I verified my LinkedIn identity. Here's what I handed over

The user verified their identity on LinkedIn through Persona, a US company, and found their passport data was shared with 17 US-based subprocessors, including AI companies. The user recommends requesting data deletion and objecting to data use under "legitimate interests" due to concerns over data protection and US law enforcement access.

What not to write on your security clearance form (1988)

A 12-year-old boy's jargon code was mistaken for a Japanese spy's and led to an FBI investigation, costing thousands of dollars. The boy's glasses case was found and the code key was kept by the FBI, compromising his secret communication scheme.

Why is Claude an Electron app?

https://www.dbreunig.com/img/sf_beach_og.jpg
Claude's desktop app uses Electron despite its downsides due to its benefits of one codebase for multiple platforms. However, coding agents struggle with the last 10% of development and increased support for native apps.

Andrej Karpathy talks about "Claws"

Andrej Karpathy discusses Claws, a new layer on top of LLM agents, enhancing orchestration and task management. Claws are AI agent systems running on personal hardware, communicating via messaging protocols.

Claws are now a new layer on top of LLM agents

Something went wrong, but don’t fret — let’s give it another shot. Some privacy related extensions may cause issues on x.com. Please disable them and try again.

AI uBlock Blacklist

https://opengraph.githubassets.com/e428dcde4b3bc2a3b2323d18ae1ca97083a364af87ad9219a2f1964fb2214a3b/alvi-se/ai-ublock-blacklist
The user maintains a list of websites that generate low-quality content using AI, which can be misleading or even dangerous, and invites others to contribute to the list. The list aims to block access to AI-generated content farms that prioritize SEO and monetization over providing useful information.

Acme Weather

https://acmeweather.com/img/intro/timeline-high-spread.png
Acme Weather addresses forecast uncertainty by providing multiple predictions and community reports to help users plan their day. The app uses various data sources and allows users to submit real-time reports of current conditions near them.

EU mandates replaceable batteries by 2027 (2023)

https://environment.ec.europa.eu/sites/default/files/styles/oe_theme_medium_no_crop/public/2023-08/Sustainable%20batteries%20image.jpeg?itok=hJooM7L7
A new European law requires batteries to have low carbon footprints, minimal harmful substances, and be collected, reused, and recycled. The law aims to support a circular economy, increase raw material security, and enhance EU autonomy by 2050.

What Is OAuth?

https://leaflet.pub/p/did%253Aplc%253A3vdrgzr2zybocs45yfhcr6ur/3mfd2oxx5v22b/opengraph-image?4c8fe174a4beabea
OAuth was designed to allow users to grant access to their accounts without sharing passwords, driven by the need for Twitter to support OpenID in 2006. The core idea of OAuth is simple: send a secret to a trusted delegate with user consent, and the delegate can make subsequent requests on the user's behalf.

Personal Statement of a CIA Analyst

The author went through multiple polygraph tests throughout their career, initially failing one due to a minor white lie about their weight, but later passing with the help of an experienced examiner. The author became increasingly cynical about the polygraph's effectiveness and eventually stopped fearing the examiners.

macOS's Little-Known Command-Line Sandboxing Tool (2025)

https://bear-images.sfo2.cdn.digitaloceanspaces.com/herman-1683556668-0.png
Sandbox-exec is a built-in macOS command-line utility that enables users to execute applications within a secure, isolated environment with limited access to system resources. It provides a powerful tool for security-conscious users and developers to create finely-tuned security environments with custom profiles tailored to specific applications and use cases.

Show HN: Llama 3.1 70B on a single RTX 3090 via NVMe-to-GPU bypassing the CPU

https://opengraph.githubassets.com/fa0dc4f65fc018e619645e75cb45d6a08162fa87cb09134820de7f0acc7a5071/xaskasdf/ntransformer
A high-efficiency C++/CUDA LLM inference engine achieves 33x speedup on consumer hardware by streaming model layers through GPU memory. The engine uses adaptive caching and optional NVMe direct I/O to bypass the CPU and achieve high performance.

CXMT has been offering DDR4 chips at about half the prevailing market rate

https://wimg.heraldcorp.com/news/cms/2026/02/20/news-p.v1.20260220.1686b9f4331c4661b835adf082563d1a_P1.png
Chinese DRAM maker CXMT is flooding the market with cheaper chips, forcing Korean rivals Samsung and SK hynix to reevaluate their strategy. CXMT's aggressive pricing is gaining traction, with US and Taiwanese firms testing its DRAM and seeking cooperation.

EDuke32 – Duke Nukem 3D (Open-Source)

EDuke32 is a free, open-source game engine and source port of Duke Nukem 3D with thousands of features and upgrades. It's the most actively developed and maintained Duke3D port, running on Windows, Linux, and more.

