Transcript
0:00 All right, welcome back. The debate over the risks and rewards of AI reaching a fevered pitch this week with the former anthropic researcher going public with his own concerns. His warnings drew a sharp rebuke from our next guest, Brad Gersonner. He's the founder and CEO of Altimeter Capital joins us now. He's an investor in both Anthropic and OpenAI, as well as all of you know by now in several of the hyperscalers. Good to have you back, especially in this to have a conversation in this important time.
0:29 well, it's good to be here, Scott. And first off, you know, I can't believe it's been 25 years. I was watching those images this morning. I know you were around I was around New York City at that point in time and and here we are 25 years later and my prayers are still with so many of my friends and families that lost people that day. So, thanks for having me on today. >> Yeah. No, it's it's a it's a pleasure to have you here. appreciate those words and of course we share all of that with you and everybody. Let's have a conversation about what I think has become the national conversation, the risk and rewards, the promise and the peril of AI which I feel like picked up this week with the the Jacob Coxin from Anthropic, the researcher who quit and posted the following on social media for those who don't remember or didn't see it. I resigned from Anthropic today, he wrote. I spent the last three years doing pre-training research at both OpenAI and Anthropic. Neither company is acting responsively. They're racing straight to self-improving super intelligent intelligence and gambling with our lives. The people building AI earnestly believe that it could kill us all by the end of the decade. You responded to that on X saying, quote, "Ridiculous hyperbole from an exjun employee who worked a total of 6 weeks at Enthropic, now calls for regulatory intervention. No thanks." And a couple of emojis a after that. I can understand why those in and around the AI ecosystem would want to both discredit and disparage this person for saying what he did. But he is not the only one who has either posted or written about concerns about AI, especially self-improving. I'm sure you saw the Paul Tudtor Jones op-ed from the Wall Street Journal where he said that AI may become the third superpower.
2:28 Quote, "The threat is real." What are the chances that very soon an AI model reinventing itself thousands of times over makes one mistake that is harmful if not devastating to humanity? Now, multiply that risk by tens of thousands of users. How would you respond to all that? Well, well, first Scott, I would say that we should always have open debate about the risk and reward of technology. And this discussion's been going on for a decade. There was nothing new this week. In fact, I've not talked to a single person in AI who says full steam ahead disregard safety. It turns out we all care about kids. We all care about the future of the country. We know the massive benefits it offers, but we also want to take the extra time to make sure that we do this safely. But what I what I didn't like this week is these hyperbolic scare tactics which I think are hiding behind a political agenda and we ignored all of the extraordinary steps already be already being taken unlike in the age of the internet unlike with social media to get ahead of these downsides. So let me just share a few.
3:35 Number one, the companies themselves are investing a massive amount of money and compute in making sure that we build sophisticated safety systems. They're deeply coordinated with the national security apparatus of the federal government to make sure that we secure our national security and other safety issues. And they're issuing transparency reports like we saw out of anthropic yesterday. Second, the Treasury and Commerce Department this year already took the unprecedented step of shutting down Mythos because they were concerned about potential cyber security hacks.
4:11 So, they hardened the systems and force them to put it on the shelf until they re-released it. They also had an EO around the model release framework, which is effectively a regulatory scrutiny that occurs before any Frontier model is released. In fact, you know, there was a lot of celebration this week about the model that cracked Navier Stokes, right? That is an Astra plus one model working inside of OpenAI. Ask yourself, why doesn't the public have access to this model? Because it's undergoing scrutiny by the lab and by the government. They're not just going to release these models. Mythos was completed at the end of January and just now is rolling out to the broader public. So, we're taking a lot of time.
4:54 And then finally, broader industry is coming together. You know, Jamie Diamond, Michael Dell, myself, many others, dozens of CEOs coming together to make sure that we're coordinated around our critical industries and making sure that we're taking safety very seriously. So, I don't think we've invested this much time and energy in safety before a new technology in my 25 years in Silicon Valley. And if you listen to the echo chamber this week, you would think that we were h hurtling ahead you know with total disregard to safety and it's simply not true.
5:30 >> But but how would you then respond to the fact let's just say you know Daario Emod himself back in January posting writings about the the warnings of of AI. Those who are leading us into the future with the models they're building are also seemingly raising the warnings that they think exist and questioning whether we're doing enough to deal with it. Sam Alman just just recently said he's that he's telling staff that OpenAI is is open to slowing AI development, that he's pushing for mandatory national AI safety requirements in the US, citing concerns that advanced AI systems could accelerate its own development. That that says to me, >> Scott, doesn't that make you feel great?
6:21 Doesn't it make you feel great? Did you ever hear that out of Mark Zuckerberg about social media? Did you ever hear him saying that I'm so concerned about what this might do to America's kids? I think it should make you feel good that America's leaders, starting with Elon Musk back in 2016, said this will be the most positive thing that we can ever build for humanity, but we have to do it safely. We've been having this conversation in Silicon Valley for a decade. The conversation continues. It's an important conversation to have, but to go on national television in a highly coordinated way and sit in the living room on NBC, on ABC, on CBS, on Fox and have a single researcher who spent 6 weeks at Anthropic tell the moms and dads of America that this can kill all of humanity in the next 3 years without any counterveailing, you know, conversation is deeply irresponsible.
