Cal Newport

A model like Mythos is going to have

Hundreds of billions, if not a trillion parameters.

And they use a three to five billion parameter off the shelf.

You could run this model on a chip inside your- Sorry, 10 trillion.

10 trillion.

Oh, okay.

10 trillion parameter.

Is that true?

Oh, my God.

10 trillion parameters is insane.

Like, you better be – that better be either gaming the stock market and creating billions of dollars a day and, like, fancy option returns or changing lead into gold.

Because to run something that has 10 trillion parameters to do almost anything else is a – it's like –

we're going to launch ourselves in this space to do something and then land every time.

That's so incredibly expensive.

But the real fear then is like, well, wait a second.

If they could do most of this stuff with a free, cheap model that I could just run on a machine at home, that's what keeps, I think, Dario Amadei up at night.

That's what keeps Sam Altman up at night.

It's the future.

Look, I've been pitching this, right?

I think the useful and the only ethical and sustainable future for AI is