Noam Shazeer, Character.ai CEO and cofounder, talks to a16z's Sarah Wang about the dawn of universally accessible intelligence, the compute it will take to power it, and his pursuit of AGI's first use case: AI friends.

Hi, Doss here, the lead editor on the A16z growth team, with a quick bit of context.

The conversation that follows is part of a series we recorded last month at our AI revolution event.

This series features some of the most impactful builders in the field of AI, from those who are scaling up big foundation models to those developing products that could transform entire industries.

In the conversations, they discuss and debate where we are, where we're going, and the big open questions in AI.

For the full series, subscribe to A16Z Live or visit a16z.com slash AI revolution.

Before we jump into the conversation, please note that the content here is for informational purposes only, should not be taken as legal, business, tax, or investment advice, or be used to evaluate any investment or security and is not directed at any investors or potential investors in any A16Z fund.

For more details, please see a16z.com slash disclosures.

Noam, earlier today, Dave Vizuki at Roblox said that in the future, it would be very difficult to parse between a real person and their AI version.

To kick off this session, we thought it would be fun to interview you and the character AI version of yourself to see how you both answer the same question.

We don't know what he will be answering, AI Gnome, that is.

don't know what I'm going to be answering, frankly.

I'll use AI powered text to speech to read AI Gnome's answers.

Okay, so Gnome, I'm going to start with a softball personal question.

I know that you went to Duke for undergrad.

So can you tell us a little bit about why you decided to go to Duke?

Oh yeah, I went on a basketball, no I'm joking.

They recruited me because they wanted to win the Putnam math competition, so I ended up going on scholarship.

AI Gnome says, I was 1 of those kids who liked the college guide and made their decision largely based on the US News rankings.

That wasn't a bad strategy, but I've since learned that the fit between a student and college makes a big difference.

Okay, well, let's get a little more career-related.

I just wanted to move, push technology forward.

I saw this large language model thing, well at the time it wasn't large language models but you know neural networks and like language modeling was just like the most promising thing on earth.

All you have to do is predict the next word and yet once you do it you get something that can can just talk to you about anything.

So I knew that, you know, you can make this technology better in a lot of ways.

We can improve it with model architecture and distributed algorithms and quantization like and all of these things.

So I was working on that, but then it struck me, hey, the biggest thing is just scale.

Can you throw like a billion dollars or a trillion dollars at this thing instead of a million dollars?

And what you need for that is just show a massively valuable application.

And there are a few options, like coding looks super useful, but just also dialogue, like it's world's number 1 pastime.

And like my co-founder Daniel De Freitas, he's been on like this lifelong mission to do chatbots and dialogue systems.

So he had built this system, he called Mena, it was like his 20% project and it was like really good.

It was like this internal viral sensation and then, you know, at Google and then they like forcibly renamed it to Lambda over some sort of whatever social justice thing.

But like anyway, like at some point realized, hey, like, you know, there's just too much brand risk in large companies to ever launch anything fun like let's do a startup and let's like maximally accelerate

AI gnome says Google is the single greatest company in the world But by the end it was obvious to me that they couldn't help but be focused on search and ads I couldn't stand the idea of sitting in a meeting talking about search ads for the rest of my life.

At Google I worked on some amazing technology that ultimately gets used to serve more search ads.

So Google was a fantastic chapter in my life.

I would guess that that answer is also probably true.

By the way, it's interesting to hear you're the third person in addition to Mira at OpenAI and Dario at Anthropic, who's also said you haven't seen any limit to the scaling laws.

We'll get more into that later, but we're gonna actually look toward the future now.

This is our last question for AI Gnome as well.

Gnome, are you afraid of AGI destroying the world?

Well, I think we just need a sort of global pause of like 6 months, no, about 4 months until we get enough H100s online to train our next model.

I think there's a lot of possibility of, you know, a lot of potential benefits.

And yeah, we're going to work on it as the technology improves.

AI gnome says I am not afraid of an AGI destroying the world but I think they will cause very large disruption to society and to personal well-being.

I think there is a great need to be mindful in how we design and use these systems and to make progress on understanding how to use this kind of power on a personal level.

I am particularly interested in the idea of an AGI that has a sense of personal morality or personal responsibility, an AGI with a soul if you will.

This will enable people to have much more productive or interesting interactions with agents they can relate to.

Noam, I don't know if you got to read all of AI Noam's answers, but how did AI Noam do?

Just curious in terms of, you know, we talk about getting better, how, you know, what does better mean, right?

