We propose patching for large language models (LLM) like software versions, a lightweight and modular approach for addressing safety vulnerability. While vendors release improved LLM versions, major releases are costly, infrequent, and difficult to tailor to customer needs, leaving released models with known safety gaps. Unlike full-model fine-tuning or major version updates, our method enables rapid remediation by prepending a compact, learnable prefix to an existing model. This "patch" introduces only 0.003% additional parameters, yet reliably steers model behavior toward that of a safer reference model. Across three critical domains--toxicity mitigation, bias reduction, and harmfulness refusal--policy patches achieve safety improvements comparable to next-generation safety-aligned models while preserving fluency. Our results demonstrate that LLMs can be "patched" much like software, offering vendors and practitioners a practical mechanism for distributing scalable, efficient, and composable safety updates between major model releases. Large language models (LLMs) have achieved remarkable advances in reasoning, generation, and multilingual capabilities (Brown et al., 2020; Wei et al., 2022; Conneau & Lample, 2019). Despite their impressive capabilities, they continue to exhibit serious safety concerns, such as the generation of toxic language (Gehman et al., 2020), biased associations that reinforce stereotypes (Dong et al., 2024), and the production of harmful or dangerous content (Mazeika et al., 2024). Addressing these risks is crucial to the broader challenge of alignment, where models are refined to better align with human values and expectations. Conventional approaches to improving safety rely on alignment techniques such as Reinforcement Learning from Human Feedback (RLHF) (Christiano et al., 2017; Bai et al., 2022; Ouyang et al., 2022), preference-based fine-tuning (Rafailov et al., 2023), domain-specific supervised fine-tuning (Li et al., 2024), etc. While these methods have proven effective, they require substantial computational resources, large-scale data curation, and careful model retraining. In practice, model providers (vendors) often release major updates to their models (major versions) on a fixed schedule, typically once or twice a year.

Self-improvement is a critical capability for large language models and other intelligent systems, enabling them to refine their behavior and internal consistency without external supervision. Despite its importance, prior approaches largely rely on empirical heuristics and lack formal guarantees. In this paper, we propose a principled framework for self-improvement based on the concept of \emph{coherence}, which requires that a model's outputs remain consistent under task-preserving transformations of the input. We formalize this concept using projection-based mechanisms that update a baseline model to be coherent while remaining as close as possible to its original behavior. We provide rigorous theoretical guarantees that these mechanisms achieve \emph{monotonic improvement}, measured by a reduction in expected Bregman divergence. Our analysis is comprehensive, covering both \emph{direct} and \emph{two-step} projection methods, and robustly extends these guarantees to non-realizable settings, empirical (finite-sample) distributions, and relaxed coherence constraints. Furthermore, we establish a general \emph{characterization theorem}, showing that any mechanism with similar provable improvement guarantees must inherently conform to a coherence-based structure. This culminates in rigidity results under the demand for universal improvement, establishing coherence as a fundamental and, in a formal sense, necessary principle for provable self-improvement.

Accurate disease interpretation from radiology remains challenging due to imaging heterogeneity. Achieving expert-level diagnostic decisions requires integration of subtle image features with clinical knowledge. Yet major vision-language models (VLMs) treat images as holistic entities and overlook fine-grained image details that are vital for disease diagnosis. Clinicians analyze images by utilizing their prior medical knowledge and identify anatomical structures as important region of interests (ROIs). Inspired from this human-centric workflow, we introduce Anatomy-VLM, a fine-grained, vision-language model that incorporates multi-scale information. First, we design a model encoder to localize key anatomical features from entire medical images. Second, these regions are enriched with structured knowledge for contextually-aware interpretation. Finally, the model encoder aligns multi-scale medical information to generate clinically-interpretable disease prediction. Anatomy-VLM achieves outstanding performance on both in- and out-of-distribution datasets. We also validate the performance of Anatomy-VLM on downstream image segmentation tasks, suggesting that its fine-grained alignment captures anatomical and pathology-related knowledge. Furthermore, the Anatomy-VLM's encoder facilitates zero-shot anatomy-wise interpretation, providing its strong expert-level clinical interpretation capabilities.

