Last session: AI is a confident guessing machine that sometimes makes things up.
So you stay in the driver's seat. Today: how to ask, and what to ask it to do.
0:45 · BRIDGE. Pick up Bill's closing line: "He just showed you it's a finish-the-sentence machine that rolls dice.
You're the team captain. A captain gives clear orders and picks the right jobs — that's the whole session."
Keys: ← → navigate · N notes · T timer (press now!) · B blank.
The plan
Two questions to master
1
How do I ask?
A lazy question gets a lazy answer. A clear question gets code you can actually run. → the 5-step prompt.
2
What do I ask it to do?
Some jobs it's amazing at. Some jobs are yours and yours alone. Learn to tell them apart.
0:45 (cum 1:30). Two acts. "First we fix HOW you talk to it. Then we sort out WHICH jobs to hand it.
Both matter — a great prompt for a job you shouldn't outsource is still a bad idea."
Act 1 · The recipe
The 5-step prompt
1 · Setting
Your gear SPIKE Prime + Pybricks
2 · Task
What to do "drive forward 30 cm"
3 · Limits
The real facts ports, wheel size
4 · ✅ Do
Good example "use DriveBase"
5 · ❌ Don't
Bad example "not EV3"
Steps 3–5 are you handing over the open book: your real robot, so it stops guessing from fuzzy memory.
1:30 (cum 3:00). This is Bill's "open book vs fuzzy memory," turned into a checklist. Walk the 5 chips fast.
Punchline: steps 3-5 are literally pasting in the truth about YOUR robot. Next 3 slides: watch it in action.
Example 1 · Robot code
"Write my robot's code"
You type
AI gives back
🎨 SettingTaskLimits✅ Do❌ Don't
2:30 (cum 5:30). Start on LAZY (default). "Looks like code! Would it run? …No — that's EV3, wrong imports, made-up ports."
Click 5-STEP: same question, now with setting+ports+wheel+do/don't. "THIS one you paste and it runs."
Callback: "You gave it the open book — your real hardware."
Example 2 · Debugging
"My code is broken"
You type
AI gives back
🎨 SettingTaskLimits✅ Do❌ Don't
2:00 (cum 7:30). LAZY = it has no idea what your robot is, so it gives a generic checklist. GOOD = real code +
the EXACT symptom ("overshoots ~15° every run") → a specific fix you can TEST. Same as Bill's helpful-vs-bossy game:
good input earns a helpful answer.
Example 3 · Innovation Project
The Innovation Project
You type
AI gives back
🎨 SettingTaskLimits✅ Do❌ Don't
2:30 (cum 10:00) ⏱ CHECKPOINT. THE IMPORTANT ONE. LAZY = cheating AND a generic idea that isn't yours — judges see
through it instantly. GOOD = your idea stays yours; AI plays skeptical judge and makes you defend it.
Land it: "Your ideas in, its polish out. Never the reverse." (Bill's ORIGINALITY pillar.) NEVER CUT.
Act 2 · Game
Whose job is it?
✅ Great job for AI
⚠️ This one's on you
2:30 (cum 12:30). Read each chip, class votes "AI job or your job?", then click it to see where it lands + why.
The ⚠️ ones lean on what Bill proved: it hallucinates (don't trust facts), and originality/effort must be yours.
"Homework" one = the honest cheating talk. Keep it moving.
Act 2 · Bonus
4 tricks worth stealing
🔟 Explain
Confusing rule or manual? Give it the text, ask for plain words.
I'm a 5th-grader in FLL.Explain this rule in plain words, with an example → [paste the rule]
🎤 Challenge
Practice your presentation Q&A before the real judges.
You're an FLL judge; my project is [topic].Ask me 5 tough questions.One at a time.
🗒 Clean up
Dump your meeting scribbles, get an organized entry.
Turn these notes into a notebook entry.Headings for Goal, Tried, Result, Next → [paste]
🎲 Brainstorm
Great for lists of options — you make the final call.
Give me 10 team-name ideas.Theme = bears + robots, short, kid-friendly.
🎨 SettingTaskLimits✅ Do❌ Don't
1:30 (cum 14:00). Quick tour — these are the everyday wins. Note trick 4 ends "then YOU pick" — brainstorming is an AI
superpower, but the choice stays human. CUT if behind: keep #1 and #2, mention the rest.
Act 3 · Wow demo
It can build you a tool
You describe the tool
Make an interactive HTML calculator for LEGO gears.I pick the driver and driven gear by number of teeth,and it shows the gear ratio, plus whether we get more speed or more torque.
What it built — try it!
Driver
12t
Driven
36t
ratio · driver : driven1 : 3
speed0.33×
torque3.00×
🎨 SettingTaskLimits✅ Do❌ Don't
▶ Change the teeth and watch it recalculate — one prompt built this.
