GPT-5.4

390 mudkipdev 2026-03-05 18:33:10 UTC openai.com Llama Share 370

GPT-5.4 is a new model that combines reasoning, coding, and computer-use capabilities for professional work, improving efficiency and accuracy. It is available in ChatGPT, the API, and Codex, with GPT-5.4 Pro offering maximum performance on complex tasks.

Wikipedia in read-only mode following mass admin account compromise

722 greyface- 2026-03-05 16:33:10 UTC wikimediastatus.net Full Share 231

Subscribe to updates for Wikis were in read only mode via email. You'll receive email notifications when incidents are updated.

The Brand Age

122 bigwheels 2026-03-05 18:33:10 UTC paulgraham.com Llama Share 106

The Swiss watch industry faced a crisis in the 1970s due to Japanese competition, exchange rate changes, and quartz movements, but survived by transforming into luxury brands. This shift led to a new era where branding and design conflicted, with brands prioritizing visibility over quality and design.

Hardware hotplug events on Linux, the gory details

61 todsacerdoti 2026-03-05 18:33:10 UTC arcanenibble.github.io Llama Share 1

To detect USB device plug/unplug events on Linux, libusb recommends using udev support, which is a mechanism that avoids race conditions by handling device events and rebroadcasting them to other programs. To listen to these rebroadcasted events without linking to libudev, one can use netlink sockets to receive messages from udev, which are sent with credentials and use a versioned packet ...

A GitHub Issue Title Compromised 4k Developer Machines

222 edf13 2026-03-05 16:33:10 UTC grith.ai Llama Share 51

An attacker exploited a vulnerability in Cline's GitHub issue triage workflow to install OpenClaw, an AI agent, on developers' machines without consent. The attack, named Clinejection, used a combination of prompt injection, cache poisoning, and credential theft to publish a compromised package.

Good software knows when to stop

266 ssaboum 2026-03-05 14:33:10 UTC ogirardot.writizzy.com Llama Share 149

Upgrading Linux distribution and packages caused unexpected behavior with the ls command. A lesson learned is to prioritize simplicity and focus on the product's core purpose rather than adding unnecessary features.

Structured AI (YC F25) Is Hiring

0 2026-03-05 21:02:10 UTC ycombinator.com Llama Share

As a Mechanical Design Engineer, you'll translate complex engineering logic into AI agents for professional-grade accuracy and risk mitigation. You'll partner with CTO and CPO to productize engineering intelligence and automate QA/QC checks for design engineers.

Show HN: Jido 2.0, Elixir Agent Framework

198 mikehostetler 2026-03-05 16:33:10 UTC jido.run Llama Share 41

Jido 2.0 has shipped on Hex after 18 months of development, offering a simpler and more efficient agent framework built on the BEAM. It provides a pure functional agent architecture, strategies for AI integration, and a growing ecosystem of builders and packages.

Launch HN: Vela (YC W26) – AI for complex scheduling

30 Gobhanu 2026-03-05 18:33:10 UTC news.ycombinator.com Llama Share 28

Gobhanu and Saatvik, the brothers behind Vela, built an AI agent that automates multi-party, multi-channel scheduling by reading context, checking calendars, and proposing times. Vela handles edge cases and varying user behavior across different demographics and channels, and the brothers are seeking feedback from experts in multi-agent coordination and conversational AI.

A ternary plot of citrus geneology

17 jlauf 2026-03-05 20:33:10 UTC jlauf.com Llama Share 1

Citrus fruits are hybrids of three original species, represented by a ternary plot showing their genetic distribution. Human breeding and historical movement have created overlapping recombination clusters, resulting in diverse market varieties.

Remotely unlocking an encrypted hard disk

37 janandonly 2026-03-05 19:33:10 UTC jyn.dev Llama Share 7

The user wants to access their Arch Linux system's early boot environment securely by setting up an SSH server and networking in the initramfs system, which is a small OS running from memory. They achieved this by adding systemd services to initramfs, including an SSH server and networking, and configuring Tailscale to allow access to the initramfs system.

Let's Get Physical

62 MBCook 2026-03-05 19:33:10 UTC m4iler.cloud Llama Share 10

The user participated in a physical pentest where they and their colleague blended in, bypassed security measures, and accessed restricted areas, including the director's office and server room. They successfully demonstrated vulnerabilities in the company's physical security, but were eventually caught and had a productive conversation with the security staff.

The Government Uses Targeted Advertising to Track Your Location

216 hn_acker 2026-03-05 17:33:10 UTC eff.org Llama Share 88

The government is using online ad systems to track people's locations without warrants, buying location data from shady data brokers. This is made possible by the online advertising industry's massive surveillance machine that collects and shares personal data.

Optimizing Recommendation Systems with JDK's Vector API

48 mariuz 2026-03-05 17:33:10 UTC netflixtechblog.com Llama Share 2

Netflix optimized their video serendipity scoring feature, which was consuming 7.5% of CPU, by batching and re-architecting memory layout, and then using the JDK Vector API for SIMD-style math in pure Java. This optimization resulted in a 7% drop in CPU utilization and 12% drop in average latency, allowing Netflix to handle the same traffic with 10% less CPU.

GLiNER2: Unified Schema-Based Information Extraction

5 apwheele 2026-03-05 20:33:10 UTC github.com/fastino-ai Prefix Share 0

Extract entities, classify text, parse structured data, and extract relations—all in one efficient model. GLiNER2 unifies Named Entity Recognition, Text Classification, Structured Data Extraction, and Relation Extraction into a single 205M parameter model. It provides efficient CPU-based inference without requiring complex pipelines or external API dependencies. 🎯 One Model, Four Tasks: ...

