David Kyle Johnson

And you could see how all the tricks are done.

Sleight of hand, misdirection, physical props, right?

As the magician's skill improves, the tricks get harder and harder to detect.

And an expert magician could be so good that even a really keen and intelligent observer cannot see how the tricks are done.

But that doesn't mean the magic is real, right?

When it gets to that point, it doesn't mean, oh, he must now be using actual magic because I can't detect the tricks.

No, he's just a really good at the illusion of magic.

So LLMs are really good at the illusion of thinking through mimicking language.

Well, it doesn't matter what kind of cake.

No, I'm just kidding.

Bob, I got him again.

Yeah, it's an illusion.

Dawkins fell, like many people have, like Blake Des Moines had on the show, right?

Fell for the illusion of thinking because they confuse sophisticated language with sophisticated thought.

But we're dealing with programs that are trained on massive amounts of data and are just really, really good at mimicking language.

They are not thinking machines.

They are language mimicking machines.

Because he's not a topic expert, Jay, and he's falling for a very powerful illusion.

How savvy is he with the actual process of LLMs?

He asked a couple of questions.