Cal Newport

But the real bottlenecks in people's work is often not the things that we're trying to aim AI at.

Like, I don't know people are super frustrated at booking a plane ticket online.

How often do you book plane tickets?

You kind of want to know, like, let me see, maybe this time will be better.

What seat's available?

It takes five minutes.

So it was a huge jump to go from a travel agent to a web interface.

But this is not a bottleneck in people's life now, where I want to give complicated- I book flights all the time.

Well, I had the same argument with social science researchers who were like, if you're geeky enough to learn coding agents...

They're like, this is revolutionizing science research because now, for example, you could have it write a program to process a data file and then format it into a plot.

And that might have taken you four hours to do.

And you work with it for a half hour and you get that result.

This is revolutionizing research.

And I'm saying, well, it's not.

The bottleneck for social science researchers is not evolution.

analyzing data and producing plots you're not sitting there doing that eight hours a day every day and if i could do this twice as fast i'll produce twice as many papers i might write one paper in a three-month period yeah in there there's like four hours i spent making a plot and sure it'd be nice if that four hours became 30 minutes but that's four hours out of like a multi-month process of sort of thinking about this paper what is a plot by the way

Like a graph.

Oh, right.

Yeah, the computer science term.

But yeah, it's like, that's nice.