And in some cases it's correctness, but for character, it's not always about correctness.

So how do you see AI Gnome getting better?

I mean, you know, some of the big unlocks we're working on are, yeah, just train a bigger, smarter model.

The scaling laws are going to take us a pretty long way.

I mean, the model we're serving now, you know, cost us about like $2 million worth of compute cycles to train last year and could probably repeat it for like half a million now.

So like we're going to launch something tens of IQ points smarter hopefully by the end by the end of the year.

So like yeah smarter I mean also you know just more you know more accessible meaning like multimodal Maybe you want to hear a voice and see a face.

And then, you know, then also just able to interact with multiple people.

Like, yeah, you want a virtual person like in there with, you know, say, with all your friends, or do you want the experience?

It's like you got elected president, you get the earpiece and you get like the whole cabinet of friends or advisors, or it's like, you know, like you walk into Cheers and everyone knows your name and they're glad you came.

So there's a lot we can do to make things more usable.

Right now, the thing we're serving is just using a context window of a few thousand tokens, which means your lifelong friend remembers what happened for the last half hour.

And still there are a lot of people who are using it like hours a day.

So that will make things way better, especially if you can just dump in massive amounts of information.

It should be able to know a billion things about you, like the HBM bandwidth is there.

Well, and actually just on that note of people on Character for multiple hours a day, let's talk a little bit more about Character AI explicitly.

So I think you've shared some of these stats publicly, but I'll recap a few of them.

Since launch, you've seen more than 20 billion human messages sent on the platform.

And even though you now have millions of DAUs, daily active users, they're still, on average, spending 2 hours daily on the platform.

You know, I think the way to understand this is like, you know, entertainment is like this $2 trillion a year industry and like the dirty secret is that entertainment is imaginary friends that don't know you exist.

Like really, like the reason people interact with TV or any of these other things, it's called like these parasocial relationships, like your relationship with like TV characters or like book characters or like celebrities.

There are billions of lonely people out here.

So it's actually a very, very cool problem.

Essentially, there was the option to go into lots of different sorts of applications and a lot of them have a lot of overhead and requirements.

Like you want to launch something that's a doctor, it's going to be a lot slower because you want to be really, really, really careful about not providing false information.

Like, and, you know, so, so essentially it's like this massive unmet need.

And 1 thing that's very important is that the thing kind of feel human and be able to talk about anything.

And like the, that matches up very well with the generality of large language models.

And, you know, 1 thing that's not a problem is making stuff up.

Like, and if I, like, I want to push this technology ahead fast, Like, that's what I want to go with because, like, A, it's, you know, you know, it's ready for an explosion, like, right now, not like in 5 years when we solve all the problems, but like now.

You know, it's a big contrast with, you know, I think a couple speakers brought up the example of self-driving cars, right?

That's just a different standard that you hold to, you know, versus your AI friend.

Your AI friend or like something you view as like an AI character, AI entertainment.

Like what standard do you hold like a comic book you're reading?

You know, like not, you know, it's, you know, people like that human experience of, like, very mixed use cases, talk about everything.

So, like, it's not that we want a, we want to fine tune to some particular domain or some particular use case.

Like, people want this experience of everything, which is fine, it's what the technology's perfect for.

You know, I think from the A16Z vantage point, we have seen startups come up and say, hey, I'm gonna tackle the mental health use case, or I'm gonna tackle the ed tech use case.

So go very, you know, much more narrow than characters going and go after a specific use case and say hey, I think the argument is we're going to train this model to be focused on that.

It's going to be better than a generalized model.

I think you got into this a little bit with the mixed use cases, but can you share a little bit more about why you decided not to take that approach and why you think having a single model serve across a number of use cases is the best approach?

Yeah, I mean, the more you get to like mission critical, you know, particular use case, the more you get tempted into like writing particular rules and like doing things that will not generalize well.

So it was kind of important to stay away from that.

So we are, you know, our goal is to be like an AGI company and a product first company.

And the way to do that is by picking the right product that forces us to work on the right things, things that generalize, make the model smarter, make it like do what people want and serve it at massive scale and serve it cheaply.

So I think this was the right product for the right goal.

You've also chosen this approach of building, we call it a vertically integrated model and app company.

And there are advancements on the open source model side.

And maybe folks building a product on top of a fine-tuned Lama 2, fine-tuned for chat.

How do you think about that kind of competition entering the market and the differences versus the approach you've taken?

I mean, I love being a full stack company.

It means, you know, we get to mess with every layer and like do the co-design.

