What San Francisco Forgot?

AI has reached a ceiling. And I think most people are not talking about the right one.

I isolated myself from the field for a couple of months.

There is nothing new anymore.

Because we hit the big ceiling.

Data.

For a long time we were told that AI's biggest problems are compute and energy.

Now we have compute.

Now we have money. Investments are flooding into this area.

Now we have everything.

So why did it stop developing?

Yes, that 1 percent improvement on some obscure benchmark they call "Humanity's Last Exam" is not real progress.

— Arzuman Abbasov

So what is the next step?

Either AI finds a way to move into the physical world, into factories, into real operations.

Or the bubble bursts.

I think the first one is more likely.

So why am I writing this?

Because AI needs to learn how to actually do things.

Unlike humans, AI needs to be shown the same thing in millions of different ways before it can do something reliably.

So the next big thing is data.

Large amounts of high quality, labeled data.

That is what will unlock the next step.

I will be writing more about this.

This is a billion dollar market.

And San Francisco has completely forgotten about it.