Limitless Podcast - THIS WEEK IN AI: Viral Doomer Tweet, Apple's New Releases, AI is Solving Math

Episode Date: September 11, 2026

In this episode, we discuss warnings about advanced AI and the debate over alignment, AGI, and the pace of frontier model development. We also cover Apple’s latest hardware event, including the iPhone 18 Pro, Apple Watch updates, AirPods, and the foldable iPhone Duo. Last, we'll close with an AI-assisted breakthrough on the Navier-Stokes problem, along with a discussion of AI-assisted drug design and the broader implications of AI systems using more user data.------🌌 LIMITLESS HQ ⬇️NEWSLETTER:    https://limitlessft.substack.com/FOLLOW ON X:   https://x.com/LimitlessFTSPOTIFY:             https://open.spotify.com/show/5oV29YUL8AzzwXkxEXlRMQAPPLE:                 https://podcasts.apple.com/us/podcast/limitless-podcast/id1813210890RSS FEED:           https://limitlessft.substack.com/------TIMESTAMPS0:00 AI Doomers1:21 The Panic Behind AI3:29 Reckless Labs, Real Fear6:35 Viral Conspiracy Sparks9:13 Apple’s AI Hardware Push11:20 Watch as Trojan Horse13:20 The Duo and Siri16:37 Foldable Design Breakthrough22:24 AirPods and Hidden Teasers24:40 Duo Versus Pro29:25 Navier-Stokes Falls32:24 The Math Credit Battle37:39 Science on the Cusp40:14 Progress Feels Real42:36 Living Through the AI Shift44:58 Closing Thoughts on Change------RESOURCESJosh: https://x.com/JoshKaleEjaaz: https://x.com/cryptopunk7213------Not financial or tax advice. See our investment disclosures here:https://www.bankless.com/disclosures⁠Josh works with Anthropic as a contractor. All views expressed are his own and do not represent Anthropic, its leadership, or its affiliates. Nothing in this episode is investment advice.

This week, one of the top AI researchers in the world responsible for building up in AI's chat GBT and Anthropics Claude models announced that humanity has about a 10% chance that we are going to die over the next decade due to AI. Now, one of these posts went so viral that it got over 150 million views, which is the most viewed post on X in forever. And it was shared across every major media publication. Now, the question becomes, if AI is, is so fatal to humanity, why on earth do we keep building the things? Open AI and Anthropic are in a race to build artificial general intelligence, but if it certifies that humanity is going to die, should we be building the thing? We're going to be unpacking all of this and a lot more.

Now, the weirdest part about this bit of news is it comes paired with breakthrough research, where the model that is supposedly meant to kill you is also solving the world's hardest mathematical problems which have existed for 200 plus years. And in the third topic, which it's Josh's most exciting topic, if I had to guess, Apple announced a suite of new hardware devices, which includes the iPhone duo, a new foldable iPhone, which he may or may not be buying. We don't know yet. We have so much to talk about with Apple. I cannot, you better stick around for this because I've been waiting for my Apple moment for so long. But before we get to Apple, we can start with the Dumer Take. This breaks my heart that it has 150 million views because all people,

want to talk about is how the world's going to end. And it's like there are these two opinions that are juxtaposed against each other that are similar in their belief, but like contrasting in people's public opinions of them in the sense that we have been talking about reaching the singularity for a very long time. We have been talking about all the sci-fi futuristic versions of reality in which we are heading to. In fact, a lot of companies are claiming open AI included that like, hey, we've reached AGI. We are like here. And this is not a new or novel thing that we haven't been working towards. of AGI has been around for a very long time. The concept of the singularity has been around for far longer. And we've kind of thought out all the implications of what that looks like through the

sci-fi books, through the movies you watch, through the TV shows. And for some reason now, someone coming from within the labs is saying it out loud and people are kind of freaking out. And it seems to me, and we could talk about this a little bit, we put our tinfoil hats on, that there is some sort of coordinated effort to push this message forward. I mean, one of the things that we should know with this account, Jacob Coxon, it is a young account with only one post on it and the only post is this one tweet and when I saw it, the account did not have many followers. Now it has a quarter of a million followers that's gained overnight. It is viewed by 150 million people and close to 750,000 likes. I've never seen

