Balaji on Why AI Raises the Cost of Verification

Balaji Srinivasan (0:00)

AI doesn't take your job, AI makes you the CEO. The problem is, AI is a shortcut, and a shortcut is good, except when it's bad. If you don't know how to go the long way around, then you can't debug the AI.

Erik Torenberg (0:10)

Do we not think that AIs are just going to be also better at taste and agency?

Balaji Srinivasan (0:13)

I don't think that's true on a short-term basis. Humans are the sensor, A is the actuator. So it's like a human-machine synthesis. What's taste? Taste the sense? And that is what AI can't yet do.

Erik Torenberg (0:23)

What happens when AI really achieves its potential? Will LLM get us to AGI in some capacity?

Balaji Srinivasan (0:28)

No, no, actually, I'll adopt it.

SPEAKER_3 (0:31)

Every tool that makes creation cheaper makes verification more expensive. The printing press made publishing easy and forgery easier. Photography made documentation instant and manipulation inevitable. In 1839, the first year a camera could capture a human face, people trusted photographs absolutely. Within a decade, courts were already debating faked evidence. The cheaper the creation, the harder the proof. AI has compressed this cycle into months. A resume that once took hours to fake now takes seconds. A slide deck that signaled competence now signals nothing. The generation cost has collapsed, but someone still has to confirm what's real, and that cost is rising fast. The result is a world that fragments into trusted groups, where AI supercharges productivity on the inside and raises walls on the outside. I speak with Balaji Srinivasan, angel investor and entrepreneur.

Erik Torenberg (1:36)

I want to start by talking about the AI economy, and I'm curious if you think it will look more like the internet economy, where applications take most of the value, or the cloud economy, where there's kind of infrastructure takes most of the value or it's more distributed. There's an argument that the big labs will take it all because they have all the capital, they have the compute, they have it integrated, but there's also an argument that, hey, maybe they won't because distillation is 98% cheaper than it is to build a model. Open source catches up and apps control the user relationship. How do you think this economy is going to play out?

Balaji Srinivasan (2:07)

Great question. So I do think that at least a very large percentage of the future is going to be distillation and decentralization because, as Anthropic said, distillation attacks work on their thing, right? And so a relatively small number of API queries helps to kind of distill a large model into something small. And it's very hard to stop that, right? Because you're stopping queries from coming back. You'd have to somehow detect that or what have you, right? And it's also hard to morally stop it because what do they do? They copy the whole internet and put it into their thing, right? So talking about stopping the copying, it's like Facebook or LinkedIn stopping someone from scraping what they scraped, right? Like Facebook scraped all these Harvard social networks or Google scraped the entire internet, build a Google index. I get why they want to do it, but it's hard to support that. Okay. So the other thing is, I think the future is personal, private, programmable because AI is so powerful that you want to use it within the trusted tribe for a variety of reasons. First is, it doesn't miss, okay, or rather, it doesn't miss small things in large datasets and things that were effectively secure through obscurity. A small example, but an important one, is the Gmail thing, right? Like the Jeffrey Epstein thing, you can query, like this guy had never thought that all of his emails would be publicly indexed and searchable by AI 10 years later or what have you, right? So you can issue queries that will synthesize information across thousands of emails or whatever, and build a story right then and there.

SPEAKER_4 (3:42)

Okay?

Balaji Srinivasan (3:44)

So what that means is it's not just surveillance, it's what the French call suvalence, surveillance from below, or even the Jeremy Bentham Panopticon, where everybody's watching each other. Any information that's in the public gets indexed and then put into these AIs where people can stalk each other and so on and so forth. And then what that means is the Commons becomes a hall of mirrors with all kinds of pseudonyms and so forth, and people retreat back to caves and tribes. Okay, so within that trusted tribe, yes, if you share all your code, within the trusted tribe, you share your whole code base, boom, you can zip along. And so AI increases productivity within the trusted tribe. But outside the trusted tribe, aren't you getting a ton of AI spam? And AI spam emails, AI spam replies, right? Low quality slide decks that are sent over. People will send me these slide decks, and I love AI. Okay, and you know my reaction is to seeing AI in a slide deck?