Search: “safety”

Strat-Reasoner: Reinforcing Strategic Reasoning of LLMs in Multi-Agent Games

arXiv:2605.04906v2 Announce Type: replace Abstract: While Large Language Models (LLMs) excel in certain reasoning tasks, they struggle in multi-agent games where the final outcome depends on the joint

Content type

All Blog X Post Paper YouTube Reddit GitHub

Clear filters

Subspace-Guided Semantic and Topological Invariant Registration for Annotation-Free Ultrasound Plane Quality Control

DGX agent

arXiv:2605.25396v1 Announce Type: cross Abstract: Reliable quality control (QC) of ultrasound images is essential for both real-time acquisition guidance and retrospective clinical audit, yet existing

safetyarxiv-cs-ai

26 May 2026

Summoning the Oracle to Slay It: Mitigating Look-Ahead Bias in Financial Backtesting with Large Language Models

arXiv:2605.24564v1 Announce Type: new Abstract: Backtesting large language models (LLMs) on historical financial data is unreliable because pre-training cuts off after the events happened. An LLM trai

TapSampling: Inference-Time Sampling with a Task-Progress-Understanding Verifier for Robotic Manipulation

arXiv:2605.25547v1 Announce Type: new Abstract: Existing embodied control research demonstrates remarkable performance improvements by scaling training data and model size. We instead explore inferenc

safetyarxiv-cs-ro

Task-Aligned Self-Supervised Learning for Medical Image Analysis: A Systematic Review and Practical Design Guidelines

arXiv:2605.23995v1 Announce Type: cross Abstract: Self-supervised learning (SSL) has emerged as a promising paradigm for addressing the annotation bottleneck in medical imaging by learning representat

temporarily putting PhD in my social media name because this tweet from Elon is so asinine and because Elon’s goons went after a friend for …

temporarily putting PhD in my social media name because this tweet from Elon is so asinine and because Elon’s goons went after a friend for supporting the poor kid that Elon so rudely attacked. @iScie

The Behavioral Credibility Trilemma: When Calibrated Autonomy Becomes Impossible

arXiv:2605.25739v1 Announce Type: new Abstract: We prove that no reinforcement learning policy with confidence-gated autonomy can simultaneously achieve maximum helpfulness, optimal calibration, and f

The Concept Allocation Zone: Tracking How Concepts Form Across Transformer Depth

arXiv:2605.24856v1 Announce Type: cross Abstract: Concept formation in transformer language models is depth-extended, not a single-layer event: concepts emerge gradually across a contiguous region of

The Implicit Bias of Adam and Muon on Smooth Homogeneous Neural Networks

arXiv:2602.16340v3 Announce Type: replace Abstract: We study the implicit bias of momentum-based optimizers on smooth homogeneous models. We show that extit{momentum steepest descent} algorithms like

The main thing that come through reading the SpaceX S-1 is how much Musk and Altman have in common.

This post draws a comparison between Elon Musk and Sam Altman based on insights from SpaceX's S-1 filing, highlighting shared characteristics or philosophies between the two tech leaders. The analysis

The Many Faces of On-Policy Distillation: Pitfalls, Mechanisms, and Fixes

arXiv:2605.11182v2 Announce Type: replace Abstract: On-policy distillation (OPD) and on-policy self-distillation (OPSD) have emerged as promising post-training methods for large language models, offer

The Path Matters: Learning a Token-Commitment Policy for Diffusion Language Models

arXiv:2605.24697v1 Announce Type: cross Abstract: Diffusion large language models promise faster generation by refining many token positions in parallel, but this parallelism introduces a hidden contr

The SpaceX IPO is a shitshow. When you peel back the layers, you realize how thoroughly corrupt it is.

The SpaceX IPO is a shitshow. When you peel back the layers, you realize how thoroughly corrupt it is. SpaceX’s unconventional corporate arrangements appear to benefit Elon Musk at the expense of othe

TopoAlign: Topology-Aware Visual Representation Alignment

arXiv:2605.25541v1 Announce Type: cross Abstract: Neural networks encode inputs as high-dimensional vectors, known as representations, that capture how models process data by encoding task-relevant st

Toward Reliable Design of LLM-Enabled Agentic Workflows: Optimizing Latency-Reliability-Cost Tradeoffs

arXiv:2605.23929v1 Announce Type: new Abstract: Modern AI systems increasingly rely on workflows composed of multiple interacting agents, some powered by large language models (LLMs) and others by con

Towards Cognitively-Faithful Decision-Making Models to Improve AI Alignment

arXiv:2509.04445v2 Announce Type: replace Abstract: Recent AI trends seek to align AI models to learned human-centric objectives, such as personal preferences, utility, or societal values. Using stand

