The ability of critique is vital for models to self-improve and serve as reliable AI assistants. While extensively studied in language-only settings, multimodal critique of Large Multimodal Models (LMMs) remains underexplored despite their growing capabilities in tasks like captioning and visual reasoning. In this work, we introduce MM-CRITIC, a holistic benchmark for evaluating the critique ability of LMMs across multiple dimensions: basic, correction, and comparison. Covering 8 main task types and over 500 tasks, MM-CRITIC collects responses from various LMMs with different model sizes and is composed of 4471 samples. To enhance the evaluation reliability, we integrate expert-informed ground answers into scoring rubrics that guide GPT-4o in annotating responses and generating reference critiques, which serve as anchors for trustworthy judgments. Extensive experiments validate the effectiveness of MM-CRITIC and provide a comprehensive assessment of leading LMMs' critique capabilities under multiple dimensions. Further analysis reveals some key insights, including the correlation between response quality and critique, and varying critique difficulty across evaluation dimensions. Our code is available at https://github.com/MichealZeng0420/MM-Critic.

Interactive preference elicitation (IPE) aims to substantially reduce human effort while acquiring human preferences in wide personalization systems. Dueling bandit (DB) algorithms enable optimal decision-making in IPE building on pairwise comparisons. However, they remain inefficient when human feedback is sparse. Existing methods address sparsity by heavily relying on parametric reward models, whose rigid assumptions are vulnerable to misspecification. In contrast, we explore an alternative perspective based on feedback augmentation, and introduce critical improvements to the model-free DB framework. Specifically, we introduce augmented confidence bounds to integrate augmented human feedback under generalized concentration properties, and analyze the multi-factored performance trade-off via regret analysis. Our prototype algorithm achieves competitive performance across several IPE benchmarks, including recommendation, multi-objective optimization, and response optimization for large language models, demonstrating the potential of our approach for provably efficient IPE in broader applications.

Abstract--Autonomous V ehicles (A Vs) are poised to revolutionize emergency services by enabling faster, safer, and more efficient responses. This transformation is driven by advances in Artificial Intelligence (AI), particularly Reinforcement Learning (RL), which allows A Vs to navigate complex environments and make critical decisions in real time. However, conventional RL paradigms often suffer from poor sample efficiency and lack adaptability in dynamic emergency scenarios. This paper reviews next-generation A V optimization strategies to address these limitations. We analyze the shift from conventional RL to Diffusion Model (DM)-augmented RL, which enhances policy robustness through synthetic data generation, albeit with increased computational cost. Additionally, we explore the emerging paradigm of Large Language Model (LLM)-assisted In-Context Learning (ICL), which offers a lightweight and interpretable alternative by enabling rapid, on-the-fly adaptation without retraining. By reviewing the state of the art in A V intelligence, DM-augmented RL, and LLM-assisted ICL, this paper provides a critical framework for understanding the next generation of autonomous emergency response systems from a Generative AI perspective. Autonomous vehicles (A Vs) are poised to transform emergency services by enabling faster, safer, and more intelligent responses. Uncrewed Aerial V ehicles (UA Vs), as key enablers within the A V ecosystem, provide rapid deployment and precise mobility. They can serve as both aerial base stations and data collectors, enhancing connectivity and information gathering for A V operations.

LLMs have achieved remarkable breakthroughs in reasoning, insights, and tool use, but chaining these abilities into extended processes at the scale of those routinely executed by humans, organizations, and societies has remained out of reach. The models have a persistent error rate that prevents scale-up: for instance, recent experiments in the Towers of Hanoi benchmark domain showed that the process inevitably becomes derailed after at most a few hundred steps. Thus, although LLM research is often still benchmarked on tasks with relatively few dependent logical steps, there is increasing attention on the ability (or inability) of LLMs to perform long range tasks. This paper describes MAKER, the first system that successfully solves a task with over one million LLM steps with zero errors, and, in principle, scales far beyond this level. The approach relies on an extreme decomposition of a task into subtasks, each of which can be tackled by focused microagents. The high level of modularity resulting from the decomposition allows error correction to be applied at each step through an efficient multi-agent voting scheme. This combination of extreme decomposition and error correction makes scaling possible. Thus, the results suggest that instead of relying on continual improvement of current LLMs, massively decomposed agentic processes (MDAPs) may provide a way to efficiently solve problems at the level of organizations and societies.

