LLM-as-a-judge has become a promising paradigm for using large language models (LLMs) to evaluate natural language generation (NLG), but the uncertainty of its evaluation remains underexplored. This lack of reliability may limit its deployment in many applications. This work presents the first framework to analyze the uncertainty by offering a prediction interval of LLM-based scoring via conformal prediction. Conformal prediction constructs continuous prediction intervals from a single evaluation run, and we design an ordinal boundary adjustment for discrete rating tasks. We also suggest a midpoint-based score within the interval as a low-bias alternative to raw model score and weighted average. We perform extensive experiments and analysis, which show that conformal prediction can provide valid prediction interval with coverage guarantees. We also explore the usefulness of interval midpoint and judge reprompting for better judgment.

We consider a real-time monitoring system where a source node (with energy limitations) aims to keep the information status at a destination node as fresh as possible by scheduling status update transmissions over a set of channels. The freshness of information at the destination node is measured in terms of the Age of Information (AoI) metric. In this setting, a natural tradeoff exists between the transmission cost (or equivalently, energy consumption) of the source and the achievable AoI performance at the destination. This tradeoff has been optimized in the existing literature under the assumption of having a complete knowledge of the channel statistics. In this work, we develop online learning-based algorithms with finite-time guarantees that optimize this tradeoff in the practical scenario where the channel statistics are unknown to the scheduler. In particular, when the channel statistics are known, the optimal scheduling policy is first proven to have a threshold-based structure with respect to the value of AoI (i.e., it is optimal to drop updates when the AoI value is below some threshold). This key insight was then utilized to develop the proposed learning algorithms that surprisingly achieve an order-optimal regret (i.e., $O(1)$) with respect to the time horizon length.

To assist users in complex tasks, LLMs generate plans: step-by-step instructions towards a goal. While alignment methods aim to ensure LLM plans are helpful, they train (RLHF) or evaluate (ChatbotArena) on what users prefer, assuming this reflects what helps them. We test this with Planorama: an interface where 126 users answer 300 multi-step questions with LLM plans. We get 4388 plan executions and 5584 comparisons to measure plan helpfulness (QA success) and user preferences on plans, and recreate the setup in agents and reward models to see if they simulate or prefer what helps users. We expose: 1) user/model preferences and agent success do not accurately predict which plans help users, so common alignment feedback can misalign with helpfulness; 2) this gap is not due to user-specific preferences, as users are similarly successful when using plans they prefer/disprefer; 3) surface-level cues like brevity and question similarity strongly link to preferences, but such biases fail to predict helpfulness. In all, we argue aligning helpful LLMs needs feedback from real user interactions, not just preferences of what looks helpful, so we discuss the plan NLP researchers can execute to solve this problem.

While large audio language models excel at tasks like ASR and emotion recognition, they still struggle with complex reasoning due to the modality gap between audio and text as well as the lack of structured intermediate supervision. To address this, we propose a unified knowledge distillation framework to transfer reasoning capabilities from a high-capacity textual teacher model to a student audio models while preserving its acoustic competence. Our method introduces two key dimensions: source-wise distillation, which leverages both textual and acoustic teachers to provide complementary modality-specific supervision; and layer-wise distillation, which aligns teacher signals with appropriate student layers to improve transfer efficiency. This dual-dimensional strategy enables fine-grained control over the distillation process, effectively bridging the gap between symbolic reasoning and speech representations. Experimental results show significant improvements in audio reasoning performance, demonstrating the effectiveness of our framework as a reasoning transfer solution for audio modeling.

Large Language Models (LLMs) excel at various tasks, including problem-solving and question-answering. However, LLMs often find Math Word Problems (MWPs) challenging because solving them requires a range of reasoning and mathematical abilities with which LLMs seem to struggle. Recent efforts have helped LLMs solve more complex MWPs with improved prompts. This study proposes a novel method that initially prompts an LLM to create equations from a decomposition of the question, followed by using an external symbolic equation solver to produce an answer. To ensure the accuracy of the obtained answer, inspired by an established recommendation of math teachers, the LLM is instructed to solve the MWP a second time, but this time with the objective of estimating the correct answer instead of solving it exactly. The estimation is then compared to the generated answer to verify. If verification fails, an iterative rectification process is employed to ensure the correct answer is eventually found. This approach achieves new state-of-the-art results on datasets used by prior published research on numeric and algebraic MWPs, improving the previous best results by nearly two percent on average. In addition, the approach obtains satisfactory results on trigonometric MWPs, a task not previously attempted to the authors' best knowledge. This study also introduces two new datasets, SVAMPClean and Trig300, to further advance the testing of LLMs' reasoning abilities.

