Postmortem: TanStack npm supply-chain compromise

480 varunsharma07 2026-05-11 21:55:03 UTC tanstack.com Llama Share 161

An attacker compromised 42 @tanstack/* npm packages by combining GitHub Actions cache poisoning and runtime memory extraction of an OIDC token, publishing 84 malicious versions on 2026-05-11. The attack was detected within 20 minutes by an external researcher and all affected versions have been deprecated, but users who installed the affected versions are recommended to rotate their credentials.

UCLA discovers first stroke rehabilitation drug to repair brain damage (2025)

198 bookofjoe 2026-05-11 17:55:03 UTC stemcell.ucla.edu Llama Share 38

UCLA researchers found a drug called DDL-920 that reproduces the effects of physical stroke rehabilitation in mice. The drug targets a brain cell called a parvalbumin neuron to help recover movement control after stroke.

If AI writes your code, why use Python?

89 indigodaddy 2026-05-11 20:55:03 UTC medium.com/@nmitchem Llama Share 96

AI has improved significantly in writing systems languages like Rust, Go, and Swift, making them easier to use. This shift changes the constraints of software development, favoring languages that are easiest for agents, not humans.

Library for fast mapping of Java records to native memory

101 joe_mwangi 2026-05-11 19:55:03 UTC github.com/mamba-studio Llama Share 23

TypedMemory is a Java library for working with contiguous off-heap memory through strongly typed views. It provides a type-safe abstraction over memory for systems, interop, graphics, and data-oriented programming.

GitLab announces workforce reduction and end of their CREDIT values

276 AnonGitLabEmpl 2026-05-11 20:55:03 UTC about.gitlab.com Llama Share 266

GitLab is undergoing a restructuring process to meet the opportunities of the agentic era, including reducing its country footprint by 30%, flattening its organization, and right-sizing roles. The company is reaffirming its Q1 and full year FY27 guidance and will reinvest the majority of savings into accelerating its progress against growth and technological initiatives.

Nullsoft, 1997-2004 (2004)

215 downbad_ 2026-05-11 15:55:03 UTC slate.com Llama Share 70

Justin Frankel, founder of Nullsoft, was a rebellious programmer who created popular music tools like Winamp and Gnutella, often challenging the music industry and his corporate parent AOL. He quit AOL and Nullsoft after releasing a private file-sharing system WASTE, which made it difficult for authorities to track copyright infringement.

Griffin PowerMate driver for modern macOS

28 classichasclass 2026-05-11 21:55:03 UTC github.com/jameslockman Llama Share 9

The PowerMate is a small device with a knob and button that can be used to scroll and click. It reports rotation and button events to the driver, which can be mapped to actions such as scrolling, clicking, and media control.

Ratty – A terminal emulator with inline 3D graphics

606 orhunp_ 2026-05-11 10:55:03 UTC ratty-term.org Full Share 194

Ratty is a GPU-rendered terminal emulator with a spinning rat cursor, multiple 3D presentation modes and support for inline 3D graphics. Inspired by TempleOS and built with Rust and Ratatui.

Can we code our way out of gentrification?

8 burlesona 2026-05-11 23:57:03 UTC freerange.city Prefix Share 5

My neighborhood in Denver is experiencing a kind of slow strangulation. Today, the neighborhood is mixed-income with a range of families from different backgrounds. A decade ago, it was lower income than it is today, and two decades ago, lower still. Aka, we’re gentrifying. But the gap between new and old residents is becoming extreme. People who have lived in the neighborhood for 10 years ...

I let AI build a tool to help me figure out what was waking me up at night

55 showmypost 2026-05-11 21:55:03 UTC martin.sh Llama Share 54

The user built a system to identify and record noise events that wake them up at night using AI tooling, a Raspberry Pi, and a smart home setup, allowing them to pinpoint and address the causes of their sleep disruptions. This project demonstrates how AI tooling has lowered the barrier for personal tooling, enabling the user to tackle small problems that were previously deemed not worth building.

Gmail registration now requires scanning a QR code and sending a text message

553 negura 2026-05-11 07:55:03 UTC discuss.privacyguides.net Llama Share 397

You created a Google account with an Italian SIM tied to your ID, then enabled 2FA and saved recovery codes. Google likely keeps a record of your used phone numbers but can't verify your account with a reassigned number.

Google says criminal hackers used AI to find a major software flaw

109 donohoe 2026-05-11 20:55:03 UTC nytimes.com Llama Share 86

A hacking group used AI to detect a previously unknown bug, posing a threat to digital security. Google thwarted the attack, highlighting concerns over AI's potential impact on cybersecurity.

Silverback Imfura took a chance, and ended up alone

25 alex000kim 2026-05-11 21:55:03 UTC gorillafund.org Llama Share 9

Silverback Imfura left the Pablo group after a series of disruptions, forming a new group with two females who had recently joined. Imfura's new group was unstable and the females eventually returned to the Pablo group, where they were welcomed back.

The rise and fall of snake oil

25 samizdis 2026-05-11 20:55:03 UTC historytoday.com Full Share 14

Establishing a secure connection... Request ID: 9f977d36d5510c51c940ff6f54185a46

Training an LLM in Swift, Part 1: Taking matrix mult from Gflop/s to Tflop/s

212 zdw 2026-05-11 13:55:03 UTC cocoawithlove.com Llama Share 11

The author optimizes a handwritten matrix multiplication code in Swift for training a Large Language Model (LLM) on Apple Silicon, exploring various techniques such as SIMD, AMX, and Metal to achieve performance comparable to C. The author concludes that while Swift can achieve similar performance to C, it requires more complex and unwieldy code, and that the fastest CPU instruction on Apple ...

