9/10/26: AI Whistleblowers Dire Warning, Mathematician Says OpenAI Stole Solution, Data Center Support Collapses

0:00This is an I Heart Podcast. Guaranteed Human. This message is brought to you by Ralph's. If you're into entertainment podcasts, you know it's all about the details that pull you in and keep you engaged. The right details make everything feel seamless. And with Ralph's, you can plan ahead by keeping your kitchen stocked with fresh groceries and essentials. So the details of your day feel seamless right from the start. Order online and pick up groceries at your local Ralph's. Serving so cal for over 150 years. This is Jacob Goldstein from What's Your Problem. Running a business is hard enough. Don't make it harder with a dozen apps that don't talk to each other. One for sales, another for inventory, a separate one for accounting. That's software overload. Odo is the all-in-one platform that replaces them all. CRM, accounting, inventory, e-commerce, HR, fully integrated, easy to use, and built to grow with your business.

1:00Thousands have already made the switch. Why not you? Try Odo for free at odo.com. That's odo.com. Did you know every 36 minutes a child in the US is diagnosed with cancer? Which means some of the bravest warriors you'll ever meet aren't professional athletes or superheroes. They're the kids in this fight. That's why I'm proud to support Hyundai Hope on Wheels. For the past 28 years with support from Hyundai, their dealers, and every new Hyundai sold, Hope on Wheels has contributed over $300 million to childhood cancer research, helping save more than 40,000 kids. But we're not done yet. Let's bring every warrior home. Join our fight at HyundaiWoyers.com. Hey guys, ready or not? 2024 is here, and we, here at Breaking Points, are already thinking of ways we can up our game for this critical election. We rely on our premium subs to expand coverage, upgrade the studio, add staff, give you guys the best independent coverage that is possible. If you like what we're all about, it just means the absolute world to have your support.

2:00But enough with that, let's get to the show. Good morning everybody. Happy Thursday. We have an amazing show for everybody today. Would you have Crystal? Did we do a lot going on in the AI world? We're going to dig deep this morning in an anthropic researcher resigned and is giving up a gigantic payday over his concerns that AI could destroy us all. He did some interviews. We also are learning about new security incidents. And we'll break all of that down for you also at the same time. Open AI is bragging that they solved this longstanding, very challenging math problem. And no one else had been able to solve before. But there are allegations. And it appears they may have cribbed some of the work from an actual human that was working or actually human beings that were working on this problem. So we will dig into the evidence there. New research says that data centers are not in fact a major boon to local economies. This comes amid increasing political fallout over AI data centers, et cetera. So we'll dig into that. We'll dig into the polling as well. Trump at his big midterm convention in Texas is now promising Americans $5,000 checks.

3:05If they vote for Republicans and keep them in the majority. So that's interesting given some of the promises that have already been made on that front. 9-11 is tomorrow and Glenn grainwild is going to join us for a bit of a retrospective looking at how that kicked off the era of master valence. And also how many of the villains of the 9-11 era and the Iraq war era are still with us today to cheerlead us into the Iran war. And we're also going to do an update on the Iran war. Of course, the who's these actually taking more territory in a stunning offensive. That is going to make it easier for them to control waterways near them. So gas prices could suffer once again. And Iran were learning from those air strikes the other night was able to successfully damage significant amount of US aircraft. Also our intervention in the bottom market did absolutely nothing. So a lot of different things going on there. Exactly. That's right. Thank you to everybody who has been signing up at breakingpoints.com. Ryan did another record yesterday with this sign up. So my new pitch is if you sign up, Ryan gets to continue to do what everyone wants.

4:06I've admitted defeat and that's it breakingpoints.com put the free month trial up there. Let Ryan be Ryan. If you love Ryan Grimm, sign up for breakingpoints.com. If you hate him too, you should also sign up or don't sign up. And what is a premium subscription? Diesel now goes for about $200 a barrel. I'm borrowing from Ryan. So our premium subscription is less than half of that. You could buy one and you could buy one. You could buy one or two premium subscriptions for the price of or for less than two barrels of diesel. I mean, come on. That's pretty incredible. That is the thing. That's my sign up. That's right around the corner guys and our premium memberships are not going to get stuck in the straight for moves. Yes, exactly. It's reliable. It's actually useful to your life. By the way, if you're watching, please hit subscribe to our YouTube channel. If you're listening to podcasts, please share an episode with a friend. So let's start with AI tons of movement in the field.

5:06Yeah, so we had this anthropic researcher who resigned and posted a long thread on Twitter. We'll get to that thread in just a minute, but he made the rounds on some shows, yes, you're including on Fox News and CNN. Let's go ahead and take a listen to a little bit of what he had to say to Brett bear about his concerns. I agree with Evan that right now the threat is minimal of actual takeover. But because of self-improvement, this AI is making themselves smarter could happen as soon as next year. Some people even say six months. So this attitude is shared across the industry where people essentially are aware that the work they are currently doing. If we don't slow down, could lead to AI takeover by the end of the decade. So what do you say to the people who look at this and say the American government could potentially put in regulations, could put in guardrails and could throttle back some companies or set up regulations. But we have no control over what China does, what India does. How do you answer that question? You heard Scott Besson in the piece before this kind of make a similar argument.

6:09I think you have to think of this as an arms race. This technology is not ordinary technology. It's not like other forms of innovation. This is essentially a super weapon. And I think that I expect the international community to be ready to make agreements about how to control the development of super weapon technology. So people would say if you say it's like an arms race, then we dealt with nuclear weapons and kind of a standoff with mutual short destruction and worked with allies and enemies to say we've got the stuff too. Don't think that you're going to get away with anything. So does that work against what you're saying? What would you like to see happen? Yeah, so unfortunately we know how to control nuclear weapons pretty much. We don't yet know how to control AI. This is possibly the most dangerous technology that humanity has ever created. It's basically out of science fiction. I think we have no other choice but to cooperate internationally because an arms race would be disastrous in a way that no other human activity has been in the past.

