Alessandro Appiani retweetledi
This dad making a sandwich with his kids' instructions is basically what happens when people use an LLM, and something goes wrong.
Kid says, "Put the peanut butter on the bread.” Dad stabs the jar into the loaf. Kid loses it. But he did what was asked.
Inexperienced people keep running into this with DAX formulas. It runs, the number comes out, looks fine. Nobody checks it. Two weeks later, someone asks why the report doesn't match finance, and the answer is often the same: the formula did what you told it, not what you meant.
People think the hard part of using AI is making it work. It's not. It's realizing you left something out of the instructions that you thought was obvious. A person would have asked, "Wait, what do you mean by last year?” The machine (often) just picks one and goes.
The annoying part is that to write a prompt that gives you the right formula, you already need to understand the problem well enough to write it yourself. At that point, you're saving time on typing, not on thinking.
I don't have a good answer for what someone starting from zero should study right now. The fundamentals still matter, though. If you understand the concepts and the semantic model, you can check what comes out.
If you can't check it, you're the kid watching the sandwich fall apart.
Video credits: "Exact Instructions Challenge" by Josh Darnit on Youtube.
English