Towards Inclusive Toxic Content Moderation: Addressing Vulnerabilities to Adversarial Attacks in Toxicity Classifiers Tackling LLM-generated Content

arXiv:2509.12672v2 Announce Type: replace Abstract: The volume of machine-generated content online has grown dramatically due to the widespread use of Large Language Models (LLMs), leading to new chal

Towards Low-Gravity Planetary Exploration using Reinforcement Learning for Walking, Jumping, and In-flight Attitude Control

arXiv:2605.24643v1 Announce Type: new Abstract: This paper presents reinforcement learning (RL) policies for dynamic quadrupedal locomotion in planetary exploration scenarios. Building on a taskoptimi

safetyarxiv-cs-ro

Towards the Connection between Activation Sparsity and Flat Minima

arXiv:2605.25612v1 Announce Type: cross Abstract: The observation that activation sparsity emerges in MLP blocks of standardly trained Transformers offers an opportunity to drastically reduce computat

Towards Understanding Adam Convergence on Highly Degenerate Polynomials

arXiv:2603.09581v2 Announce Type: replace Abstract: Adam is a widely used optimization algorithm in deep learning, yet the specific class of objective functions where it exhibits inherent advantages r

Trait-Aware Policy Optimization for Autoregressive Multi-Trait Essay Scoring

arXiv:2605.25731v1 Announce Type: new Abstract: Multi-trait essay scoring aims to provide fine-grained evaluation of writing quality across multiple dimensions. However, how to effectively post-train

Trust-Aware Joint Feature-Prediction Discrepancy for Robust Domain Adaptation

arXiv:2605.25119v1 Announce Type: cross Abstract: Domain adaptation aims to mitigate performance degradation caused by distribution shifts between a labeled source domain and an unlabeled or sparsely

Uncertainty-DTW for Sequences and Visual Tokens

arXiv:2605.25110v1 Announce Type: cross Abstract: Aligning structured data is a fundamental problem in computer vision and machine learning, underlying tasks such as time series analysis, human action

Unifying Value Alignment and Assignment in Cross-Domain Offline Reinforcement Learning with Heterogeneous Datasets

arXiv:2605.24862v1 Announce Type: new Abstract: Cross-domain offline reinforcement learning (RL) aims to learn a policy in the target domain with a limited target domain dataset and a source domain da

UniRank: End-to-End Domain-Specific Reranking of Hybrid Text-Image Candidates

arXiv:2603.29897v2 Announce Type: replace-cross Abstract: Reranking is a critical component in many information retrieval pipelines. Despite remarkable progress in text-only settings, multimodal reran

Universal Boosts, Specific Suppressors: Sparse Autoencoder Steering of Medical Vision-Language Models

arXiv:2605.24977v1 Announce Type: cross Abstract: Medical vision-language models (VLMs) often hallucinate findings when generating chest X-ray reports: they fabricate findings that are not present in

University of California STEM professors want standardized tests back due to severe math deficiencies among students: “We now observe prepar…

University of California STEM professors want standardized tests back due to severe math deficiencies among students: “We now observe preparation gaps so severe that instructors must reteach middle sc

“Unserious, empty, hallucinatory, and borderline dishonest” - the prospectus for a company that the S&P 500 is about to jam down your throat…

Gary Marcus criticizes a company's prospectus as containing unserious, empty, and potentially dishonest claims before its inclusion in the S&P 500 index. The post appears to highlight concerns about i

VEN-VL: A Visual Ensemble MoE Framework for Effective and Efficient Multi-Modal Understanding

arXiv:2605.25952v1 Announce Type: cross Abstract: Despite the remarkable progress achieved by recent efficient methods in accelerating multimodal understanding, they still suffer from noticeable perfo

Vision-Guided Outdoor Flight and Obstacle Evasion via Reinforcement Learning

arXiv:2605.24449v1 Announce Type: cross Abstract: Although quadcopters boast impressive traversal capabilities enabled by their omnidirectional maneuverability, the need for continuous pilot control i

What Gets Cited: Competitive GEO in AI Answer Engines

arXiv:2605.25517v1 Announce Type: new Abstract: AI answer engines generate answers from retrieved pages but cite only a few sources. This makes visibility depend not just on ranking, but on being cite

When Does Multi-Agent RL Improve LLM Workflows? Workflow, Scale, and Policy-Sharing Tradeoffs

arXiv:2605.24202v1 Announce Type: new Abstract: Multi-agent LLM workflows route inference through specialized roles to lift end-task accuracy, but jointly training those roles with reinforcement learn

