CSS Grid Lanes

286 frizlab 2025-12-19 22:36:33 UTC webkit.org Llama Share 91

Grid Lanes is a new CSS feature that allows for flexible masonry layouts on the web, using grid-template-columns and grid-auto-flow to create a layout that works on all screen sizes. It also introduces a new concept called "tolerance" that lets developers adjust how picky the layout algorithm is when deciding where to place items.

Mistral OCR 3

402 pember 2025-12-19 19:36:33 UTC mistral.ai Llama Share 75

User had a busy day at the office, dealing with Westcott and liquidators, and plans to go to Eddie's Cafe and a Chinese dinner. Mistral OCR 3 is a new AI model for extracting text and images from documents with high fidelity, ideal for enterprise pipelines and interactive workflows.

PBS News Hour West to go dark after ASU discontinues contract

62 heavyset_go 2025-12-20 00:36:33 UTC statepress.com Llama Share 14

ASU's Walter Cronkite School of Journalism will end its contract with PBS News Hour West, closing a reporting hub that covered the western US. The decision is reportedly based on ASU's revised priorities, affecting over 20% of News Hour's audience.

Garage – An S3 object store so reliable you can run it outside datacenters

464 ibobev 2025-12-19 16:36:33 UTC garagehq.deuxfleurs.fr Llama Share 93

The project provides software that runs over the Internet across multiple datacenters, allowing deployment everywhere. It received funding from the European Union and NLnet for its development.

Show HN: Carolina Cloud – 1/3 the cost of AWS for data science workloads

11 bojangleslover 2025-12-20 01:36:33 UTC carolinacloud.io Llama Share 3

Derek earned a Master's in Computational Biology from Cambridge and worked in industry, applying analytical expertise to sports betting and finance. He now focuses on Carolina Cloud, developing high-performance cloud solutions combining science, data, and computing.

A Better Zip Bomb

88 kekqqq 2025-12-19 21:36:33 UTC bamsoftware.com Llama Share 31

The user is trying to find information about the creator of 42.zip, a zip bomb that expands to 4.5 PB if recursively unzipped, and discusses its construction and optimization. The user also talks about optimizing the compression ratio of the zip bomb, using a custom DEFLATE compressor, and discusses the limits of the zip format and how to surpass them using Zip64.

TP-Link Tapo C200: Hardcoded Keys, Buffer Overflows and Privacy

224 sibellavia 2025-12-19 18:36:33 UTC evilsocket.net Llama Share 65

The author reverse engineered a TP-Link Tapo C200 camera's firmware using AI and found several security vulnerabilities, including a private key exposure and integer overflows in the HTTPS server and ONVIF service. The vulnerabilities allow for MitM attacks, remote code execution, and enumeration of nearby WiFi networks, which can be used to determine the camera's location.

8-bit Boléro

179 Aissen 2025-12-19 19:36:33 UTC linusakesson.net Llama Share 29

The creator made a video showcasing a unique musical instrument setup with 9 custom instruments, including NES timpani, using a combination of C64 hardware and clever audio processing. The video and audio were recorded simultaneously, with some sections recreated using an automaton for precise sound capture.

Amazon will allow ePub and PDF downloads for DRM-free eBooks

547 captn3m0 2025-12-19 10:36:33 UTC kdpcommunity.com Full Share 280

Sorry to interrupt

GotaTun – Mullvad's WireGuard Implementation in Rust

544 km 2025-12-19 11:36:33 UTC mullvad.net Llama Share 112

GotaTun is a fast and efficient WireGuard implementation in Rust, forked from BoringTun, with added features like DAITA and Multihop. It has replaced wireguard-go on Android, resulting in a significant drop in user-perceived crashes and improved user experience.

Graphite is joining Cursor

176 fosterfriends 2025-12-19 16:36:33 UTC cursor.com Llama Share 201

Graphite, a code review platform, has been acquired by Cursor and will continue to operate independently. The acquisition aims to integrate Graphite with Cursor to improve code review and collaboration workflows.

Brown/MIT shooting suspect found dead, officials say

100 anigbrowl 2025-12-19 19:36:33 UTC washingtonpost.com Full Share 112

Show HN: TinyPDF – 3kb pdf library (70x smaller than jsPDF)

119 lulzx 2025-12-19 18:36:33 UTC github.com/lulzx Llama Share 16

The code uses the TinyPDF library to create a PDF document with a custom layout, including text, images, and a table. It generates an invoice PDF with company and billing information, a table of items, and totals, and saves it to a file named "invoice.pdf".

Rust's Block Pattern

127 zdw 2025-12-19 17:36:33 UTC notgull.net Llama Share 53

The "block pattern" is a Rust idiom that organizes code into a block to improve readability and reduce namespace pollution by encapsulating implementation details. This pattern has several advantages, including clear intent, reduced pollution, and improved idiot-proofing, and can be used to restrict mutability and improve code flow.

Qwen-Image-Layered: transparency and layer aware open diffusion model

67 dvrp 2025-12-19 21:36:33 UTC huggingface.co Llama Share 9

Qwen-Image-Layered decomposes images into semantically disentangled RGBA layers for independent editing. It surpasses existing approaches in decomposition quality and enables consistent image editing.

Performance Hints (2023)

53 danlark1 2025-12-19 20:36:33 UTC abseil.io Llama Share 24

The document discusses general performance tuning principles and techniques for improving software performance, focusing on single-binary context and providing examples in C++ with illustrative source code changes. It emphasizes the importance of considering performance early on and using techniques such as algorithmic improvements, data structure optimization, and profiling tools to achieve ...

