Claude Opus 4.7 autonomously built a complete Rust text-to-speech engine from scratch—neural model, SIMD kernels, browser demo—then fed its own output through a speech recognizer to verify it matched the Python reference.
AI从零构建完整系统并进行自我验证的能力令人震惊，这展示了AI从代码生成向系统级工程设计的转变，'months of senior engineering, delivered autonomously'这一表述揭示了AI生产力革命的潜力。
I enjoyed this podcast but got the feeling they see PKM as a kind of grueling Fordist production line. The process in your book seems a lot less like a grind and a lot more like fun!
Some of the framing goes back to using the card index as a means of overcoming the eternal problem of "information overload" [see A. Blair, Yale University Press, 2010]. I ran into an example the other day in David Blight's DeVane Lectures at Yale in which he simultaneously shrugged at the problem while talking about (perhaps unknown to him) the actual remedy: https://boffosocko.com/2024/09/16/paul-conkins-zettelkasten-advice/
It's also seen in Luhmann claiming he only worked on things he found easy/fun. The secret is that while you're doing this, your zettelkasten is functioning as a pawl against the ratchet of ideas so that as you proceed, you don't lose your place in your train of thought (folgezettel) even if it's months since you thought of something last. This allows you to always be building something of interest to you even (especially) if the pace is slow and you don't know where you're going as you proceed. It's definitely a form of advanced productivity, but not in the sort of "give-me-results-right-now" way that most have come to expect in a post-Industrial Revolution world. This distinction is what is usually lost on those coming from a productivity first perspective and causes friction because it's not the sort of productivity they've come to expect.
In reply to writingslowly and Bob Doto at https://discord.com/channels/992400632390615070/992400632776507447/1285175583877103749<br /> Conversation/context not for direct attribution