AMD GPUs offer state-of-the-art compute and memory bandwidth; however, peak performance AMD kernels are written in raw assembly. To address the difficulty of mapping AI algorithms to hardware, recent work proposes C++ embedded and PyTorch-inspired domain-specific languages like ThunderKittens (TK) to simplify high performance AI kernel development on NVIDIA hardware. We explore the extent to which such primitives -- for explicit tile-based programming with optimized memory accesses and fine-grained asynchronous execution across workers -- are NVIDIA-specific or general. We provide the first detailed study of the programming primitives that lead to performant AMD AI kernels, and we encapsulate these insights in the HipKittens (HK) programming framework. We find that tile-based abstractions used in prior DSLs generalize to AMD GPUs, however we need to rethink the algorithms that instantiate these abstractions for AMD. We validate the HK primitives across CDNA3 and CDNA4 AMD platforms. In evaluations, HK kernels compete with AMD's hand-optimized assembly kernels for GEMMs and attention, and consistently outperform compiler baselines. Moreover, assembly is difficult to scale to the breadth of AI workloads; reflecting this, in some settings HK outperforms all available kernel baselines by $1.2-2.4\times$ (e.g., $d=64$ attention, GQA backwards, memory-bound kernels). These findings help pave the way for a single, tile-based software layer for high-performance AI kernels that translates across GPU vendors. HipKittens is released at: https://github.com/HazyResearch/HipKittens.

Latent diffusion models such as Stable Diffusion achieve state-of-the-art results on text-to-image generation tasks. However, the extent to which these models have a semantic understanding of the images they generate is not well understood. In this work, we investigate whether the internal representations used by these models during text-to-image generation contain semantic information that is meaningful to humans. To do so, we perform probing on Stable Diffusion with simple regression layers that predict semantic attributes for objects and evaluate these predictions against human annotations. Surprisingly, we find that this success can actually be attributed to the text encoding occurring in CLIP rather than the reverse diffusion process. We demonstrate that groups of specific semantic attributes have markedly different decoding accuracy than the average, and are thus represented to different degrees. Finally, we show that attributes become more difficult to disambiguate from one another during the inverse diffusion process, further demonstrating the strongest semantic representation of object attributes in CLIP. We conclude that the separately trained CLIP vision-language model is what determines the human-like semantic representation, and that the diffusion process instead takes the role of a visual decoder.

LLMs demonstrate performance comparable to human abilities in complex tasks such as mathematical reasoning, but their robustness in mathematical reasoning under minor input perturbations still lacks systematic investigation. Existing methods generally suffer from limited scalability, weak semantic preservation, and high costs. Therefore, we propose MSCR, an automated adversarial attack method based on multi-source candidate replacement. By combining three information sources including cosine similarity in the embedding space of LLMs, the WordNet dictionary, and contextual predictions from a masked language model, we generate for each word in the input question a set of semantically similar candidates, which are then filtered and substituted one by one to carry out the attack. We conduct large-scale experiments on LLMs using the GSM8K and MATH500 benchmarks. The results show that even a slight perturbation involving only a single word can significantly reduce the accuracy of all models, with the maximum drop reaching 49.89% on GSM8K and 35.40% on MATH500, while preserving the high semantic consistency of the perturbed questions. Further analysis reveals that perturbations not only lead to incorrect outputs but also substantially increase the average response length, which results in more redundant reasoning paths and higher computational resource consumption.

Enterprise adoption of agentic AI systems requires reliable evaluation methods that reflect real-world deployment scenarios. Traditional LLM benchmarks suffer from training data contamination and fail to measure agentic capabilities such as multi-step tool use and decision-making under uncertainty. We present the Kamiwaza Agentic Merit Index (KAMI) v0.1, an enterprise-focused benchmark that addresses both contamination resistance and agentic evaluation. Through 170,000 LLM test items processing over 5.5 billion tokens across 35 model configurations, we demonstrate that traditional benchmark rankings poorly predict practical agentic performance. Notably, newer generation models like Llama 4 or Qwen 3 do not always outperform their older generation variants on enterprise-relevant tasks, contradicting traditional benchmark trends. We also present insights on cost-performance tradeoffs, model-specific behavioral patterns, and the impact of reasoning capabilities on token efficiency -- findings critical for enterprises making deployment decisions.

The rapid advancement of generative adversarial networks (GANs) and diffusion models has enabled the creation of highly realistic deepfake content, posing significant threats to digital trust across audio-visual domains. While unimodal detection methods have shown progress in identifying synthetic media, their inability to leverage cross-modal correlations and precisely localize forged segments limits their practicality against sophisticated, fine-grained manipulations. To address this, we introduce a multi-modal deepfake detection and localization framework based on a Feature Pyramid-Transformer (FPN-Transformer), addressing critical gaps in cross-modal generalization and temporal boundary regression. The proposed approach utilizes pre-trained self-supervised models (WavLM for audio, CLIP for video) to extract hierarchical temporal features. A multi-scale feature pyramid is constructed through R-TLM blocks with localized attention mechanisms, enabling joint analysis of cross-context temporal dependencies. The dual-branch prediction head simultaneously predicts forgery probabilities and refines temporal offsets of manipulated segments, achieving frame-level localization precision. We evaluate our approach on the test set of the IJCAI'25 DDL-A V benchmark, showing a good performance with a final score of 0.7535 for cross-modal deepfake detection and localization in challenging environments. Experimental results confirm the effectiveness of our approach and provide a novel way for generalized deepfake detection. Our code is available at https://github.com/Zig-HS/MM-DDL.

