Hacking OpenAI

278 Handy-Man 2026-09-18 03:21:32 UTC hacktron.ai Llama Share 98

Hacktron discovered two critical vulnerabilities in OpenAI's services, including a remote code execution (RCE) exploit that allowed them to access internal repositories and employee accounts. The vulnerability was patched within 72 hours of reporting.

Jemalloc 5.4.0

49 gkfasdfasdf 2026-09-18 04:21:32 UTC github.com/jemalloc Llama Share 9

This release includes over 160 commits focusing on technical debt cleaning, bug fixes, and performance improvements. It also introduces new features such as pinned memory tracking and OS abstraction layer.

The scourge of x86 emulation

61 dagmx 2026-09-18 04:21:32 UTC fex-emu.com Llama Share 2

The x86 Total Store Ordering (TSO) memory model poses significant challenges for emulating on ARM-based systems, with issues including alignment faults, unaligned atomic operations, and poor performance due to write-combined memory. Despite progress in addressing these problems through extensions like FEAT_LRCPC1/2/3 and hardware support for coherent cachelines, edge-cases remain unresolved, ...

Astra for Law

448 vertigoruntime 2026-09-17 20:21:32 UTC openai.com Llama Share 467

OpenAI introduces Astra for Law, a foundation for law firms and legal technology companies to build AI products around their expertise. It combines GPT-6 with settings and tools tailored for professional legal work.

Bonsai 2 27B: Near-Lossless Compression in a 9x Smaller Footprint

384 JonSchneider 2026-09-17 21:21:32 UTC prismml.com Llama Share 116

The Ternary Bonsai 2 27B model is released, offering a more capable and energy-efficient version of the previous Bonsai 27B model. It retains 98.2% of aggregate benchmark performance while being nearly 9x smaller in memory footprint.

Bend – A language that blocks AI mistakes via proof, on CPU and GPU

419 nicolas-siplis 2026-09-17 21:21:32 UTC bend-lang.com Llama Share 201

The Bend system uses laws and proofs to ensure AI agents implement human intents correctly, verifying code changes in real-time. With LAWS.bend, developers can declare rules that AIs must follow, preventing bugs from being shipped into production.

Pre-Greek: The lost language hidden within Ancient Greek

53 axiologist 2026-09-18 03:21:32 UTC linguisticdiscovery.com Llama Share 20

The article discusses the mystery language that influenced Ancient Greek, known as Pre-Greek. This language is thought to have been spoken in the Aegean region prior to the arrival of Indo-European speakers and has left a significant impact on the vocabulary of Ancient Greek. The authors analyze various linguistic features, including suffixes and prefixes, to identify potential loanwords from ...

Hister: A private search engine for the pages you visit and the files you keep

567 bookofjoe 2026-09-17 17:21:32 UTC github.com/asciimoo Llama Share 151

Hister is a private search engine that indexes web pages and local files for personal use. It allows searching from various interfaces, including the browser, terminal, and AI assistants, without relying on third-party providers or cloud services.

Qwen 3.8 Omni Flash

161 jjcm 2026-09-17 23:21:32 UTC qwen.ai Full Share 50

Qwen Studio offers comprehensive functionality spanning chatbot, image and video understanding, image generation, document processing, web search integration, tool utilization, and artifacts.

Wax motor

350 mhb 2026-09-17 18:21:32 UTC en.wikipedia.org Llama Share 62

A wax motor is a device that converts thermal energy into mechanical energy using phase-change waxes, providing smooth and gentle operation. Wax motors are used in various applications such as aerospace, washing machines, and hydronic heating systems due to their reliability and cost advantages.

Fujitsu launches made-in-Japan next-generation CPU FUJITSU-MONAKA

571 my123 2026-09-17 12:21:32 UTC global.fujitsu Full Share 216

We're verifying your browser Website owner? Click here to fix

Shapelearn Qwen 3.8 27B (13.1 GB VRAM)

44 syntaxing 2026-09-18 02:21:32 UTC byteshape.com Llama Share 4

The ShapeLearn models have been benchmarked and found to outperform the original Lite set, with GPU-5 being the default recommendation for its high aggregate score. The smaller model, ShapeLearn-Lite, also performed well despite its lower KLD ranking.

How to Write with an LLM

115 joeriddles 2026-09-17 22:21:32 UTC sockpuppet.org Llama Share 74

A writer shares two rules to use Large Language Models (LLMs) effectively, avoiding "LLM-creep" that can make writing sound unnatural. By using LLMs as a copyeditor rather than a ghostwriter and following these rules, writers can improve their work without losing their unique voice.

Waymo in Singapore

94 ramanan 2026-09-18 04:21:32 UTC waymo.com Llama Share 85

Waymo plans to introduce an all-electric autonomous ride-hail fleet in Singapore, aiming to provide a safe and accessible new mobility option. The company will work closely with government officials to support Singapore's green journey and vision for sustainable development.

Telstra outage: The night a network decided the year was 2006

45 TMWNN 2026-09-18 01:21:32 UTC netnod.se Llama Share 17

A Telstra mobile network in Australia experienced a failure due to a single GPS receiver incorrectly believing the year was 2006, causing time-related issues with voice calls and other services. The incident highlights the importance of precise timekeeping in modern cell phone protocols.

