AI and suffering

Because I like to "suffer", or maybe that's the only way to explain it, I began a doctorate that specializes in GenAI. 

It's got to be a love of suffering to combine that with being a dad to 4 kids, working a job, running business  - and trying to keep my music skills intact! 😅

It is quite fascinating to go behind the scenes, even behind the more familiar world of code to start to poke at "what makes AI tick?"

I'll share some things I am learning over the next couple of days. 

This way, when people are talking about AI at the coffee machine, you'll say something that is not about prompting. 

One question.

So… how do you actually make AI remember stuff?🤔

I’ll be real with you — most people think ChatGPT (or any large language model (LLM)) is this all-knowing machine brain. But here’s the thing: it actually doesn’t know anything about your business.

Ask it about your company’s policies? Blank stare.
Your product manuals? Nope.
The 500-page PDF your legal team insists employees must read? Forget it.

That’s where RAG — Retrieval Augmented Generation — comes in.



Imagine this:

You’ve got a super-smart friend (the AI) who can talk about anything in the world.

Now, you hand them a binder — your private knowledge base.
The catch? They don’t memorize it.
Instead, every time you ask a question, they flip to the right page, pull out the answer, and then explain it in their own words.

That’s R.A.G: Retrieval Augmented Generation

Simple:
1. Retrieve → Find the relevant info in your documents.
2. Augment → Pass that info into the AI as context.
3. Generate → AI responds, using your private knowledge + its general smarts.



Think of an airline. A passenger asks:
✈️ “Can I bring my snowboard bag on the plane?”

The old way: Customer service digs through a giant policy PDF or throws generic “check our website” answers.

The RAG way: AI instantly retrieves the exact line in the baggage policy and replies, “Yes, but it counts as one piece of checked luggage. Dimensions must be under 158 cm.”

Fast. Accurate. Human-like.
And — bonus — the AI doesn’t hallucinate nonsense because it’s tethered to real data with a control known as "temperature". 

High temperature, a more "crazy" and "creative" AI  with higher deviation from the material. 🤪

Low temperature. Chilled, "grounded" AI. 😎


It’s basically open-book exams.
Without RAG → The student (AI) is tested without notes.
With RAG → The student brings the textbook, flips to the right page, and answers confidently.

No teacher on earth would call that “cheating.”
That’s just being resourceful.


👉 In the next post of this series, I’ll break down how you actually build this: from chunking text to using vector databases without needing a PhD.

But before I go — I want your thoughts:
If you had a personal AI that could instantly search through your team’s knowledge base…
What’s the first problem you’d throw at it?

Looking forward to your answers.

I read everyone. 

John