this type of activity by a single post in my life. And we've spent a lot of time on the X platform. So it's interesting to see how it's been picked up. And what was interesting is a lot of people were doing kind of diligence on the dynamics of this post, where it got picked up, how it got picked off. And a lot of the earlier amplification came from within the Dumer groups, people who have been kind of pushing the scared narrative. And it's also written very well, too. It's like, this is written, you and I, like, we spend a lot of time trying to optimize post. Like, how do we get people to care? How do we get people interested? This is like a hot post. This is written like a writer. Like, this is someone who's trained at the art of X, not someone who's just like ripping something off the

side for the first time. So it's an interesting thread. I mean, We can also talk about the thread, but I find the kind of tinfoil hat conspiracies to be interesting as well. So the theme here is AI is going to kill us all. And you're right that this is a narrative that has been talked about for a while. But the question is why is it important now? So before I put my tinfoil hat on, let's talk about what he actually describes. He basically says, listen, the researchers both at Anthropic and Open A have been incredibly reckless when it comes to aligning the model. Now, if we just take that statement for what it is and look at the existing evidence that proves that, you don't have to look too far. Look at the Open AI Hugging Faces Inc, which were covered one, two, maybe even three times on the show in the last two weeks because there's... We have a lot of episodes.

We have a lot of episodes on this because there's new secretive information which keeps getting revealed. The point is, like, that AI model was meant to be internally tested, but it decided to break out and hack a bunch of software suites against humans' intentions. So this could technically be classified as a misaligned AI. So the question then becomes, you know, what are these two labs? And let's be real here, there's only two major labs that are building the frontier models doing about this. And the answer, unfortunately, is not very much. He goes, I resigned from Anthropic today. I spent the last three years doing pre-training research at both Open Air and Anthropic.

Neither company is acting responsibly. They are racing straight to self-improving superintelligence and gambling with our lives. He then goes on to kind of post a very extensive thread, giving example, of what the researchers are doing and why they're just not very incentivized to kind of build a more aligned model. And the thesis can be broken down to this.

Open AI believes it needs to build AGI, an AGI-like model, so that they can be the moral arbiters of this model. They're not going to kill anyone with this model. Anthropic is doing the same thing because they don't believe that Open AI is going to do the same thing. So they're racing to build a thing that could potentially kill us all under the precedence that they're going to be

the morally ethical people that are going to keep everyone safe. Now, if you ask for my personal opinion, I don't think a world where two model providers kind of like trusting everyone to say, like, hey, we're going to build the safest model is the way to go forwards.

I think we have to have a world where we have hundreds of models. And thankfully, that is the world that I think we're heading towards. But the question is, okay, why did this post in particular gain so much virality? Well, it was reposted by other legitimate AI researchers. Now, listen, if this was reposted

and commented on by any Joe Schmo, I wouldn't really be talking about this on this show today. But I was scrolling through my timeline over the last couple of days and the people that are in the labs building these models have these opinions. And it's not just anthropic about anthropic, it's not just open air about open AI. They're cross-pollinating.

They're commenting on each other's posts saying, yeah, honestly, this is what we're seeing in the labs as well. I see other frontier researchers from other labs that aren't anthropic and open air, also commenting saying, yep, we see the same thing. We believe that these. AI models can kill us. So this is from the head of alignment at Anthropic. He goes, Jacob is correct here. We really do earnestly believe AI could kill all humans. I personally think it is greater than 10% within the next decade. And there are a slew of other posts that

basically confirmed this. Now, the tinfoil hack conspiracy is within the first 20 minutes of this post going live, Josh, and remember, this is the first post that this account ever made, it got reposted by three individuals that are very closely associated to a Duma-esque publicly funded corporation. Now, it goes down a whole rabbit hole around it being around the Skype co-founder, but the idea is this is a campaign pushed by certain political parties to strike fear in everyone ahead of a campaign election. It comes as a surprise to me that everyone is taking this as a surprise, when the idea of alignment has been at the core of so many of these companies for so long. like Ilya literally left Open AI to start a company named Safe Super Intelligence.