7:15So that was Jacob Coxen. Let me go and put his thread up on the screen, which went absolutely viral. This has now been viewed more than 146 million times on Twitter, which is interesting because he is not the first AI researcher to sound the alarm to resign to put on a statement. Another one previously said like I'm out of here. I'm quite right poetry. Good luck to humanity effectively. So let me read what he says here and then we can talk some about why this particular warning has gone as viral as it did. I'm going to read the whole threads that just bear with me says I resigned from anthropic today. I spent the last three years doing pre training research at both open AI and inthropic. So we've seen the inside of both of these major companies. Neither company is acting responsibly. They are racing straight to self improving super intelligence and gambling with our lives. Do not underestimate the power of this technology. These will soon be super human systems that can hack anything revolutionize any field overnight and acquire real power and resources. We've all witnessed the progress in each of these domains and progress is not slowing.

8:16The people building AI earnestly believe that it could kill us all by the end of the decade. This is not a marketing stunt. If anything many executives and senior researchers will couch their freezing in the press to sound sensible. But I hear the same people express fear privately. No other human activity poses this level of danger. A common response is if they truly believe this why are they still building it at open AI. Many have not deeply internalized the civilizational stakes and inthropic the stakes are well understood but they are not. They are well understood but they are locked in a race to get their first. They believe no one else will act responsibly so they must do it themselves despite the risk. That by the way is like a classic prisoner is dilemma. Accepting this race and entering the end game is a heubristic gamble that should not be launched from a private company slack. Tempting a speed run alignment should require extraordinary confidence that there are no better trajectories available. I am optimistic about the potential for cordon mages. That makes one of us warning shots like the hugging face attack have made pacing agreements between us labs more viable. I don't feel like we're on track to prevent a global race which may require costly actions such as a temporary ban on improving model capabilities.

9:21If you are a lab researcher I urge you to consider what the next few years will actually feel like. Do you want to kick off a super intelligent RL run without a rigorous understanding of its mind. Should you put your head down because quote it's happening anyway or take this moment to call for different conditions. This is set off a flood of other people from inside the industry. Many of them who are still at their post at Anthropic at OpenAI will get to some of that in a second saying yeah he is right. And yeah I do think that there's about a 10% risk that super intelligent AI is going to literally destroy all of humanity. So quite a stunning moment and clearly sort of elites and the public at large were ready to hear this message of the time. So it was the most viral incident and what in my opinion really broke containment because and the way that you know that is exactly how Jake what Jacob was on Fox News. He was on CNN Ted Cruz will show you reacted to it.

10:21It was on the view obviously it's the A block of our show. If you have heard any warning from the AI industry outside of let's say some 2025 pronouncements from Dario Amade. Probably a year or so now at this point where he set many similar things you and I we cover this day in and day out. None of this was surprising. So the story here is about how much this breach containment and I also think how much that this has also been agreed upon by the very people in the industry itself. Let's put the next one up here on the screen because you could say okay Jacob Cox and the guy barely worked there for a couple weeks. He worked to open AI. Oh he's just a agitator an activist or something like that. Well here is Evan Hubinger. He is the current alignment science lead at Anthropic right so a current employee at the company effectively his boss. He says Jacob is correct here. We really do earnestly believe AI could kill all humans. I personally think it is a greater than 10% chance within the next decade.

11:24I believe Anthropic is trying its best but we do not yet have a plan to solve alignment for super intelligence and are not clearly on track to. Now this is extraordinary. This is a company which is going public in the next six months. Many people in the AI industry were saying oh this is a comms disaster. No this is a reality disaster and in fact it is a reality disaster which has been in plain English from the CEO of Anthropic. From the CEO of open AI. From the employees themselves. Ryan and I covered the Anthropics total addressable market where they said it's 30 trillion dollars. That is the entire US economy where they said wow our plan for our total addressable market is 30 trillion dollars. Effectively replace all aspects of commerce in human society. So you are like okay and then you continue to look not just at the warnings within the company, not just at the past statements, not just at Jacob.

12:24But his bosses of his counterparts at open AI and all of them ultimately come back to the same thing. It all seems like we're on this track to make it so that we will create this technology of which we have no control. I do think though and I would really love we've tried to make contact with Jacob. Jacob if you're listening please we really would love to interview you more than five minutes of Brett Bear with a stick mic. I'm going to touch with him. Yeah. Okay so here's the thing we really need a longer conversation because you know what if you're going to make extraordinary claims, require extraordinary evidence. So you say you're going to kill humanity how? Like what is the exact mechanism? Is it taking down the energy grid? Is it bioengineering targeted virus? Is it nuclear weapons launch which are all controlled by technology? Like when you're going to say something like this you actually owe it to us to tell us like how exactly do you mean it would kill all of us. But the second thing and this is why we focus so much on the cybersecurity part.

13:24Is it is so obvious now through multiple incidents that the infrastructure of which our modern lives depend on is not prepared for this whatsoever? And that the companies who allegedly have all these safety and alignment and all these things these models continue to break containment. And that breaking containment is actually a feature not a bug because it means that it is developing this recursive intelligence and its ability to improve itself to reason to work outside the bounds of what humans tell it to do so that it can break containment and it can reach further. Ultimately that's what we have decided the value here in the United States is all about why because of brute capitalist enterprise. The last thing I just want to say and this is a much longer conversation around China is everybody says we need to beat China. So I actually spent a lot of time reading primary source documents and translations. What do the Chinese believe? They are not quote AGI PILD and I'm quoting directly from one of their top analysts who says the Chinese Communist Party is not going to be able to do this.

14:26The Chinese Communist Party is looking at this with deep suspicion with some interest. They have a absolute bedrock philosophy as if this thing ever got to a point where it could supplant the Communist Party. We're shutting this shit down period. They're way that they are currently developing their AI industry is deep seek is their only company which is deeply interested in AGI. And even then the deep seek CEO is very different than our top CEOs. He says we do not believe in profit. We do not believe that this is a capitalist enterprise. They look at it a little bit more as a research project. I guess is probably the best way to describe it. They also are willing to sacrifice profit by making sure all of their models are open source and a lot of it is resource competition with the United States over our continued export controls on their GPUs. But the second thing is that they don't have a counterpart, at least from what I could tell, of all of these other people who are racing towards AGI. The way that the Chinese are developing their AI industry is they have given a Dictat basically to the entire market where they're like your job is to create AI that has capabilities to enhance directives which are important to the state, aka manufacturing, aka robotics.

