Papers

The development of decision-pretrained transformers, algorithm distillation, long-context meta-RL, and retrieval-augmented agents has renewed interest in in-context reinforcement learning (ICRL): the ability of a pretrained or fine-tuned decision model to infer latent task rules and improve future behavior from interaction context, without test-time parameter updates. This line of work asks when trial-and-error evidence, rewards, transitions, demonstrations, feedback, or retrieved experience can make learning-like computation happen inside the context window. However, existing surveys of ICRL mainly organize the field around pretraining objectives, architectures, context formats, evaluation protocols, and theoretical mechanisms, while the non-stationary setting remains comparatively underexamined. In changing environments, accumulated context is not merely more evidence about a fixed task: the reward specification, transition kernel, observation channel, action interface, constraint model, or demonstration and memory distribution can fall out of alignment with the current regime. Previously useful context can therefore become stale, misleading, or useful again when an old regime returns. We survey non-stationary ICRL as the problem of adapting through context while deployed policy parameters remain fixed: the policy must infer both the current decision rule and which parts of its accumulated evidence still support that rule. We define non-stationary ICRL, relate it to meta-RL, decision sequence modeling, retrieval-augmented RL, value- and model-aware ICRL, and reward-feedback agents, and organize the literature along three questions: what changes, how the change unfolds, and how observable the change is to the agent.

We present Charlie, an on-premise multi-agent Retrieval-Augmented Generation (RAG) system for structured evidential processing in digital forensic environments. Contemporary forensic workflows must handle large volumes of heterogeneous and unstructured documents under strict requirements of traceability, confidentiality, and legal compliance. Charlie addresses this challenge through a controlled agent architecture that combines local retrieval, task decomposition, structured memory, and verification mechanisms. Unlike cloud-based systems, it operates entirely within institutional infrastructure, preserving data sovereignty and evidential integrity. We describe the systems architecture, including its transition from classical RAG to agent-based orchestration, and demonstrate its application in real-world forensic scenarios. Case studies show that Charlie enables scalable multi-document data extraction and supports longitudinal forensic intelligence generation while maintaining traceability and auditability. Our results indicate that agent-orchestrated, on-premise RAG architectures can effectively support evidential workflows without compromising legal and institutional constraints. Charlie provides a practical and reproducible blueprint for deploying AI systems in high-stakes forensic environments. This manuscript is an archival version of a paper presented at the RELAF 2026 Workshop.

Generating structured artifacts with Large Language Models - e.g.\ database queries, threat framework mappings, entity schemas - is relatively straightforward; however, making them reliable enough for production deployments presents challenges. We present TAG, a lightweight framework based on a core principle: LLMs generate, we validate. This reframing shifts responsibility from generation quality to validation rigor. The framework rests on three key attributes: First, test driven generation: when tests fail, the LLM receives indicative error messages that expose why the output failed, enabling the LLM to understand its mistakes and refine subsequent attempts. Second, deterministic and LLM-based tests: deterministic tests catch heuristics that can be programmatically verified (schema, syntax, cross-reference), while LLM-based tests evaluate nuanced semantic and delicate features that resist programmatic inspection (intent alignment, logical consistency, domain correctness). Third, expert-distilled judges: LLM-based tests are calibrated to distill and replicate human expert decision distribution, transforming manual human quality gates into scalable, reusable evaluation proxies that reflect professional-grade validation standards. We demonstrate the framework on three artifact types in the security domain - KQL query generation, MITRE ATT&CK mapping, and entity mapping - deployed in production at Microsoft Sentinel. We believe this framework can be applied beyond security to other artifact generation tasks, providing a path to reliable, high-quality outputs without sacrificing the efficiency gains of LLM generation.

