Anthropic Is Now Worth Nearly $1 Trillion - Warning Shots #44

Added: 2026-06-01

[00:00:00]Hey everybody, welcome back to Warning Shots. I am John Sherman, your host, joined by my friend Michael of the Lethal Intelligence YouTube channel by my friend Lon Shapiro of the Doom Debates YouTube channel. Every week we get together to talk about the headlines and AI risk and AI harm. Talk about what's warning shot, what's not. We are three dads running three YouTube channels trying to make a safer world out here. Big week as always. I feel like they're all big weeks, but uh this week we got a big thing with the Pope coming in with his uh encyclical on AI and as one would I don't know if you guys read it. Did you guys actually read the entire text of this encyclical thing? I did not. Michael, did you I see no maybe a little bit >> sum summaries of it, you know, >> I did the trick where I pasted it into chat GBT and I was like, give me the two-page version.

[00:00:45]>> As did I. As did I. I dropped it into chatgbt and I was like, "Well, you know everything about me, so tell me what is important to me in this encyclical." And I'll just read you all what it came up with here. Just top just sort of topline. It said, "The Vatican explicitly rejects the idea that intelligence is just information processing. Uh, the Pope is framing AI as a labor issue, not primarily a technology issue. The Vatican is quietly attacking the idea that AI is neutral."

[00:01:10]Uh, and the Vatican may be the first global institution treating AI as a spiritual threat. And then it says the bottom line you it it would hook the segment around is the greatest danger of AI may not be from may not be that machines become human. It may be that humans become more machine-like. So that is what chat GBT told me the pope said obviously a big deal because you know a whole lot of Catholics on this planet we need uh spiritual leadership in this. So I don't know I was thrilled to see it.

[00:01:37]Lon what was your sort of top line about the pope?

[00:01:40]>> Yeah uh well you know that's such a highdimensional space of views right?

[00:01:43]It's always interesting when somebody comes in with their own position, you know, their own angle. So, obviously, I'm an atheist, right? So, I I don't necessarily think that the Pope's um I don't know, philosophical authority on this is that high, but uh I I will I'll give you my most positive spin on this.

[00:01:58]Out of all the things you could say about this, I think it's good that somebody who represents people outside of Silicon Valley, right, Catholics, right, or whatever group he represents, is coming out there saying, "Hey, here's my position on this. I feel like my group, my large group also has a right or a license to help be part of this process, right? And I do think Silicon Valley is like running so fast, avoiding all regulations that when other people stand up and it's like, "Hey, can I have a seat at the table?" I like that aspect of it.

[00:02:29]>> Yeah. Yeah. Uh Michael, uh you're a little bit closer to the Vatican out there in Greece. Not not, you know, a little closer.

[00:02:35]>> Not really.

[00:02:37]Uh I think that's uh what really stood out with me is that he described investments in AI powered weapons as feeding a spiral of annihilation a phrase that lands re with real weight.

[00:02:47]What if you think about it? So he pointed out how military budgets are ballooning especially in Europe while education and healthcare get squeezed.

[00:02:55]So all to enrich a small group of elites who have little stake in the common good. And uh the the pope is right to sound the alarm and he's doing it from a moral platform that reaches billions.

[00:03:06]It's significant. Uh but the deeper danger that the pope's warning points to us is like this is the training ground for something far worse. Now imagine giving a super intelligent mind, one that can think millions of times faster than any human general um that has perfect recall of every past war and every possible strategy the keys to the world's arsenals. And so not because we're planning for it, but because each side keeps building just little smarter, you know, autonomous system to stay ahead. It starts with helpful targeting assistance. it will end up with an entity whose goals are not no longer fully understandable or controlled by its creators. So that's the spiral of annihilation.

[00:03:42]>> He's real concerned about them that military stuff. You know the p the focus on the military stuff is really interesting because I think we tend to sort of look away from you know it's not the first concern of extinctionist people.

[00:03:52]>> Exactly. And I mean the pope's voice is a powerful reminder that this technology is not neutral. It amplifies whatever values or lack of values we bake into it. So if we don't deliberately steer it towards preserving human dignity and peace, it will steer itself and the spiral will will get worse, will tighten. I got one thing for what do you guys think about this? Have you read this at all? So within US law and I want someone one of our viewers. Maybe try this and let us know in the comments how it goes. If you are employed in an American workplace because of the Pope's encyclical, if you were required by your employer to use AI, you could voice a religious objection and say, "Uh, based on my religious objection, I will not and cannot use AI in my employment.

[00:04:37]Also, you must keep my me employed here or you'll be firing me over illegal religious discrimination." Who wants to be the first to try that?