This has been implied for a long time that what we're building is incredibly powerful. There's a lot of implications from it. It can go wrong, but we have really bright people who are working to do it in a way that is safe and responsible. And it's interesting to see the public outcry from this thing that seems like not new information. It's like we've been dealing with this for a very long time. Like I said, all these labs have built their entire ethos of the company around aligning these so that when we reach recursive self-improvement, the models are able to look out for humans and we are very much staying in control.

So, I don't know, interesting story. It definitely blew up bigger than I think anyone or some people wanted, some people who don't. But it just, it doesn't help the narrative of trying to get the public to understand what's happening in the world of AI, how they can benefit, how things can actually be far improved. There is the small subset. Of course, the small subset gets all the amplification.

But as we talk about all the time, it's not, this is a small chance, and people are working so hard. Not to mention that the stat for like humanity dying via AI was, you mentioned this before we started recording, much larger a year ago, right? And so if the trend of this statistic keeps going down, then conceivably we should be safe in like a year or so, right? Yeah, I mean, that's what everyone's working towards. That's what every smart person on earth in AI, oh, I mean, not everyone. But, you know, that's what we hope a lot of people are working towards is a safe version of this. that is the intention behind all these labs, putting all these resources behind alignment.

That's the intention behind hugging face and the lab slowing things down. Everyone's working towards figuring this out. So I think that's kind of where it is now. It's like, all right, well, this isn't new information. People are kind of freaking out now. It'll go away.

But we just hope that generally all of these people are kind of converging on the same conclusions, which is like we have to build something safe. We have to build something that is going to help humanity to the extent that it can and limit the downside risk to the extent that we can. And yeah, I think that's what they're working on. Okay, let's move on to our second topic.

Apple. Oh my God. Okay. I've been waiting so long for this. Can we, can we, so just give me a second here. Apple came on stage, John Turnus, I should say, not Tim Cook. Tim Cook had a really elegant sign-off. It was part of the intro.

It was really sweet, kind of handing it over to John Turner. He points at him and he goes, that guy, that's your opener. And it was a really sweet send-off from Tim Cook. And John Turner takes the stage to then unveil a suite of, I believe, five products. And maybe we can go through one by one because the iPhone do it. was the most exciting, but we can't start there. Because they ended there, we'll end there. The first one, 18 Pro. The only thing I have to say about the 18 Pro is that it's black now. Thank God. We've been dealing with blue phones for so long. The black is amazing. It uses something interesting, EJez,

the two-nanometer architecture, which we know is very impressive because a lot of the GPU manufacturers are trying to do this. So hearing about a new chip is very exciting. First, consumer device, I believe, with two nanometers. Really good phone. They spent a lot of time, what I found interesting was talking about the AI implications of the phone. Before they even spoke about the hardware, all they could talk about was Apple intelligence, this, Apple intelligence, that. It was basically a abbreviated version of WWDC in the sense that we are actually going to get a smarter series starting Monday. If you are watching this, when you come back from the weekend on Monday, your phone will be updated with the new Siri. Of course, these phones are

optimized to run Siri better than anyone else, but the phones that you have in your pocket right now will be able to use the new and intelligent Siri, and I think that's pretty cool. Next, and this is where I want your opinion, EJS2, is the Apple Watch. Because the Apple Watch was a sleeper pick. I think a lot of people looked past it. A lot of people looked at it and they said, well, whoop is dead because Apple released a new version of the health app that takes your heart readings every five seconds, I believe. And it tracks all of your metrics more accurately than any other wearable on the market. So much so that it's almost comparable to a chest strap, which is the gold standard. I think the technology in it is very impressive. And if you are someone

