I'm getting good results with "overexplain" and "hedging language"

Have you tried these keywords? What has your experience been? Are there more meaningful alternatives? Is this old news?

My results feel like a more specific and nuanced output than most my other keywords. Chat seems to provide followup examples to broad claims, and read a lot more “human” by properly using words like “may”.

I’ve gotten more mixed results previously with words like detailed, specific, humble, nuanced, and honest.

1 Like

Hey there! Welcome to the community!

You’re definitely off to a good start! At the end of the day, any system that works for you is a meaningful method if it generates the responses you want.

I will say, there’s a lot more depth to explore than keywords. It’s not a google search; there’s much more possibility and potential the more detailed and intricate you can make your responses, and use larger conceptual frameworks to get more interesting and desirable results.

What you’re learning is more of the true capabilities of the system on a broader level. LLMs are exceptionally well at “fuzziness”, meaning that it’s really good at exploring the “maybes”, and is quite skilled for when you need responses that require some variance in it (like fuzzy logic).

Understanding linguistic concepts, like semantics and pragmatics also helps to make good prompting techniques, although as mentioned earlier, all of us power users / advanced long-time users simply learn to develop the best techniques that work for us. Everyone is gonna have a slightly different approach. Since my background is in linguistics, I tend to favor applying those concepts to practice a lot when I work with these models. It seems like you’re starting to pick up on similar stuff.

I think the keyword that probably is gonna help you the most here is “complexity”. Complexity has a more specific meaning when it comes to NLP/AI stuff, but one that seems to fit what you’re attempting with the keywords you have. Overexplain might point to a similar degree of complexity in a response, but if you try and ask it to maximize its complexity in a response directly (while maintaining accuracy), you’re likely to get the kind of detail-rich answers you like.

Just remember; you can guide its responses beyond mere keywords. Please, by all means, explore, experiment, and ask questions like this, but don’t limit yourself yourself either! You can experiment with the way in which you ask a query as much as you can experiment with the words in those queries to generate vastly different results.