safety

Use the Online Network If You Can: Towards Fast and Stable Reinforcement Learning

arXiv:2510.02590v2 Announce Type: replace Abstract: The use of target networks is a popular approach for estimating value functions in deep Reinforcement Learning (RL). While effective, the target net

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Safety

Video Reconstruction using Diffusion-based Image-to-Video Generation with Trajectory Guidance

DGX agent

arXiv:2605.16420v1 Announce Type: new Abstract: This paper addresses the problem of reconstructing missing or dropped frames in top-down drone video of autonomous surface vehicles performing structure

safetyarxiv-cs-cv

19 May 2026

View-Aware Semantic Alignment for Aerial-Ground Person Re-Identification

arXiv:2605.18192v1 Announce Type: new Abstract: Aerial-Ground Person Re-Identification (AGPReID) remains highly challenging due to drastic viewpoint variations between drones and fixed cameras. Existi

Vision Transformer-Conditioned UNet for Domain-Adaptive Semantic Segmentation

arXiv:2605.16393v1 Announce Type: cross Abstract: Semantic segmentation is essential for analysing anatomical features in biomedical research, yet a performance gap remains for Vision Transformers (Vi

Visual Sculpting: Visually-Aligned Planning Representations for Long-Horizon Robot Clay Sculpting

arXiv:2605.17556v1 Announce Type: cross Abstract: Clay sculpting is a nuanced, artistic task involving dexterous manipulation with long-horizon planning to achieve high-level goals. As a robotics prob

VLM-AutoDrive: Post-Training Vision-Language Models for Safety-Critical Autonomous Driving Events

arXiv:2603.18178v2 Announce Type: replace-cross Abstract: The rapid growth of ego-centric dashcam footage presents a major challenge for detecting safety-critical events such as collisions and near-co

Voices in the Loop: Mapping Participatory AI

arXiv:2605.16827v1 Announce Type: new Abstract: Participatory approaches to artificial intelligence are increasingly documented across public, civic, and humanitarian settings, but evidence about how

VolTA-3D: Self-Supervised Learning for Brain MRI using 3D Volumetric Token Alignment

arXiv:2605.16775v1 Announce Type: cross Abstract: Self-supervised learning (SSL) has advanced medical image analysis be enabling learning form large unlabelled data. However, in brain magnetic resonan

Weak-to-Strong Elicitation via Mismatched Wrong Drafts

arXiv:2605.17314v1 Announce Type: cross Abstract: We consider whether off-policy experience from a smaller, weaker model can elicit capability in a stronger learner that on-policy RL fine-tuning (e.g.

What we find most useful about CNA is that the intervention is simple yet powerful. The steering is a multiplicative ablation on a sparse se…

What we find most useful about CNA is that the intervention is simple yet powerful. The steering is a multiplicative ablation on a sparse set of MLP neurons, which makes CNA a clean addition on top of

safetynous-research--x

When Dynamics Shift, Robust Task Inference Wins: Offline Imitation Learning with Behavior Foundation Models Revisited

arXiv:2605.17017v1 Announce Type: cross Abstract: Behavior Foundation Models (BFMs) enable scalable imitation learning (IL) by pretraining task-agnostic representations that can be rapidly adapted to

When Is Rank-1 Steering Cheap? Geometry, Granularity, and Budgeted Search

arXiv:2605.16362v1 Announce Type: cross Abstract: Activation steering offers a lightweight way to control LLMs without retraining, but its effectiveness varies sharply across concepts. Prior work ofte

When Vision Speaks for Sound

arXiv:2605.16403v1 Announce Type: new Abstract: Despite rapid progress in video-capable MLLMs, we find that their apparent audio understanding in videos is often vision-driven: models rely on visual c

Where Pretraining writes and Alignment reads: the asymmetry of Transformer weight space

arXiv:2605.16600v1 Announce Type: cross Abstract: Cross-entropy pretraining and preference alignment update the same transformer weights, but leave geometrically distinct traces. We characterise this

Whispers in the Noise: Surrogate-Guided Concept Awakening via a Multi-Agent Framework

arXiv:2605.18150v1 Announce Type: new Abstract: Diffusion models (DMs) are widely used for text-to-image generation, but their strong generative capabilities also raise concerns about unsafe or undesi