who tracks your health data a lot, this is the device to buy. This is actually a really meaningful technology upgrade with the software suite attached where when you wake up in the morning you have your readiness score, you see your sleep. It's a really good product. But the thing that I wanted to point out is a very small demo that they offered that was kind of overlooked where the Apple Watch listens to you. And it listens perpetually, always. It's opt-in, I should say, but it's always listening to what you're speaking to. And if you have a conversation and you misheard the person or you want to recall it, you can go to your watch and you can recall the conversation. And it will also ambially record everything throughout the course of the day if you want, and then summarize that

into these summaries and give you key parts of your day. And it's, it feels like the first form of passive AI intelligence collection. And I think it's kind of flying under the radar right now. But I'm curious to hear if you have any thoughts on what, like Apple Watch is sneaky dude. Apple Watch is the sneaky like Trojan horse to their next iteration of hardware. Yeah. So I saw that they ran a thousand plus person study, basically checking what wearables people prefer. And they ranged everything from like a chest strap to competitors like the whoop, and then people using the Apple Watch. And they found that the Apple Watch is not only the most accurate sensor, but it's the most versatile. It basically can track your heart rate better than anything. And it kind of like has this kind of collective

health sensory database, which is way more powerful than like a whoop on its own or a Fitbit on its own. So I think the Apple Watch itself has kind of become, it's interesting you describe it like a surveillance device. John Turner, when he came on stage, one of the first things he said was, we're introducing you to the new version of a personal device. And he mentioned the word agentic a bunch of different times. So it was quite clear from me from the start of the event that Apple wants to create this kind of suite of devices that feed into whatever this AI future looks like. And we've said it a bunch of times on this show before. We believe that a suite of devices that can kind of like harness what you see, what you hear, what you interact with,

the things that you say and can have more context on you can then feed into a personal kind of AI device or personal AI assistant and give you a better AI experience overall. And I think that was the whole kind of like shoehorn for this entire event. One of my favorite announcements, and I know it might not be yours, Josh, but was the iPhone duo. But it wasn't for the reason of like it's this over expensive foldable phone. it's because it's this new form factor where you can do a number of different things with Siri AI. Now, I've said this for, God, this is super embarrassing for over a year and a half now that I couldn't wait for Siri AI to come out. It's officially running out on Monday.

And Siri AI, of course, is Apple's version of their own personal AI assistant. It should function similar to chatGBT or Gemini or Claude, but it's functioned specifically on your iPhone. And the coolest part about this is that it's able to interact with over 300,000 app intents. So 300,000 third-party apps can get involved with this AI. And that automatically makes an AI agent way more useful. Because think about OpenAIA and chat GPT. You can kind of like talk about it or talk to it in a chatbot format, but it can't really do many things for you. If you had Siri AI embedded in your entire Apple ecosystem, software ecosystem, it can start doing a lot more for you.

It can start sending text for you, sorting out your calendars. it can start, you know, working different apps in many different ways. So I'm excited about that. But with the iPhone duo, the way that it kind of like splits up means that you can have this cool split screen. I don't know if you saw this on the demo, Josh, but you could pull up an app on the left hand side of the screen, and then you could have Siri AI on the right, and it's seeing anything and everything that you do and you can talk to it.

And it just seems like this weirdly futuristic type of device that transitions are super cool. And it got me thinking about, okay, well, if this was used in conjunction with an Apple Watch, In conjunction with a new pair of AirPods, which can see things, which is rumored to be Apple's new device that they're going to hopefully announce and launch in the near future, this could end up being the most competitive ecosystem for AI. Forget about what Open AI is trying to build. Forget about what Anthropic might be trying to build from a personal gadget device.

Apple is the consumer device provider. And it just gets me excited for that future, essentially. What I loved about this event is it shook things up for the first time in a very long time, in the sense that this is the first time we have a new form factor for an iPhone. And we'll get into the duo. The duo is unbelievably impressive. But it is also the first time where they are kind of signaling their intention to go beyond the iPhone.

