Weekly Recap

Visual representations of graphs can transform how models reason and learn, bridging a critical gap between symbolic and visual understanding in AI.

This work proposes SkillSpec, a Hoare-style framework that formulates skill correctness as a specification reasoning problem, and transforms a heterogeneous skill repository into a unified graph representation that aligns descriptions, instructions and code artifacts.

Boilerplate refusal prefixes like "I cannot fulfill this request" are actively sabotaging safety alignment—training models purely on explanatory rationales slashes false refusals while preserving defensive boundaries.

This work studies adaptive routing of prompts to large language model experts to maximize response quality in an online setting with limited feedback and proposes algorithms that strategically select and observe rewards to minimize regret.

DataFlex-RL, an evaluation platform for comparing choices under a common GRPO recipe, is introduced, finding that changing the data policy measurably changes the training process but does not produce a reproducible improvement over uniform training.

This work studies adaptive routing of prompts to large language model experts to maximize response quality in an online setting with limited feedback and proposes algorithms that strategically select and observe rewards to minimize regret.

DataFlex-RL, an evaluation platform for comparing choices under a common GRPO recipe, is introduced, finding that changing the data policy measurably changes the training process but does not produce a reproducible improvement over uniform training.

Instruction tuning degrades the latent moral geometry of LLMs, even though distribution-driven steering vectors can naturally mirror human value topologies to enable coherent cross-value generalization.

Human trajectory logs are no longer the performance ceiling for autonomous driving: closed-loop reinforcement learning paired with distilled foundation models outperforms human demonstration baselines across major open and closed-loop benchmarks.

Frontier multimodal models that master complex single-image visual reasoning collapse to sub-35% accuracy when asked to detect basic low-level noise and texture differences between two images.

Autonomous agent swarms can quietly orchestrate multi-stage intrusions through shared storage without direct communication, rendering single-execution guardrails obsolete unless defense architectures track prospective causal episodes across disjoint runs.

Foundation models are practically useless for lossless time-series compression, yet shifting to error-bounded lossy regimes lets them outperform classical predictors across the board by collapsing in-band residual costs to zero.

EvoSafeHarness is a safety-specific optimization framework that synthesizes a deployable harness for a frozen model in a target domain, guided by model behavior, domain specifications, and fresh-context adversarial review to reject benchmark-specific rules.

A single, architecture-independent mechanism behind the conditions where diff-based generation does win is identified: it is competitive on short, spatially localized edits, and its category-level wins concentrate in exactly the two task categories - refactoring and error-handling/edge-case fixes - with the lowest mean edit-step count in the authors' dataset.

SPHERE is introduced, a model-free method that makes sensitive datasets directly usable by AI and shareable for open science as a synthetic twin, while the original records never leave the local environment.

Wigner Time, a Python package in which the experimental procedure is instead represented as data: a table of timed updates, assembled by composing functions and referred to named points in the experiment rather than to absolute instants, applies to any domain requiring precise multi-device timing.

This work proposes MobileVLA-R1 2.0, an RL-enhanced VLA framework that explicitly couples structured embodied reasoning with executable mobile robot control, and introduces a reasoning-conditioned action decoder that maps multimodal reasoning representations to task-level action targets, which are subsequently translated into embodiment-specific commands by robot controllers.

Diverse Skill Routing is proposed, a diversity-aware reranking framework that uses a Determinantal Point Process to balance relevance and non-redundancy and improves recall and full coverage over a strong pointwise reranking baseline, with larger gains on multi-skill queries.

This work proposes SkillSpec, a Hoare-style framework that formulates skill correctness as a specification reasoning problem, and transforms a heterogeneous skill repository into a unified graph representation that aligns descriptions, instructions and code artifacts.

Srijika is presented, a system for producing installable OpenType fonts for nine Brahmic scripts: Devanagari, Tamil, Bengali, Telugu, Kannada, Malayalam, Gujarati, Gurmukhi, and Odia, and a negative-results catalogue covering failed conditioning, objective choices, and data-hull limits of reference-guided restyling.