15:44And even then I don't know if you saw this but the humanoid robots which a lot of people have been worried about they just smashed that industry over the head. They're like this seems to be going too far. This is creating some problems just yesterday they issued a regulatory authority basically wiping out or at least dramatically restricting a lot of the humanoid robot development that's been happening inside of China. So they have a much much more different they have a very different system. They're not going towards this artificial and recursive super intelligence the ability for the models to train themselves. They don't seem to care about that as much. They're trying to use AI to enhance the actual capabilities of the Chinese economy in manufacturing and in a productivity sense. So they they're very much in the school of develop the tools not the supply plantations. We are in the opposite right where we have not seen as much of the enterprise enterprise application for AI. And instead we're racing forward with all these data centers and we're going sprinting towards AGI with some belief that it'll magically create some great new economic opportunity.

16:47Yeah, so their approaches to actually focus on the applications. Yeah, there's actually a big article about how automation is replacing all of these blue car workers in China. Yeah, they have an economic problem on their hands because of that. But that's been much more of the focus. The other thing that I've talked to to Garrison about and then he's made the case for here too is what the Chinese are able to do very effectively is fast follow. So wherever we push the bleeding edge of the frontier, they're able to follow behind very quickly. Well, what does that mean? That means that the further we push the frontier, the more capability the Chinese are going to have to. So the whole like, well, the Chinese will do it thing really misunderstand the nature and the dynamic of what is happening here. In any case, you know, getting back to Jacob, I think the reason why this broke through at this moment is because there have been enough concern and coverage over the hugging face. And even the anti-hack and crime spree and civilizational build up of these AIs that people were kind of paying attention so that when he said, OK, guys, I really have concerns so much so that and again, I want to reiterate this. He's leaving and he is giving up all of the money that he could have made in the Anthropic IPO like he was due for a massive payday, one of the largest in the history of capitalism.

18:00And he walked away from all of that in order to sound this alarm. So again, once again, the objections I'm hearing, which are sort of unfalseifiable and have just really are increasingly devolving to the land of QAnon conspiracy theory is that this is all some elaborate marketing hype play. OK, there is first of all no evidence for that second of all, we've had many whistleblowers leave the companies and give up their paydays to come out and say, I'm genuinely deeply concerned. Have we had any whistleblowers come out and say, I'm coming forward because they're lying to you about the capabilities of this. They're lying to you about the labor replacement. They're lying to you about the existential risk. No, we see no evidence for that whatsoever. And as we discussed earlier this week, that whole idea, this is a hype cycle and it's just marketing. It should have been put to bed by the fact that we now know open AI tried to cover up some of the safety incidents and AI agents escaping the sandbox and make sure that that did not become public.

19:02So they're not going to be able to do that with their hand was forced and they've continued to constrain access to researchers to be able to tell exactly what the hell has been going on at their company. So if it was all a hype cycle, wouldn't they want to get all of this out there? Oh my god, our AIs are so dangerous. No, that is not the way they've been behaving. So I need people who had that view, which again, I understand where it came from because you should not trust AI executives. You should not trust. So I just want to say I'm not sure what the hell was going on at this time, but I really would implore you to reconsider your priors if that is where you have been because we now have evidence that contradicts that. This message is brought to you by Ralph's. If you're listening for a laugh or a fresh take on everyday life, you know it's often the little things that make everything more enjoyable and easier. And with Ralph's, you can plan ahead by stocking your kitchen with fresh groceries and everyday essentials. So the little things in your day feel easier right from the start. Order online and pick up groceries at your local Ralph's. Ralph's.

20:02Serving so cal for over 150 years. This is Jacob Goldstein from What's Your Problem. Running a business is hard enough. Don't make it harder with a dozen apps that don't talk to each other. One for sales, another for inventory, a separate one for accounting. That's software overload. Odo is the all-in-one platform that replaces them all. CRM, accounting, inventory, e-commerce, HR, fully integrated, easy to use, and built to grow with your business. Thousands have already made the switch. Why not you? Try Odo for free at odo.com. That's odo.com. Did you know every 36 minutes a child in the U.S. is diagnosed with cancer? Which means some of the bravest warriors you'll ever meet aren't professional athletes or superheroes, they're the kids in this fight. That's why I'm proud to support Hyundai Hope on Wheels. For the past 28 years with support from Hyundai, their dealers, and every new Hyundai sold, Hope on Wheels has contributed over $300 million to childhood cancer research, helping

21:06save more than 40,000 kids. But we're not done yet. Let's bring every warrior home. Join our fight at HyundaiWayers.com. You had not just Jacob's boss effectively come out and say, yeah, he's right. We do actually think there is a 10% chance of that these will lead to human extinction. You have a huge number of people from within the industry. Go ahead and put a four up on the screen. This Twitter user, Taylor Poppillars did the Lord's work here and compiled a bunch of quotes from people leading scientists in the industry coming out and being basically saying, yeah, we agree. So he starts with Jacob Coxen. Then it goes to Samuel Marx, scalable oversight lead ad inthropic, quote, AI developers believe their tech could cause human extinction or similarly bad outcomes. This could happen in the next few years. Next one, Evan Huberger. We read that one earlier. Alex Turner, former research scientist at Google Deep Mind says, yes, I left Google Deep Mind in June. Jacob is right. Many researchers believe they're building something that could kill everyone on the planet.

22:08It was literally my day job to think about how to stop that. Another chief scientist at OpenAI, Jacob Pachoki. This is a time that calls for extreme caution. I'm concerned no one is prepared for the consequences of a continued rapid rise in machine intelligence. Jason Wolf, a researcher at OpenAI. I don't know what my probabilities are on literal extinction, but I think there are a number of ways AI could go poorly for humanity. At the current, frankly, terrifying pace humanity will be quite lucky if we manage to find and stay on the narrow path between all the bad outcomes. Jonathan Richard Swartz, former senior research scientist at Google Deep Mind. I left Deep Mind after seven years and since rejected offers to join the other two labs due to severe concerns about the concentration of power these labs represent what is currently happening in this field is deeply unhealthy for society. Go to the next page. You can see just an image of some of the tweets that have put out and every time I refresh my timeline on Twitter, I see some new researcher, OpenAI, Anthropic, Deep Mind coming out and saying, yes, the risks are real. We are genuinely concerned about this. This isn't just some marketing hype cycle.

