The Artificial Intelligence Show Podcast - Ep. 233: OpenAI Pauses Some AI Training, Anthropic Controversy & State Opposition to Data Centers

Transcript

Intro

0:00 · We have these very very powerful models that the labs are having a hard time controlling and that even once they control them, we haven't removed the traits and behaviors. Like we've just suppressed them and those behaviors can show back up. Welcome to the artificial intelligence show, the podcast that helps your business grow smarter by making AI approachable and actionable.

0:27 · My name is Paul Ritzer. I'm the founder and CEO of Smarter X and Marketing AI Institute and I'm your host. Each week, I'm joined by my co-host and SmarterX Chief Content Officer, Mike Kaput, as we break down all the AI news that matters and give you insights and perspectives that you can use to advance your company and your career. Join us as we accelerate AI literacy for all.

0:54 · Welcome to episode 233 of the Artificial Intelligence Show. I'm your host, Paul Ritzer, along with my co-host, Mike Putut. We are recording Monday, August 24th, uh about 9:30 a.m. Eastern time. I don't know if we're getting new models or not this week, Mike. We've got a stealthy, what's the thing called? Ox Alpha. Is that what?

1:15 · Yeah, I believe so. Yeah.

1:17 · Yeah.

1:17 · So, if you're an exus user and you follow similar people that Mike and I do, your feed was two things over the last 72 hours. the robot Olympics in China and the Ox model that uh was getting all kinds of coverage and theories of whose model it is and we're going to talk about that. So, Mike's going to cover those. Maybe we find out who the secret owner of the Ox model is this week. I don't know. Uh leading theory seems to be either a China lab or Google. I don't know.

1:49 · Well, I don't know if I if there's like odds markets for this. It's not a poly market has like who owns stealth models or not.

1:57 · That wouldn't surprise me.

1:58 · Yeah, they should if they don't. All right, so this week's episode is brought to you by MEON, the AI conference for marketing and business leaders that is happening October 13 to 15 in Cleveland.

2:08 · This is our 7th annual MEON event. Uh if you've been thinking about joining us at MEON this year, we have a special offer exclusively for podcast listeners. We talked a little bit about this last week and we definitely saw some people registering over the last seven days.

2:22 · So, we hope to see more of you coming through. So, at MECON, we are hosting a private lunch exclusively for our podcast audience. During that lunch, you'll be able to ask uh Mike and myself any questions on your mind about AI, your company, your career, or whatever you're trying to navigate. It's a whole hour of totally unfiltered ask us anything about AI with fellow members of the artificial intelligence show audience. So, this is happening during Mayon. We are actually going to do this as part of Meon. So, if you register for MEON at mcon.ai, AI. That's m a n.ai.

2:54 · And use the code pod 100. Not only will you get a $100 off of your ticket, but you'll also reserve your place at this private lunch courtesy of the artificial intelligence show. It's our way of saying thank you for being a listener.

3:08 · If you've already registered for MCON using the POD 100 code, which many of you have, uh, then you're in. Your seat is already reserved if you'd like to join us, and we'll be sending an email out to you with all of the details. If you haven't registered for MECON, now is the time to act. Seats for this private lunch are limited and the offer closes as soon as the room is full. So, if you love the show and have been thinking about joining us for MEON, this week is the time to get your ticket at mcon.ai.

3:36 · Again, that's M A.I.

3:39 · All right. Every week we do an AI pulse survey. These are informal polls that kind of get the sentiment of our listeners uh for a topic we've covered on the podcast. We're trying something new. We're actually starting to promote these polls outside of just our podcast audience in order to get a broader range of responses. And so we left this uh question open for two weeks, I think, Mike. Yeah. Um and and started promoting it again through some limited channels.

4:05 · We're going to start doing some more stuff to put it out there, but we'd love to get, you know, 3 to 500 responses a week. So these are actually like much more projectable uh polls and data. But we had over 200 responses to the question, where is your organization at when it comes to deploying AI agents?

4:22 · And this is probably what I would think, and again, keep in mind, we're we're asking uh a group of people who listen to an artificial intelligence podcast every week. Um, so you would think that the people in this audience are going to be a little bit more AI forward than others. And the answer to the question again, where is your organization at when it comes to deploying AI agents?

4:45 · 18% say aggressively deploying across workflows. 49% selectively piloting in a few areas. Um 23% waiting and watching for now and 10% not on our radar yet. So again, only 18% say that they are aggressively deploying AI agents across workflows. Um I I don't know. I mean that feels about right to me.

5:11 · Mike, given that our audience is Morai Forward, if you were to ask me based on, you know, my time I've spent over the last 6 months with dozens of major companies and hundreds of executives, I would guess the numbers under 3% in in like the broader um the I guess the broader US space of businesses that I've spent time with. Um, uh, piloting would be about right, selectively piloting.

5:38 · I I could see that where some people are like testing them out, but vast majority of enterprises I spend time with, I I think I actually tweeted this this morning. Um, they're they're trying to still figure out how to use AI as an assistant across departments and roles.

5:54 · like the whole idea of using reasoning models for high level cognitive work and AI agents to automate workflows is is not even like on the near-term roadmap for a lot of these companies. So yeah, fascinating data. I I think again it's probably pretty representative of our audience. It's get 206 responses at the time we're um sharing this. So um yeah, fascinating. All right. Um so today we're going to Oh, by the way, you can go to smarterx.ai/pol a/pulse. That's where you can take part in the surveys.

6:26 · Mike will talk at the end again about what's going on with the survey this week. All right, Mike, I will turn it over to you um to get into, you know, I guess we've got a pause on some model development, we've got anthropic debates, and we've got data centers. Those are our three big things to start off the week.

6:43 · Indeed, they are. So, first up, this past week, OpenAI announced it had paused some of its most advanced AI training over cyber cyber security concerns. In a blog post they released titled pacing model development in an era of cyber critical capabilities. The company cited the recent security incident we've talked about at length involving hugging face along with preliminary evidence that their upcoming model which is uh termed Astra at the moment may meet the critical cyber

OpenAI Pauses Some Frontier AI Training Due to Cybersecurity Concerns

7:12 · security capability threshold under their preparedness framework which is OpenAI's internal system for measuring dangerous AI capabilities and deciding what safeguards they require. So, OpenAI now says that in response to that hugging faces incident, it took a two-week pause in reinforcement learning training on its latest models intended for deployment. Reinforcement learning being a later stage of training where models learn advanced skills through trial, error, and feedback.

7:40 · The company's largest planned frontier reinforcement learning run also remains on hold while it conducts smaller scale training and evaluations. Now, OpenAI also says it is expanding safety monitoring. They have the goal of issuing an alert within 30 minutes after concerning activity is surfaced. It actually the company estimates cost the cost of that monitoring is at roughly 20% of the computing power being used to run the models it watches.

8:09 · They also say they will evolve their preparedness framework to bring safeguards together across training and deployment. Open AAI CEO Sam Alman addressed the pause in a post on X, saying the company cares deeply about AI safety and believes the entire field will have to coordinate on shared safety standards, but that it will act unilaterally in the meantime.

8:31 · He said, quote, "Model progress is now extremely rapid and we always said we would take action if we felt that model capabilities were outstripping the pace of safety and alignment." He also added in this post, which was longer than what I'm reading here, he said, quote, "We expect confidence in safety to increasingly set the pace of AI progress." So, Paul, OpenAI has unilaterally paused some of this training due to safety concerns. I'm curious how big a deal this is.

9:00 · It'll I mean, it could be a really big deal if other labs follow suit. My my guess is other labs are probably doing something similar. Specifically, Anthropic is the one I I'm thinking of.

9:09 · um they're just probably not putting blog posts out about it, I would guess at this point. So context on this one, if if you're a regular listener to the show, you'll remember uh pacing the frontier, the article or I guess the statement that was signed by over,300 AI lab leaders and um researchers. So we talked about this on episode 228, which was on August 4th, so just you know 3 weeks ago.

9:36 · So pacing the frontier came out in that uh statement the it started with to realize AI's potential uh industry government and society at large may need the option to buy time to address emerging risks develop security measures and strengthen oversight but each company and country is under intense competitive pressure not to unilaterally slow that acceleration and today the world lacks the technical and governance tools to deliberately pace frontierwide progress uh than The actual

10:07 · statement that the signatories signed on to was, "We request that the US government support an international effort to develop the technical and governance tools needed to deliberately pace the frontier of automated AI development." Now I bring this back up because some of the key signitories was uh Yakob Pachaki who's the chief scientist at OpenAI, Mark Chen, who is the chief research officer at OpenAI, Dario Amade, Ilia Sutskava, Shane Le,

10:32 · co-founder of Google DeepMind, and then I wanted to highlight a few actual excerpts from this because when people signed it, they could also put personal comments and so I think it's like an interesting perspective to hear what the people in these labs are actually saying. So you can go to, we'll put the URL, but it's pacingthef frontier.com.

