Cybersecurity arXiv cs.AI

Execution-Grounded Security Testing for Coding Agents in Software Engineering Pipelines

A new red-team testing framework for coding agents that uses execution-grounded evidence to identify security vulnerabilities in software engineering pipelines.

AI/ML arXiv cs.AI

Reference Feature Atlases for Mechanistic Auditing of Language Models

Introduction of 'reference feature atlases', a sparse feature library that enables more efficient and stable auditing of internal features across different language models.

AI/ML arXiv cs.AI

SCAIR: Schema-Conditioned Agentic Iterative Reasoning for Enterprise Knowledge Graphs

SCAIR is presented as a training-free framework that improves KG-RAG performance in enterprise settings by incorporating schema-conditioned structural priors.

AI/ML arXiv cs.AI

Schema-Aware Localisation (SAL): Live Schema Grounding and Hallucination Validation for Oracle NL2SQL

SAL is a lightweight middleware for Oracle NL2SQL that reduces hallucinated columns and aliases by using live schema grounding and a validation index.

Other arXiv cs.AI

PhononBench-MP40: a spectrum-resolved benchmark dataset for phonon stability

Release of PhononBench-MP40, a spectrum-resolved benchmark dataset for phonon stability in computational materials screening.

Tech Business/VC TechCrunch

eBay reaches $56M settlement with e-commerce newsletter writers it terrorized in 2019

eBay has reached a $56 million settlement with e-commerce newsletter writers who were victims of a cyberstalking campaign in 2019.

Other TechCrunch

These App Store hidden gems prove there’s still room for great software in the AI era

An exploration of new App Store applications that demonstrate the continued viability of traditional software development in the age of AI agents.

Other The Verge

New Yorkers will soon need to verify their age to access algorithmic feeds

New York City is implementing rules requiring age verification for users to access algorithmic feeds on social media platforms to protect minors.

AI/ML arXiv cs.AI

MIITA: Memory-Induced Inference-Time Adaptation for Continual Learning with Small Language Models

The MIITA framework introduces memory-induced inference-time adaptation for small language models to enable continual learning without catastrophic forgetting.

AI/ML arXiv cs.AI

Codifying the Judge: Scalable Evaluation via Program Distillation

The PAJAMA system uses program distillation to create programmatic judges for scalable and transparent LLM evaluation, reducing costs and latency.

AI/ML arXiv cs.AI

SF-AMS: Strategic Forgetting for Structured Memory in LLM Agent

SF-AMS is a framework for LLM agents that uses strategic forgetting and utility-driven memory management to improve long-context reasoning.

AI/ML arXiv cs.AI

Synthetic Scenario Generation for Evaluation of Industry 4.0 Agents

A new pipeline, ScenarioGeneratorAgent, allows for the synthetic generation of industry-standard scenarios to evaluate agents in Industry 4.0 environments.

AI/ML arXiv cs.AI

Loss-Aware Feature-Map Pruning in Convolutional Neural Networks Using Multi-Armed Bandits

A study on using multi-armed bandits to prune redundant feature maps in CNNs, maintaining accuracy while reducing computational cost.

AI/ML arXiv cs.AI

DSTFView: Multi-View Cloud-Edge Workload Forecasting with Dual-Input Spatio-Temporal-Frequency Modeling

DSTFView is a multi-view workload forecasting framework designed for collaborative cloud-edge environments to improve AI inference efficiency.

AI/ML arXiv cs.AI

MedLoCoMo: A Long-Context Multi-Session Medical Dialogue Benchmark for Large Language Models

MedLoCoMo is a new long-context multi-session medical dialogue benchmark for evaluating LLMs' ability to perform longitudinal clinical reasoning.

Other Hacker News

iPhone Upgrade Program

Discussion regarding the iPhone Upgrade Program on Hacker News.

Hardware/Chips Hacker News

Using a Gaming PC's RTX 5070 from a separate Linux workstation

A discussion on using a gaming PC's RTX 5070 GPU from a separate Linux workstation.

Other TechCrunch

HBO Max embraces vertical video with a new ‘Shorts’ feed

HBO Max is introducing a 'Shorts' feed featuring vertical video to improve content discovery and align with short-form content trends.

Tech Business/VC TechCrunch

Saudi prince buys 5% stake in Lucid Motors

A Saudi prince has acquired a 5% stake in Lucid Motors amid speculation about the company going private.

Other The Verge

Tile’s best Bluetooth tracker is down to its lowest price of the year

The Tile Pro Bluetooth tracker is currently available at its lowest price of the year.