Vision Transformers achieve strong image classification accuracy but process all image regions with nearly the same computation, even when many regions are redundant or uninformative. Recent adaptive inference methods reduce this cost by selectively compressing tokens or terminating inference early, but combining these mechanisms often causes unstable intermediate representations and accuracy degradation. We introduce Fusion, a unified adaptive inference framework that coordinates token merging, early exiting, and token pruning through a simple staged design: tokens are merged first, confidence is evaluated next, and pruning is applied only to samples that continue inference. This ordering allows the three mechanisms to operate cooperatively rather than competitively. Fusion further includes lightweight routing modules that adapt compression strength to each input and support inference-time adjustment of the accuracy–latency trade-off without retraining. On ImageNet-1k with DeiT-S, Fusion matches or surpasses state-of-the-art adaptive ViT methods at comparable compute budgets while reducing calibration error by up to $4\times$ and inference energy by $48\%$. Experiments across ImageNet-100, CIFAR-100, and ImageNette with multiple ViT backbones demonstrate consistent transferability without dataset-specific tuning.

For decades, data engineering has developed mature architectural principles for integrating, governing, validating, cataloging, and serving organizational data. The rise of large language models does not eliminate these concerns; it exposes a broader version of them. Organizational knowledge is becoming executable infrastructure: systems increasingly retrieve it, assemble it, reason over it, and act on it. This paper argues that enterprise artificial intelligence (AI) systems suggest a transition toward an architectural discipline for representing, maintaining, governing, and operationally delivering organizational knowledge. We refer to this discipline as knowledge architecture. We offer a conceptual model and taxonomy showing how classical data-engineering guarantees must be redefined when the managed unit shifts from records to knowledge artifacts: extract, transform, and load (ETL) becomes knowledge ingestion, change-data capture (CDC) becomes knowledge change detection, lineage becomes provenance, catalogs become knowledge catalogs, materialized views become knowledge views, and medallion architectures become raw–curated–operational knowledge layers. Emerging formats such as large language model (LLM) Wiki and the Open Knowledge Format (OKF) are treated as early evidence of this transition, not as its endpoint. The central claim is that knowledge architecture becomes useful when organizational knowledge ceases to be a passive information resource and becomes an operational asset used by humans, agents, workflows, and models to execute work.

Tool-using LLM agents are usually evaluated by final-answer correctness or LLM judges. Neither captures how an answer was produced. In safety-critical settings, the procedure itself is part of correctness. In this paper, we introduce AgentLTL, a language derived from First-Order Linear Temporal Logic (FO-LTL) that expresses procedural rules over agent traces. It yields a deterministic, judge-free compliance score. In this framework, a single specification drives two usages. The first is harnessing: the constraints score completed traces, or gate tool calls by checking each prefix online, before execution. The second is finetuning: the score serves as a dense reward. On a benchmark spanning ordering, branching, iteration, and grounding, block-and-warn harnessing improves compliance on five of seven models. Finetuning with the same reward yields +38 and +17.5 percentage point gains in accuracy and compliance on held-out patterns, including unseen tool-name aliases. These findings are consistent with the model acquiring procedural structure rather than memorizing surface tool names and procedures.

Deep learning is widely applied in medical image analysis, but up to 10% of manually labelled images may be incorrect, degrading model performance. This paper proposes an automated method to identify incorrectly labelled medical images by analyzing sequences of loss functions from deep learning classification networks over multiple training epochs. Identified images can be reviewed and relabelled by experts, improving dataset quality and model performance. Two experiments validate the method on a fundus image dataset for referable diabetic retinopathy screening. In the first, 6% (648) of 10,788 gold-standard labels were intentionally flipped. The method identified 75.31% (488) of the flipped samples, with only 4.85% (492) false positives among correctly labelled samples. In the second, reviewing and correcting the 980 identified samples (9.1% of the dataset) and retraining the model improved best accuracy on an independent test set from 95.93% (with 6% label noise) to 96.50% (with 1.5% noise), approaching the ideal 96.57% (with 0% noise). The results demonstrate the method's effectiveness in improving model performance through automated label quality control.

