Recent work has demonstrated state-of-the-art results in large language model (LLM) hallucination detection and mitigation through consistency-based approaches which involve aggregating multiple responses sampled from a single LLM for a given prompt. These approaches help offset limitations stemming from the imperfect data on which LLMs are trained, which includes biases and under-representation of information required at deployment time among other limitations which can lead to hallucinations. We show that extending these single-model consistency methods to combine responses from multiple LLMs with different training data, training schemes and model architectures can result in substantial further improvements in hallucination detection and mitigation capabilities beyond their single-model consistency counterparts. We evaluate this "consortium consistency" approach across many model teams from a pool of 15 LLMs and explore under what conditions it is beneficial to team together different LLMs in this manner. Further, we show that these performance improvements often come with reduced inference costs, offsetting a significant drawback with single-model consistency methods.

With the proliferation of the internet and the rapid advancement of Artificial Intelligence, leading technology companies face an urgent annual demand for a considerable number of software and algorithm engineers. To efficiently and effectively identify high-potential candidates from thousands of applicants, these firms have established a multi-stage selection process, which crucially includes a standardized hiring evaluation designed to assess job-specific competencies. Motivated by the demonstrated prowess of Large Language Models (LLMs) in coding and reasoning tasks, this paper investigates a critical question: Can LLMs successfully pass these hiring evaluations? To this end, we conduct a comprehensive examination of a widely used professional assessment questionnaire. We employ state-of-the-art LLMs to generate responses and subsequently evaluate their performance. Contrary to any prior expectation of LLMs being ideal engineers, our analysis reveals a significant inconsistency between the model-generated answers and the company-referenced solutions. Our empirical findings lead to a striking conclusion: All evaluated LLMs fails to pass the hiring evaluation.

Large language models (LLMs) have demonstrated impressive reasoning capabilities, but scaling their performance often relies on massive reasoning datasets that are computationally expensive to train on. Existing data selection methods aim to curate smaller, high-quality subsets but often rely on costly external models or opaque heuristics. In this work, we shift the focus from external heuristics to the model's internal mechanisms. We find that complex reasoning tasks consistently activate a sparse, specialized subset of attention heads, forming core reasoning circuits. Building on this insight, we propose CircuitSeer, a novel data selection method that quantifies the reasoning complexity of data by measuring its influence on these crucial circuits. Extensive experiments on 4 models and 9 datasets demonstrate CircuitSeer's superiority. Notably, fine-tuning Qwen2.5-Math-7B on just 10% of data selected by our method achieves a 1.4-point gain in average Pass@1 over training on the full dataset, highlighting its efficiency and effectiveness.

Integrating visual-language instructions into visuomotor policies is gaining momentum in robot learning for enhancing open-world generalization. Despite promising advances, existing approaches face two challenges: limited language steerability when no generated reasoning is used as a condition, or significant inference latency when reasoning is incorporated. In this work, we introduce MoTVLA, a mixture-of-transformers (MoT)-based vision-language-action (VLA) model that integrates fast-slow unified reasoning with behavior policy learning. MoTVLA preserves the general intelligence of pre-trained VLMs (serving as the generalist) for tasks such as perception, scene understanding, and semantic planning, while incorporating a domain expert, a second transformer that shares knowledge with the pretrained VLM, to generate domain-specific fast reasoning (e.g., robot motion decomposition), thereby improving policy execution efficiency. By conditioning the action expert on decomposed motion instructions, MoTVLA can learn diverse behaviors and substantially improve language steerability. Extensive evaluations across natural language processing benchmarks, robotic simulation environments, and real-world experiments confirm the superiority of MoTVLA in both fast-slow reasoning and manipulation task performance.

Humans do not just find mistakes after the fact -- we often catch them mid-stream because 'reflection' is tied to the goal and its constraints. Today's large language models produce reasoning tokens and 'reflective' text, but is it functionally equivalent with human reflective reasoning? Prior work on closed-ended tasks -- with clear, external 'correctness' signals -- can make 'reflection' look effective while masking limits in self-correction. We therefore test eight frontier models on a simple, real-world task that is open-ended yet rule-constrained, with auditable success criteria: to produce valid scientific test items, then revise after considering their own critique. First-pass performance is poor (often zero valid items out of 4 required; mean $\approx$ 1), and reflection yields only modest gains (also $\approx$ 1). Crucially, the second attempt frequently repeats the same violation of constraint, indicating 'corrective gains' arise largely from chance production of a valid item rather than error detection and principled, constraint-sensitive repair. Performance before and after reflection deteriorates as open-endedness increases, and models marketed for 'reasoning' show no advantage. Our results suggest that current LLM 'reflection' lacks functional evidence of the active, goal-driven monitoring that helps humans respect constraints even on a first pass. Until such mechanisms are instantiated in the model itself, reliable performance requires external structure that enforces constraints. Our code is available at: https://github.com/cruiseresearchgroup/LLM_ReflectionTest

