There's a set of stock phrases I mark down on sight: "in today's...", "with the development of...", "it is worth noting that", "in summary", "crucial", "indispensable", "empower", "grip/lever" (抓手), "closed loop" (闭环), "underlying logic" (底层逻辑), "delve into" (深入探讨). One point off for each.
They all have the same problem. They say almost nothing. But they show up so often in training text that models reach for them constantly.
A few ideas I picked up recently. Writing them down.
One is about failed attempts. Say something cost you money and cost you time. Before you beat yourself up over it, ask: if that time hadn't gone there, was there something of higher value it could have gone to?
If there was, st…
Read more — 28. A Few Excellent Ways of Thinking I Picked Up
DeepSeek recently open-sourced an interesting project: dsh-superagent-prototype.
https://github.com/deepseek-ai/dsh-superagent-prototype
The idea is pretty straightforward: listen for GitHub events, have a Main Agent delegate tasks to Worker Agents, open PRs, and automatically fo…
Read more — 15. When Code Gets Cheap, Where Does a Developer's Value Go
https://zhuanlan.zhihu.com/p/2094373708773311272
[This 10,000-word piece by Claude's lead completely changed how I use AI, prompts included! — Ai 学习的老章]
It did change how I use AI, honestly. One point stuck with me: ambition decides how good your agent's output can be.
The deeper…
Read more — 25. Ambition Decides Whether Your Agent Produces Quality Work
Something here has always struck me as odd.
The more developed society gets, the less anyone writes by hand once they start working. Typing replaced the pen, and handwriting fell off a cliff.
Yet for the first ten or twenty years of our lives — all of them pointed at jobs where w…
Read more — 26. Why Do We Still Have to Learn With Pen and Paper?
New essays on AI, software, and long-term thinking. Delivered to your inbox.
Checking your saved subscription…
Unsubscribe anytime. Privacy