--The unabated growth in AI workload demands is driving the need for concerted advances in compute, memory, and interconnect performance. As traditional semiconductor scaling slows, high-speed interconnects have emerged as the new scaling engine, enabling the creation of larger logical GPUs by linking many GPUs into a single, low-latency, high-bandwidth compute domain. While initial scale-up fabrics leveraged copper interconnects for their power and cost advantages, the maximum reach of passive electrical interconnects (approximately 1 meter) effectively limits the scale-up domain to within a single rack. The advent of 3D-stacked optics and logic offers a transformative, power-efficient scale-up solution for connecting hundreds of GPU packages (thousands of GPUs) across multiple data center racks. This work explores the design tradeoffs of scale-up technologies and demonstrates how frontier LLMs necessitate novel photonic solutions to achieve aggressive power and performance targets. We model the benefits of 3D CPO (Passage) enabled GPUs and switches within the scale-up domain when training Frontier Mixture of Experts (MoE) models exceeding one trillion parameters. Our results show that the substantial increases in bandwidth and radix enabled by 3D CPO allow for an 8X increase in scale-up capability. The race to build larger, more sophisticated AI models is pushing the limits of existing infrastructure. At the chip and package level, GPUs are constrained by shoreline, yields and power. These challenges have led to the development of large high-bandwidth, low-latency scale-up pods. These pods effectively combine hundreds of GPUs into a single logical GPU to facilitate a variety of parallelism strategies (e.g. Approaches like Mixture of Experts (MoE) [1] have pushed scale-up networks to their limits due to copper reach (1 meter), which constrains the number of GPUs that can be connected within a single network hop. With MoEs, an ensemble of specialized sub-networks work together through sparse activations to increase model capacity without significantly increasing computational requirements. The output of the selected experts are combined to create the final result.

As large language models (LLMs) continue to scale, multi-node deployment has become a necessity. Consequently, communication has become a critical performance bottleneck. Current intra-node communication libraries, like NCCL, typically make use of a single interconnect such as NVLink. This approach creates performance ceilings, especially on hardware like the H800 GPU where the primary interconnect's bandwidth can become a bottleneck, and leaves other hardware resources like PCIe and Remote Direct Memory Access (RDMA)-capable Network Interface Cards (NICs) largely idle during intensive workloads. We propose FlexLink, the first collective communication framework to the best of our knowledge designed to systematically address this by aggregating these heterogeneous links-NVLink, PCIe, and RDMA NICs-into a single, high-performance communication fabric. FlexLink employs an effective two-stage adaptive load balancing strategy that dynamically partitions communication traffic across all available links, ensuring that faster interconnects are not throttled by slower ones. On an 8-GPU H800 server, our design improves the bandwidth of collective operators such as AllReduce and AllGather by up to 26% and 27% over the NCCL baseline, respectively. This gain is achieved by offloading 2-22% of the total communication traffic to the previously underutilized PCIe and RDMA NICs. FlexLink provides these improvements as a lossless, drop-in replacement compatible with the NCCL API, ensuring easy adoption.

Modern chip physical design relies heavily on Electronic Design Automation (EDA) tools, which often struggle to provide interpretable feedback or actionable guidance for improving routing congestion. In this work, we introduce a Multimodal Large Language Model Assistant (MLLMA) that bridges this gap by not only predicting congestion but also delivering human-interpretable design suggestions. Our method combines automated feature generation through MLLM-guided genetic prompting with an interpretable preference learning framework that models congestion-relevant tradeoffs across visual, tabular, and textual inputs. We compile these insights into a "Design Suggestion Deck" that surfaces the most influential layout features and proposes targeted optimizations. Experiments on the CircuitNet benchmark demonstrate that our approach outperforms existing models on both accuracy and explainability. Additionally, our design suggestion guidance case study and qualitative analyses confirm that the learned preferences align with real-world design principles and are actionable for engineers. This work highlights the potential of MLLMs as interactive assistants for interpretable and context-aware physical design optimization.

The design of complex machines stands as both a marker of human intelligence and a foundation of engineering practice. Given recent advances in large language models (LLMs), we ask whether they, too, can learn to create. We approach this question through the lens of compositional machine design: a task in which machines are assembled from standardized components to meet functional demands like locomotion or manipulation in a simulated physical environment. With this simplification, machine design is expressed as writing XML-like code that explicitly specifies pairwise part connections. To support this investigation, we introduce BesiegeField, a testbed built on the machine-building game Besiege, which enables part-based construction, physical simulation and reward-driven evaluation. Using BesiegeField, we benchmark state-of-the-art LLMs with agentic workflows and identify key capabilities required for success, including spatial reasoning, strategic assembly, and instruction-following. As current open-source models fall short, we explore reinforcement learning (RL) as a path to improvement: we curate a cold-start dataset, conduct RL finetuning experiments, and highlight open challenges at the intersection of language, machine design, and physical reasoning.