[00:04:50]like you know I think someone could really do that >> maybe people don't even are not even familiar what the ensyclical is even means this is basically primarily used to define the doctrine or offer moral guidance uh on significant social political and moral issues. So you know it's literally if you are a believer and it's your right you know to say okay because my my lead my religious leader has is talking about this in that in that certain way I'm I'm not comfortable to to do this. It's like asking a Muslim to eat, you know, >> yeah, pig. You know what I mean? But it's so something important is shifting actually.

[00:05:24]>> So religious objections are a start.

[00:05:26]They're a signal that uh even ancient institutions are waking up and saying this feels wrong in my bones. Um but they're also a symptom of how weak our real safeguards are because this not this is not real safety, right? We're not going to be um Anyway, I'll come back to this soon. Maybe Lon wants to share a few. Lon, what? Lero, you have a lot of employees. What do you do if one of them says, "Uh um, Mr. Shabir, Mr. Spear, >> I'm out on that AI stuff. I'm a Catholic."

[00:05:56]>> Yeah, I definitely filed this under uh uh tough issues with having employees because I used to have a lot more employees across many states than I do now. And man, it's the craziest stuff, man. you know, that they're asking the employer to like help garnish their wages for like their alimony for for like years ago and it's like $10 for paycheck and it's just like all this stuff piling up that you know all these states like this random city that some employee is working from wants you to do a bunch of stuff for them. Um being an employer with employees is not that great of an experience. So luckily I don't have that many employees these days. We're running a leaner operation.

[00:06:29]But yeah, I mean if one of them was like, "Hey, >> I can't uh use AI or whatever." Um, you know, it's like how much am I gonna compensate the job? I mean, the the problem is that it's like, okay, so now you're just going to be a lot less productive, right? So, I don't know how much I would accommodate that as an employer.

[00:06:45]>> Yeah. And I man uh uh uh Shapiro versus employee lawsuit uh in coming.

[00:06:54]>> Seriously, >> we'd have to see it. All right. Someone out someone one of our viewers out there, give it a shot. Let us know.

[00:07:01]People people wonder why employees don't get paid enough and then employers are dealing with like all this overhead. You know, that's where the money is going to to paying employees to deal with this crap.

[00:07:09]>> Especially the few employees uh Leon has left. I mean, I'm sure you're watching this, so now you know what to do.

[00:07:20]All right. All right. Let's move on. So, we have next thing up is is a lot of bad news for business related to AI. I'm just going to tick off a couple things here. Microsoft canceled its clawed code licenses, citing the cost. Uber burned through its entire 2026 AI budget in 4 months. Uh Uber COO publicly said AI costs are harder to justify. A Fortune 20 CEO ordered token spending dramatically slashed. One company rumor is it's Amazon spent a half a billion dollars in a single month on Clawude because there were no usage limits and the stocks remained at an all-time high.

[00:07:55]But uh a lot of um implementation drama happening. There's another one where a Pizza Hut franchisee is suing uh uh the uh OpenAI I believe because they screwed up all the pizza orders and it screwed up a bunch of orders at Starbucks as well. Michael, everybody says we need to get this in all of our businesses and it's going to save our businesses. We're going to be operating so well, but it turns out it's actually a little bit harder than just sort of plug and play here for a lot of folks, >> right? And um I'm just going to use an analogy. So imagine like a group of engineers frantically building the the tallest tower imaginable on the found on the foundation that's already cracking.

[00:08:32]So every floor they add it makes the structure more impressive and more but also more unstable. So the view from the top like the AGI level capabilities looks revolutionary but the ground below is giving way under the sheer weight of the concrete. So that's today's ice investment story. You know the technology is real, the infrastructure is real, the eventually useful applications will definitely be real.

[00:08:53]But this this changes everything eventually and this pays for itself on schedule are two very different claims.

[00:09:00]And right now the economics are screaming that the second claim is in trouble. But uh actually this this pressure this economic pressure you see is uh is a reckless spring towards super intelligence systems is basically it's just bad news on in terms of cutting corners on safety. That's the dangers because desperation is a terrible engineer if if the if the money tap starts tightening what's going to happen just going to cut you know cut corners on safety testing alignment research just to keep the hype alive and the valuations propped up >> so the pressure to to ship faster doesn't disappear when the economics get ugly it gets worse >> and I mean Lon I know you love this stuff I know you you you you genuinely love the AI and so uh how do you feel about this is this is this uh do Do you think they're just screwing it up or or is there a real problem here?

