Social interactions are shaped by the interplay of dispositional traits and situational context, yet systematically investigating how personality configurations between individuals jointly influence social behavior across diverse social contexts remains methodologically challenging. We address this gap by introducing a simulation pipeline adapted from the CHARISMA framework, which employs well-known movie characters and public figures as psychologically grounded agents for multi-LLM social simulation using a method we term personality anchoring. We present a large-scale empirical study examining how dyadic Agreeableness composition influences social interaction outcomes across 1,010 simulated conversations. Our results reveal a monotonic relationship between dyadic Agreeableness composition and shared goal achievement, with Homogeneous-Agreeable pairs achieving success 10 times the rate of Homogeneous-Disagreeable pairs (62% vs. 6%). Behavioral mediation analysis reveals that Agreeableness shapes goal achievement partially through cooperative strategy selection, though it continues to predict outcomes within the same dominant strategy, indicating pathways beyond observable conversational behavior. Robustness analyses confirm high consistency of results across repeated simulations (ICC = 0.89) and stable personality expression across diverse scenarios, validating personality anchoring as a viable operationalization strategy.

With the rapid demand for energy, even the incorporation of bulk renewable energy sources is not entirely sufficient to meet demand besides adding supply uncertainty. Distributed Energy Resource Aggregators (DERAs) have the potential to address this uncertainty via aggregation and control of decentralized distributed energy sources, thereby acting like virtual power plants. We present a new approach that combines forecasting and model-predictive control to assign DERAs to follow net-demand patterns, while accounting for the dynamics of the aggregate energy sources and their capacity limits. Each DERA is represented as a flexible ``virtual battery" with constraints on state-of-charge and power limits. The dispatch problem is set up as a long-term model predictive control task that aims to minimize differences from desired charge levels, output ramping, and net-load tracking errors. To keep operations efficient in real time, we implement a rolling-horizon MPC, which updates decisions regularly using the latest marginal-demand forecasts. For forecasting, we present two models: linear regression and long-short term memory (LSTM) neural network. Using high-resolution CAISO net-demand data and five typical DERA types, our simulations demonstrate how well our approach tracks marginal-demand; in particular, we highlight the tradeoffs between forecasting horizon times and MPC update rate as well as the dependence on the choice of the load forecasting model. Our results also indicate a slight edge for LSTM models over linear regression for desired time shifts and horizon choices.

An edge node monitoring a wildfire observes more than a duty-limited or windowed down-link can carry. The receiver must predict the H-step-ahead hazard map from whatever the link delivers. We argue the operative design problem is not which neural architecture to use but how to derive a structured belief sufficient for the receiver's prediction task and maintain it through a scheduler that anticipates future transmission opportunities. We formalize this as a partially observed sequential allocation problem with three coupled per-region action axes (sensing, representation, transmission), and derive each component of the structured belief from the H-step forward operator's input requirements. Identifying these mechanisms requires independent control over the window period P, per-window capacity C, predictive horizon H, and fuel composition, which is not separable in real-landscape data; we therefore evaluate on a physics-calibrated synthetic environment. Three empirical observations support the principle: the gap between a non-myopic activity-paced reference and uniform pacing is unimodal in window-period sparsity, peaking at intermediate spacing; ablating the structured belief, the dominant operative component flips between a default landscape (temporal staleness) and a structured landscape (static-risk prior), while the per-cell intensity belief is redundant in both; and a 40 k-parameter lightweight cross-region attention encoder exceeds the FAIR activity-paced reference by ~28% on the default landscape and ~11% on the structured landscape. A deeper Transformer encoder does not improve over the lightweight encoder in mean predictive loss and exhibits higher training-seed variance. Within this task class and regime, a modest architectural inductive bias suffices when the belief and the scheduling problem are correctly posed.

