ChatGPT is starting with intent-based ads and firm limits

Transcript

Transcribed by machine (Whisper) from the episode’s audio, so a word may be off here and there.

1. 42:34

   And that's the added benefit of putting that in the hands of tons of people for free. Maybe a bit of a more technical question, but I think it speaks to how people will feel about the product. You've kind of got two directions ahead of you. You could go the Facebook, which you helped build, the Facebook ads route, or you could go the Google ads route. and to most people they probably don't understand the difference but for Facebook it's you know I tell Facebook I want to reach moms in Ohio who like Target speaking from experience yes that's our podcast that's your side hustle and and Facebook just goes and does it and they use the social graph and all that you know this you help build this there's also the Google approach which is keyword based auction based and is not tied to identity in the same way and is not like we go

2. 43:17

   figure it out for you. Which direction are you guys going to go? Because I mean, your ex-Facebook, Vijay, who's helping you build the ads team is ex-Facebook. I mean, there's a lot of ex-Facebook people at OpenAI. So are you going to go the more Google route or the more Facebook route, do you think? Well, first off, we have a lot of ads experience, but we all came here to build a very different ads model than what we built in the past that's appropriate for the product we have in front of us. And I think if you look at ChatGPT, the Google model is a lot more relevant.

3. 43:46

   like you're talking about something you're you're clearly having some kind of intent and like chat gpt is helping you and so you can imagine ads like being related to that content performing a lot better and being a lot more relevant for the user so well definitely starting there you can imagine over time that there might be surfaces that are more like less intent driven and more open to you know yeah exactly polls where it might be you know closer to the facebook model uh but we're very much starting with like max level of relevance for the content that you're talking about.

4. 44:20

   This has just become such a conversation in tech. I mean, I was just at Davos and it was like, everyone was talking about ads and AI. And I was even catching up with Boz, one of your former colleagues about it. And he was saying that, that basically the Google model makes the most sense in a chatbot context, but also that, you know, people are rightfully going to be still worried that the ad is somehow tied to the LLM's output, no matter what you do to design it.

5. 44:46

   And Demis from Google was kind of saying a similar thing. And I'm curious, like, do you think you can, even if you're saying in the background, this is how it works, do you think people will still, there's going to be a challenge with people still thinking, no, this is actually like influenced by what I'm telling ChatGPT. I think that's definitely a risk. I mean, you know, Facebook is a good example.

6. 45:06

   Facebook didn't sell your data to advertise, but tons of people think they did. She doesn't listen to your microphone and tons of people think they do. And so is there a perception risk? For sure, we're doing everything we can in the products through education to educate people. The thing that I think is also helpful is because we have a conversational product, I think people will actually talk to the model and realize what's going on.

7. 45:30

   So I'll give you an example. Because the model is not aware of the ads, if you ask the model, like, hey, I just see some ads on my screen. Can you tell me more about them? The model is going to tell you, I actually don't see the ads because we're not surfacing that. But if you want me to see them, click this button. And that reinforces to people like, oh, by default, the model is not seeing that. So that's a good example.

8. 45:55

   And then on top of that, once it says, OK, you can actually look at the ads and tell me what you think, the model is going to respond something that is completely neutral. So that means that in some cases, the model is going to say, well, I see the ads for these two companies. That one is good, but that one is actually not really good. And I think that's going to reinforce neutrality. Now, I think that some brands are going to get spooked by this.

9. 46:19

   That's going to piss some brands off, yeah. But I also think that we live in a world where transparency and authenticity is what counts. And so if a brand has a bad reputation so that chat GPT ends up saying something bad, like they should fix that instead of blaming us. So I actually think the discussion with the model will reinforce the neutrality in a way that nothing else could. I think the placement of the ads inside of a product that is so personal.

10. 46:48

    I definitely am a believer that ads can be helpful and show you more of what you might be interested in. After all, products do solve problems in your life. I think I'm going to argue that the reason people are so apprehensive and concerned about this is almost just on principle. Like, why does arguably the best product in human history? Are you saying that's ChessieBT? Maybe consumer AI. I'll take it.

11. 47:13

    Consumer AI. All right. One of the best products in human history. Why does that need ads, you know? And I think there is a difference with Facebook with the cost of providing the service. you know the cost of burning all those tokens as opposed to running a photo news feed or a text news feed or something like that but would you have fewer users long term if you didn't have ads given that the argument is about accessibility i feel like the same number of people would use it just because of how good and how important it is well this brought me back to the apple thing like uh turns out more people use facebook than when the whole apple war over att started right so it's like people do actually use free services with ads.

12. 47:52

    Yeah, like, you know, ads have a bad rep, but ultimately people like end up, you know, enjoying services with ads. And the thing I would say is like, you're saying that in the current world, but think about how fast intelligence is progressing. It's going to become more and more like important to have the highest levels of intelligence in the free tier. And I think if you don't offer ads, what's going to end up happening is you're going to have a divergence where the people on paid tiers are going to get tons of great intelligence, are going to be, you know, empowered with these superpowers.

13. 48:25

    And then people in the free tier are going to have, you know, like okay experience, but with a pretty big gap. And so that's a gap we don't want to create. We want having a stripped down version of the library or the internet or. Yeah. And, or just like, you know, older models, lower intelligence, lower tool use, like all of these things. And so, you know, to me, it's like, yeah, maybe we still have a lot of users, but I want those users to have as much intelligence as possible so they can do things in their life.

14. 48:55

    And I think having ads to support that will be neutral to an enhancement if we do our job well. I just kind of wish that all of you, Anthropic included, would just actually say the business reasons for this too more. I get the democratization access thing a lot, but I mean, you guys got to make money. Most of your users, they cost you money. right because they're on the free tier so we spend a fortune yeah so this was inevitable that this was going to happen and by the way anthropics thing i noticed they say we may do ads in the future and the thing where they're kind of yeah i mean as elise was saying you know people leave themselves some openness and the funny thing the funny thing in this is like we went out and announcing ads with very strong principles they went out criticizing ads but leaving the door

15. 49:40

    very open for them to do it and saying they're going to do commerce and shopping in that same blog post. But like, isn't there an argument to just saying like, look, we have to pay for this. And like, this is the way we're going to pay for this is we're going to do ads. And you're not going to have billions of people paying for ChatGPT. Just no consumer, you know this well, Facebook, like no consumer product reaches that scale without ads. And so just kind of embracing that and saying like, that's what it is. And like, yeah, Sam called it a last resort a couple years ago. But like, you all are a huge company, and you're going to probably IPO and you need to make