Cybersecurity Hackaday

This Week in Security: Escaping Linux VMs, Vulnerable Solar, Confusing AI (Again), and Confusing NPM Malware

A roundup of security vulnerabilities, including the Januscape KVM escape and NPM malware.

Software Engineering Hacker News

The footgun of right-to-left decorative characters

A discussion on the security and usability risks associated with right-to-left decorative characters in digital text.

Other Hacker News

Silent speech with ultrasound

An exploration of using ultrasound for silent speech communication.

AI/ML arXiv cs.AI

Accurate Portraits of Scientific Resources and Knowledge Service Components

Research on creating accurate profiles of scientific resources using knowledge graphs, text representation learning, and entity extraction.

AI/ML arXiv cs.AI

Knowledge Graph and Accurate Portrait Construction of Scientific and Technological Academic Conferences

A review of technologies for building knowledge graphs and portraits for academic conferences to help researchers find information more efficiently.

AI/ML arXiv cs.AI

Retrieval of Scientific and Technological Resources for Experts and Scholars

Research on building expert and scholar databases using text relation extraction and vector retrieval to improve matching between talent and social needs.

AI/ML arXiv cs.AI

The Contribution of XAI for the Safe Development and Certification of AI: An Expert-Based Analysis

An expert-based analysis of the role of eXplainable AI (XAI) in the safe development and certification of AI systems, highlighting both its utility and limitations.

AI/ML arXiv cs.AI

ParamMute: Suppressing Knowledge-Critical FFNs for Faithful Retrieval-Augmented Generation

Introduction of ParamMute, a framework that improves RAG faithfulness by suppressing specific feed-forward networks (FFNs) that cause unfaithful generation.

AI/ML arXiv cs.AI

Simulator Ensembles for Trustworthy Autonomous Driving Systems Testing

Introduction of MultiSim, an ensemble-based simulator approach for identifying generalizable, simulator-agnostic failures in autonomous driving systems.

AI/ML arXiv cs.AI

Concept-as-Tree: A Controllable Synthetic Data Framework Makes Stronger Personalized VLMs

Introduction of the Concept-as-Tree (CaT) framework for generating controllable synthetic data to improve the personalization of Vision-Language Models (VLMs).

AI/ML arXiv cs.AI

ToDMA: Large Model-Driven Massive Token Communications for Semantic Multiple Access

Proposal of ToDMA, a semantic multiple access scheme using large models to enable massive token communications with reduced latency and improved recovery.

Open Source Hacker News

FreeCAD in the Browser

FreeCAD, a popular open-source 3D parametric modeler, is now being made available to run directly in the web browser.

Tech Business/VC Hacker News

Meta pulls new AI image feature after days of backlash

Meta has removed a newly introduced AI image feature following significant user backlash.

Homelab/Self-Hosting Hacker News

An iroh powered smart fan

A developer has created a smart fan powered by Iroh, demonstrating the network's capabilities for decentralized device communication.

AI/ML Hacker News

Choosing the Right AI Agent Memory Strategy: A Decision-Tree Approach

A guide proposing a decision-tree approach for selecting the most effective memory strategy for AI agents.

Cybersecurity TechCrunch

US cyber agency CISA had to build its incident playbook during the incident, agency reveals

CISA admits it failed to prepare a security incident playbook before a major event, necessitating on-the-fly development during the crisis.

AI/ML arXiv cs.AI

Retrieval-Augmented Generation Must Move Beyond Factual Grounding to Represent Diverse Opinions

Researchers propose O-RAG, an Opinion-Aware RAG architecture designed to represent diverse opinions rather than just factual grounding.

AI/ML arXiv cs.AI

The Power of Power Law: Asymmetry Enables Compositional Reasoning

Research indicates that power-law data distributions, rather than uniform ones, actually improve a model's ability to perform compositional reasoning.

AI/ML arXiv cs.AI

SHARP: Sleep-based Hierarchical Accelerated Replay for Long Range Non-Stationary Temporal Pattern Recognition

SHARP is a new framework that uses a 'sleep-based' replay mechanism to improve long-range temporal pattern recognition in streaming data.

AI/ML arXiv cs.AI

Deployment-Time Memorization in Foundation-Model Agents

Study examines how memory design choices in foundation-model agents affect personalization utility versus the risk of adversarial data extraction.