"What does this mean practically? Large language models can often give useful answers to broad questions — questions on topics they’ve seen articulated millions of times — but they become increasingly unreliable as you ask more specific questions."
https://vectorverse.beehiiv.com/p/james-gosling-ask-one-question-about-gen-ai
Mastodon Source 🐘
I've also read similar how LLMs have a disproportionately beneficial impact for novices which tapers off for experts. I suspect experts are asking much more targeted context-specific questions (if they ask them at all) relative to novices.