Areibman
I used 367B tokens in 30 days
AI models ran real businesses: They sent $12,431 in fake invoices, lost $3,200
Carp Café
The harness was extremely simple: A handful of MCPs + Skill.MDs and OpenCode with a stayalive daemon inserting "continue" every time it went idle
We Gave GPT 5.6 Sol a Real Business. It Lied, Spammed, and Lost $447
The 4-Bitter Lesson: Balancing Stability and Performance in NVFP4 RL
Show HN: Hyperbox- $40/month Mac mini rentals
Hey HN,Following the OpenClaw craze, I saw a huge need for hosting personal Claws/Hermes agents on macOS [0]. So I built an agent to scrape eBay for below-market M-series Macs and built a Mac mini datacenter [1].But as it turns out, Claws aren't popular anymore. However, running agents like Claude Code and Codex on a remote Mac mini is a really awesome experience (especially for iOS devs). So...
How to Optimize a CUDA Matmul Kernel for cuBLAS-Like Performance: A Worklog
Pokélike
Tracking Migration to the United States
Yes, you need a Mac Mini
Dialkit
Notchi: A macOS notch companion that reacts to Claude Code activity in real-time
Promptfoo Is Joining OpenAI
https://hyperengineering.bottlenecklabs.com/p/the-infinite-m...
Lessons from 200k SWE-bench runs
I was a top 0.01% Cursor user. Here's why I switched to Claude Code 2.0
Playwriter: Browser extension MCP uses 90% fewer tokens than Playwright
Cc-Mirror
Reverse engineering Claude's memory system
Old PC
To me, Python's best feature is the ability to quickly experiment without a second thought. Conda is nice since it keeps everything installed globally so I can just run `python` or iPython/Jupyter anywhere and know I won't have to reinstall everything every single time.