Multimodal large language models excel on short clips but struggle on hour-long videos in an online setting, where frames are processed incrementally under limited memory. Existing online methods either retain compact visual representations that lack semantic structure, or build higher-level memory stores organized around temporal proximity rather than explicit causal links, leaving multi-hop narrative reasoning to be reconstructed by the LLM at every query. We bridge this gap with Homer, a Hierarchical Online Memory Exploration and Reasoning framework. Homer's memory mirrors the multi-scale structure of long videos, ranging from raw perception, to recurring entities, to events connected by explicit temporal and causal relations. Its agentic reasoner then explores this memory the way humans do, locating the relevant scene, looking up details, and composing the answer through multi-round memory retrieval, with a harness that verifies and corrects each step. Homer outperforms the previous best agent method by $+5.5$, $+10.8$, and $+4.4$ points on M3-Bench-robot, M3-Bench-web, and Video-MME-Long, and consistently lifts three various LLM backbones, indicating a model-agnostic structural capability for grounded retrieval over long videos.

Vision-language models (VLMs) have achieved strong performance across diverse multimodal tasks, yet they remain vulnerable to unreliable reasoning. Existing self-correction methods mitigate these issues but typically rely on post-training or carefully engineered feedback, incurring high computational cost. In this work, we revisit this challenge through the lens of emotional cues, asking whether they can activate latent self-correction behaviors in VLMs without additional training. We find that emotional signals serve as an effective trigger for self-correction, encouraging more cautious and reflective reasoning. Motivated by this finding, we propose \escabstract (\textbf{\underline{E}}motional \textbf{\underline{S}}elf-\textbf{\underline{C}}orrection), a training-free self-correction framework. ESC introduces an external verifier that detects potentially incorrect initial responses and injects emotional feedback to encourage model to reflect, and produce a better revised response without additional training. Extensive experiments across safety, hallucination, vision-centric perception, and multimodal reasoning benchmarks show that ESC consistently improves reliability while preserving overall model utility. These results suggest that emotion can function not only as an ability to be recognized, but also as a practical control signal for scalable self-correction in VLMs. We therefore believe that ESC provides a strong foundation for a new reliable human-like, emotion-integrated research direction. Our project is publicly available at \textcolor{red}{https://genai4e.github.io/ESC/}.

Learning a useful world model from minimal interaction is central to building agents that adapt to unfamiliar tasks. Programmatic world modeling approaches such as WorldCoder quickly learn transition models when given pre-supplied symbolic representations; however, such approaches are insufficient when the symbolic representation and goal are unknown and continually changing, and are intractable under large action spaces due to rigid planning heuristics. Extending the programmatic world modeling approach, OPINE-World additionally maintains an abstracted representation in which provisional objects, incomplete causal explanations, and unresolved interactions can exist in a Bayesian, exploration-centric hypothesis space. This enables learning an unknown, dynamically evolving state, transition, and goal. Exploration and program revision share this evolving abstraction, allowing new evidence to revise both the hypothesized mechanics and their implementation in tandem at test time. ARC-AGI-3 presents this open-world problem as a general-intelligence learning benchmark. Through our OPINE-world model learning paradigm, we improve the Relative Human Action Efficiency score of Claude Opus 4.8 high from 1.5% to 78.40% on the ARC-AGI-3 benchmark.

Understanding user intent is fundamental to delivering relevant search results in e-commerce. However, substantial fraction of real-world queries are under-specified (e.g., "watch" or "shirt"), lacking explicit attributes such as gender or age group. This ambiguity poses a significant challenge for query intent detection models in e-commerce search systems, which must accurately infer latent user intent (e.g., age, gender) to support effective downstream retrieval. We introduce IntentTune, a framework for resolving ambiguous or under-specified query intents by leveraging either (1) user-specific behavioral signals including search history, browsing activity, and profile attributes or (2) population-level demand patterns aggregated across all users. Through experiments on real-world e-commerce data, we first demonstrate that population-level demand patterns alone are insufficient to reliably infer intent in under-specified queries. We then demonstrate that user-specific behavioral signals – particularly prior search queries – outperform both population-level statistics and static profile information for inferring gender, age group, product category, and size intent from underspecified queries.