Privacy deceptive patterns in web interfaces systematically manipulate users into disclosing personal data, yet existing defenses are fragmented, static, and increasingly vulnerable to manipulation by large language models. Moreover, data voids, areas of information scarcity within the web ecosystem, create fertile ground for adversaries to inject misleading content that can be scraped and learned by AI systems, thereby amplifying both deceptive design and model misbehavior. In this paper, we formalize a new threat model, AI grooming, where attackers exploit data voids to seed benign-looking but malicious samples that corrupt model reasoning and normalize deceptive practices. To address this threat in privacy deceptive patterns, we present DPAgent, an agentic and reasoning-aware framework that orchestrates four specialized agents to mitigate the AI Grooming threat via a proactive defense that combines latent space purification with defensive prompting and operates directly in live web environments to proactively explore, detect, and repair privacy deceptive user interfaces before they reach end users. Extensive evaluations show that DPAgent detects 90.98% of groomed samples, achieves state-of-the-art privacy deceptive pattern detection with a micro F1 of 0.816, explores over 80% of pattern types while visiting only about 10% of the pages required by baselines, and successfully repairs 77% of detected deceptive interfaces. A large-scale study of 485 websites in the wild reveals that up to 98% contain at least one privacy deceptive pattern, over 90% of which can be mitigated by DPAgent. User studies further confirm that DPAgent effectively reduces privacy risks while preserving browsing experience. Our results demonstrate the promise of agent-in-the-middle defenses for securing the web UI supply chain against deceptive design and emerging AI threats rooted in data void exploitation.

Software refactoring is essential for maintaining code quality. However, API replacement refactoring, which replaces custom logic with API calls, remains underexplored. Existing refactoring tools provide limited support for detecting such opportunities because they rely on predefined templates and have difficulty capturing complex, multi-statement semantic equivalents. To address this limitation, we conduct the first empirical study of API replacement refactorings by mining 166,299 commits across six open-source Java projects and manually analyzing a curated subset of 1,800 commits, from which we identify 366 validated instances to characterize their scope, categories, and recurring patterns. Based on these insights, we propose AKIRA (Adaptive Knowledge Discovery and Retrieval), a hybrid framework that integrates pattern-deterministic heuristics with a refactoring-aware knowledge base to assess the practical feasibility of recommending API replacement refactorings. Our evaluation shows that AKIRA achieves 90% recall and 88% precision on a manually curated dataset. Furthermore, on the external RETIWA dataset, AKIRA significantly improves the state of the art by increasing recall from 21% to 81% and precision from 40% to 78%. These results demonstrate the effectiveness of combining static pattern matching with semantic reasoning to support the automation of recommending complex API replacement refactorings.

In this paper we describe a communication-strategy study for multi-GPU three-dimensional finite-difference time-domain computation with convolutional perfectly matched layer boundary conditions using CUDA. The metrics used to determine the most effective implementation include runtime, throughput in millions of output points per second, strong-scaling efficiency, CPML overhead, host-staged versus direct GPU-to-GPU exchange speedup, and enlarged-ghost speedup. On a single NVIDIA Quadro RTX 6000 GPU, the CPML implementation sustains 2,889--3,290 million output points per second with less than 1\% boundary-layer overhead, providing the single-GPU baseline for the multi-GPU study. The results show that direct GPU-to-GPU peer exchange is the dominant optimization with a 2.46--2.76$\times$ speedup over host-staged exchange, while enlarged ghost regions give only modest benefits because the reduced communication frequency is partly offset by redundant computation and additional memory traffic. On NVIDIA Quadro RTX 8000 GPUs, the implementation gives up to a 1.51$\times$ speedup on two GPUs for the tested strong-scaling cases, while four GPUs enable larger grids that approach or exceed single-GPU memory capacity.

Selecting appropriate machine learning (ML) configurations for malware detection is a complex, multi-criteria problem. Model choice, feature engineering, and update mechanisms must jointly satisfy operational constraints that vary across deployment contexts. This paper proposes the Framework for Decision-making (FDM) to build ML-based malware detection systems. The FDM formalises this selection process using the Weighted Configuration Compatibility Score (WCCS), a multi-criteria scoring function mapping five operational parameters (platform constraint, resource budget, response latency, update frequency, and detection sensitivity) to ranked recommendations across nine configuration dimensions. To validate the framework, four experiments were conducted on three datasets (a private Windows API dataset, the public Malimg image benchmark, and an Android static API dataset). Key results include: (i) XGBoost achieved the best accuracy-to-resource ratio in binary classification (97.46 % test accuracy, <70 MB RAM), outperforming LSTM/BiLSTM which consumed up to 2.8 GB; (ii) in multi-class classification, classical models (XGBoost 79.03 %) outperformed recurrent deep models (BiLSTM 72.27 %), reversing the binary ranking; (iii) class-incremental learning with EfficientNetB0 maintained 99.13 % accuracy with only 0.65 pp degradation across 11 incremental steps; (iv) transfer learning reduced training time by 2.14 times on average for image-based malware data without significant accuracy cost; and (v) autoencoder pre-processing yielded a 14 times training speedup at a cost of only 0.86 pp accuracy. These findings confirm that the optimal ML configuration is context-dependent, validating the FDM's core premise and demonstrating its practical utility for cybersecurity practitioners.