White-Box Sensitivity Auditing with Steering Vectors

arXiv:2601.16398v2 Announce Type: replace-cross Abstract: Algorithmic audits are essential tools for examining systems for properties required by regulators or desired by operators. Current audits of

safetyarxiv-cs-cl

Why Do Safety Guardrails Degrade Across Languages?

arXiv:2605.17173v1 Announce Type: cross Abstract: Large language models exhibit safety degradation in non-English languages. Standard evaluation relies on Jailbreak Success Rate (JSR), which confounds

World Model-Enabled Causal Digital Twins for Semantic Communications in Physical AI Systems

arXiv:2605.16547v1 Announce Type: new Abstract: Semantic communication has emerged as a promising paradigm for enabling goal-oriented networking. However, most existing semantic communication solution

Zero-Shot Textual Explanations via Translating Decision-Critical Features

arXiv:2512.07245v2 Announce Type: replace Abstract: Textual explanations make image classifier decisions transparent by describing the prediction rationale in natural language. Large vision-language m

ZeroSiam: An Efficient Asymmetry for Test-Time Entropy Optimization without Collapse

arXiv:2509.23183v3 Announce Type: replace Abstract: Test-time entropy minimization helps adapt a model to novel environments and incentivize its reasoning capability, unleashing the model's potential

A Differentiable Measure of Algebraic Complexity: Provably Exact Discovery of Group Structures

arXiv:2511.23152v3 Announce Type: replace Abstract: Discovering discrete algebraic rules from data is a fundamental challenge in machine learning. We formalize this problem through Cayley-table comple

A Generative AI Framework for Intelligent Utility Billing CO 2 Analytics and Sustainable Resource Optimisation

arXiv:2605.16250v1 Announce Type: cross Abstract: Distribution utilities are now expected to deliver bills that customers can actually read attach a defensible carbon number to every kWh sold and sche

A Split-Client Approach to Second-Order Optimization

arXiv:2510.15714v3 Announce Type: replace-cross Abstract: Second-order optimization methods offer superior convergence rates but are often bottlenecked by the wall-clock cost of Hessian computation an

a very procedural end. we will never know what the world might be like had OpenAI been forced to fully follow its original mission.

a very procedural end. we will never know what the world might be like had OpenAI been forced to fully follow its original mission. Breaking News: A jury rejected Elon Musk’s lawsuit accusing OpenAI o

safetygary-marcus--x

Accelerated Gradient Descent for Faster Convergence with Minimal Overhead

arXiv:2605.16017v1 Announce Type: new Abstract: In this paper, we present CT-AGD (Curvature-Tuned Accelerated Gradient Descent), an optimization method for non-convex optimization problems in deep lea

ActiveDPO: Active Direct Preference Optimization for Sample-Efficient Alignment

arXiv:2505.19241v2 Announce Type: replace-cross Abstract: The recent success in using human preferences to align large language models (LLMs) has significantly improved their performance in various do

Ada-Diffuser: Latent-Aware Adaptive Diffusion for Decision-Making

arXiv:2605.16054v1 Announce Type: cross Abstract: Recent work has framed decision-making as a sequence modeling problem using generative models such as diffusion models. Although promising, these appr

Adaptive Outer-Loop Control of Quadrotors via Reinforcement Learning

arXiv:2605.16015v1 Announce Type: cross Abstract: Deep Reinforcement Learning (DRL) for quadrotor flight control typically relies on Domain Randomization (DR) for sim-to-real transfer, resulting in ov

AI Consciousness and Existential Risk

arXiv:2511.19115v2 Announce Type: replace Abstract: In AI, the existential risk denotes the hypothetical threat posed by an artificial system that would possess both the capability and the objective,

AI-Mediated Communication Can Steer Collective Opinion

arXiv:2605.16245v1 Announce Type: cross Abstract: Generative artificial intelligence (AI) is increasingly integrated into the online platforms where humans exchange opinions; large language models (LL

Always Learning, Always Mixing: Efficient and Simple Data Mixing All The Time

arXiv:2605.15220v1 Announce Type: cross Abstract: Data mixing decides how to combine different sources or types of data and is a consequential problem throughout language model training. In pretrainin

“Americans are now more comfortable living near a nuclear power plant than an AI data center” -@RachelBitecofer The AI oligarchs took a winn…

“Americans are now more comfortable living near a nuclear power plant than an AI data center” -@RachelBitecofer The AI oligarchs took a winning hand, and with a mixture cigarette-industry level greed

safetygary-marcus--x

An Algebraic Exposition of the Theory of Dyadic Morality

arXiv:2605.16153v1 Announce Type: new Abstract: This paper provides an algebraic exposition of the theory of dyadic morality (TDM), a psychological model of moral judgment grounded in a simple two-nod

