All entries

When Benign Inputs Lead to Severe Harms: Eliciting Unsafe Unintended Behaviors of Computer-Use Agents

arXiv:2602.08235v2 Announce Type: replace-cross Abstract: Although computer-use agents (CUAs) hold significant potential to automate increasingly complex OS workflows, they can demonstrate unsafe unin

Model Releases

When Do Local Score Models Extrapolate Across Size? A Diagnostic Theory and Benchmark

DGX agent

arXiv:2606.09705v1 Announce Type: new Abstract: Scientific generative modeling often requires size transfer, where models trained on small systems are evaluated on larger ones. While translation-invar

model-releasesarxiv-cs-lg

9 Jun 2026

When Does Delegation Beat Majority? A Delegation-Based Aggregator for Multi-Sample LLM Inference

arXiv:2606.08098v1 Announce Type: new Abstract: Majority voting over sampled answers is the dominant unsupervised aggregator for multi-sample LLM inference. We show that piping the signals every sampl

researcharxiv-cs-ai

When No Answer Is Correct: Diagnosing Absent Answer Detection for MLLMs in Video Understanding

arXiv:2606.08239v1 Announce Type: new Abstract: Multimodal large language models (MLLMs) have made substantial advancements in video understanding, yet the reliability of their responses remains under

researcharxiv-cs-ai

Agents

When Should an AI Scientist Stop? Verifiable Experiment Steering and Refusal for Autonomous Discovery

arXiv:2606.07576v1 Announce Type: new Abstract: We present CARTOGRAPH, a verification layer for AI scientists that couples unresolved-subspace experiment steering (select), explicit ambiguity closure

agentsarxiv-cs-lg

Agents

When Video Misreads: Closed-Loop Distillation of Reading Heuristics for Exploratory Manipulation Trace QA

arXiv:2606.08542v1 Announce Type: cross Abstract: Exploratory manipulation often turns an apparent failed attempt into the key evidence for what to do next. For example, a robot pulls a locked cabinet

agentsarxiv-cs-ai

Local Ai

When Vision Misleads, Let Location Speak: A Worldwide Image Geo-Localization Method via Location Attention Mechanism and Large Multimodal Models

arXiv:2606.08918v1 Announce Type: new Abstract: Worldwide image geo-localization aims to determine the capture location of an image on a global scale. Existing methods often mislocalize images by matc

local-aiarxiv-cs-cv

Where Does the Answer Come From? Benchmarking View-Level Visual Evidence Identification in Multi-View MLLMs for Autonomous Driving

arXiv:2606.09644v1 Announce Type: cross Abstract: Multimodal large language models (MLLMs) achieve strong results on visual reasoning benchmarks, but answer accuracy alone does not indicate whether a

Where Instruction Hierarchy Breaks: Diagnosing and Repairing Failures in Reasoning Language Models

arXiv:2606.07808v1 Announce Type: new Abstract: Reasoning language models deployed in agentic workflows must follow an instruction hierarchy: when instructions from different sources conflict, the mod

Where the Score Lives: A Wavelet View of Diffusion

arXiv:2606.08309v1 Announce Type: cross Abstract: Score-based generative models have had remarkable success over the last decade in generating a diverse set of visually plausible images. A variety of

researcharxiv-cs-cv

WhiFlash: Accelerating Speculative Decoding with Token-Level Cross-Paradigm Routing

arXiv:2606.07710v1 Announce Type: cross Abstract: The autoregressive nature of large language models (LLMs) remains a significant bottleneck for inference, particularly in complex agentic workloads. W

safetyarxiv-cs-ai

Who Earns the Safety? Intervention-Aware Quantum Predictive Control with Safety Attribution

arXiv:2606.09778v1 Announce Type: cross Abstract: Hard safety filters are increasingly placed downstream of learned controllers to guarantee constraint satisfaction at run time. Yet a filtered control

Why are governments in the dark about the threat posed by superintelligent AI? Because the industry is spending to keep them there. In ABC's…

Why are governments in the dark about the threat posed by superintelligent AI? Because the industry is spending to keep them there. In ABC's new Four Corners documentary, ControlAI's US Director Conno

safetyconnor-leahy--x

Local Ai

why does ollama unloads models automatically?

Ollama automatically unloads inactive models from memory based on inactivity parameters, with models being removed after a specified duration of non-use . By default, Ollama unloads a model after 5 mi

local-air-ollama

Why Limit the Residual Stream to Layers and Not Tokens? Persistent Memory for Continuous Latent Reasoning

arXiv:2606.07720v1 Announce Type: new Abstract: Large language models (LLMs) have demonstrated remarkable reasoning abilities on mathematical and multi-hop planning tasks. The CoCoNuT (Chain of Contin

researcharxiv-cs-ai

Wispy to Voluminous: Prior-free Multi-view Capture of Strand-level Facial Hair

arXiv:2606.08041v1 Announce Type: cross Abstract: Facial hair is a defining trait of personal identity, yet remains a critical bottleneck for digital avatars. Recent volumetric methods achieve photore

applicationsarxiv-cs-cv

Wordle 1,815 6/6 ⬛⬛🟨⬛⬛ ⬛🟨⬛⬛⬛ ⬛🟩⬛🟩⬛ 🟩🟩⬛🟩⬛ 🟩🟩🟩🟩⬛ 🟩🟩🟩🟩🟩

model-releasesanthropic--x

This post shows a completed Wordle game (#1,815) solved on the sixth and final attempt, with the emoji grid indicating which letters were correct, misplaced, or absent in each guess. Anthropic shared