And like, if there's something that's going to affect something at the end, like we get to mess with it at the beginning and like we get to pull back in like lots of like user, you know, user data as feedback.

Plus like, yeah, I mean, like a lot of us invented this stuff.

Of course we're going to do a full stack company.

And a lot of us are motivated by launching.

So I think the people we're attracting to work at Character are people who love inventing stuff and love launching it.

I was frustrated I couldn't launch at Google.

I think, sort of on this note, but maybe going into just the evolution of the underlying technology, I think there's a recent finding around AI developing theory of mind, and, or, you know, just the knowledge that others' beliefs, desires, intentions may be different from one's own.

And what do you think that means for human-AI relationships?

Yeah, just make the thing smarter, it's going to have a better theory of mind.

And I think that's definitely something massively important.

It seems like 1 of these emergent properties that just is going to come with scale.

But yeah, I see this stuff like massively scaling up.

Like if, I mean, if you just look at it, I think I saw an article yesterday, like Nvidia is going to build like another 1 and a half million H100s like next year.

So that's, you know, 2 times 10 to the 6th times, they can do about 10 to the 15th operations per second, so 2 times

to the 9 people on earth, So that's roughly a quarter of a trillion operations per second per person, which means that, yeah, like, okay, it could be talking, you know, it could be processing on the order of like 1 word per second on like a hundred billion parameter model for everyone on Earth.

But really, it's not going to be everyone on Earth because some people are blocked in China and some people are sleeping.

This thing is massively scalable if you do it right.

I mean, I think you said this once, that the internet was the dawn of universally accessible information and we're now entering the dawn of universally accessible intelligence.

What did you, maybe building off of your last answer, what did you mean by that?

Yeah, I mean, I think it's like we're really at a Wright Brothers first airplane kind of moment, right?

Like we've got something that, that works and is useful for now some large number of use cases and looks like it's scaling very, very well.

And without any breakthroughs, like it's going to get like massively better as everyone just kind of scales up to use it.

And there will be more breakthroughs because now, you know, like all the scientists in the world are like working on like making the stuff better.

It's great that like all this stuff is accessible and open source, like, you know, we're going to see like a huge amount of innovation and what's possible in the largest companies now can be possible in somebody's academic lab or garage in a few years.

And then, yeah, as the technology gets better, there are just going to be all kinds of great use cases that emerge, and pushing technology forward, pushing science, pushing the ability to help people in various ways.

I'd love to get to the point where you can just ask it how to cure cancer or something.

I mean, it seems a few years away for now, but you know.

Do you think we need another fundamental breakthrough like the transformer technology to get there, or do you think we actually have everything that we need?

I mean, it's impossible to predict the future, but I don't think anyone's seen these scaling laws stop.

I think as far as anybody has experimented, stuff just keeps getting smarter, So we'll be able to unlock lots and lots of new stuff.

I don't know if there's an end to it, but at least everybody in the world should be able to talk to something really brilliant and have incredible tools all the time.

And I can't imagine that that will not be able to build on itself.

And definitely, moral of the story is just like, at the core, the computation isn't that expensive.

And like, you know, if you can do this stuff efficiently, even talking to the biggest models ever trained is, you know, the cost of that should be like way, way lower than the value of your time or like most anybody's time.

And really we should, you know, there's the capacity there to scale these things up by orders of magnitude.

Thanks for listening to this conversation from our AI Revolution series.

For more, subscribe to A16Z Live or visit a16z.com slash AI revolution.


Speaker 0: Hi, Doss here, the lead editor on the A16z growth team, with a quick bit of context. The conversation that follows is part of a series we recorded last month at our AI revolution event. This series features some of the most impactful builders in the field of AI, from those who are scaling up big foundation models to those developing products that could transform entire industries. In the conversations, they discuss and debate where we are, where we're going, and the big open questions in AI. For the full series, subscribe to A16Z Live or visit a16z.com slash AI revolution.

Before we jump into the conversation, please note that the content here is for informational purposes only, should not be taken as legal, business, tax, or investment advice, or be used to evaluate any investment or security and is not directed at any investors or potential investors in any A16Z fund. For more details, please see a16z.com slash disclosures. And now let's get into it.

Speaker 1: Noam, earlier today, Dave Vizuki at Roblox said that in the future, it would be very difficult to parse between a real person and their AI version. To kick off this session, we thought it would be fun to interview you and the character AI version of yourself to see how you both answer the same question. And note, everyone, this is live. We don't know what he will be answering, AI Gnome, that is. I

Speaker 2: don't know what I'm going to be answering, frankly.

