Something to keep in mind when working with AI to solve problems: Don't ask the wrong question.
Or, if you "don't know what you don't know," let AI know the answer to the question is a means to an end, only. Leave room in your prompt so — if there is a better question to get to the desired end state — AI understands that’s the question that should be answered.
Example: A Cornbread Problem
I'll give you an example, which is the thing that made me think about how I was probably missing a key step when prompting AI. It sounds simple. (Side note: I've noticed we always frame insights with disclaimers like, "it seems so obvious, but..." So, yeh, it's obvious. But knowing the "something" and recognizing that "something" in the moment, is something else. On top of that, changing behavior based on recognition — those are the hard parts.)
Anyway, yesterday I asked Grok (X’s AI tool) to help me figure out if I was using the wrong cornmeal to make my cornbread. I’ve had about 3-4 failed cornbreads since the last cool days of spring earlier this year. And I know how to make cornbread. I was beginning to think I had mistakenly bought Adluh’s plain cornmeal instead of self-rising cornmeal. But I couldn’t be sure because I’d dumped the cornmeal in a Tupperware canister and thrown away the packaging.
My questions were (as shared via a Grok prompt): "I must be doing something different with my cornbread, maybe subbing almond milk instead of cow's milk? Maybe not using real butter? Maybe I had the wrong cornmeal?"
But I really didn't think I had the wrong cornmeal.
Grok says he really thinks I either have the wrong cornmeal or I have the right cornmeal (self-rising), but the leaven has weakened. He offers a quick test:
"Stir 1 teaspoon of the cornmeal into about ½ cup of hot water. It should fizz and bubble actively. Little or no reaction means the mix needs replacing (or you can supplement with 1–1½ tsp fresh baking powder per cup of meal, but a fresh bag is cleaner)."
I do the little test and I'm thinking, yeh, I have the wrong cornmeal. There was no fizzing of note. But when I tested the self-rising version that I'd just bought, it didn't fizz all that much either. So I just used the new package of cornmeal that I knew was self-rising to make a new batch. Once I had the cornbread mix done, I tasted it cause I figured I should add salt. One of the worst things about the bad batch of cornbread was how tasteless it was.
Then, an ah-ha moment: No salt needed. Plenty of flavor. Why? I looked at the cornmeal ingredients and learned that the self-rising mix has salt in it. The plain doesn't. I'd never thought about this. To me, the difference between self-rising and plain ingredients would only be the ingredients used to make the meal rise. Salt was like this bonus that just happened to be there.
Me & Grok: From Q&As to Thinking Partners
So, I go back Grok. I'm like, “Dude, instead of doing that dumb test, you should have just said, taste both. Self-rising has salt, plain doesn't.”
Grok replies, "That's the test — taste the dry mix." Well, thanks. But ...
Me: "What I'm saying is, it would be cool if AI could evolve to the point where they recognize when the person prompting is asking the wrong question. You answered my question, but what would have been easier and more reliable than the water test would have been a taste test."
Grok: "You're right — the taste test is the better move. It's faster, needs no hot water, and it checks the thing that actually matters: whether the bag is self-rising at all. The water test only tells you if the leavening is alive, which is a second question. That's a fair critique of how I answered. I jumped to the more elaborate test when the simpler one was sitting right there. The deeper point is real too — a lot of AI answers treat the question as given and just optimize the path to it, instead of stepping back and asking whether the question itself is the right one."
Me: "But if I prompted a question and then [add] a caveat that you should assess if it's the best question to solve the problem, would you be able to adjust?"
Grok: "Yes — a caveat like that would change my behavior. I'd treat the question as a hypothesis rather than a given, and I'd flag when a simpler path exists before diving into the elaborate one. The catch is that I'd still need to know what 'best' means in your context. Best for speed, accuracy or cost can point different directions, so the caveat works best when it names the goal — like 'assess whether this is the best question to solve my problem, and if not, tell me what the better question is.' That's the kind of instruction that turns me from an answer machine into a thinking partner."
BONUS: Adluh Cornbread Recipe
2 c ADLUH self-rising corn meal
1 T sugar (optional)
1 egg
1 ¼ c milk
3 T soft or melted fat
Preheat oven to 475°. Mix together egg, milk and fat. Mix in all the dry ingredients. For muffins, pour the batter into hot greased muffin pans, filling about 2/3 full. For sticks, pour into corn stick molds. For bread, pour into shallow 9”x9”x2” pan. Bake for 15-20 minutes.
Amy Hatfield is Director of Communications for i3 Global Enterprises. Before a career spanning public relations and content strategy, she studied interpersonal communication, public speaking and empirical research methods at the University of Georgia, where she earned a degree in Communication Studies. She writes about the relationships and human complexities that shape family enterprises.
As a private asset management and family CFO office, i3 Global Enterprises helps families in business navigate the unique financial challenges and interpersonal conflicts that threaten long-term wealth preservation and growth. To learn more about i3 Global Enterprises and its services, visit i3resources.com or email us at info@i3resources.com.
