safety

A unified perspective on fine-tuning and sampling with diffusion and flow models

arXiv:2605.00229v1 Announce Type: cross Abstract: We study the problem of training diffusion and flow generative models to sample from target distributions defined by an exponential tilting of a base

Adaptive Equilibrium: Dynamic Weighting Framework for Generalized Interruption of DeepFake Models

DGX agent

arXiv:2605.00443v1 Announce Type: cross Abstract: The advancement of generalized deepfake disruption is constrained by the interruption imbalance, a fundamental bottleneck inherent to the generation o

safetyarxiv-cs-cv

4 May 2026

Agent Capsules: Quality-Gated Granularity Control for Multi-Agent LLM Pipelines

arXiv:2605.00410v1 Announce Type: new Abstract: A multi-agent pipeline with N agents typically issues N LLM calls per run. Merging agents into fewer calls (compound execution) promises token savings,

Almost every warning that I have issued over the last several years has come true. You better hope to God or Darwin or whoever you believe i…

Almost every warning that I have issued over the last several years has come true. You better hope to God or Darwin or whoever you believe in that this one is wrong. Accidental nuclear war is a potent

🌐At the @UN, @Yoshua_Bengio briefed @antonioguterres on AI governance, followed by a reception with @CanadaUN & @DavidLametti opened by @Eg…

🌐At the @UN, @Yoshua_Bengio briefed @antonioguterres on AI governance, followed by a reception with @CanadaUN & @DavidLametti opened by @EgriseldaL As AI risks grow, global coordination matters. 🔗AI D

safetyyoshua-bengio--x

Augmented Lagrangian Multiplier Network for State-wise Safety in Reinforcement Learning

arXiv:2605.00667v1 Announce Type: new Abstract: Safety is a primary challenge in real-world reinforcement learning (RL). Formulating safety requirements as state-wise constraints has become a prominen

Beyond Prompt-Induced Lies: Investigating LLM Deception on Benign Prompts

arXiv:2508.06361v4 Announce Type: replace Abstract: Large Language Models (LLMs) are widely deployed in reasoning, planning, and decision-making tasks, making their trustworthiness critical. A signifi

Beyond Suffixes: Token Position in GCG Adversarial Attacks on Large Language Models

arXiv:2602.03265v2 Announce Type: replace Abstract: Large Language Models (LLMs) have seen widespread adoption across multiple domains, creating an urgent need for robust safety alignment mechanisms.

Bias in Large Language Models: Origin, Evaluation, and Mitigation

arXiv:2411.10915v2 Announce Type: replace Abstract: Large Language Models (LLMs) have revolutionized natural language processing, but their susceptibility to biases poses significant challenges. This

BlenderRAG: High-Fidelity 3D Object Generation via Retrieval-Augmented Code Synthesis

arXiv:2605.00632v1 Announce Type: new Abstract: Automatic generation of executable Blender code from natural language remains challenging, with state-of-the-art LLMs producing frequent syntactic error

BOLT: Online Lightweight Adaptation for Preparation-Free Heterogeneous Cooperative Perception

arXiv:2605.00405v1 Announce Type: new Abstract: Most existing heterogeneous cooperative perception methods depend on prior preparation like offline joint training or tailored collaborator-model adapta

Can Small Language Models Handle Context-Summarized Multi-Turn Customer-Service QA? A Synthetic Data-Driven Comparative Evaluation

arXiv:2602.00665v3 Announce Type: replace Abstract: Customer-service question answering (QA) systems increasingly rely on conversational language understanding. While Large Language Models (LLMs) achi

Confirmed 2.5 years later, by Greg Brockman.

Confirmed 2.5 years later, by Greg Brockman. Remember how Sam Altman told the US Senate he had no “direct” investment in OpenAI? and how they gushed over his apparent selflessness? 👉He didn’t mention

Conformalized Quantum DeepONet Ensembles for Scalable Operator Learning with Distribution-Free Uncertainty

arXiv:2605.00330v1 Announce Type: new Abstract: Operator learning enables fast surrogate modeling of high-dimensional dynamical systems, but existing approaches face two fundamental limitations: quadr

Data Deletion Can Help in Adaptive RL

arXiv:2605.00298v1 Announce Type: new Abstract: Deploying reinforcement learning policies in the real world requires adapting to time-varying environments. We study this problem in the contextual Mark

Debate-Enhanced Pseudo Labeling and Frequency-Aware Progressive Debiasing for Weakly-Supervised Camouflaged Object Detection with Scribble Annotations

arXiv:2512.20260v5 Announce Type: replace Abstract: Weakly-Supervised Camouflaged Object Detection (WSCOD) aims to locate and segment objects that are visually concealed within their surrounding scene