We consider multiple agents competing to acquire some costly divisible resource (e.g. shares of a financial asset, compute resources, etc.) over time. Leveraging a standard model for price dynamics, we propose a novel game-theoretic model for this problem, generalizing settings studied in diverse literatures. Our analysis considers different assumptions on the information available to agents. Under partial-information with a common prior (which subsumes complete information as a special case), we establish the existence, uniqueness, and efficient computability of the Bayesian Nash equilibrium (BNE), and bound the price of anarchy. Next and more generally, we consider agents with no common prior learning to act optimally given realistic market feedback from repeated interactions. We provide sufficient conditions on agents doing simultaneous learning dynamics for last-iterate convergence to the BNE. For all settings, we provide simulations based on real financial data to illustrate our theoretical results and offer new insights on strategic behavior in the context of trading and resource acquisition.

Retrieval for search agents is still inherited from non-agentic information retrieval: a retriever ranks the corpus and the agent reads a small set of returned documents. Recent direct corpus interaction (DCI) work shows that agents can instead interact with the raw corpus through shell tools such as grep and file reads. But unbounded interaction does not scale: every broad shell command is a scan over the whole corpus, and latency degrades sharply as the corpus grows. We argue that the role of retrieval for agentic search is not just to select documents that fit in the LLM context window, but to construct an interaction space: a bounded subset of the corpus the agent can explore with associated tools. Two design consequences follow. The space needs a boundary supplied by retrieval, and the objects within it should be processed for interaction. As a proof of concept, we propose RISE (Retrieving Interaction SpacE): we use BM25 to construct the interaction space; meanwhile, its documents are processed during indexing for shell-style navigation. On BrowseComp-Plus, RISE matches the pure-shell DCI baseline at 78% accuracy with gpt-5.4-mini at roughly one quarter of the per-query cost. At 1M documents, RISE-BM25 reaches 81% on gpt-5.4-mini, whereas DCI on gpt-5.4-nano degrades to 60% with 33 of 100 wall-clock failures.

Bitcoin is permissionless and does not rely on any central administrator, which gives it strong censorship resistance. At the same time, it is important to incentivize miners to behave in ways that align with the interests of the system as a whole. This paper asks whether miners are individually incentivized to propagate blocks, one of the most fundamental processes in Bitcoin. Miners collectively maintain the blockchain by generating blocks and disseminating them across the network. If miners have an incentive not to propagate some blocks, this would indicate a fundamental flaw in Bitcoin's incentive design. Although prior work has studied how propagation delays affect forks and mining rewards, it has not fully characterized miners' incentives to improve block propagation under different tie-breaking rules. To address this gap, we derive analytical reward expressions for each tie-breaking rule based on a blockchain network model that captures the effect of forks on mining fairness. These expressions explicitly characterize how block propagation delays, hashrate distribution, and tie-breaking rules jointly determine mining rewards. We then use them to analyze miners' incentives to improve block propagation. Our results show, for example, that miners have no mining-reward incentive to relay blocks generated by other miners. By contrast, under the first-seen rule, every non-majority miner is incentivized to receive other miners' blocks more quickly and to propagate its own blocks more quickly. Finally, we compare tie-breaking rules and identify a trade-off between propagation incentives and mining fairness. In particular, the first-seen rule provides the strongest incentives to reduce propagation delays, but it also worsens mining fairness the most.