Large language models (LLMs) are increasingly used as role-playing agents, yet their capacity to faithfully and consistently portray version-specific characters -- for example, superheroes across comic and cinematic universes -- remains underexplored. Superhero canons such as Marvel and DC provide a rich testbed: decades of storytelling yield multiple incarnations of the same character with distinct histories, values, and moral codes. To study this problem, we introduce Beyond One World, a benchmark for character-grounded roleplay spanning 30 iconic heroes and 90 canon-specific versions. The benchmark comprises two tasks: (i) Canon Events, which probes factual recall of pivotal life stages, and (ii) Moral Dilemmas, which confronts models with ethically charged scenarios. We score responses for canonical accuracy and reasoning fidelity under a framework that separates internal deliberation ("thinking") from outward decisions ("acting"). We further propose Think-Act Matching, a metric that quantifies alignment between reasons and actions and serves as a proxy for model trustworthiness. Experiments across reasoning- and non-reasoning-oriented models yield three findings: (1) chain-of-thought prompting improves narrative coherence in weaker models but can reduce canonical accuracy in stronger ones; (2) cross-version generalization within a character remains a major obstacle; and (3) models often excel at either thinking or acting, but rarely both. Beyond One World exposes critical gaps in multiversal consistency and reasoning alignment, offering a challenging evaluation for role-playing LLMs.

Large language models (LLMs) often solve problems using step-by-step Chain-of-Thought (CoT) reasoning, yet these intermediate steps are frequently unfaithful or hard to interpret. Inspired by the Uniform Information Density (UID) hypothesis in psycholinguistics -- which posits that humans communicate by maintaining a stable flow of information -- we introduce entropy-based metrics to analyze the information flow within reasoning traces. Surprisingly, across three challenging mathematical benchmarks, we find that successful reasoning in LLMs is globally non-uniform: correct solutions are characterized by uneven swings in information density, in stark contrast to human communication patterns. This result challenges assumptions about machine reasoning and suggests new directions for designing interpretable and adaptive reasoning models.

Test-time scaling has enabled Large Language Models (LLMs) with remarkable reasoning capabilities, particularly in mathematical domains, through intermediate chain-of-thought (CoT) reasoning before generating final answers. However, the specific sources and mechanisms underlying these reasoning capabilities remain insufficiently understood. Optimization reasoning, i.e. finding extrema under constraints, represents a fundamental abstraction that underpins critical applications in planning, control, resource allocation, and prompt search. To systematically evaluate this capability, we introduce ExtremBench, a benchmark dataset for solving mathematical extremal problems, curated from inequality exercises used for Chinese Mathematical Olympiad and transformed into $93$ standardized extrema-finding problems. We conduct extensive evaluations across various state-of-the-art open-source model families, including the Qwen3, GPT-OSS, and DeepSeek. Our results reveal that LLMs' extremal-solving reasoning capabilities do not always align with those of current mathematical benchmarks such as AIME25 and MATH-500, with some models showing strong general mathematical reasoning but poor extremal-solving skills, and vice versa. This discrepancy highlights a critical gap in current evaluation practices and suggests that existing benchmarks may not comprehensively capture the full spectrum of mathematical reasoning abilities.

Surveys provide valuable insights into public opinion and behavior, but their execution is costly and slow. Large language models (LLMs) have been proposed as a scalable, low-cost substitute for human respondents, but their outputs are often biased and yield invalid estimates. We study the interplay between synthesis methods that use LLMs to generate survey responses and rectification methods that debias population estimates, and explore how human responses are best allocated between them. Using two panel surveys with questions on nutrition, politics, and economics, we find that synthesis alone introduces substantial bias (24-86%), whereas combining it with rectification reduces bias below 5% and increases effective sample size by up to 14%. Overall, we challenge the common practice of using all human responses for fine-tuning, showing that under a fixed budget, allocating most to rectification results in far more effective estimation.

Large language models (LLMs) often generate reasoning traces that appear coherent but rest on unsupported assumptions, leading to hallucinated conclusions. Prior work mainly addresses factual hallucinations or relies on post-hoc verification, leaving reasoning-induced hallucinations largely unaddressed. We propose Audit-of-Understanding (AoU), a framework that constrains inference to validated premises through three phases: (1) decomposing a query into candidate assumptions, (2) auditing their support, and (3) conditioning inference only on the validated subset. Formally, AoU is \emph{posterior-constrained inference}, connecting to selective prediction and rejection learning. Our contributions are threefold: (i) theoretical guarantees under perfect validation, (ii) excess-risk bounds under imperfect audits, and (iii) tractability analysis. Empirically, AoU improves both accuracy and faithfulness on GSM8K, MultiArith, and SVAMP, achieving up to +30% gains on GSM8K, +45% on MultiArith, and consistent +20--28% improvements on SVAMP over Chain-of-Thought, Self-Consistency, and CoT-Decoding. Code is available at https://anonymous.4open.science/r/audit-of-understanding-E28B.

Criminal Court View Generation (CVG) is a fundamental task in legal artificial intelligence, aiming to automatically generate the "Court View" section of a legal case document. Generating court views is challenging due to the diversity and complexity of case facts, and directly generating from raw facts may limit performance. In this paper, we present ShiZhi, the first large language model (LLM) specifically designed for court view generation. We construct a Chinese Court View Generation dataset, CCVG, of more than 110K cases, each containing fact descriptions paired with corresponding court views. Based on this dataset, ShiZhi achieving 70.00 ROUGE-1 and 67.85 BLEU-1 on court view generation, as well as 86.48\% accuracy with 92.75\% macro F1 on charge prediction. Experimental results demonstrate that even a small LLM can generate reasonable and legally coherent court views when trained on high-quality domain-specific data. Our model and dataset are available at \href{https://github.com/ZhitianHou/ShiZhi}{https://github.com/ZhitianHou/ShiZhi}.