Cloudflare outage on February 20, 2026

https://cf-assets.www.cloudflare.com/zkvhlag99gkb/9f2k63fiixI2YXDgsnbGq/4d156aa83c47ff6cecf3bd5c747126ef/unnamed__1_.png
Cloudflare experienced a 6-hour service outage on February 20, 2026, due to a change in how the network manages IP addresses onboarded through the Bring Your Own IP (BYOIP) pipeline, unintentionally withdrawing customer prefixes. The incident was caused by a software bug in the Addressing API and was exacerbated by incomplete testing and a lack of controlled rollouts, but Cloudflare is ...

Parse, Don't Validate and Type-Driven Design in Rust

https://www.harudagondi.space/uploads/parse-dont-validate/tingey-injury-law-firm-veNb0DDegzE-unsplash.jpg
The user discusses the concept of encoding invariants in Rust's type system to avoid runtime errors and improve code robustness, using examples like NonZeroF32 and NonEmptyVec to demonstrate this approach. By using newtypes to represent valid states, the user argues that validation can be done at compile-time, reducing the risk of runtime errors and making code more maintainable and resilient ...

Cord: Coordinating Trees of AI Agents

https://www.june.kim/assets/default-icon.png
Current AI models are good at planning and breaking down tasks into subproblems, but existing frameworks require developers to predefine the coordination structure. A new framework, Cord, uses a protocol with five primitives and allows agents to understand and use coordination tools they've never seen before, enabling them to decompose tasks and parallelize work on their own.

Meta Deployed AI and It Is Killing Our Agency

https://mojodojo.io/wp-content/uploads/2023/10/Untitled-2.png
A high-value Meta ad customer is experiencing repeated account bans for its employees, despite following standard professional practices and having a long history of advertising on the platform. The customer is seeking a manual onboarding pathway for verified agencies and a human-accessible support channel to resolve the issue.

zclaw: personal AI assistant in under 888 KB, running on an ESP32

https://raw.githubusercontent.com/tnm/zclaw/main/docs/images/lobster_xiao_cropped_left.png
zclaw is a C-based firmware for ESP32 boards with a 888 KB budget, supporting tasks, GPIO, and secure flash storage. It's tested on ESP32-C3, S3, and C6, with other variants likely compatible.

LibreOffice blasts OnlyOffice for working with Microsoft to lock users in

The Document Foundation accuses OnlyOffice of being "fake open-source" for partnering with Microsoft and defaulting to proprietary file formats. This move is seen as a replay of Microsoft's past attempts to enforce vendor lock-in through proprietary standards.

Toyota Mirai hydrogen car depreciation: 65% value loss in a year

https://static0.carbuzzimages.com/wordpress/wp-content/uploads/2025/06/2021-2025-toyota-mirai14.jpg?q=49&fit=crop&w=360&h=202&dpr=2
The Toyota Mirai's rapid depreciation is due to limited hydrogen infrastructure and the rise of battery-electric vehicles. Its value has dropped dramatically due to geographical limitations and maintenance complexities.

24 Hour Fitness won't let you unsubscribe from marketing spam, so I fixed it

https://ahmedkaddoura.com/images/24hf/spanish-error.jpeg
24 Hour Fitness has a broken unsubscribe page that links to a Spanish error message. They were informed in November 2025 but have not fixed it, violating the CAN-SPAM Act.

Over 80% of 16 to 24-year-olds would vote to rejoin the EU

Evidence of the bouba-kiki effect in naïve baby chicks

Coccinelle: Source-to-source transformation tool

https://opengraph.githubassets.com/403df72a8ed1ae124458ba369ff510d69328b5b05271ea975e5fc377fa28a526/coccinelle/coccinelle
Coccinelle is a tool for writing source-to-source transformations on C code, allowing for refactorings and more. It can be installed from source or downloaded pre-compiled, with instructions found at https://coccinelle.gitlabpages.inria.fr/website/download.html.

Canvas_ity: A tiny, single-header <canvas>-like 2D rasterizer for C++

https://raw.githubusercontent.com/a-e-k/canvas_ity/main/.github/icon.png
This C++ library, called canvas_ity, is a single-header implementation of the W3C HTML5 2D canvas specification, providing high-quality rendering and ease of use with a compact size. It supports various features like path building, fills, strokes, gradients, and compositing operations, and is designed to be widely portable and easy to understand.

Inputlag.science – Repository of knowledge about input lag in gaming

https://inputlag.science/assets/logo.png
Input lag in gaming systems is a growing issue due to increased complexity and developers' lack of understanding of its causes. This website aims to help developers and consumers understand and tackle latency issues.