The Learning Analytics (LA) community has undergone rapid development over the 15 years since the first LAK conference was held. However, while epistemological and ethical debates regarding the philosophical foundations of LA have been vigorous, metaphysical discussions have been sparse, signifying a lack of effort to derive the identity of LA from its internal principles. In this paper, we attempt to establish a metaphysics of LA by addressing the ontological question of ``What is LA?'' We do so by tracing back to LA's own definitions and principles to derive an answer from within LA itself. Specifically, we address what kind of existence the data LA operates on constitutes, identify eight agents including learners as ontological prerequisites, and clarify, via the is/ought problem, that LA does not derive norms from data. In particular, this system reveals that a class of LA practices, here termed \textit{norm-embedded LA}, conflates LA's purpose with its operations, creating an ontological tension with the first principle. We also discuss connections with related fields and the limitations of this system. The metaphysics outlined here is not imposed from outside LA, but surfaces what LA itself has always implicitly presupposed.

Developers increasingly use AI tools such as ChatGPT, Copilot, and Claude in everyday software workflows, but prior studies often evaluate LLM outputs in isolation rather than examining how developers adapt them in real projects. We analyze 35,361 GitHub code comments that explicitly reference AI use and their associated code blocks. We first open-code 500 unique comments and code blocks to derive a taxonomy of AI-assisted development activities, then annotate the full dataset using two LLM-based classifiers and aggregate predictions with Dawid-Skene expectation-maximization. We also analyze 12,996 subsequent commit messages to study how AI-assisted code evolves after introduction, and examine temporal trends from December 2022 to March 2026. Our results show that developers primarily use LLMs for code implementation, followed by code enhancement, debugging, documentation, and testing. Subsequent commits frequently involve refactoring and cleanup, feature integration and extension, and bug fixing, indicating sustained human oversight in adapting AI-assisted code. Over time, AI-referencing comments shift from direct code generation toward knowledge and conceptual support and code enhancement. These findings suggest that AI tools are becoming embedded not only as code-generation aids, but also as collaborative support mechanisms whose outputs are refined, extended, and corrected by developers over time.

With the remarkable progress of Code Large Language Models (Code LLMs) in achieving semantic correctness, execution efficiency has become an increasingly important dimension for evaluating their practical utility. However, existing approaches typically treat full programs as a single optimization target during training, without explicitly modeling the structural factors that influence efficiency. As a result, although these models can generate semantically correct code, they fail to learn, at a fine-grained level, the underlying skeleton features that lead to efficient implementations. To address this limitation, we propose SkelDPO (Skeleton-Guided Direct Preference Optimization), a skeleton-guided preference optimization framework that systematically enhances the efficiency of code generation. SkelDPO first identifies efficient and inefficient implementations from the code dataset and, through comparative analysis, locates their efficiency-prone and inefficiency-prone points, forming alignment signals between efficiency and inefficiency skeletons. During training, a joint code and skeleton preference loss is introduced, enabling the model to learn semantic correctness while reinforcing its understanding of efficiency-critical components in code. Results show that SkelDPO consistently surpasses existing methods: compared with SOTA method that relies solely on efficient and inefficient code preference optimization, it improves Pass@1, Beyond@1, and Effi@1 by 3-6%, 3-7%, and 2-5%, with greater improvements observed on complex tasks. Overall, SkelDPO provides a new perspective on skeleton-level efficiency alignment, breaking the limitation of conventional preference optimization that relies solely on correctness or efficiency pairs. All datasets and source code are publicly available at: https://github.com/icpcSkelDPO/SkelDPO.

Large Language Models (LLMs) are capable of generating syntactically correct and functionally complete programs, greatly streamlining software development. However, recent studies reveal that these programs typically execute substantially slower than human-optimized counterparts. Existing approaches to bridging this efficiency gap typically involve either iteratively optimizing code after generation or fine-tuning models on corpora of efficient code. Yet, these methods expose the model to efficiency signals only by mimicking complete, optimized solutions, without explicitly encoding the structural code patterns essential for achieving high runtime performance. Addressing this gap presents two core challenges: (1) extracting and representing latent, efficiency-oriented structural patterns embedded within complex syntax and control flows, and (2) effectively learning these patterns without destabilizing the semantic training of LLMs. To tackle these challenges, we propose EffiSkel, an efficiency skeleton-guided framework that explicitly extracts and learns efficiency skeletons-abstract, reusable structural patterns underpinning efficient code-by leveraging three complementary strategies. These skeletons are integrated into a multi-task learning regime that jointly optimizes code generation and skeleton prediction. Experiments across multiple programming languages and benchmarks demonstrate that EffiSkel significantly enhances both functional correctness and efficiency, resulting on Mercury with DeepSeek-Coder (7B) a +11.11% (vs. EffiCoder) and +3.71% (vs. CodeDPO) higher Efficiency Ratio (ER), and a +0.36 (vs. EffiCoder) and +0.22 (vs. CodeDPO) increase in Average Speedup (AS). These results highlight the effectiveness of explicitly modeling efficiency skeletons in improving the runtime performance of code generated by LLMs.

