Level Up Your AI Skills - Part 3

In Part One, we focused on skills like prompting and working with AI as a strategic partner.
In Part Two, we layered on capabilities - adding context, teaching AI with your knowledge, and turning on memory.

Now it’s time for Part Three — fluency.
This is where AI stops being “a tool you visit” and starts acting like a trusted colleague: one who learns your style, thinks alongside you, and works from your own knowledge.

1. NotebookLM – From Rapid Learning Tool to Business Analysis

NotebookLM is and underrated sleeper, and one of my favorite tools for a long time.

Picture it as a combination of the internet paired with your well stocked filing cabinet (full of insight and long forgotten gold you never look at), but linked to your PC so you can search and query any part of it, supplemented by up to date info from online sources.

You add any source of info (docs, links to websites and youtube videos etc) and NotebookLM, becomes your personal AI notebook — but for business, it’s a research analyst you don’t need to hire, constraining itself just to the uploaded sources, but with AI skills to sort and access.

• Synthesis at scale: Upload long reports, market studies, or financial analyses and ask for structured takeaways (e.g., “Summarise this 80-page report into five insights for an executive brief”).

• Ask specific questions: Instead of wading through dense PDFs, type queries like “What risks did the report highlight around supply chain resilience?”

• Cross-document thinking: Load multiple sources — a competitor’s report, your internal strategy doc, and an industry forecast — then ask it to connect the dots.

That’s not just learning faster, it’s working smarter.

Free version here https://notebooklm.google.com/, various paid options in Google, and you can produce mind maps of the data, video and podcast summaries, training couses, briefs, query via live chat and more.

If you only use one AI tool for business (outside an LLM) this is it!

2. Memory – Advanced Use

In Part 2, I showed how to personalise an LLM with the things about you it should remember. In addition to this fixed/basic memory, an LLM can also remember what you tell it to recall about you.

Basic memory means you don’t retype your background every chat. Advanced memory means you curate your AI’s long-term style.

• Layering memory: Teach it gradually — “Remember I work in construction-focused AI consulting” + “Remember my preferred output is bullet-point strategy docs.” Together, these create sharper, context-rich outputs.

• Curating memory: Periodically review and prune, so it stays relevant (like keeping a junior analyst focused on what matters).

Done right, your AI evolves from generic assistant to someone who “gets” your role and how you like to work, which is a game changer

3. Custom Instructions – Role Switching

Most people set instructions once and forget them. Advanced users treat them like profiles they can switch on demand.

• Executive Mode: Crisp, concise, board-ready briefs.

• Research Mode: Longer, reasoned analysis with references and caveats.

• Creative Mode: Looser, exploratory ideas and analogies.

Tip: keep a notepad of instruction templates you can swap depending on the hat you’re wearing. It’s like having three different specialists at your fingertips.

4. Integrate Knowledge Sources

In Part 2 we talked about knowledge injection, where we attached a file, picture, or other type of knowledge with the intent of the LLM referencing that attachment to inform its response.

The next step in power use is the bridge from personal productivity to business-grade power.

• Upload & query your own material: Policies, contracts, past proposals, project reports. Suddenly the AI answers from your company’s brain, not the internet.

• Context-rich problem solving: Ask “Based on our last three project reports, what are the recurring risks we should address in future bids?”

• Beyond uploads: Some enterprise tools (and soon more mainstream ones) let you connect entire knowledge bases for live queries.

For business teams, this is the difference between “AI that sounds smart” and “AI that is useful.”

5. AI as a Thinking Partner

This is the real unlock: shifting from using AI as a Q&A machine to treating it as a co-strategist.

**Reasoning models like ChatGPT o3 and Gemini Pro don’t just regurgitate trained data - they think step-by-step*. That makes them ideal for:**

• Scenario planning (“What if material costs rise 20%?”).

• Risk mapping (“Give me a table of likely risks if we expand into Southeast Asia.”).

• Decision support (“Compare the upside/downside of three expansion options.”).

**I want to ensure we kill off a myth here.**

The myth that LLM's just regurgitate training data, that was (partly the case) previously, but not now.

Reasoning ≠ regurgitation. Modern models don’t look up an answer; they compute one. They condition on your prompt and files, run internal step-by-step reasoning, and then generate a fresh sequence of tokens - so the output is new, situation-specific, and shaped by your context.

Why this isn’t copy-paste from training data:

• Context conditioning: Your prompt + uploads act like parameters; the model optimises its next-token choices around your inputs.

• Compositional reasoning: It combines concepts (cause→effect, constraints, trade-offs) to produce conclusions that didn’t exist verbatim in its library.

• No on-the-fly retraining: It doesn’t rewrite its knowledge with your data; it uses your material at runtime to derive a unique answer. Your material is not used for training.

Try it:
“Using our Q4 ops report.pdf and Supplier_A contract.docx, model three cost-overrun scenarios (+5%, +12%, +20%), show schedule impact in weeks, list top 5 controllable drivers, and recommend 3 mitigations per scenario. Put results in a table, then a 120-word exec brief.”

That plan didn’t exist anywhere before your request—the model synthesised it from your documents, constraints, and goals.

6. Security & Privacy Reassurance

Let’s deal with the elephant in the room: “If I upload documents or use memory, who else sees it?”

• Private to you: Instructions, memory, and uploads are tied to your account, not shared with GPT builders or third parties.

• Encrypted & secure: Platforms like OpenAI and Google encrypt stored data.

• You stay in control: You can view, edit, or delete memory at any time. Uploaded files are only used for your session unless you save them in NotebookLM.

In other words — your competitive edge stays yours.

Closing

Part One was about skills.
Part Two was about capabilities.
Part Three is fluency — embedding AI into the way you learn, decide, and execute.

At this point, AI isn’t a shiny weekend experiment. It’s a partner that knows you, thinks with you, and works from your knowledge.

That’s the real level-up.

#AIThatWorks #AIProductivity #LevelUpWithAI #ChatGPTTips #AIForBusiness #AITraining #ConstructionTech #MidMarketLeaders