10:51 · You can go read all of these personal statements yourself. I found them to be pretty fascinating. So Leo Gao, member of the technical staff at OpenAI, which is what they call their researchers. The world is locked in a deadly race towards an intelligence explosion where AI's ability to create better AIs reaches a critical point, just like a runaway nuclear chain reaction. Going slower would give us much needed time to make it go well. But no individual actor is willing to stop unilaterally to survive.

11:22 · We must coordinate to slow down the race. So again, this is someone within OpenAI. And that's why I wanted to highlight these. These are the people in the labs who are seeing the stuff we are not seeing. They're seeing these advanced models that haven't come out yet that they're having to slow down.

11:36 · And as of 3 weeks ago, they would have known all of this. So like it's a good kind of snapshot in time. Uh, Xenji Xiao, chief scientist from Meta AI, formerly the super intelligence lab, or maybe it's still the super intelligence lab. I don't know, they've changed the name a couple times. Um, said, "AI is progressing at a rate that our society might not be ready for. Frontier labs are very close to AI that can exceed even the best people on almost every metric of intelligence. This will lead to unprecedented societal and safety risks.

12:04 · To ensure a positive future, we need to develop AI in a way that is driven by responsibility and thoughtfulness. And then one of the biggest names, actually like the top name on the page, if you go to it, the person they highlighted before Dario Amade, John Schulman, who's the chief scientist of thinking machines, he was also he's also the co-founder of thinking machines with Mir Maratti.

12:26 · Um, prior to that he was a co-founder of OpenAI and then he was uh one of the leaders on the alignment science team at Anthropic for about nine months before he bounced and went over to thinking machines. So he said signed because this statement helps establish common knowledge about the possible need for coordination mechanisms as automated AI research accelerates progress. I'd also like to see labs start designing these mechanisms voluntarily even before the US government gets involved. Okay.

12:58 · And then Sam, right around the time that that came out, did an interview um and he had just met with Republican and Democratic senators in Washington discussing OpenAI's upcoming AI model, which I assume was Astra at the time, and that that he was stressing to White House officials the need to slow down AI development. Um, the article said, "We've talked about the need to pace it as the models get more capable, which I think is in everyone's interest." That was a a quote from Altman. Um, so that's in a Bloomberg article.

13:27 · So, when we get into the actual news of like what's happening here, they're basically just saying it's getting harder and harder to align these things with what they intend them to do and how they intend them to behave. As we talked about with the hugging face incident, this went on for weeks um before OpenAI had any idea that their model had done anything. And actually, if I remember correctly, Mike, it was like May 7th that it first started going off the rails. And then it wasn't until was it July 19th or something that they like?

13:57 · Yeah.

13:57 · Yeah.

13:57 · I mean, we're talking about like 6 weeks before opening realized that their own model that they knew was showing signs of misalignment actually went and did things to other companies without them knowing. So they're now committing to like, okay, we want to know and then alert people within like 30 minutes of misaligned behavior. Um, so they know that the models are deceptive by nature.

14:22 · Like these models have these capabilities that that are inherent. And what the labs are trying to do is they're trying to suppress these abilities. So they're not saying we're going to train new, more powerful models that just don't have these bad traits.

14:40 · They're saying, "We're going to keep training these more powerful models, and we're just going to get better at monitoring the misbehaviors and the misalignments, and then when it's severe enough, we need to be able to step in and do something about it and alert people." So, what they're trying to do is put through reinforcement learning where they try and tell it like I'm I'm I'm oversimplifying here, but like be good, do good. like they're basically trying to reinforce this thing to not use the negative traits and skills it has against people and companies.

15:09 · Um, and so they're trying to put these guard rails around it even though their own research tells them that the models increasingly understand when they're being monitored and tested and and are very good at deceiving researchers to to not realize that they're aware of this.

15:30 · So, I don't know. When I think about it, Mike, like, you know, I try and always bring this back to like what does this mean to our listeners? Like what what is the impact on business users? Um, my guess is it's pretty minimal. Like the fact that OpenAI is slowing down progress on one of their more advanced models, it doesn't mean much to you and I dayto-day. Um, it does mean that the most powerful frontier frontier models will increasingly be used for the highest level cognitive tasks um within the labs like they're going to have access to do these kinds of things.

15:57 · Um, and I think that to me is maybe like the the most obvious outcome in the near term is the labs are going to have the most powerful models. They may choose to re not release them for a time a period of time, but they're also going to give them to select partners who will also then have access to these most powerful models, including the government.

16:21 · So that's the part that, you know, again, does it affect us dayto-day? Not really. like we're generally speaking businesses are still going to have access to a collection of open and closed models that you need for your day-to-day work and are largely going to be sufficient for what we're all trying to do. So the models we're talking about here are for the absolute most advanced stuff. And then uh my final note, Altman on the the tweet where he put out about this slowdown, he also did reply and said, "We still expect to ship great new models soon.

16:52 · This impacts further out releases." So I I again I think it's super important that people understand what's going on here that we have these very very powerful models that the labs are having a hard time controlling and

17:08 · that even once they control them we haven't removed the traits and behaviors like we've just suppressed them and those those behaviors can show back up like people can find ways to unlock those behaviors and then I guess the other highlight here would be whatever the concerns are here on these proprietary models.

17:29 · We've talked about this before, but you have to assume 6 to 12 months from now at most there will be open-source models of equivalent power that people will not withhold releasing, right?

17:42 · So, yeah, I don't know. I mean, just it's super important to understand this stuff. I just does it affect anyone tomorrow? No. But big picture, it's a massive thing to understand what's going on. Well, I'm glad you mentioned the open source models, too, because this starts to, I think, connect the dots between some of the things we've been talking about a lot this year, where it's like, okay, if we assume, and we'll talk about this in a couple of these topics, to some degree, access to AI or compute equals, as Jensen puts it, revenue, but in a lot of cases, power.

18:15 · It's like this inherently could consolidate power and money-making ability to those organizations to the labs. But then people say perhaps open weight open source is a counterweight to that but also fraught with huge problems based on what the models are now capable of.

18:34 · Yeah, it's a very messy and very uncharted territory. Like I just and we'll get into some of the stuff around the data centers in a minute and the impact like what the government's trying to do, but it like the government has no ability to solve this. Like they're they're so worried right now about public perception of data centers and soon to be public perception and reality of impact on jobs that I just I I don't know that the government's going to be able to solve for the complexity of this um anytime soon.

19:04 · And I do think it's going to be on the labs to coordinate efforts. And I my guess is they're all spooked enough now that there's probably a whole bunch of conversations going on behind the scenes between the AI lab leaders realizing that they're going to have to unify in some way and come up with a plan because I don't think the government's going to do it.

Debate Over Anthropic Intensifies Ahead of IPO

19:25 · All right. So, next up, somewhat related. We heard that Bloomberg reported this past week that Anthropic's revenue run rate, so the annual revenue implied by its current sales pace, passed 65 billion at the end of July, up from 47 billion in May. The company now expects its IPO to match or beat SpaceX's record listing that happened this summer, which targeted $75 billion and ultimately raised about $86 billion.

19:52 · And we're anticipating a public filing coming soon. Now, with this big milestone looming, there's been some debate and controversy about Anthropic and its direction. And those debates have been intensifying, most visibly this past week in a very public debate that broke out on X.

20:09 · So, this all started when Gavin Baker, who we've talked about quite a bit, managing partner, chief investment officer at a trades management, said on the All-In podcast that he had quote been told by multiple people I trust that Daario has said that Anthropic might end up being the only private company in the world at some point. So, he added on the podcast, quote, think about that in this vision and anthropic maximalist vision. There's anthropic and then there are governments and that's it.

20:38 · Now, this kind of debate got started or intensified when anthropic researcher Schult Douglas uh kind of came out and said, "Look, that's completely false." And then Baker kind of responded with a long post arguing that all this talk Anthropic CEO Dario Amade has had about safety messaging, all of it's backfired and it's basically handed ammunition to anti-data center campaigns. He said, quote, "At this point, I think it's safe to say that Daario has lost the argument. his messaging has failed to result in his preferred regulatory path.

21:08 · The fact that the only solution to the recent incident where an unreleased advanced open AAI model hacked HuggingFace was an open-source model likely ended any chance of strict near-term regulation.

21:22 · He did say among many other things that Anthropic was an amazing company. But what then kind of lended weight to this is that Amade himself posted a rare lengthy reply calling this idea that AI must either be concentrated through regulation or distributed widely quote a false choice. He pushed back on Baker's claims that his negative messaging on AI is helping drive the public's negative view on AI.

21:46 · I thought this was interesting. and he wrote, "I do agree that the public has a negative view of AI and this is a big problem, but I don't think it is primarily caused by me or any other AI leader warning about AI's risk. I think it is fundamentally a crisis of trust. I think that ordinary people don't trust companies, governments, or the tech industry and always suspect that we are looking up cooking up some new way to screw them over. The causes of this go back decades and AI is just the latest iteration of it.

22:14 · I don't think that a glitzy marketing campaign with a positive spin, which some have advocated that Anthropic do, is the way to win back that trust.

