Codex on AWS bedrock bug causing 10x charges

116 TheP1000 2026-08-21 03:27:29 UTC github.com/openai Llama Share 32

Native Codex CLI requests to Amazon Bedrock Mantle cannot opt into GPT-5.6 explicit prompt caching, resulting in high costs and cache-write tokens. The issue is related to agentic coding workloads with long stable instructions/tool definitions followed by changing content.

The August 17 outage

469 0xedb 2026-08-20 19:27:29 UTC github.blog Llama Share 536

GitHub experienced a 7-hour and 47-minute outage on August 17, affecting multiple services due to capacity failure. The company is accelerating work to improve reliability after two significant incidents in August.

I like 'em thick: an apology to my English teachers

693 Ariarule 2026-08-20 15:27:29 UTC experimental-history.com Llama Share 287

The author reflects on how great literature and art can be misunderstood as lacking substance, but in reality, they are complex and layered. Thick works of fiction require attention and effort to fully appreciate their depth.

HTML Can Do That

727 encyclopedism 2026-08-20 13:27:29 UTC chrisburnell.com Llama Share 180

This page showcases modern HTML features that can achieve dynamic functionality without JavaScript, highlighting browser implementation limitations and accessibility needs. The examples include popovers, modal dialog boxes, color pickers, range inputs, and more.

Malicious Rust crate Arrayref runs a build-time payload

475 abhisek 2026-08-20 13:27:29 UTC safedep.io Llama Share 405

A compromised Rust crate, arrayref, was released on crates.io with a malicious dependency proc-macro1. The malicious code ran at build time and fetched a remote binary without validation.

Ox Alpha

116 mtokmak06 2026-08-21 00:27:29 UTC openrouter.ai Llama Share 85

Ox Alpha is a stealth model for coding, agentic work, and production workloads with a 1,048,576-token context window. It accepts text, images, and video inputs and returns text outputs through OpenRouter's unified API.

I should have loved biology (2020)

257 tyre 2026-08-20 18:27:29 UTC jsomers.net Llama Share 100

The author reflects on their biology education, feeling that it was too dry and lacked real-world context. They credit Lewis Thomas's writing style for making the subject more engaging and understandible. The author believes that learning about biology should start with questions and hands-on projects rather than just memorizing facts.

Version Control for Everything

15 evakhoury 2026-08-21 05:27:29 UTC tyoverby.com Llama Share 3

AI-assisted coding has reached a point where it's hard to use without version control, which is lacking in non-programming use cases. Integrating AI into the development process could greatly improve productivity, but current tools and workflows hinder its adoption.

The Religious Experience of Philip K. Dick by R. Crumb (1986)

18 wise_blood 2026-08-21 06:27:29 UTC philipdick.com Full Share 6

I have added a link to a .cbr file (Comic Book Archive) to download and read on your electronic device if that would be easier for your enjoyment of the story. The Religious Experience of Philip K.

Stop Making TUIs

11 underdeserver 2026-08-21 06:27:29 UTC sockpuppet.org Llama Share 7

The author is re-evaluating the relationship between terminal and command line interfaces, suggesting that native user interfaces are becoming more accessible. They've built several macOS applications using SwiftUI, including a Markdown viewer and a calculator-style frontend for SageMath.

Japan tried to build an operating system for the world, the US intervened

53 rdmuser 2026-08-21 06:27:29 UTC xda-developers.com Llama Share 22

TRON, a Japanese operating system developed in the 1980s, was abandoned due to US government concerns over its adoption in schools and telecom networks. Despite this, its embedded variant ITRON became widely used globally.

Show HN: Argentic – An L402 Lightning toll booth for AI scraping agents

4 Ag0146 2026-08-21 06:27:29 UTC argentic.network Full Share 0

Launch HN: Vendo (YC S26) – Let users build features on top of your product

34 yousefh409 2026-08-20 16:27:29 UTC github.com/runvendo Llama Share 18

Vendo is an open-source customization layer for B2B SaaS teams, allowing users to build features and micro-apps on top of their product without touching the source code. It acts as a signed-in user through the product's API, rendering UI in a sandboxed surface.

CIA funding helped keep NeXT afloat in the 80s

392 EwanG 2026-08-20 13:27:29 UTC wsj.com Full Share 234

Please enable JS and disable any ad blocker

Why aren't smart people happier? (2022)

155 rafaelc 2026-08-20 19:27:29 UTC experimental-history.com Llama Share 209

The author argues that intelligence is not just about solving well-defined problems, but also about navigating poorly defined problems in life. They suggest that people who excel at solving well-defined problems may not necessarily be happier or more fulfilled.