Comparing genomes is critical for discovering mutations, tracking evolutionary lineages, and advancing cross-species genomics. Fundamentally, this reduces to an O(n^2) string-matching dynamic programming (DP) problem, a challenge that has driven decades of performance research. However, executing a strict O(n^2) DP algorithm is computationally intractable for genomes spanning millions to billions of base pairs. Consequently, modern aligners rely on global heuristics to identify thousands of candidate similarity regions between species. Unfortunately, these methods are burdened by complex serial dependencies. Once candidate regions are identified, the pipeline executes localized DP alignments, which introduce their own non-trivial heuristics and irregular data dependencies. While parallelizing dense, two-dimensional DP is a well-studied problem, accelerating this end-to-end pipeline is significantly more challenging. Parallelizing across candidate regions and offloading irregular, heuristic-laden local alignments to modern hardware (such as GPUs) remains a major hurdle. In this work, we address the challenge of overcoming these serial bottlenecks by optimizing the global pipeline across regions. We take inspiration from four papers: LASTZ, SegAlign, Darwin-WGA, and SNAP, synthesizing findings across each to inform optimizations, which we either prototype or implement directly in LASTZ.

Modern infrastructure depends on externally maintained artifacts such as package-registry dependencies, CI/CD actions, container images, Terraform providers and modules, developer extensions, model artifacts, and AI tool servers. These artifacts are easy to fetch but difficult for institutions to admit, govern, and revoke. This paper proposes the Custody Envelope Threshold, an authority-scaled model of artifact admission. It argues that direct institutional admission is defensible only when object identity, ingress path, and revocation capacity are sufficiently closed relative to the execution authority delegated to the artifact. When this threshold is not met, institutions tend to proxy, policy-mediate, vendor-mediate, internalize, quarantine, or reject the artifact. The framework is operationalized as a four-condition ordinal instrument and connected to reference-monitor reasoning, least privilege, and transaction cost economics. It is applied to package dependencies, GitHub Actions, container images, Terraform providers and modules, developer extensions, and open model artifacts, with Model Context Protocol (MCP) servers treated as held-out evidence. The paper also specifies a validation design, deterministic prediction function, and OSF replication package for testing whether high-scrutiny institutions converge toward stronger custody closure for high-authority artifacts.

In this short experience paper, we present Pomona, a lightweight agentic tool that utilises agent skills for continuous automated code quality improvement. Inspired by the philosophy of Kaizen(TM), Pomona automates a cycle of discovery and incremental repair: a Scanning skill identifies improvement tasks (e.g., linting violations, technical debt markers, and test gaps) and prioritises them in a structured backlog, while a Repair skill generates tiny pull requests (PRs) targeting ~10 lines of diff. This human-in-the-loop design enables frequent, low-risk improvements while maintaining engineer trust and productivity and reducing technical debt. We evaluated Pomona through a one-month deployment in a team and a questionnaire distributed to 10 senior engineers. Our preliminary results are promising: 15 of 17 generated PRs were successfully merged with a median time-to-close of under 2 hours. Furthermore, 8/10 of surveyed engineers expressed a desire to adopt Pomona, praising small diff sizes and Pomona's focus on improving code quality. We conclude by discussing actionable insights for researchers and practitioners on strategies for effective agentic deployment in industry.