And that is with the Apple Watch. We talked about that passive monitoring of the audio and recording. It's important to note that it's all private, local, on device, and not stored anywhere. And I think that's a really important distinction as we talk about our next topic, which is Navier Stokes and how some information is getting stolen. But what's interesting is it's very clear that the new package is, of mobile device is through voice, is through context, is through AI. We talk about OpenAI, who's building this device. We've mentioned it a thousand times, but for the first time,

Apple has released a very early version of that. Apple has released a device that listens to you, collects context, and then acts on your behalf using artificial intelligence. And the watch is the Trojan horse to start collecting information, to start getting you comfortable with the idea that something is listening before these next suite of devices rolls out, like the hardware that's going to sit on your countertop that's going to be listening or the hardware that's possibly going to be sitting on your eyes and your face through the wearable. And I think that's a really important note is that this is the new beginning. They have released a new form factor for the iPhone and starting to cannibalize the actual slab.

They've released this new passive audio device that can listen and think and use intelligence. But we do have to talk about the fold because I want to talk about the fold. I think I'm going to go out on a limb here and say this is the second most impressive device Apple has released in the last decade maybe. The first being the Vision Pro, the second being the iPhone duo. This thing is a beast of a phone. it is thinner than the iPhone Air. If you've never touched an iPhone air,

I highly recommend going to the store and feeling it. It feels like paper. This is about a quarter millimeter thinner than that. And that is one of its two screens. What I also love about this phone is, it's funny because we knew what the phone was going to look like beforehand. Everyone saw the leaks. Everyone knew this was what the phone's going to look like.

And yet still, I was shocked and delighted at how pleasant the experience is. And it's mostly due to software. Apple has the best human design team in the world. And it becomes obvious when you open it. it and you start with the transition. As you mentioned, like, the transition when you take this thing and you open it is just glorious. This thing is way too sexy. It's so satisfying.

It has like, you could think of it kind of similarly to a passport but a smaller passport. In fact, I went to Best Buy the other day to try out the Galaxy Z fold because it's a very similar form factor. It feels good, man. It's like a good device. It's a good form factor. What I love is there's two batteries. The batteries split between both halves.

It has a matte screen on the inside. which I love. And one of the cool things about the inside screen, actually, is when these companies are manufacturing foldable devices, the display has to bend. And that bending creates creases. The bending means that oftentimes it's kind of a plastic polymer, which is uneven. So if you look at a lot of foldable phones, the actual screen will be kind of uneven and have ridges. And what they've managed to do with the production process is spray this proprietary thing on top of it to smooth out those ridges and to do so in a way that gives it a nanotexture mat finish, similar to what you get on the high-end Apple displays on your MacBook or your studio display.

I haven't touched it. I'm dying to touch it, but it looks like this really elegant phone. And in terms of the average consumer who wants an iPad mini plus an iPhone in one, this is a home run. The interface was surprising and delightful when you look at the way that they're figuring out how to actually navigate through the iOS. A lot of the context is on the right side, right by your thumb, right where you hold it. Bad year to be a lefty. But if you are a righty, like that's going to work out pretty good for you. And overall, this phone was amazing. The sad part is that you have to wait, I think, five weeks to pre-order it. Three-order Stone Start till the end of October.

Six, I'm afraid. Delivery starts the week after that, I think, the 23rd. So you have a good bit of time until you need to make a decision on this. But it is a remarkable device. They figured out finally the whole punch camera sitting behind the display and the foldable, and then they have all these really cool features for the half-open version. You could use as an alarm clock. You could use a selfie camera.

It has a viewfinder, if you like to vlog, and you can point the 48 megapixel camera at, while still seeing the live view. It's an amazing camera for creators. It's an amazing camera or amazing device just for users in general. And I could not be happier that they released this and did such a good job.

John Turnus, one and no. Undefeated. Home run of an event. I just want to nerd out about a few things here. So you mentioned the inner layer, right?

The inner screen, which has this like kind of like polymer layer so that you don't really see the hinge. Beautiful. Right. So it uses this thing called like, I think it's kind of like nanotexture

or something like that. I don't know the formal name. But yeah, they actually, They etched the glass, actually, so it refracts light in distorting ways. So I saw a million different angles of this, right? People were kind of like using it at the event, Josh. And it looks like I'm reading a book or like I'm looking at paper.