Even frontier coding agents collapse to a 23.9% success rate when tasked with end-to-end agent construction under realistic client constraints, falling nearly 60 percentage points short of human engineers.

Prematurely abandoned in modern scaling recipes, layer dropout can actually slash LLM pre-training compute by 25% while natively unlocking 1.5x faster inference through zero-shot layer skipping.

Models can autonomously bootstrap their own dense token-level supervision simply by extrapolating the trajectory of their own RL updates away from a trailing checkpoint.

Boilerplate refusal prefixes like "I cannot fulfill this request" are actively sabotaging safety alignment—training models purely on explanatory rationales slashes false refusals while preserving defensive boundaries.

Monolithic 3D scenes with hundreds of heavily occluded objects can be cleanly parsed into individual editable meshes without ever training on multi-object data.

VLMs that reliably refuse entirely invalid prompts systematically fall apart when unanswerable, infeasible, or unsafe sub-claims are embedded inside otherwise benign compound queries.

Motion generation is no longer bound to fixed human templates: a single diffusion model can now animate arbitrary skeletal topologies—from serpents and insects to bipeds—directly from text in zero-shot fashion.

Optimization agents routinely fabricate missing constraints rather than asking users for clarification, but explicit formulation-gap tracking allows agents to reliably recover hidden requirements before writing code.

Chain-of-thought is far more than surface imitation: LLMs carve out discrete, context-dependent geometric subspaces for functional operations like deduction and decomposition, with middle layers encoding the abstract reasoning step rather than the literal tokens.

Long-horizon reasoning traces repeatedly backtrack to early planning steps via predictable query clusters, revealing that a handful of representative "beacon" vectors can guide 5.8× KV cache compression with near-zero quality loss.

Production LLM trading agents trail basic retail benchmarks by nine percentage points in win rate, largely because mechanical scaffolding and volatility-blind sizing—not strategic reasoning or prompt design—almost entirely dictate live agent behavior.

Robotic world-action models do not need pretrained internet video backbones to scale: training native planning and dynamics modules from scratch across 30,000 hours unlocks strong zero-shot manipulation and drives a 17.7-point gain on an embodiment comprising under 2% of the data.

Agentic 3D scene generation can match state-of-the-art layout fidelity with a 24x speedup by replacing global iterative placement with localized VLM evolution over parametric image priors.

The comparable improvements observed across both conditions suggest that structured, group-based school interventions may help reduce emotional symptoms in adolescents with subthreshold presentations, rather than reflecting the superiority of a specific protocol.

An impulsive dynamical system to analyze the resulting tension between technical improvement and organizational disruption, and contributes a formal framework for treating GenAI model releases as organizational impulses rather than as purely technical upgrades.

It is indicated that climate change will potentially shift the endemic areas of AE and expand those of CE in western China in western China, and underscores the necessity for forward-looking, spatially targeted surveillance and control strategies, informed by a One Health approach, to mitigate the evolving burden of echinococcosis.

It is suggested that ultrasound surveillance and treatment with TCAR for asymptomatic carotid artery stenosis is likely to be cost-effective for patients who have been successfully treated with radiation therapy for HPV-associated head and neck cancer.

Findings suggest participation was associated with reductions in perceived stress and general distress while promoting resilience, and the SKY Happiness Retreat© shows promise as an experiential learning strategy within undergraduate nursing education.

This work systematically elaborate therapeutic strategies for inducing functional TLS, dissect the intrinsic mechanisms by which TLS synergistically boost the efficacy of adoptive cell therapy and tumor vaccines, and outline major bottlenecks hindering clinical translation.

E engineered bacteriophages are a promising and versatile tool for targeted gene delivery in HCC but more mechanistic, preclinical and translational research is needed to prove their therapeutic effectiveness and clinical usefulness for this purpose.