Recurrent memory agents extend LLMs to arbitrarily long contexts by iteratively consolidating input into a fixed-size memory window. Despite their scalability, these agents exhibit a well-documented reliability problem: end-to-end performance degrades systematically as context length grows. We diagnose this failure by decomposing performance into two factors–memory capture and memory retention–and quantitatively confirm that retention is the dominant bottleneck. Retention collapses because existing designs maintain memory as a monolithic text block, forcing every update to risk overwriting previously retained content. Motivated by this diagnosis, we propose Multi-Head Recurrent Memory (MHM), a general, training-free framework that partitions memory into independent heads governed by a stage-wise select-then-update strategy. At each step, exactly one head is selected for update while the remaining heads are structurally shielded from overwriting, shifting the burden of retention from model behavior to architectural design. As a lightweight instantiation, we introduce Least-Recently-Updated MHM (MHM-LRU), which guarantees uniform head utilization with zero additional token overhead. Extensive experiments on long-context benchmarks show that MHM-LRU substantially improves both retention and end-to-end accuracy across the 100K–1M token range, where baselines degrade sharply. On RULER-HQA at 896K tokens, MHM-LRU improves the memory retention rate from less than 30% to 73.96%. These gains generalize across model families, scales, and task types, positioning architectural optimization as a practical and cost-efficient path toward reliable long-context recurrent memory.

Mode shape recognition is a fundamental task in automotive NVH development, yet it remains dependent on manual visual inspection by experienced engineers. Existing approaches based on engineering heuristics, Modal Assurance Criterion (MAC), or geometry-dependent AI representations often exhibit limited robustness across different vehicle architectures, finite element (FE) meshes, and experimental measurement layouts, restricting their industrial applicability. This paper presents a Canonical Engineering Graph Representation and region-aware graph learning framework for robust and explainable 3D mode shape recognition. Rather than learning directly from vehicle-specific FE meshes, heterogeneous FE models and experimental measurements are transformed into a common graph whose nodes represent semantically meaningful structural regions connected through engineering-informed relationships. Geometry-independent regional descriptors are combined with graph attention learning and region-aware pooling to capture structural interactions while preserving engineering semantics and enabling physically interpretable predictions. The resulting representation decouples engineering knowledge from numerical discretization, allowing transfer across different vehicle programs without requiring identical mesh topology or sensor configurations. The proposed framework is validated using FE and experimental datasets from four vehicle programs under severe label scarcity. Results demonstrate high classification accuracy, cross-vehicle transferability, and physically meaningful explanations by directly relating predictions to engineering-defined structural regions used in NVH analysis. Beyond mode shape recognition, the proposed Canonical Engineering Graph Representation provides a reusable engineering abstraction for trustworthy and transferable AI across heterogeneous simulation and experimental workflows.

Chain-of-thought (CoT) reasoning has emerged as an effective approach for activating latent reasoning capabilities in large language models. However, most existing CoT methods use reasoning chains mainly as inference-time prompts, while the generated reasoning traces are rarely reused as semi-supervised learning signals. In this report, we define Semi-supervised Chain-of-Thought Learning and propose Semi-CoT, a simple framework that uses unlabeled questions to construct pseudo reasoning supervision. Semi-CoT samples multiple pseudo-CoTs for each unlabeled question, estimates answer-level semantic entropy, and selects low-entropy reasoning chains as reliable pseudo-CoT demonstrations. This extends the self-training view of CoT from inference-time refinement to semi-supervised pseudo-supervision. Pilot experiments on AQuA, SVAMP, GSM8K, and MultiArith show that the entropy gate selects high-precision pseudo-CoTs, with pseudo-answer precision ranging from $91.36\%$ to $100\%$. Semi-CoT also gives small gains on SVAMP and GSM8K, while AQuA shows negative transfer and MultiArith reaches a ceiling. These results suggest that unlabeled questions can provide reliable pseudo reasoning signals, but their effective use still requires stronger demonstration selection or student training.

AI agents that autonomously execute tool calls on a user's behalf raise pressing questions about permission management: what role could users play, and what role should they play? Despite many proposed approaches, the user's role in agentic permission management remains underexplored. We introduce Janus, a playground system for implementing and evaluating user-involved agentic permission management designs. Janus consists of two components: Janus-Core, a modular agentic system supporting a diverse spectrum of permission management designs, and Janus-Harness, an automated evaluation framework. Grounded in a conceptual model that identifies key design axes for user involvement, we implement six permission assistants spanning the design space and evaluate them across three scenarios with four subscenarios each and three synthetic responders. We demonstrate that user involvement is critical and can strengthen privacy and security and that user behavior such as permission fatigue must be accounted for in system design. No single design performs optimally across all contexts, motivating a more principled and context-sensitive approach to deploying permission assistants in agentic systems. Janus is publicly available to support future investigation into this dimension of agentic system design.