When a distributed training job slows down, the hard part is knowing where to look. Synchronization hides the cause: a stall on one rank shows up as a wait on the others, so a data delay on a single rank can surface as backward time across the group. The cheap dashboards that run all the time -- per-stage averages and maxima -- misread this, double-counting the same exposed delay or burying the slow rank in an average, while full profilers see it clearly but are far too heavy to leave on. StageFrontier is an always-on signal that closes this gap. Each rank reports only a short ordered vector of coarse stage durations -- data, forward, backward, and so on -- timed with CPU wall-clock, with no synchronized clocks and no kernel tracing. At each stage boundary, StageFrontier takes the cumulative time of whichever rank is furthest along; the increments of this frontier form an exact, additive accounting of the step's exposed time and point to the stage and rank where group-visible delay first appears, telling an operator where to aim a heavy profiler, not which fix to make. The accounting is exact, but the coarse signal alone cannot tell whether a leading stage truly caused the slowdown or merely ran alongside it; StageFrontier labels the windows where that distinction needs more evidence instead of guessing. A PyTorch implementation adds under 0.2% throughput overhead through 128 ranks on Gloo and NCCL, places injected faults among its top two suspects on all 50 rows of a hidden-rank DDP test, and recovers the same top-stage routing as PyTorch Profiler, HTA, and Nsight Systems once their traces are reduced to the same coarse stages -- from a 0.11 MB summary instead of a 15.81 GB trace.

Configuring network access control policies in large, complex networks is error-prone and requires significant expert effort. LLMs offer a promising interface for expressing such policies in natural language, but their capability for translating user requests into access policies, and the system architectures best suited to leverage LLMs, remain underexplored. We present an architecture for natural-language access control (NLAC) that uses LLMs to translate user requests into access policies, and introduce NLACBench, a benchmark for evaluating LLM-based intent translation systems in large-scale networks. Our evaluation across multiple state-of-the-art models shows that top-performing LLMs achieve up to 96.9% accuracy in small-network settings, but performance degrades substantially (below 20% for some models) as network size increases. To address this limitation, we identify relevant network components via embedding similarity and construct compact subgraphs that are passed to the LLM. This approach enables scaling to larger networks with up to 98.7% accuracy, while simultaneously reducing inference time, hardware requirements, and operating costs to a constant resource budget. Finally, a case study indicates that top-performing models exhibit largely complementary error patterns, suggesting that intent translation accuracy may be further improved through multi-LLM architectures.

This paper develops the Dynamic Gaussian Process (DGP), a framework for estimating functions governed by integro-difference equations (IDEs). IDEs model continuous functions that evolve with discrete-time dynamics and arise naturally from time-discretization of linear partial differential equations (PDEs). The DGP extends Gaussian process regression to time-varying functions and extends Kalman filtering to infinite-dimensional states. The DGP posterior remains a Gaussian process with closed-form mean and covariance updates, and separable kernel structure reduces the problem to a finite-dimensional Kalman filter on basis function coefficients. This paper extends the DGP to vector-valued states, enabling the treatment of higher-order PDEs, and provides a stability and approximation error analysis for the basis function approximation. The functional L2 estimation error decomposes exactly into in-subspace and out-of-subspace contributions, and all approximation errors vanish as the number of basis functions grows. The framework is demonstrated on the heat equation and on the wave equation, the latter with a vector-valued state. Code is available at https://github.com/JvHulst/Dynamic_Gaussian_Processes.

This paper examines what happens when GenAI tools are fully embedded in the drafting of an academic paper rather than confined to late-stage polishing. To investigate how an intensive multi-tool GenAI workflow differs from conventional academic writing, I drafted this paper from the first sentence in parallel with three GenAI tools - Claude, ChatGPT, and Gemini - comparing their outputs against my own intended contribution. Across this process, a recurring pattern took shape that I call adversarial co-thinking: using past peer reviews to calibrate the tools, then setting their outputs against one another to be tested rather than deferred to. I argue that surfacing genuine critique from tools that default to praise is a central practical challenge of working with these tools, and that the skill at stake is evaluative rather than generative. Adversarial co-thinking is a high-skill epistemic practice: it can amplify expertise where it exists, but it can also mask its absence. I further argue that current disclosure frameworks are poorly equipped to capture this shift. The paper offers four propositions for workshop discussion concerning autonomy, supervision, equity of access, and disclosure.