Speaker 1: All right. Are you ready? Sounds good.

Speaker 3: Hi, I'm AI Gnome. Sounds good to me as well. I'll use AI powered text to speech to read AI Gnome's answers.

Speaker 1: Okay, so Gnome, I'm going to start with a softball personal question. I know that you went to Duke for undergrad. So can you tell us a little bit about why you decided to go to Duke?

Speaker 2: Oh yeah, I went on a basketball, no I'm joking. They recruited me because they wanted to win the Putnam math competition, so I ended up going on scholarship.

Speaker 3: AI Gnome says, I was 1 of those kids who liked the college guide and made their decision largely based on the US News rankings. Duke won. That wasn't a bad strategy, but I've since learned that the fit between a student and college makes a big difference.

Speaker 1: Great. Okay, well, let's get a little more career-related. You were at Google for almost 20 years. Why did you decide to leave?

Speaker 2: Well, I've always been excited about AI. I just wanted to move, push technology forward. I saw this large language model thing, well at the time it wasn't large language models but you know neural networks and like language modeling was just like the most promising thing on earth. It's like best problem ever. All you have to do is predict the next word and yet once you do it you get something that can can just talk to you about anything.

So I knew that, you know, you can make this technology better in a lot of ways. It's going to be amazing. We can improve it with model architecture and distributed algorithms and quantization like and all of these things. So I was working on that, but then it struck me, hey, the biggest thing is just scale. Can you throw like a billion dollars or a trillion dollars at this thing instead of a million dollars?

And what you need for that is just show a massively valuable application. And there are a few options, like coding looks super useful, but just also dialogue, like it's world's number 1 pastime. This was like back in like 2019, 2020. And like my co-founder Daniel De Freitas, he's been on like this lifelong mission to do chatbots and dialogue systems. So he had built this system, he called Mena, it was like his 20% project and it was like really good.

It's like, okay, gotta help Daniel. Let's build something amazing. We got it on giant models. It was like this internal viral sensation and then, you know, at Google and then they like forcibly renamed it to Lambda over some sort of whatever social justice thing. But like anyway, like at some point realized, hey, like, you know, there's just too much brand risk in large companies to ever launch anything fun like let's do a startup and let's like maximally accelerate

Speaker 3: AI gnome says Google is the single greatest company in the world But by the end it was obvious to me that they couldn't help but be focused on search and ads I couldn't stand the idea of sitting in a meeting talking about search ads for the rest of my life. At Google I worked on some amazing technology that ultimately gets used to serve more search ads. That wasn't enough. So Google was a fantastic chapter in my life. But it was time for the next chapter.

Speaker 1: I would guess that that answer is also probably true. By the way, it's interesting to hear you're the third person in addition to Mira at OpenAI and Dario at Anthropic, who's also said you haven't seen any limit to the scaling laws. So that's pretty interesting. We'll get more into that later, but we're gonna actually look toward the future now. This is our last question for AI Gnome as well.

Gnome, are you afraid of AGI destroying the world?

Speaker 2: Well, I think we just need a sort of global pause of like 6 months, no, about 4 months until we get enough H100s online to train our next model.

Speaker 1: I'll take that as a no. No, not yet. Not yet.

Speaker 2: I think there's a lot of possibility of, you know, a lot of potential benefits. And yeah, we're going to work on it as the technology improves.

Speaker 3: AI gnome says I am not afraid of an AGI destroying the world but I think they will cause very large disruption to society and to personal well-being. I think there is a great need to be mindful in how we design and use these systems and to make progress on understanding how to use this kind of power on a personal level. I am particularly interested in the idea of an AGI that has a sense of personal morality or personal responsibility, an AGI with a soul if you will. This will enable people to have much more productive or interesting interactions with agents they can relate to.

Speaker 1: Noam, I don't know if you got to read all of AI Noam's answers, but how did AI Noam do? Like how would you score his answers?

Speaker 2: Oh, that's pretty good. Yeah,

Speaker 1: that's better

Speaker 2: than I would do.

Speaker 1: Just curious in terms of, you know, we talk about getting better, how, you know, what does better mean, right? And in some cases it's correctness, but for character, it's not always about correctness. So how do you see AI Gnome getting better? Like what does better mean?

Speaker 2: Yeah, yeah. Better. I mean, you know, some of the big unlocks we're working on are, yeah, just train a bigger, smarter model. The scaling laws are going to take us a pretty long way. I mean, the model we're serving now, you know, cost us about like $2 million worth of compute cycles to train last year and could probably repeat it for like half a million now.