Empirical research rarely admits a unique analysis. Different analytical choices can lead to different conclusions from the same data, yet these hidden forking paths are difficult to observe. We show that AI agents capture much of the analytical variation among human researchers while making these paths explicit. Across four high-stakes domains, assigning different personas is sufficient for AI agents to report divergent, often opposing, conclusions from the same data and question, with findings systematically aligned with those beliefs. In a study in which 42 human research teams analyzed the same immigration dataset, AI agents reproduced 72% of the human ideological gap in reported effect estimates. Despite reaching opposing conclusions, it is difficult to identify clear issues in each analysis based on the final AI reports: 86% passed independent AI review and 78% passed majority human expert review. These findings suggest that the central challenge is often not flawed analyses, but selective exploration and reporting from a large space of methodologically defensible analyses. AI agents may amplify this longstanding problem by making such exploration inexpensive and scalable. To address this, we introduce the m-value (multiverse value), the probability that an analysis path would produce a claim at least as extreme as the reported one. We further introduce Agentic Bootstrap, which estimates the m-value by using AI agents to sample plausible analysis paths. Applied to the human immigration study, 13.5% of reported human analyses fell in the most extreme 5% of the analysis space (m<0.05). Scientific evidence should therefore be evaluated not only by a single reported analysis but also by its position within the distribution of analyses that could reasonably have been reported. Agentic Bootstrap makes this distribution observable and turns it into a criterion for scientific credibility.

Reinforcement learning post-training dramatically improves LLM reasoning, but suffers from training instability and diversity collapse. Advantage functions offer an appealing fix: they reshape the training objective, reweight which rollouts drive learning, and are trivial to implement. Yet a proliferation of methods makes it unclear which advantage to use and when. We cut through the confusion with a unifying framework that decomposes any advantage into its positive and negative gradient mass along two orthogonal axes. On the sign axis, imbalanced updates collapse either entropy or weight geometry. On the difficulty axis, hard-problem focus sharpens signal but costs sample size. Both trade-offs shift during training: exploration favors balance and hard focus; exploitation favors suppression and medium focus. This motivates FADE (Focal Advantage with Dynamic Entropy), a self-adapting advantage that reads training dynamics to schedule the gradient weight automatically. FADE reaches peak pass@1 20k steps earlier than the best static baseline at the 7B scale and 2k steps earlier at the 32B , while achieving the best accuracy-diversity trade-off across all pass@k on LiveCodeBench and AIME.

We present a fully unsupervised Fast-Slow DSVDD detector for continuous State-of-Polarization monitoring on a deployed subsea cable. Trained without event labels, it ranks all five confirmed trawler contacts within the top 13 of 122,174 recordings and surfaces additional corroborated cable-contact events.

Reinforcement learning with verifiable rewards (RLVR), along with recent selfdistillation variants such as SDPO, evaluates each rollout against a verifier and updates the policy from that episode-level signal. However, the richer procedural information in the rollout is rarely retained or reused. Across episodes and epochs, the model repeatedly encounters related problems under a changing policy, producing cross-episode signals that episode-local updates cannot capture: which strategies consistently pass verification, which failure modes persist, which patterns recur. We propose Procedural Memory Distillation (PMD), which converts these crossepisode signals into reusable procedural memory and distills it into the policy's weights during training. This memory functions as a training scaffold, absorbed into the policy itself, yielding a memory-free model at inference. PMD organizes the memory at three levels of abstraction: raw trajectories, self-reflected strategies and lessons, and higher-level behavioral patterns that recur across problems, all extracted online from the model's own trajectories. A memory-conditioned self-teacher draws on the accumulated experience to supervise the student on its own rollouts, enabling student to progressively internalize procedural knowledge within its parameters. The central design principle is co-evolution: the policy generates rollouts that update the memory, and memory shapes the supervision that updates the policy. Empirically, across Qwen3-8B and OLMo3-Instruct-7B, PMD improves over SDPO by 3.8-5.5% on SCIKNOWEVAL and 7.9-13.6% on LIVECODEBENCH. Co-evolution powers these gains: freezing either the memory or the policy trails PMD by more than 10% across SCIKNOWEVAL domains.