23:11So then the natural question is, what the hell are you people doing? Like why are we doing this? And I think that Jacob is taking the right course of action here by walking away and saying I'm not going to be involved. However, it is not hard to imagine the very human and very capitalist driven psychology that leads to them staying. We've seen a version of it in the Trump administration. Well, if it's not me there and I'm genuinely concerned and I'm a responsible actor, I can help shape things from the inside than someone else who has less groupples is going to come in and they're just going to do the same thing. Or the other thing that you hear from politicians all the time is, ah, but if not us, then China. So they are locked in this very obvious prisoner's dilemma. Again, I think the right thing to do would be to come forward the way that Jacob has, but it is not hard to imagine the human psychology that has led to this completely bizarre and insane situation where the majority of the people who are building the product think the product could have an absolutely catastrophic outcome. Right.

24:11And the one thing that you didn't add in there, which is in the White House case of like politicians or others, that's just power. Here we're talking about generational wealth. We're talking about the ability where you can stay and invest your shares and literally not have to work again and day in your life. I'm pretty sure Jacob is like 27 years old. I mean, I don't know the exact number. I'd be willing to bet it's tens of millions of dollars. And he just went way through. I don't mean him that he probably walked away from. So I mean, ask yourself that question too from a family perspective. I'm like, is it that easy to walk away from something like that? Or what if you never had to work again? What would you be willing to sacrifice? I want to go be a mass exodus after the share is fast. Yeah, probably. I mean, there almost always is in normal times. So then I'll imagine with something like this, let's go to the next part. Please, please, just to show you this actually came on the heels of Jacob's announcement. In my opinion, that's got nearly an amount of coverage. It deserves, which is anthropic literally on the same day had to put out a statement,

25:13quote, an alignment assessment of recent cybersecurity incidents. And you will be shocked to learn that what they discovered after auditing some of their systems is that they had multiple incidents of a scan of their transcripts where Claude could have obtained internet access through a cyber evaluation, given the volume, this missed a set of transcripts, et cetera. And we scanned these identified a fourth incident from January of 2026 involving an early version of Claude. And we have notified all effective parties. They basically went through and audited all of their systems. And they discovered that even though they had told these AIs, which were training in an allegedly non-internet connected system, that it was able to break out and hack into different parts, or the very least hack their own internal system and gain access to the internet to be able to do whatever it wanted. I mean, these again are the frontier models where you have, I mean, these people are paid

26:14exorbitant amounts of money. Like, I can't even describe to you what the money race right now in Silicon Valley is. Like to put it in perspective, the second tier players are making 10 to 15 to $20 million a year. Imagine what these guys are making who are on the actual cutting edge, 150 IQ machine learning engineers. I mean, these people are printing to the, I can't even, again, like, and I'm talking just cash, not even a lot alone in the shares of what they are being able to take away. That is the extent of what we're dealing with. And the talent pool who are around this entire thing, who are trying their best to align the technology or any of that. But I think what they're trying to tell us is that it is unalignable in the current fashion without government regulation. I just come back to, there is literally no read. I mean, again, my China reading, every model has to be registered and evaluated by the Chinese government. The Chinese government itself, I encourage everybody to go read Dan Wang's book to the

27:15extent that the CCP and the, is the China is a society of elites, its elites are engineers. There are many pluses and minuses to that. But one of the pluses within the Chinese system is they have genuine expertise at the government regulatory level and their ability to inform the Chinese political leaders so that the two of them come up with a strategy and their regulators are just as knowledgeable as many of the researchers, at least again, from what I understand. Many of their Pentagon bureaucracy, again, others, they don't have to be able to do that. Democracy, they have permanent governments so these people can stay there for 15, 20 years. Their deep state, as long as it's intelligent, like it's a very well-run system. So for them, they're able to guide the technology in the way that they want, look with interests, read the research papers, register the models. We have none of that zero whenever it comes to our models. That's the last part here is that Jacob called for international cooperation. I mean, in a sane world, Jacob would be called along with all the CEOs of these companies

28:20to testify for hours before Congress. They would testify. There would be a working group immediately to develop bipartisan legislation. This would be with guidance from the White House, from industry leaders. And we would all, you know, there would be some democratic input. None of that is happening. Trump is out here being like, I'm not that worried about it. What did he say? Is it at all work out in the end? Something like that. I mean, it's great. Yeah, we have the thought. Take a listen. A8. Some people say the worst case scenario with the AI is that the robots, the machinery learns to, obviously, it thinks for itself, that's what it does. And that could turn against humanity. I just, do we have the guardrails? It's going to be fine. We'll always have something to stop on, right? We'll have a little gear. I really hope so. I don't like that. I really don't like that robot. We'll stop. Yeah. I see the extent of the thinking of Trump. And look, geriatric leader who doesn't stand technology, maybe we could say, oh, that's just a one-off surprise. That's the whole system. Literally, what Chuck Grassley's going to regulate AI?

29:20I think I was like, not the year's old. I mean, in fairness, Bernie has like been the most intelligent on this, and he's also very aged. Okay. But, yeah, that he is an outlier. It's extreme outliers to the number one. I mean, Trump sold his presidency to the AI tech oligarchs. That is entirely clear at this point. Elon given him $250 million for his campaign and remember at the inauguration who was standing behind him. It wasn't that cabinet members or elected politicians. It was the richest tech overlords on the planet. That is who is running the show. He's going to do what they want him to do. And that's what his presidency is. So that's part of why it's so important to just slow things down because we're not getting anything out of the Trump administration. They've set the course. They're all on board for the like no holds barred, no regulation whatsoever. And so, I mean, my final plea to people would just be, let's say that you're, you feel like maybe there's a 1% chance that some fraction of this destruction occurs.