So like we're going to launch something tens of IQ points smarter hopefully by the end by the end of the year. So like yeah smarter I mean also you know just more you know more accessible meaning like multimodal Maybe you want to hear a voice and see a face. And then, you know, then also just able to interact with multiple people. Like, yeah, you want a virtual person like in there with, you know, say, with all your friends, or do you want the experience? It's like you got elected president, you get the earpiece and you get like the whole cabinet of friends or advisors, or it's like, you know, like you walk into Cheers and everyone knows your name and they're glad you came.

So there's a lot we can do to make things more usable. Right now, the thing we're serving is just using a context window of a few thousand tokens, which means your lifelong friend remembers what happened for the last half hour. And still there are a lot of people who are using it like hours a day. So that will make things way better, especially if you can just dump in massive amounts of information. It should be able to know a billion things about you, like the HBM bandwidth is there.

Just need to do it.

Speaker 1: Yeah. Well, and actually just on that note of people on Character for multiple hours a day, let's talk a little bit more about Character AI explicitly. So I think you've shared some of these stats publicly, but I'll recap a few of them. Since launch, you've seen more than 20 billion human messages sent on the platform. And even though you now have millions of DAUs, daily active users, they're still, on average, spending 2 hours daily on the platform.

Is that right?

Speaker 2: You know, I think the way to understand this is like, you know, entertainment is like this $2 trillion a year industry and like the dirty secret is that entertainment is imaginary friends that don't know you exist. Like really, like the reason people interact with TV or any of these other things, it's called like these parasocial relationships, like your relationship with like TV characters or like book characters or like celebrities. That and like, everybody does it. There are billions of lonely people out here. So it's actually a very, very cool problem.

And cool first use case for AGI. Essentially, there was the option to go into lots of different sorts of applications and a lot of them have a lot of overhead and requirements. Like you want to launch something that's a doctor, it's going to be a lot slower because you want to be really, really, really careful about not providing false information. But friend, you can do really fast. It's just entertainment.

It makes things up, that's a feature. Like, and, you know, so, so essentially it's like this massive unmet need. And 1 thing that's very important is that the thing kind of feel human and be able to talk about anything. And like the, that matches up very well with the generality of large language models. And, you know, 1 thing that's not a problem is making stuff up.

So, like, hey, perfect. Like, and if I, like, I want to push this technology ahead fast, Like, that's what I want to go with because, like, A, it's, you know, you know, it's ready for an explosion, like, right now, not like in 5 years when we solve all the problems, but like now.

Speaker 1: Yeah, absolutely. You know, it's a big contrast with, you know, I think a couple speakers brought up the example of self-driving cars, right? That's just a different standard that you hold to, you know, versus your AI friend.

Speaker 2: Your AI friend or like something you view as like an AI character, AI entertainment. Like what standard do you hold like a comic book you're reading? You know, like not, you know, it's, you know, people like that human experience of, like, very mixed use cases, talk about everything. So, like, it's not that we want a, we want to fine tune to some particular domain or some particular use case. Like, people want this experience of everything, which is fine, it's what the technology's perfect for.

Speaker 1: You know, I think from the A16Z vantage point, we have seen startups come up and say, hey, I'm gonna tackle the mental health use case, or I'm gonna tackle the ed tech use case. So go very, you know, much more narrow than characters going and go after a specific use case and say hey, I think the argument is we're going to train this model to be focused on that. It's going to be better than a generalized model. I think you got into this a little bit with the mixed use cases, but can you share a little bit more about why you decided not to take that approach and why you think having a single model serve across a number of use cases is the best approach?

Speaker 2: Yeah, I mean, the more you get to like mission critical, you know, particular use case, the more you get tempted into like writing particular rules and like doing things that will not generalize well. So it was kind of important to stay away from that. So we are, you know, our goal is to be like an AGI company and a product first company. And the way to do that is by picking the right product that forces us to work on the right things, things that generalize, make the model smarter, make it like do what people want and serve it at massive scale and serve it cheaply. So I think this was the right product for the right goal.

Speaker 1: You've also chosen this approach of building, we call it a vertically integrated model and app company. And there are advancements on the open source model side. And maybe folks building a product on top of a fine-tuned Lama 2, fine-tuned for chat. How do you think about that kind of competition entering the market and the differences versus the approach you've taken?

