Personal behavior suggests chat is increasingly substituting for search, raising deeper questions about how knowledge gets filtered, curated, and audited.
LLMs can expose their own internal conflicts so vividly that even trivial tasks sometimes read like miniature moral dramas.
Using one coding model to refine prompts for another improves clarity, reduces review effort, and increases the chance of getting a usable one-shot result.
Custom instructions let the author turn ChatGPT into a terser, more challenging, more curiosity-inducing collaborator that better matches his own preferred working style.
Using LLMs live in meetings is powerful, but presenting the filtered answer as your own judgment often works better socially than disclosing the exact source every time.