Interfaze: A new model architecture built for high accuracy at scale

107 yoeven 2026-05-11 16:55:03 UTC interfaze.ai Llama Share 28

Interfaze is a new model architecture that outperforms other models in OCR, vision, and structured output tasks. It offers a balance between performance and price, making it suitable for high-volume tasks.

Abstract Machines for Logic Programs

6 surprisetalk 2026-05-11 23:55:03 UTC chrisistyping.bearblog.dev Prefix Share 1

[Based on conversations with Rob Simmons. Reader prerequisites: comfortable with inference rules, peano notation, state machines.] The following inference rules, it can be argued, define addition: whenever plus N M P holds, the numbers N and M sum to P. ----------------- pz plus 0 N N plus N M P -------------------- ps plus (s N) M (s P) ----------------- pz plus 0 2 2 --------------- ps plus ...

Interaction Models

75 smhx 2026-05-11 21:55:03 UTC thinkingmachines.ai Llama Share 8

Researchers at Thinking Machines are developing a new type of AI model called interaction models that can handle real-time conversations and collaboration with humans. These models can perceive and respond to humans in real-time, across multiple modalities such as audio, video, and text, and can be used for a variety of applications including customer service, education, and healthcare.

Show HN: OpenGravity – A zero-install, BYOK vanilla JS clone of Antigravity

43 ab613 2026-05-11 20:55:03 UTC github.com/ab-613 Llama Share 16

OpenGravity is a lightweight, BYOK recreation of Google Antigravity UI with a live terminal and proactive agent. It needs community contributions to add features like top menu bars, settings UI, and basic functionality.

Bild AI (YC W25) Is Hiring Founding Product Engineers

0 2026-05-11 17:55:03 UTC bild.ai Full Share

Bild AI Jobs

AMÁLIA and the future of European Portuguese LLMs

117 johnbarron 2026-05-11 14:55:03 UTC duarteocarmo.com Llama Share 57

The Portuguese government invested 5.5 million euros in AMÁLIA, a large-scale LLM for European Portuguese. The model's architecture is based on EuroLLM, but with modifications to increase European Portuguese data.

CUDA-oxide: Nvidia's official Rust to CUDA compiler

355 adamnemecek 2026-05-11 16:55:03 UTC nvlabs.github.io Llama Share 107

cuda-oxide is a Rust-to-CUDA compiler for safe GPU programming. It compiles Rust code to PTX and supports async GPU programming with Rust's type system and ownership model.

The Boston library where you still can borrow a giant puppet

47 gnabgib 2026-05-11 17:55:03 UTC binj.news Llama Share 7

Sara Peattie runs a free puppet library in Boston's basement, where puppets of all shapes and sizes can be borrowed by the public. Peattie, a Boston-based puppeteer, creates her own puppets using papier-mâché and fabric, often making them look more complicated than they are.

Counting Fast in Erlang with:counters and:atomics

69 malmz 2026-05-11 18:55:03 UTC andrealeopardi.com Llama Share 3

The speaker discussed two new OTP data structures, :atomics and :counters, which provide fast and atomic operations for shared memory spaces. They compared these to ETS, a traditional shared memory space, and found that :atomics and :counters offer better performance and tighter semantics for certain use cases.

Show HN: E2a – Open-source email gateway for AI agents

18 mnexa 2026-05-11 20:55:03 UTC github.com/mnexa-ai Llama Share 2

Authenticated email gateway for AI agents with inbound and outbound email processing, verification, and approval features. It supports local and cloud agents, with a shared domain for instant slug-based onboarding and managed deliverability.

Building a web server in aarch64 assembly to give my life (a lack of) meaning

99 theanonymousone 2026-05-11 12:55:03 UTC imtomt.github.io Llama Share 32

The author built a web server called ymawky in AArch64 assembly for macOS to understand how a web server works and to strip away convenience layers. The project involved parsing HTTP requests, handling edge cases, and implementing security features such as request timeouts and path sanitization in a low-level and manual way.

Linux Terminal Memory Usage

40 speckx 2026-05-11 19:55:03 UTC gilesorr.com Llama Share 38

User uses multiple terminals across multiple desktops, switching between xterm, rxvt, lxterminal, alacritty, and kitty due to various features. User discovers smem, which shows that kitty consumes excessive swap memory, leading to a switch back to xterm and then to alacritty. User devises a test to measure memory usage of various terminals and finds st to be the least memory-hungry option.

Hardware Attestation as Monopoly Enabler

2076 ChuckMcM 2026-05-11 00:55:03 UTC grapheneos.social Llama Share 702

Google and Apple are expanding hardware-based attestation to lock out non-approved devices and OSes, presented as a security feature but actually anti-competitive.

Software engineering may no longer be a lifetime career

352 movis 2026-05-11 14:55:03 UTC seangoedecke.com Llama Share 589

Some software engineers think AI use makes them less intelligent by not learning tasks, but this is a bad argument. They should adapt to AI like construction workers do to heavy lifting, even if it's detrimental to their long-term cognitive abilities.

From Buffon's Needle to Buffon's Noodle

24 _alternator_ 2026-05-11 19:55:03 UTC mbmccoy.dev Llama Share 7

A needle of length L is dropped onto a hardwood floor with floorboards of width W. The average number of lines it crosses is 2L / pi W.