Nvidia, Anthropic Weigh In On AI Safeguards

Video Transcript

Let's start with Jensen. I'm curious. Do you have a takeaway there?

Yeah, I do. I mean, not to be disparaging of the interview, but, you know, he answered the question he was asked, which is, if a customer was using uh Nvidia's technology, which is the the underlying compute for both training and running models to do something that was bad, would you take them away? And his answer was, well, that is a hypothetical. But yes, if they were doing something bad. And, you know, what he goes on to talk about is that that's not how it works, right? You know, I think that what's been amazing is tracking Nvidia so closely over a a long period of years is that it was never Nvidia selling directly to the world. It is never worked like that. Basically, Nvidia sold its technology through companies like Dell. So, you would go out and buy a server, and you would do that business with Dell. Um you wouldn't be like, hey, Jensen, I need like 100,000 GPUs, stat. It's just not how the industry was developed. But it's kind of become that way. And so that's what he was asked. He was asked, if you had a customer, a big user of your compute platforms that was doing something bad, would you take them away? And his answer was like, of course we would. But that's not how how those technology companies are set up.

Okay.

Okay.

What about China, Ed? Okay. Because my goodness. That's the real question, right? Like, wouldn't that be a good example of a a a country that a lot of people argue is an adversary when it comes to this technology? Yet, investors want Nvidia to sell its products to China.

Yeah, I mean, we're we're talking specifically in China in the context of GPU access, right? Not as opposed to the models. So, look, um there there is a range of views held by this administration and by industry in this country, which is at one extreme, we shouldn't send any uh chips to China, no matter how um deprecated they are versus the cutting edge, or whether we do send the cutting edge. There there is that school of thought and at the other end, it was like, if you don't send American technology to China, then they will come up with their own, and then they will start exporting their own technology to the gap in between, which is other markets excluding the US. But that's distinct from the current export controls that this administration put in place and built upon from the Biden administration. Those export controls were designed to stop China having access to cutting-edge chips out of the concern they would use them to develop AI in the military context, not as an economic adversary. Slightly distinct and separate arguments. But there is there is a difference between economic opportunity school of thought that's like, what happens if we don't sell our best tech to China? And there is the, what happens if we did give them our best tech? How would they use that in a way that was counter to the safety and interest of this nation?

Hey, listen, before we wrap up with you, Anthropic, uh, their investors meeting. What are we going to find out realistically here today? And I know you're already kind of on edge about the AI trade right now again.

Yeah, I think I mean, so the reporting is that there is an investor meeting happening here in San Francisco. And I think like a lot of of what people want to understand is the real financials, right? Real revenues from the third quarter, which is the period ending September, or an updated revenue run rate. So, revenue run rate as of July was 65 billion. That is, if Anthropic kept booking revenue at this pace for the next 12 months, they'd end up with 65 billion dollars of revenue. And the next layer beyond that is that if you are a student of what Anthropic is doing with its products right now, they've dropped prices, token prices really significantly of late. What does it cost them to produce a token relative to what they charge? And that then spits out what the bottom line looks like for this company. Um, and the last thing I'd point out is they have a lot of money on their balance sheet. So we're waiting for this massive IPO, but don't forget that they raised like tens and tens or hundreds of billions of dollars earlier in the year privately.

How much does it cost to produce a token?

Well, so so this is the thing. I've been posting about this on X and like I I don't want to go through all the numbers, but it's basically halved um what Anthropic charges for any given model from the summer until about last week when they did Haku 5.5 and the latest variant of Sonic, they they cut token prices. In other words, a dollar per million input tokens, a dollar per million output tokens. But at the same time, we don't know how much it costs Anthropic in terms of running compute to generate the tokens. The numbers I gave you were what it charges customers. And so that's what all the analysts and investors want to model for. Like the economics of the entire business that the biggest IPO in history so far probably is going to be predicated on. And that's fun. Um, but you need some data to to inform that.

Data would help here. Uh, we want to know more.

All right, Ed Ludlow, we always learn a lot with you. Uh, host of Bloomberg Tech, catch him Monday through Friday at 11:00 a.m. Wall Street time on Bloomberg. Let's uh stay with AI.