7:17 Look at what people said about it. Jensen Huang yesterday called it deeply untrue. Andrew Feldman, the founder of Cerebras, who spent decades studying this, called it utter horseshit. Elon himself, right, who's concerned about safety, said it was a coordinated SCOP. So my only concern here is that we have a balanced conversation. This week has not been a balanced conversation. Everybody who's invested in AI, everybody who's leading these companies, the government takes AI safety very seriously. That should make everybody at home feel very comfortable about let me say one other thing. We act like there's no risk in the alternative.
8:00 So I would ask these, you know, these folks who are causing this alarm, what would they like us to do, right? What what is the alternative? What are the things that we're not doing that they want to see happening? My sense is they don't have a very good understanding of all of the things that are happening in government, all the things that are happening in the labs. But if the idea here is that we need to unilaterally disregard to all of our potential adversaries around the world who by the way are not going to slow down or stop, right?
8:35 Roll the clock back 70 years. We may not have wanted to pursue the Manhattan Project, but we knew our adversaries would whether or not we did. This is a matter of national economic security. It's a matter of national security. We have great track records. Despite the hyperbolic claims, remember all the claims around nuclear that it was going to kill all the kids, that Ralph Nater that it would wipe out the city of Cleveland. Not a single person in this country died or was injured by nuclear energy. But we shut down 67 nuclear reactors. We unilaterally disregard disarmed while China's building 100 fision reactors. Why did we do that?
9:15 Because of alarmist activists that went uncountered. And all I'm asking here is for a balanced conversation that looks at both the risk and the rewards of AI. And I think we're we're moving forward safely. I >> I understand. But when you're in a in an environment where if you look at the polling and they suggest one after another that a growing number of Americans don't want data centers in their backyard, so to speak, don't trust the technology.
9:45 Certainly, they're not all fools, who don't understand where all of this may be going. They see an industry that's racing ahead. They hear us talk about IPOs. They see people getting rich. They worry about this technology that's coming into their backyards and they think causing their water bills, their electricity bills to go up and god forbid take my job. And that's why there's this growing level of distrust. >> So, and correct. Correct. And by the way, we need to do a much better job in a very reasoned and fact-based way of explaining what in fact is happening.
10:24 Right. The Economist is out with a story just this week that despite all the doomers who said 10% of all jobs would have gone away by now, in fact, AI has created a million new jobs in America, right? We learn that all of this catastrophizing about data centers may in fact have had support from the CCP. Right? So, I want a little investigation. Who is behind all of this negativity? Is it coordinated? Right? Where is the money coming from? I don't see anybody who's saying, Scott, like you did, you know, full steam ahead, disregard safety, etc. I think it's a false narrative. I think it's a straw man argument. It's simply not true.
11:07 Every single person in Silicon Valley has kids. We take this very seriously. We want to usher in a safe world of AI. But that doesn't mean that we need to stop, shut down, overregulate because of the fears, the hyperbolic fears of a few who have their own political agenda. We've seen this in every technological revolution, right? And like I said before, we've shut down really important industries, giving up massive upside to humanity, you know, before it even takes flight.
11:40 I want to address the the role that some suggest that technology the industry has played in in getting to the point where we're having this national conversation and where this level of distrust I think we can can both agree seems to be only on the rise. I want you to listen to what the Treasury Secretary had to say at the G20 recently and we can react to that on the other side. Here's Scott Bessen. I think that the AI companies, whether it is the builders of the data centers, whether it is the labs themselves, have done a horrendous job, horrendous job of explaining themselves to the American people. And I think we need a big reset on this. They're they're going to have to take some of the blame and they are going to have to convince the American people that all the benefits will not acrue to a small group.
12:36 >> How do you respond to that? I totally agree with him. you know, listen, that it brings up a totally separate issue, Scott, which I think may be behind some of the distrust and behind some of the discontent, frankly, with AI, which is this idea that we have this everexpanding wealth gap. The benefits are going to a small group, and they're stuck with the anxiety that maybe their kid won't be able to get a job. The answer to that, as you know, what I've been working on for a long period of time here is we have to get everybody into the game of capitalism. They all have to share in the upside of AI. We haven't done a good enough job, but stay tuned. We've created 9 million capital accounts for kids across the country, as you saw, Michael Dell, 6.25 billion. I think all the AI labs will make huge contributions to this. I think we'll expand it to include all 70 million kids under the age of 18. That's an important step to make people feel like they have skin in the game. They're aligned with the upside of America. It's not the only step, but we need to take concrete actions to address these concerns. One of the concerns is an economic concern.