[00:09:48]>> I can give you the economic analysis. A lot of the news that you're hearing is just like this short-term bubble because we just got to the point where the AI is just automating more and more. Like agents are really working, you know, more and more of a programmer's job is being automated. All these other, you know, analysts that put together reports, they're being more and more automated, you know, from 90% to 98% to 99%. And this is all new to the extent that actually the tokens are still somewhat expensive, especially if you use fast mode as I do. Uh, so the tokens are kind of expensive. So the latest gotcha of all the people who are like skeptical about AI, like the Edetrons of the world, they come in and they're like, "Oh, okay. Well, it's too expensive. There's no profit model because you have to spend like $10,000 a month per person on tokens." But what these people don't see, maybe willfully don't see, is like, "Yeah, okay. Give it a couple months, right? This is like a new technology being introduced with very little optimization." And in fact, funny enough, 24 hours ago, Anthropic just announced the latest optimization.

[00:10:41]and they said, "Hey, people like me who are using Cloud Code fast mode and they're paying like $10,000 a month plus less to to go faster." Uh, they just slashed the price uh in thirds. So now, oh great, I don't have to pay $10,000 a month. I have to pay $3,000 a month.

[00:10:53]Okay, great. And guess what I'm going to be paying in a in a couple months?

[00:10:55]Probably $1,000 a month. You know, because there are optimizations. They're still lowhanging fruit. So, if you want to just extrapolate, if you want to go a year from now, all these stories about like, oh my god, Amazon, Uber, they're spending so much on tokens. They're going to spend less than 10% and they're going to get the same output. Okay?

[00:11:10]Okay. So, they just temporarily don't want a big line item on their budget.

[00:11:14]That's all this is. This is a blip. This is nothing. This is a distraction.

[00:11:17]>> But doesn't it, you know, let's say two years from now, if everything goes nicely and we're all around, the usage is going to go up dramatically, right?

[00:11:24]Like the number of users, the amount they're using, it goes up dramatically.

[00:11:28]So, the token bills, even though they may be, you know, per token getting less, the overall spend, I think, keeps pushing up as implementation increases.

[00:11:37]And and so if they're not really seeing dollars and cents on the other side, I think that's I think that's a part of the problem. Like what what I think like >> Let me just be very clear about my claim. Okay. So I I can confidently claim, you know, my opinion that uh the amount of uh value you're getting per dollar or let's say per $100,000 is going to be measured in increasing amounts of humans, right? So like today you pay $100,000 in the programming domain. In my opinion, for $100,000, you're already getting like two or three senior programmers worth of value. Even if it's not a drop in replacement, you need one manager, but that one manager now doesn't need to hire two or three people under them. Uh, so let's call it two to three. I think if you wait a year, it could easily be 10 or 20 and then it it'll go from there. And and I'm talking dollar for dollar. Even as Nvidia has trouble manufacturing chips and prices go up due to supply constraints, I claim that even as we have supply pressure, we're still going to see a huge surge in the amount of human equivalent employees that you're getting per dollar.

[00:12:36]>> I'll agree. But you also need to see the revenue increase as well, right? Because if you're just adding on more employees, but you're not actually, you know, or whether they're whether let's say they're they're virtual employees, they're AI employees and they're doing a ton of productive work and you're spending a lot more on tokens, but I need to see it pay off in the bottom line as well. I don't know, Michael, what do you think? We're in the is the age of token maxing uh are we are we coming to the end of it or where are we in the token spend world here? M >> no I I I think optimization would be insane actually and uh we see glimpses already with all these papers that come from the best universes like MIT they're doing research so that optimizes it very very very much want to mention like many people are talking about the bubble and they're drawing parallels with the com bubble h but uh the problem here is that in the com big companies like search engines and you know different internet products and maybe they got busted but here if we mess up safety it's going to be exactly the same Because you know if we actually go far and we create a superent agent even if it's too expensive and it's a bubble but if it's a super diligent agent that can plan deceive self-improve and operate at a planetary scale is not the same as a search engine or a social network the failure mode is not going to be some companies go bust. It's something that's far harder to recover from. Hey guys, John and Dolly out here on Baltimore's Inner Harbor telling you if you like what we're doing at AI Risk Network, be sure to check out our Substack. a lot of unique content coming on Substack all the time these days. Check it out.

[00:14:06]Everybody I talk to, like I have a friend who's in like VC world and and uh they are just all so enamored with Anthropic and and the way it's generating cash and the you know how much money it's bringing in and it seems to me a lot of it is fueled by everybody just going hog wild with these tokens and just spending like drunken sailors and nobody's watching anything and and we it's all new and like you know it's the wild west and and I think we're seeing the sort of first like notes of can we really sustain the you know it there needs to be a relationship between the bottom line and what we're spending on AI. And it seems like right now they're totally disconnected. Everyone is like just go nuts on AI and we'll see it pay off on the bottom line eventually. And as we sort of get into that gap period, people are asking questions. The bosses are asking questions. Is that fair, Leon?