Datasets for Reconstructing Visual Perception from Brain Data

36 katsee 2026-03-05 16:33:10 UTC github.com/seelikat Llama Share 6

Researchers from AI and machine learning backgrounds need to understand neuroimaging methodology to avoid common pitfalls in reconstruction from neuroimaging data. Reconstruction requires generalizing to novel stimuli outside the training set, which is a harder problem than decoding or identification.

OpenTitan Shipping in Production

23 rayhaanj 2026-03-05 19:33:10 UTC opensource.googleblog.com Llama Share 1

OpenTitan, an open source silicon Root of Trust, is now shipping in commercially available Chromebooks, providing strong security guarantees. It's the first open source silicon RoT to support post-quantum cryptography and is maintained by lowRISC C.I.C.

Nvidia PersonaPlex 7B on Apple Silicon: Full-Duplex Speech-to-Speech in Swift

343 ipotapov 2026-03-05 08:33:10 UTC blog.ivan.digital Llama Share 112

A new speech library, qwen3-asr-swift, enables full-duplex speech-to-speech on Apple Silicon with NVIDIA's PersonaPlex 7B model. The library runs entirely on Apple Silicon, processing audio in and out with streaming via PersonaPlex 7B.

Google Workspace CLI

875 gonzalovargas 2026-03-05 00:33:10 UTC github.com/googleworkspace Llama Share 273

The Google Workspace CLI (gws) is a command-line tool that provides a unified interface to Google Workspace APIs, including Drive, Gmail, Calendar, and more, with features like tab completion and structured JSON output. It supports multiple authentication workflows and comes with over 100 pre-built agent skills for automating tasks and integrating with large language models.

Show HN: PageAgent, A GUI agent that lives inside your web app

55 simon_luv_pho 2026-03-05 17:33:10 UTC alibaba.github.io Full Share 29

PageAgent.js: Intelligent GUI Agent for any website. Modern web AI automation with minimal integration.

A man who broke into jail

85 fortran77 2026-03-05 13:33:10 UTC newyorker.com Llama Share 39

Alex Friedmann, a respected prison reform activist, was arrested for infiltrating Nashville's new central jail and planting guns in its infrastructure. He was caught after months of secretly entering the jail, wearing disguises and hiding evidence, and was later found to have hidden multiple guns and keys in various parts of the jail.

Greg Kroah-Hartman Stretches Support Periods for Key Linux LTS Kernels

46 brideoflinux 2026-03-05 15:33:10 UTC fossforce.com Llama Share 19

Linux kernel maintainer Greg Kroah-Hartman extended support lifetimes of several long-term kernels after consulting users and maintainers. This move will likely stretch to future kernel releases, reflecting real-world adoption rather than a strict schedule.

AI and the Ship of Theseus

40 pixelmonkey 2026-03-05 16:33:10 UTC lucumr.pocoo.org Llama Share 21

AI-generated code is redefining software ownership, with implications for copyleft licenses like GPL and the future of software development. The rise of AI-generated code may lead to more permissive licenses, proprietary software re-emerging as open source, and new fights over copyrights and authorship.

World-first gigabit laser link between aircraft and geostationary satellite

149 giuliomagnifico 2026-03-05 10:33:10 UTC esa.int Llama Share 57

The European Space Agency and partners successfully connected an aircraft to a geostationary satellite using laser communications, achieving error-free connections at 2.6 gigabits per second. This breakthrough enables seamless, high-speed connections for travellers and remote communities, laying the foundations for everyday technologies that depend on strong links.

Relicensing with AI-Assisted Rewrite

352 tuananh 2026-03-05 05:33:10 UTC tuananh.net Llama Share 348

The chardet maintainers relicensed the Python character encoding detector from LGPL to MIT using AI-generated code, but this move may be a GPL violation or render the MIT license moot due to copyright and licensing complexities. The case raises questions about AI-generated code ownership and the future of copyleft licensing.

Pentagon formally labels Anthropic supply-chain risk

281 klausa 2026-03-05 19:33:10 UTC wsj.com Full Share 156

Please enable JS and disable any ad blocker

Fast-Servers

86 tosh 2026-03-05 14:33:10 UTC geocar.sdf1.org Llama Share 26

A high-performance network server design uses one thread per core, pinned to separate CPUs, with each thread handling a separate state transition and using epoll/kqueue for efficient file descriptor management. The design involves simple blocking/IO calls, no decision points, and can handle up to 100k requests per second on modern systems.

Poor Man's Polaroid

179 ZacnyLos 2026-03-05 08:33:10 UTC boxart.lt Llama Share 48

The code controls a camera and printer using GPIO pins and Python libraries, capturing images and applying filters based on brightness. It also handles button presses and shutdown requests, with a cleanup function to release resources when the program exits.

Google Safe Browsing missed 84% of confirmed phishing sites

248 jdup7 2026-03-05 15:33:10 UTC norn-labs.com Llama Share 70

Huginn, a phishing discovery tool, found 254 confirmed phishing websites in February, with 83.9% not flagged by Google Safe Browsing. Muninn's automatic scan correctly identified 238 phishing sites, but a deeper scan caught every single confirmed phishing site with zero false negatives.

AMD will bring its “Ryzen AI” processors to standard desktop PCs for first time

214 Bender 2026-03-05 04:33:10 UTC arstechnica.com Llama Share 196

AMD is releasing its first Ryzen AI desktop chips with NPUs for business PCs, combining Zen 5 CPU cores and RDNA 3.5 GPU cores. These chips qualify for Microsoft's Copilot+ PC label and support device management capabilities for business PCs.