30:25Okay. Well, what's the cost from just slowing down and allowing a democratic process to play out and putting some safeguards in place? What's the harm in doing that? You don't lose anything from doing that. Whereas on the other side, you know, the risk that's being sketched down here is so catastrophic. And to Saggar's point, I've seen a lot of people say, okay, well, how exactly would it happen? You know, it's very hard to know, right? You can't exactly predict the future. But there has been a lot of thinking about scenarios that could theoretically unfold. And so if you want to actually dig into one of the use cases, we had on the one of the co-authors of the book, if anyone builds it, everyone dies and they lay out. Okay, they lay out one potential scenario for how this could all unfold. Just to help you wrap your mind about it because I understand it sounds insane. Like I feel insane talking about it genuinely sounds like sci-fi. But just ask yourself the question, you know, when has it ever been the case that you've had a superior intelligence that really cares about the lower intelligence creature and like, you know, allows them to continue to do their thing and doesn't impose their will

31:28on it. So what they're trying to create, what they're trying to create is something that surpasses human beings in terms of intelligence in all meaningful ways. That is a completely different situation that we have never experienced on this planet before. What is the likely fallout of that? I mean, at the very least, we should take some time to think that through. We should have, you know, requirements, anthropic puts out this analysis of this latest safety incident. They do it out of, I guess, the goodness of their hearts. We're depending on these people to come clean. And we already have instances where they have it. So what else has happened? We don't even know about how could it be that it is easier to, it is easier to build a potentially job and life and, you know, quality of life threatening AI than it is to open a food truck, which is the example that Greg Cazar gave. So, you know, that's, that's where we are. And, you know, there have been, there is starting to be what I'm encouraged by is the level of attention and concern because previously these warnings have just sort of flown under

32:30the radar. They haven't reached a mass population. So I'm glad to see that this is being taken seriously now, but both on the domestic front and on the international cooperation front, you know, it's very discouraging where we are at the moment. Let me read, because you talk a lot about the doom case. Let me read about the positive case. This is a friend of mine, Tanner Greer, it goes by scholar's stage, really insightful guy here in Washington. And he wrote this. Most politicos do not realize that the quote, good outcome, the outcome that does not involve instinction instead involves the following, a transhumanist future for our species see Altman's the merging blog post where we meld with machines. A future where all meaningful agency is given to machines on whose grace we will there after depend. The general eclipse of humanity by a successor species, but they will realize these things and will realize them soon. But the industry does not realize is that to most Americans, all of those futures are horrific and disgusting. Few outside of the valley want gilded slavery. Fewer still yearned to merge with a machine.

33:32And I think that the point that he makes theirs, we've just spent an entire 30 minutes talking about the destruction and the cybersecurity and bio labs. It's like, yeah, but this is the bull case. The bull case for it is that we do become these transhumanists who are able to turn our entire lives over to these agents. Their theory is that this will unleash human flourishing, whereas all evidence exists to the contrary about how much it will actually supplant what it means to be human itself. So I wanted to put that in there too. People say, oh, you don't cover the bull case of it. That's the bull case right there. I don't want that. I certainly don't. And you shouldn't either. All right. Let's get to the next part about math. This is Jacob Goldstein from What's Your Problem? Doing a business is hard enough. Don't make it harder with a dozen apps that don't talk to each other. One for sales, another for inventory, a separate one for accounting. That's software overload. O2 is the all-in-one platform that replaces them all. CRM, accounting, inventory, e-commerce, HR, fully integrated, easy to use, and built

34:38to grow with your business. Thousands have already made the switch. Why not you? Why Odo for free at oduo.com? That's oduo.com. Did you know every 36 minutes a child in the US is diagnosed with cancer, which means some of the bravest warriors you'll ever meet aren't professional athletes or superheroes. They're the kids in this fight. That's why I'm proud to support Hyundai Hope on Wheels. For the past 28 years, with support from Hyundai, their dealers, and every new Hyundai sold, Hope on Wheels has contributed over $300 million to childhood cancer research, helping save more than 40,000 kids. But we're not done yet. Let's bring every warrior home. Join our fight at HyundaiWoyers.com. I'm Glenn Washington. This is Snap Judgment from KQED. When everything is falling apart, what do you do? Hurricane Helene shut everything down. No roads, no cars, no way through. But his daughter was getting married. And somehow some way.

35:38He was going to walk her down that aisle. All systems down. A new Snap Judgment mini-series from KQED, wherever you get your podcasts. All right. So let's talk a little bit about the ethical conduct of these AI companies today, focusing in on OpenAI in particular. So put B1 up on the screen. They put out this big announcement that they solved a new frontier math problem. They say we're sharing a solution to the Navier Stokes Millennium Prize problem. One of the deepest problems at the frontier of mathematics, the proof was produced by a group of agents using an OpenAI next generation model significantly more capable than GPT-6 Aster. That's the one they just put out. The problem concerns whether the description of smooth three-dimensional fluid motion modeled by the Navier Stokes equations can break down. It is remained unresolved for roughly 90 years. So we have some other instances of AI solving some of these very challenging longstanding

36:41math problems. So first glance, oh, wow, another breakthrough. Not so fast. This one seems to have some caveats involved. Good. Put B2 up on the screen. So this is from TechCrunch. I'm going to read you a little bit of this article. So there were two other mathematicians among many others, but there were two other mathematicians who were working on this particular Navier Stokes problem. And we're making significant progress towards a proof. So the allegation here from TechCrunch is opening a dirty on career-making math problem says NYU mathematician. This math professor Tristan Buckmaster announced three proofs on Tuesday with a preliminary finding on one of the major unsolved problems in theoretical mathematics. The findings made in collaboration with anthropic mathematician Levin Alpohs and using both codecs, which is OpenAI. Keep that in mind. And Claude AI models are significant in themselves, but they're also accompanied by an unusual controversy surrounding OpenAI's attempts to solve the same problem.

37:42There is another part of the story, Buckmaster wrote in a statement announcing the proofs. One that honestly, I very much wish I did not have to be concerned with. Corn in the statement, a parallel effort by OpenAI built on their work before it became public led to a tangle of academic rivalries and conflicting claims. So here is the TLDR. These two mathematicians were using Claude AI and codecs, which is again OpenAI. They were using this to work on a direction of the problem that other mathematicians had sort of pushed on and they were pushing forward using this AI. Very close to completion on this. And suddenly, they find out that OpenAI is announcing that they have solved this problem and they used the same direction to solve this problem that these two mathematicians were using. And according to the math world, this was a very unusual way to go about solving this particular problem. So lo and behold, they get in this exchange with OpenAI where they're saying basically

38:46like, what's going on here? And asking them, hey, did you use our sessions on your product? Did you take those and then throw what apparently now we know was like $20 million of compute at pushing through the final stages of this problem to be able to claim yourself that you did it. Of course, they deny that that is the case. Although they say they cannot go out, that the model didn't use their work for its own training. So then, and this is by the way, just a thing that you have to sign up for when you use these products, that what you do on their platforms will be used in their training. So they said, well, we can't say that they didn't use the training of your work on our platform in order to come up with this answer. So this is where things effectively stand. There's some wild things in here too, sovereign, like for example, when they were kind of going back and forth and there was the mathematician Tristan, he was like, well, look, I'm going