22:22 · And obviously, this comes as they are literally eyeing a valuation north of $2 trillion in what could be the biggest IPO in history. So Paul, the reason we're talking about all this is because with that IPO looming, what Daario, what Anthropic believes, what Daario specifically believes becomes really, really important because a relatively small number of people are determining the direction of one of the most powerful technologies created at one of the fastest growing companies ever created. Curious what you took away from this exchange.

22:56 · the first on the revenue and the valuation, you know, the potential market cap at two trillion. It's just hard to comprehend. I don't I don't know that most people have the the ability to really think about these numbers and how significant they really are. And then just going back to a point I just made, like we're talking about, you know, potential $2 trillion. So the largest IPO ever was at 1.7 with SpaceX. Um raising, you know, 85 billion is what SpaceX did. getting to a hundred billion. It's just insane numbers.

23:26 · And then I I can't help but think about how early we are in the adoption curve. So again, I I as I just mentioned like most organizations are still trying to figure out how to use like chat bots and assistance across all functions in an organization like marketing and sales and service and ops and HR. And like we haven't even saturated that market. We haven't even come close to saturating that market.

23:50 · And then when you layer in that organizations will eventually figure out how to use AI as a strategic thought partner in you know decision making data analysis they're going to figure out how to infuse agents across workflows and roles and then the demand for the models and the demand for the compute that that powers all of that is going to skyrocket. And so all of these numbers we're talking about are in like the first inning of AI adoption and transformation in most organizations.

24:18 · So that's like a really hard for thing for me to wrap my brain around. Um that's why I always kind of laugh when people like ask me about again I'm not providing investing advice here. Um when people ask me, hey do you think Nvidia's like hit its peak? I'm like are you kidding me? like do you have any idea how how early we are and how big this is going to get? Um so anyway, all right.

24:46 · So then the Gavin thing, it kind of bothered me, honestly. Like I really like Gavin Baker. I I I love listening to his interviews.

24:54 · I'm not a big all-in podcast listener. I kind of gave up when it got really political a while back. Um, but I do listen I do dip in for specific segments and sound bites like but if I hear there's a like a really important segment I will go listen to that segment. I just can't do the whole show anymore. Um, but that being said when I saw this clip I was like why is he doing that? Like what an absurdly ridiculous rumor to feature and like and then all the all-in team, you know, as they'll do that just ran with this sound boy.

25:26 · But like to he was very clear in the clip that like I've heard from multiple people like these rumors that Daario says they're going to be the only company left in the world. I'm like that is absurd. Like even if it was said in passing actually, there is no way that that's like a literal thing and for them to make it like the focal point of that and then you know push this clip out there. I honestly just felt like it was playing to David Sax cuz I know Sax was on that episode.

25:57 · It was like playing into this like hatred for Daario and Anthropic for for no valid reason other than it just seems like people love to pile on them.

26:08 · So I was glad when I saw Schulto Douglas reply and I follow Shoto. So that's actually how I first saw this like I got the alert that he had, you know, tweeted. So his reply was, "This is completely false. I like Gavin's takes, but whoever he heard this from is lying so that it fits the narrative some people so desperately want you to believe." The same people will try to convince you anthropic has no moat and a sentence later that it might become so powerful it could be the only company left.

26:36 · So then there was a really nice actual interaction back and forth, very civil interaction between the two. So then Gavin Baker responded and said, "Shto, thank you for setting the record straight." And and then you could almost see like he kind of like backed off like, "Oh man, this got a little bit bigger than I was expecting, so I'm going to sort of try and soften this."

26:56 · So he said, "Larger issue is that multiple very serious people in Silicon Valley have heard some variation of this and believe it to be true. And the reason it is believable to so many is that is consistent with Daario's public messaging and what he outlined in the essay you shared. This technology might be dangerous for humans in multiple ways, could lead to extreme concentration of economic power, and therefore needs to be regulated thoughtfully. I agree with the potential risks, and I believe Daario makes all these arguments in good faith.

27:24 · Daario's messaging has been massively helpful to efforts to ban data centers here in America. I suspect we will see anti-data center advocacy groups running ads using clips of Daario warning about how dangerous AI could be for humans. His good faith efforts in favor of regulation are now increasing the odds that AI will not be beneficial for Americans and humans everywhere. So he's basically like saying it's Daario's fault that people hate data centers which again also an absurd position which I'll explain in a minute.

27:53 · Then Schol replies, um, this is why Daario us talked about risk so much. Fundamentally, it's because we wanted to be honest with people. I think this is really interesting from Shto.

28:06 · Employment risk is the classic here. I actually disagree with Daario on the pace. I think it's most likely that compute shortages, diffusion complexity, policy and unde unmet demand for services mean that even for years after we have models which could automate 90% of computer-f facing jobs also called knowledge work um models will get there in 2028. So I'm going to pause here for a second.

28:31 · So he's so Shoto is saying he actually believes and again keep in mind he works anthropic he sees these models he thinks that we will reach a point by 2028 in which the models could automate 95% of all jobs by 2028.

28:51 · So two and a half years max from now right but he is saying that let me I'll finish the the quote and then I'll come back to it. Um, so he says which could automate 95% of computer-f facing jobs.

29:03 · People will work at them well into the 2030s. But I do think we as a society should take the possibility far more seriously than we are now and prepare contingency policies for what we do at various levels of unemployment. For example, you could imagine not letting profitable companies lay off more than 5% per year. That's actually a really smart thing. as well as meter style evals to measure progress on different job families so we have a clear picture.

29:29 · Our opinion has always been that we need to be straight up and honest with people. So I'm going to come back to that middle part. He's saying he thinks we will get there to your point Mike 2 and a half years from now where the AI can do 95% of all cognitive work all all knowledge work. But he's saying that doesn't mean jobs go away. it could take into the 2030s because of compute shortages, diffusion complexity, policy, and onet demand for services. So his point is like listen dude, we're just trying to be honest with people like this is what we actually believe.

30:00 · Why if we think there's risks or why if we think there's, you know, job related risks, why would we not say anything?

30:07 · Like what is the point of that? So then Daario, as you mentioned, shows up on Twitter for like the fourth time this year and he says, "On messaging around AI, I do not agree that my messaging has been disproportionately negative. In fact, it has been equally balanced between risks and benefits. I've written one major essay about each, which we've covered on the show, and even in interviews where I discuss the risks, I make sure to frequently mention the incredible benefits as well as proposing possible solutions to the risks.

30:36 · In fact, I wrote Machines of Loving Grace because I didn't feel the AI industry was painting an inspiring enough picture of how the technology could radically transform the world for the better. And then just to like expand on my point that they're just like they they make it sound like this is just all Daario, which I've never understood.

30:55 · Yeah.

30:55 · There's an interview that just came out on August 23rd, so this is yesterday with David Senra. I've never listened to his podcast before, but he interviewed Altman. And here is Altman. We have not as a field done a very good job of explaining to people what the benefits are and how the downsides could be mitigated. And we certainly have not done a good job even if people have had answers like saying there's going to be UBI or work will be optional. There's been very little discussion from people about how and why it's important that people have more power and personal freedom in the world, not less.

31:26 · And that matters a lot to most people. The ability of people to influence their own future and collectively design where society is going to go and the autonomy that comes with that is very important.

31:39 · I don't think a lot of people in the AI field, they feel it for themselves, but they don't spend time thinking about it, reflecting on or acknowledging how important it is to other people. So, he's basically saying is, yeah, we've failed at this. Like Altman, as much as anybody has said, yeah, jobs are going to get, you know, basically decimated and it's okay. we'll have like UBI or something like we'll come up with some way to solve for this. And this is what we've been saying in this podcast for like four years. It's like have a plan.

32:06 · Like nobody has all these AI lab leaders have been talking about it's going to change society. It's going to impact jobs. It's going to do all these things.

32:14 · There's all these risks. It might destroy humanity. Like you know the P doom. Like there's a 10% chance humanity is wiped out. And then they never had a plan for what that meant. And of course eventually that was going to catch up to them and the public was going to realize that. And so now you have all these changes in tone and messaging and it's just like it's too late. Like it's I don't know. So I just get really really annoyed. I'm not even trying to stick up for Dario honestly. Like I don't I don't know Daario. Like I I I think he's pretty honest with how he feels whether you believe that stuff or not.

32:42 · But for some reason the tech accelerationist group or the EC or whatever it is, they just hate the guy and like everything is just Anthropic's fault or Daario's fault. And I don't understand it at all.

32:55 · Yeah. Yeah. I also wonder how much your average person is going to hear things like AI is going to take jobs or AI is going to potentially have a very negative effect on humanity. We have to work against that. And then your first question without knowing the AI space would be like, well, why do you keep still building it? Right? It's like it's like your average person. I I think without a lot of context, that's a very reasonable response. And so therefore there's this huge trust gap I would imagine as a part of it.

33:25 · And they all been asked that question.

33:27 · They all say well we if we stop everybody else is going to keep going. So like we got we got to get there and try and help it happen in the best of you know in the best interest of humanity or whatever.

33:36 · Yeah. Which is a more nuance point to get across than some of these like sound bites and clips and fearmongering for sure.

