In March 2026, AI-generated code vulnerabilities surged past 2025's total, compelling Dawnguard—a startup co-founded by former security leads at major tech companies—to address this escalating issue. With $3.3 million in new funding, Dawnguard aims to enhance cybersecurity by bridging gaps in current practices.

Salt is a novel programming language designed for high performance and safety. It utilizes Z3 for contract verification at compile time, avoiding runtime costs. With features like arena-based memory management and seamless integration for various applications, Salt achieves performance metrics on par with C, proving its efficiency and reliability across multiple benchmarks.

Reinforcement learning (RL) is becoming essential for developing specialized AI agents in various workflows. This guide explores practical RL training techniques, including strategies for defining success and improving model behavior. It outlines when to use different approaches and provides a workflow for implementing RL effectively in agent development.

The ProbeLab team's Optimistic Provide optimization significantly enhances IPFS content publishing speed, reducing upload latency from over 13 seconds to under 1 second. This improvement allows publishers to operate almost in real-time, streamlining processes for developers and users in the evolving decentralized network environment.

OOMWOO is a build-it-yourself robot vacuum project designed for makers. It's fully open-source, with a focus on local control, no cloud dependency, and modular design. Users can follow along as the project progresses, sourcing parts and contributing to software development from the start.

A new strain of malware disguised as the "Android Developer Verifier" is infecting billions of Android devices, risking user autonomy by blocking unapproved software. This controversial Google initiative has sparked widespread opposition, raising concerns about its implications for open software development and user control in the Android ecosystem.

A recent study reveals that while experienced developers feel they work faster with AI tools, they are actually performing about 19% slower in reality. This gap suggests that reliance on perceived speed could be misleading, highlighting the importance of measuring actual productivity rather than subjective feelings.

Elon Musk envisions a future where AI and data centers operate in orbit, potentially revolutionizing space computing. However, logistical, technical, and economic challenges suggest this ambitious plan may take years to realize, raising questions about practicality, costs, and environmental impact.

Vite+ has entered beta, offering a streamlined toolchain for web development by unifying essential tools like Vite and Vitest. With consistent commands and built-in features, it aims to enhance developer experience and simplify project setups. Vite+ is open-source, framework-agnostic, and encourages community feedback for future improvements.

This article explores the complexities of managing AI agents through a detailed experiment involving 11 language models. It addresses the challenges of a "god node" in code design, comparing proposals and evaluations from American and Chinese models to uncover the best strategies for code reorganization.

ZeroFS presents a log-structured filesystem for S3-compatible buckets, enabling access through NFS, 9P, and NBD protocols. It features immutable segments for efficient storage, built-in encryption and compression, and robust caching mechanisms. The design emphasizes compatibility, performance, and straightforward testing across various operating systems.

Flow is a programming language designed to enhance concurrency in C++11 by introducing actor-based messaging and asynchronous processes. It effectively balances high performance, development productivity, and testing support through tools like futures and promises, streamlining the process of building reliable systems such as FoundationDB.

This study evaluates the reliability of Vision-Language Models (VLMs) for assessing medical image quality, particularly under conditions of image corruption and bias. Findings reveal significant performance drops with pixelation and highlight the influence of contextual metadata on scoring, raising concerns about objectivity and privacy in clinical settings.

OntoLearner is a new Python library designed for ontology learning using large language models. It provides a unified framework for accessing ontologies, learning pipelines, and benchmarking across multiple domains. The study highlights challenges in ontology complexity, offering insights for effective knowledge model construction.

This article presents Task-Agnostic Pretraining (TAP) for Vision-Language-Action (VLA) models, addressing the challenge of limited expert data. TAP improves motor skills through self-supervised learning before incorporating language, resulting in enhanced performance with less labeled data and greater resilience in real-world applications.