AI/ML arXiv cs.AI

Gauging, Measuring, and Controlling Critic Complexity in Actor-Critic Reinforcement Learning

Introduces a method to measure and control 'critic complexity' in actor-critic reinforcement learning using spectral effective-rank entropy.

AI/ML arXiv cs.AI

A Multi-Resolution Finite-Volume Inspired Deep Learning Framework for Spatiotemporal Dynamics Prediction

Introduces MuRFiV, a deep learning framework that combines finite-volume methods with neural networks for stable long-term spatiotemporal dynamics prediction.

AI/ML arXiv cs.AI

Predicting Lethal Outcome (Cause) And Understanding Key Biomarkers Linked With Acute Myocardial Infarction Using Deep Artificial Neural Network And Ensemble Of Machine Learning Methodologies

Develops an automated machine learning ensemble to predict lethal outcomes and identify biomarkers associated with acute myocardial infarction.

Cybersecurity arXiv cs.AI

Beyond the Prompt: Jailbreaking Function-Calling LLMs via Simulated Moderation Traces

Presents SMT, a black-box attack framework that jailbreaks function-calling LLMs by simulating moderation traces, exposing vulnerabilities in stateful tool-enabled environments.

AI/ML arXiv cs.AI

PAPA: Online Personalized Active Preference Alignment

Introduces PAPA, a method for online personalized preference alignment in diffusion models using real-time user feedback without a parameterized reward model.

AI/ML arXiv cs.AI

MindEdit-Bench: Benchmarking Object-Level Counterfactual Spatial Reasoning in VLMs from In-the-Wild Photos

Introduces MindEdit-Bench, a benchmark to evaluate object-level counterfactual spatial reasoning in Vision-Language Models using in-the-wild photos.

AI/ML arXiv cs.AI

BaseRT: Best-in-Class LLM Inference on Apple Silicon via Native Metal

Presents BaseRT, a native Metal inference runtime for LLMs on Apple Silicon that achieves best-in-class throughput by optimizing for unified memory and kernel fusion.

AI/ML arXiv cs.AI

Cross4D-JEPA: Dense Cross-modal Correspondence Distillation for 4D Point Cloud Representation Learning

Proposes Cross4D-JEPA, a method for learning 4D point cloud representations by distilling dense cross-modal correspondences from 2D and video foundation models.

Tech Business/VC Hacker News

Meta building cloud business to sell excess AI capacity

Meta is reportedly planning to establish a cloud business to monetize its excess AI compute capacity.

Tech Business/VC Hacker News

Nvidia offers startup customers chance to swap compute power for revenue share

Nvidia is offering startup customers the option to exchange compute power for a share of their future revenue.

Software Engineering The Verge

BitTorrent’s disastrous, legendary, and controversial story

A retrospective on the 25-year history of BitTorrent and its disruptive impact on file sharing and the entertainment industry.

Other Ars Technica

Editorial: It's time to step up and have your say for science

An editorial urging public participation in science policy to prevent political interference in scientific research.

AI/ML arXiv cs.AI

Holographic Quantum Transformer: A Generalist Neuro-Symbolic Architecture for Solving Frustrated Systems via Generative Attention

Introduction of the Holographic Quantum Transformer (HQT), a neuro-symbolic architecture for simulating frustrated quantum systems with zero-shot size extrapolation.

AI/ML arXiv cs.AI

The Illusion of High Utility in Safety Alignment of Text-to-Image Diffusion Models

Researchers identify a 'utility illusion' in safety-aligned T2I models and propose SAGE, a regularization method to preserve semantic fidelity.

AI/ML arXiv cs.AI

EO-VGGT: Orbital Ray-Conditioned 3D Foundation Models for Satellite Multi-View Reconstruction

EO-VGGT is a new framework for satellite multi-view 3D reconstruction that integrates physical orbital geometry into frozen perspective-driven models.

AI/ML arXiv cs.AI

Learning Gait-Aware Quadruped Locomotion with Temporal Logic Specifications

A framework for quadruped locomotion using Signal Temporal Logic (STL) to define gait constraints, implemented on Google's Barkour robot.

AI/ML arXiv cs.AI

Search-Based Spatiotemporal and Multi-Robot Motion Planning on Graphs of Space-Time Convex Sets

A new algorithmic framework using graphs of space-time convex sets (ST-GCSs) for efficient multi-robot motion planning in dynamic environments.

AI/ML arXiv cs.AI

VideoSearch-R1: Iterative Video Retrieval and Reasoning via Soft Query Refinement

VideoSearch-R1 is an agentic framework that uses Soft Query Refinement (SQR) and GRPO for iterative video retrieval and reasoning.

Homelab/Self-Hosting Hacker News

Creating a Personalised Bin Calendar

A discussion on creating a personalized bin calendar, likely focusing on automation or scheduling.

Tech Business/VC The Verge

OpenAI floats giving Trump administration 5 percent cut of AI boom 

OpenAI is reportedly considering offering the US government a 5% ownership stake to mitigate political tensions and public backlash.