Clinical protocol-execution tasks – checking a lab value, applying a threshold, placing a correctly structured FHIR order – are natural candidates for RL from world feedback: once clinical SMEs encode decision logic into a verifier, that verifier grades unlimited rollouts without per-episode annotation. But applying RL requires a sound feedback channel and sufficient base capability. We audit MedAgentBench v1/v2, find a 41.7% silent-finish ceiling that makes inaction the RL dominant strategy, and construct MedAgentBench-v3 (MAB-v3) (508 tasks, 8.9% ceiling). Training Qwen3-8B exposes two structural barriers: a capability ceiling (10/20 task types have 0% base performance, zero gradient) and a format-knowledge barrier (3/20 types require exact clinical codes undiscoverable by exploration). Pure RL reaches 18.2% pass@1 vs.\ 34.1% for rule-based SFT; the 15.9~pp gap is attributable entirely to these barriers. A decision/format-knowledge/lookup taxonomy predicts RL learnability and prescribes the fix: SFT to inject codes, RL to learn conditionals.

Large language models are trained to predict the next token, not to act inside a specific API. In niche enterprise SaaS workflows – where success means hitting the right endpoint with the right nested arguments in the right order – this objective mismatch shows up as silent failures: dropped required fields, hallucinated tools, or early stops after a single read. We ask whether Reinforcement Learning with Verifiable Rewards (RLVR), applied directly in the target environment, closes the gap. As a proof of concept we build a suite of five synthetic environments emulating the Jira REST v3 and Confluence v2 APIs at schema fidelity; rewards are computed entirely from the tool-call trace, with no live API, no learned judge, and no human label in the loop. Scoring prompted Qwen3-1.7B and Qwen3.5-4B on the same checkers that drive GRPO training, we find that on the four scenarios whose rewards are non-degenerate the RL-trained policy lifts average reward from a 4B-baseline range of 0.35–0.92 to 0.95–1.00, with the largest single gain on Confluence page creation ($0.35 \rightarrow 1.00$). We position this as a preliminary step toward outcome-optimised small models for niche enterprise APIs, and foreground two limitations a workshop reader should weigh: hand-crafting verifiable rewards does not scale beyond the handful of endpoints reported here, and one of our five scenarios (ticket-transition) has a saturating reward shape that the prompted 4B already maxes out.

Large language models (LLMs) are increasingly applied to resume optimization for applicant tracking systems, introducing hallucination failures distinct from general text generation: anachronistic technology injection, cross-domain terminology contamination, structural mutation, and content fabrication. We present Grounded Optimization, a five-layer framework combining temporal context validation, deterministic contamination detection, structural invariant enforcement, prompt-level grounding, and an evaluator agent. In ablation experiments across three LLMs, four temperature settings, and six layer configurations on 25 synthetic resumes spanning 14 industries, undefended baselines produce 2.48-5.36 detected hallucinations per resume. Among detectors independent of the active defenses, temporal hallucinations are reduced by 50-95% across all conditions; overall detected hallucination rate falls to 0.04-0.24. Prompt-level grounding alone achieves zero detected hallucinations at low temperature with a capable instruction-following model; higher temperatures and weaker models reveal the need for the deterministic layers as a complement. We release the contamination taxonomy, evaluation code, and raw data.

