (ii) align the resulting reasoned queries to candidate videos without task-specific finetuning. To evaluate reasoning in CoVR, we also propose CoVR-Reason, a benchmark that pairs each (reference, edit, target) triplet with structured internal reasoning traces and challenging distractors that require predicting after-effects rather than keyword matching. Experiments show that our zero-shot method outperforms strong retrieval baselines on recall at K and particularly excels on implicit-effect subsets. Our automatic and human analysis confirm higher step consistency and effect factuality in our retrieved results. Our findings show that incorporating reasoning into general-purpose multimodal models enables effective CoVR by explicitly accounting for causal and temporal after-effects. This reduces dependence on task-specific supervision, improves generalization to challenging implicit-effect cases, and enhances interpretability of retrieval outcomes. These results point toward a scalable and principled framework for explainable video search. The model, code, and benchmark are available at https://github.com/mbzuai-oryx/CoVR-R.

We present ReMoT, a unified training paradigm to systematically address the fundamental shortcomings of VLMs in spatio-temporal consistency -- a critical failure point in navigation, robotics, and autonomous driving. ReMoT integrates two core components: (1) A rule-based automatic framework that generates ReMoT-16K, a large-scale (16.5K triplets) motion-contrast dataset derived from video meta-annotations, surpassing costly manual or model-based generation. (2) Group Relative Policy Optimization, which we empirically validate yields optimal performance and data efficiency for learning this contrastive reasoning, far exceeding standard Supervised Fine-Tuning. We also construct the first benchmark for fine-grained motion contrast triplets to measure a VLM's discrimination of subtle motion attributes (e.g., opposing directions). The resulting model achieves state-of-the-art performance on our new benchmark and multiple standard VLM benchmarks, culminating in a remarkable 25.1% performance leap on spatio-temporal reasoning tasks.

Background. Biomedical language models should improve performance on biomedical text while retaining general-language-modeling fluency. For Mamba-based models, this trade-off has not been systematically studied across biomedical literature and clinical text. Methods. We developed BioMamba, a family of biomedical Mamba2 models at five scales obtained by continued pretraining of released public Mamba2 checkpoints on a balanced 80%/10%/10% mixture of PubMed abstracts, the Colossal Clean Crawled Corpus (C4), and Wikipedia. The contribution is the adaptation recipe and the accompanying open-weight checkpoints. Results. Across five scales, BioMamba consistently lowered PubMed perplexity, improved Wikipedia-style held-out perplexity by 1.46-4.72 PPL, and left C4 perplexity essentially unchanged. On six out-of-domain multiple-choice benchmarks, BioMamba stayed within +/-3 percentage points of Mamba2 with no systematic regression. After supervised fine-tuning, BioMamba+SFT matched or exceeded Mamba2+SFT on MIMIC-IV note completion and discharge summary generation at every evaluated scale, and improved PubMedQA at every scale. The strongest model (BioMamba-2.7B) reached a PubMed perplexity of 5.28 and accuracies of 90.24% and 73.00% on BioASQ and PubMedQA, respectively. Conclusions. A balanced domain-adaptive continued pretraining recipe strengthens Mamba2 language models on biomedical literature and clinical text while preserving general-language-modeling fluency.

Multi-robot formation control has various applications in domains such as vehicle troops, platoons, payload transportation, and surveillance. Maintaining formation in a vehicle platoon requires designing a suitable control scheme that can tackle external disturbances and uncertain system parameters while maintaining a predefined safe distance between the robots. A crucial challenge in this context is dealing with the unknown/uncertain friction forces between wheels and the ground, which vary with changes in road surface, wear in tires, and speed of the vehicle. Although state-of-the-art adaptive controllers can handle a priori bounded uncertainties, they struggle with accurately modeling and identifying frictional forces, which are often state-dependent and cannot be a priori bounded. This thesis proposes a new adaptive sliding mode controller for wheeled mobile robot-based vehicle platoons that can handle the unknown and complex behavior of frictional forces without prior knowledge of their parameters and structures. The controller uses the adaptive sliding mode control techniques to regulate the platoon's speed and maintain a predefined inter-robot distance, even in the presence of external disturbances and uncertain system parameters. This approach involves a two-stage process: first, the kinematic controller calculates the desired velocities based on the desired trajectory; and second, the dynamics model generates the commands to achieve the desired motion. By separating the kinematics and dynamics of the robot, this approach can simplify the control problem and allow for more efficient and robust control of the wheeled mobile robot.