When Self-Belief Misleads: Active Label Acquisition for Reinforcement Learning with Verifiable Rewards

arXiv:2605.25864v1 Announce Type: cross Abstract: Large Language Models (LLMs) have achieved remarkable advancements in reasoning capabilities empowered by Reinforcement Learning with Verifiable Rewar

when you think about it SpaceX has cumulatively lost a lot less than OpenAI so maybe it’s a bargain? or maybe we just shouldn’t value compan…

when you think about it SpaceX has cumulatively lost a lot less than OpenAI so maybe it’s a bargain? or maybe we just shouldn’t value companies at a trillion dollars until they actually show evidence

Workflow cleanup tools for ComfyUI: Visual Fold, group folding, and node alignment

Visual Fold is a tool for simple visual organization of ComfyUI workflows that does not turn selected nodes into a subgraph or change workflow logic. Group folding and node alignment features enable c

safetyr-stablediffusion

X-DiffVLA: X-Embodied Diffusion Action Heads for Vision-Language-Action Models

arXiv:2605.25044v1 Announce Type: new Abstract: Learning universal policies from cross-embodied data remains a fundamental challenge in robotics. Although Vision-Language-Action (VLA) models are pre-t

safetyarxiv-cs-ro

XRPO: Pushing the limits of GRPO with Targeted Exploration and Exploitation

arXiv:2510.06672v3 Announce Type: replace Abstract: Reinforcement learning algorithms such as GRPO have driven recent advances in large language model (LLM) reasoning. While scaling the number of roll

100% agree. and this is important. and it will affect you, personally.

100% agree. and this is important. and it will affect you, personally. The SpaceX IPO is the most brazen retail fleecing in modern market history. NASDAQ has REWRITTEN the index rules specifically for

4DThinker: Thinking with 4D Imagery for Dynamic Spatial Understanding

arXiv:2605.05997v2 Announce Type: replace Abstract: Dynamic spatial reasoning from monocular video is essential for bridging visual intelligence and the physical world, yet remains challenging for vis

safetyarxiv-cs-cv

ALIVE: Awakening LLM Reasoning via Adversarial Learning and Instructive Verbal Evaluation

arXiv:2602.05472v2 Announce Type: replace Abstract: The quest for expert-level reasoning in Large Language Models (LLMs) has been hampered by a persistent extit{reward bottleneck}: traditional reinfor

ARMS: Automatic Reward Shaping for Sparse-Reward Multi-Agent Reinforcement Learning

arXiv:2605.23562v1 Announce Type: cross Abstract: Sparse rewards are a major bottleneck in multi-agent reinforcement learning (MARL), where simultaneous learning induces non-stationarity and makes rew

As one of the first people to warn about a possible AI backlash—years ago—let me tell you this: it’s going to get much, much worse. It break…

As one of the first people to warn about a possible AI backlash—years ago—let me tell you this: it’s going to get much, much worse. It breaks my heart that AI—something I spent my whole life thinking

Assessing Predictive Models for Fairness Based on Movement Patterns

arXiv:2605.23234v1 Announce Type: new Abstract: Assessing the spatial fairness of predictive models involves establishing whether they are statistically penalizing (favoring) individuals associated wi

B-GRTO: Bootstrapped Group Relative Tool Optimization for Referring Segmentation

arXiv:2605.23500v1 Announce Type: new Abstract: Segmentation is a fundamental task in computer vision, underpinning pixel-level scene understanding and serving as a cornerstone for applications rangin

safetyarxiv-cs-cv

Beyond Binary Edits Robust Multimodal Knowledge Editing with Adversarial Subspace Alignment

arXiv:2605.23780v1 Announce Type: new Abstract: Multimodal large language models (MLLMs) need efficient mechanisms to update knowledge without degrading existing capabilities. While intrinsic multimod

Beyond VLM-Based Rewards: Diffusion-Native Latent Reward Modeling

arXiv:2602.11146v2 Announce Type: replace-cross Abstract: Preference optimization for diffusion and flow-matching models relies on reward functions that are both discriminatively robust and computatio

Bridging AI and Clinical Reasoning: Abductive Explanations for Alignment on Critical Symptoms

arXiv:2602.13985v2 Announce Type: replace Abstract: Artificial intelligence (AI) has demonstrated strong potential in clinical diagnostics, often achieving accuracy comparable to or exceeding that of

CapTrack: Multifaceted Evaluation of Forgetting in LLM Post-Training

arXiv:2603.06610v2 Announce Type: replace Abstract: Large language model (LLM) post-training enhances latent skills, unlocks value alignment, improves performance, and enables domain adaptation. Unfor