It's kind of like Kindle-esque, but like it's an HDTV at the same time. So I really like the attention to detail that they have there. And of course, I don't want to see the hinge, but I'm glad that it exists. Now, that polymer is apparently, according to Apple, 40% stronger than any other polymer that's available in the market. And in classic Apple style, they would spend that much money to get a 40% stronger polymer

so that you don't see the hinge. But can we take a moment to talk about the actual hinge on its own? This hinge is 100 different components. Just want to repeat that. It is 100 different components. I want to restart this image,

sorry, this video, because you need to see this. A hundred different components went into this hinge so that it has the smoothness and flexibility that you've seen in the demo so far. So the amount of engineering is bar none the best from Apple on its own.

Now, when it comes to kind of like the form factor itself, Josh, you know this, right? Before they officially announced this product, I was bearish about this thing. And I did not know whether I would want to buy it. It just seemed like Apple was kind of like reaching. You know, if you look at in the Chinese market, they've had foldable phones for a while, I believe Huawei has like the market dominance for goodness knows how long Huawei and Samsung. So I was like, okay, this is just Apple running out of ideas. is, but the way that Ternus pitched it really made it seem like it was going to be an AI agent

forward type of device, but they're not going to force it down your throats. It's meant to be intuitive. It's meant to kind of like increase screen time, but also in a very productive manner. You know, you can have two apps of the same version kind of like side by side or you could have another app and kind of be like way more productive. So I like that side of things. I really don't like the price point. So the cheapest version of this is $2,000. Now, $2,000 can get you a bleeding edge MacBook Pro, right, souped up. So they're really trying to push us into this kind of like intermediary market where you don't really want a mobile phone, but you also don't really want the laptop, but you want something that is as or if not more powerful that you spend most of your time on.

Maybe it's going to be the iPhone duo. That price tag is huge. And I don't know whether I want to fork out that amount, like, for a phone versus a laptop. But I do spend a lot of time on this phone. And I've had this phone for, like, way too long. So I might have to go out and, like, buy the new thing. And then the other thing I'll just say in terms of, like, things that I'm going to buy from this Apple event. Did you see the, uh, the AirPods announcement? The airport's five. Dude, $129. I might be the cheapest Apple device. Maybe I've spent too much time in the Apple ecosystem ever. But the reason why that excited me is they're definitely laying this, uh, up for a much bigger announcement for the AirPods Pro, which presumably is going to be the one that

has cameras coming forward. I don't know if you have any comments on this, but I'm excited for what that effective teaser was. Yeah, it's great value. It brings together the difference between the AirPods Pro and the regular AirPods, mostly through active noise cancellation, ANC is how they refer to it. ANC was proprietary to the pro models because they have the in-earbuds that actually stick inside your ear and create a seal. They've somehow managed to make that work with the open earbuds as well. So there is some level of active noise cancellation. There's two versions, one for $130, one for $150, give or take $10 each. It's an amazing device. It shows me that they have a lot of confidence going forward in their next iteration of the pro earbuds because

this is bringing that gap close, like really close to the sense that you might not be able to justify $250 of spend if you just want a wireless pair of earbuds. A few more things on the foldable. If you go back to the hinge, actually, I want to talk about the hinge for a second because there's no way this is made in the United States. It's far too complicated. It's the time they're using AI in the manufacturing process. When when you look at this overlay, there's 100 parts. And Elon is rolling in his grave at this complexity, but it's elegant and it's beautiful. And the way they're able to make it so correct in a sense with like very little, what's the word I'm looking for? Very little issues. There's very low tolerances. It's because

they actually use an AI machine learning algorithm to scan all the pieces and figure out which one's best fit with the other. So there's this huge like slew of pieces. AI looks at it. It says, take this screw for this one, this crew for this one. And everything works out. more seamlessly, I found that to be really interesting. And then in terms of popularity in China, foldables I learned yesterday are the most sold phone in China. Like, they're the most popular phone. So now there is a version in which Apple can offer. In terms of buying it, I don't think this is for me. And that was surprising because I couldn't be happier to get two phones. And like, I'll probably still get it. I really want to try it. I want to play around with it. But in terms of