Yeah, we promise we'd do a deep dive here. I want to bring in Jacqueline Corbelli, CEO of the Interactive TV app platform Brightline.

She's also advised global leaders on AI policy, including at the UN and in the Vatican, and is heading to meet with King Charles of England to discuss policy next month. She joins us here in the Bloomberg Interactive Brokers studio. Welcome, welcome. What are these leaders asking you when you meet with them?

They're asking for perspective. Look, I think one of the things that stands out when you're when you're advising global leaders is the, you know, the ability to think in practical terms, in pragmatic terms um about the change ahead and what is required from a policy standpoint now and give them something that they can reflect on as they really for the most part these days struggle um to develop policies that balance the in in the in the case of AI, balance out the security issues with with the with the potential for progress and commercial upside.

It's it's not an easy thing for them. Um, there are a lot of people analyzing these things. It's a question of how to synthesize it. My role is really to bring in the sort of practical on the ground experience. Um, the sort of real life considerations and uh yeah it's a it's a it's a privilege to do it.

Your work though, you know, with the Vatican really kind of caught our attention. I was you know, um, in our planning for this, our team, the New York Times recently wrote about how religious scholars have been meeting with Anthropic and that for months Anthropic, this is again from the New York Times, has been hosting private meetings, shuttling in dozens of religious scholars from across the world, papering them with non-disclosure agreements and demanding that key aspects of many conversations remain confidential. Um, but they're talking about stuff. Um, what is it that and I I know you probably can't reveal too much, but curious about that component in what is probably one of the biggest debates and technological disruptions of of our time.

Yeah. I mean, I think at the center of all of that is this idea, a couple things really that are all rooted in the same thing. It's it's humanity per se, um, human agency, uh, what is it going to mean to the dignity of people? Uh, and

But money always clouds sometimes those. So how do we how do we reconcile it? If you're you're dealing with leaders who are thinking about, we think about it too. If you're public, every quarter we're going to be looking at your financials.

That's right. You know, and then there's like the heart and what you maybe should be doing, morality. Yeah, which uh which investors don't care about. Which investors don't care about. Absolutely. And and by the way, you know, world organizations don't necessarily trust and and it's uh, you know, it's not it's not unfair.

Uh the a lot of the decisions that are coming out of Anthropic, uh OpenAI and others, don't necessarily come from a place of ethics, right? They come from a place of sort of commercial viability ultimately, uh fear, right? They're motivated by fear and this is something that is not lost on uh religious uh religious uh leaders as well.

What do you mean when you say the biggest risk, global risk is outsourcing human agency to artificial intelligence? What does that mean?

I I what I mean by that and I actually don't think it is as much of a risk as many do. Um but to, you know, to put it to put it bluntly, we rely on technology so much these days and we're caught in a moment where we are not prepared as humans to get out in front of the agents, to get out in front of AI. and that's where human the ability for us to be human and for us to um direct this in a human way become questionable, right? The change is just so fast. And by the way, it's a huge time of change management. We haven't really shown a real adeptness as as a race as a species for changing very rapidly and this is requiring that.

Yeah, there's race and then there's leaders. And I think about the responsibility of leaders, whether it's government leaders, whether it's business leaders. and again, um I've never felt so much where doing the right thing is not a priority. And whether it's the gaps between what the sea the sea suite makes versus what most workers make, the K-shaped economy that we talk about, maybe the market's at highs, but not everybody enjoys it. What does that mean now when it comes to some of the work you're doing and especially as it relates to AI?

I I think the reality is that we are getting a uh projection of social responsibility from uh tech billionaires who uh are showing some kind of concern for where this is all going, but we also all know that there's some question as to what that's really rooted in. Um from the standpoint of where this is all headed, it's a question of what are what are we prepared to do? So, as AI charges forward, we need the legislation, the institutions, the international cooperation to move with it. And that's where I think it's going to be difficult.

Just um Jacqueline, just got about 25 seconds. you know, you are dealing, you know, you know, whether it's King Charles or you're headed to the Vatican and then you've got corporate leaders.

Um, you know, I do wonder who has more willingness to really make to make sure that there are the guard rails really quickly. Is it more religious leaders or the corporate world? Real quickly.

Religious leaders. Religious leaders, world leaders and the corporate world.