Large Language Models (LLMs) have impressive data fusion and reasoning capabilities for autonomous driving (AD). However, training LLMs for AD faces significant challenges including high computation transmission costs, and privacy concerns associated with sensitive driving data. Federated Learning (FL) is promising for enabling autonomous vehicles (AVs) to collaboratively train models without sharing raw data. We present Federated LLM-based Autonomous Driving (FLAD), an FL framework that leverages distributed multimodal sensory data across AVs in heterogeneous environment. FLAD has three key innovations: (1) a cloud-edge-vehicle collaborative architecture that reduces communication delay and preserving data privacy; (2) an intelligent parallelized collaborative training with a communication scheduling mechanism that optimizes training efficiency, leveraging end-devices otherwise having insufficient resources for model training; and (3) a knowledge distillation method that personalizes LLM according to heterogeneous edge data. In addition, we prototype FLAD in a testbed with NVIDIA Jetsons, overcoming practical implementation challenges including CPU/GPU memory sharing in resource-constrained devices, dynamic model partitions, and fault-tolerant training.Extensive experimental evaluation demonstrates that FLAD achieves superior end-to-end AD performance while efficiently utilizing distributed vehicular resources, opening up new possibilities for future collaborative AD model training and knowledge sharing.

Large Language Models perform well at natural language interpretation and reasoning, but their inherent stochasticity limits their adoption in regulated industries like finance and healthcare that operate under strict policies. To address this limitation, we present a two-stage neurosymbolic framework that (1) uses LLMs with optional human guidance to formalize natural language policies, allowing fine-grained control of the formalization process, and (2) uses inference-time autofor-malization to validate logical correctness of natural language statements against those policies. When correctness is paramount, we perform multiple redundant formalization steps at inference time, cross checking the formalizations for semantic equivalence. Our benchmarks demonstrate that our approach exceeds 99% soundness, indicating a near-zero false positive rate in identifying logical validity. Our approach produces auditable logical artifacts that substantiate the verification outcomes and can be used to improve the original text. The content generation and reasoning capabilities of Large Language Models (LLMs) continue to advance rapidly, demonstrating unprecedented improvements in coherence and analytical accuracy (Wei et al., 2022; Y ao et al., 2023; Lewis et al., 2021). Despite these advances, their probabilistic nature and tendency to generate plausible but incorrect information (hallucinations, cf. Xu et al. 2024b) remain barriers to widespread adoption in regulated sectors. Industries such as healthcare, financial services, and legal practices have legal and regulatory obligations for accuracy and auditability that current LLM technology has yet to meet (Haltaufderheide & Ranisch, 2024). Companies develop institutional policies to ensure compliance with applicable laws and regulations. Such policies are typically captured in natural language (NL) documents that define rules, procedures, or guidelines. A challenge thus emerges when organizations look to deploy LLMs to answer questions about such documents: can we develop guardrails to ensure that LLM outputs conform to institutional policies? Consider an airline implementing a chatbot to assist customer service representatives in navigating refund policies: if the chatbot incorrectly claims that a customer is eligible for a refund when they are not, this could lead to legal exposure and loss of customer trust. An effective guardrail would help representatives decide if they can rely on a chatbot response without spending additional human effort to verify it. The key concern would be to ensure that when the guardrail reports an answer is valid, it actually is.