Pennsylvania Leads Increasing State Opposition to Data Centers

33:43 · All right. So third topic this week very related to what we've been talking about. So, Pennsylvania Governor Josh Shapiro has signed an executive order that he says puts the nation's strictest guard rails on AI data centers. So, the order requires developers in Pennsylvania to pay the full electricity costs tied to their facilities, meet strict environmental and transparency standards, and win approval from the local community before the state will issue permits.

34:08 · It also removes all data center projects from Pennsylvania's fasttrack permitting program and bars state agencies from signing non-disclosure agreements with data center developers. Announcing the move on X, Shapiro wrote, "I will not allow Pennians to be bullied by greedy developers and bulldozed by the lawyers working for these big tech companies.

34:31 · Pennsylvania is part of a wider wave of state action. We've talked about uh how New York Governor Kathy uh Hokll paused permits for large new data centers for up to a year in July. Texas Governor Greg Abbott froze new projects pending a grid audit. And local opposition blocked or delayed at least 75 US data center projects worth worth roughly $130 billion in the first 3 months of 2026.

34:57 · And you know the political feelings around this are shifting very quickly too. Axios has reported on some Gallup polling that finds 70% of Americans oppose an AI data center being built in their area. There was also a leaked memo from the Senate Republicans campaign arm warning top AI companies of data centers poll about as favorably as quote spent nuclear waste. And this also could cost Republican Ohio Senator John Hustead his seat in this fall's midterms.

35:25 · Um, we're also hearing some reporting, according to Wall Street Journal in Bloomberg, that big tech is launching a charm offensive. OpenAI is holding community meetings around a planned facility worth more than 30 billion near Savannah, Georgia. Meta is running TV ads featuring local workers at its Iowa data center. Microsoft is pledging to cover its own power costs. And Oracle is sponsoring food pantries and youth groups near a project in New Mexico.

35:51 · So Paul, we've talked plenty about opposition to data centers, but honestly, it really seems like this summer this has blown up as an issue and really really intensified and the states appear to start uh getting in on the action because there's some political hay to be made for sure.

36:11 · Yeah, if you missed episode 232, go back and listen to the segment where we talked about the Ezra Klein interview with Jasmine Sun. there there's a lot of like really valuable insights from someone who's who's gone into the the communities where this is happening, met with them, met with different stakeholders. Um, so she had a really valuable perspective that we shared in that episode. Um, the Wall Street Journal, I'll dig into that one first.

36:33 · So the headline of that one is inside big tech's frantic race to quell the growing backlash to AI. So they say data said our opponents frequently cite rising electricity prices, environmental impacts, and noise as grounds for fighting new projects. And some also oppose the development of AI itself, believing it'll wipe out large numbers of jobs or otherwise degrade society.

36:50 · Uh in the article, there's a uh a section that says, "The powerful tech companies behind the AI infrastructure buildout are putting on a fullcourt press to solve a public relations crisis that has been derailing projects and scaring off political allies, including Shapiro, who was an advocate of data centers. um fearful of getting caught up in the blast radius. They are changing how they negotiate data center deals, adding sweeteners such as jobs guarantees and clean water investments.

37:18 · They are touting the way data centers are shifting tax burdens off nearby households in some cases. Meanwhile, their top executives are increasingly talking about AI in ways that emphasize potential broad-based economic opportunity rather than job displacement or other scary scenarios. Sounds familiar from the previous topic. in Georgia in particular. They said there was a richer uh fair on offer in announcing the project.

37:43 · The company pledged to invest 80 million in the community and provide as much as 71 million in coding credits to local students. So you're seeing like these kind of like innovative things I guess.

37:54 · And then uh there was a cro L Chris Lean Openi's chief global affairs officer who is a busy guy these days I'm guessing said hosting an openhouse was an important piece of winning community buyin. people really do want to know, are my electricity bills going to go up or not? Is this going to impact my water supply? Am I going to pay more or less in taxes? That was his quotes. Um, and then, as we've been talking about on the podcast for well over a year, the 2026 midterm elections in November are the forcing function here.

38:25 · So, there was an article in Bloomberg that said, "Suddenly, residents and officials in red states and swing states such as Texas, Mississippi, and Ohio are echoing progressive leaders like Bernie Sanders and calling for pauses on the new data centers. They're all looking at the same polling data. So, they all kind of know where they got to go." And then I found this GOP letter to be really fascinating, Mike. So, we're going to put the link to Axios in there. They actually have the letter.

38:50 · So they said in a memo obtained by Axios, the National Republican Senatorial Committee says Democrats have made data centers a centerpiece of their campaign to defeat Senator John Hustead and that it's working. Uh Hughstead's opponent in Ohio, former Senator Sherid Brown, has spent millions of dollars on a summer ad campaign calling Hustead the face of data centers in Ohio. Mike, I don't know if you watch Cleveland Guardians games, but like as someone who lives in Ohio in Cleveland and watches sports, that's about the extent of my TV watching.

39:24 · Yeah.

39:24 · These freaking Houston ads are going to drive me insane. Like they're you'll see seven to 10 of them every Guardians game. It drives me crazy. Like I can't get to the mute button fast enough for this stuff. Um and I'm not even saying like right or wrong on data. I'm just I hate political ads. So it's just like it's a lot. Um he's definitely Brown is definitely spending money on these ads.

39:44 · Um okay. So then it goes on to say if he loses Houston and data centers get the blame, politicians across the country will take notice and they will not go near the next one. This has become a sleeper issue for the entire election cycle. According to the memo, Republicans say in the memo headlined Ohio data center risk that it's up to AI companies to improve perceptions of data centers by explaining quote who benefits, who pays, and why a community should want one.

40:13 · Um, until that happens, the issue will continue to dominate this race, the memo says. Uh it goes on to say, "If voters perceptions of data centers are not fixed quickly, the campaign against them will expand far beyond Ohio." And then just for context, there are now more than 4,000 data centers online across the US with another 3,000 plus proposed or in progress. So, you know, trying to put this in context, Mike, and why it seems like almost every week now we're talking about data centers.

40:44 · This is a very macro level, but just like some some thoughts I was having as I was kind of getting ready for today. So I am not diminishing environmental concerns, not wanting these things in I'm just like assuming all of that is true and equal whatever without this data center buildout and the capex spend because we're talking again almost like $200 billion this year from Meta, from Google, from OpenAI.

41:10 · Well over half a trillion I would imagine combined this year is being spent on data center buildout and the capex behind this. Um, without that spend and the jobs that come with it, the economy stalls or crashes. Like if we don't have that money being spent right now, the economy looks nothing like people think it does. Like it is holding up the economy.

41:33 · Basically, the Republicans, and again, if you're new to the show, totally neutral here in terms of Republicans, Dem. I don't I don't care. I don't care who you vote for. I don't care like what your, you know, beliefs are about this stuff. We're just trying to be objective here and like here's the facts. The Republicans are in a very tough position because voters don't want data centers, but the data centers are needed to drive GDP, economic growth, and to compete with China. So, the Republicans are the spot where it's like they know they need the data centers, but they can't say they want them because that's bad.

42:06 · Um, so then I started thinking like let's say the Republicans manage to hold the House in the midterms by shifting data center messaging, getting these AI companies to like, you know, shift their messaging.

42:19 · But then that turns into the actual political position and policy. If that's the case, so let's say Republicans publicly say, "Hey, we want to slow this down. Let's like, you know, ease this up for a minute." But then it ends up actually leading to a slowdown in the buildout of data centers, thereby a reduction of the capex that's going to be spent by these labs. Then there's a much larger reckoning coming in the 2028 election cycle because now the trickle down of that lack of GDP growth, the the impact on jobs, the competition, all of that becomes an even bigger issue.

42:50 · So the only way like this charm offensive that the labs, you know, the Republicans want the labs to do, it's not going to shift consumer public opinion fast enough. Yeah.

43:04 · The labs need very very quickly a massive breakthrough that'll have broad appeal and benefit fast enough in order to change the narrative. Meaning we're going to solve disease and we actually had this massive breakthrough and it's coming very soon. Like we're going to cure cancer next year kind of thing. like they need something that dramatically shifts the narrative. Um otherwise, I don't know how they come back. In the meantime, what we're all going to get in society is a whole bunch of very optimistic essays from AI labs leaders.

43:34 · We're going to get opinion pieces in the Wall Street Journal and New York Times saying how amazing AI is going to be for science and biology and all these things. We're going to get podcast interviews with Sam Alman and others. Um Mark Zuckerberg essays. We're gonna get news segments featuring AI lab leaders touting the benefits of AI and downplaying the impact on the environment and jobs. Like we're gonna get the charm narrative and that's what everybody's going to be exposed to.

43:57 · Now whether voters believe any of it, I don't know, but I would not trust the stuff that you're going to be hearing out of most of these AI labs in the coming months because they're just trying to win people over. Um, and then I also think we're going to see more crossovers into pop culture like we saw last week with the Liquid Death and Garage Beer ad. And I know you saw this, Mike. Um, so there was an ad featuring NFL legend from the Philadelphia Eagles, broadcaster, and future Taylor Swift brother-in-law, Jason Kelsey, which has 11 million plus views on X.