Tabular prediction has traditionally relied on gradient-boosted decision trees and specialized deep learning models, which excel within tasks but provide limited interpretability and weak transfer across tables. Reasoning large language models (LLMs) promise cross-task adaptability with trans- parent reasoning traces, yet their potential has not been fully realized for tabular data. This paper presents TabR1, the first reasoning LLM for tabular prediction with multi-step reasoning. At its core is Permutation Relative Policy Optimization (PRPO), a simple yet efficient reinforcement learning method that encodes column-permutation invariance as a structural prior. By construct- ing multiple label-preserving permutations per sample and estimating advantages both within and across permutations, PRPO transforms sparse rewards into dense learning signals and improves generalization. With limited supervision, PRPO activates the reasoning ability of LLMs for tabular prediction, enhancing few-shot and zero-shot performance as well as interpretability. Comprehensive experiments demonstrate that TabR1 achieves performance comparable to strong baselines under full-supervision fine-tuning. In the zero-shot setting, TabR1 approaches the performance of strong baselines under the 32-shot setting. Moreover, TabR1 (8B) substantially outperforms much larger LLMs across various tasks, achieving up to 53.17% improvement over DeepSeek-R1 (685B).

Vision-Language-Action (VLA) models extend vision-language models to embodied control by mapping natural-language instructions and visual observations to robot actions. Despite their capabilities, VLA systems face significant challenges due to their massive computational and memory demands, which conflict with the constraints of edge platforms such as on-board mobile manipulators that require real-time performance. Addressing this tension has become a central focus of recent research. In light of the growing efforts toward more efficient and scalable VLA systems, this survey provides a systematic review of approaches for improving VLA efficiency, with an emphasis on reducing latency, memory footprint, and training and inference costs. We categorize existing solutions into four dimensions: model architecture, perception feature, action generation, and training/inference strategies, summarizing representative techniques within each category. Finally, we discuss future trends and open challenges, highlighting directions for advancing efficient embodied intelligence.

Transformer models have driven breakthroughs across various language tasks by their strong capability to learn rich contextual representations. Scaling them to improve representation, however, often demands substantial memory and compute costs, such as the Key-Value (KV) cache used during auto-regressive decoding. Skip connections offer a promising way to improve representation without bloating resource usage, yet most prior works either improve expressivity while leaving KV costs unchanged, or reduce memory at the cost of weaker representation. In this work, we propose SkipV1Former, a Transformer variant that uses skip connections from the first layer's Value heads to strengthen model representation and reduce KV cache. Specifically, from the second block onward, each layer reuses half of its Value heads from the very first layer, while computing the other half as usual-cutting Value projections and V cache by nearly 50 \%. Theoretically, we show that routing uncompressed first-layer Values into deeper layers restores information lost to compression and accelerates the model's implicit mesa-optimization-a key pattern of Transformer in auto-regressive tasks. Empirically, across different model scales, SkipV1Former delivers consistent reductions of approximately 25 \% in KV cache while improving perplexity relative to standard Multi-Head Attention (MHA) Transformers and some advanced variants. Moreover, we propose a recipe for uptraining existing MHA Transformer checkpoints to SkipV1Former with only 10-15\% additional compute. Finally, SkipV1Former can seamlessly combine advanced methods like Group-Query Attention and Multi-Latent Attention to achieve further KV cache savings and performance improvement. When combined with YOCO, it cuts KV cache size by nearly 50 \% while still improving performance.

Reinforcement Learning (RL) has become a compelling way to strengthen the multi step reasoning ability of Large Language Models (LLMs). However, prevalent RL paradigms still lean on sparse outcome-based rewards and limited exploration, which often drives LLMs toward repetitive and suboptimal reasoning patterns. In this paper, we study the central question of how to design exploration for LLM reasoning and introduce MERCI (Motivating Exploration in LLM Reasoning with Count-based Intrinsic Rewards), a novel RL algorithm that augments policy optimization with a principled intrinsic reward. Building on the idea of count-based exploration, MERCI leverages a lightweight Coin Flipping Network (CFN) to estimate the pseudo count and further epistemic uncertainty over reasoning trajectories, and converts them into an intrinsic reward that values novelty while preserving the learning signal from task rewards. We integrate MERCI into some advanced RL frameworks like Group Relative Policy Optimization (GRPO). Experiments on complex reasoning benchmarks demonstrate that MERCI encourages richer and more varied chains of thought, significantly improves performance over strong baselines, and helps the policy escape local routines to discover better solutions. It indicates that our targeted intrinsic motivation can make exploration reliable for language model reasoning.

Most existing underwater instance segmentation approaches are constrained by close-vocabulary prediction, limiting their ability to recognize novel marine categories. To support evaluation, we introduce \textbf{MARIS} (\underline{Mar}ine Open-Vocabulary \underline{I}nstance \underline{S}egmentation), the first large-scale fine-grained benchmark for underwater Open-Vocabulary (OV) segmentation, featuring a limited set of seen categories and diverse unseen categories. Although OV segmentation has shown promise on natural images, our analysis reveals that transfer to underwater scenes suffers from severe visual degradation (e.g., color attenuation) and semantic misalignment caused by lack underwater class definitions. To address these issues, we propose a unified framework with two complementary components. The Geometric Prior Enhancement Module (\textbf{GPEM}) leverages stable part-level and structural cues to maintain object consistency under degraded visual conditions. The Semantic Alignment Injection Mechanism (\textbf{SAIM}) enriches language embeddings with domain-specific priors, mitigating semantic ambiguity and improving recognition of unseen categories. Experiments show that our framework consistently outperforms existing OV baselines both In-Domain and Cross-Domain setting on MARIS, establishing a strong foundation for future underwater perception research.