39:47to go forward publicly with all of this. And this open AI guy response to him, well, I don't know why you'd want to ruin your career like that. I mean, just sort of like mob boss threatening tactics. There was also a lot of upset over the fact that the other mathematician involved works for Anthropic. Now, they say he wasn't doing this in any sort of official capacity for Anthropic. They were doing this on the side. There was no contract within Thropic or anything like that. But that was a problem for open AI as well. So it's pretty wild situation being alleged here that they either saw that this work was going on that it was almost close to completion and then they threw tens of millions of dollars of compute at it to get to the problem. Or they didn't realize that their platforms were stealing the work of these mathematicians. And so when they asked, okay, let's work on the Navier Stokes problems, like, oh, these guys are doing it. I'll just use their work and we can finish it this way. That's put B3 on the screen. Tristan Buckmaster actually released some of this in a written statement. So here's what he says.

40:48He says, I asked whether the model had been trained on or had access to our sessions in codex into which we've been putting all of our drafts for the whole of this project. I was told the model did not look up user data. I asked again about training. I did not get an answer. The first two proposals were offered to me. The first was that we post our result and that open AI post its result the next day. The second was that after posting, I alone write a paper representing the result, acknowledging that an internal open AI model had resolved it. They twice asserted that he wanted them to remove from authorship one of his co-authors and said it would be simple. If only it were not the case and then it was so annoying that the other person he was working with works at Anthropic, it was also said if open AI post it after us, they would say that we deserve the prize and were the closest humans to the problem. I declined both of these offers. I said that if open AI was at least its results in the way proposed, I would go public with what happened. The reply was, why would you ruin your career? I replied, I am an academic and asked why he thought going public would ruin my career.

41:53The reply was, if you don't want me to be nice, then I don't have to be nice. I mean, yeah, this is genuine, genuine like thuggery. And look, it's a math thing, no offense to the math people. I don't think my life is going to change very much because this problem has been solved. But as a history guy, this is all of the hallmarks of, this is how I know we are truly back in the Guilden age because the hallmarks of the Vanderbilt, I encourage everybody to go read it. The first tycoon, one of my favorite books I've ever read, it is a biography of Cornelius Vanderbilt. It literally is just a tale of these Guilden age cutthroat barbarians who just like steal each other's stuff, they're constantly price cutting and warring and stealing each other's employees and then coming in and threatening the other person and suing them in court. Some of our most important court decisions actually stem from their thievery of each other

42:53and from breaking the law. I mean, that was how the steamship and eventually railroad industry itself was born. So when I've given that famous example of railroaded, these people weren't that smart, they were ultimately, in many cases they made Titanic mistakes but they were just cutthroat ruthless men who built it on the backs of a lot of their workers. When I saw this, I said, wow, like I know exactly where this comes from. Like this, like why would you want to be an awful bad thing, you know, for that to happen? Here's a thing, if you're a mathematician or you're a scientist, even if you're a writer and you're writing drafts through these things, like they're stealing your shit. I mean, that's what it appears to be tonight. Yeah, allegedly. But I mean, but they don't even really deny it because we know that they are using whatever you input, they're using that for training. That's how these things have been built is off of the sum total of human knowledge. So that you should think about moving forward. And I don't know how to get around it because they really are powerful tools in terms of,

43:57you know, if you're doing something that is brute force or cars a lot of data or trying a whole bunch of different solutions and just sort of like going through them and putting them in one by one and seeing which one works out, there's really no other way to go about it. But they, I doubt this will be the last case and it may not even be the first case where they either intentionally or accidentally just take your thing and are like, oh, we got it. We did it over here at, you know, with our AI and no real human involvement. So congratulations to us. It's really wild. I was saying it's sort of like if you wrote a novel in Microsoft Word, and then you're like on the last chapter in Microsoft analysis, like, hey guys, we wrote this incredible novel. Isn't it great? I was like, whoa, whoa, whoa, whoa, how did that happen? Yeah, so that's kind of what was being into that here. A alleged here is kind of the idea. So it really is, it's a very wild situation and it's hard to imagine what we'll prevent. Future situations like this also in the future.

44:59I can't, yeah, I mean, who knows? I also was thinking about it in terms of, because my parents are academics, so I totally get it where, you know, these guys, it's not like you make a lot of money being a mathematician. Yeah. In fact, the only way that you get to keep your job is that you go out and you get billionaires and companies like OpenAI and Anthropic to give you money to fund your research. And I'm assuming that's kind of what he really was saying there was, hey, you know, the math with a math billionaire world isn't that big? Maybe I'm just going to dry up all your funding. You know, I mean, that's scary stuff. If you want an analogy, it's very similar to what happened with gain of function, where any of the academics that were warning Fauci about gain of function or who were whistle-blowing about the dangers, how because Fauci was the head of the NIAID, which controls all of the scientific funding that he could have just cut his own opponents off. When you have that ability, this is bigot, which funds all of these projects and different things that are happening in academia, you have massive control over people's careers.

46:01So just again, to give you examples. So yeah, I see very much hints of the gilded age in all of this. And that's how you know these people really are ready to play. Not a good thing for humanity. Let's get to data centers. This is Jacob Goldstein from What's Your Problem. Running a business is hard enough. Don't make it harder with a dozen apps that don't talk to each other. One for sales, another for inventory, a separate one for accounting. That's software overload. Odo is the all-in-one platform that replaces them all. CRM, accounting, inventory, e-commerce, HR, fully integrated, easy to use, and built to grow with your business. Thousands have already made the switch. Why not you? Try Odo for free at odo.com. Did you know every 36 minutes a child in the US is diagnosed with cancer, which means some of the bravest warriors you'll ever meet aren't professional athletes