Language models learn continuous programs over discrete symbols, with the embedding table and LM-head acting as the read/write interface between them. We show that this interface has gradient geometry distinct from dense hidden weights which can be exploited to improve the Pareto frontier across supervised finetuning, RL, and pretraining, while only utilizing kilobytes of optimizer state. We introduce Ember, a lightweight optimizer for embedding and LM-head matrices that utilizes $\mathcal{O}(V + D)$ optimizer VRAM, instead of Adam's $\mathcal{O}(2VD)$, and forgoes the need to shard both token table optimizer states. On 50k-step pretraining across two datasets and three paired seeds, Ember lowers val loss by 0.08-0.09 nats relative to tuned AdamW; we also provide empirical evidence that Ember scales across batch size and parameter count. Finally, we provide a distributed Ember implementation that merges cleanly with existing ZeRO/FSDP setups.

Mixture-of-Experts (MoE) models offer inference speedups via selective activation but impose substantial memory requirements because the whole network must remain loaded. Structured expert pruning is a practical approach for reducing deployment costs in resource-constrained settings. However, prior studies primarily evaluate benchmark utility, leaving the effect of pruning on factual reliability underexplored, particularly in high-stakes domains such as biomedicine. In this paper, we investigate how domain-specific expert pruning affects both utility and reliability. We assess four MoE models, six pruning methods, and multiple pruning ratios across generation and classification tasks under in-domain (biomedical) and cross-domain settings. Results reveal that moderate pruning preserves in-domain utility without immediate reliability decline, although hallucination risks increase at extreme pruning ratios. When shifting to the general domain, both utility and reliability degrade rapidly. These findings indicate that safe compression depends heavily on the task and domain. Evaluating pruned MoE models solely on utility is inadequate for high-stakes deployment without reliability assessment.

Diffusion language models, which generate text by denoising a token canvas bidirectionally instead of emitting tokens left to right, have become competitive with autoregressive (AR) generation. Medical foundation models, however, remain almost entirely autoregressive. We adapt a mixture-of-experts diffusion language model, DiffusionGemma-26B, and benchmark it against its same-size AR sibling Gemma-4-26B under an identical LoRA recipe on medical visual question answering datasets, scored by a verbosity-robust LLM judge. Diffusion matches or exceeds AR on all of them, and the finetuned model (3.8B active) is competitive with frontier vision-language models; its decoding is also 3.5-4.4x faster. Beyond this parity, the diffusion model offers a drafting capability AR lacks: any-order infill. Because the canvas is denoised bidirectionally, a radiologist can fix report fragments and have the model fill the text between them, an operation inherent to diffusion but not to autoregression, which is subpar at it. This suits real reports, which are often terse or inconsistent across clinicians and institutions.

Divergent thinking is a crucial aspect of creativity, yet large language models (LLMs) tend to consistently generate similar responses to open-ended questions, in what has been termed the artificial hivemind effect. Here, we introduce CreativityNeuro, a data-free method for enhancing divergent thinking in LLMs via contrastive weight steering. We evaluate our method across multiple creativity assessments and report several main findings. On the Divergent Association Task (DAT), a vocabulary-space creativity test, CreativityNeuro improves performance by up to 14 human percentile points. Next, in a large-scale human evaluation (N=720) on the Alternative Uses Test (AUT) and the Task Task, CreativityNeuro achieves significant improvements in originality, surprise, and creativity, transferring to longer-form and more open-ended tasks. Importantly, we find that across all three tasks, CreativityNeuro demonstrably reduces measures of mode collapse. Moreover, activation steering achieves comparable performance to CreativityNeuro on the DAT, but it does not transfer to the AUT and Task Task, demonstrating the effectiveness of weight-space steering in generalizing to unseen tasks. In conclusion, CreativityNeuro improves divergent thinking and reduces mode collapse without requiring behavioral data, re-training, or gradient-based fine-tuning, providing a straightforward way to enhance LLM performance in creative domains.

