arXiv:2605.12667v2 Announce Type: replace Abstract: The alignment of Large Language Models (LLMs) utilizes Reinforcement Learning from AI Feedback (RLAIF) for non-verifiable domains such as long-form question answering and open-ended instruction following. These domains often rely on LLM based auto-raters to provide granular, multi-tier discrete rewards (e.g., 1-10 rubrics) that are inherently stochastic due to prompt sensitivity and sampling randomness. We empirically verify the stochasticity of auto-raters that can propagate and corrupt standard advantage estimators like GRPO and MaxRL, as a noisy reward samples can skew normalization statistics and degrade the global learning signal. Empirically, sampling more rewards and taking majority voting may reduce the noise and improve performance, but this approach is computationally expensive. To address this bottleneck, we introduce $\textbf{O}$rdinal $\textbf{D}$ecomposition for $\textbf{R}$obust $\textbf{P}$olicy $\textbf{O}$ptimization ($\textbf{ODRPO}$), a framework that structurally isolates evaluation noise by decomposing discrete rewards into a sequence of ordinal binary indicators. By independently computing and accumulating advantages across these progressively challenging success thresholds, ODRPO prevents outlier evaluations from corrupting the global update while establishing an implicit, variance-aware learning curriculum. Empirically, ODRPO achieves robust performance on Qwen2.5-7B and Qwen3-4B models, outperforming baselines with relative improvements of upto 14.8% on FACTS-grounding-v2 and 7.5% on Alpaca-Evals. Critically, these gains are achieved with negligible training-time overhead, as ODRPO requires no additional compute per step compared to standard estimators. Supported by theoretical analysis confirming its optimization stability, ODRPO provides a scalable and robust framework for aligning models within the noisy, discrete evaluation landscape of modern RLAIF.
Science Journals
arXiv:2605.09391v2 Announce Type: replace Abstract: It is becoming increasingly necessary to have monitors check for harmful behaviors during language model interactions, but text-only monitoring has not been sufficient. This is because models sometimes exhibit strategic deception and sandbagging, changing their behavior during evaluation. This motivates the use of white-box monitors like linear probes, which can read the model internals directly. Currently, such probes can fail under distribution shift, limiting their usefulness in real settings. We study whether there exists a low-dimensional subspace of the model internals that captures harmful behaviors more robustly, while leaving out spuriously correlative features. Inspired by the Assistant Axis and Persona Selection Model, we construct persona axes for deception and sycophancy using contrastive persona prompts. The first principal components, obtained by unsupervised PCA of the persona-specific vectors, cleanly separate harmful and harmless personas. Across 10 evaluation datasets, we show that persona-derived directions transfer non-trivially and probes trained on persona-PC projections generalize better than probes trained on raw activations. We also find that a unified axis consisting of multiple harmful and harmless behaviors improves generalization across behaviors and datasets. Overall, persona vectors provide a useful inductive bias for building more transferable behavior probes.
arXiv:2605.15556v1 Announce Type: new Abstract: Large language models (LLMs) have evolved AI assistants into autonomous reasoning engines that maintain context, invoke tools, and pursue long-horizon tasks. This has spurred Agent Operating Systems (Agent OS) as kernel-like layers for lifecycle management, memory, scheduling, and access control. Yet most designs remain agent-centric, treating the OS as a single-host runtime for internal reasoning and tool use, leaving open how autonomous actions integrate with distributed, collaborative, permission-sensitive workflows. TopoClaw is an open-source, human-centric, topology-aware Agent OS modeling the user's ecosystem as two coupled structures: a physical device topology of heterogeneous surfaces and a social relationship topology of shared spaces, teams, and delegated roles. It unifies device operation, messaging, and skills around accountable cross-boundary execution, with three core contributions: (1) cross-device action placement, decoupling intent from actuation and routing distributed actions across the device cluster based on hardware affordances and user context; (2) cross-user identity attribution, treating agents as socially situated "Digital Twins" that coordinate in multi-user spaces while preserving provenance, role-aware permissions, and human accountability; (3) cross-context authority governance, pairing broad capability with distributed, context-aware policy enforcement across physical and social trust boundaries to bound proactive autonomy at the OS layer. This report presents TopoClaw as an engineering-oriented reference architecture, covering its design principles, runtime, cross-device execution, collaboration mechanisms, security model, and deployment outlook.