X-OP: Cross-Morphology Whole-Body Teleoperation via MPC Retargeting

arXiv:2606.07934v1 Announce Type: new Abstract: Whole-body teleoperation is essential for scalable robot data collection in loco-manipulation tasks, yet existing approaches relying on exoskeleton suit

safetyarxiv-cs-ro

X-Palm: Paired Multispectral-to-Smartphone Dataset for Cross-Domain Palmprint Authentication

arXiv:2606.08437v1 Announce Type: cross Abstract: Palmprint modality offers a privacy-preserving biometric solution, yet its deployment is hindered by the domain gap between controlled enrollment and

applicationsarxiv-cs-cv

Tutorials

XAInomaly: Explainable and Interpretable Deep Contractive Autoencoder for O-RAN Traffic Anomaly Detection

arXiv:2502.09194v1 Announce Type: cross Abstract: Generative Artificial Intelligence (AI) techniques have become integral part in advancing next generation wireless communication systems by enabling s

tutorialsarxiv-cs-ai

XCR-Bench: Benchmarking Cross-Cultural Reasoning in LLMs via Culture-Specific Items and Hall's Triad

arXiv:2601.14063v2 Announce Type: replace-cross Abstract: Cross-cultural competence in large language models (LLMs) requires understanding and adapting Culture-Specific Items (CSIs) across varying cul

Tools

Yesterday at London Tech Week we announced the Billion Pound Build competition. Teams can enter to secure a share of £1M in Computer credits…

Yesterday at London Tech Week we announced the Billion Pound Build competition. Teams can enter to secure a share of £1M in Computer credits by using Perplexity Computer to build their company. The pi

toolsperplexity--x

You can try Claude Fable 5 as part of Devin Cloud’s Ultra agent. Devin Ultra is our smartest and most capable agent, which excels at long-ho…

model-releasescognition-ai--x

You can try Claude Fable 5 as part of Devin Cloud’s Ultra agent. Devin Ultra is our smartest and most capable agent, which excels at long-horizon tasks and debugging. We tuned the harness so Ultra cos

you gotta love these juxtapositions.

Gary Marcus highlights interesting contrasts or contradictions, likely related to AI, technology, or cognitive science based on his expertise in these areas. The post format suggests he's pointing out

safetygary-marcus--x

Your Model Already Knows: Attention-Guided Safety Filter for Vision-Language-Action Models

model-releasesarxiv-cs-lg

arXiv:2606.09749v1 Announce Type: cross Abstract: Vision-Language-Action (VLA) models have demonstrated impressive end-to-end performance across a variety of robotic manipulation tasks. However, these

Your Self-Play Algorithm is Secretly an Adversarial Imitator: Understanding LLM Self-Play through the Lens of Imitation Learning

arXiv:2602.01357v2 Announce Type: replace Abstract: Self-play post-training methods has emerged as an effective approach for finetuning large language models and turn the weak language model into stro

safetyarxiv-cs-lg

Zero and Few Shot Load Forecasting with Large Language Models

arXiv:2411.11350v2 Announce Type: replace Abstract: Deep learning models have shown strong performance in load forecasting, but they generally require large amounts of data for model training before b

applicationsarxiv-cs-lg

Zero-Flow Encoders

arXiv:2602.00797v3 Announce Type: replace-cross Abstract: Flow-based methods have achieved significant success in various generative modeling tasks, capturing nuanced details within complex data distr

applicationsarxiv-cs-lg

Zero-Parameter Geometric Gating for Temporally Stable Low-Altitude UAV Video Semantic Segmentation

arXiv:2606.09162v1 Announce Type: new Abstract: Video semantic segmentation for low-altitude UAVs requires temporal consistency, yet dense optical flow introduces spatially structured noise in the pla

Zero-Shot Learning in Industrial Scenarios: New Large-Scale Benchmark, Challenges and Baseline

arXiv:2606.07965v1 Announce Type: new Abstract: Large Visual Language Models (LVLMs) have achieved remarkable success in vision tasks. However, the significant differences between industrial and natur

Zero-Shot Semantic Re-Identification for Autonomous Driving: A VLM Baseline Study

arXiv:2606.09362v1 Announce Type: new Abstract: Re-Identification (ReID) in autonomous driving is typically formulated as a visual matching problem, where observations of vehicles, pedestrians, and cy