Apple detectives solved mystery of ancient tree and rewrote the history of fruit

35 nkurz 2026-09-18 02:21:32 UTC scientificamerican.com Llama Share 3

A team of apple detectives, including John Bunker and Todd Little-Siebold, identified a lost apple tree in Maine that was a direct ancestor of many American varieties. The tree, known as Drap d'Or de Bretagne, had been brought to North America by French colonists and had thrived in the region.

Flet 1.0 – Build cross-platform apps in Python

99 absqueued 2026-09-17 21:21:32 UTC flet.dev Llama Share 47

Flet is a Python framework that turns code into interactive apps with various controls and tools. It offers features like layout, navigation, forms, and dialogs with customizable colors and themes for building desktop, mobile, and web applications.

Fixing an NZXT Signal 4K30 part 2: the green/pink video bug

26 zdw 2026-09-18 03:21:32 UTC downtowndougbrown.com Llama Share 8

The user repaired an NZXT Signal 4K30 USB capture device with a bad solder joint and found it didn't work with one HDMI source, but was able to fix the issue by patching firmware using reverse engineering tools.

Ask A Monk – A digital wilderness for thoughts with no immediate answer

28 13613288957 2026-09-18 02:21:32 UTC askamonk.online Full Share 18

Ask anonymously. A stranger—not AI—will answer, because some things are easier to tell someone who doesn't know you.

Diplodocus, Long Thought Exclusively American, Turns Up in Spain

58 embedding-shape 2026-09-17 21:21:32 UTC sci.news Llama Share 35

Paleontologists have discovered the first confirmed fossils of Diplodocus outside North America in Spain. The 150-million-year-old dinosaur, similar to its American relatives, was found with well-preserved tail vertebrae and chevron bones near El Castellar.

The most important product decision is what you don't build

90 ChrisArchitect 2026-09-17 21:21:32 UTC liamnugent.me Llama Share 29

The author shares two common pitfalls in consumer-facing financial services apps: over-engineering and unnecessary complexity, which can lead to poor user experience. To avoid these issues, the author advises being extremely choosy about what is added to a system and militant about removing unnecessary elements.

CrowdSec Source Code Leak

147 eccgecko 2026-09-17 16:21:32 UTC crowdsec.net Llama Share 44

A source code leak occurred in CrowdSec's GitHub repository in May 2026, but the leaked code is mostly outdated and does not pose a significant threat. The team is investigating the incident and monitoring for any abnormal activity.

How Uber Protects Against Retry Storms

86 iscmt 2026-09-17 21:21:32 UTC uber.com Llama Share 33

Uber's context-aware mechanism in shared infrastructure handles errors more efficiently by establishing error ownership, which prevents retry storms and cascading failures. This approach limits retries to services that can genuinely resolve the issue, preventing exponential fan-out of requests.

How do we prevent mathemathics from devolving into the Medieval Era of secrecy?

108 jjgreen 2026-09-17 16:21:32 UTC mathoverflow.net Llama Share 91

Mathematicians are now facing a new era of secrecy due to the advent of large AI companies competing for solutions to famous problems, potentially stifling open science and collaboration. The shift in incentives may lead mathematicians to hold onto partial results, fearing that others will use powerful tools like AI to solve the problem before they can publish their work.

Why I didn’t sign the Fields medallists’ letter

241 simianwords 2026-09-17 11:21:32 UTC gowers.wordpress.com Llama Share 342

The author reflects on the impact of AI on mathematics, disagreeing with a letter signed by 25 Fields medallists that emphasizes the importance of conceptual understanding over problem-solving. The author believes that both aspects are valuable and that a spectrum of attitudes exists among mathematicians.

Show HN: Snapdrop: Instantly share files between devices. No setup, no signup

67 Capira 2026-09-17 21:21:32 UTC snapdrop.me Full Share 31

Allowing microphone access helps Chrome connect to nearby devices. Snapdrop will briefly activate your microphone and stop it immediately. No audio is recorded or sent.

Khipu (Quipu) Field Guide

10 SanjayMehta 2026-09-18 04:21:32 UTC khipufieldguide.com Llama Share 1

The Khipu Field Guide is a project by Ashok Khosla that democratizes khipu research, providing detailed symbolic renderings and interactive graphics of over 600 OKR khipus. The guide offers a comprehensive resource for understanding the Inka's use of cloth to communicate and record commitments.

Code Scans

17 geoffbp 2026-09-18 03:21:32 UTC devin.ai Llama Share 4

Code Scans is a tool that helps engineers investigate and improve their codebase by identifying areas for improvement based on specific goals. It uses Agentic MapReduce to break down large investigations into smaller batches, making it more efficient and effective.

Speeding up gearhash on ARM64

5 pranitha_m 2026-09-18 05:21:32 UTC sam.dev LocalLLM Share 0

gearhash now includes a NEON backend for ARM64, doubling speed on typical chunk sizes. It automatically selects NEON on aarch64, improving performance without requiring user changes.

Infinite-Parameter LLMs: Generating and Adapting Weights from Live Data

136 Betelbuddy 2026-09-17 17:21:32 UTC arxiv.org Llama Share 38

A conventional model cannot learn from live interaction due to frozen weights, but an Infinite-Parameter LLM can by generating weights from run-time data. This is achieved through a compact hypernetwork that updates the effective weight online using Bayesian belief over the generator's latent code.