Dan Nottingham

But the discussion became quite different.

Now, I had learned something and I could actually delve into it more deeply, understand why I didn't understand what really happened, how I had been misled.

And it was a kind of a growth experience for me.

And I'm hoping that's the experience that others will have using this.

At first, I thought this was going to be something that was going to be very easy.

You just prompt a large language model properly and it's going to come back.

And it's just saving people the trouble of creating a prompt.

But there's not a big technical hurdle here.

And I tried using ChatGPT and Gemini and a few others.

And thought, okay, if I just prompt it correctly, it'll get me back.

Some answers.

But then it turns out when you really start hammering it with all kinds of different sorts of information that you might want to get out of it, we discover that it's just not that simple.

AI will kind of go off and it'll hallucinate.

It'll do things wrong.

It doesn't behave quite the way you want it to.

So it required almost a year's worth of work, of understanding how the engines work in this particular context.

AI is an emerging technology, so it's not always going to behave the way you think it's going to behave.

So we had to do a lot of different techniques.

A lot of proprietary code was written to help the AI engines behave in a way that was going to give us the results we wanted.

what the, the big moment for me was when I was putting in, uh, just silly things just to, just to check it to just simple things like, uh, you know, the earth is flat or something like that and see what, uh,