👋 New Wednesday, new newsletter!

Do you wish ChatGPT could remember what you teach it?

Most people use AI like a calculator — they ask something, get an answer, and move on. But what if I told you that ChatGPT could become your personal assistant, learning exactly your way of working? 🎯

Today I’ll show you how you can “train” your AI assistant using a simple method. No programming and no paid versions required.

How does AI actually learn? 🤔

Imagine you’re teaching a dog a new trick:

• When it does something right → gets a treat
• When it makes a mistake → gets nothing
• Over time → learns exactly what you want

This is exactly how “reinforcement learning” works in AI. Except instead of treats, it uses “rewards” in the form of positive feedback.

How to use this in practice? 💡

Let’s say you want to improve your customer email responses.

1. First: Give it a basic task “Write a response to a customer asking about product delivery time.”
2. Then: Improve its responses like:
   1. “No! This sounds too robotic. I need a more personal approach.”
   2. “Good! But add how the customer can track their order status.”
3. Key trick: Save good examples - Copy the best responses to a special document (e.g., Notion, Google Docs, or any other…)
4. Next time start a new conversation like: “I want a response in this friendly style: [paste example]”

🔥 It’s like having a customer support assistant that learns your personal approach!

Why does this work?

OpenAI does support “memory” between your conversations (if you enable it), but:

• If you give it good examples → it knows exactly what you expect
• If you give it feedback → it can improve its response
• If you save the best results → you always have a better starting point

Concrete steps to start 🚀

1. Choose one recurring task that doesn’t contain private data (e.g., writing emails, reports, analyses)
2. Start with a basic prompt
3. Improve responses with feedback
4. Save the best examples
5. Use these examples as a starting point for next time

Why is this approach so successful?

Lately, I’ve been thinking a lot about and studying how we can improve our results with AI.

And what I described above is actually very similar to what the best AI programmers do!

When they improve AI models, they use similar processes, just automated.

They use “reinforcement learning” for training and “fine-tuning” for adapting models to specific tasks. Both processes are quite expensive. Let’s look at the costs of fine-tuning:

But we can achieve a similar result — for our specific tasks — practically for free! 🎯

Ok, that’s it for this week. I hope this trick will help you with your work! 😉

💬 If you know AI can help you with a task but don’t know exactly how to start, email me and I’ll help you.

Talk to you soon,
Primož