1:30 (cum 15:30). "AI doesn't just write text — it builds working tools. One prompt, and you get a calculator
you can actually click." Let a kid press the buttons. Teacher tidbit: real LEGO gears come in 8/16/24/40 teeth
(straight-cut) and 12/20/28/36 (double-bevel) — the default 12t + 36t is a real meshing pair. The +/– stepper can
land on counts LEGO doesn't make; that's fine, the math still holds.
Act 3 · Wow demo
Outline → slides
You type
Turn my bullet points into a clean slide deck.One idea per slide — keep my exact words.
# How LEGO gears work
- two gears meshed = a gear train
- small drives big → more torque, slower
- big drives small → more speed, weaker
- count the teeth to get the ratio
- example: 12t + 36t = 1 : 3
It formats real slides
gears-deck.html
Gummy Bears Robotics
How LEGO Gears Work
⚙️⚙️
Torque vs Speed
small → big = more torque, slower
big → small = more speed, weaker
Find the ratio
12t + 36t
1 : 3
🎨 SettingTaskLimits✅ Do❌ Don't
▶ Your thinking stays yours — the bullets. AI just does the formatting.
1:00 (cum 16:30). "Your brain does the thinking — the bullets. AI does the boring formatting.
Your ideas in, its polish out — same rule as the Innovation Project."
Act 3 · Wow demo
Words → flowchart
You describe it in plain words
Draw a flowchart.The robot reads the color sensor. If it sees the black line, it stops. If not, it drives straight. Then repeat.
It draws the flowchart
🎨 SettingTaskLimits✅ Do❌ Don't
▶ You explain the logic in one sentence; it draws the picture.
1:00 (cum 17:30). Flowcharts + wiring diagrams = huge for the engineering notebook and presentations.
"You explain the logic in one sentence; it draws the picture."
Act 3 · Bonus showcase
AI builds tools for every subject
🤖 Robotics
Plan the run · Degrees → distance
💻 Computer Science
Bubble sort
🧪 Chemistry
Heat it up · states of matter
🎲 Math & Statistics
Random → patterns
⚡ Electronics & Optics
Resistor network · Light bending
🎵 Music
Compose a tune
Same 5-step recipe. Any subject. One clear prompt → a working tool.
Optional bonus section. 8 demos across 6 subjects — cherry-pick or skip if short on time. The point: prompting isn't just for robots.
Candidate 13 · Strategy
Plan the run
You type
We have 2 minutes 30 seconds and 5 missions on the mat.Make a route planner: I click missions in order and it drives the robot.Add up driving + stop time, and let me bend the path around obstacles.
⏱ Stop / mission 3s
Click missions on the mat →
What it built — click missions; ✏️ on → click gaps to bend around obstacles
🎨 SettingTaskLimits✅ Do❌ Don't
▶ Strategy in numbers: driving isn't the whole story — time spent AT each mission counts too. Bend around obstacles and watch the 2:30 clock.
Pitch: replay is 3× speed but reports real time (20 cm/s + the stop seconds). Turn on "Add bends" and click empty mat to route around the gray obstacles; raise "Stop / mission" and watch the total blow past 150 s. Ask "can we afford M3?" and let the numbers answer.
Candidate 6 · Calculator
Degrees → distance
You type
Our robot has LEGO wheels and a motor that turns in degrees.Build a calculator: pick the wheel size and the motor degrees, show how far it drives.Animate the wheel rolling along a ruler in cm.
Wheel:
Degrees:
What it built
🎨 SettingTaskLimits✅ Do❌ Don't
▶ The math kids avoid: distance = π × wheel × degrees ÷ 360 — now they can SEE it. Pairs with the gear slide.
Pitch: the #1 real calculation in FLL robot programming. 56 mm and 88 mm are real LEGO tire sizes.
Candidate 2 · Sorting
Bubble sort
You type
Show bubble sort working on a row of bars.Highlight the two it compares, swap only when out of order, and let me step or play.
What it built
🎨 SettingTaskLimits✅ Do❌ Don't
▶ The simplest sort: compare neighbors, swap if out of order, repeat. The biggest value "bubbles" to the end each pass.
Pitch: each pair takes TWO steps — orange = comparing ("is A bigger than B?"), then either red = swap or teal = keep (already in order); green = locked in place at the end. Step through slowly so kids predict swap-or-keep before the act phase. Ties into Big-O (this is O(n²)).
Candidate 5 · States of matter
Heat it up
You type
Animate water particles as I drag a temperature slider.Packed and shaking when solid, flowing as liquid, flying free as gas — and label the state.