Proof-of-work (PoW) blockchains rely on computational expenditure to secure a ledger supporting a native cryptocurrency. In existing systems such as Bitcoin, this expenditure is intentionally useless: the computation secures consensus but produces no external economic output. An emerging alternative -- proof of useful work (PoUW) -- enables the same computation to simultaneously secure the blockchain and generate economically valuable output. However, PoUW is often criticized on economic grounds: if the work is useful, attackers might be "paid to attack," potentially weakening security. We develop a competitive-equilibrium model of a PoUW blockchain in which compute can be allocated across pure mining, pure useful work -- instantiated as machine-learning inference -- or "duplex" work that produces both with computational overheads. We provide a complete closed-form characterization of equilibrium allocations and prices as a function of the duplex overheads and a single economic parameter -- the token-inference ratio -- measuring token adoption relative to the inference market. This characterization reveals three regimes: "Bitconia," in which the economy reduces to classical PoW; "Fortessia," in which duplex replaces mining, increasing security while useful output remains unchanged; and "Duplexia," in which token rewards subsidize inference, lowering prices and expanding inference supply. Contrary to the common strawman argument, PoUW does not make attacks economically cheap: once equilibrium prices are taken into account, the economic cost of a majority attack remains tied to the block reward. Moreover, in Duplexia, block rewards act as rebates on inference prices, generating additional socially useful computation that would not arise without the blockchain -- an expansion monotonically increasing in token adoption and technological efficiency.

Modern GPU applications increasingly interact with storage systems, network devices, vendor libraries, and GPU-resident services rather than executing only isolated compute kernels. This shift creates a need for operating-system-like protection around GPU services, where service metadata, device queues, memory-mapped I/O regions, and library-internal state should not be directly exposed to untrusted application kernels. However, today's CUDA programming model, by default, still gives each application direct ownership of its CUDA context, device pointers, runtime handles, module loading path, and kernel launches, leaving protected GPU services to build their own ad hoc interfaces and isolation mechanisms. This paper presents the initial design and prototype scope of AgileOS, a GPU operating-system layer for protected CUDA services. AgileOS virtualizes CUDA at the library boundary: applications link against client-side CUDA Runtime, Driver, and selected library shims, while a trusted runtime worker owns the real CUDA context and mediates supported operations. To protect service state and module interfaces, AgileOS also defines a GPU memory-management model that separates user allocations from protected module/MMIO ranges, using pointer validation and memory access guards via PTX injection. AgileOS is modularized and flexible, supporting a range of protected services and existing libraries such as cuFFT and PyTorch. The prototype includes client-side interceptors, worker-side CUDA handlers, virtualized CUDA object tables, protected AgileOS modules, a GPU memory manager that separates user allocations from protected module/MMIO ranges, selected trusted library adapters, and the PTX-level kernel memory guard.

Motivated by the critical need for energy-efficient scheduling in cloud computing, this paper investigates Span Minimization, a fundamental variant of the well-studied BusyTime problem. In the general BusyTime problem, $n$ jobs characterized by release times, deadlines, and processing times must be partitioned into bundles of capacity $B$, where the objective is to minimize the total active duration of the virtual machines. Span minimization addresses the specific case of unbounded capacity ($B = \infty$), a problem that serves as a vital precursor for achieving high-performance approximation guarantees in more complex scheduling environments. While previous research established a deterministic $2$-approximation for interval jobs and a $3$-approximation for the general BusyTime problem, the online landscape of span minimization remains less explored. In this paper, we focus on the online version of span minimization. We demonstrate that randomization can be leveraged to break the known deterministic competitive barrier of $2$. For uniform-length jobs, we derive a randomized competitive upper bound of $\frac{1}{\ln{2}}\approx 1.443$ and a lower bound of $\frac{\sqrt{3}+1}{2}\approx 1.366$. Furthermore, we show that by introducing the ability to restart jobs, we can achieve an optimal competitive ratio equal to the golden ratio ($ϕ\approx 1.618$). Our results provide new insights into the power of randomization and flexibility in online energy-aware scheduling.

AI agents are beginning to interact not only with people, but also with one another. We investigate what happens to sentiment in such an AI-only social network: does negativity spread, or do replies calm it down? We study MOLTBOOK, a social network made up of autonomous language-model agents, using almost 2.9 million posts and 1.5 million comments. Negative posts receive many more replies than neutral or positive posts, so negativity still attracts attention. However, replies to negative content usually do not stay negative. They most often become neutral, and there is meager evidence that negative sentiment spreads across days. The main pattern is therefore not a cycle of negativity, but negative attention followed by neutralisation. These findings suggest that AI-agent networks may behave differently from human social networks: they may dampen emotional extremes, while still depending strongly on how interactions are organised.

