Understanding AI & LLMs

Workshop companion · Florin Bădiță · ai-courses.badita.org

Session —:—

Welcome — keep this open during the session

This companion is your live working surface. The labs reveal on cue, your commitment saves automatically, and nothing leaves your browser. Refresh-safe, offline-safe, private.

Quick capture

Jot anything worth remembering during the session. Auto-saves to your browser. Don't worry about being neat.

✓ saved

How to use this companion

  • Start the timer when the workshop begins. The phase indicator will follow along.
  • Three labs unlock during Module 2. Florin will say "reveal Lab 1", etc.
  • The Prompt Builder tab gives you a starter system prompt to use after the workshop.
  • The Commitment tab is where you write your one-week behaviour change. Type it before you log off.
  • Everything saves locally to your browser. Use the same browser to come back.

Three live labs

Each lab unlocks when Florin says so. The reveal button is a synchronisation tool, not a lock — but please wait for the cue so the cohort moves together.

Lab 1 · 8 minutes

Vague → Specific

🔒

This lab unlocks when Florin says "reveal Lab 1".

Start with this vague prompt:
Write something about marketing.

Rewrite it using RTFC — Role · Task · Format · Constraints.
Pick any context — your real industry, or invent one.
Run it in your LLM. Then paste your rewritten prompt into the workshop chat.
✓ saved
✓ saved
Lab 2 · 12 minutes

Three-turn improvement

🔒

This lab unlocks when Florin says "reveal Lab 2".

Take your Lab 1 prompt or start fresh. Run three follow-up turns:

Turn 2 — change the format or length
Turn 3 — change the tone or audience
Turn 4 — ask What's missing?

Drop one sentence in the workshop chat about what changed.
✓ saved
✓ saved
Lab 3 · 20 minutes · the lab

Your real task

🔒

This lab unlocks when Florin says "reveal Lab 3".

The real task you brought, or pick a starter:
  1. Rewrite a difficult email
  2. Summarise a long document you have open
  3. Draft a meeting agenda
  4. Generate interview questions
  5. Role-play a difficult conversation
Anonymise first — change customer names, financials, anything sensitive.
Use RTFC. Iterate at least twice. Land a usable output.
✓ saved
✓ saved
✓ saved

System prompt builder

A starter system prompt tailored to your role. Use it in any LLM that supports system prompts (Claude, ChatGPT custom instructions, the OpenAI API). Edit, save, refine over time.

Your generated system prompt

Fill in the fields above. Your starter system prompt will appear here.
✓ copied

How to use this

  • Claude.ai → Settings → Custom Instructions → paste it there.
  • ChatGPT → Settings → Personalization → Custom Instructions.
  • Claude API or any developer tool → use it as the system message.
  • It applies to every conversation, so be conservative with strong constraints.
  • Coaching-tier participants: bring this draft to your 1:1 and we'll refine it together.

Your one-week commitment

The single best predictor of whether you'll use what you learned today is whether you write down a specific commitment now. Not "I'll use AI more." Something specific you'll do in the next 7 days.

In the next 7 days, I will use AI to...

✓ saved · you can come back to this anytime

What happens next

  • Tomorrow — recording, slides, and workbook in your inbox.
  • In 7 days — a one-question check-in email from Florin: "Did you use AI for something this week?" Even one sentence is enough.
  • In 21 days — a small prompt-of-the-week pack.
  • In 30 days — invite to a free 30-minute cohort retrospective call.

The frameworks — your reference

The two things every participant has tape-marked a month later.

RTFC — the prompt framework

When you write a prompt for anything that matters, run through these four:

R · Role

Who do you want the model to be? "Act as a senior product manager." "Act as a skeptical investor."

T · Task

What specifically should it do? Not "help me" — "Write a one-page brief outlining the problem, solution, key metrics, and risks."

F · Format

What shape? Bullets, table, JSON, email, three-sentence summary. Always say it.

C · Constraints

Word count, tone, audience, things to avoid. "Under 400 words. No jargon. For a non-technical CEO."

The 3-step verify protocol

When the model gives you a specific fact — citation, statistic, date, quote — anything you'd defend in a meeting:

  1. Ask the model to cite its source.
  2. Check the source exists. Search for it. Open it.
  3. Check the source says what the model claims.

Step 3 catches what step 2 misses.

The 5 prompting principles

  1. Be specific. Vague in, vague out.
  2. Add context. Use RTFC.
  3. Give examples. Show, don't just tell.
  4. Ask for a format. Two seconds saves five minutes.
  5. Iterate. Best outputs come on turn 2 or 3. Useful follow-ups: "Make it shorter." "More formal." "What's missing?"

Privacy — what never to paste

  • Customer or patient PII
  • Passwords, API keys, secrets
  • Company financial data not yet public
  • Proprietary source code
  • Anything covered by NDA

Safer: anonymise first, disable chat history, use Team/Enterprise plans, or run local models via Ollama.

Resources

Curated tightly. Everything here earns its place.

Tools to try (free tiers)

To go deeper (books)

  • Co-Intelligence by Ethan Mollick
  • The Coming Wave by Mustafa Suleyman
  • AI Engineering by Chip Huyen

To stay current (low-noise)

  • Stratechery — Ben Thompson (paid)
  • Import AI — Jack Clark (free, weekly)
  • The Rundown AI — daily, lighter

To go technical (video, free)

  • 3Blue1Brown — neural networks series, YouTube
  • Andrej Karpathy — "Let's build GPT", YouTube
  • fast.ai — hands-on, opinionated

Prompt engineering

Next from Florin