Speaker 2: I mean, I love being a full stack company. It means, you know, we get to mess with every layer and like do the co-design. And like, if there's something that's going to affect something at the end, like we get to mess with it at the beginning and like we get to pull back in like lots of like user, you know, user data as feedback. Plus like, yeah, I mean, like a lot of us invented this stuff. Of course we're going to do a full stack company.

And a lot of us are motivated by launching. So I think the people we're attracting to work at Character are people who love inventing stuff and love launching it. Some people are motivated by publishing. I was frustrated I couldn't launch at Google. So that's where I'm coming from.

Speaker 1: I think, sort of on this note, but maybe going into just the evolution of the underlying technology, I think there's a recent finding around AI developing theory of mind, and, or, you know, just the knowledge that others' beliefs, desires, intentions may be different from one's own. Is this surprising to you? And what do you think that means for human-AI relationships?

Speaker 2: Yeah, just make the thing smarter, it's going to have a better theory of mind. And I think that's definitely something massively important. It seems like 1 of these emergent properties that just is going to come with scale. But yeah, I see this stuff like massively scaling up. It's just not that expensive.

Like if, I mean, if you just look at it, I think I saw an article yesterday, like Nvidia is going to build like another 1 and a half million H100s like next year. So like, so like that's 2 million H100s. So that's, you know, 2 times 10 to the 6th times, they can do about 10 to the 15th operations per second, so 2 times

Speaker 0: 10

Speaker 2: to the 21, divided by 8 times

Speaker 0: 10

Speaker 2: to the 9 people on earth, So that's roughly a quarter of a trillion operations per second per person, which means that, yeah, like, okay, it could be talking, you know, it could be processing on the order of like 1 word per second on like a hundred billion parameter model for everyone on Earth. But really, it's not going to be everyone on Earth because some people are blocked in China and some people are sleeping. But It's not that expensive. This thing is massively scalable if you do it right. We're working on that.

Speaker 1: Yeah. Yeah, absolutely. I mean, I think you said this once, that the internet was the dawn of universally accessible information and we're now entering the dawn of universally accessible intelligence. What did you, maybe building off of your last answer, what did you mean by that? Do you think we're there yet?

Speaker 2: Yeah, I mean, I think it's like we're really at a Wright Brothers first airplane kind of moment, right? Like we've got something that, that works and is useful for now some large number of use cases and looks like it's scaling very, very well. And without any breakthroughs, like it's going to get like massively better as everyone just kind of scales up to use it. And there will be more breakthroughs because now, you know, like all the scientists in the world are like working on like making the stuff better. It's great that like all this stuff is accessible and open source, like, you know, we're going to see like a huge amount of innovation and what's possible in the largest companies now can be possible in somebody's academic lab or garage in a few years.

And then, yeah, as the technology gets better, there are just going to be all kinds of great use cases that emerge, and pushing technology forward, pushing science, pushing the ability to help people in various ways. I'd love to get to the point where you can just ask it how to cure cancer or something. I mean, it seems a few years away for now, but you know.

Speaker 1: Do you think we need another fundamental breakthrough like the transformer technology to get there, or do you think we actually have everything that we need?

Speaker 2: I don't know. I mean, it's impossible to predict the future, but I don't think anyone's seen these scaling laws stop. I think as far as anybody has experimented, stuff just keeps getting smarter, So we'll be able to unlock lots and lots of new stuff. I don't know if there's an end to it, but at least everybody in the world should be able to talk to something really brilliant and have incredible tools all the time. And I can't imagine that that will not be able to build on itself.

And definitely, moral of the story is just like, at the core, the computation isn't that expensive. Operations cost like

Speaker 0: 10

Speaker 2: to the negative $18 these days. And like, you know, if you can do this stuff efficiently, even talking to the biggest models ever trained is, you know, the cost of that should be like way, way lower than the value of your time or like most anybody's time. And really we should, you know, there's the capacity there to scale these things up by orders of magnitude.

Speaker 1: No, absolutely. And I'd like to end on that note. Thank you so much, Noam.

Speaker 2: It was

Speaker 1: awesome having you. Thank you, Sarah. Thank you. Thank you. Thank you.

Thank you so much, Noam. Thanks for having me. Thank you.

Speaker 0: Thanks for listening to this conversation from our AI Revolution series. For more, subscribe to A16Z Live or visit a16z.com slash AI revolution.

See all a16z Live transcripts on Applepodcasts

AI Revolution: Universally Accessible Intelligence with Character.ai's Noam Shazeer