Today's autonomous agents, largely driven by foundation models (FMs), can understand natural language instructions and solve long-horizon tasks with human-like reasoning. However, current human-robot interaction frameworks largely follow a one-way master-apprentice technique where the embodied agent passively executes commands without reciprocal learning. This neglects the co-adaptive, multi-turn nature of everyday human-to-human interactions. We introduce symbiotic interactive learning (SIL), a bidirectional co-adaptation framework in a shared latent task space, where both the human and the agent maintain joint belief states that evolve with the interaction history. This enables proactive clarification, adaptive suggestions, and shared plan refinement. SIL leverages FMs for spatial perception and reasoning, together with a triplet-loss-trained neural encoder that grounds the FMs' outputs into task-specific latent representations. To support long-term stability as tasks evolve, SIL utilises episodic and semantic memory architectures, regularised via elastic weight consolidation to mitigate catastrophic forgetting. We evaluate SIL on simulated and real-world embodied tasks, including instruction following, information retrieval, query-oriented reasoning, and interactive dialogue, achieving a $90.4\%$ task completion rate and a belief alignment score of $ρ\approx 0.83$, an absolute improvement of about $20$ percentage points over the best ablations. Demos and resources: https://linusnep.github.io/SIL/.

Recent advances in large multimodal models (LMMs) have enabled impressive reasoning and perception abilities, yet most existing training pipelines still depend on human-curated data or externally verified reward models, limiting their autonomy and scalability. In this work, we strive to improve LMM reasoning capabilities in a purely unsupervised fashion (without any annotated data or reward distillation). To this end, we propose a self-evolving framework, named EvoLMM, that instantiates two cooperative agents from a single backbone model: a Proposer, which generates diverse, image-grounded questions, and a Solver, which solves them through internal consistency, where learning proceeds through a continuous self-rewarding process. This dynamic feedback encourages both the generation of informative queries and the refinement of structured reasoning without relying on ground-truth or human judgments. When using the popular Qwen2.5-VL as the base model, our EvoLMM yields consistent gains upto $\sim$3\% on multimodal math-reasoning benchmarks, including ChartQA, MathVista, and MathVision, using only raw training images. We hope our simple yet effective approach will serve as a solid baseline easing future research in self-improving LMMs in a fully-unsupervised fashion. Our code and models are available at https://github.com/mbzuai-oryx/EvoLMM.

Text-to-image generation models have advanced rapidly, yet achieving fine-grained control over generated images remains difficult, largely due to limited understanding of how semantic information is encoded. We develop an interpretation of the color representation in the Variational Autoencoder latent space of FLUX.1 [Dev], revealing a structure reflecting Hue, Saturation, and Lightness. We verify our Latent Color Subspace (LCS) interpretation by demonstrating that it can both predict and explicitly control color, introducing a fully training-free method in FLUX based solely on closed-form latent-space manipulation. Code is available at https://github.com/ExplainableML/LCS.

Visual instruction tuning is crucial for improving vision-language large models (VLLMs). However, many samples can be solved via linguistic patterns or common-sense shortcuts, without genuine cross-modal reasoning, limiting the effectiveness of multimodal learning. Prior data selection methods often rely on costly proxy model training and focus on difficulty or diversity, failing to capture a sample's true contribution to vision-language joint reasoning. In this paper, we propose CVS, a training-free data selection method based on the insight that, for high-quality multimodal samples, introducing the question should substantially alter the model's assessment of answer validity given an image. CVS leverages a frozen VLLM as an evaluator and measures the discrepancy in answer validity with and without conditioning on the question, enabling the identification of samples that require vision-language joint reasoning while filtering semantic-conflict noise. Experiments on Vision-Flan and The Cauldron show that CVS achieves solid performance across datasets. On Vision-Flan, CVS outperforms full-data training by 3.5% and 4.8% using only 10% and 15% of the data, respectively, and remains robust on the highly heterogeneous Cauldron dataset. Moreover, CVS reduces computational cost by 17.3% and 44.4% compared to COINCIDE and XMAS.

High-throughput Mamba-2 inference is usually tied to fused CUDA and Triton kernels, limiting portability across accelerator backends. We show that the state space duality (SSD) recurrence has a compiler-friendly structure: diagonal per-head dynamics, fixed-size chunking, einsum-dominated compute, and static control flow. Expressing this structure in standard JAX primitives gives a single-source inference path with no custom kernels, a registered JAX PyTree cache, and a compiled on-device autoregressive loop. On a single Google Cloud TPU v6e, batch-1 prefill reaches approximately 140 TFLOPS, or 15% model FLOP utilisation (MFU), the roofline ceiling for this regime, and cached decode reaches up to 64% hardware bandwidth utilisation (HBU). At a 4096-token context, cached decode is 27x--36x faster than full-prefix recomputation across five Mamba-2 checkpoints from 130M to 2.7B parameters. The same source runs unmodified on NVIDIA L40S, where cached decode remains sequence-length independent across all model scales. WikiText-103 validation perplexity matches the Triton reference mamba_ssm v2.2.2 within +/-0.0005 points, and hidden states agree to float32 rounding tolerance. Code is available at https://github.com/CosmoNaught/mamba2-jax.

