How I Use AI — And Why You Should Never Trust It Blindly

I genuinely enjoy working with AI, and everything I share here comes from my own experiences, curiosities, surprises, and frustrations — not from any attempt to teach a general “how to use AI” lesson.

The Standard Conversation About AI

I recently attended the presentation of a world-renowned researcher. Someone asked him about the use of AI in research, and his answer was the most diplomatic you can imagine. AI is immensely helpful, he said, but it also hallucinates, and the links it provides sometimes lead nowhere or to completely wrong sources.

This is the standard answer professionals give when asked about AI: it hallucinates, it’s not precise, double-check everything.

I, however, don’t need to be diplomatic. I can be honest: I love using AI as it is.

Ironically, its weaknesses make it more human. It is not perfect, which means we need to show some intelligence too.

Writing Good Prompts: The First Skill You Need

When I write a prompt, I think it through carefully. I always say I try to explain the issue to a 5-year-old. I ask my question precisely and in detail so I get a high-quality answer.

AI is powerful support — but only if you use it responsibly.

If you take ChatGPT’s answers as unquestionable truth, shame on you.
If you submit assignments with links you haven’t checked, shame on you again.

And to use it responsibly, you need to understand how it works.

A Real Example: When ChatGPT Pretended to Translate

A small example: a friend received a four-page Hungarian legal document and asked for my help. I speak Hungarian and Slovenian, but I didn’t feel like bothering too much with it. So I fed it to ChatGPT and asked for a translation into Slovenian. A perfect-looking “translation” appeared in seconds.

Before sending it to her, I skimmed it — and immediately noticed statements that exist in Slovenian law but absolutely not in Hungarian law.

That’s when I realised what had happened: ChatGPT didn’t translate.
It searched for similar Slovenian documents and simply served content from those, neatly placed into the original paragraph structure.

It pretended to translate.

This is why you must evaluate whether what you get is reasonable.
If it looks reasonable, then check it meticulously.
And always verify the sources — especially with AI hallucinations, which are confident, fluent, and sometimes completely wrong.

Using AI in Research: A Great Start, But Never the Full Picture

Another example: ChatGPT is truly helpful for research because it saves you from scrolling through hundreds of papers on Google Scholar. It can offer key articles on a topic immediately — but often not the full picture. Sometimes it doesn’t even show the most important research.

Still, those first articles you receive already contain reference lists, and nothing beats that old-fashioned academic detective work: follow the references and chase the trail.

AI can help you start, but it cannot replace the thinking part.

Evaluating AI Answers: The Second Skill You Need

People often say the real skill in AI is writing a good prompt.

True. A good prompt shapes the quality of the answer.

But an equally important skill is evaluating whether the answer makes sense.
You must be able to tell if something is factually correct or at least reasonable.

When we receive different kinds of work (school assignments, strategy proposals, article drafts), our next challenge is to be able to tell whether something is AI-generated or human-written.

One major question is still wide open:
Where do we draw the line between AI-generated work and human-generated work?
We will need to define this sooner rather than later.

Full Disclosure

With full disclosure, this blog was written in the following way:

I dictated the full text verbally to a dictation app, and then I copied it into ChatGPT with the following prompt:

“This is a raw dictation of an article. Rephrase and enhance it stylistically and grammatically, but do not modify the content. Do not remove and do not add anything. Make it SEO-friendly for my website as a blog. Use the correct SEO keywords, place them in the text, and after that prepare a full SEO meta description.”

So, is this blog human- or AI-generated?