Rob Wiblin

We're getting to the end eventually.

You might be under the impression that AI companies are selling access to their AI models at below what it costs to supply them in order to pull in customers, get them hooked, a bit like Uber did back in the day.

But that's not the case.

The companies are clearly making money on each additional paying user who signs up.

Now, there's also the essential issue of fixed costs.

They need to make enough profit from those customers to pay the fixed costs of their sales and marketing team, the research and development teams.

The staff can be very expensive.

And of course, the cost of the compute to train the models in the first place, which is one of their biggest costs.

but they're making a profit on each new customer and they're growing revenue five fold a year.

That sounds like they're on a reasonable track to become profitable to me.

And that's even setting aside the possibility that they might fully replace human labor or build a super intelligent machine god at some point, which I think would surely change the game.

So that is all a lot to digest.

I've been trying to give an overview of the things that we learned last year, but where does it leave us overall?

Well, personally, I would be pretty shocked if we got fully automated AI research and development next year.

2028, I guess it's imaginable, but it's gonna require some surprising breakthroughs or an acceleration beyond what we're seeing right now.

2029, 2030, it begins to feel plausible.

At least we can't rule it out if current trends continue and we don't hit any significant new roadblocks.

But there's also a very real chance that we're in for a significantly longer and slower takeoff.

But let's put this into a broader context.

Previous guest of the show, Helen Toner, wrote an article last year titled, Long Timelines to Advanced AI Have Gotten Crazy Short.