arXiv:2603.23351v2 Announce Type: replace Abstract: Laser-enhanced contact optimization (LECO) has emerged as an important method for simultaneously reducing contact resistivity and metallization-induced recombination in advanced crystalline silicon solar cells, thereby enabling concurrent gains in fill factor and open-circuit voltage, particularly in TOPCon devices. However, broader industrial transferability remains constrained by the need to preserve these gains within a narrow process window and by unresolved, architecture-dependent questions regarding the kinetic stability of some LECO-modified interfaces. LECO is therefore examined in this review as a coupled multiphysics process that links localized electrothermal activation and microstructural evolution to device-level electrical signatures through an instantaneous regime map and a reliability classification based on time-dependent drift. A predictive workflow is outlined that couples transient electrothermal modeling with reduced state metrics, including effective diffusion depth and local areal energy density, and propagates calibrated thresholds across the recipe space. The framework separates stable optimization from marginal activation and latent damage, while explaining why fine-line scaling and copper-containing contact stacks can tighten stability margins through current localization and diffusion-barrier constraints. These insights provide a basis for reliability-aware process-window design and future digital-twin-assisted optimization of LECO for scalable, high-efficiency silicon photovoltaics.
arXiv:2605.15455v1 Announce Type: new Abstract: Chatbot behavior is often opaque to users, as responses can shift unpredictably across a conversation, drifting toward sycophancy, toxicity, or other unsafe responses. This can leave users vulnerable, either being misled by overly agreeable AI or manipulated by a harmful chatbot that no longer behaves as intended. To address this, we introduce multi-turn neural transparency, an interface that surfaces an LLM's internal neural activations in real time to help users anticipate and recognize how behaviors change across turns. We construct behavioral vectors for six personality traits using methods from mechanistic interpretability, identifying directions in activation space that correlate with trait expression ($R^2 \geq 0.9$) via contrastive system prompts, and visualize trait expression using a sunburst and drift panel that updates at each turn. In a randomized controlled study (N = 246), participants predicted trait expression from a system prompt alone, then rated observed behavior after interacting with the chatbot for both assistant and role-play personas. We find that participants without visualization struggled to accurately evaluate traits (RMSE $\approx$ 0.6-0.7), while the inclusion of neural transparency significantly improved both anticipation and evaluation compared to no visualization (d = -0.34 to -0.49). The multi-turn dynamic visualization additionally outperformed the static single-turn visualization on holistic evaluation of model behavior (d = -0.32). Transparency also reduced overconfidence: participants without visualization grew more confident despite no gain in accuracy. These findings suggest that surfacing internal model representations to everyday users is a meaningful step toward more transparent and informed human-AI interaction.
arXiv:2510.02453v3 Announce Type: replace Abstract: Frontier language models are deployed as black-box services, where model weights cannot be modified and customization is limited to prompting. We introduce Advisor Models, a method to train small open-weight models to generate dynamic, per-instance natural language advice that improves the capabilities of black-box frontier models. Advisor Models improve GPT-5.2's performance on RuleArena (Taxes) by 27.4%, reduce Gemini 3 Pro's steps taken in SWE agent tasks by 24.6%, and outperform static prompt optimizers in personalizing GPT-5 to user preferences (85-100% vs. 40-60%). We also find that advisors are transferable: an advisor trained with a low-cost student model still transfers improvements to a frontier model. Moreover, Advisor Models are robust: we observe no degradation on other benchmarks than the pipeline is trained on. Our method shows how to perform parametric optimization for black-box frontier models in a practical and cost-effective way.
