Re:Use AI like a $10M consultant instead of like a search en...
by gumboshoes on Sunday June 15, 2025 @10:12AM
(#44282386)
I think there's some straw-manning happening in the Reddit comment. I don't know anyone who uses LLMs daily who writes such facile prompts. It's simply the same advice as everyone else is giving: GIGO. Garbage in, garbage out. The quality of your results depends on the quality of your prompt. And even then it depends on which LLM you're using.
If there's any prompt-writing advice that is less commonly seen that is worth sharing more than anything in OP's remarks, it's this: instead of asking for something or giving commands, start your LLM-related work with an open query like: "Here's {a collection of facts and circumstances outside my usual skillset}. What do you think?" With the big LLMs you typically get big-picture responses of the good vs. bad variety, and most of the time, I get ideas I hadn't considered. It's the rubber ducky tactic, basically, but if you're careful not to inject much presupposition in the original prompt, you actually hear back from ducky more than you told it.
If there's any prompt-writing advice that is less commonly seen that is worth sharing more than anything in OP's remarks, it's this: instead of asking for something or giving commands, start your LLM-related work with an open query like: "Here's {a collection of facts and circumstances outside my usual skillset}. What do you think?" With the big LLMs you typically get big-picture responses of the good vs. bad variety, and most of the time, I get ideas I hadn't considered. It's the rubber ducky tactic, basically, but if you're careful not to inject much presupposition in the original prompt, you actually hear back from ducky more than you told it.