Large language models (LLMs) can answer religious knowledge queries fluently, yet they often hallucinate and misattribute sources, which is especially consequential in Islamic settings where users expect grounding in canonical texts (Qur'an and Hadith) and jurisprudential (fiqh) nuance. Retrieval-augmented generation (RAG) improves grounding, however, a single retrieve-then-generate pipeline is insufficient for diverse Islamic queries, including verbatim scripture, citation-grounded guidance, and rule-constrained computations such as zakat and inheritance. To address these challenges, we present Fanar-Sadiq, a bilingual Arabic-English Islamic QA system built on a multi-agent, tool-augmented architecture. It is a core component of the Fanar AI platform. Fanar-Sadiq routes Islamic queries to specialized modules within an agentic tool architecture. It supports intent-aware routing, retrieval-grounded fiqh answers with normalized citations and verification traces, exact verse lookup with quotation validation, and deterministic Sunni zakat and inheritance calculators with madhhab-sensitive branching. We evaluate the end-to-end system on public Islamic QA benchmarks and show strong effectiveness and efficiency. It is publicly accessible through an API and Web application and has received over 1.9M accesses in less than a year (https://api.fanar.qa/docs).

Multilingual large language models (LLMs) are increasingly used in socially sensitive mental health contexts, including support chatbots, screening, and content moderation. This raises a reliability question: do semantically equivalent mental health inputs elicit comparable evaluations across languages, or systematic shifts consistent with language-associated social and cultural contexts? We examine this question in an English-Chinese setting with GPT-4o and Qwen3-32B using a two-level framework: construct-level evaluative orientation, measured by psychometric stigma instruments, and decision-level behavior, measured by binary stigma detection and four-class depression severity classification. Across instruments and models, Chinese prompts elicit higher stigma-related scores than English prompts. At the decision level, Chinese prompts reduce sensitivity to stigmatizing content and produce more conservative depression severity judgments, leading to more under-estimation errors. These findings show that prompt language can shift both evaluative orientation and downstream behavior in LLM-based mental health evaluation. They highlight the need to evaluate multilingual LLMs not only for aggregate performance, but also for whether they apply comparable evaluative standards across languages in socially sensitive domains.

Scalable sequence models, such as Transformer variants and structured state-space models, often trade expressivity power for sequence-level parallelism, which enables efficient training. Here we examine the bounds on error and how error scales when models operate outside of their expressivity regimes using a Lie-algebraic control perspective. Our theory formulates a correspondence between the depth of a sequence model and the tower of Lie algebra extensions. Echoing recent theoretical studies, we characterize the Lie-algebraic class of constant-depth sequence models and their corresponding expressivity bounds. Furthermore, we analytically derive an approximation error bound and show that error diminishes exponentially as the depth increases, consistent with the strong empirical performance of these models. We validate our theoretical predictions using experiments on symbolic word and continuous-valued state-tracking problems.

Intermediate token generation (ITG), where a model produces output before the solution, has become a standard method to improve the performance of language models on reasoning tasks. These intermediate tokens have been called \say{reasoning traces} or even \say{thinking traces} -- implicitly anthropomorphizing the traces, and implying that these traces resemble steps a human might take when solving a challenging problem, and as such can provide an interpretable window into the operation of the model's thinking process to the end user. In this position paper, we present evidence that this anthropomorphization isn't a harmless metaphor, and instead is quite dangerous -- it confuses the nature of these models and how to use them effectively, and leads to questionable research. We call on the community to avoid such anthropomorphization of intermediate tokens.

We present SAM, a State-space Audio-language Model that integrates an audio encoder with a Mamba-2 backbone. SAM-2.7B achieves 21.1 mAP on AudioSet and 17.6 SPICE on AudioCaps, matching or surpassing larger 7B transformer-based models with fewer parameters. We further provide the first systematic, representation-level analysis of how SSMs interact with audio encoder outputs: (1) joint audio encoder finetuning is essential, supported by accuracy gains and observed adaptation of token representation rank and similarity across different SSM sizes; (2) despite linear scaling, SSMs benefit more from compact, information-rich audio token representations than from excessively long token sequences; and (3) incorporating instruction-following supervision substantially improves reasoning ability, boosting MMAU-Sound accuracy from 22.8 to 56.8. Through comprehensive experiments and analysis, we establish practical design principles for SSMs as strong, scalable backbones for audio-language models.