Decentralized Proximal Stochastic Gradient Langevin Dynamics

arXiv:2605.00723v1 Announce Type: cross Abstract: We propose Decentralized Proximal Stochastic Gradient Langevin Dynamics (DE-PSGLD), a decentralized Markov chain Monte Carlo (MCMC) algorithm for samp

Disentangled Safety Adapters Enable Efficient Guardrails and Flexible Inference-Time Alignment

arXiv:2506.00166v2 Announce Type: replace-cross Abstract: Existing paradigms for ensuring AI safety, such as guardrail models and alignment training, often compromise either inference efficiency or de

@dromanocpm @GaryMarcus Two stood up. I know that one told the truth. Gary Marcus and Sam Altman, Senate Hearing on Oversight for Artificial…

Gary Marcus and Sam Altman testified before the Senate regarding AI oversight, with Marcus noting that two individuals stood up during the hearing and asserting that one of them told the truth. This r

Dynamic-TD3: A Novel Algorithm for UAV Path Planning with Dynamic Obstacle Trajectory Prediction

arXiv:2605.00059v1 Announce Type: new Abstract: Deep reinforcement learning (DRL) finds extensive application in autonomous drone navigation within complex, high-risk environments. However, its practi

safetyarxiv-cs-ro

Estimating LLM Grading Ability and Response Difficulty in Automatic Short Answer Grading via Item Response Theory

arXiv:2605.00238v1 Announce Type: new Abstract: Automated short answer grading (ASAG) with large language models (LLMs) is commonly evaluated with aggregate metrics such as macro-F1 and Cohen's kappa.

Exploring LLM biases to manipulate AI search overview

arXiv:2605.00012v1 Announce Type: cross Abstract: Modern large language models (LLMs) are used in many business applications in general, and specifically in web search systems and applications that ge

Fair Dataset Distillation via Cross-Group Barycenter Alignment

arXiv:2605.00185v1 Announce Type: new Abstract: Dataset Distillation aims to compress a large dataset into a small synthetic one while maintaining predictive performance. We show that as different dem

Fairness of Classifiers in the Presence of Constraints between Features

arXiv:2605.00592v1 Announce Type: new Abstract: In Machine Learning, an accepted definition of fairness of a decision taken by a classifier is that it should not depend on protected features, such as

Foundation AI Models for Aerosol Optical Depth Estimation from PACE Satellite Data

arXiv:2605.00678v1 Announce Type: new Abstract: Aerosol Optical Depth (AOD) retrieval is essential for Earth observation, supporting applications from air quality monitoring to climate studies. Conven

FreeRet: MLLMs as Training-Free Retrievers

arXiv:2509.24621v2 Announce Type: replace Abstract: Multimodal large language models (MLLMs) are emerging as versatile foundations for mixed-modality retrieval. Yet, they often require heavy post-hoc

@GaryMarcus to yann - yank overhyped AI claims after they've been marcussed https://julian-goldstahl.blogspot.com/2025/09/dictionairy.html

Gary Marcus criticizes overhyped AI claims, with the term 'marcussed' referring to his practice of debunking exaggerated AI assertions. The post appears to reference a discussion with Yann LeCun about

🚨 GREG BROCKMAN JUST CONFESSED UNDER OATH Q: You have an ownership interest in this cap profit company. Brockman: That is accurate. Q: And …

🚨 GREG BROCKMAN JUST CONFESSED UNDER OATH Q: You have an ownership interest in this cap profit company. Brockman: That is accurate. Q: And you invested 0 in order to acquire that interest. Correct? Br

Greg Brockman’s diary is well on its way to being the most famous, impactful diary of the 21st century.

Greg Brockman’s diary is well on its way to being the most famous, impactful diary of the 21st century. takeaway from Greg Brockman testimony at Elon vs. OpenAI trial today is that no grown man should

Hey filthy rich AI warlords who have alienated most of the population, don’t say I didn’t try to warn you.🤷‍♂️

Hey filthy rich AI warlords who have alienated most of the population, don’t say I didn’t try to warn you.🤷‍♂️ “Bubble or not, the AI backlash is validating what one researcher and critic has been say

How 'emotion AI', the use of facial and sentiment analysis tools to track workers' moods, is seeping into white-collar jobs amid concerns over privacy and bias (Ellen Cushing/The Atlantic)

Ellen Cushing / The Atlantic: How “emotion AI”, the use of facial and sentiment analysis tools to track workers' moods, is seeping into white-collar jobs amid concerns over privacy and bias — The good

safetytechmeme

How Language Models Process Out-of-Distribution Inputs: A Two-Pathway Framework

arXiv:2605.00269v1 Announce Type: new Abstract: Recent white-box OOD detection methods for LLMs -- including CED, RAUQ, and WildGuard confidence scores -- appear effective, but we show they are struct

