From an earlier video
This is material from an earlier video — the interfaces and features of AI services may have changed since. The idea and the approach still hold.
Most people know ChatGPT as an AI that writes text.
But to use it to the fullest, you need to understand how it works
ChatGPT is based on the GPT model — Generative Pretrained Transformer.
It’s a transformer neural network, pre-trained on huge amounts of text to predict the next piece of text (called a token) with the highest probability.
Example of tokens:
Put simply, ChatGPT doesn’t “know” the truth.
It guesses which text looks logical in response to your request.
It’s not connected to a knowledge base or a search engine — if you don’t give it a source, it doesn’t verify anything on its own.
❓ That’s why it sometimes:
- invents authors that don’t exist
- mixes up facts or dates
- gives generic, obvious advice
Prompt engineering.
It’s an approach where you don’t just ask a question — you describe:
- Who it’s playing (the role)
- What format you want (table, list, checklist)
- What style you prefer (friendly, scientific, concise)
- What context matters (goal, niche, audience)
That’s exactly when ChatGPT starts giving not just an answer, but the one that actually helps you.
Keep going?