47:01or superheroes. They're the kids in this fight. That's why I'm proud to support Hyundai Hope on Wheels. For the past 28 years, with support from Hyundai, their dealers, and every new Hyundai sold, Hope on Wheels has contributed over $300 million to childhood cancer research. Helping save more than 40,000 kids. But we're not done yet. Let's bring every warrior home. Join our fight at HyundaiWoyers.com. England, Washington. This is Snap Judgment from KQED. When everything is falling apart, what do you do? Hurricane Helene shut everything down. No roads, no cars, no way through, but his daughter was getting married. And somehow some way, he was going to walk her down that aisle. All systems down. A new Snap Judgment mini-series from KQED, wherever you get your podcasts. Turning now to data centers is a lot of talk about data centers. There's a lot of propaganda you might have heard. It's, oh, really helps all the local communities,

48:02their property taxes go down, they're flooded with money. These are all the things that were told. All of it was from industry sources. Well, luckily for our purposes, there's actually now been a systematic review of all the evidence. Let's put it up here on the screen. I read this entire paper. I encourage you to, if you want to, it's in the social sciences journal. It's literally public. Maybe we'll have been put a link in the description. You can read it for yourself. So it's called Capital Without Labor, Data Centers, and the Local Economy. So let me read you some of the findings. What they ultimately found through, again, this review. And I'll even explain the methodology, because it is actually pretty interesting. So the takeaway is basically that they were able to look at counties with data centers, and then compare them to counties without. But you can't just compare apples to oranges. You want to try and compare apples to apples. So they actually looked at counties with data centers, and then compared them to counties which had been heavily considered for data centers, but ultimately the project did not go forward.

49:04So we know that it was a desirable place to put a data center. They looked at this over, I think, a 20-year period. Now, obviously, their previous data centers, like the YouTube type data centers, smaller ones, but they ended their data set in 2025. The data center boom began in 2023, so a lot of the data is actually pretty good. I will give you some of the caveats later, but I'm just setting it up. So Capital Without Labor, here are the key findings as I summarize them. Number one, there was no increase in the overall financial health of these communities. Okay? Number two, there was no increase in local employment or new business formation. I actually think that's the single most important one because network effects of companies coming in to your county is the single reason why you want it. And everybody knows this. So, for example, whenever factories would close down in Michigan, one of the things that would devastate the town was not that the factory was gone and the jobs. It was the local supplier to the factory also went away. It was the laundromat,

50:04which does all of the laundry for the guys who work there closes down. It was the seat leather manufacturer 20 miles away. He had no longer, like that's what devastates a community. This, in this case, and that's why the paper is called Capital Without Labor, is they find that while it may take 1,000 plus people to build a data center, the permanent number of employees there is only 20 to 30 people. And so the network effect of having it there doesn't do very much. Okay, so that's number two. Number three, and this is the most interesting finding to me, is that local government borrowing costs actually went up. And that is because housing price growth in the towns, where they build one, it goes down, because people don't want to live around them. And that when people don't want to live around them, and housing prices still go down, then schools actually end up having to squeeze more from the property tax base. Because if your house value is not going to continue to rise in the way that it normally has, that means you need to increase the overall proportion

51:06of property tax on the not decreasing asset, but the less appreciating asset of housing. So ultimately, put it all together. No increasing the overfinancial health of the county, no increasing local employment, local government borrowing costs increase, housing price growth slows, schools end up squeezing more from property tax. There's a water element to this, which I want to note. It's a little complicated, but basically in the counties where there is water scarcity. The borrowing and the bond yields on water bond specifically went up, meaning that they saw more risk to the overall water infrastructure. I'm being very careful in my language. I'm not saying they're guzzling up all the water or any of that. Basically, they're putting strain on the infrastructure, requiring more money to be built in. Now, here are the caveats, which I would be foolish not to mention. This is a cutoff in 2025. We don't have long term data. We don't have the promises of the industry will take five to 10 years to materialize.

52:06But I am just saying that, and by the way, not a single data center proponent came and said, this is BS, this is bad research or any of that. They gave the caveats that I just gave you about the time frame and about how it doesn't capture all that stuff. And there are some minute things that you can quibble with, but nobody is disputing that this is the finding as of right now. Interesting. So I mean, I feel like this is one of the most important papers for the discourse. And instead, everybody just wants to cite New York Times articles about how one place, Loudon County, is doing so well. And it's like, well, here's the evidence. From thousands of data centers across the nation, I have it right here in front of you. Loudon, already a wealthy county. By the way, these are people have filled the rich defense contractors. Yeah, we're just, it's crazy. It's worth noting too, that they may not be representative of the average data center experience. It's almost like you're saying that if you support data centers in your area, you want to be backwards and poor. It's almost like you're saying that. No, I mean, it's pretty devastating when you see this,

53:07because this is the one thing that the data center proponents have really had. And I think a lot of people are just sort of looking at her like, well, what would, what would be the problem? Why would you want this economic job creation in your town and all of this additional tax revenue? And they say they're going to deal with the electricity issue. Don't worry about it. The water issues are overblown. This is just economic development. And it's kind of amazing how sometimes the just natural gut instinct of the American people is more aligned with reality, not always, but sometimes than the, you know, the, what the intellig- the intelligentsia, the beltway intelligentsia want to sell to you. I wanted to also pull up, there's some polling here about Bernie and Greg Kizarra put for this bill to pause AI development, which was framed as this very radical idea in the beltway and sort of like legacy media and among intellectuals here in DC. Well, as I predicted, and I'm sure you would have expected as well, Sagar, this is a very common sense proposal

54:08to the vast majority of Americans who are like, I don't know where this is leading, but I know that I don't trust these tech oligarchs. So maybe we should just take a pause and figure this thing out. And what is going on? What is the impact going to be? And who is going to have control? And is there going to be some public ownership? How is this all going to work? So that specific bill to pause AI development and permanently ban the creation of AI superintelligence is supported by 68% of voters. 72% of Democrats, 70% of Independents, and even 63% of Republicans. You only have 25% of voters who oppose this and the remainder are undecided. So 68% say, yeah, for the pause, I don't know why we would want to create superintelligence. I'm good with that. And only 25% who are opposed, and only 9% by the way, who strongly oppose. So that's where the politics of this thing reside right now.

