Anthropic deleted 80% of Claude Code's system prompt and the model got smarter.
Boris Cherny explained the whole method on stage at YC in July. 15 minutes, no slides, all of it actionable: 1. every new model, they delete the system prompt and bring it back one line at a time.
A fan built a Raspberry Pi that makes GPT, Claude, and Gemini debate every question before answering
It's inspired by a supercomputer from a 1995 anime. the MAGI system from Neon Genesis Evangelion had three supercomputers vote on every decision. Melchior. Balthasar. Casper. no
The anatomy of a second brain that writes itself
One Obsidian vault, three layers, and a loop that files every good answer back into a wiki you never touch the structure: CLAUDE.md + AGENTS.md - the constitution. same rules whether Claude or Codex runs the loop raw/ -
Woke up to this. While I slept, my system watched a League of Legends final
Saw one team up 11–6 at 26 min while the market hadn't moved off the favourite, bet $5 against the favourite, and 3 hours later settled +$9.71. Nobody pressed anything. Not me, not an approve button. One
You have reached the end of the archive
All of Claude Opus 5