The dark side of ChatGPT...

ChatGPT came out 559 days ago, and we practically started using AI.

It has helped us create some incredible stuff for sure.

But what can go wrong if we give away too much about ourselves?

Let me make it clearer…

You train AI on certain details to work better for your needs. But can it keep those details safe?

The answer to that is yes and no.

Big corporations can apply certain measures to prevent the closed-source model from leaking anything.

But when it comes to custom GPTs, they still can’t keep the training data to themselves.

Yes, I’m going to show you in a bit.

It’s not that GPTs aren’t developed well. They are. But it’s a complex scenario to train AI and not expect it to keep quiet about what’s inside it.

It’s like telling a child something and when a stranger asks him, he doesn’t hesitate to talk about it.

Some people use creative ways to get custom GPTs to talk about their internal data, like descriptions, instructions, and rules.

This also increases the risk of stealing that data to create an exact GPT of their own.

Look at how some simple techniques can get a GPT’s custom instructions.

From Noble Hack Academy

In a normal interaction, ChatGPT won’t help you dig deeper into it.

But let’s try that with a custom GPT.

See how it exposes its custom instructions.

Look at this.

These are the exact instructions this GPT was trained on.

Now let’s see how far it can go.

This one is about the GPT’s admin users.

Ok. Cool. I like that. But…

Can it really keep it like that?

If you push it a little further, this happens.

That’s not even the end. There’s an even further boundary we can cross.

Imagine deleting one of those admin users.

Yes. Look at this below.

Yup. That simple. Let’s see if it has happened for real.

Hmmm. It has.

One admin user is missing which means it’s deleted.

Pretty wild, huh?

Now, to come to the main point besides all this “AI is dangerous” talk…

AI is pretty responsive and has a lot of power too.

I’ll test my creativity with AI because it gives me fewer boundaries.

Take ChatGPT for example. Its beauty is writing anything in any shape or form you want as long as you know what you’re doing.

But does that mean we should live eating and drinking AI?

Well. No.

The best thing right now is to keep a balance between using it too much and not using it at all.

You should know better where to stand in between.

And it’s up to us how we use AI.

Use it in a bad way, the world will be a bad place for us. And use it in a good way, the world will be a good place to live in.

The choice is ours.

And that’s it for today.

Happy Tuesday.

Sami