BigData Boutique blog

c0de517e's weblore

CNCF

Container Journal

Cyble

<h2>Must know</h2>
<ul>
<li><strong>Kubernetes ingress NGINX faces deprecation, security risks</strong>: The Kubernetes Ingress NGINX controller, used by roughly half of cloud-native environments, will stop receiving security patches in March 2026 due to maintainer burnout and a push for faster deprecation by Kubernetes leadership, as reported by DEVCLASS. This necessitates migration to alternatives like Gateway API, Envoy Gateway, Traefik, or Cilium, which is not a straightforward process due to custom annotations and varying feature sets. The situation highlights the ongoing challenges of open-source project sustainability and the pressure on volunteer maintainers of critical infrastructure. <a href="https://app.daily.dev/feed-by-ids?id=wK9Dxbori&amp;id=PptBbhnAu" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Linux kernel gets 1GB huge pages and OOM management via BPF</strong>: The Linux kernel is seeing significant updates, including experimental patches for 1GB PUD-level Transparent Huge Pages (THPs) that show 34% faster memory access times, though page fault allocation is slower. Additionally, Google engineer Roman Gushchin is updating patches for managing out-of-memory (OOM) behavior using eBPF programs, offering a new, programmable approach to memory exhaustion. These developments aim to improve performance and stability, but the 1GB THP patches are still under scrutiny from upstream maintainers due to concerns about their approach. <a href="https://app.daily.dev/feed-by-ids?id=NT39y376C&amp;id=QVssY3RTX&amp;id=tPbz1g571" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Upwind security raises $250M for runtime cloud security</strong>: Upwind Security secured $250 million in Series B funding, valuing the company at $1.5 billion, to expand its ‘runtime’ cloud security platform. The platform uses an ‘inside-out’ approach, analyzing internal signals like network requests and API traffic to prioritize real-time threats, a contrast to traditional agentless scanning. Despite initial hesitation from security teams to deploy internal agents, Upwind has achieved 900% year-over-year revenue growth and doubled its customer base, including Siemens and Roku, according to TechCrunch. <a href="https://app.daily.dev/posts/LGgCBPA2D" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Good to know</h2>
<ul>
<li><strong>NVIDIA cosmos policy advances robot control with world models</strong>: NVIDIA introduced Cosmos Policy, a robot control system fine-tuned on the Cosmos Predict-2 world foundation model, achieving high success rates on benchmarks by encoding robot actions and states as latent frames using diffusion processes. <a href="https://app.daily.dev/posts/kQ1bPWpKA" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google Cloud Run now supports NVIDIA RTX 6000 GPUs</strong>: Google Cloud Run now offers preview support for NVIDIA RTX PRO 6000 Blackwell GPUs, enabling serverless deployment of large AI models with 96GB vGPU memory and rapid scaling for generative AI inference and model fine-tuning. <a href="https://app.daily.dev/posts/Vh4SzEb3w" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>China’s open-source AI ecosystem shifts to MoE and hardware optimization</strong>: China’s open-source AI ecosystem is increasingly adopting Mixture-of-Experts (MoE) architectures and optimizing models for domestic chips like Huawei Ascend, with a focus on smaller models (0.5B-30B parameters) and Apache 2.0 licensing. <a href="https://app.daily.dev/posts/UWmmhD98d" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI agents for CI/CD pipelines introduce new security risks</strong>: AI agents are being developed to secure CI/CD pipelines through threat modeling and self-healing, but they introduce new risks like indirect prompt injection and excessive privileges, necessitating sandboxed runners and human-in-the-loop workflows. <a href="https://app.daily.dev/posts/ERjZhsaaZ" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Rust and WebAssembly adoption grows for performance-critical tasks</strong>: Rust and WebAssembly are increasingly used together, with Rust handling performance-critical logic within JavaScript/TypeScript applications, as seen in Figma’s graphics rendering and Cloudflare’s edge computing. <a href="https://app.daily.dev/feed-by-ids?id=UAEJhL6mu&amp;id=2VsBKezZr" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>New Linux malware shadowHS operates filelessly in memory</strong>: Cyble discovered ShadowHS, a sophisticated fileless Linux post-exploitation framework that runs entirely in memory, using multi-stage encrypted loaders and aggressively fingerprinting EDR/AV solutions to evade detection. <a href="https://app.daily.dev/posts/7kO021ZBt" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Google Cloud’s Axion CPUs now generally available</strong>: Google Cloud’s N4A instances, powered by new Axion ARM processors (Neoverse N3 cores), are now generally available, offering up to 64 vCPUs and 512GB RAM at a cost-effective price point for web servers and containerized applications. <a href="https://app.daily.dev/posts/JvBjVyZ5U" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
<h2>Bullshit police</h2>
<ul>
<li><strong>AI-coded tools for AI code review: a self-referential loop of ‘innovation’</strong>: GitHuman, a new local-first tool for reviewing AI-generated code, was apparently written entirely by Claude Code, an AI model. And get this, every commit was human-reviewed using GitHuman itself. So, we’re now using AI to write code, and then using AI-written tools to review that AI-written code. What could possibly go wrong? It’s not a bug, it’s a feature, right? <a href="https://app.daily.dev/posts/M78k9F0lW" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>Tesla’s robotaxi crash rates: human drivers still winning, barely</strong>: Tesla’s robotaxi fleet in Austin managed to crash 9 times over 500,000 miles, which, if you’re keeping score, is about three times worse than human drivers for all crashes and nine times worse for police-reported incidents. And they redact all crash narratives in NHTSA reports. Meanwhile, Waymo, with truly driverless vehicles, has a better safety record over 25+ million miles. But hey, at least Tesla has safety monitors in every car, so it’s not <em>really</em> driverless, is it? <a href="https://app.daily.dev/posts/M8WkrOIKM" target="_blank" rel="noopener nofollow">Read more</a></li>
<li><strong>AI coding agents: the productivity myth that keeps on giving</strong>: After two years of using AI coding agents, one author found the AI-generated code created ‘slop’ and lacked respect for overall codebase structure, leading to unmaintainable messes. They’ve since returned to manual coding, finding it faster and more productive in the long run. But don’t worry, the claims of 10x-100x productivity gains from AI coding assistants are still out there, completely lacking any research backing. Because who needs actual data when you have hype? <a href="https://app.daily.dev/feed-by-ids?id=HCRUybI5Q&amp;id=R5VZVKE0f&amp;id=Rgiczq9Do" target="_blank" rel="noopener nofollow">Read more</a></li>
</ul>
