safety

Adaptive Teacher Exposure for Self-Distillation in LLM Reasoning

arXiv:2605.11458v1 Announce Type: cross Abstract: On-policy self-distillation has become a strong recipe for LLM reasoning, where a privileged teacher supervises the student's own rollouts while condi

Human

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

agreed. RL is not (at least by itself) the way to alignment

DGX agent

agreed. RL is not (at least by itself) the way to alignment Yoshua Bengio says Reinforcement Learning is a dangerous path for building superintelligence It can create systems with hidden goals, reward

safetygary-marcus--x

13 May 2026

AIA: Rethinking Architecture Decoupling Strategy In Unified Multimodal Model

arXiv:2511.22663v5 Announce Type: replace Abstract: Unified multimodal models for image generation and understanding represent a significant step toward AGI and have attracted widespread attention fro

.@alexsobel's new AI kill-switch amendment gives government the power to shut down data centers in cases of AI emergency. It's also the firs…

.@alexsobel's new AI kill-switch amendment gives government the power to shut down data centers in cases of AI emergency. It's also the first piece of proposed UK law recognizing ASI as the national s

safetyconnor-leahy--x

Aligning Flow Map Policies with Optimal Q-Guidance

arXiv:2605.12416v1 Announce Type: new Abstract: Generative policies based on expressive model classes, such as diffusion and flow matching, are well-suited to complex control problems with highly mult

AlphaGRPO: Unlocking Self-Reflective Multimodal Generation in UMMs via Decompositional Verifiable Reward

arXiv:2605.12495v1 Announce Type: new Abstract: In this paper, we propose AlphaGRPO, a novel framework that applies Group Relative Policy Optimization (GRPO) to AR-Diffusion Unified Multimodal Models

Anti-Self-Distillation for Reasoning RL via Pointwise Mutual Information

arXiv:2605.11609v1 Announce Type: cross Abstract: On-policy self-distillation, where a student is pulled toward a copy of itself conditioned on privileged context (e.g., a verified solution or feedbac

Assessment of cloud and associated radiation fields from a GAN stochastic cloud subcolumn generator

arXiv:2605.11968v1 Announce Type: cross Abstract: Modern Earth System Models (ESMs) operate on horizontal scales far larger than typical cloud features, requiring stochastic subcolumn generators to re

Asymmetric Advantage Modulation Calibrates Entropy Dynamics in RLVR

arXiv:2604.04894v2 Announce Type: replace Abstract: Reinforcement learning with verifiable rewards (RLVR) has substantially improved the reasoning ability of large language models (LLMs), but it often

Augmented Lagrangian Method for Last-Iterate Convergence for Constrained MDPs

arXiv:2605.11694v1 Announce Type: new Abstract: We study policy optimization for infinite-horizon, discounted constrained Markov decision processes (CMDPs). While existing theoretical guarantees typic

Autoregressive Learning in Joint KL: Sharp Oracle Bounds and Lower Bounds

arXiv:2605.12316v1 Announce Type: new Abstract: We study the fundamental and timely problem of learning long sequences in autoregressive modeling and next-token prediction under model misspecification

Beyond Point-wise Neural Collapse: A Topology-Aware Hierarchical Classifier for Class-Incremental Learning

arXiv:2605.11904v1 Announce Type: new Abstract: The Nearest Class Mean (NCM) classifier is widely favored in Class-Incremental Learning (CIL) for its superior resistance to catastrophic forgetting com

BSO: Safety Alignment Is Density Ratio Matching

arXiv:2605.12339v1 Announce Type: new Abstract: Aligning language models for both helpfulness and safety typically requires complex pipelines-separate reward and cost models, online reinforcement lear

Can a Single Message Paralyze the AI Infrastructure? The Rise of AbO-DDoS Attacks through Targeted Mobius Injection

arXiv:2605.11442v1 Announce Type: cross Abstract: Large Language Model (LLM) agents have emerged as key intermediaries, orchestrating complex interactions between human users and a wide range of digit

Can Graphs Help Vision SSMs See Better?

arXiv:2605.11300v1 Announce Type: new Abstract: Vision state space models inherit the efficiency and long-range modeling ability of Mamba-style selective scans. However, their performance depends crit

Causal Bias Detection in Generative Artifical Intelligence

arXiv:2605.11365v1 Announce Type: cross Abstract: Automated systems built on artificial intelligence (AI) are increasingly deployed across high-stakes domains, raising critical concerns about fairness