[00:15:00]>> Yeah, it's fair. But it's like I I guess the the the central claim here is what I mentioned on the show where I I claim the pie is growing, right? So if people are looking at like the current balance sheet of companies and being like where's all this value is going to created and I claim it's because every company is going to actually deliver more products and services, you know, everybody's going to be doing pulling forward their expansion plans. Uh and you could be like no it's all in your head like none of that's going to happen. But I personally have seen enough evidence to feel confident about that thesis. I mean I'll give you an example. If you told me, hey, go create another company from scratch. Um, I could create two companies from scratch right now as easily as I could have created one company from scratch a year ago, right? Because I'm using so many AI assistants, I could l Yes. I could do twice as much stuff.

[00:15:43]>> But, so here's my question. So, let's just say there's a consumer out there and they have $1 and they were going to spend it at your first company, but now you created two companies. Where do they come up with the second dollar?

[00:15:55]>> This this is this is how the economy works, right? because they're going to be at an employer which is also more productive. Right. Every time the pie grows, you could ask >> the employer going to increase their salary, the employer is going to give them that extra dollar to buy the increased productivity result.

[00:16:07]>> Yes. Yes. Their salary is going to increase. Correct. On average.

[00:16:11]>> Okay. So everybody's >> been the trend, John. The trend of the last 100 years since the industrial revolution 200 years is that everybody makes more real dollars and buys more stuff. That's the trend. And and the same trend is going to increase. I mean, remember this show is warning shots and we do say that we're going to fly too close to the sun and all die. Let's not forget that. Okay. But I claim that until until we get to that point. I do actually claim that the economic bulls are correct. So, it's kind of funny. I'm like the most extreme, right? I'm the most extreme saying we're going to fly close to the sun and I'm the most extreme saying and then we're going to burn our wings off and and go to hell. I I will make both of those claims.

[00:16:44]>> Yeah. Michael, what where does the second dollar come from? Do you see this problem?

[00:16:51]Look, I agree the the pie is growing, but I don't understand how how it will reach the person if the person is is useless because we were talking AI agents are going to do the work, right?

[00:17:01]So, I mean uh are we talking now like some kind of UBI or how how are we going to get the dollar is going are going to be taxed like the companies or what's what's going on?

[00:17:10]>> Uh okay. Well, if I mean if your job is actually adding value to the AI. So like if you're managing an army of AIs, but you actually do type useful prompts, which is kind of like my relationship to AI today is I don't know how long it'll last, but I you know I am still adding value for now. If that's you, you're definitely going to make more money, right? Because you're you're going to do the work. You're just going to deliver more and you're going to uh affect the bottom line of your company more through what you're delivering. So they they can actually afford to pay you more. But then you can be like, okay, but who's getting unemployed, right? And I definitely think the first people who are going to get unemployed are people who their head dips below the waterline of getting 100% replaced by AI. Right?

[00:17:46]So there's like a huge difference between like, okay, well I still do 5 to 10% of my job, which is what I feel like today, right? I feel like I'm doing 10% of my old job. Like I do one AI prompt and the AI writes a thousand lines of code and I do one prompt and I get a thousand lines of code. So I'm doing like 10% of my old job. If the 10% goes to 0% then you're screwed, right?

[00:18:03]Because you're not getting a paycheck at all. Instead of getting a huge paycheck, you're getting like no paycheck. But then there might be a saving grace for a while where it's like, okay, well, you're this other layer that's like connecting AIs to the world. So like we just need a warm body here to like translate what the the genius stuff that the AI is doing. Like we need to like put it in the real world or like you know, like do something physical, whatever, be a handyman, but you know, but do all these like ambitious projects because we're going to have more building projects, right? So there's this this new layer of jobs that get created, but then of course the endgame is like, okay, well now there's robots and now like we're all unemployed and then yeah, like I hope there's universal basic income. So, so like if if you fire 20,000 people maybe you for what you're describing which is totally valid 100%.

[00:18:40]H how many are you going to need? You're not going to need let's say not even as you said 10% or even less maybe 2% of of that. So all the rest we're not going to be able to um spend on the economy. So this smells like a like a you know depression kind of that's why even though the the pie could be growing I mean there there's a serious depress depression pressure downwards in a way because people don't spend a lot of people. Yeah, I I don't see I don't L I don't see how the salaries go up, right? So to Michael's saying, so we had 20,000 people. We laid off 18,000 of them because the AI is doing their job.