Show HN: Huzzah – a novel approach to coding with AI

288 danielvaughn 2026-08-20 19:27:29 UTC danielvaughn.dev Llama Share 152

A software engineer is dissatisfied with the current state of coding with AI, seeking a better balance between automation and control. They're developing an experimental editor called Huzzah to address these issues by using pseudocode prompts that are declarative and persistent.

Captain Zilog

43 rbanffy 2026-08-21 01:27:29 UTC zilog.com Full Share 5

Make a 6-Tesla-class high-temperature superconducting dipole magnet at 4.2 K

36 supermagnet 2026-08-21 01:27:29 UTC journals.aps.org Full Share 7

Seed: Minimal, self-modifying agent harness

5 gandalfgeek 2026-08-21 05:27:29 UTC github.com/vivekhaldar LocalLLM Share 0

A self-contained agent system using seed.py to initialize a single tool and system prompt, with all growth managed session-by-session in self/. Sessions are ephemeral, with transcripts saved for later study, allowing different agents to emerge from each directory.

Vomit: Clean up Claude 5's token output with a separate LLM

236 Bluestein 2026-08-20 16:27:29 UTC github.com/zachahn Llama Share 241

Vomit converts Claude's tokens into English by piping them through a local LLM, which is fully local and has no external dependencies. It can be used in non-invasive mode to translate tokens for a specified session or follow the latest one.

Linux 7.2

239 mariuz 2026-08-20 16:27:29 UTC igalia.com Llama Share 81

This kernel cycle saw significant improvements, including cache-aware scheduling and runtime power management for GPUs on Raspberry Pi devices. The DRM scheduler fair policy was also improved but remains opt-in due to a last-minute regression report.

AliExpress runs silent WebAudio fingerprinting that breaks Bluetooth multipoint

950 emctech 2026-08-20 10:27:29 UTC blog.laserphile.com Llama Share 298

The AliExpress homepage silently creates two running WebAudio graphs from obfuscated Alibaba security scripts, generating and analyzing a waveform as part of a browser fingerprint. Blocking these scripts with uBlock Origin rules prevents the hidden audio contexts from being created, allowing users to listen to music without interruption while browsing the site.

Git at any scale

322 meetpateltech 2026-08-20 07:27:29 UTC cursor.com Llama Share 102

Git repositories are notoriously difficult to host at scale due to their distributed nature, which makes it hard for a single server to manage the data. A centralized approach is often necessary, but this can be complex and unreliable. The author of the article presents an alternative solution called Continuity, which stores Git repositories in S3-compatible object storage using a write-ahead ...

Speeding Up (Small) Ruby Hashes

47 arto 2026-08-20 21:27:29 UTC byroot.github.io Llama Share 0

The author discusses their struggles with writing blog posts, particularly the first sentence, and how it helps them think about problems. They then dive into optimizing Ruby's Hash class by implementing a more efficient data structure called ar_table.

AI companies destroy physical books – let's scan rare books before it's too late

275 Cider9986 2026-08-21 03:27:29 UTC annas-archive.gl Llama Share 205

AI companies are secretly buying and destroying physical books to train their models, threatening human knowledge. Anna's Archive is calling on volunteers worldwide to scan and upload books before this cultural heritage disappears forever.

Anti-AI fonts are useless and harmful

149 speckx 2026-08-20 15:27:29 UTC blog.yaros.ae Llama Share 108

Creating anti-AI fonts is an inaccessible solution that may lead to centralized identity verification systems and filtering content, ultimately benefiting those who want to censor the web. The effort will likely be futile as AI systems will continue to improve and make publicly available information accessible.

SpacetimeDB: a short technical review

85 hurrrr 2026-08-20 19:27:29 UTC strn.cat Llama Share 17

SpacetimeDB's new version 2.0 has a unique approach with benchmarks that mock competitors, but these are not honest and have technical flaws. The database is an all-in-one solution with application server capabilities, but its scalability and availability are limited by the CPU and RAM capacity of the machine it runs on.

Stop Anthropomorphizing Intermediate Tokens as Reasoning/Thinking Traces (2025)

224 nunodonato 2026-08-20 07:27:29 UTC arxiv.org Llama Share 144

Researchers argue that anthropomorphizing intermediate tokens in language models can be misleading and confusing. They call for the community to avoid such metaphors to effectively use these models.

Consumer Rights Wiki

269 gregsadetsky 2026-08-20 18:27:29 UTC consumerrights.wiki Llama Share 50

The wiki has received a major update with new features, including improved feedback and award systems, enhanced moderation tools, and better performance. The update also includes temporary anonymous accounts for edits and improvements to search functionality on Google.

Project Cybersyn (2022)

59 cassepipe 2026-08-20 18:27:29 UTC bactra.org Llama Share 45

The Chilean government attempted to implement a cybernetic system called Cybersyn in the 1970s, aiming to improve economic management through data analysis and early warning systems. However, the project ultimately failed due to technical issues and lack of decentralization.