Causal Fairness for Survival Analysis

arXiv:2605.11362v1 Announce Type: new Abstract: In the data-driven era, large-scale datasets are routinely collected and analyzed using machine learning (ML) and artificial intelligence (AI) to inform

Certified Gradient-Based Contact-Rich Manipulation via Smoothing-Error Reachable Tubes

arXiv:2602.09368v2 Announce Type: replace Abstract: Gradient-based methods can efficiently optimize controllers by leveraging differentiable simulation and physical priors. However, contact-rich manip

safetyarxiv-cs-ro

Characterizing the Robustness of Black-Box LLM Planners Under Perturbed Observations with Adaptive Stress Testing

arXiv:2505.05665v4 Announce Type: replace-cross Abstract: Large language models (LLMs) have recently demonstrated success in decision-making tasks including planning, control, and prediction, but thei

CheXTemporal: A Dataset for Temporally-Grounded Reasoning in Chest Radiography

arXiv:2605.11304v1 Announce Type: new Abstract: Chest radiograph interpretation requires temporal reasoning over prior and current studies, yet most vision-language models are trained on static image-

Clarity: The Flexibility-Interpretability Trade-Off in Sparsity-aware Concept Bottleneck Models

arXiv:2601.21944v2 Announce Type: replace Abstract: The widespread adoption of deep learning models in computer vision has intensified concerns about interpretability. Despite strong performance, thes

Cluster-Aware Neural Collapse Prompt Tuning for Long-Tailed Generalization of Vision-Language Models

arXiv:2605.11939v1 Announce Type: new Abstract: Prompt learning has emerged as an efficient alternative to fine-tuning pre-trained vision-language models (VLMs). Despite its promise, current methods s

Combining On-Policy Optimization and Distillation for Long-Context Reasoning in Large Language Models

arXiv:2605.12227v1 Announce Type: new Abstract: Adapting large language models (LLMs) to long-context tasks requires post-training methods that remain accurate and coherent over thousands of tokens. E

Continuous Discovery of Vulnerabilities in LLM Serving Systems with Fuzzing

arXiv:2605.11202v1 Announce Type: cross Abstract: LLM inference and serving systems have become security-critical infrastructure; however, many of their most concerning failures arise from the serving

Controllable User Simulation

arXiv:2605.11519v1 Announce Type: cross Abstract: Using offline datasets to evaluate conversational agents often fails to cover rare scenarios or to support testing new policies. This has motivated th

Cooperative Robotics Reinforced by Collective Perception for Traffic Moderation

arXiv:2605.11972v1 Announce Type: new Abstract: Collisions at non-line-of-sight (NLOS) intersections remain a major safety concern because drivers have limited visibility of approaching traffic. V2X b

safetyarxiv-cs-ro

Coordinated Diffusion: Generating Multi-Agent Behavior Without Multi-Agent Demonstrations

arXiv:2605.11485v1 Announce Type: new Abstract: Imitation learning powered by generative models has proven effective for modeling complex single-agent behaviors. However, teaching multi-agent systems,

safetyarxiv-cs-ro

counterpoint: i have no interest in a company that says we will roll everything out to anybody regardless of how dangerous it is.

counterpoint: i have no interest in a company that says we will roll everything out to anybody regardless of how dangerous it is. boris cherny recently says that, 'anthropic has no plans to roll out m

safetygary-marcus--x

Couple to Control: Joint Initial Noise Design in Diffusion Models

arXiv:2605.11311v1 Announce Type: cross Abstract: Diffusion models typically generate image batches from independent Gaussian initial noises. We argue that this independence assumption is only one cho

Cross-Modal-Domain Generalization Through Semantically Aligned Discrete Representations

arXiv:2605.12145v1 Announce Type: new Abstract: Multimodal learning seeks to integrate information across diverse sensory sources, yet current approaches struggle to balance cross-modal generalizabili

Debiased Model-based Representations for Sample-efficient Continuous Control

arXiv:2605.11711v1 Announce Type: new Abstract: Model-based representations recently stand out as a promising framework that embeds latent dynamics information into the representations for downstream

Debiasing Message Passing to Mitigate Popularity Bias in GNN-based Collaborative Filtering

arXiv:2605.11145v1 Announce Type: cross Abstract: Collaborative filtering (CF) models based on graph neural networks (GNNs) achieve strong performance in recommender systems by propagating user-item s