Show HN: Iron-Wolf – Wolfenstein 3D source port in Rust

https://camo.githubusercontent.com/04504e7ef451663eacba5e31cc4a2f83f68983d5df21d05a7fecdf0ed2fe3078/68747470733a2f2f776f6c662e69726f6e6d756c652e6465762f67682f7765625f707265766965772e706e67
The goal is to recreate Wolfenstein 3D in Rust, specifically the shareware version, with mod-friendliness in mind. A web version is available at https://wolf.ironmule.dev, and a local version can be run with a config file for optional settings.

How an inference provider can prove they're not serving a quantized model

https://tinfoil.sh/blog-images/2026-02-03-proving-model-identity.jpg
Modelwrap is a system that guarantees a specific model is being used by verifying its weights through a Merkle tree and dm-verity, a Linux kernel subsystem. This approach ensures the model weights are not tampered with and provides a strong guarantee that the same model is used every time, even in a cloud environment.

Password managers less secure than promised

https://ethz.ch/en/news-and-events/eth-news/news/2026/02/password-managers-less-secure-than-promised/_jcr_content/articleLeadImage/image.imageformat.carousel.1156498316.jpg
Researchers from ETH Zurich discovered serious security vulnerabilities in three popular cloud-based password managers, allowing hackers to view and change stored passwords. The team found 25 attacks on Bitwarden, LastPass, and Dashlane, compromising millions of users' sensitive data.

Trunk Based Development

https://trunkbaseddevelopment.com/trunk1b.png
Trunk-Based Development is a model where developers collaborate on a single branch called 'trunk', avoiding long-lived branches and merge hell. It enables Continuous Integration and Delivery, allowing teams to release code on demand with a high-throughput and quality.

DialUp95 – A 90s inspired nostalgia hit

https://dialup95.com/dial1.png
Server type: PPP: Windows 95, Windows NT 3.5, Internet

Excessive token usage in Claude Code

https://opengraph.githubassets.com/6137cf198e1522c947909fdda844f94af24c315a46afbb1745150dda0097c76c/anthropics/claude-code/issues/16856
User experienced sudden and excessive usage of CC 2.1.1, hitting 5h limits much faster than usual. This issue started after updating to CC 2.1.1 and is affecting Opus but not Heiku.

The Nekonomicon – Nekochan.net Archive, Updated

http://nekonomicon.irixnet.org/static/nekonomicon-c.png
The collected works of the Nekochan forums.

The Software Development Lifecycle Is Dead

https://boristane.com/assets/blog/the-software-development-lifecycle-is-dead/og.png
The software development lifecycle (SDLC) is collapsing due to AI agents that can generate code, tests, and deploy it simultaneously, eliminating traditional stages like design, testing, and code review. The new skill is context engineering, and the new safety net is observability, which feeds directly back into the agent loop to ensure faster and safer shipping.

Colorado proposal moves age checks from websites to operating systems

https://d1sr9z1pdl3mb7.cloudfront.net/wp-content/uploads/2025/02/14100643/teens-phones-1024x683.jpg
Colorado lawmakers are revisiting online age verification, focusing on how to protect children without burdening adults' rights. A new bill, SB 26-051, shifts responsibility to operating system providers and app developers to verify users' ages.

Gamedate – A site to revive dead multiplayer games

Schedule sessions for dead games, niche mods, and retro netplay. No account needed.

When etcd crashes, check your disks first

https://nubificus.co.uk/images/cluster.png
The user set up a cloud-edge continuum testbed for a computer vision demo using MLSysOps, a framework that enables custom policies to adapt application deployment and runtime behavior. A subtle problem arose due to etcd timing out due to slow storage, which was resolved by optimizing ZFS settings on the NUC, teaching the user that distributed systems have strong opinions about their infrastructure.

How Taalas "prints" LLM onto a chip?

https://www.anuragk.com/blog/staticFiles/Taalas/GPUInference.png
Taalas released an ASIC chip running Llama 3.1 8B at 17,000 tokens per second, 10x cheaper and 10x less electricity than GPU based systems. The chip hardwires model weights, sidestepping memory bandwidth bottlenecks and using a "magic multiplier" for efficient multiplication.

Large Language Model Reasoning Failures

https://arxiv.org/static/browse/0.3.4/images/arxiv-logo-fb.png
Researchers present a comprehensive survey on Large Language Models' reasoning failures, categorizing them into embodied, informal, and formal types, and addressing root causes and mitigation strategies. The survey aims to unify research efforts and guide future development of more reliable and robust reasoning capabilities in LLMs.