What it built
🎨 SettingTaskLimits✅ Do❌ Don't
▶ Same molecules, more energy: heat doesn't change the water — it just makes the particles move faster and spread apart.
Pitch: water melts at 0 °C, boils at 100 °C. Watch the particles break free as you cross each line.
Candidate 15 · Probability
Random ≠ patternless
You type
Roll two dice over and over and chart the totals.Add a “roll ×100” button so the pattern shows up — 7 should win most.
🎲
What it built
🎨 SettingTaskLimits✅ Do❌ Don't
▶ Session-1 callback: the AI "rolls weighted dice" — and even dice make patterns. 7 wins because it has the most ways to happen.
Pitch: ask "which total will win?" BEFORE rolling. Roll ×100 twice — the triangle emerges every time. Ties randomness (Session 1) to data (this session).
Candidate 5 · Resistor networks
Series vs parallel
You type
Two resistors in a circuit.Show how they combine in series versus parallel.Let me change each resistor and show the total with the formula.
R1: 100 Ω
R2: 200 Ω
What it built
R1 + R2300 Ω
🎨 SettingTaskLimits✅ Do❌ Don't
▶ Counterintuitive twist: put resistors in parallel and the total drops below either one — more paths for current.
Pitch: series adds (R1+R2); parallel is product-over-sum and always smaller than the smallest. The "more lanes" analogy lands well.
Candidate 7 · Optics
Light bends into water
You type
A light ray going from air into water.Show how the ray bends — refraction.Let me drag the angle, and use the real numbers (air 1.0, water 1.33).
▶ Why a straw looks broken: the ray bends toward the normal entering the denser water.
Physics: Snell's law n1·sinθ1 = n2·sinθ2 → θ2 = asin(sinθ1/1.33). Always bends toward normal into water.
Candidate 9 · Creative
Compose a victory tune
You type
Make a little music grid to compose a victory beep-tune for the robot's speaker.Click notes, press play, and give me a preset tune too.
What it built — click cells
🎨 SettingTaskLimits✅ Do❌ Don't
▶ The sleeper hit: Pybricks really does hub.speaker.beep() — compose here, then AI writes the beep code.
Pitch: needs speakers on. Bridge line: "AI can turn this grid into the actual Pybricks code that plays it on the hub."
Your turn · Game
Build a prompt that works
Your prompt so far
2:00 (cum 18:30). Whole-class build: each row has a STRONG choice (green) and a VAGUE one (red). Pick the vague ones on purpose first —
the prompt underlines them and the badge stays "⚠️ the AI still has to guess…". Then swap to the strong picks and watch it flip to green:
"✅ the AI can actually run this." That contrast IS the lesson. Reset and let a different kid drive.
Trivia time
Did it stick?
~2:30, game-show pace — this IS the recap. Read the question, kids shout A/B/C, click the crowd's answer.
Wrong picks shake red and stay — let them regroup and pick again. The green "why" line is the recap; read it aloud.
Tight on time? Do 3 questions and jump to the big questions.
No right answers
Chew on these
?
If the AI writes your Innovation Project, whose project is it — and could a judge tell?
?
It sounds just as sure when it's wrong. How would you catch a mistake on a topic you don't know?
?
One job you'd happily give AI forever — and one you'd never hand over?
?
Using AI on homework: cheating, studying, or both? Where's the line?
Flexible — the Q&A springboard. Pick ONE, 30 sec think-pair-share, then take answers. No right answers here.
Closer line: "You now know how to use this thing better than most adults. Use that responsibly."
You're still the captain.
ASK CLEARLY
Give it the open book: your real gear, facts, do's & don'ts.
CHECK IT
Confident ≠ correct. Test the code. Verify the facts.
KEEP IT YOURS
Your ideas in, its polish out. Never the reverse.
gummybearsrobotics.org
Now go build something.
0:30 (cum 19:00). DONE — Q&A after. Three rules, mirror Bill's curiosity/judgment/originality.
Seed Qs if silent: "What's one job you'd NEVER hand the AI?" / "How would you fact-check its homework answer?"
Driving the AI — take-home card
The 5-step prompt
Setting — your gear (e.g. "SPIKE Prime + Pybricks MicroPython")
Task — what to do ("drive forward 30 cm, then turn right 90°")
Limits — the real facts (ports A & E, wheel Ø 56 mm, axle track 112 mm)
✅ Do — a good example ("use PrimeHub + DriveBase")
❌ Don't — a bad example ("do NOT use EV3 or spike/hub modules")
Steps 3–5 hand it the open book — your real robot — so it stops guessing.
Whose job is it?
✅ AI's job: generate/debug/explain code · charts from your data · diagrams & flowcharts · polish your slides. ⚠️ Your job: do your own homework · fact-check everything · invent your own Innovation Project idea · keep personal info private.