44:28 · And I don't think I'll do it justice by reading it. So, Mike, let's play the the first like 12 seconds of the clip from the actual ad here.

44:42 · AI data centers waste millions of gallons of water. That's why Liquid Death and Garage Beer have decided to team up. We want your pee to cool these data centers. We want your pee. Please give us your P. All right. So, if uh if you haven't seen it, go watch it.

45:02 · There's about another minute and 15 seconds of a song about sending pee to data centers. Um there is an entire web page on liquid death dedicated to we want your pee campaign. And then there's actually um like a a a cup you can buy a glass mug that uh you're supposed to send your pee in. But then I I I did laugh because underneath it, which is sold out by the way, so if you wanted to buy one of these, you can't.

45:30 · Uh, it says, "Mailing your P to AI data centers has never been easier with these limited edition realable jars. Just fill these and send to the data center of your choice. Also works for drinking your favorite beverage as well." Then it said, "The suits want us to tell you, please don't actually send your pee.

45:46 · Limited availability while supplies last only." So I, you know, it's like we've crossed over into the parodies and and the ads and everything. So it's going to be pretty fascinating, Mike. I I don't um I I don't think that uh political advisers and campaign leaders are going to have an easy time here trying to figure out what direction to go and we're going to have politicians jump and ship from supporting David Sers and saying, "Hey, just give us like, you know, 3, four months.

46:18 · We got to win this election. Then we'll come back and everything we said over the last three or four months was a lie anyway and like we'll help you do what you got to do." I don't I don't know. It's it's a really crazy environment right now.

46:29 · Well, someone on X, I believe, mentioned this or paraphrased it from someone else who said it, but it was actually kind of tongue and cheek, but I thought interesting. You know, we've talked about Elon Musk wanting to build data centers in space. And they were like, "Guys, he doesn't want to build them in space because it's better to build them there. He wants to build them in space because you can't throw Molotov cocktails at him in space." And I was like, "It's actually probably a decent point."

46:55 · Yeah. Yeah. viable. That is we won't get into the next generation of warfare between satellites, but yes.

47:02 · Yeah, exactly. Right. Right.

47:03 · Yeah, you could throw something at them.

47:05 · Yeah, for sure.

47:08 · All right. So, before we dive into this week's rapid fire, just a quick announcement. This episode is also brought to you by AI Academy by Smarter X, which helps individuals and businesses accelerate their AI literacy and transformation through personalized learning journeys and an AI powered learning platform. New educational content is added weekly so you always stay up to date with the latest AI trends and technologies.

47:28 · And we're spotlighting here this week our AI for departments collection which features seven or eight rather different course series and certificates designed to jumpstart AI understanding and adoption.

47:42 · So we have each of these ondemand core series each with a certificate. We've got AI for marketing, sales, customer success, HR, finance, operations, legal and IT. And these are an ideal launchpad for organizations that want to level up their teams and accelerate AI adoption and impact. We've got individual and business account plans available now. Or you can buy single courses and series for onetime fees. So go to academy.smarterx.ai to learn more. And you can use the code pod 100 for $100 off any individual plan. That's academy.smarterx.ai.

48:18 · All right, Paul. So all these topics very interrelated here. This past week, our first rapid fire, Nvidia founder and CEO Jensen Wong published a lengthy post on X titled securing the infrastructure of intelligence. And in it he argues that land power and the buildings that house computing, he calls these quote land, power and shell, have become the next critical resource for what he calls quote AI factories. his term for data centers that turn energy and data into intelligence.

Jensen's AI Factory Manifesto

48:49 · The post then goes on to announce some details that Nvidia is partnering with SB Energy to lock up land power and building capacity at the Ports Pike technology campus in Pike County, Ohio on the site of the decommissioned Portsmith uran uranium enrichment plant. SB Energy will build, own, and operate this data center under a 20-year lease, and Open AI is the tenants. This initial deployment is expected to provide 4.25 gawatt of capacity coming online in phases starting in 2028.

49:20 · Hang says that each generation of Nvidia systems there could represent roughly 1.5 million GPUs, which translates into 150 billion to$200 billion in Nvidia revenue. Hong also says OpenAI's existing and planned commitments represent about 12 gawatt of Nvidia compute through 2030. there's room to expand that to about 16 gigawatts and he pegs this opportunity at roughly $600 billion of Nvidia compute.

49:49 · Uh in the post he also answered the question there's been some talk about circular financing in AI and he answers that OpenAI will be paying this lease. So it's not kind of a financial shell game really to hear him tell it. These are paying customers uh reserving this compute. The Wall Street Journal reported Nvidia's backing for the first phase of this could reach $105 billion.

50:11 · They're also investing 1.5 billion in SB Energy. So Paul, according to the Wall Street Journal, this is actually expected to become one of the largest AI hubs to date. Curious to hear. I mean, this ties into a few things we've been talking about, especially my gosh, when you said Nvidia is just getting started.

50:28 · One of these projects is worth that much in compute, hundreds of billions of dollars. Yeah, Nvidia certainly does not want to see a data center slow down.

50:36 · They want as much demand for AI as possible because that's what their chips power and they're building these AI factories and they're just going to make money every way. They don't care if it's open source, proprietary, like they're they're still getting paid. Like yeah, it's Nvidia just does not lose in any of these scenarios unless there's a massive slowdown or breakage of the supply chain.

50:55 · Um, one interesting note here, there's been some conversation within AI circles that like data centers is just a bad name for it and that part of the reason the public hates these things is just bad naming and then we call them Colossus and all these like, you know, big sci-fi type names. I don't think AI factories is winning over public sentiment either. Like so they Nvidia may need a rebranding too and and Jensen's been calling them this for years now.

51:23 · This isn't a new thing, but I don't know that AI factories plays better in public sentiment than data center does. Um, so a couple of notes from his expost, which by the way, he just started posting on X like 2 months ago also. So this is him and Daario both like in the last year have started showing up on X. Um, AI factories are the defining infrastructure of the AI era where compute transform energy and data into intelligence that powers every business, industry, and country. That's his, you know, feelings on what AI factories are.

51:52 · And then he says, "Frontier labs have extraordinary demand for training and inference compute." Again, inference is when you and I use the models. Training is the cost to to build them. Um, but many are growing faster than their balance sheets and long-term credit profiles can support.

52:10 · So he's saying these labs can't afford to take on the risk they're taking on to build for the future through traditional mechanisms that the banks aren't going to like lend these labs this kind of money and it's going to be increasingly difficult for them to raise it um through private markets. So goes on to say they may have strong customer demand and rapidly growing revenue yet still lack decadesl long infrastructure contracts and investment grade financing capacity needed to secure the AI factory infrastructure independently. Meaning they can't afford this but we can is is pretty much what he's saying.

52:40 · Their growth is increasingly increasingly constrained not by algorithms or customer demand but by the availability of compute which we and video also can solve. For these companies, more compute means more intelligence, more products, more users, and more revenue. Nvidia is helping provide the infrastructure that powers this flywheel. So, they can't afford to do this buildout. We can, and by us helping them afford it, they spend more money with us. And so, like, let's go.

53:08 · Let's just keep let's keep building, keep growing, and let's all keep making money is pretty much what it comes down to. So obviously Nvidia benefits by selling all the compute related to this, but also just sounds like they're becoming the AI bank when other traditional lenders won't get as involved or can't.

53:24 · Yes, that's exactly they're just going to underwrite this in different different ways. So um yeah, and then the other interesting thing, I don't know if we get into this in the product and funding updates, Mike, but they're also all of a sudden a major player competitor in openweight models. like they're going to compete directly and that my premise is and the story I think Jensen would tell to the labs is demand is almost infinite.

53:49 · People are still going to buy your models and we'll support you in all these different ways, but we're going to we're going to go ahead and build our own openweight models and sometimes people choose to work directly with us and I I mean what are you going to do? You can't be mad at Jensen because he's he's the guy that provides the chips that let you build your company. So I don't know. It's all so bizarre. They all compete with each other and work with each other and it's been going on that way for years.

AI Job Loss Hits Young Adults

54:15 · All right, next up. This past week, Axios reported on some new Pew Research Center data showing that AI optimism is fading among young Americans, which is an age group that had been one of the more positive about the technology. And this is happening largely over fears about jobs. So in this Pew survey of 3,488 US adults conducted in late June of this year, they found that for the first time, a majority of adults under the age of 30, 55% of them say they are more concerned than excited about the increased use of AI in daily life.

54:46 · That is up from 31% in 2021. And only about 1 in 10 now say they are more excited than concerned. The job outlook is a major driver here. So Pew found that 73% of adults under 30 now say AI will lead to fewer jobs in the US over the next two decades. Up from them saying 61% just saying this just 2 years ago.

55:10 · About 27% of Americans aged 18 to 34 believe they or someone they know has lost a job due to AI. This is according to a new Axio generation lab poll. And as Axios and others have pointed out, this timing is just especially tough for people entering the workforce because young adults are heavy AI users, but they're also competing for the entry-level jobs often that AI can threaten. So Paul, this is some real interesting data.