An Introduction to Deep Reinforcement and Imitation Learning

arXiv:2512.08052v3 Announce Type: replace-cross Abstract: Embodied agents, such as robots and virtual characters, must continuously select actions to execute tasks effectively, solving complex sequent

ASRU: Activation Steering Meets Reinforcement Unlearning for Multimodal Large Language Models

arXiv:2605.15687v1 Announce Type: cross Abstract: Multimodal large language models (MLLMs) may memorize sensitive cross-modal information during pretraining, making machine unlearning (MU) crucial. Ex

AstraFlow: Dataflow-Oriented Reinforcement Learning for Agentic LLMs

arXiv:2605.15565v1 Announce Type: cross Abstract: Reinforcement learning (RL) is increasingly used to improve the reasoning, coding, and tool-use capabilities of large language models, but agentic RL

Best-of-Both-Worlds for Heavy-Tailed Markov Decision Processes

arXiv:2602.01295v3 Announce Type: replace Abstract: We investigate episodic Markov Decision Processes with heavy-tailed losses (HTMDPs). Existing approaches for HTMDPs are conservative in stochastic e

Beyond Objective-Based Improvement: Stationarity-Aware Expected Improvement for Bayesian Optimization

arXiv:2601.21357v2 Announce Type: replace Abstract: Bayesian Optimization (BO) is a principled framework for optimizing expensive black-box functions, with Expected Improvement (EI) among its most wid

Beyond Performance Disparities: A Three-Level Audit of Representational Harm in CelebA

arXiv:2605.15312v1 Announce Type: cross Abstract: Large-scale facial datasets like CelebA are widely used in computer vision, yet the cultural biases embedded in their labels remain underexplored. Fai

Blending Supervised and Reinforcement Fine-Tuning with Prefix Sampling

arXiv:2507.01679v3 Announce Type: replace-cross Abstract: Existing LLMs-post-training techniques are broadly categorized into supervised fine-tuning (SFT) and reinforcement fine-tuning (RFT). Each par

Can we all agree that Dario played the “Ooh! AI scary!” card one time too many?

Can we all agree that Dario played the “Ooh! AI scary!” card one time too many? “Americans are now more comfortable living near a nuclear power plant than an AI data center” -@RachelBitecofer The AI o

safetygary-marcus--x

Constrained MPC-Based Motion Planning for Morphing Quadrotors in Ultra-Narrow Passages under Limited Perception

arXiv:2605.15999v1 Announce Type: new Abstract: This paper introduces a motion planning framework to plan morphology and trajectory for morphing quadrotors under extremely constrained environments. We

safetyarxiv-cs-ro

Controllable Molecular Generative Foundation Models

arXiv:2605.15354v1 Announce Type: new Abstract: Despite the success of foundation models in language and vision, molecular graph generation still lacks a unified framework for heterogeneous design tas

CTF4Nuclear: Common Task Framework for Nuclear Fission and Fusion Models

arXiv:2605.15549v1 Announce Type: cross Abstract: The demand for clean energy is ever increasing, with new nuclear technologies presenting a complementary solution to renewable energies. However, desi

DebiasRAG: A Tuning-Free Path to Fair Generation in Large Language Models through Retrieval-Augmented Generation

arXiv:2605.16113v1 Announce Type: cross Abstract: Large language models (LLMs) have achieved unprecedented success due to their exceptional generative capabilities. However, because they depend on kno

Decomposed Vision-Language Alignment for Fine-Grained Open-Vocabulary Segmentation

arXiv:2605.15942v1 Announce Type: cross Abstract: Open-vocabulary segmentation models often struggle to generalize to unseen combinations of object categories and attributes, because fine-grained desc

Deep Double Q-learning

arXiv:2507.00275v2 Announce Type: replace-cross Abstract: Double Q-learning is a classical control algorithm that mitigates the maximization bias of Q-learning. To do so, it explicitly trains two inde

DeltaPrompts: Escaping the Zero-Delta Trap in Multimodal Distillation

arXiv:2605.15532v1 Announce Type: cross Abstract: Distillation enables compact Vision-Language Models (VLMs) to obtain strong reasoning capabilities, yet the prompts driving this process are typically