[00:19:15]>> The salaries of those 2,000 people now competing against the 18,000 unemployed for those jobs.

[00:19:23]Yeah. Well, consider this. Okay. There's all of these anybody who's touching the AI supply chain right now is desperate to turn money into building more AI stuff. You know, like we talk about data center projects and I point out how like, you know, you could give everybody like half a million dollars who's like in that 2,000 person town and you'd be fine because the companies are so desperate to get these data centers built because they're expecting such huge returns on these data centers.

[00:19:45]Returns for what? Whatever. Building more data centers, going to Mars, like they they have reasons why they're willing to spend. So, you're going to have a money fire hydrant being pointed at like random people who have minor roles at like getting out of the way or just helping get these data centers built, doing literally anything. But you could be competing with so many people who are laid off. So like if you ask for a lot of uh money, I'm going to go find someone cheaper. Someone who asked for nothing peanuts because you're what all you're doing is promoting. Basically, you're almost useless. So I'm just going to find anyone like there's 20,000 begging for work.

[00:20:15]>> Yeah.

[00:20:16]>> So I'm just going to pick one at random for peanuts.

[00:20:18]>> Uh yeah, I mean look, I I agree with you when we get to the end game. Okay. I agree that there's eventually going to be like we can argue about when, right?

[00:20:24]I think the end game comes a little bit later than you do, but like if you fast forward, let's say 10 years. I do think like 10 years from now there will be plenty of data centers built and even your salary for like helping build the next data center will dry up and then yeah I agree and and this is basically called gradual disempowerment right like when most of us have like literally nothing to add because robots can do everything 100%. Then yeah I agree that's like a scary situation and then also the AIs can just like you know use nanobots to wipe us out. I'm still hung up on the salaries going up thing because it's like let's say we have a 100 doctors, we lay off 95 of them, there's five doctors employed and those 95 doctors are trying to get those five doctors jobs. They're going to say I'll take less.

[00:21:01]>> I agree that a lot of people are experiencing what you're going to say.

[00:21:04]Um the the key reason the salary goes up is in areas where there's suddenly a lot of new demand. And like I said, the first area I'd point to is like building the AI itself, right? Like anybody who's touching AI right now is like trying to throw money at the problem and having difficulty doing it. You know, like everybody's like, "Take my money to help me build more of this AI stuff."

[00:21:19]>> Right. I mean, what you're saying is true with like electricians building data centers. They are making more money than they were before data centers were a thing. Their salaries have gone up. If you're an electrician that's working on data centers, but I just think it's such a hyper specialized um you know, tiny part of the to overall workforce.

[00:21:35]>> Yeah. I don't think medical doctor salaries are going to go up for at least not for long, right? We're still in this transition period where you still kind of need the doctor to sign off, but like in five years or whatever, Yeah. I don't think doctors >> doctors, lawyers, accountants, all all of those highly paid professional service people are going to be replaced by machines and be unemployed and have large salaries and mortgages and all the things that go with it. Uh, and I don't, you know, that's a problem that I don't know how we get out of. uh you know like like we we've talked about on the show before, but if people who are are the highest rated mortgage payers can't pay their mortgages in mass, that causes a major problem for the whole financial system. All right. All right. Let's keep going. Let's keep going. Let's let's get to anthropic raising $65 billion on a >> $965 billion post money valuation.

[00:22:27]They're just they're just anybody will if you say you're anthropic, they'll There we go. you know, uh, make it rain with the cash. If you're anthropic, the world is your oyster. You can get as much money from you want from anyone you want. I don't know what is I mean, what does this mean? I don't know. What does this mean? It just means it means the beast is just getting stronger and stronger and stronger.

[00:22:51]>> Yeah. I mean, I know this is kind of what we talked about, but the reason why people are saying an, you know, 985 billion, I think, you know, basically a trillion was the official valuation and they're saying, "Okay, Anthropic is worth a trillion dollars." Generally, if a company is worth a trillion dollars, it's because you expect them to make like let's say 50 billion a year in profit and you give it like a 20x multiple. That's like very roughly how the math works out. They're making like very a very small positive amount of profit right now, but they're still early stage. So, I guess the question is why are people so confident that they can still be make like 50 billion or 100 billion a year of profit plus, right?