CDI also offers a plausible approach for automatically making sense of competing arguments in a way that accords with the features enumerated here. This paper is part of an argument that it is now feasible to computationally instantiate a reasonable approximation of a coherence theory of truth [64]: the recent benchmark [12] provides additional quantitative evidence in this direction. By "hard-coding" acceptance of conclusively established propositions, this theory can furthermore be anchored in a correspondence theory of truth [65]. In other words, coherence computations can be required to incorporate privileged information that also coheres with observed reality. While it is easy to imagine attempts to try the same thing with privileged information that does not cohere with observed reality, lies cannot persist when they can easily be unraveled. Even with flawless technology (which this will not be), obstacles will be manifold. For example, in a pluralistic society, legal coherence may actually require sacrificing fairness in some ways [66]. Ultimately, people must decide matters for themselves. It is only reasonable to hope that technology can serve as a reliable tool to help people make their decisions more coherent.

This paper explores how deliberate mutations of reward function in reinforcement learning can produce diversified skill variations in robotic manipulation tasks, examined with a liquid pouring use case. To this end, we developed a new reward function mutation framework that is based on applying Gaussian noise to the weights of the different terms in the reward function. Inspired by the cost-benefit tradeoff model from human motor control, we designed the reward function with the following key terms: accuracy, time, and effort. The study was performed in a simulation environment created in NVIDIA Isaac Sim, and the setup included Franka Emika Panda robotic arm holding a glass with a liquid that needed to be poured into a container. The reinforcement learning algorithm was based on Proximal Policy Optimization. We systematically explored how different configurations of mutated weights in the rewards function would affect the learned policy. The resulting policies exhibit a wide range of behaviours: from variations in execution of the originally intended pouring task to novel skills useful for unexpected tasks, such as container rim cleaning, liquid mixing, and watering. This approach offers promising directions for robotic systems to perform diversified learning of specific tasks, while also potentially deriving meaningful skills for future tasks.

Recent announcements from frontier AI model labs have highlighted strong results on high-school and undergraduate math competitions. Yet it remains unclear whether large language models can solve new, simple conjectures in more advanced areas of mathematics. We propose the Gödel Test: evaluating whether a model can produce correct proofs for very simple, previously unsolved conjectures. To this end, we study the performance of GPT-5 on five conjectures in combinatorial optimization. For each problem, we provided one or two source papers from which the conjecture arose, withheld our own conjecture, and then assessed the model's reasoning in detail. On the three easier problems, GPT-5 produced nearly correct solutions; for Problem 2 it even derived a different approximation guarantee that, upon checking, refuted our conjecture while providing a valid solution. The model failed on Problem 4, which required combining results from two papers. On Problem 5, a harder case without a validated conjecture, GPT-5 proposed the same algorithm we had in mind but failed in the analysis, suggesting the proof is more challenging than expected. Although our sample is small, the results point to meaningful progress on routine reasoning, occasional flashes of originality, and clear limitations when cross-paper synthesis is required. GPT-5 may represent an early step toward frontier models eventually passing the Gödel Test.

As large language models (LLMs) become increasingly powerful, the sequential nature of autoregressive generation creates a fundamental throughput bottleneck that limits the practical deployment. While Multi-Token Prediction (MTP) has demonstrated remarkable benefits for model training efficiency and performance, its inherent potential for inference acceleration remains largely unexplored. This paper introduces FastMTP, a simple yet effective method that improves multi-step draft quality by aligning MTP training with its inference pattern, significantly enhancing speculative decoding performance. Our approach fine-tunes a single MTP head with position-shared weights on self-distilled data, enabling it to capture dependencies among consecutive future tokens and maintain high acceptance rates across multiple recursive draft steps. By integrating language-aware dynamic vocabulary compression into the MTP head, we further reduce computational overhead in the drafting process. Experimental results across seven diverse benchmarks demonstrate that FastMTP achieves an average of 2.03x speedup compared to standard next token prediction with lossless output quality, outperforming vanilla MTP by 82%. FastMTP requires only lightweight training and seamlessly integrates with existing inference frameworks, offering a practical and rapidly deployable solution for accelerating LLM inference.

Handshapes serve a fundamental phonological role in signed languages, with American Sign Language employing approximately 50 distinct shapes. However,computational approaches rarely model handshapes explicitly, limiting both recognition accuracy and linguistic analysis.We introduce a novel graph neural network that separates temporal dynamics from static handshape configurations. Our approach combines anatomically-informed graph structures with contrastive learning to address key challenges in handshape recognition, including subtle interclass distinctions and temporal variations. We establish the first benchmark for structured handshape recognition in signing sequences, achieving 46% accuracy across 37 handshape classes (with baseline methods achieving 25%).