Teaching resource
The best answer is true, concrete, and one layer deeper than the asker already knows. That holds for students, for staff, and for the AI assistants we build.
Beginners and experts are more alike than different. Both want the real reason, both learn from something concrete, and both notice at once when they are talked down to. This guide is for faculty, trainers and managers bringing people into AI: how to pitch an explanation at the right depth, and how to recover when it goes wrong.
Simple, never false
The same truth can be told at every depth. Only the detail changes. Pick a question, then move from a first exposure to an expert answer. Each answer is correct, and each one leaves room for the next.
Why did the AI make that up?
A chat model learned patterns from a huge amount of text. When you ask something, it predicts a plausible answer from those patterns, not from a list of checked facts. If the real answer wasn’t in what it learned, or the question is unusual, it can produce something that sounds right and isn’t. Giving it your own documents to answer from cuts this down a lot.
Adds where the patterns come from, and the most useful fix.
Start one layer above where the learner is, keep every simplification true, and make the next layer easy to ask for.
The difference
A new colleague asks about a wrong answer from an AI tool. Both replies below are accurate. Only one keeps them learning.
Accurate, and closes the door
New colleague
Why did the AI make that up?
Reply
It hallucinated. They all do that. Just double-check everything.
True, and it ends the conversation. They learn that the tool is unpredictable and that asking was pointless. Many stop using it; some stop checking.
Just as true, and opens a door
New colleague
Why did the AI make that up?
Reply
It doesn’t look facts up. It predicts what a good answer would sound like, so when it doesn’t know, it still sounds sure. That’s why we give it our own documents and ask it to cite them. Want to see the difference?
Now they know when to trust it, what to check, and what to ask next.
For anyone who teaches or trains
Give the true answer at the learner’s level, then offer the next layer. Simple and true beats complete and lost.
Try “Short version: it predicts likely words. Want the longer one?”
A question learners can try for themselves sticks far longer than one you answer for them.
Try “Ask it the same question twice and compare the answers.”
It shows that not knowing is where finding out begins, and that experts keep learning too.
Try “I don’t know. Let me find out and follow up with you.”
Anxious beginners and impatient experts both settle faster once the feeling has been named out loud.
Try “New tools are frustrating at first. Want a hand, or a minute?”
If you are the expert, narrate the steps you would normally skip. That is where everyone else learns.
Try “First I checked the source it cited, then I compared it with our policy.”
People learn a new tool fastest on a low-stakes task. Let them practise before the real work.
Try “Before the real task, ask it something you already know the answer to.”
When it gets hard
Try “That’s where the research is still open. What’s your read?”
Try “You’re right. Thanks for catching that.”
Try “Let’s pick this up tomorrow morning. I want to do it properly.”
Try “I got this wrong three times before it clicked.”
Try “Can you find the three questions our assistant handles worst?”
Try “Want more detail, or is that enough for now?”
Ridian’s design rules
An AI assistant sits at the same table. The people it answers range from a newcomer who wants one line to an expert who wants the reasoning. The same rules apply.
Answer at the asker’s level, and keep every simplification true.
Say why a step was taken or a question was asked, in one plain sentence.
Lead with the answer, and keep the detail one click away rather than in the way.
One question per missing piece, in the plainest words, never three versions of it.
“I couldn’t find that in your documents” beats a confident guess.
Respect the question, whoever asks it.
See it live
The student version of this guide puts the ladder next to a live orbit you can push too slow, just right, or too fast. Ridian built it for Ada’s Kids.
Live version at AdasKids.ai/curious-table (opens in a new tab)
Ridian Technologies