#2010 - Marc Andreessen

Marc Andreessen (0:13)

Fantastic, thanks.

Joe Rogan (0:14)

You are in the middle of this AI discussion.

You're right in the heat of this thing. But I think you have a different perspective than a lot of people do. A lot of people are terrified of AI, me included.

Marc Andreessen (0:27)

Oh, okay.

Joe Rogan (0:28)

For all the wrong reasons.

Marc Andreessen (0:29)

Of all the things to worry about.

Joe Rogan (0:31)

For me, my terror of it is all the way. It's kind of fun terror.

I'm not really freaking out. But I am recognizing that this is an emerging technology that is so different than anything we've ever experienced before. Particularly like what's chat GPT, what's happening with that right now. It's really fascinating and a lot of advantages. Like we were just talking last night, someone in the green room brought up the fact that there was this, they're using it for medical diagnoses, and it's very accurate, which is incredible.

There's a lot of good things to it.

Marc Andreessen (1:03)

Yeah, so you probably remember last time I was on, we spent quite a bit of time talking about this, and this is when these check bots were running inside Google, but the rest of us didn't have access to them yet. And that guy had come out and said that he thought that they were self-aware.

And the whole thing was like this big mystery of what's going on. And now the world gets to use these things, right? And everybody since then, everybody kind of has access.

Joe Rogan (1:23)

Really quickly, that was a short amount of time.

Marc Andreessen (1:25)

Yeah, it's been great. And then look, these things, when I say this, it's like chat GPT, and then Microsoft has a version called Bing. Google has a version called Bard now that's really good. There's a company, Anthropic, that has a thing called Claude.

You know, if you just run the comparison, they're basically as good as a doctor. They're as good as the average doctor at this point at being a doctor. They're as good at being a lawyer as the average lawyer.

You kind of go through basically anything involving knowledge work, you know, anything involving information, synthesizing, reporting, you know, writing legal briefs, anything like this. In business, they're actually already really good. They're as good as the average management consultant.

Joe Rogan (1:58)

No, the way they acquire data, they're essentially scouring the internet, right? Sort of.

Marc Andreessen (2:05)

It's more like they're fed the internet.

Joe Rogan (2:07)

They're fed the internet.

Marc Andreessen (2:08)

And I say it makes a difference because the company that produces the AI determines what data goes into it, and that determines a lot of how it works and what it does or won't do.

Joe Rogan (2:17)

So in that regard, is there a concern that someone could feed it fake data? Yeah.

Marc Andreessen (2:23)

Well, you may have noticed that people over time have said a lot of fake things.

Joe Rogan (2:26)

Yes, I have noticed that.

Marc Andreessen (2:29)

So that's all in there. So the way to think about it basically is it's being trained, the full version of these things are being trained on basically the sum total of human written expression.

Right? So basically everything people have ever written. There are some issues and you got to get all, you know, somehow we got to figure out how to get all the books in there. Although all the books prior to 1923 are in there because they're all out of copyright, but more recent books are a challenge. But anything that you can access on the internet that's text, right? Which is, you know, a staggeringly broad, you know, set of material is in there. By the way, both nonfiction and fiction, right? So a lot of stories are in there.

And then the new versions of these that are being built right now are what are called multimodal.

And so that means you can feed them not only text, but you can also feed them images, you can feed them videos, right? So they're going to be trained on all of YouTube, right? They're going to be trained on all podcasts, right? And they're going to be trained kind of equivalently between text and images and video and all kinds of other data. And so they're going to, they already have very comprehensive knowledge of human affairs, but it's going to get very complete.

Joe Rogan (3:25)

So if it's scouring the internet, it's getting all this data from both fiction and nonfiction, how does it interpret data that's kind of satire? Like what does it do with like Hunter S. Thompson, like gonzo journalism?