Graph Transformers (GTs) have shown strong empirical performance, yet current architectures vary widely in their use of attention mechanisms, positional embeddings (PEs), and expressivity. Existing expressivity results are often tied to specific design choices and lack comprehensive empirical validation on large-scale data. This leaves a gap between theory and practice, preventing generalizable insights that exceed particular application domains. Here, we propose the Generalized-Distance Transformer (GDT), a GT architecture using standard attention that incorporates many advancements for GTs from recent years, and develop a fine-grained understanding of the GDT's representation power in terms of attention and PEs. Through extensive experiments, we identify design choices that consistently perform well across various applications, tasks, and model scales, demonstrating strong performance in a few-shot transfer setting without fine-tuning. Our evaluation covers over eight million graphs with roughly 270M tokens across diverse domains, including image-based object detection, molecular property prediction, code summarization, and out-of-distribution algorithmic reasoning. We distill our theoretical and practical findings into several generalizable insights about effective GT design, training, and inference.

Understanding complex biomolecular mechanisms requires multi-step reasoning across molecular interactions, signaling cascades, and metabolic pathways. While large language models(LLMs) show promise in such tasks, their application to biomolecular problems is hindered by logical inconsistencies and the lack of grounding in domain knowledge. Existing approaches often exacerbate these issues: reasoning steps may deviate from biological facts or fail to capture long mechanistic dependencies. To address these challenges, we propose a Knowledge-Augmented Long-CoT Reasoning framework that integrates LLMs with knowledge graph-based multi-hop reasoning chains. The framework constructs mechanistic chains via guided multi-hop traversal and pruning on the knowledge graph; these chains are then incorporated into supervised fine-tuning to improve factual grounding and further refined with reinforcement learning to enhance reasoning reliability and consistency. Furthermore, to overcome the shortcomings of existing benchmarks, which are often restricted in scale and scope and lack annotations for deep reasoning chains, we introduce PrimeKGQA, a comprehensive benchmark for biomolecular question answering. Experimental results on both PrimeKGQA and existing datasets demonstrate that although larger closed-source models still perform well on relatively simple tasks, our method demonstrates clear advantages as reasoning depth increases, achieving state-of-the-art performance on multi-hop tasks that demand traversal of structured biological knowledge. These findings highlight the effectiveness of combining structured knowledge with advanced reasoning strategies for reliable and interpretable biomolecular reasoning.

LLMs have made significant progress in the field of mathematical reasoning, but whether they have true the mathematical understanding ability is still controversial. To explore this issue, we propose a new perturbation framework to evaluate LLMs' reasoning ability in complex environments by injecting additional semantically irrelevant perturbation sentences and gradually increasing the perturbation intensity. At the same time, we use an additional perturbation method: core questioning instruction missing, to further analyze the LLMs' problem-solving mechanism. The experimental results show that LLMs perform stably when facing perturbation sentences without numbers, but there is also a robustness boundary. As the perturbation intensity increases, the performance exhibits varying degrees of decline; when facing perturbation sentences with numbers, the performance decreases more significantly, most open source models with smaller parameters decrease by nearly or even more than 10%, and further increasing with the enhancement of perturbation intensity, with the maximum decrease reaching 51.55%. Even the most advanced commercial LLMs have seen a 3%-10% performance drop. By analyzing the reasoning process of LLMs in detail, We find that models are more sensitive to perturbations with numerical information and are more likely to give incorrect answers when disturbed by irrelevant numerical information. The higher the perturbation intensity, the more obvious these defects are. At the same time, in the absence of core questioning instruction, models can still maintain an accuracy of 20%-40%, indicating that LLMs may rely on memory templates or pattern matching to complete the task, rather than logical reasoning. In general, our work reveals the shortcomings and limitations of current LLMs in their reasoning capabilities, which is of great significance for the further development of LLMs.

Multi-character role-playing aims to equip models with the capability to simulate diverse roles. Existing methods either use one shared parameterized module across all roles or assign a separate parameterized module to each role. However, the role-shared module may ignore distinct traits of each role, weakening personality learning, while the role-specific module may overlook shared traits across multiple roles, hindering commonality modeling. In this paper, we propose a novel HyCoRA: Hyper-Contrastive Role-Adaptive learning framework, which efficiently improves multi-character role-playing ability by balancing the learning of distinct and shared traits. Specifically, we propose a Hyper-Half Low-Rank Adaptation structure, where one half is a role-specific module generated by a lightweight hyper-network, and the other half is a trainable role-shared module. The role-specific module is devised to represent distinct persona signatures, while the role-shared module serves to capture common traits. Moreover, to better reflect distinct personalities across different roles, we design a hyper-contrastive learning mechanism to help the hyper-network distinguish their unique characteristics. Extensive experimental results on both English and Chinese available benchmarks demonstrate the superiority of our framework. Further GPT -4 evaluations and visual analyses also verify the capability of HyCoRA to capture role characteristics.

Direction-of-Arrival (DOA) estimation is critical in spatial audio and acoustic signal processing, with wide-ranging applications in real-world. Most existing DOA models are trained on synthetic data by convolving clean speech with room impulse responses (RIRs), which limits their generalizability due to constrained acoustic diversity. In this paper, we revisit DOA estimation using a recently introduced dataset constructed with the assistance of large language models (LLMs), which provides more realistic and diverse spatial audio scenes. We benchmark several representative neural-based DOA methods on this dataset and propose LightDOA, a lightweight DOA estimation model based on depthwise separable convolutions, specifically designed for mutil-channel input in varying environments. Experimental results show that LightDOA achieves satisfactory accuracy and robustness across various acoustic scenes while maintaining low computational complexity. This study not only highlights the potential of spatial audio synthesized with the assistance of LLMs in advancing robust and efficient DOA estimation research, but also highlights LightDOA as efficient solution for resource-constrained applications.