AI + Cybersecurity + Charities

1637 Descartes “No.” 1666 Leibniz “Yes. Calculemus.” 1950 Turing Change the question. 1997 Deep Blue Beats Kasparov. 2017 Transformers “Attention is all you need.” 2022 ChatGPT 100M users in 60 days.

How it works.

The internet.
Computer chips.
A new way to read.
L
LARGE
Read everything humans wrote.
L
LANGUAGE
Words in. Words out.
M
MODEL
A pattern of word to word.

You already use it.

Your phone guessed. That’s the whole trick.

Now — a million times bigger.

1,000,000,000,000 1 10,000 1,000,000 1,000,000,000,000
guesses

Not looking up. Not thinking.

chatgpt.com
ChatGPT
What is the capital of France?
C
The72% A18% The capital81% city12% capital of95% in3% of France88% Europe6% France is92% was5% is Paris.87% Lyon.7% Paris.
Message ChatGPT… ↑
K
Andrej Karpathy @karpathy

Hallucination is all LLMs do.

…dream machines.

…we direct their dreams with prompts.

…looks like a bug, but it’s just the LLM doing what it always does.

8:42 PM · Dec 8, 2023

Arup

Hong Kong · Jan 2024

00:00
mooZ — Quarterly Finance Review
David Thompson
CFO
Deepfake
Maria Santos
Finance Director
Deepfake
James Liu
Regional Lead
Deepfake
Alex Park
Finance Clerk
Only Real Person

$25.6 million.

15 wire transfers. One video call.

Cybersecurity

Stay safe. Five rules.

1
Turn on two-factor everywhere.
Password alone is not enough. One extra tap stops most attacks.
2
Verify money asks out-of-band.
Before any wire or gift card, hang up and call back on a number you already have.
3
Use a password manager.
One password per account. Let the manager remember. You only remember one.
4
Keep everything updated.
Turn on auto-update. Skipping updates leaves doors open.
5
Report the weird, fast.
Strange email, odd call, surprise download — tell someone before you click.

Most AI tools are one prompt on top of a model.

C
Canva Magic Write
AI
Write a flyer headline for our food drive.
Three options — warm, local, specific…
g
Grammarly
Pro
Make this donor email sound warmer.
Here’s a gentler rewrite…
M
Microsoft Copilot
365
Summarize Monday’s board minutes.
Five decisions, three follow-ups…

All ChatGPT underneath. Different wrapper.

Good prompts have five parts.

Role · Objective · Limits · Examples · Steps.

R Role Who the AI pretends to be.
O Objective What you want.
L Limits The fences.
E Examples The specifics.
S Steps Shape of the answer.

Not a script — a shape.

Same five parts — your work.

R Role “Grant writer at a small rural food charity in southwestern Ontario.”
O Objective “500-word letter of intent to the Trillium Foundation — Grow stream.”
L Limits “Plain English. Under 500 words. One specific ask. No ‘empowering,’ no ‘impactful.’”
E Examples “Weekend backpacks: 120 kids, three schools, food bank partnership. $38K last year.”
S Steps “Need → program → one outcome with a number → specific ask → thank-you.”

Your charity. Your numbers. Your ask.

Try it. Keep humans in the loop.

AI drafts. You decide.

01
Start small.
02
Your name goes on it.
03
Save what works.

Thank you.

What will you try first?

AI-drafted · human-decided