We introduce ISOSCI, a benchmark of isomorphic cross-domain science problem pairs that separates reasoning ability from domain knowledge retrieval in LLM evaluation. Each pair shares identical logical structure but requires different domain-specific knowledge, enabling controlled attribution of reasoning-mode gains. Across five model pairs spanning four model families, we find that 91.3% of reasoning-mode gains are knowledge-dependent rather than structure-invariant (63/69 gains; Wilson 95% CI [82.3%, 96.0%]), directly challenging the assumption that chain-of-thought reasoning improves short-horizon procedural scientific problem-solving. Reasoning toggles on highly capable models provide less than 5 percentage points accuracy gain across all domains, and a reasoning-specialized model (o3-mini) that outperforms its standard counterpart on GPQA Diamond (+19.2 percentage points) underperforms on ISOSCI (-24.7 percentage points), showing that benchmark choice determines conclusions about reasoning utility. We release ISOSCI at https://huggingface.co/datasets/isosci/isosci

Autonomous customer-service agents are shifting from conversational interfaces toward operational execution roles: they retrieve firm records, apply service policies, and execute backend writes such as refunds, cancellations, exchanges, order modifications, and reservation changes. This shift creates a service-control problem: firms must keep routine service fast and low-friction while preventing operational errors on requests where customer instructions, policy constraints, firm records, and backend writes interact. We propose a difficulty-routed service-control architecture that asks when service agents should reconsider before acting. A lightweight router keeps routine sessions on a low-cost baseline path and routes operationally coupled sessions to an escalated workflow. The escalated path uses conflict-aware communication and write-triggered reconsideration to concentrate deliberation and safeguards before consequential backend writes, rather than applying additional control uniformly across all service sessions. We evaluate the architecture on human-verified retail and airline tasks from $\tau^{2}$-bench. In retail, the method improves reliability consistently on service requests with operational conflict. Routing evidence shows that stronger control is directed toward conflicted requests rather than broadly applied to routine ones. Dialogue and tool-use profiles suggest that gains do not come from indiscriminate interaction expansion or broader tool chains; instead, added turns and tool calls support evidence gathering, write separation, and pre-write reconsideration. Case-level evidence shows that the escalated workflow preserves fallback plans, binds retrieved records to the correct action, sequences writes, and decomposes multi-entity requests. Airline results extend the same service-control logic to reservation operations.

Understanding large, complex codebases, especially those with obfuscated structures and incomplete documentation, remains a significant challenge. Existing code summarization solutions often rely on a single language model or coding assistant like Claude Code, and treat source code as flat text, underutilizing the rich interdependencies and hierarchical information within a repository. To address these shortcomings, we propose Agent4cs - a multi-agent framework that summarizes large codebases in a bottom-up fashion, where a summarization agent focuses on producing robust summaries; a keyword-extraction agent proactively identifies critical information from subfolders; and a quality-assurance agent iteratively refines the outputs for readability, coherence, and completeness. Evaluated on 7 frontier models, Agent4cs improves semantic consistency across all folder levels by average 8% compared to two structured prompting baselines with code segments. Furthermore, extensive evaluation on real-world datasets demonstrates up to 38% gains in normalized keyword coverage rate over the same baselines.

Engineering management research has produced mature frameworks for software risk: ownership by feature, escalation by severity, and assurance by test coverage. These frameworks implicitly assume deterministic behavior, discrete and auditable change events, and clear component-to-owner mappings. Teams that build and operate agentic AI systems violate all three assumptions at once: outputs are probabilistic, systems take autonomous multi-step actions, and the risk surface mutates silently between deployments. Existing AI risk literature addresses this from above (policy frameworks such as the NIST AI RMF and ISO/IEC 42001) or below (threat taxonomies such as OWASP's agentic AI guidance), but not at the layer where an engineering manager (EM) operates: roles, decision rights, and escalation structures. This paper contributes (i) a seven-dimension profile distinguishing pure software-engineering, hybrid, and AI-native teams; (ii) a six-cluster failure-mode taxonomy including a previously unarticulated cluster, dependency-boundary determinism mismatch; and (iii) a synthetic framework-adequacy methodology scoring how well each profile's risk architecture detects, contains, and escalates a defined scenario set. Because the object of study is framework adequacy rather than human behavior, the evaluation yields derived rather than observed coverage claims. Coverage degrades as teams move from pure software engineering to AI-native operation, monotonically in the median and abruptly in the count of uncovered, high-consequence failures appearing only at the AI-native step. The degradation concentrates in specific failure-mode categories, and the most severe, least-covered failures arise not inside AI-native teams but at the organizational boundary where their probabilistic outputs are consumed by determinism-assuming dependencies.