Generative AI Fundamentals�Meet your oh-so clever, but ‘doh’ co-worker…

Presented by: Natalia Rosa | Founder & CEO, Big Ambitions

• AI refers to intelligent behaviour exhibited by machines.
• AI systems learn from data to make predictions, decisions, or generate content.
• Generative AI can create new content: text, images, videos, code, etc.
• Revolutionising how we work: from writing to translation to idea generation.
• Major players include ChatGPT, Gemini, Claude, and others.

​

​

​

Powered by Large Language Models

These AI tools are powered by something called LLMs, or Large Language Models. You don't need to remember the term — just think of them as the engines behind the tools.

1

ChatGPT (by OpenAI)

Great for writing content, summarising, asking questions, brainstorming, and building custom assistants (called GPTs).

2

Claude (by Anthropic)

Known for being really good at reading long documents and writing clearly. You can personalise with settings and long-term memory.

3

Gemini (by Google)

Integrated with the Google tools you already use. You cannot personalise in settings but can set up Gems (Projects).

4

CoPilot

Your AI layer across Microsoft 365 — not a separate tool, but one that lives inside the apps you already use:

Word, Excel, Outlook, PowerPoint, and Teams. You cannot personalie with long-term memory.

5

And there are others

Such as Llama, Grok and DeepSeek.

Who are the head honchos?

The quality of any outcome by another depends on the clarity of our instructions.

​

Vague requests lead to vague results.

Specific, structured communication creates sharper, more useful outcomes.

Today’s exercise will show how this applies – both with people and with AI.

​

Clear instructions (prompts) → better results. The same is true for AI.

​

This is the least capable AI you will ever use…

Stages of

Artificial Intelligence

It’s only getting smarter…

The evolution MidJourney

/imagine High quality photograph of a young Japanese woman smiling, backlighting, natural pale light, film by camera

Prompt:

Step aside MidJourney…

Where AI models fall short

• Understanding vs. Mimicking
• Contextual and Complex Reasoning Challenges
• Reliance on Training Data
• Bias in AI responses
• Responding in a literal way

• Layer 1: Assistant – AI does tasks you hand over (emails, reports, summaries).
• Layer 2: Identifier – AI helps you spot where it can add value (what to offload, what to keep).
• Layer 3: Thought partner – AI works with you to unpack how to use it, not just what for.

​

Short-term memory vs long-term memory

Short-Term (Regular Chat):

Long-Term (GPTs, Projects & Gems):

Why it matters:

• Short-term chats are fine for quick tasks.

• But if you want smarter, faster, more personalised help – long-term memory is where the magic happens.

• Like talking to someone who nods along but forgets what you said five minutes ago.
• After a few prompts, it loses track and starts making things up.

• Like having a teammate who remembers your goals, your tone, and your history.
• It holds onto your instructions and your own information (your files, examples, and way of working) – so outputs stay consistent and relevant.

Personalising your ChatGPT assistant

The Perfect �Prompt

My ‘Recipe’

• R - Role: Assign the AI a specific role.
• Example: "You are a market analyst."
• I - Instructions: Clearly state what you want the AI to do.
• Example: "Analyze the current trends in the renewable energy market."
• S - Specifics: Provide detailed steps and context.
• Example: "Start by gathering data from the last quarter, focus on solar and wind energy sectors, compare growth rates against fossil fuels."

• E - End goal: Define what success looks like.
• Example: "Create a detailed report that identifies growth opportunities and risks. The report should be comprehensive enough to assist in investment decisions."
• N - Narrowing: Specify the format and style of the output.
• Example: "Present your findings in a structured report, using British English, in a formal tone, limited to 800 words, include charts and graphs.�

• We censor ourselves when we type
• We edit, overthink, and try to "sound right" — which can dilute clarity
• You'll say more than you'd write — and that's a good thing
• Voice prompts tend to give richer, more usable context
• And editing the follow-up outputs are easier

​

Dictation (Asynchronous Voice Prompting)

Hit the mic, speak your thoughts → AI transcribes and repliesGreat for messy thoughts, rambling ideas, or first drafts

Voice Mode (Synchronous Prompting)

ChatGPT talks with you in real time. More like a coaching session than a task

Projects = The House

GPTs = A Room with a Specialist Inside

• Think of a project like a full house.
• Each room stores different types of info and has different functions.
• Every conversation you have inside that house has access to everything you have added – the full family history.

• A GPT is like walking into one room – say, the social media room – and talking to the expert who only does that.
• It’s focused, efficient and knows that topic inside out.
• It doesn’t allow you to save your conversations although it does remember them.

Speak instead of type

• Use voice mode or dictation – it helps you explain naturally, like talking to a colleague.

• You'll give more detail and context when you're not worrying about getting the wording perfect.

Let ChatGPT do the heavy lifting

• Once you've explained what you need, ask it to write the instruction for you.

• It's great at turning messy thoughts into a clear, structured brief.

Don't worry about getting it perfect the first time

• Review what it gives you.

• Tweak, add, or ask it to revise until it sounds just right.

Your AI Assistant gets better the more you work with it, but only if you help it learn!

You can always update your GPT or Project

• Edit the instruction if your needs change.
• Upload new or better riles anytime – tone updates, new examples, background information.

Need to share something just once?

• Add it in a single chat (short-term memory)
• Useful for one-off requests like “Use this client email to write your email response”

Best practice: Close the loop

• Once you’re happy with the final version of whatever has been produced, paste it back into the GPT or Project chat window and say: “This is the version I used – just for your reference and learning – no need to respond”
• That helps the AI learn what “good” looks like.

Practical Considerations

• The copyright belongs to AI, unless you’ve been instrumental in the creation of the content
• Privacy (sensitive information)
• It is not possible for AI to plagiarise. They predict the next word and do not copy large portions of text from the Internet
• It can make things up – hallucinate – so don’t accept everything as fact.

Generative AI and Data Privacy

• AI tools don’t “know” your data unless you share it
• Public tools may store or reuse inputs → risk to proprietary info & client data
• GDPR requires protection of personal data (names, emails, bookings, payments)
• Risks: loss of client trust, legal penalties, exposure of business IP

​

How to safeguard your data

• Use enterprise/business AI tools with GDPR compliance
• Never paste personal, financial, or contract details into public AI
• Safe for: brainstorming, generic copy, summarising, translation
• Anonymise data before use where possible
• Create simple team rules: what’s in / what’s out

You Don't Need to Master AI. You Just Need to Start

Start small

One prompt. One task. One result. That's all it takes to build momentum.

Stack your skills over time

You'll get faster, clearer, and more confident — just like training a team member

Name your assistant

Mine's called Allie. When you give your AI a name, you'll brief it like a person — and it'll respond like one

Want to get started?

Have a conversation with your version of Vusi. Tell him about your day to day and ask him to ask you questions that you can answer that will help him understand where he can make the greatest impact on freeing you up to do your highest value work.

natalia@bigambitions.co.za

Linkedin: @nataliarosaza