13:46 Another concern and anxiety is about their kids and jobs. So, we need to make sure that we're doing job retraining. We need to make sure that they know and understand a million new jobs have been created. There are shortage of trades workers all around this country because we have this massive building boom going on with data centers across the country. Electricians in Texas and Wyoming are making $250,000 a year because we don't have enough to build the capabilities to, you know, that will power the next wave of economic growth in this country. So, I agree with Scott. We need to do a lot better job. But that is a very different argument, like I said, than going on national TV this week with a very coordinated effort to say that AI is going to kill everybody in 3 years. By the way, who pushed back on them?
14:35 Explain how that is going to happen. You know, Bill Gurley posted something about the five W's. You know, walk us logically link that together. They are lost in a scientific future, a sc a sci-fi future while we're facing real and present challenges today. And so I'm not saying that there aren't real risks ri there there are like with any new technology, but I think there's a lot more being done around those risks than we've done in the prior super cycles.
15:06 And I want people to be aware of it. the the threats to some aren't going to sound that hyperbolic that have been articulated this week. When you read what Anthropic itself put out just yesterday, 154 pages detailing instances and efforts to misuse AI. They said they blocked possible efforts to build biological weapons. That a group in Yemen used Claude to try and develop missiles.
15:37 That doesn't sound like a politically motivated hyperbolic threat. That sounds like >> Hold on, let me finish. I I'll give you I'll give it I'll give it to you. I'll give it to you. >> Great. That sounds to me like holy, you know what? Do we fully as a society as as humanity understand what is possible and capable of AI >> and how do we deal with that both on a governmental standpoint and an individual standpoint?
16:11 Well, I first I hope you know and understand that these same efforts have been going on with Google for 25 years. Don't you think that our adversaries looked up on Google how to make a bomb, how to better target rockets, how to do all these things? Of course they did. This has been going on for a long period of time. And before Google, they were using other mechanisms. So, am I surprised that adversaries are doing adversarial things trying to use these technologies? I'm not surprised at all.
16:40 Does it make me feel great that the company has safety systems that identified them? That the company has the transparency systems that opens them to the public for scrutiny. Those things should give us deep comfort. But let me take on one of the things they highlighted in the report specifically and this is Alibaba undertaking one of the largest distillation industrial scale distillation. They weren't the only one, right? But massive distillation going on. I think if we could do one additional thing, right, that the government ought to be focused on when they're looking at these frontier models, I do worry of these massive distillation efforts because I don't believe the Chinese are going to scrutinize themselves to the same safety standards that the United States is going. And I don't want to hand our adversaries capabilities that we shut down ourselves.
17:35 Right? This is a world that has all sorts of risks. Yes, there are risks that AIs can get out of sandboxes. Yes, there are risks that AIs can jailbreak. But there are also risks that if we allow our adversaries to develop AI that is far more capable than our AI, that our kids die on battlefields in the Middle East or around the world. And so this is about balancing those different risks, being responsible about them. And I take comfort from the fact that this company has such sophisticated systems that they identified those potential hacks. They stopped them and then they had the confidence to share them with all of us.
18:14 >> And you and you don't have any any doubts going forward, for example, that Anthropic and others are going to continue to have the ability to block what they say they just did as the technology continues to evolve and get even smarter than it is today. Of course I have concerns, Scott. Right. The future is a distribution of unknown probabilities and we can never eliminate all risks. We haven't. Right. Remember when we split the atom? We were told for 50 years that it was going to wipe out humanity. I grew up climbing under my grade school desk because we were going to have a thermonuclear attack that was going to wipe out the country. Okay. We live in a world of inherent risk. I do not believe that standing down on AI unilaterally withdrawing from the world of AI, I don't think that makes any sense. And when we have activists that go out with these scare tactics, that's the danger. I don't believe he understands how Washington works. Look at the calls in Washington this week in response to somebody who worked at Anthropic for 6 weeks, right? Look at the calls for a moratorum to shut down AI and all of this regulation which we all know the government has a very poor track record of right just because you have an FAA doesn't mean that you don't have plane crashes right just because you have a nuclear regulatory agency doesn't mean you don't have right a a a a threemile island and so for me I want balanced regulation I think we have incredible regulation the I I mentioned the model the model frontier framework that we have or the frontier model framework that we have that's already scrutinizing the models.
19:58 That's why mythos was in fact shut down. So I just want to have a balanced approach. I didn't think the conversation this week was particularly balanced. so I want to come on here today and offer the other side. I want to give people comfort that the labs themselves take it seriously. All the hyperscalers take it seriously. Everybody building the silicon takes it seriously. Everybody in government takes it seriously. The broader industry, I'm talking the people who run our airlines, the people who run our infrastructure, the people who run our financial services industry, they are all talking and coordinating in a way I haven't seen before. It can be even better. I'm working every day to help them make it even better. but mom and dad at home, my sister and brother who called me after seeing this on the nightly news and were worried about this, I told them the same thing I told everybody else. I think it's deeply irresponsible, very hyperbolic. His goal and aim may have been to get people's attention. but he did it in a way that I think is is is dangerous and and wasn't needed.
21:03 >> I'm glad we had the conversation. I know we'll have more like this in the future. I appreciate the time very much, Brad. Thanks so much. Thanks for having me on, Scott.