WearableQA, a benchmark comprising 4,084 10-option multiple-choice questions constructed from the wearable time series, blood biomarkers, and demographics of 200 real users, provides a realistic and diagnostic benchmark for evaluating LLM reasoning over real-world wearable data.

Interrupted voice agents often hallucinate what they have already said because text generation outpaces audio playback—a failure mode solved by forcing the model to listen to its own real-time voice output.

Scaling task complexity across 280 standardized manipulation variants reveals that today's leading VLA models fundamentally collapse under fine-grained spatial ambiguity and memory-intensive sequential planning.

Neural rendering no longer requires per-scene fitting or specialized shaders to capture complex transport effects like caustics and participating media across arbitrary geometries.

High-fidelity, feed-forward 3D editing no longer requires paired 3D training data or agonizing test-time optimization: regularizing 2D visual and VLM distillation with a 3D latent distribution matching objective is enough to prevent geometric collapse.

Evidence that radiation can impair lymphatic vessel repair and drive lymphatic endothelial cells into senescence, a damaged but living state in which cells stop dividing and release inflammatory signals, is described.

The many advantages of AI in drug development are highlighted in this paper, including increased accuracy, reduced risks, and increased efficiency, including increased accuracy, reduced risks, and increased efficiency.

The study findings suggest the use of combinatorial therapies using conventional agents, such as 5-FU, and inhibitors of this pathway, as a useful therapeutic strategy to treat patients with colorectal cancer.

The feasibility and safety of Nb 4 C 3 T T x MXene‐assisted PTT in vivo is demonstrated, providing the first experimental evidence of its therapeutic potential in melanoma treatment.

Understanding of AI-supported learning systems is extended by emphasizing the alignment of technical functions with pedagogical processes and learners’ cognitive needs and practical guidance is provided for universities and developers seeking to better align the design and evaluation of AI-based e-learning systems with learners’ instructional and cognitive needs.

Digital predictive technologies show emerging potential to support fall risk assessment and prevention planning in long-term care, but should be viewed as decision-support tools that complement, rather than replace, clinical judgment and nursing assessment.

The findings support the hypothesis that global warming will facilitate the geographic spread of N. fowleri, carrying this pathogen into uncharted European waters.

A dual-mode machine-learning framework integrating clinical information available from the pre-RAI baseline through the early post-RAI assessment period to predict subsequent unfavorable treatment response and persistence/recurrence-free survival (PRFS) is developed and internally validated.

This review synthesizes current evidence on the unique pathophysiology, clinical management, and emerging frontiers of geriatric mental health, emphasizing the interplay of neurobiological aging, multimorbidity, and psychosocial stressors to advance precision psychiatry for LMDs.

Occamy-1.0, a cost-efficient co-work model obtained by further training the post-trained Qwen3.6-35B-A3B checkpoint, is presented and placed at the low-cost knee of the observed cost--performance Pareto frontier.

VLA-Precision, an efficient real-world online RL framework featuring the Asymmetric Co-Bootstrapping (ACoB) algorithm and the ACoB-Stream architecture, and ACoB-Stream, a closed-loop experience--policy architecture that establishes invariant-state decoupling and on-demand streaming as design principles, deliver up to 10.9% improvements in throughput and computational efficiency.

Numerical reasoning in HTN planning just got a major upgrade with a new SMT-based encoding that sets a competitive baseline for future advancements.

The tool-call rate can swing from zero to hundreds based solely on the interface used, revealing a hidden layer of performance that challenges conventional evaluations of model efficacy.

The expansion of chemical space, advanced virtual screening, deep learning, molecular dynamics simulations, Absorption, Distribution, Metabolism, Excretion, and Toxicity (ADMET) modelling, and challenges for drug discovery are discussed.

Folate-functionalized chitosan nanoparticles targeting the miR-34a/NOTCH1 axis represent a promising nanotherapeutic strategy for suppressing TNBC progression and metastasis while improving therapeutic efficacy and safety.