The alignment of Large Language Models (LLMs) for multi-turn conversations typically relies on reward signals derived from the content of the text. This approach, however, overlooks a rich, complementary source of signal: the dynamics of the interaction itself. This paper introduces TRACE (Trajectory-based Reward for Agent Collaboration Estimation), a novel reward signal derived from the geometric properties of a dialogue's embedding trajectory--a concept we term 'conversational geometry'. Our central finding is that a reward model trained only on these structural signals achieves a pairwise accuracy (68.20%) comparable to a powerful LLM baseline that analyzes the full transcript (70.04%). Furthermore, a hybrid model combining interaction dynamics with textual analysis achieves the highest performance (80.17%), demonstrating their complementary nature. This work provides strong evidence that for interactive settings, how an agent communicates is as powerful a predictor of success as what it says, offering a new, privacy-preserving framework that not only aligns agents but also serves as a diagnostic tool for understanding the distinct interaction patterns that drive successful collaboration.

Rotary Positional Encodings (RoPE) have emerged as a highly effective technique for one-dimensional sequences in Natural Language Processing spurring recent progress towards generalizing RoPE to higher-dimensional data such as images and videos. The success of RoPE has been thought to be due to its positional equivariance, i.e. its status as a relative positional encoding. In this paper, we mathematically show RoPE to be one of the most general solutions for equivariant positional embedding in one-dimensional data. Moreover, we show Mixed RoPE to be the analogously general solution for M-dimensional data, if we require commutative generators -- a property necessary for RoPE's equivariance. However, we question whether strict equivariance plays a large role in RoPE's performance. We propose Spherical RoPE, a method analogous to Mixed RoPE, but assumes non-commutative generators. Empirically, we find Spherical RoPE to have the equivalent or better learning behavior compared to its equivariant analogues. This suggests that relative positional embeddings are not as important as is commonly believed, at least within computer vision. We expect this discovery to facilitate future work in positional encodings for vision that can be faster and generalize better by removing the preconception that they must be relative.

Hybrid Quantum Classical (HQC) algorithms constitute one of the most effective paradigms for exploiting the computational advantages of quantum systems in large-scale numerical tasks. By operating in high-dimensional Hilbert spaces, quantum circuits enable exponential speed-ups and provide access to richer representations of cost landscapes compared to purely classical methods. These capabilities are particularly relevant for machine learning, where state-of-the-art models especially in Natural Language Processing (NLP) suffer from prohibitive time complexity due to massive matrix multiplications and high-dimensional optimization. In this manuscript, we propose a Hybrid Quantum Classical selection mechanism for the Mamba architecture, designed specifically for temporal sequence classification problems. Our approach leverages Variational Quantum Circuits (VQCs) as quantum gating modules that both enhance feature extraction and improve suppression of irrelevant information. This integration directly addresses the computational bottlenecks of deep learning architectures by exploiting quantum resources for more efficient representation learning. We analyze how introducing quantum subroutines into large language models (LLMs) impacts their generalization capability, expressivity, and parameter efficiency. The results highlight the potential of quantum-enhanced gating mechanisms as a path toward scalable, resource-efficient NLP models, in a limited simulation step. Within the first four epochs on a reshaped MNIST dataset with input format (batch, 784, d_model), our hybrid model achieved 24.6% accuracy while using one quantum layer and achieve higher expressivity, compared to 21.6% obtained by a purely classical selection mechanism. we state No founding