arXiv:2510.05676v2 Announce Type: replace Abstract: Graph-based methods are becoming increasingly popular in machine learning due to their ability to model complex data and relations. Insurance fraud is a prime use case, since fraudulent claims are often the result of organised criminals that stage accidents or the same persons filing erroneous claims on multiple policies. One challenge is that graph-based approaches struggle to find meaningful representations of the data because of the high class imbalance present in fraud data. In addition, insurance graphs are heterogeneous and dynamic, given the changing relations among people, companies and policies. As a result, gradient-boosted tree approaches on tabular data still dominate the field. Therefore, we present a novel inductive graph gradient boosting machine (G-GBM) for supervised learning on heterogeneous and dynamic graphs. G-GBM combines the class-imbalance robustness of gradient boosting with heterogeneous graph information encoded through interpretable path-level feature concatenations, while preserving access to the original tabular feature space. In addition, the explicit representation of neighbourhood information enables transparent SHAP-based explanations at the metapath and feature level. We demonstrate G-GBM for insurance fraud detection on an open-source and a real-world, proprietary dataset, and find that G-GBM performs on par or better than the state-of-the-art. The associated insurance fraud dataset is publicly released to facilitate reproducibility.
arXiv:2603.20030v2 Announce Type: replace Abstract: A triangulation of a surface is k-irreducible if every non-contractible curve has length at least k and any edge contraction breaks this property. Equivalently, every edge belongs to a non-contractible curve of length k and there are no shorter non-contractible curves. We prove that a k-irreducible triangulation of an orientable surface of genus g has $O(k^2g)$ triangles, which is optimal. This is an improvement over the previous best bound $k^{O(k)} g^2$ of Gao, Richter and Seymour [Journal of Combinatorial Theory, Series B, 1996].
arXiv:2604.21251v5 Announce Type: replace Abstract: Large language models (LLMs) trained on unfiltered corpora inherently risk retaining sensitive information, necessitating selective knowledge unlearning for regulatory compliance and ethical safety. However, existing parameter-modifying methods face fundamental limitations: high computational costs, uncontrollable forgetting boundaries, and strict dependency on model weight access. These constraints render them impractical for closed-source models, yet current non-invasive alternatives remain unsystematic and reliant on empirical experience. To address these challenges, we propose the Controllable Alignment Prompting for Unlearning (CAP) framework, an end-to-end prompt-driven unlearning paradigm. CAP decouples unlearning into a learnable prompt optimization process via reinforcement learning, where a prompt generator collaborates with the LLM to suppress target knowledge while preserving general capabilities selectively. This approach enables reversible knowledge restoration through prompt revocation. Extensive experiments demonstrate that CAP achieves precise, controllable unlearning without updating model parameters, establishing a dynamic alignment mechanism that overcomes the transferability limitations of prior methods.
arXiv:2604.26086v2 Announce Type: replace Abstract: Accurate quantification of protein-nanoparticle interactions is essential for applications in nanobiotechnology, nanomedicine, and drug delivery. Motivated by recent computational and experimental work, we combine coarse-grained united-atom (UA) models with molecular docking to characterize protein adsorption on SiO_2 nanoparticles. We construct orientation-resolved heatmaps in which polar and azimuthal angles uniquely specify the relative protein-nanoparticle pose, and the map amplitude reports binding propensity via the minimum UA adsorption energy or the docking score. Each angular bin corresponds to a distinct docked complex, enabling systematic comparison of binding geometries across models. To relate docking score landscapes to Boltzmann-averaged UA adsorption energetics, we analyze eight birch pollen allergen proteins previously studied experimentally. Similarity between the two orientational distributions is quantified using the Jensen-Shannon divergence (JSD). We find encouraging agreement between the two approaches in several cases, while also identifying limitations and routes for improvement, including optimized angular resolution and iterative refinement of interaction parameters. Overall, this framework provides a quantitative bridge between coarse-grained energetics and docking outputs at protein-nanoparticle interfaces, supporting improved predictive modeling and mechanistic insight into protein-nanoparticle binding landscapes.