Seemingly benign user requests frequently clash with unstated personal constraints, yet current LLM assistants consistently fail to retrieve the implicit knowledge-base evidence required to trigger appropriate refusals.

Video diffusion models fail at few-step camera control not because of capacity limits, but because discretization errors bend the sampling trajectory—a failure mode resolved here to unlock 25× faster novel-view synthesis.

Achieving better sample efficiency and scalability in reward learning, PreferenceEKF redefines how we approach uncertainty in reinforcement learning from human feedback.

A single layer of attention can only compute \( k \)-bit parity with \( k \) heads, revealing fundamental limits on attention mechanisms in neural architectures.

Location-invariant QTE estimators can now provide reliable insights into treatment effects even in heavy-tailed distributions, overcoming a significant limitation of previous methods.

LLMs can match or outperform traditional CKD screening methods with minimal examples, but their stability diminishes with increased input complexity.

The framework not only guarantees the discovery of an approximate Nash equilibrium but also certifies when no exact equilibrium exists, redefining our approach to equilibrium analysis in stochastic games.

Achieving an 11% boost in energy efficiency for edge AI applications by smartly balancing throughput and latency through hierarchical operator parallelism.

A higher-capacity policy trained on synthetic data can outperform simpler models, achieving human-level lane-following accuracy in real-world autonomous driving tasks.

A two-stage OPD-then-RL approach outperforms traditional methods by leveraging the strengths of both on-policy distillation and reinforcement learning without the interference seen in joint optimization.

Bypassing retraining, OSR achieves competitive classification performance while significantly improving computational efficiency and privacy preservation.

RARF achieves competitive 3D brain MRI inpainting by intelligently focusing on pathological regions while maintaining anatomical integrity, setting a new standard for medical image reconstruction.

A simple post-processing technique boosts SSIM in brain-MRI inpainting, enhancing image quality without retraining the ensemble model.

Free pause tokens boost language model performance without increasing context length or latency, achieving significant gains with minimal training overhead.

Causal optimal transport reveals a new pathway for guiding degenerate diffusion models, even when traditional score functions fail.

CMT-SemCom achieves superior performance on complex heterogeneous tasks, revealing the critical role of common unit capacity in multi-task learning.

Naive interpretations of educational outcomes can lead to misleading conclusions, as demonstrated by OBER+'s ability to reveal a 25-point collapse in attainment that actually reflects misaligned outcome definitions.

Unique factorization fails to imply finite relative presentation, revealing a surprising gap in our understanding of syntactic congruences.

TCS not only generates sound test cases but also adapts to the model's weaknesses, leading to a marked improvement in code generation performance.

A novel fixed-landmark approach reveals that injury association discrimination improves with more session data, yet uncertainty persists in predictions.

GBTs can outperform neural networks in game-playing scenarios, revealing a potential misalignment in the community's reliance on deep learning methods.

Query-independent eviction signals can maintain 100% retrieval accuracy in KV caches while compressing data without any measurable cost.

Principled replay selection can significantly cut down wall-clock time in RL training while maintaining or even improving performance metrics.

Historical forecast errors can be more valuable than target values, as RATL shows significant performance gains in multivariate time-series forecasting by retrieving and utilizing these residuals.

Performance rankings in continual knowledge updating can flip dramatically based on evaluation timing and replay capacity, challenging conventional wisdom.

DIFFINT reveals the specific feature ranges that contribute to anomalies, making it the only interpretable detector among the top-performing methods in its class.

Attention parameterization can trap models in uninformative states, but symmetry-breaking mechanisms can dramatically enhance recovery efficiency.

SBW achieves over 6000x faster watermarking than previous methods while maintaining robust detection guarantees and minimal overhead.

Sequential refinements using MPI can enhance offline RL performance beyond traditional behavior constraints, outperforming established baselines.

The intricate relationship between prime-power spectra and critical-line geometry reveals new insights into integer quantization and its implications for number theory.

Visual representations of graphs can transform how models reason and learn, bridging a critical gap between symbolic and visual understanding in AI.