Hacker News Reader

Agree, the application of LLM often feels too literal. I like when LLM inform the UX but a layer or two removed for example like you said by giving feedback on things produced by the user or combining multiple things that happened over time and acting on that. That's where the magic lies and where I see users actually be happy having AI in their workflow.

View on HN

I have decided that I don't want an LLM, or Joe at the bar, to write anything for me. What I do use them for, at different times, is feedback. "Here's a draft of X; where is it weak, and where is it strong?" Or, "Give me your best counterargument." I read or listen to what comes, accepting some parts, defending against others, and dismissing the rest. And then close the tab or walk away. There's time and space between those exchanges and the next time I so much as look at the draft.

Reply on news.ycombinator.com