[00:23:21]Why why do people feel like that's a defensible position for them? And I think the simple answer is look at how much people get paid to do work and now the AI is going to get paid to do that work. And that is in fact trillions of dollars. So if anything, I'm on the team that says, "Yeah, one trillion is like a low valuation, right? Like maybe it should be 1.5 or or two trillion." So So >> you think they get give them more money.

[00:23:41]Give them more. Keep it going. They just they deserve it.

[00:23:44]>> I think it'll pay off as an investment, right? If Google if Google is worth five trillion, then it makes sense that Anthropic has a chance of, you know, being of of having like a not a monopoly, but a big share of the pie of like replacing human labor. And so they're in like the same ballpark of Google even though they're much earlier along and they don't have the same profit engine that Google does. The idea that they're worth like a fifth of Google or a fourth of Google is calibrated in my opinion.

[00:24:08]>> Yeah. And Michael, my dream is that like the bottom falls out of the money game of this whole thing and the investors realize they're not going to make their money back in the time and way they think and and that's how we slow this thing down. But this appears to be cutting in entirely the opposite direction where the investors are just flooding money into the system. Uh I don't know. What do you think? I think there's a part of the story we're not really talking that much about which is basically the incentives are all wrong right now. So a valuation this high creates enormous pressure to ship faster, deploy wider, you know, capture more market share and safety becomes a thing you do after you hit the next milestone even if >> I mean all the focus about competition how you know how you going to otherwise open AI or Google will lap you. So when the price a trillion dollar company and the risk is existential, I mean the rational move for any single actor is still to push forward and hope for the alarming problem to get solved in time and that's why I keep raising the alarm because it's not like I hate AI. I think it would be incredible but because we're treating the most transformative technology in history just like another app update you know uh when the downside isn't the app crashes the downside is we lose the ability to course correct at all. So this series round is just a news about it's not just about the the comment getting reached. It's another data point of showing how fast the train is moving.

[00:25:24]>> The doom train >> and um you know who who's going to be driving the doom train? I don't know.

[00:25:28]It's going to be humans driving you know if I keep going like that.

[00:25:31]>> Yeah. And and Lyron it seems a little bit like Anthropic has they have separated themselves from the other labs. I feel like do you think this is this continues and Anthropic continues to be the lead horse and just creates distance or is it going to get caught?

[00:25:45]>> I mean in terms of valuation they slightly surpassed OpenAI but they're in a very similar ballpark. I wouldn't read too much into the valuation difference but they've got more momentum because they just like committed earlier to this idea of like we're going to make these tool these agents for these companies and I think I've mentioned this on a previous show. Um, OpenAI was going all in on like the $20 a month per user type of model and Anthropic is like, how about how about $20 every 10 minutes, right? So that's businesses just have a a bigger way to pump your pump the fire hose of money at you. Uh, but OpenAI is very much catching up. I mean, GPT 5.5 is roughly the same as Cloud Opus 4.7 or even 4.8. So I wouldn't write out OpenAI at this point. Open AAI versus anthropic. It's they're very similar and my best guess is they'll stay neck and neck for for a while. I mean, I'm happy to talk about the horse race, but that's not really what we care about, right?

[00:26:31]And then Google is is like the third horse, right? So, these three horses, Google, Anthropic, OpenAI, they may be tied for a while. I think the relevant prediction is just one company might hit that recursive self-improvement threshold, which sadly they're explicitly trying to do, right? That's how they're trying to end the game.

[00:26:46]They're trying to say, "Hey, we're going to do the kill move where our AI starts improving our AI, and then we run it overnight and we come back and it's already like 5 years ahead because it was like improving itself and the improved version improved itself." And from my perspective, it's like, "Oh, great." So that's that's that's the end game in terms of like game over. Like we don't survive an event like that.

[00:27:03]>> Yeah. You know what's something I like?

[00:27:05]And this is in car racing, not horse racing, but you guys are familiar with like the caution flag, right? When everybody's racing like 200 plus miles an hour, and then there's something on the track, they throw out the flag, and everybody gets down to like super slow for several laps. We really need like the equivalent in the AI race of the caution flag, which is like we need to go from 280 to 10 miles an hour for some indeterminate period until we judge that it's safe to open it back up again. If you like the content we're making here on the AI Risk Network, but you're not a subscriber, what are you thinking? We need everyone to be a subscriber. Do it.

[00:27:40]Next two things we're going to talk about are kind of related, but we'll do them one at a time. This I think is is a bigger deal than people are going to make out of it or than it thinks because it seems like it's a natural thing to happen. But Apple wants to put cameras in your AirPods. The new AirPods are going to come out with cameras in them.