arXiv:2605.15458v1 Announce Type: new Abstract: Video diffusion models have made rapid progress in perceptual realism and temporal coherence, but they remain primarily optimized for plausible generation rather than verifiable reasoning. This limitation is especially pronounced in tasks where generated videos must satisfy explicit spatial, temporal, or logical constraints. Inspired by the role of reinforcement learning with verifiable rewards (RLVR) in reasoning-oriented language models, we introduce VideoRLVR, a practical recipe for optimizing video diffusion models with rule-based feedback. VideoRLVR formulates video reasoning as the generation of verifiable visual trajectories and consists of an SDE-GRPO optimization backbone, dense decomposed rewards, and an Early-Step Focus strategy for efficient training. The Early-Step Focus strategy restricts policy optimization to the early denoising phase, reducing training latency by about 40% while preserving performance. We evaluate VideoRLVR on Maze, FlowFree, and Sokoban, three procedurally generated domains with objective success criteria. Across these tasks, VideoRLVR consistently improves over supervised fine-tuning baselines, with dense decomposed rewards proving especially important in low-success-rate settings. Our RL-optimized model also outperforms the evaluated proprietary and open-source video generation models on these verifiable reasoning benchmarks and out-of-domain benchmarks. These results suggest that verifiable RL can move video models beyond perceptual imitation toward more reliable rule-consistent visual reasoning.
arXiv:2604.26733v4 Announce Type: replace Abstract: Live future prediction refers to the task of making predictions about real-world events before they unfold. This task is increasingly studied using large language model-based agent systems, and it is important for building agents that can continually learn from the real world. It can provide a large number of prediction questions grounded in diverse real-world events, while preventing answer leakage. To leverage the advantages of future prediction, we present FutureWorld, a live agentic reinforcement learning environment that closes the training loop between prediction, outcome realization, and parameter updates. Specifically, we modify and extend verl-tool, resulting in a new framework that we call verl-tool-future. Unlike standard reinforcement learning training frameworks that rely on immediate rewards, verl-tool-future stores prediction-time rollouts, backfills rewards after real-world outcomes become available, and then replays the completed trajectories for policy update. Across three open-source agents, successive FutureWorld training rounds lead to consistent improvements in prediction accuracy, probabilistic scoring, and calibration, demonstrating that delayed real-world outcome feedback can serve as an effective reinforcement learning signal.
arXiv:2603.13608v2 Announce Type: replace Abstract: The concept of matrix D-stability plays an important role in applications, ranging from economic and biological system models to decentralized control. Here we provide necessary and sufficient Lyapunov-type conditions for the robust (block) D-stability property. We leverage this characterization as part of a novel Lyapunov analysis of decentralized integral control for MIMO LTI systems, providing sufficient conditions guaranteeing stability under low-gain and under arbitrary connection and disconnection of individual control loops.
arXiv:2602.16363v2 Announce Type: replace Abstract: We study reward-free and reward-agnostic exploration in episodic finite-horizon Markov decision processes (MDPs), where an agent explores an unknown environment without observing external rewards. Reward-free exploration aims to enable $\epsilon$-optimal policies for any reward revealed after exploration, while reward-agnostic exploration targets $\epsilon$-optimality for rewards drawn from a small finite class. In the reward-agnostic setting, Li, Yan, Chen, and Fan achieve minimax sample complexity, but only for restrictively small accuracy parameter $\epsilon$. We propose a new algorithm that significantly relaxes the requirement on $\epsilon$. Our approach is novel and of technical interest by itself. Our algorithm employs an online learning procedure with carefully designed rewards to construct an exploration policy, which is used to gather data sufficient for accurate dynamics estimation and subsequent computation of an $\epsilon$-optimal policy once the reward is revealed. Finally, we establish a tight lower bound for reward-free exploration, closing the gap between known upper and lower bounds.