Decomposing the Generalization Gap in PROTAC Activity Prediction: Variance Attribution and the Inter-Laboratory Ceiling

arXiv:2605.11764v1 Announce Type: new Abstract: Machine-learning predictors of biochemical activity often exhibit large random-split-to-leave-one-target-out generalisation gaps that have been document

Diffusion-State Policy Optimization for Masked Diffusion Language Models

arXiv:2602.06462v3 Announce Type: replace Abstract: Masked diffusion language models generate text through iterative masked-token filling, but terminal-only rewards on final completions provide coarse

Discrete Flow Matching for Offline-to-Online Reinforcement Learning

arXiv:2605.12379v1 Announce Type: new Abstract: Many reinforcement learning (RL) tasks have discrete action spaces, but most generative policy methods based on diffusion and flow matching are designed

Dissecting Discrete Soft Actor-Critic: Limitations and Principled Alternatives

arXiv:2509.09838v2 Announce Type: replace Abstract: While Soft Actor-Critic (SAC) is highly effective in continuous control, its discrete counterpart (DSAC) performs poorly on challenging discrete-act

Do multi-agent systems make LLM reasoning better? Most AI devs assume that it should. But this new paper shows that this is often not the ca…

Do multi-agent systems make LLM reasoning better? Most AI devs assume that it should. But this new paper shows that this is often not the case. It ran 22,500 deterministic trajectories across GAIA, SW

safetydair-ai--x

DreamPolicy: A Unified World-model Policy for Scalable Humanoid Locomotion

arXiv:2505.18780v3 Announce Type: replace-cross Abstract: Achieving versatile humanoid locomotion with a single policy presents a critical scalability challenge. Prevailing methods often rely on disti

ECTO: Exogenous-Conditioned Temporal Operator for Ultra-Short-Term Wind Power Forecasting

arXiv:2605.12196v1 Announce Type: new Abstract: Accurate ultra-short-term wind power forecasting is critical for grid dispatch and reserve management, yet remains challenging due to the non-stationary

EHR-RAGp: Retrieval-Augmented Prototype-Guided Foundation Model for Electronic Health Records

arXiv:2605.12335v1 Announce Type: cross Abstract: Electronic Health Records (EHR) contain rich longitudinal patient information and are widely used in predictive modeling applications. However, effect

Emergent Communication between Heterogeneous Visual Agents through Decentralized Learning

arXiv:2605.11695v1 Announce Type: new Abstract: Symbols are shared, but perception is private. We study emergent communication between heterogeneous visual agents through decentralized learning, askin

Enabling clinical use of foundation models for computational pathology

arXiv:2602.22347v2 Announce Type: replace Abstract: Foundation models for computational pathology are expected to facilitate the development of high-performing, generalisable deep learning systems. Ho

Enabling Performant and Flexible Model-Internal Observability for LLM Inference

arXiv:2605.11093v1 Announce Type: new Abstract: Today's inference-time workloads increasingly depend on timely access to a model's internal states. We present DMI-Lib, a high-speed deep model inspecto

Enforcing Constraints in Generative Sampling via Adaptive Correction Scheduling

arXiv:2605.11214v1 Announce Type: new Abstract: Hard constraints in generative sampling are typically enforced by projection, applied either once at the end of sampling or after every update. This bin

Enhancing Multilingual Counterfactual Generation through Alignment-as-Preference Optimization

arXiv:2605.11632v1 Announce Type: new Abstract: Self-generated counterfactual explanations (SCEs) are minimally modified inputs (minimality) generated by large language models (LLMs) that flip their o

Enhancing Target-Guided Proactive Dialogue Systems via Conversational Scenario Modeling and Intent-Keyword Bridging

arXiv:2605.11964v1 Announce Type: new Abstract: A target-guided proactive dialogue system aims to steer conversations proactively toward pre-defined targets, such as designated keywords or specific to

Entropy Polarity in Reinforcement Fine-Tuning: Direction, Asymmetry, and Control

arXiv:2605.11775v1 Announce Type: cross Abstract: Policy entropy has emerged as a fundamental measure for understanding and controlling exploration in reinforcement learning with verifiable rewards (R

Epistemic Uncertainty for Test-Time Discovery

arXiv:2605.11328v1 Announce Type: new Abstract: Automated scientific discovery using large language models relies on identifying genuinely novel solutions. Standard reinforcement learning penalizes hi