Instant AI Response

chat jimmy LLM web interface

Claude Code's compaction discards data that's still on disk

https://opengraph.githubassets.com/d711d4a759e02ee8c56f2ae04a427dc6f921fd8208d3703e74298d5825d444bb/anthropics/claude-code/issues/26771
The current auto-compaction process in Claude irreversibly loses user-provided data when the context window fills up, causing Claude to hallucinate or ask for re-pasted data. A proposed solution is to tag compacted entries with transcript line-range annotations, allowing Claude to surgically recover specific content on demand and reducing token cost.

Happy Zelda's 40th first LLM running on N64 hardware (4MB RAM, 93MHz)

https://opengraph.githubassets.com/77115941910b4607fa7f9189a5fe050c8c2e9923a66f559e9a6014f64c3cc3c1/sophiaeagent-beep/n64llm-legend-of-Elya
A nano-GPT language model runs on a 1996 Nintendo 64's VR4300 CPU, generating responses in real-time. This is the first neural language model to run live inference on N64 hardware, enabling game mechanics previously impossible in the cartridge era.

MeshTNC is a tool for turning consumer grade LoRa radios into KISS TNC compatib

https://private-user-images.githubusercontent.com/662932/500575652-ca4e8caf-5eff-44d3-8ff0-c9d57bfc6ca3.png?jwt=eyJ0eXAiOiJKV1QiLCJhbGciOiJIUzI1NiJ9.eyJpc3MiOiJnaXRodWIuY29tIiwiYXVkIjoicmF3LmdpdGh1YnVzZXJjb250ZW50LmNvbSIsImtleSI6ImtleTUiLCJleHAiOjE3NzE3MDkyNjMsIm5iZiI6MTc3MTcwODk2MywicGF0aCI6Ii82NjI5MzIvNTAwNTc1NjUyLWNhNGU4Y2FmLTVlZmYtNDRkMy04ZmYwLWM5ZDU3YmZjNmNhMy5wbmc_WC1BbXotQWxnb3JpdGhtPUFXUzQtSE1BQy1TSEEyNTYmWC1BbXotQ3JlZGVudGlhbD1BS0lBVkNPRFlMU0E1M1BRSzRaQSUyRjIwMjYwMjIxJTJGdXMtZWFzdC0xJTJGczMlMkZhd3M0X3JlcXVlc3QmWC1BbXotRGF0ZT0yMDI2MDIyMVQyMTIyNDNaJlgtQW16LUV4cGlyZXM9MzAwJlgtQW16LVNpZ25hdHVyZT0zNWNiMTA2MzNkYTgzMzZlOWJlMTJlN2FmZmIyMzk0ZWRkZmY2MWU5Njc4OTA5NmRjZjZmNDg5NzA4YTU4ZTkzJlgtQW16LVNpZ25lZEhlYWRlcnM9aG9zdCJ9.xNF9QNxj-TwUNPG0iDGvd59pXgOq9Ib_CcJWX-0zUXA
MeshTNC can be flashed using MeshCore Flasher, OEM tools, or VS Code, and supports most hardware variants. It has a CLI with KISS mode for LoRa radio compatibility and can be used with APRS tools and AX.25 interface.

Online Pebble Development

https://cloudpebble.repebble.com/static/root/images/cloudpebble.png

Mathematics in the Library of Babel

https://images.squarespace-cdn.com/content/v1/57bf2a6de3df281593b7f57d/d71932d3-8dca-41d5-8b76-fee9b8de77d9/ai_dungeon.jpeg
The author, a mathematician, discusses the rapid progress of AI models in mathematics research, citing the First Proof project where models successfully solved 6-8 out of 10 lemmas, and argues that this technology is likely to be more impactful than the invention of the computer. The author believes that the main obstacle to automation is not the models' lack of knowledge, but rather their ...

Show HN: Script Snap – Extract code from videos

https://script-snap.com/_next/image?url=%2Fassets%2Fvisual-truth-table.png&w=3840&q=75
Script Snap is a video-to-blog engine with 99% accuracy for technical terms like Kubernetes and LLM. It validates against strict engineering ontologies to ensure correctness, not just confidence.

Show HN: Agent Passport – OAuth-like identity verification for AI agents

Agent Passport is an open-source identity verification layer for AI agents, preventing impersonation and data exfiltration. It uses Ed25519 authentication, JWT tokens, and a risk engine to securely verify agent identities.

DJB's Cryptographic Odyssey: From Code Hero to Standards Gadfly

Daniel J. Bernstein ("DJB") made his name in the late 2000s by upending established cryptographic primitives. In 2006 he unveiled Curve25519, an elliptic curve meant to be faster and safer than NIST's curves, and shortly after introduced complementary algorithms like the ChaCha stream cipher and Poly1305 MAC for authenticated encryption. He even packaged these into a high-level ...