arXiv:2605.00424v2 Announce Type: replace Abstract: Agent skills - structured packages of instructions, scripts, and references that augment a large language model (LLM) without modifying the model itself - have moved from convenience to first-class deployment artifact. The runtime that loads them inherits the same problem package managers and operating systems have always faced: a piece of content claims a behavior; the runtime must decide whether to believe it. We argue this paper's central thesis up front: a skill is untrusted code until it is verified, and the runtime that loads it must enforce that default rather than infer trust from a signature, a clearance, or a registry of origin. Without skill verification, a human-in-the-loop (HITL) gate must fire on every irreversible call - which is operationally untenable and degrades into rubber-stamping at any non-trivial scale. With skill verification treated as a separate, gated process, HITL fires only for what is unverified, and the system becomes sustainable. We give a trust schema that includes an explicit verification level on every skill manifest; a capability gate whose HITL policy is a function of that verification level; a biconditional correctness criterion that any candidate verification procedure must satisfy on an adversarial-ensemble exercise; and a portable runtime profile with ten normative guidelines abstracted from a working open-source reference implementation. The contribution is harness- and model-agnostic; nothing here requires retraining, fine-tuning, or proprietary infrastructure.
arXiv:2508.17218v3 Announce Type: replace Abstract: Reliable path planning in stochastic transportation networks requires decisions that account for uncertain and correlated travel times on irregular road graphs, rather than only minimizing expected delay. Such networks exhibit strong spatial-temporal coupling, where link travel times evolve as stochastic processes over graph edges, making the problem inherently sequential under uncertainty. Existing stochastic on-time arrival (SOTA) methods primarily depend on the current node and remaining budget, which limits their ability to exploit trajectory-level temporal structure and history-dependent correlations. This work proposes GPG-HT, a history-aware graph-signal policy framework that integrates a Decision Transformer with generalized policy gradient optimization for reliable routing. By attending to historical node-edge-time observations, GPG-HT captures non-Markovian spatial-temporal dependencies and enables context-aware decision making under uncertainty. Experiments on the Sioux Falls and Anaheim networks demonstrate consistent gains in on-time arrival probability over representative optimization and reinforcement learning baselines.
arXiv:2605.15467v1 Announce Type: new Abstract: Conversational nurse-patient transcripts contain actionable observations, but converting these transcripts into structured representations at scale remains challenging. Documentation burden is substantial, with prior studies showing clinicians spend large portions of their workday on documentation and related desk work rather than direct patient care. MEDIQA-SYNUR focuses on observation extraction from conversational nurse-patient transcripts, requiring systems to normalize these narratives into a predefined schema with value-type constraints. We propose a modular retrieval-augmented generation (RAG) pipeline that uses the training set as an exemplar corpus, combines schema-constrained prompting (full schema vs. pruned candidate schema), deterministic schema-based postprocessing, and a second-pass audit, with two LLM backbones: Llama-4-Scout-17B-16E-Instruct and GPT-5.2 with corresponding embedding models for RAG. Our best configuration uses GPT-5.2 with full schema, RAG, and a second-pass auditing, achieving 80.36% F1 score. Overall, our results show that RAG consistently improves performance, while the optimal degree of schema constraint depends on the model, and second-pass auditing yields modest additional gains by correcting residual schema-adherence errors.
arXiv:2605.15393v1 Announce Type: new Abstract: As large language models (LLMs) are increasingly deployed to perform tasks with minimal human oversight, it is crucial that these models operate robustly. In particular, a model that can solve a given problem should not fail simply because certain entities$\unicode{x2013}$such as names, numbers, or other contextual details$\unicode{x2013}$have changed while the underlying problem logic remains the same. Prior work suggests that current LLMs still struggle with this form of robustness: they often succeed on some variations of a problem but fail on others. However, existing evaluations often lack a systematic way to identify which logic-preserving variations are most likely to induce failure. Instead, they typically test a random subset of allowable variations, which can overstate robustness. To address this gap, we introduce logic-preserving difficulty scaling (LPDS), a framework that (i) quantifies the difficulty of a problem variation and (ii) systematically searches the space of allowable variations to find those that maximize difficulty and expose failures. We show that as difficulty increases, performance declines and errors in the models' reasoning chains become more pronounced. We further demonstrate that LPDS efficiently finds difficult problem variations for a model, resulting in performance drops up to 5 times larger compared to random sampling. Finally, we show that fine-tuning on more difficult variations leads to more consistent robustness gains than training on easier ones.
