Show HN: Simple algorithm and color space to generate diverse skin tones

75 automatoney 2026-08-04 15:21:51 UTC toneyalexander.github.io Llama Share 20

A custom color picker based on a color space was created to represent human skin tones digitally, aiming to identify a broad and inclusive range of colors that correspond to plausible skin tones. The color space was developed using principal component analysis and manual fitting of spherical equations to a dataset of labeled skin tones, resulting in a color picker with three independent ...

Ray Bradbury's "There Will Come Soft Rains" is set today (2026-08-04)

396 askvictor 2026-08-04 10:21:51 UTC short-stories.co Llama Share 177

A house stood alone in a post-apocalyptic city, its automated systems still functioning. The house's voices sang and announced the time, but chaos erupted when a fire broke out, destroying the house and its inhabitants.

DeepSeek V4 Flash on a Single AMD MI300X

253 zhoutong 2026-08-04 10:21:51 UTC github.com/ryanzhou Llama Share 58

This repository contains a production configuration for running DeepSeek-V4-Flash-0731 on AMD MI300X with fixes for FP8 format, MoE routing, and other issues. It achieves 6,988-7,019 tok/s on fresh prompts with a 2,048-token scheduler budget.

Germany Records Historic 12B KWh Solar Feed-In in July 2026

56 johnbarron 2026-08-04 14:21:51 UTC solarquarter.com Llama Share 24

Germany's renewable energy transition reached a milestone in July 2026 with 12 billion kWh of solar electricity fed into the public grid. This growth is attributed to investments by millions of households and businesses over the past two decades, making photovoltaics a key pillar of Germany's electricity supply.

Agent skills that bring team coding standards to Claude Code and Codex

55 kanfilior 2026-08-04 15:21:51 UTC github.com/tikalk Llama Share 26

ADLC Team Skills is an open-source Team Layer of the Twelve-Factor Agentic SDLC that turns AI agents into compliant team members, while Team AI Directives holds the team's version-controlled context modules and skills manifest. The ADLC framework provides tools for structured planning, evaluation, and directive compliance, aiming to move from a conversational model to a contract model and ...

Dates That Don't Exist (2015)

48 EndXA 2026-08-04 13:21:51 UTC blog.yossarian.net Llama Share 27

The Catholic world transitioned from the Julian to the Gregorian calendar in 1582, removing 10 days to correct a 11-day drift. This resulted in October 5-14, 1582, not existing in the Gregorian calendar system used today.

Keyv and friends compromised in active Shai-Hulud supply chain attack

115 cimi_ 2026-08-04 11:21:51 UTC aikido.dev Llama Share 47

Attackers compromised the GitHub account of keyv's maintainer, injecting credential-stealing malware into keyv and other caching utilities with 2 billion monthly installs. The malware harvests secrets from the victim's environment and exfiltrates them to a public GitHub repository.

LLMs reward expertise

1217 MaxMussio 2026-08-03 21:21:51 UTC seangoedecke.com Llama Share 502

Using large language models like LLMs requires domain knowledge to get the most out of them. A skilled prompter like Terence Tao can steer the model with expertise in the domain, but this skill is not easily replicable by following tips.

Buckminster Fuller: everything I know

77 simonebrunozzi 2026-08-04 12:21:51 UTC bfi.org Llama Share 25

Buckminster Fuller's 42-hour lectures cover his inventions, discoveries, and design approach, spanning topics from architecture to economics. The lectures are now available as a transcript, inviting public participation in editing and refining the content.

Xbox goes down. You can't play games you own on disc

361 surprisetalk 2026-08-04 12:21:51 UTC birchtree.me Llama Share 404

An Xbox outage blocked users from playing both digital and disc-based games. The author prefers PC gaming because it allows for maintaining access to games despite digital issues.

Online ad giant Adform was hacked, proving once again why ad blockers are needed

45 speckx 2026-08-04 15:21:51 UTC this.weekinsecurity.com Llama Share 6

Adform, an online ads provider, was hacked and served malicious ads containing code that stole crypto wallet addresses. The hack compromised devices of users visiting websites that used Adform, allowing malware to steal crypto.

Show HN: Fine-tune an 8B model on a 4 GB laptop GPU

76 MakazhanAlpamys 2026-08-04 11:21:51 UTC github.com/makazhanalpamys Llama Share 8

Soup is a tool for fine-tuning large language models with a simple workflow, using one config and one command. It supports various models and training tasks, including LoRA, quantization, and batching.

Harness Engineering for Self-Improvement

204 tosh 2026-08-04 06:21:51 UTC lilianweng.github.io Llama Share 36

Harness engineering is a crucial component of recursive self-improvement (RSI) in AI, enabling models to improve their own performance and adapt to new tasks through workflow design, context management, and optimization. Recent research has focused on developing automated design of agentic systems, meta-harnesses, and evolutionary search methods to improve harness engineering, with ...