Large Language Models (LLMs) have demonstrated remarkable success in conversational systems by generating human-like responses. However, they can fall short, especially when required to account for personalization or specific knowledge. In real-life settings, it is impractical to rely on users to detect these errors and request a new response. One way to address this problem is to refine the response before returning it to the user. While existing approaches focus on refining responses within a single LLM, this method struggles to consider diverse aspects needed for effective conversations. In this work, we propose refining responses through a multi-agent framework, where each agent is assigned a specific role for each aspect. We focus on three key aspects crucial to conversational quality: factuality, personalization, and coherence. Each agent is responsible for reviewing and refining one of these aspects, and their feedback is then merged to improve the overall response. To enhance collaboration among them, we introduce a dynamic communication strategy. Instead of following a fixed sequence of agents, our approach adaptively selects and coordinates the most relevant agents based on the specific requirements of each query. We validate our framework on challenging conversational datasets, demonstrating that ours significantly outperforms relevant baselines, particularly in tasks involving knowledge or user's persona, or both.

Existing paper review methods often rely on superficial manuscript features or directly on large language models (LLMs), which are prone to hallucinations, biased scoring, and limited reasoning capabilities. Moreover, these methods often fail to capture the complex argumentative reasoning and negotiation dynamics inherent in reviewer-author interactions. To address these limitations, we propose ReView-Graph (Reviewer-Author Debates Graph Reasoner), a novel framework that performs heterogeneous graph reasoning over LLM-simulated multi-round reviewer-author debates. In our approach, reviewer-author exchanges are simulated through LLM-based multi-agent collaboration. Diverse opinion relations (e.g., acceptance, rejection, clarification, and compromise) are then explicitly extracted and encoded as typed edges within a heterogeneous interaction graph. By applying graph neural networks to reason over these structured debate graphs, ReViewGraph captures fine-grained argumentative dynamics and enables more informed review decisions. Extensive experiments on three datasets demonstrate that Re-ViewGraph outperforms strong baselines with an average relative improvement of 15.73%, underscoring the value of modeling detailed reviewer-author debate structures.

Best-of-$N$ reasoning improves the accuracy of language models in solving complex tasks by sampling multiple candidate solutions and then selecting the best one based on some criteria. A critical bottleneck for this strategy is the output diversity limit, which occurs when the model generates similar outputs despite stochastic sampling, and hence recites the same error. To address this lack of variance in reasoning paths, we propose a novel unsupervised activation steering strategy that simultaneously optimizes the steering vectors for multiple reasoning trajectories at test time. At any synchronization anchor along the batch generation process, we find the steering vectors that maximize the total volume spanned by all possible intervened activation subsets. We demonstrate that these steering vectors can be determined by solving a Riemannian optimization problem over the product of spheres with a log-determinant objective function. We then use a Riemannian block-coordinate descent algorithm with a well-tuned learning rate to obtain a stationary point of the problem, and we apply these steering vectors until the generation process reaches the subsequent synchronization anchor. Empirical evaluations on popular mathematical benchmarks demonstrate that our test-time Riemannian activation steering strategy outperforms vanilla sampling techniques in terms of generative diversity and solution accuracy.

The transition from human-centric to agent-centric software development practices is disrupting existing knowledge sharing environments for software developers. Traditional peer-to-peer repositories and developer communities for shared technical knowledge and best practice have witnessed dramatic drops in participation in a short period of time. At the same time, agentic functional equivalents are yet to emerge leaving AI agents, which already generate a significant proportion of all new software code produced, without access to repositories of valuable shared learning. In this paper, we introduce Spark, a novel shared agentic memory architecture which is designed to emulate the collective intelligence and know-how of human developer communities. Spark enables AI coding agents to both contribute to and draw from a persistent and continuously evolving experiential memory. Agents operating in the same general problem space use the Spark shared memory as a repository of new knowledge to achieve collective continual learning. We evaluate Spark as a coach for AI coding agents performing software development tasks. We demonstrate that recommendations made by Spark improve the quality of code generated by generic code generation models at varying sizes and capability tiers. Boosted by Spark, a small open-weights model with 30 billion parameters was able to match the code quality afforded by a much larger state-of-the-art model. Separately, we measure the intrinsic quality of recommendations generated by Spark against a wide range of criteria inspired by software development best practice, and achieve helpfulness levels of up to 98.2% in the top two (out of five) qualitative helpfulness bands.