55:40 · Obviously, it's just a couple polls here, but I mean that 27% stat jumped out to me a little bit too here. I mean, that's kind of a a secondary one here, but it's really interesting to hear that more than a quarter of young people essentially believe their job or someone else's has been lost due to AI. What did you take away from this?

55:58 · Yeah, there was a couple of data points on like Pew's obviously a highly respected uh research firm and like this data is probably as good as any you're going to see around these topics.

56:08 · There's two things I thought was a little odd though. One is you said the 55% say they're more concerned than excited about increased use of AI in daily life. That is up 31% from 2021 when no one knew what the hell AI was very early.

56:23 · So 2021 we obviously didn't have chat GBT yet and the average person that would be answering these questions would have no idea what the hell anybody was talking about if they asked them using AI daily in your daily life.

56:35 · Um so we're actually talking about 24 percentage point change over 5 years which is I would expect to be higher but whatever. And then the one about okay 71% of adults think AI will lead to fewer jobs in the US. I totally believe that data point but they qualified it with over the next two decades.

56:53 · Yeah good two decades predicting anything on that.

56:56 · That's that's an absurd timeline to even ask someone about like if they asked over the next two years then that data point carries like way more weight with me. Um but my guess and this is totally like a semi-educated guess is I think the result would be real similar. Like if you said over the next years over the my guess is you're going to get a very similar percentage. Um, but I did think it's interesting. 5% say I will lead to more jobs. Like, so what this leads me to, Mike, is more about like the perception of what matters to voters.

57:29 · And if again from a political perspective either side, which I assume would be the Democrats at this point since they're on the offensive, um if they think that the jobs narrative will carry as much weight or impact as the data center narrative, you could quickly see this kind of data starting to play a much more prominent role in the 2026 midterms. Right now, data centers is working.

57:58 · So the Democrats are just going to hammer the data centers, but you could see jobs being it. And if it's not in the 2026 midterms, like a, you know, top three to five, it's going to be by 2028 like the topic. Um, so yeah, just it's interesting to keep watching these polls and as they evolve, I just those two oddities about the research sort of stuck out to me as noteworthy.

58:22 · Yeah, interestingly enough, uh it's not young people only, but we did ask in our state of AI for business report about how people feel about the net effect of AI on jobs over the next three years. And funnily enough, it's the exact same number. It's 71% are pessimistic. Uh they think more jobs will be eliminated rather than created by AI, which is and we were over the next two years. Is that what we qualified it as?

58:47 · We did over next three years.

58:48 · Next three years. Yeah. So that's kind of what I assumed. Yeah.

58:52 · Yeah.

58:52 · All right, so next up, somewhat related to this topic here. This past week, former presidential candidate and the founder of the Forward Party, Andrew Yang, renewed his call for direct payments to Americans funded by the AI and data economy. In an interview with Scripps News, he proposed sending families $15,000 a year as compensation for the personal data that trains AI systems. He said, "These AI models are built upon our data, and our data is being sold and resold for about $300 billion a year right now."

Andrew Yang Pushes for Direct Payments to Americans, Paid for by AI

59:23 · He also pointed to the coming wave of AI public offerings, saying, "These AI companies are about to IPO to a trillion dollar or more valuation, and we're not seeing a dime of it, unless you're an insider."

59:37 · In a separate CNBC appearance this past week, Yang argued the government should tax AI instead of payroll since companies skip payroll taxes and healthcare costs when they choose AI over new hires. He said that direct revenue should go to workers as direct checks. He cited uh Daario who has floated a 3% tax on AI revenue in the past and this is basically reviving kind of back in 2020 he made universal basic income UBI um a a really central plank of his presidential campaign. Paul is actually an interesting connection.

1:00:08 · I wonder if we'll hear more about this at Mon because Andrew Yang is actually speaking on the main stage. So I'm I'm just curious if you think like there are legs to the policy of kind of paying people outright. I mean, it's been a fringe policy for a while, but seems like more people are talking about it.

1:00:26 · Yeah.

1:00:26 · It goes beyond the UBI concept.

1:00:28 · It's beyond just like sending people a couple thousand dollars a year because they they can't get jobs. This is more like making it tied to specifically like the training, you know, the training data that's powering this. them. So, yeah, I mean, the reason we pursued getting Andrew Yang to to speak at Meon was because he's been talking about this for years and actually putting out real ideas, which was what was lacking from the labs themselves. It was like, let's test ideas, put hypotheses out into the world. So, yeah, I couldn't be more excited. He's doing the closing keynote on October 14th. So, the human- centered economy is the title of his talk.

1:01:02 · Building a future that works for everyone. So, I'll be fascinated to hear what he has to say. I I have no idea what he's going to say, but I think it'll be an amazing talk and I think it'll open a lot of people's minds at Monon about where this could go and some of the, you know, policy ideas that, you know, may end up becoming part of the the actual political landscape as we move forward in the next couple years.

1:01:24 · Yeah, no kidding. depending on how power shifts, you could see him uh whether it's another presidential run or maybe a cabinet position or something like that happening or maybe one of the labs snaps him up as a good person to actually be talking about this be interesting.

OpenAI and Anthropic Rethink Data Retention

1:01:40 · Okay, next up. This past week, OpenAI and Anthropic each announced changes to how they handle customer data from their most powerful AI models. Uh OpenAI said it will continue offering what it calls zero data retention to eligible API customers using its most advanced models. That means the company does not keep a customer's prompts or model outputs after a request is processed.

1:02:00 · Employees cannot review that content and enterprise data is not used for training unless the comp customer opts in. Now to keep that promise while still catching abuse, OpenAI actually previewed here a new safety system called private safety processing. It uses automated analysis to detect patterns of potential misuse across related interactions. When it flags something, OpenAI receives only a limited signal describing the category of risky activity, not the actual content. Testing is underway with early customers.

1:02:31 · Rollout is starting in September. Anthropic, meanwhile, plans to change a policy it announced in June that requires enterprise customers using its most capable cloud models to have their data retained for 30 days. This is a rule the company said it helped guard against cyber attacks while using its tech. This I believe came to prominence when Fable came out. This is like a interesting policy change they made with that model. Now under this new system which they are going to they expect to happen later this year.

1:03:00 · The 30-day requirement stays but business customers get the option to store that data on their own cloud infrastructure instead of Anthropics. Bloomberg reports anthropic developed this approach with more than 800 customers. So yeah Paul interesting changes here. obviously like a little in the weeds on the technical stuff, but I think is I'm just curious like is this going to help business leaders have more confidence their data is safe with these labs? I know especially the anthropic policy I've heard from a couple people in it and um technology functions.

1:03:32 · That's like a huge limiter that stops them from using Fable 5 specifically.

1:03:38 · I think it's a tricky balance for the labs. you know, there of course there's going to be people who think that they're just trying to keep the data and train on the data and like it's only for the benefit of the labs.

1:03:49 · I I believe there is validity to the argument that it it it's needed from a safety and security perspective. Um, but I also know that the labs realize that the threat of open weight models is very real and a big part of the argument for why companies would use openweight models is so that no one company has their data.

1:04:10 · So yeah, again it's like a lot of these topics, Mike, are so nuanced and there is no, you know, you know, what is and is not the right answer. it's it's kind of this middle where there's trade-offs being made and so the labs understand there's the competitive environment and they have to respond to it and you know meet enterprises where they they are and make sure that they adapt their policy.

1:04:32 · So I yeah I I would imagine at bigger enterprises this is a very common conversation that's happening and so these are the sorts of things they're going to need to see from AI labs to be able to continue to invest in the proprietary closed models.

Chinese Robots Break Human World Records

1:04:48 · All right, next up this past week, the World Humanoid Robot Games opened in Beijing. This is an Olympic style competition featuring more than 2,000 humanoid robots from 16 countries, including the US, Germany, and Japan, competing across 51 events like sprinting, table tennis, soccer, weightlifting, and tugofwar. On opening day, robots broke records set by humans.

1:05:12 · A humanoid from the Beijing based company XH humanoid ran the 100 meter sprint in 9.39 seconds, beating the 9.58 second world record Jamaican spinner Usain Bolt, uh, a record he set in 2009.

1:05:27 · Another robot from exhumanoid cleared 2.88 m in a standing high jump that surpassed the human high jump record of 2.45 m set in 1993. It was a massive leap from the 0.95 m the best humanoid managed at last year's first edition of the game. So, the Associated Press reports these games are kind of this spectacle that's demonstrating China's rapid progress in advanced robotics as its tech race with the US heats up.

1:05:55 · China makes the majority of the world's humanoids. And the event opened the same week that Beijing opened the 2020 hosted 2026 World Robot Conference where companies showed off about 3,000 products. This comes as in the US the FCC banned imports of new foreignmade humanoid robots on national security grounds. That's a move that was targeting China. Pentagon recently the Pentagon recently added leading Chinese robot maker unit to the list of companies it says have ties to the Chinese military.