Where .env Went Wrong

57 domenkozar 2026-08-04 14:21:51 UTC secretspec.dev Llama Share 27

The .env file is useful for delivering strings to a process but lacks a formal specification and has various parser disagreements. SecretSpec defines a durable project declaration and resolves the same declaration regardless of value storage or delivery, making it a better choice for teams.

Roame (YC S23) Is Hiring Lead Engineer

0 2026-08-04 12:21:51 UTC ycombinator.com Llama Share

Roame is a flight and hotel search engine for points and miles that helps travelers find deals in seconds. The company is looking for a senior engineer to lead its engineering team and drive its move to a modern Go service architecture.

"Clean" Code, Horrible Performance (2023)

68 FrojoS 2026-08-04 10:21:51 UTC computerenhance.com Llama Share 31

The author of the Performance-Aware Programming series argues that following "clean code" guidelines can significantly harm performance, citing a 10x speed difference between a "clean" code implementation and a more efficient one. By violating certain "clean code" rules, such as using polymorphism and avoiding internal knowledge, the author demonstrates a 10x to 15x speed improvement in a ...

Why Large Language Models Fail at Tabular Prediction

65 sbulaev 2026-08-04 10:21:51 UTC arxiv.org Llama Share 23

Large language models fail at predictive analytics over tabular data due to their inability to handle high dimensionality. In low dimensions, LLMs behave like local, distance-based methods, but in higher dimensions, their capability dissolves and no classical model can replicate their predictions.

Looking inside a 1970s PROM chip that stores data in microscopic fuses (2019)

12 Jimmc414 2026-08-04 13:21:51 UTC righto.com Llama Share 4

The MMI 5300 was a PROM chip from the 1970s that stored 1024 bits in tiny fuses, programmed once by blowing fuses. It had a 33x33 array of diodes and fuses, with address decode circuitry and output drivers to select and output the bits.

Rebuilding and analysing 4 years of Wordle stats from WhatsApp chat logs

13 surprisetalk 2026-08-04 14:21:51 UTC blog.omgmog.net Llama Share 3

The user has played 1,552 Wordle games, winning 99.42% of the time, with a longest streak of 278 days and current streak of 107 days. Data from WhatsApp shares reveals a win rate dip on Fridays and Sundays, and losses tend to occur on these days, but the clock, not the day of the week, is the main factor in determining when the user plays.

Show HN: Run an 80B Qwen in 4.3 GB of RAM on a Mac, and a 35B on an iPhone

271 leonickson 2026-08-04 02:21:51 UTC github.com/leonickson1 Llama Share 120

Swiftlet is a Swift + Metal runtime for Qwen3-Next and Qwen3.5/3.6 models, allowing them to run on iPhone with 2.5 GB of RAM. It streams weights from storage and activates only 3B parameters per token.

Why etymologies matter: How tracing words can illuminate history (2024)

58 benbreen 2026-08-04 10:21:51 UTC resobscura.substack.com Full Share 12

Ten advances in mathematics and theoretical computer science

600 milkshakes 2026-08-03 17:21:51 UTC openai.com Full Share 883

Apple says more ex-employees may have taken confidential data to OpenAI

16 thewebguyd 2026-08-04 15:38:51 UTC techcrunch.com Llama Share 1

Apple is requesting expedited discovery from OpenAI and former employees over alleged theft of intellectual property. Apple claims it has uncovered new evidence of misconduct beyond the original complaint, including meetings and sharing of confidential information.

Devtools must be open source

683 bryanmikaelian 2026-08-03 16:21:51 UTC blog.exe.dev Llama Share 226

Software engineers used to rarely write personal software, but now agents can easily manage customizing software, making it easier to personalize and maintain. Agents can build and manage custom software with minimal programming, allowing for more efficient and flexible use of software.

Amazonian civilization had estimated 3M people in 3% of forest area

231 marojejian 2026-08-04 00:21:51 UTC science.org Full Share 174

AI-Generated Images Discourage Me from Reading Your Blog

596 meysamazad 2026-08-04 12:21:51 UTC nelson.cloud Llama Share 351

You dislike AI-generated images in blogs, especially in indie blogs, and prefer authentic content from real humans. You value the unique perspective of individual bloggers over potentially AI-generated content.

Homebench – Benchmark local LLMs for speed, memory, and quality

34 davai-g 2026-08-04 10:21:51 UTC github.com/david-g-3654 Llama Share 1

homebench is a local-first tool that benchmarks LLMs on your laptop, measuring speed, memory, and quality. It provides a live leaderboard and supports various models and providers.

There Will Come Soft Rains (1950) [pdf]

277 pmg101 2026-08-04 00:21:51 UTC users.wpi.edu Full Share 99

Archaeologists Find Ancient Glyphs in the Amazon

36 wglb 2026-08-04 08:21:51 UTC nytimes.com Full Share 11

Please enable JS and disable any ad blocker

Mosh in a Lift (2012)

32 gavide 2026-08-04 12:21:51 UTC mosh.org Llama Share 12

The writer got stuck in a lift due to a faulty alarm system and was unable to contact the outside world until they used the mosh program to send an email. The fire brigade was called and rescued the writer, who then took the stairs to avoid using the lift again.