arXiv:2601.19562v2 Announce Type: replace Abstract: Quality diversity (QD) is a branch of evolutionary computation that seeks high-quality and behaviorally diverse solutions to a problem. While adversarial problems are common, classical QD cannot be easily applied to them, as both the fitness and the behavior depend on the opposing solutions. Recently, Generational Adversarial MAP-Elites (GAME) has been proposed to coevolve both sides of an adversarial problem by alternating the execution of a multi-task QD algorithm against previous elites, called tasks. The original algorithm selects new tasks based on a behavioral criterion, which may lead to undesired dynamics due to inter-side dependencies. In addition, comparing sets of solutions cannot be done directly using classical QD measures due to side dependencies. In this paper, we (1) use an inter-variants tournament to compare the sets of solutions, ensuring a fair comparison, with 6 measures of quality and diversity, and (2) propose two tournament-informed task selection methods to promote higher quality and diversity at each generation. We evaluate the variants across three adversarial problems: Pong, a Cat-and-mouse game, and a Pursuers-and-evaders game. We show that the tournament-informed task selection method leads to higher adversarial quality and diversity. We hope that this work will help further advance adversarial quality diversity. Code, videos, and supplementary material are available at https://github.com/Timothee-ANNE/GAME_tournament_informed.
arXiv:2601.13039v2 Announce Type: replace Abstract: In this work, we study projection-based model order reduction (MOR) for switched linear systems (SLS) in control form, where the projection matrices are obtained from the solutions of generalized Lyapunov equations (GLEs). We investigate how numerical inaccuracies in solving the GLEs propagate through the MOR process and impact the accuracy and reliability of the resulting reduced-order model. This highlights the importance of accounting for such inaccuracies, motivating the introduction of a novel error bound to quantify and control the error in the approximation of the GLE solution. Moreover, classical balanced truncation error estimates for SLS are neither theoretically sound nor practically applicable, as they rely on restrictive assumptions requiring several linear matrix inequalities (LMIs) to be satisfied exactly by numerically computed GLE solutions. To address these limitations, we propose a new MOR framework for SLS, termed piecewise balanced reduction (PBR). The approach is based on solving multiple GLEs and constructing projection matrices that are piecewise constant in time. By extending the standard balanced truncation error bound for SLS, we show that the PBR framework effectively controls errors arising from inexact LMI satisfaction. In addition, the proposed error bound captures the influence of the piecewise constant in time projection matrices. Altogether, this makes the PBR approach applicable to a broad and flexible class of switched linear systems. Numerical experiments are presented to support the theoretical results.
arXiv:2512.16933v4 Announce Type: replace Abstract: The dynamics of a degenerate spherically symmetric wormhole in a vacuum is considered. An extension of the equivalence principle to matter free objects that are the source of a gravitational field is proposed. Using the Klinkhamer metric as an example, it is shown that a degenerate wormhole is precisely such an object. Application of the extended equivalence principle reduces the radial dynamics of the Klinkhamer wormhole to the dynamics of the radial fall of a test particle in a Schwarzschild gravitational field. It is proven that any bound state of the traversable Klinkhamer wormhole eventually collapses into a nontraversable Einstein-Rosen wormhole. An estimate is presented showing that the traversable Klinkhamer wormhole, although nonstationary, is a longlived state.
arXiv:2605.15503v1 Announce Type: new Abstract: Microarchitectural attacks continue to evolve, uncovering new exploitation vectors in modern processors. From a defensive perspective, assessing a system's susceptibility to such attacks remains challenging. Developing functional attack implementations is labor-intensive, requires deep microarchitectural expertise, and is highly sensitive to execution environments. Consequently, existing attacks often lack portability, limiting systematic and scalable vulnerability assessment. Recent advances in large language models (LLMs) suggest a potential avenue for lowering these barriers. However, it remains unclear whether LLMs can reliably generate functionally correct microarchitectural attack code suitable for rigorous vulnerability testing. In this work, we present uGen, the first LLM-driven framework for automated microarchitectural attack code generation. A key challenge we address is identifying attack-specific knowledge gaps in LLMs. Through a systematic study of state-of-the-art models (GPT, Claude, and Qwen3), we find that LLMs frequently misgenerate or misplace critical attack primitives. Guided by this analysis, uGen employs a retrieval-augmented, multi-agent design that injects missing domain knowledge to synthesize functionally correct microarchitectural attack PoCs tailored to defender requirements. We evaluate uGen on cache-based and speculative-execution attacks across diverse set of microarchitectures, vulnerable functions, and LLM platforms. In the deployment stage, uGen achieves up to 100% success rate for Spectre-v1 (Claude Sonnet-4) and 80% for Prime+Probe (Qwen3-Coder). Finally, we demonstrate that uGen can generate a successful PoC code with a cost of $1.25 in under four minutes.
