How I use Claude Code: Separation of planning and execution

441 vinhnx 2026-02-22 00:30:57 UTC boristane.com Llama Share 259

The user's workflow with Claude Code involves reviewing and approving a written plan before letting Claude write code, preventing wasted effort and ensuring better results. They use a disciplined pipeline that separates thinking from typing, involving research, plan annotation, and implementation, to ensure accurate and efficient coding.

Japanese Woodblock Print Search

72 curmudgeon22 2026-02-22 03:30:57 UTC ukiyo-e.org Llama Share 12

Ukiyo-e Search offers a unique feature to search Japanese woodblock prints by taking a picture of an existing print. It also allows users to see similar prints across multiple collections.

How Taalas "prints" LLM onto a chip?

88 beAroundHere 2026-02-22 02:30:57 UTC anuragk.com Llama Share 39

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.

A Botnet Accidentally Destroyed I2P

93 Cider9986 2026-02-22 01:30:57 UTC sambent.com Llama Share 49

I2P was hit with a massive 700,000-node Sybil attack on February 3, 2026, overwhelming its normal 15,000 to 20,000 active devices by a factor of 39 to 1. The attack was caused by the Kimwolf botnet, which accidentally disrupted I2P while attempting to use it as backup command-and-control infrastructure.

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

225 xaskasdf 2026-02-21 21:30:57 UTC github.com/xaskasdf Llama Share 53

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.

Two Bits Are Better Than One: making bloom filters 2x more accurate

81 matheusalmeida 2026-02-22 01:30:57 UTC floedb.ai Llama Share 15

We use bloom filters in Floe to speed up SQL queries by reducing the number of rows to probe in hash joins, achieving 2x fewer false results with a fixed 256KB bloom filter per join. By optimizing bloom filter performance and accuracy, we can avoid decompressing rows that will not be needed, resulting in significant performance gains and improved efficiency.

How far back in time can you understand English?

512 spzb 2026-02-21 13:30:57 UTC deadlanguagesociety.com GoogleT5 Share 274

A travel blogger takes to the train from london to wulfleet and writes about the journey. he writes that 'when i was first come to wullflete, i did not see the harbour' compare the very first line with the line from the 1800s section, and you'll see what i mean. the reasons for the difficulty are two-fold: grammar and spelling - the 1500s and 1600s

Gamedate – A site to revive dead multiplayer games

78 msuniverse2026 2026-02-22 01:30:57 UTC gamedate.org Full Share 7

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

Evidence of the bouba-kiki effect in naïve baby chicks

121 suddenlybananas 2026-02-21 22:30:57 UTC science.org Full Share 33

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

179 todsacerdoti 2026-02-21 20:30:57 UTC harudagondi.space Llama Share 43

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 ...

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

157 tosh 2026-02-21 19:30:57 UTC github.com/tnm Llama Share 84

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.

Scientists discover recent tectonic activity on the moon

47 bookmtn 2026-02-22 01:30:57 UTC phys.org Full Share 3

This page will redirect in a moment...

Claws are now a new layer on top of LLM agents

276 Cyphase 2026-02-21 12:30:57 UTC twitter.com/karpathy Full Share 721

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.

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

195 phront 2026-02-21 15:30:57 UTC koreaherald.com Llama Share 181

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.

Coccinelle: Source-to-source transformation tool

93 anon111332142 2026-02-21 18:30:57 UTC github.com/coccinelle Llama Share 28

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.

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

136 iancmceachern 2026-02-21 19:30:57 UTC carbuzz.com Llama Share 311

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.

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

90 PaulHoule 2026-02-21 19:30:57 UTC github.com/a-e-k Llama Share 33

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.

Carelessness versus Craftsmanship in Cryptography

15 ingve 2026-02-22 04:30:57 UTC blog.trailofbits.com Llama Share 0

Two popular AES libraries, aes-js and pyaes, have a default IV that leads to key/IV reuse bugs, affecting thousands of projects. The maintainer of aes-js took a cavalier approach, while the maintainer of strongSwan provided a model response to a similar bug.

Forward propagation of errors through time

24 iNic 2026-02-22 00:30:57 UTC nicolaszucchet.github.io Llama Share 0

We investigate a fundamental question in recurrent neural network training: why is backpropagation through time always ran backwards? We derive an exact gradient-based algorithm that propagates error forward in time, showing that this does not necessarily need to be the case. However, the algorithm suffers from critical numerical stability issues as the network forgets information faster.

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

1251 ColinWright 2026-02-21 08:30:57 UTC thelocalstack.eu Llama Share 432

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.

The Human Root of Trust – public domain framework for agent accountability

11 3du4rd0v3g4 2026-02-22 03:30:57 UTC humanrootoftrust.org Llama Share 4

The Human Root of Trust is a framework for cryptographic accountability in autonomous systems, ensuring a human is accountable for AI actions. It's a public domain framework for security engineers, cryptographers, and implementers to build upon.

A16z partner says that the theory that we’ll vibe code everything is wrong

125 paulpauper 2026-02-21 10:30:57 UTC aol.com Llama Share 190

Anish Acharya says AI-assisted coding is not worth it for all business functions, citing low cost savings and risks. He believes software stocks are oversold and companies should focus AI on core businesses or optimizing remaining costs.

Keep Android Open

2054 LorenDB 2026-02-21 08:30:57 UTC f-droid.org Llama Share 694

F-Droid users were misled into thinking Google canceled plans to lock down Android, but the plans are still scheduled. F-Droid is raising awareness about the issue with a banner in their app and website.

A New Perspective on Drawing Venn Diagrams for Data Visualization

7 IdealeZahlen 2026-02-22 05:30:57 UTC arxiv.org Prefix Share 3

We introduce VennFan, a method for generating $n$-set Venn diagrams based on the polar coordinate projection of trigonometric boundaries, resulting in Venn diagrams that resemble a set of fan blades. Unlike most classical constructions, our method emphasizes readability and customizability by using shaped sinusoids and amplitude scaling. We describe both sine- and cosine-based variants of ...

EDuke32 – Duke Nukem 3D (Open-Source)

182 reconnecting 2026-02-21 20:30:57 UTC eduke32.com Llama Share 65

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.

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

436 wizardforhire 2026-02-21 17:30:57 UTC milk.com Llama Share 193

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.

Finding forall-exists Hyperbugs using Symbolic Execution

36 todsacerdoti 2026-02-21 22:30:57 UTC dl.acm.org Full Share 2

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

82 akyuu 2026-02-21 20:30:57 UTC inputlag.science Llama Share 14

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.

“Playmakers,” reviewed: The race to give every child a toy

14 fortran77 2026-02-22 04:30:57 UTC newyorker.com Llama Share 2

Morris Michtom, a Jewish immigrant, created the Teddy bear in 1902, sparking a craze that led to the development of the American toy industry. Jewish immigrants played a significant role in shaping American childhood through their innovations in toys, despite facing poverty and cultural alienation.

Wave Twisters (2001)

6 hyperific 2026-02-22 05:30:57 UTC youtube.com Embed Share 3