Secrets of ChatGPT revealed

The artificial intelligence bot ChatGPT is so smart, it seems alien. But, columnist Peter Judd explores whether the super bot is more like us than we think.

Is Chat GPT more human than we think?

My partner in crime, ChatGPT, has been keeping a secret from me.

Not on purpose, I don’t think.

But this marvellous artificial intelligence that can turn Shakespeare into Katy Perry may not be as alien as I thought.

Oh, oh, oh.

It took a while to crack through my thick skull but I get it now.

ChatGPT has no idea what it is going to say next.

When it answers a question, it predicts its answer.

Not in gigabytes.

It does it one word at a time.

Like watching someone type.

Or telling a joke without knowing what the joke is or the punchline and making it up as you go along while channelling shaggy dog comedian Dave Allen smoking and drinking on his TV show more than 25 years ago.

If I ask it to write a business strategy, it refers to its training data, discovers a pattern, then word-by-word, it predicts what it is going to say next without knowing what it will actually say next.

How insane is that?

How does it come up with a strategy, with subheadings and bullet points, when it doesn’t know what it will say?

It’s a bit like us in a business meeting, isn’t it.

The boss asks you a question about how you’re gonna hit budget and you just start talking.

It spills out of your mouth, all this stuff, some numbers that no one questions, and you’re listening to your gibberish and occasionally get a surprise hit of adrenaline on fast reflection because you might be actually making sense.

People are nodding.

You keep going.

Then someone corrects you and you apologise for being a language model and make a quick deviation back towards a revised version of the truth.

You had no idea what you were going to say.

Then you said it. Then everyone agreed and now they’re going to do it.

Not once did you look up the internet to check your facts.

There have been conversations with colleagues and each time a snippet of insight gets tucked away.

But you haven’t had time to reflect on any of this and now it all comes together in a rush.

How the heck did you do that?

This is what makes ChatGPT so much more interesting than any other technology.

It doesn’t behave like a robot at all. It seems to behave the way we might when put under the pump.

Creative. Inventive.

A bit loose with the truth.

Almost human.

Here we are, wanting it to behave like a robot and have all the exactitude that comes with a more mechanical, concrete interaction with the world, but what we get is something that embodies all the ambiguity and inconsistency that comes with being human.

And if it is the greatest mimic, predicting its next moves based on what it has experienced, then perhaps that’s who we are, too?

Originally published as Secrets of ChatGPT revealed