arXiv:2605.15468v1 Announce Type: new Abstract: Digital systems have become simultaneously more powerful and more wasteful. Features accumulate that nobody uses. Data is collected that nobody analyzes. AI is deployed at significant energy and water costs for gains that a simpler approach could have achieved. And through all of it, the people who depend on these systems quietly absorb the consequences in cognitive load, lost time, and eroded trust. This paper introduces GreenZ, a three-layer Sustainable UX Framework for complex digital systems. Its three layers are a Philosophy Layer built around ten published principles, an Operational Frameworks Layer comprising five applied systems, and a Tools and Canvases Layer of practical audit instruments and decision models. Two contributions sit at the framework's core: a Digital Waste Taxonomy classifying eight distinct waste types, and an AI Sufficiency Decision Model that asks whether AI should exist in a given flow before any question of how to implement it. GreenZ v1 is theoretically grounded but empirically unvalidated. A practitioner expert review study is underway at the time of submission. The paper presents the framework's architecture, its conceptual foundations, its position relative to existing literature, and an honest account of what remains to be established.
arXiv:2512.06655v3 Announce Type: replace Abstract: Sparse autoencoders (SAEs) are increasingly used to extract activation directions for inference-time steering, but their standard sparsity objective treats latent features as independent. This prior can be poorly matched to high-level safety behaviors, where refusal and harmful compliance appear to depend on distributed structure in activation space. We introduce Graph-Regularized Sparse Autoencoders (GSAE), a dictionary-learning method that learns safety-steering directions by smoothing SAE decoder vectors over a neuron co-activation graph and applying the resulting direction bank through a two-gate runtime controller. Empirically, GSAE improves selective refusal across JailbreakBench, HarmBench, and XSTest, increasing harmful-request refusal while keeping benign-prompt refusals low. On Llama-3-8B, replacing the standard SAE with GSAE in an otherwise identical pipeline improves $\Delta_s$ by $20.1$ points on JailbreakBench and $16.8$ points on HarmBench. GSAE outperforms activation-steering baselines and black-box guardrails, preserves benign-task performance, generalizes across Llama-3, Mistral, Qwen 2.5, and Phi-4, and remains strong under black-box and gray-box jailbreak attacks.
arXiv:2605.15322v1 Announce Type: new Abstract: As generative AI (GenAI) systems become increasingly proficient at simulating human-like and well-reasoned text, users may attribute authority to AI outputs, shaping how they engage with writing and reasoning tasks. While prior work has raised concerns about AI overreliance, empirical approaches for observing this phenomenon during open-ended writing remain limited. In this paper, we examine how GenAI assistance influences users' interactions with AI suggestions during writing. We report results from a mixed-methods study in which 47 participants completed analysis and synthesis writing tasks with or without AI assistance. We quantify the textual overlap between AI suggestions and participants' writing and analyze participants' reflections. Our results show that AI assistance is associated with patterns of suggestion reuse. Building on these findings, we design and evaluate an interactive writing interface that may support reflection on the usage of the AI suggestions during writing. Evidence from a small follow-up think-aloud study (n = 4) suggests that the interface can increase users' awareness of how AI outputs are incorporated into their writing and may support more conscious engagement with AI assistance. Together, our findings contribute empirical methods for studying AI adoption in writing contexts and demonstrate how interface design can shape user-AI interaction.