[00:27:56]And uh oh boy, think of all the training data Apple's going to get from putting cameras in everybody's ear pods uh AirPods. Um I don't I'm really like personally torn about it. like I I uh I don't want to get behind. I feel like I don't want to needlessly be overly cautious with my privacy. And yet, the idea of having this thing when I'm not even paying attention, just logging everything that's going on all the time is unsettling. Michael, you're wearing them right now, so we'll go to you first.

[00:28:33]There are no cameras in your AirPods, but if there were, how would you feel about it?

[00:28:39]Yeah, I mean on the surface it sounds helpful like giving your assistant a pair of glasses so it stops guessing. Uh but u I mean what this actually is this these cameras aren't just another gadget. They're another high resolution hose pumping real world visual data straight into AI training pipeline. So Apple and every other big tech company racing ahead needs this data to make their model smarter. more eyes, more training, faster progress towards system that don't just assist us. They they understand us better than we understand ourselves. So we are racing towards AI that could one day think circles around every human on earth. And the question won't be oh can Siri see my fridge? It will be like Siri what happens next?

[00:29:22]What is your plan for me next? Will you allow me to stay alive? Will you keep providing me for me to be happy and stuff? Uh, so you know, we sleepwalk into giving an ever smart ever smarter stranger permanent front row seats to our lives.

[00:29:37]>> Yeah, that's what it is. It's a front row seat inside your life. Uh, Lon, are you are you going to get the camera pods?

[00:29:45]>> I mean, I I spent like an hour today with AI watching me and I'm I'm for it.

[00:29:50]You know, I think this is I'd f I'd you know, I draw a distinction between like what is uh you know, the bad part of AI versus the good part. I think AI has a lot of value to add giving us advice, you know, like it it is quite insightful and specifically the one of the main times I use cameras besides like handyman projects like, "Hey, take a look at this in my house. What should I build?" I also I just used it a lot in the gym. And Mark Andre actually uh was telling Joe Rogan the other day that his friend, I think it's Mark Zuckerberg, puts up cameras in his gym when he does jiu-jitsu. And and then the AI just gives him like a list of tips after the session. And I'm like, "Yeah, that's basically what I do when I when I work out." And it's been like a game changer.

[00:30:27]It's been like having a personal trainer like watching me like literally every day. And like I I think privacy is overblown. Like yeah, you don't want to like tell every reveal everybody's secrets, but like what is privacy worth compared to like getting a bunch of utility? I feel like generally utility is worth more than privacy.

[00:30:41]>> Will you will you will you wear them uh will you wear them during sexy time?

[00:30:45]>> Well, I would just not wear it in my bedroom. I mean, how hard is that, right? I feel like that's like a real slippery slope argument, right? Like would you like put it up your butt? No, I wouldn't.

[00:30:56]Never say never.

[00:30:57]>> All right. This is the show we need.

[00:30:59]This is where this show's been heading for this whole time.

[00:31:03]>> Um, man, I don't know. I just I just, you know, every they're all fighting over who can take pictures of us more.

[00:31:10]Like, uh, this is another I read an article about how Walmart bought Vizio, uh, so that they could basically own the TVs that they sell in the store. And there's a setting you cannot take off if you buy a TV from Walmart now that takes a picture of your uh screen every 10 seconds and sends it back to Walmart. So like they are judging, you know, they they everybody wants to see our behavior. Everybody wants to document it to get that data flow. Uh and I just don't like the idea of someone watching me all the time. I don't know. I guess that's I mean that we're inextricably heading that way. I guess it's what it is. The last story relates exactly to this. So, let's just get right to it.

[00:31:50]Open AI says Apple AirPods camera hold my beer. We're actually going to put cameras in New York City in people's houses, in their kitchens, and their living rooms and document all of the human behavior happening. So, OpenAI has been running some sort of program in New York where they are putting cameras all over people's houses, uh, recording everything that happens and then obviously training models based on that data. Um, Lon, I'm I'll go to you again.

[00:32:17]you would you welcome the open AI?

[00:32:20]>> This is this isn't my this isn't my beef, okay? Like this is not this is straying too far from what I worry about, right? I just worry about AI disempowering humanity. The only thing I could say is like, okay, maybe it'll get smarter faster if it has more data. But I think the the impact of that is so small compared to like the larger issue of like, hey, let's pause the training, right? Let's get ready to hit the stop button. But the idea of like, oh my god, it's got more visual data of what people are doing. I'm like, whatever. Sounds useful. Michael, will you put OpenAI's cameras in your living room and your kitchen and your bedroom? I have I have this uh you know, just imagine you were a creature and instead of having two eyes, you had like billions of eyes like being able to connect pictures and dots from millions of homes, you know, just imagine like a headful of eyes, just millions of eyes watching everything.