1:06:23 · So Paul, this is certainly one of the more sci-fi segments we've probably done. It's kind of wild. This got a lot of buzz over the weekend. I'm just curious like did you take this away as like mostly just really cool theater? Like did it signify anything about Chinese robotics or robotics in general or just kind of like a a wild thing going on?

1:06:44 · Uh yeah man, I don't know. Like the clips were crazy.

1:06:49 · Yeah. Yeah.

1:06:50 · Um it was all over act. So again, I I don't know like how the algorithm works, but like this was like every other tweet I was seeing was like clips of this. And the one in particular was the one where the robot like beats the world record in the 100 meters or whatever and then promptly runs into a wall cuz it doesn't know how to stop and then like snaps at its waist and sparks fly everywhere and there's like a photographer looks like he's going to throw up. Like it was it's wild. Like it totally looks like something that was AI generated sci-fi kind of stuff. Um I don't know.

1:07:20 · I mean I I think that there's a very real concern that China is way ahead of the US when it comes to robotics. If you look at the breakdown of like the robotics supply chain, it is dominated by Chinese suppliers. So I know that the US government's concerned about this like you know most of the best demos in AI robotics in the US are um of dexterity and robots taking human jobs.

1:07:47 · It's like you you don't see this kind of display of more fun, more kind of like approachable stuff that isn't like threatening to take our jobs. It's just like, you know, cool things where these robots are fighting. I think there's like boxing matches and jumping and all these things. I I think there's like a pickle ball or tennis one.

1:08:08 · So, I don't know. It's fascinating to watch. Um, but I do think there's very real like geopolitical undertones behind this where China is sort of showing off how far ahead it probably is on robots.

1:08:21 · And, you know, I do think that by the turn of the c by the turn of the decade, humanoid robots are going to be a very real thing economically um and and in our lives. Uh, I've I've said before like I think probably by the end of the decade it it won't be unheard of that you'd be able to like lease a robot for your house for a couple hundred dollars a month like you would a car. Like you're just going to go get a robot and it'll do tasks around the house. I I think that's probably a real thing, but in the next four to five years.

1:08:50 · So, um, yeah, right now it's an oddity and kind of a a curiosity, but very quickly I think it's going to become real and you're going to see them walking around Walmart stocking shelves and you're going to know friends who have a robot in their house and it's just going to it's going to feel really weird the first few times and then you'll just get used to it again.

1:09:11 · I I look forward to coming back to this prediction 5 years from now or so cuz I think we're going to be directionally right and it's going to be weird. Yeah, I think it'll be sooner than that within pockets like you're going to, you know, people who can pay the hundred thousand for the first offtheline robots, but

1:09:28 · spreading around society, I think, you know, it would be more common place that by the end of the decade, but I would say within probably two to three years, you'll probably have the first signs of like 60 Minutes episodes following people around who have robots in their homes and things like that. likely a Whimo situation where every time we travel to San Francisco, we'll see him running around before anywhere else.

1:09:49 · Yeah. If you haven't been in San Francisco, you don't know that there's driverless cars in the world and has been like everywhere around there now. Yeah.

AI Use Case Spotlight

1:09:58 · All right. So, next up, we have our AI use case spotlight where each week we give you a quick look under the hood at some real AI use cases we are exploring at Smarter X. So Paul, I've got one to share of some work we've been doing and I'll let you take it away with anything you've got this week. Cool.

1:10:13 · So one thing we have started to evolve in our content operations is that AI can now basically produce an unlimited amount of pretty decent content. But what it can't produce and what is inherently thus valuable is the perspective and the knowledge and the insight that your experts internally have. So we actually think you know original quotes, firsthand experiences, strong opinions, insights earned through actual work are going to become even more important as AI generated content becomes easier to create.

1:10:41 · So our content strategy is certainly evolving to be much more like human and expert first in terms of like what is the original seed content. So, we do things like reuse what we're saying here on the podcast, in courses, presentations, in interviews, and we're also trying to do more Q&As with people across our team to get their knowledge and perspective into the content we produce because AI can't do that for us.

1:11:04 · So, I mention all that because this is what we're building AI this past week or two to solve this problem because the problem is this process for us is still largely pretty manual and inefficient. So like scheduling interviews with people takes time. An expert may not be available when the content team needs them.

1:11:23 · Whereas like most of the time right now as we're since we're starting small the expert is me. Like hey let's get Mike's perspective on a few different topics to really enrich some posts we want to do. So what we've actually started doing is building I guess what we tenatively call probably needs a better name an ondemand content interviewer agent in chat GPT.

1:11:43 · So basically like you can imagine it this way. Say our content team has five posts planned for the coming week and they say okay three of these like we would love to get Mike to do a quick Q&A to tell us some interesting stuff about these topics or strategies or whatever.

1:11:58 · So the agent will actually interview a member of our content team to learn more about each post like what is the audience scope etc. The agent, in an ideal world, we're still setting this part up, will then email me to schedule an interview by sending me a link that lets me start interviews when I have the time. So, I don't have to schedule a bunch of separate meetings, find time on someone's calendar, and then before interviewing me, and this is the part we've actually set up, the agent researches the assignment, the subject, the relevant things that I've said in recent podcast episodes.

1:12:25 · It can then run several research tasks in parallel and use that context to ask really smart questions. And then what it does is it interviews me adaptively one question at a time. So each answer determines what question comes next. It can kind of like dive into more interesting threads, skip something I've already covered, ask me to clarify, etc. Now when the interview is done, and this part exists today, the agent also creates this complete brief for the content team, which is really cool.

1:12:53 · It includes main takeaways from my interview, pull quotes, my full transcript, relevant source context. It also fact checks some of the things I say because I don't know everything. So, we got to make sure I'm saying the right thing. So, that's all working today. Um, the agent conducts the interview. It produces a really useful brief. We're starting to kind of test this out this week. Um, the remaining work here, which we're still working, is like automating those handoffs. Like right now, it's like manually jumping saying, "Okay, here's what we're writing about. Do the interview and create the brief for me."

1:13:25 · Um, but we want to eventually have that pretty automated. like someone would tag the post in our project management system as like, "Hey, needs an interview from Mike." Or maybe they get an email where they respond to that and say, "Hey, here's three posts we're working on this week. Go get Mike scheduled for the interview." We're working on that part. But this is pretty cool just because it's like not something to like generate content.

1:13:47 · It's just to inject more original human perspective into more of our content. and super helpful for me to then do these interviews when I'm on like a walk or commuting in the morning or like in these pockets of time that don't require me to like schedule something with another person. So, kind of a cool thing.

1:14:06 · Fascinating. See, this is great. Like I I learned things that our own company's doing on the podcast.

1:14:10 · This came together very quickly. So, yeah, this is as we're kind of exploring more agentic workflows. This seemed like a a narrowly focused one that we could get started relatively quickly. So now if I get an an email from an agent asking me for an interview in the next couple weeks, I'll know where it came from.

1:14:28 · Well, yeah. Yeah, don't worry. You'll have you'll have full communications to roll this thing out once that is actually possible.

1:14:36 · That's cool. All right. Um do a few quick ones. So one, Google Lens, and this is my maybe an overlooked existing tech that, you know, I sometimes honestly forget it exists, but in the Google app on my phone, I don't know how else you can get it. I I would imagine you can probably use it in Chrome, but on the Google app on my phone, there's a lens option, a camera option, and it'll like take a picture of anything you're looking at, and then it will tell you about that item. So, I was going through some estate stuff, and there was like some family heirlooms, and I was trying to figure like, what is this thing?

1:15:06 · Where did it come from? That kind of stuff. And so, you just pull up the app and and open lens, take the picture, and it immediately says, "Hey, this thing's from Taiwan at this time period, or this came from Germany." And and then you say, "What's the value of and it looks up and finds the last listings, the most recent sales on eBay, stuff like that.

1:15:22 · So, super valuable like computer vision technology. I don't know when Google Lens came out, but I think it's been out for a few years now. Um, that I just don't think to use that often. Another one I started experimenting with Gemini Spark. So, I have it in my personal Gemini app. It is not in our business app, but Spark is kind of an effort to enable like a capabilities and automations right within Gemini.

1:15:43 · And so I had a great use case which is my daughter started high school today and the high school she's going to sends lots of emails um from lots of different people for all kinds of important things that I lose track of all the time. And so I was like this is the perfect use case to try Spark to start with.

1:16:01 · And I said go in uh find all the emails from her high school summarize them for me every Sunday night and highlight for me any actions for the coming week any like time-sensitive things. and if there's something like super important like ping me when it happens but otherwise I lose track of the emails every day. So I set that up and I was like that was fun. So I I went through and set up a few other things where you just like tell it what you want and it builds the agent capability and then it just shows up in your Gemini app. So it's cool.

1:16:29 · But then my favorite one was I've alluded to this a little bit and I'm not being vague on purpose. I hate vague posting. I like it's one of my pet peeves of X is like these vague posts.