55:09And it is one more issue where there is just such a massive disconnect between DC elected officials and the broad sensibilities of the public. It is just hanging out there as this incredibly polarize, incredibly powerful issue for someone to be able to grab onto and really speak to the concerns of American people. And I think it ties in with the data center concerns, although people have a variety of issues and concerns. Yeah, and I also wanted to just show the wide range of how rapidly things are now moving in America. So let's start with Rokana, who immediately put out a long statement about multiple proposals in reaction, both not just the data center issue, but about AI. Let's take a listen to him. We need action. Here are five things we must do. First, I've been calling for an AI agency, a federal AI safety agency, just like we have an agency to regulate nuclear power

56:11and to regulate our airlines. Second, there has to be a safety check to make sure that any agentech AI system is contained before it goes on the open network. So you could see that's a flavor of Congressman Rokana to give you. Always worth remembering. This man represents Silicon Valley. Right. So if he can take a stand on this, anyone else has an excuse in Washington, whatsoever, because the tech oligarchs and many of the workers with these companies do are literally his constituents. And then you just to give you a taste too. Here's one of the most pro free market conservatives. Ted Cruz reacting actually to this. C7, let's take a listen. I can tell you there's legislation that I'm working on with Amy Klobuchar and John Thune to try to address catastrophic risk. It is dangerous. It is frightening. How, explain how that would happen. For AI that would implicate biological issues

57:11or nuclear issues that have the potential to have catastrophic risk, it would require government supervision and approval before they can be implemented. Will it pass? It's not clear. So Amy and John are negotiating right now. I mean, look, this is, it is possible to move bipartisan legislation. And actually the Commerce Committee, which I chair, has moved more bipartisan legislation than any committee in the Senate. So it is possible. But it ain't easy. This is a very divided time. And is it, is the, is the horse out of the barn already? Look, that is a good question. And one of the real challenges is China is going, going full speed ahead. Yep. And whatever we do here, China is not going to stop. So I've said somewhat tongue in cheek, but not entirely. If they're going to be killer robots, I'd rather they be American killer robots than Chinese killer robots. All right. So you could see still the Chinese point on ironically. Killer robots. In saying, I don't really care whether I get murdered by a Chinese over American killer robot. Personally, it all seems bad to me, soccer.

58:12Right. And I mean, that's my last point before we move, before to the midterms, is there is not actual evidence that the Chinese want any sort of quote, autonomous robot. In fact, there is only one paper that has ever been published by the People's Liberation Army talking about AI warfare. I read it yesterday. And in it, it's honestly very nuanced. They're like, yeah, this will present us opportunities, autonomous weapons and autonomous drone systems, have tremendous capabilities on the battlefield as we've seen in Ukraine. However, significant questions remain about control and about our ability to make sure that we have the, you know, to, like, to have any sort of say. And for these weapon systems to then turn on on, this is the official, again, the official People's Liberation Army. It's like two kernels, I think, or writing about this in Chinese. Any of you can go and read it right now this English translation, which has been made available by the RAN Corporation. And reading through it, I was like,

59:13I don't see a rainy endorsement of autonomous robots. I mean, look, I could be naive and that they're going all in. But I look at the way that they're controlling. This is some of the fact that that is the only published piece yet so far. I just feel like we've created this Schrodinger's China, where it's like, you know, it is or it isn't this dramatic threat. And nobody is ever, nobody's actually looking at like, what are they doing? You know, they're not racing towards they're not in some mission to race towards AGI. They have one company which is. They have a lot of companies which aren't. Their government seems very skeptical and semi curious about the way that things are happening, but very ready to shut things down. I don't see a lot of evidence that it's some Sputnik, Soviet style, star city, if you watch that show, type a program where they're like, we're going to pursue this at all costs. I do not see that whatsoever. Just Elizabeth C3B up on the experience my last point with regard to the politics. Right now, Democrats are being far more outspoken

1:00:13than Republicans not hard to understand why. Number one, Trump is all in on AI. So this is a tabulation at least 22 politicians responded directly to the tweet about, you know, the AI researcher who resigned and said, like, this is an existential threat. You had 22 politicians, three of them were Republicans. The rest of them were Democrats. It's not hard to understand why you have Trump who was all in on this. You have a Republican party that is very much in league with him. And you also have, you know, the tech world used to be seen as very like liberal democratic, especially Obama era. They were all given money to Obama and doing stuff at the White House and partnering with him even in his campaign. They have completely switched and gone all in for Trump. And so now the top echelons of the tech companies are very much, they're right winged, they're right wing coded, like that is. So they have less way with both the Democratic Party elites because they're not getting the money they used to get from tech world and also with the Democratic Party base.

1:01:14So if Democrats do take control of the House and or the Senate, then it's going to be okay. How does the rubber meet the road? They're saying they're going to do a lot of investigations into this that they're going to be hearings. The bigger question is, okay, who runs in 2028? What does the platform look like there? But that's going to be a major political question going forward is whether or not the Democratic Party actually has the stones to stand up to these titans of oligarchy and actually push back on the major capital class at this point because that's what is going to be required in order to rein this stuff in. Even in some of them pose like, oh, we would so love regulation. Look at how they spend their lobbying cash. Look at who the money goes to from their super PACs and it paints a very different picture of what they actually want out of Washington. Very, very true.

1:02:22Some people use chatGPT to answer questions. Others use it to get real work done. With chatGPT work, you bring the goal plus real inputs like briefs, notes, files, feedback, data, and project plans. It can help turn them into an analysis, tracker, deck outline, prototype, or workflow you can review and keep shaping. You stay in control, giving direction and choosing the final output. Put chatGPT to work on your most ambitious ideas and projects. Get started at chatGPT.com by selecting Work Mode, available on plus and pro plans. This episode is supported by FX's The Drop, a snowfall saga. Wanda and Leon are back, but this time they're chasing a different kind of legacy. As West Coast rap begins to take over 90s Los Angeles, Wanda is betting on a new generation of artists, but in a city where every opportunity comes with a price, they'll have to decide what they're willing to risk to leave their mark. FX is The Drop. Now streaming on Hulu. I love the little everyday moments with my dog,

1:03:24especially treat time, but dental care, not always so fun. Until we found Greeny's dental treats, these are more than just tasty. They're vet-recommended and designed to support the four areas Vets Check the Most, plaque, charter, gums, and breath. Their unique texture cleans down to the gum line and supports cleaner teeth after 28 days when fed daily. It's a simple daily habit that supports their health and gives us another moment to enjoy together. Greeny's Make Health a Treat. Shop now at Greeny's.com. This is an I Heart Podcast. Guaranteed Human.