[00:33:09]So, it's really unsettling. It's kind of an alien thing to even relate to this, right? like as a human experience from a personal perspective. Let's say um picture your your kitchen. Yeah. Not as your private space but as a live training data set of a super digent robot. So every time you you chop your vegetables or load the dishwasher or argue with your partner, the camera captures it, you know, in high resolution. And that footage doesn't just reach a robot doesn't just tell it how to how to do laundry. It teaches the system how to how to be more how to understand humans, how to actually behave um you know uh when how to think about this everything. So, zoom out, scale that to millions of homes and then you get a a picture where again how powerful this alien technology is going to be and how completely out of our league like imagine like a little ping-pong ball and like a planet like z you know a huge planet in comparison of um of how rich experience I will have processing all this data and uh yeah >> doing stuff with it >> and just one of the things like this is this is it's childhood memories Because at the moment it's very you know like a little infant or you can imagine like that >> and um imagine now we're raising a child and um imagine had a child with hidden cameras in every room you know um and the child will learn ethics and you know at some point when it becomes an adult we give him the keys to everything the economy the decision making everything so it's just so weird isn't it it's just so weird >> I don't know think like because a lot of these people are very poor because they give them some peanuts to to put the camera on, right? And they imagine there might be like domestic violence going on. I don't know, whatever. Like just crazy situations, you know, all kinds of situations. And this is actual training data. It's being fed 24/7 inside this alien intelligence >> and uh just growing growing to be more powerful and understand things at a deeper level. So I don't know. Wow.

[00:35:08]>> Wow. That's wild actually. That is okay.

[00:35:10]So it is a totally different thing to like read every book about human relationships. Like let's just take marriages for example. Like to read every book written about marriages. But then to actually 24/7 monitor the conversation of a marriage.

[00:35:24]That is unique proprietary data that I don't believe is accessible anywhere on the internet. Like you could watch every movie about marriage and everything but there's nothing like actually documenting it over a long time horizon.

[00:35:36]>> All the marriages of the earth real time.

[00:35:38]>> All the marriages of the earth man. I mean all right. All right. Now, I'm going to go to Lon real quick cuz you probably learned something real interesting if you if you had a 24/7 camera on all the marriages uh and you fed it into some genius model would become like the best marriage counselor ever on the planet or something. Some benefit later on.

[00:35:57]>> I mean, it's already pretty good at marriage counseling, right? Like if you compare AI versus human at therapy, it's it's already comparable. You can say the human's like a little better, but there's also plenty of human therapists who aren't good at their job. So I think it's fair to say it's not in the worst quartortile compared to humans. So it's like okay give every data of every marriage. Okay great. Yeah. So then it becomes like a little bit better but I think that therapists are already good and I think their job is already straightforward. It's like okay talk out your problems. What does this person think? What does that person think?

[00:36:24]Right. It's like who cares? Just give them more data. It's like I think you guys are getting obsessed here.

[00:36:28]>> I I actually think if you had this documentation device and you gave it to the human therapist that would be of tremendous value, right?

[00:36:37]>> Like it's not just like he comes in and says what he says. She comes in and says what he he says, but like here's objectively what actually happened.

[00:36:43]>> Yeah.

[00:36:44]>> Um >> Whoa.

[00:36:50]I don't know what just happened right there.

[00:36:53]>> All right. That's the signal. That's the signal that it's time to end the show.

[00:36:56]That's what it is. That That must be what it is. Just a fireworks. Uh I don't even >> I don't know what the hell just happened. Riverside.

[00:37:04]>> Some gesture that triggered it.

[00:37:05]>> Some gesture that triggered it.

[00:37:06]>> I was like, "All right. All right.

[00:37:08]Anyway, >> yeah, >> it depends on the software. It's Riverside.

[00:37:13]>> Riverside doing something there. All right, guys. It is a pleasure as always.

[00:37:17]Uh, what are we going to do here as we go out? Uh, you know, uh, but I'm I might maybe a a prayer for the Pope. Uh, I don't know what what do you guys got?

[00:37:25]Uh, >> I'm just going to use my my AirPods to get that camera going.

[00:37:30]>> Perfect. Might take a picture with his ear.

[00:37:32]>> All right. Here, look. I got my own sound effects here.

[00:37:39]Scary.

[00:37:40]Oh my god.

[00:37:41]>> Excellent. Excellent. Excellent. All right, guys. A pleasure as always. Have a great weekend. I'll see you next week.

[00:37:47]>> Bye, everybody.

[00:37:48]>> Bye.