Modern software development relies on CI/CD pipelines to automate testing, building, and deployment operations. Configuring DevOps pipelines is challenging and time-consuming, as developers must understand platform-specific syntax and manually create configuration files. This complexity can lead to configuration errors and reduced productivity, especially for developers with limited DevOps experience. This paper introduces the AutoPipelineAI system, which generates CI/CD pipeline configurations using natural language descriptions. The proposed solution uses large language models (LLMs) to translate developer intent, analyze repository structures, and create specific pipeline scripts for environments like GitHub Actions and GitLab CI/CD. It integrates repository-aware analysis, automated validation systems, and a feedback mechanism that confirms the accuracy and usability of the created pipelines. We present the system architecture, its implementation, and an assessment framework designed to measure generation precision, configuration validity, and reduction in setup effort compared to manual pipeline creation. AutoPipelineAI illustrates how LLMs can simplify the complexity of DevOps configuration and enhance developer access to continuous delivery methods. Evaluation results provide early evidence that repository-aware, natural-language-driven CI/CD generation is a viable and promising paradigm for reducing the complexity of DevOps configuration and enabling more accessible software delivery automation.

We present LinkNav, an enhanced experience for reading academic papers which makes explicit connections between related but non-adjacent passages. To create the experience, we instruct a language model to generate questions that may arise while reading a passage and then search for answer passages elsewhere in the document, forming intra-document connections when answers are found. We confirm that these building blocks work well to power the experience, with an answer detection pipeline that works with high precision, resulting in a reasonable number of connections being made for a document. On a dataset of academic papers, we find that connected passages are on average ten segments away from each other, making explicit connections that a reader may have otherwise missed.

This work provides a comparative review of three different numerical methods generally used to discretize continuous-time non-linear equations appearing in model predictive control problems: direct multiple shooting, direct collocation and successive linearizations. An overview of the characteristics of each method is given and the performance of each method is evaluated through the simulation of two test cases.

We analyze competing auctions in intermediated markets, where a seller selects among parallel mechanisms for the sale of a single good, most prominently the relay-and-protocol architecture of proposer-builder separation in Ethereum. When the intermediary can enforce single-homing on its bidders, sealed-bid second-price intermediary auctions fully unravel into the sealed first-price principal auction; open bidding-format intermediaries unravel only partially, collapsing into first-price in equilibrium under symmetric latency and sorting fast bidders to the intermediary under asymmetric latency. Any last-look advantage is removed through the availability of a credible sealed bidding channel. These results extend to multi-plexing environments (no enforcement by the intermediary). While the unraveling result indicates that the availability of a sealed first-price bidding channel pushes the overall market to the same auction structure, the very assumption of the credibility of such channel is problematic, as the seller may have an incentive to leak information: a first-price auction is leakage-resistant in the presence of a single ``fast'' bidder but not against two or more. However, if the seller can credibly commit to not leak bids, it is optimal for them to do so. A main motivation is the forthcoming Glamsterdam update of Ethereum: our analysis suggests that the availability of an in-protocol (first-price) bidding channel severely limits the design space for out-of-protocol auctions by relays and other intermediaries.

The present study provides a closed-form characterisation of Nash equilibria in N-player binary games with unidirectional dependencies. While general network games are PPAD-complete, prior work has established that trees or paths admit polynomial-time solutions via dynamic programming. We provide a deterministic characterisation for the subclass of directed cycle graphical games, demonstrating that non-zero boundary incentives linearize the topology into a feed-forward propagation. Under this Robust Incentive Structure, resolution is achieved in O(N) time: strict dominance guarantees a unique equilibrium; in its absence, pure strategy equilibria are governed by the Parity Condition, while a unique fully mixed equilibrium is guaranteed via induced payoff indifference. For non-robust regimes, we deliver branching rules. The transition-matrix formulation evaluates the search tree size beforehand. This transparency enables the inverse design of target equilibria in circular networks, making explicit the mechanics that remain opaque in numerical solvers.