ISB AI Founders · portal
1
Week / Day 1 · Wed Aug 5 / AI Foundations
Studio · taught

AI Foundations

You already use AI daily. Now: where it fails, what it does to the world you graduate into, and three scored challenges.

Open this session's deck →
9:20
start
70m
duration
2
exercises
3
things you leave with
Welcome

You are fluent users already; nobody here needs ChatGPT explained. This session is about what fluent users still get wrong (confident lies, agreeable flattery, fake voices) and the bigger picture: what this technology does to careers, truth, school, and power, in the world you graduate into. Then you drive: three scored challenges that are harder than they look.

You walk out with
✓A calibrated distrust: you made it lie and caught it
✓The four parts of a strong ask, used twice
✓A fixed answer that proves prompting is a skill, not luck
Why this session exists

This week AI is your co-founder, and a co-founder who flatters you and occasionally lies needs managing. Learn to break it, cross-check it, and direct it under pressure now, and every session after this gets faster.

The arc, minute by minute
9:20
The honest poll, then the mechanism in five minutes flat
9:30
The traps: hallucination with real stakes, sycophancy, deepfakes
9:40
The big questions: work, truth, school, power, alignment + foot-vote debate
9:55
The anatomy of a strong ask: the four parts, weak vs strong
10:03
The famous fails: the strawberry count and the one-dollar Chevrolet
10:11
Break it, together: students dictate, the room watches it lie live
10:18
Fix it, together: a volunteer repairs it with the five moves
10:25
The bar for the week + the handoff into brainstorming
Exercises
01
Break it, together
No laptops: students dictate the attack, Phillip types it on the big screen, the room watches the model commit to a confident lie.
02
Fix it, together
A volunteer dictates the repair using the five moves: context, constraints, push back. The room names which move fixed it.
prompt · give this to the room verbatim
Attack my thinking. Here is my plan: [X]. Do not tell me what is good about it. Find the three weakest assumptions, say what evidence would prove each one wrong, and rate how likely each is to survive contact with a real customer.
Tools in play
Claude
Materials to prep
ProjectorOne presenter machine, no student laptops
The checkpoint · nothing counts unless it is live
You broke a model and fixed one, and you can name the four parts of the ask that did it.
The handoff · what the next session assumes
Leave withYou broke a model and fixed one: the four parts of a strong ask
DecidedYour two-model pair for the week
Bring nextThe attack prompt: in S2 you run it on your own problem ideas, unprompted
next →Brainstorming problems