HyCOP: Hybrid Composition Operators for Interpretable Learning of PDEs

arXiv:2605.00820v1 Announce Type: cross Abstract: We introduce HyCOP, a modular framework that learns parametric PDE solution operators by composing simple modules (advection, diffusion, learned closu

i love the recent explosion of interactive data visualizations. more plz

i love the recent explosion of interactive data visualizations. more plz Who actually shapes AI policy in the U.S.? We mapped 1,812 entities: 745 people, 918 organizations, 2,925 relationships. Fronti

safetyyohei-nakajima--x

If you need proof of why AI output can't be trusted by default watch that page for more than 1 minute for any given LLM that produces output…

If you need proof of why AI output can't be trusted by default watch that page for more than 1 minute for any given LLM that produces output in there. An amazing time capsule of the unevenness of curr

Impact of Task Phrasing on Presumptions in Large Language Models

arXiv:2605.00436v1 Announce Type: new Abstract: Concerns with the safety and reliability of applying large-language models (LLMs) in unpredictable real-world applications motivate this study, which ex

Import AI 455: AI systems are about to start building themselves.

This newsletter entry discusses advancements in automating AI research and development processes, exploring how artificial intelligence systems are becoming capable of autonomously designing and impro

safetyimport-ai

InpaintSLat: Inpainting Structured 3D Latents via Initial Noise Optimization

arXiv:2605.00664v1 Announce Type: new Abstract: We present a training-free approach for controllable 3D inpainting based on initial noise optimization. In the structured 3D latent diffusion framework,

Intelligent Elastic Feature Fading: Enabling Model Retrain-Free Feature Efficiency Rollouts at Scale

arXiv:2605.00324v1 Announce Type: cross Abstract: Large-scale ranking systems depend on thousands of features derived from user behavior across multiple time horizons. Typically requires model retrain

Jensen Huang said Nvidia's market share of AI accelerators in China has 'now dropped to zero' and that US export policy 'has already largely backfired' (Anton Shilov/Tom's Hardware)

Anton Shilov / Tom's Hardware: Jensen Huang said Nvidia's market share of AI accelerators in China has “now dropped to zero” and that US export policy “has already largely backfired” — US export restr

safetytechmeme

Last-Iterate Convergence of General Parameterized Policies in Constrained MDPs

arXiv:2408.11513v2 Announce Type: replace Abstract: This paper focuses on learning a Constrained Markov Decision Process (CMDP) via general parameterized policies. We propose a Primal-Dual based Regul

Learn where to Click from Yourself: On-Policy Self-Distillation for GUI Grounding

arXiv:2605.00642v1 Announce Type: cross Abstract: Graphical User Interface (GUI) grounding maps natural language instructions to the visual coordinates of target elements and serves as a core capabili

Learning Coarse-to-Fine Osteoarthritis Representations under Noisy Hierarchical Labels

arXiv:2605.00718v1 Announce Type: new Abstract: Knee osteoarthritis (OA) assessment involves a natural but often underused label hierarchy: a coarse binary OA decision and a fine-grained Kellgren--Law

Learning How and What to Memorize: Cognition-Inspired Two-Stage Optimization for Evolving Memory

arXiv:2605.00702v1 Announce Type: new Abstract: Large language model (LLM) agents require long-term user memory for consistent personalization, but limited context windows hinder tracking evolving pre

Learning physically grounded traffic accident reconstruction from public accident reports

arXiv:2605.00050v1 Announce Type: cross Abstract: Traffic accidents are routinely documented in textual reports, yet physically grounded accident reconstruction remains difficult because detailed scen

Learning while Deploying: Fleet-Scale Reinforcement Learning for Generalist Robot Policies

arXiv:2605.00416v1 Announce Type: new Abstract: Generalist robot policies increasingly benefit from large-scale pretraining, but offline data alone is insufficient for robust real-world deployment. De

safetyarxiv-cs-ro

Linking Behaviour and Perception to Evaluate Meaningful Human Control over Partially Automated Driving

arXiv:2605.00556v1 Announce Type: cross Abstract: Partial driving automation creates a tension: drivers remain legally responsible for vehicle behaviour, yet their active control is significantly redu

safetyarxiv-cs-ro

MemRouter: Memory-as-Embedding Routing for Long-Term Conversational Agents

arXiv:2605.00356v1 Announce Type: new Abstract: Long-term conversational agents must decide which turns to store in external memory, yet recent systems rely on autoregressive LLM generation at every t