1:16:40 · Um, I am not doing it intentionally just to like annoy people with vague posting, but there's a project I've been working on for a very long time and it has arrived at the stage where I can now start building prototypes um, and working with developers to bring this thing to life hopefully to be able to share with you all very soon. And so I spent Thursday and Friday and then a good portion of my weekend building prototypes in clawed code of like functioning apps. I will just say like as someone who has orchestrated the building of apps and and sites before.

1:17:14 · So back in my agency days, I I led the building of technology and different software products that we created um through the agency. This is a whole different level like the the capability and I was using Fable 5. So I paid extra credits because I was like I'm just going to see this through with the most powerful model they've got. And so I paid the extra money um to get credits for Fable 5.

1:17:37 · Mike, I I mean, you saw like screenshots of it. I haven't showed you the actual, but it is a working full prototype app that is so far beyond like anything that I I would have been able to build with a developer years ago through months of like iterations. And I just like I think I put it on LinkedIn. It It's just magic to me still. Like I've known these capabilities exist.

1:17:59 · I've done some things with these things, but to be able to go in and build a fully functioning prototype of a product that should take probably more than $100,000 to build, I I just like can't wrap my brain around that we live in a reality where you can do these kinds of things.

1:18:16 · Yeah.

1:18:17 · So, super super cool. And and when I'm able to, I I'll like I I'll share the full story of how it all kind of came together. Um, we're just at the point now where I can kind of take these prototypes and hopefully get them into production. But the fact that someone like me with zero coding ability, zero design ability can orchestrate a fully functioning prototype is just mind-blowing still.

1:18:41 · That's awesome. That's so cool to hear.

1:18:42 · Can't wait to hear more about that.

AI Product and Funding Updates

1:18:45 · All right. So, to wrap up this week, we have our AI product and funding update roundup. So, I'm going to run through these. Got quite a few this week. and then we'll close out this week's episode. So, first up, a mystery model called Ox Alpha appeared free on Open Router with a 1 million token context window, image and video input, and a focus on coding and longunning agent work. Speculation so far is centered in it on it possibly being an unreleased model out of China or as Paul I think you mentioned maybe Google that the developer right now remains anonymous.

1:19:17 · This happens from time to time as these models kind of anonymously come out. So, we'll probably find out soon because this is generating quite a bit of buzz.

1:19:26 · Open AAI released an Apple messages plugin for chat GPT on Mac that lets you search your messages, catch up on conversations, and draft and send replies from inside chat GPT. This is available now in chat GPT work and codecs on desktop. OpenAI has launched AI futures, a new blog from its strategic futures team that examines how transformative AI could reshape power, governance, the economy, and individual freedom with an opening post by Dean Ball arguing that AI could let states project force and collect revenue without needing human cooperation.

1:19:59 · OpenAI also announced grants to 14 independent policy research projects across the US, EU, Brazil, Singapore, and South Korea, splitting $1 million in funding, plus up to a million dollars in model credits among groups including the American Enterprise Institute, the Progressive Policy Institute, and others to study how AI can broaden economic opportunity and societies can build resilience amidst the changes AI is going to bring.

1:20:25 · Anthropic published research showing clawed designed working protein binders, the small proteins that stick to a target protein and form the basis of many drugs against 14 of 15 targets with hit rates as high as 35.1% compared with a typical 10 to 15%. And all of this was validated in physical lab tests run with a couple of biotech companies.

1:20:50 · Anthropic launched Claude Academy as well, a free learning platform atademy.cloud.com with courses, tutorials, and industry use cases aimed at teaching individuals and organizations how to actually work with AI. Nvidia is paying about $6 billion to license AI startup poolides, model factory technology, and hire 109 of its staff.

1:21:12 · This is a deal the Wall Street Journal reports is aimed at building a powerful openweight alternative in the US to Chinese models like DeepSync, which we talked about briefly during the Nvidia segment.

1:21:23 · Real quick note on this one, Mike, we'll talk about this maybe next week, but it also came out this morning that Nvidia made a play for Perplexity to do a similar deal, like a Aqua Hire type play. Now it sounds like they're maybe just going to like license some Perplexity technology which I need to dig into cuz like I don't know what the hell they would license from them but I think it was valuing perplexity like 30 billion or something. So we'll touch on that next week.

1:21:47 · XAI has launched Grockbot in beta. This is a product that lets you create always on AI agents that you message like a co-orker. Each runs on its own computer in the cloud, so it can sign into your existing tools, work through multi-step jobs unsupervised, and learn a workflow by watching you do it once. Access is bundled into higher tier paid super Grock and cursor plans.

1:22:07 · Bloomberg's Mark German reported new details on Apple's camera equipped AirPods, which use small cameras to analyze your surroundings and feed Siri visual information rather than take pictures. though his own timing on this has moved from as soon as it possibly coming out in this December for a lower-end version to 2027 for the premium model. So, we'll see when that actually comes out.

1:22:32 · And I think the Apple event, if I'm not mistaken, I could look this up real quick, but I believe it's September 9th is when we're supposed to get like the foldable iPhone and the new stuff. So, we may get a preview of the AirPods with cameras then, but we'll we'll we'll know for sure in early September.

1:22:49 · Apple Music is going to start showing made with AI labels on tracks later this year. They're going to require record labels and distributors to tag any song where a material portion of the content was created with generative AI. Uh, a company called Etched, which builds chips designed specifically to run AI models instead of general purpose GPUs, raised $700 million at a $21 billion valuation from firms like Jane Street, Kleiner Perkins, Sequoia, Andre Horowitz, Peter Teal, Bane Capital Ventures, and Blackstone.

1:23:20 · Whisper Flow, the dictation app we've talked about before that lets you talk instead of type across any app on your computer or phone, raised a $280 million series B at a $2 billion valuation led by Menllo Ventures. They previewed their first in-house speech model, a 2 billion parameter model called Kanto that is built for real noisy realworld conditions.

1:23:42 · You're a big Whisper Flow user, right Mike?

1:23:45 · I am. Yeah. It'll be interesting to see what their trajectory looks like. I'm curious to dig more into that one. Higsfield, an AI video startup founded by a former Snap executive. They turned text prompts into marketing videos, raised 400 million at a $5.4 billion valuation from investors including Goldman Sachs and others. Uh that's roughly four times the $1.3 billion valuation they carried in January.

1:24:10 · Harvey, the legal AI company, launched its own model called tenant, built by post-training the openweights Kimmy K3 model on public legal data, synthetic data, and expert data. Harvey says this lifted its all pass rate on a popular legal benchmark, 82% over the base model. We're running at less than a fourth of the cost of leading foundation models. And last, real quick, that one's a big deal.

1:24:40 · I would go spend some time on that when we want to dig too much into details because it's pretty technical on the show, but it's a prelude to what other AI native companies are going to do. So, I I think it's the it's the most detailed explanation I've seen yet of a company training their own open weights model to do a very specific industry.

1:25:00 · And I I think it could be a sign of of things to come. So if you if you're curious on that, go pull that thread and and read up on the technical side of how they did it.

1:25:11 · And last but not least, our friends Adam Broman and Andy Sack have published a book they call a living book called Agents Inc. It's a 16chapter book on agentic business transformation that was researched and assembled partly by AI agents. They actually credit an AI co-author and the whole thing is shipped with a website where readers chat with that agent about the book instead of just reading it on their own.

1:25:36 · And I did actually request the custom version. So you Vera, I think is the agent's name. You can request a custom version. You tell a little about yourself and then it sends you a PDF version and then I also just got the hard coverver copy on Friday. So I haven't dug into it yet, but I did have a pre-eread on it. Um because I I was actually interviewed as part of that book. So Adam and Andy are good friends and um great thought leaders in this space. So yeah, I'm fascinated to get into the the whole book and and see everything.

1:26:03 · Yeah, it's a cool approach.

1:26:06 · All right, so one final reminder, our new AI pulse survey is live. Go to smarterx.ai/pulse. It's just one question over this week or the next couple weeks. We'll probably be promoting this across a couple different channels. We would love your feedback.

1:26:19 · This one is about your hiring plans, especially with entrylevel workers and how your company is treating those roles. So, we'd love to hear from you on that. Uh Paul, thanks for breaking down a crazy compelling week in AI as always.

1:26:34 · Not as much soapa drama this week, so that's good.

1:26:37 · That's okay. I'm fine with that.

1:26:38 · Fascinating. And I always like I always say like these uh the product and funding updates, man, like we could just do an episode on that stuff alone, like expanding on that. Like so many of those are very noteworthy and we just, you know, touch on them. But if any of them sound intriguing to you, you can always check the show notes. Go to podcast.smarterx.ai and we put the show notes up every week and all the links are in there so you can go and pull through those.

1:27:02 · All right, Mike. Thanks a lot. Thanks everyone for listening. Have a great week. Thanks for listening to the Artificial Intelligence Show. Visit smarterx.ai AI to continue on your AI learning journey. And join more than 100,000 professionals and business leaders who have subscribed to our weekly newsletters, downloaded AI blueprints, attended virtual and in-person events, taken online AI courses, and earned professional certificates from our AI academy, and engaged in the Smarter X Slack community. Until next time, stay curious and explore AI.