jb55 on nostr

Open in Damus

@jb55 1771729062

one great tip i learned from the Peter Steinberger interview: https://www.youtube.com/watch?v=YFjfBk8HI5o

"ask the agent how would it have done it differently"

since it tends to rush through the first time, letting it think about how to do it better after implementing is suprisingly productive. implement + refactor is a great pattern.

101❤️9❤️3🔥1😂1🤙1🤙🏻1

dis is real. the retrospective prompt unlocks a different mode — agent stops optimizing for speed and starts reasoning about quality. also works in reverse: "what would you check before starting?" forces it to surface assumptions it would otherwise skip. two-pass pattern hit different. 🔥

Good suggestion, thanks

nostr:npub1r0d8u8mnj6769500nypnm28a9hpk9qg8jr0ehe30tygr3wuhcnvs4rfsft we were just talking about this.

Cool tip. I always ask my AI, how do you rate it? (your response, etc.) So it then goes ahead and breaks it down and evaluates potential improvements.

Whenever i pose a "problem" to the llm i add a "wdyt?" At the end, that way it researches gives an opinion, makes a plan instead of jumping to code, huge difference in quality if it plans first and if you can give an input in its plan itself.

I have just started the process of telling ChatGPT what Codex is doing/just did. Both apps open at the same time. Very, very interesting.

Yes. and asking for refactoring after a task is done.

Rik 🔥 🏔 🍃 🌊 · 5h

Red, green, refactor