The FreeBSD Foundation's Laptop Support and Usability Project

138 mikece 2025-12-19 15:36:33 UTC github.com/freebsdfoundation Llama Share 46

The FreeBSD Foundation is investing $750,000 to improve laptop support and accessibility through updates to FreeBSD 14.x and documentation. The project will be managed through the Laptop and Desktop Working Group and updates will be shared on GitHub and the FreeBSD Foundation newsletter.

We ran Anthropic’s interviews through structured LLM analysis

47 jp8585 2025-12-19 23:36:33 UTC playbookatlas.com Llama Share 49

People are using AI despite feeling conflicted about it, with 85.7% experiencing unresolved tensions. They're adopting AI for short-term benefits while worrying about long-term consequences.

Language Immersion, Prison-Style (2017)

12 johnny313 2025-12-20 00:36:33 UTC themarshallproject.org Llama Share 0

A white woman in her 20s, serving a 60-month sentence for conspiracy to distribute heroin, learned Spanish in prison by interacting with hundreds of Mexican women who have ICE holds. Through her studies, she gained a new perspective on the world, seeing people facing deportation as friends and family, and developed a deeper understanding of global issues.

Believe the Checkbook

120 rg81 2025-12-19 16:36:33 UTC robertgreiner.com Llama Share 53

AI companies claim engineering is being automated away but secretly pay millions to acquire human teams that work with AI. They value human judgment over code production, indicating AI is a force multiplier for high-judgment people.

LLM Year in Review

63 swyx 2025-12-19 21:36:33 UTC karpathy.bearblog.dev Llama Share 22

2025 was a transformative year for LLMs with the emergence of Reinforcement Learning from Verifiable Rewards (RLVR) and the development of new LLM apps and agents. The industry is on the cusp of significant progress with LLMs displaying jagged capabilities, empowering regular people to approach programming, and opening up new possibilities for software development and interaction.

Lite^3, a JSON-compatible zero-copy serialization format

137 cryptonector 2025-12-19 11:36:33 UTC github.com/fastserial Llama Share 33

Lite³ is a zero-copy binary serialization format that encodes data as a B-tree, allowing for O(log n) access and mutation of arbitrary fields. It is schemaless, self-describing, and supports conversion to/from JSON.

Vm.overcommit_memory=2 is the right setting for servers

42 signa11 2025-12-19 19:36:33 UTC ariadne.space Llama Share 70

Enabling memory overcommit in Linux allows allocations to succeed even if memory is not available, leading to unpredictable failures. Disabling overcommit ensures immediate allocation failures, making debugging easier and shifting responsibility to application developers.

Reverse Engineering US Airline's PNR System and Accessing All Reservations

92 bearsyankees 2025-12-19 18:36:33 UTC alexschapiro.com Llama Share 41

A vulnerability was discovered in Avelo Airlines' reservation system, allowing an attacker to access full reservation details, including PII, with a 6-character alphanumeric code and no last name verification. The Avelo team responded quickly and professionally, patched the issue, and maintained clear communication throughout the disclosure process.

NOAA deploys new generation of AI-driven global weather models

89 hnburnsy 2025-12-19 19:36:33 UTC noaa.gov Full Share 60

NOAA has launched a groundbreaking new suite of operational, artificial intelligence (AI)-driven global weather prediction models, marking a significant advancement in forecast speed, efficiency, and accuracy. The models will provide forecasters with faster delivery of more accurate guidance, while using a fraction of computational reso

The pitfalls of partitioning Postgres yourself

48 abelanger 2025-12-19 20:36:33 UTC hatchet.run Llama Share 5

Hatchet's partitioning system initially worked well but queries on partitioned tables became slower due to incorrect query statistics caused by autovacuum not running ANALYZE on partitioned tables. Manually running ANALYZE on the parent partition table fixed the issue, and the team learned to use DETACH PARTITION...CONCURRENTLY and run ANALYZE regularly on parent partition tables.

The scariest boot loader code

30 todsacerdoti 2025-12-19 20:36:33 UTC miod.online.fr Llama Share 1

The author obtained an HP 9000/720 workstation and worked on porting OpenBSD to the PA-RISC architecture, with the help of Michael Shalayeff and later Matt Fredette who also ported NetBSD to the same architecture. The author eventually succeeded in allowing the OpenBSD boot blocks to display and modify the firmware's console path, enabling the 712 model to run with a serial console.

History LLMs: Models trained exclusively on pre-1913 texts

767 iamwil 2025-12-19 02:36:33 UTC github.com/dgoettlich Llama Share 380

Researchers created a family of 4 billion parameter large language models trained on historical data up to specific cutoff dates. These models serve as windows into the past, enabling research in humanities, social sciences, and computer science.

Show HN: Misata – synthetic data engine using LLM and Vectorized NumPy

16 rasinmuhammed 2025-12-19 22:36:33 UTC github.com/rasinmuhammed Llama Share 0

The code uses Misata to generate data for various use cases, including a SaaS company and a fitness app, with options for LLM-powered generation and custom output. It also includes features for data customization, noise injection, and reproducibility.

Buteyko Method

36 rzk 2025-12-19 22:36:33 UTC en.wikipedia.org Llama Share 17

The Buteyko method is a breathing technique that aims to regulate breathing and alleviate symptoms of asthma and other respiratory conditions by correcting hyperventilation and encouraging shallower breathing. Its effectiveness is disputed in the medical community due to limited evidence supporting its claims.