Zero Touch Predictive Orchestration: Automating Time-Series Models for the Cloud-Edge Continuum

arXiv:2606.09787v1 Announce Type: new Abstract: The Cloud-Edge Continuum (CEC) enables latency-critical applications by distributing resources to the far edge, but its extreme volatility makes proacti

researcharxiv-cs-lg

ZIPP:Zero-shot Image Personalization from Personas

arXiv:2606.08841v1 Announce Type: new Abstract: Text-to-image diffusion models are increasingly deployed in open-ended creative contexts, yet their outputs remain impersonal, optimized for aggregate a

Zscaler launches AI Broker and Endpoint AI Security for agents

model-releasessiliconangle

Zscaler Inc. today unveiled a set of products designed to secure autonomous artificial intelligence agents, with the cybersecurity company claiming it has built the industry’s first complete zero-trus

100p! in our recent intelligence per watt (ipw) paper, @JonSaadFalcon & i find that 71.3% of real world chat and reasoning queries can be sh…

applicationsclem-delangue--x

100p! in our recent intelligence per watt (ipw) paper, @JonSaadFalcon & i find that 71.3% of real world chat and reasoning queries can be shifted from frontier lms to local lms! link to ipw paper in c

Tools

1,500 teams entered the Billion Dollar Build and have been building for 7 weeks. 8 finalists present live on June 9 to our judges @LewisHami…

1,500 teams entered the Billion Dollar Build and have been building for 7 weeks. 8 finalists present live on June 9 to our judges @LewisHamilton (7x World Champion Driver), @AravSrinivas (CEO, Perplex

toolsperplexity--x

Local Ai

[3090] Gemma4 QAT + MTP quick TPS numbers [TLDR 1.2-1.8x better]

I'd need to search for this specific Reddit post to provide accurate details about the actual findings and technical specifics of this GPU performance benchmark. This post likely discusses throughput

local-air-localllama

3DMorph: Single-Image-Guided Local 3D Shape Editing and Morphing

arXiv:2606.07115v1 Announce Type: new Abstract: Despite recent progress in 3D generation, intuitive editing of existing shapes remains limited. Unlike images, which benefit from well-established inpai

A Causal Probabilistic Framework for Perception-Informed Closed-Loop Simulation of Autonomous Driving

arXiv:2606.07186v1 Announce Type: new Abstract: Software-in-the-loop (SIL) simulation is a cornerstone for the validation of modern automotive safety functions. However, many current frameworks utiliz

safetyarxiv-cs-ro

A Comprehensive Anatomy of Human and DeepSeek-R1 LLM Mathematical Reasoning

arXiv:2606.07410v1 Announce Type: cross Abstract: The emergence of 'Aha moments' in large language models, particularly DeepSeek-R1-0120, has raised the question of whether these systems genuinely rea

A Conformation-Centric Generative Foundation Model for Linear Polymer Modeling and Design

arXiv:2510.16023v2 Announce Type: replace Abstract: Linear polymers, macromolecules formed from monomers covalently bonded into continuous chains, underpin countless technologies and are indispensable

researcharxiv-cs-lg

A Cross-view Fusion Framework for Robust 6-DoF Grasp Pose Estimation

arXiv:2606.06878v1 Announce Type: cross Abstract: In this paper, we propose a cross-view fusion framework that enhances the robustness of 6-DoF grasp pose estimation in corner views. Our framework all

A Dynamic Self-Evolving Extraction System

arXiv:2603.06915v2 Announce Type: replace Abstract: The extraction of structured information from raw text is a fundamental component of many NLP applications, including document retrieval, ranking, a

applicationsarxiv-cs-cl

Industry

A Falcon 9 booster turns 5 years old—and just set a remarkable reuse record

SpaceX has broken its rocket reuse record by launching Falcon 9 booster B1067 for its 35th time , marking a significant achievement in the company's reusability program. The Falcon 9 rocket carried 29

industryars-technica

A federal judge blocks President Trump's policy imposing a $100,000 fee for employer H-1B visa applications (CNBC)

CNBC: A federal judge blocks President Trump's policy imposing a $100,000 fee for employer H-1B visa applications — Choose CNBC as your preferred source on Google and never miss a moment from the most

safetytechmeme

A Four-Condition Diagnostic Protocol for Evidence Utilization in Long-Context and Retrieval-Augmented Language Models

model-releasesarxiv-cs-cl

arXiv:2606.06758v1 Announce Type: new Abstract: Final-answer accuracy, retrieval recall, and citation overlap do not by themselves identify whether a long-context or retrieval-augmented language model

A French engineer who lives quietly in Paris has spent 30 years writing software that the entire internet now runs on without knowing his na…

model-releasesfrancois-chollet--x

A French engineer who lives quietly in Paris has spent 30 years writing software that the entire internet now runs on without knowing his name. He wrote the code that streams every YouTube video, ever

A Geometric Account of Activation Steering through Angle-Norm Decomposition

arXiv:2606.06735v1 Announce Type: new Abstract: Linear activation steering has gained popularity as a simple and empirically effective way to control language model behavior. More recently, spherical

safetyarxiv-cs-ai