doorss

METR finds that many AI-generated pull requests that technically pass the SWE-bench benchmark would not actually be accepted by human maintainers. This raises important questions about whether current AI coding benchmarks truly measure useful software engineering ability.

The Verge explores the growing trend of AI-conducted job interviews, highlighting the dehumanizing and often surreal experience of being evaluated by a bot. The piece raises concerns about AI's expanding role in hiring decisions and its impact on job seekers.

A rigorous critique arguing that evaluating generative AI models based on 'vibes' is insufficient, proposing clearer criteria for when these models actually provide value. The post pushes back against hype-driven adoption of AI tools in software development.

A thoughtful analogy comparing selective, principled use of generative AI to vegetarianism—arguing you don't have to go fully AI-free, but being intentional about when and how you use it matters. The post explores the ethical and practical dimensions of AI adoption.

Security researchers detail how they found serious vulnerabilities in McKinsey's AI platform, exposing risks in enterprise AI deployments. The writeup serves as a cautionary tale about the security surface area introduced by rushing AI products to market.

Hackaday reports on the discovery of a covert secure communication channel hidden within a seemingly innocuous news application. The technique highlights creative approaches to steganographic messaging and raises privacy and surveillance questions.

Krebs on Security reports that an Iran-backed hacking group claims responsibility for a destructive wiper attack on medical technology giant Stryker. The attack underscores the growing threat of state-sponsored cyberattacks on critical healthcare infrastructure.

An insightful breakdown of how WireGuard functions both as a VPN protocol and a kernel network interface, and why understanding this distinction matters. The article clarifies common misconceptions about WireGuard's architecture and security model.

Microsoft's BitNet demonstrates a 100-billion-parameter language model using 1-bit weights that can run on local CPUs without GPUs. This represents a significant step toward democratizing access to large language models by dramatically reducing hardware requirements.

A new open-source tool that adds context-aware permission controls to Claude Code, letting users fine-tune what the AI agent can and cannot do. This addresses growing concerns about giving AI coding assistants too much unsupervised access to codebases and systems.

Bloomberg engineers chronicle the nearly decade-long effort to bring the Temporal API to JavaScript, finally replacing the notoriously broken Date object. The post details the technical and political challenges of landing a major new API in a language used by billions.

Mozilla outlines efforts to elevate WebAssembly from a compilation target to a full first-class web language with improved debugging, DOM access, and developer tooling. This could significantly expand what's practical to build with Wasm on the web.

The SQLite team discloses a WAL-reset bug that can cause database corruption under specific conditions. Given SQLite's ubiquity across billions of devices, this is a critical finding with wide-reaching implications.

Hacker News has formally updated its guidelines to prohibit AI-generated or AI-edited comments, reaffirming that the forum is meant for authentic human conversation. The policy reflects growing concern about AI-written content degrading the quality of online discourse.

A PNAS paper documents how organized entities are enabling scientific fraud at industrial scale, including paper mills and fake peer review networks. The research highlights a systemic threat to the integrity of scientific publishing.

Google has officially completed its acquisition of cloud security startup Wiz, marking one of the largest cybersecurity acquisitions in recent memory. The deal significantly bolsters Google Cloud's security offerings as competition with AWS and Azure intensifies.

Gruber reviews Apple's latest MacBook Neo, analyzing its design philosophy and positioning within Apple's laptop lineup. The piece examines how Apple continues to iterate on its hardware strategy in the post-M-series chip era.