Although Large Language Models (LLMs) show exceptional fluency, efforts persist to extract stronger reasoning capabilities from them. Drawing on search-based interpretations of LLM computation, this paper advances a systematic framework for understanding LLM reasoning and optimization. Namely, that enhancing reasoning is best achieved by structuring a multi-agent pipeline to ensure a traversal of the search space in a gradual, incremental, and sequential (GIS) manner. Stated succinctly, high-quality reasoning is a controlled, incremental search. To test this framework, we investigate the efficacy of recursive refinement (RR)--an iterative process of self-criticism, adversarial stress-testing, and integrating critical feedback--as a practical method for implementing GIS search. We designed an experiment comparing a simple, linear pipeline against a complex, explicitly structured pipeline leveraging a recursive refinement layer. The multi-agent models were constructed to reflect the historical personas of three US Founding Fathers (Hamilton, Jefferson, and Madison) using RAG-powered corpora and were prompted to generate responses to three contemporary political issues. Model performance was evaluated using a two-tiered approach: a quantitative score from an LLM arbiter agent and qualitative human judgment. Our results revealed that the complex model consistently outperformed the simple model across all nine test cases with an average arbiter-outputted score of 88.3 versus 71.7. The complex model's arguments were superior in analytical depth, structural nuance, and strategic framing. We conclude that recursive refinement is a robust architectural feature for enhancing LLM reasoning via GIS search.

Emotional support is a core capability in human-AI interaction, with applications including psychological counseling, role play, and companionship. However, existing evaluations of large language models (LLMs) often rely on short, static dialogues and fail to capture the dynamic and long-term nature of emotional support. To overcome this limitation, we shift from snapshot-based evaluation to trajectory-based assessment, adopting a user-centered perspective that evaluates models based on their ability to improve and stabilize user emotional states over time. Our framework constructs a large-scale benchmark consisting of 328 emotional contexts and 1,152 disturbance events, simulating realistic emotional shifts under evolving dialogue scenarios. To encourage psychologically grounded responses, we constrain model outputs using validated emotion regulation strategies such as situation selection and cognitive reappraisal. User emotional trajectories are modeled as a first-order Markov process, and we apply causally-adjusted emotion estimation to obtain unbiased emotional state tracking. Based on this framework, we introduce three trajectory-level metrics: Baseline Emotional Level (BEL), Emotional Trajectory Volatility (ETV), and Emotional Centroid Position (ECP). These metrics collectively capture user emotional dynamics over time and support comprehensive evaluation of long-term emotional support performance of LLMs. Extensive evaluations across a diverse set of LLMs reveal significant disparities in emotional support capabilities and provide actionable insights for model development.

Recent advances in large reasoning models have been driven by reinforcement learning and test-time scaling, accompanied by growing interest in latent rather than purely textual reasoning. However, existing latent reasoning methods lack mechanisms to ensure stable evolution of latent representations and a systematic way to interleave implicit and explicit reasoning. We introduce SpiralThinker, a unified framework that performs iterative updates over latent representations, enabling extended implicit reasoning without generating additional tokens. A progressive alignment objective combined with structured annotations maintains coherence between latent and textual reasoning. Across mathematical, logical, and commonsense reasoning tasks, SpiralThinker achieves the best overall performance among latent reasoning approaches, consistently surpassing previous methods across all benchmarks. Detailed analyses reveal that both iteration and alignment are indispensable, the numbers of latent tokens and iterations exhibit dataset-specific optima, and appropriate alignment proves critical for an effective iterative process. Overall, SpiralThinker bridges iterative computation and latent reasoning, demonstrating that aligned iterative updates can reliably steer reasoning in the latent space.

We present Bayesian Mixture of Experts (Bayesian-MoE), a post-hoc uncertainty estimation framework for fine-tuned large language models (LLMs) based on Mixture-of-Experts architectures. Our method applies a structured Laplace approximation to the second linear layer of each expert, enabling calibrated uncertainty estimation without modifying the original training procedure or introducing new parameters. Unlike prior approaches, which apply Bayesian inference to added adapter modules, Bayesian-MoE directly targets the expert pathways already present in MoE models, leveraging their modular design for tractable block-wise posterior estimation. We use Kronecker-factored low-rank approximations to model curvature and derive scalable estimates of predictive uncertainty and marginal likelihood. Experiments on common-sense reasoning benchmarks with Qwen1.5-MoE and DeepSeek-MoE demonstrate that Bayesian-MoE improves both expected calibration error (ECE) and negative log-likelihood (NLL) over baselines, confirming its effectiveness for reliable downstream decision-making.