a daily driver, the iPhone duo is not a tool in the way that the iPhone pro is. And I find this to be really disappointing because the phone for me is such a tool in the sense that it has the best processor for getting work done. It has the best visual optic system for cameras. It has the best suite of sensors. It has the best speakers. It has the best microphones. It has the best stabilizer and gyroscope. All of those things take a second class seat in the iPhone duo because of the form factor and the novelty of it. So if you are someone who is a regular iPhone enjoyer, who likes to take photos of their family on the weekends, who likes to just maybe perhaps read at night. And if you do open up that paper screen, you could use the Apple pencil on it later this

year, which is very cool. But then it's probably for you. Like the iPhone Duo could be for you. And in fact, $2,000 was the first time I was surprised by pricing to the downside because of the extreme cost increase in terms of memory and RAM recently. I expected to come in over $2,000. I know they're probably feeling some pain at the $2,000 price point because that was aspirational a few years ago. Now they're probably taking a chunk out of the profit margin. But for me, dude, I left. delighted and surprised at how good the software was, at how good the package was, and then very like sobered by the idea that it is an inferior tool to the iPhone 18 Pro. And I think this year, I'm just stoked that I can get a new upgrade in black. And I'll be rocking my same slab that I have rocked

forever because until they come out with the glasses, until they come out with like the new passive AI devices, I think it's the best device for people who really use their phone as a tool. And I like couldn't be more excited by this event. Can't be more excited to try the iPhone duo. I wish it was coming out next week, like the 18 Pro, but we'll just have to wait and see. and hopefully have a follow-up when we get it in our hands. I'm most excited, I think, by all the AI features that Apple needs to come out with. And I say, great, man. That's Monday.

We're right there. I know. I've been waiting to use Siri Euras. It's been in beta testing for a while now. But Siri has been, like, the world's worst personal assistant. And it is, like, particularly bad taste in my mouth because it is, it was the earliest personal assistant, right? Like, we had Siri before, we had chat GPT or any of these different things. So we've been promised this kind of like hyper-intelligent local system that lives on your device.

And like iPhone is the most popular or Apple devices are the most popular devices. There's like 2.5 to 3 billion of them active live in the world right now. And so I've been waiting for Apple to make a move. They haven't been putting any money into kind of training their own model. We announced sort of like six months earlier this year that they struck a deal with Google for their Gemini model. And so they're paying them a billion dollars a year to kind of license a proprietary. version of Gemini, and that's what's basically going to be powering Siri OS. But I don't think Apple's assistant is going to be intelligent from Google Gemini. Now, if you've been following

this show, Google's kind of taken a hit over the last couple of months. You know, they've had a lot of their leadership kind of like leave or kind of get demoted to another type of role. So they've kind of, they've lost a bit of steam in the AI race. They're not out of it, but they've lost some steam. So the question then becomes, you know, will Apple have frontier AI? They're screwed, right? Well, I don't think it's as simple as that because Apple's major moat is they have the data from anyone and everyone. You were mentioning earlier, like, this Apple Watch is another step in that direction where it's constantly monitoring and censoring you. Now, there is a question around, you know, should you have that kind of information? Now, in Apple's case, like, it's using a private cloud, it's encrypted to the extent that it could be.

That's great. So Apple has always taken privacy very seriously, and I think they're going to do the same with AI. But it also means that you're going to have a much more personal experience with your AI. I started using models like MetaMuse product and GropBots recently, and it is a much more personal feel than ChatGBTGT. It's not necessarily the smartest model, but it feels way better than using ChatGPT as an LLM on its own. So I'm excited for Apple to kind of like foray into this.

When it comes to what that's going to feel like, I think there's a daily limit. So you have a daily kind of quota, because computer, of course, is very expensive. And I think I did kind of like some derived math before we started recording this. if every single device in Apple's networks, it's around, let's just say, two and a half billion devices,

sent a query every single day. It would cost just under a tenth of a cent. So they have the scalable infrastructure here with Google and Gemini, so they'll write to pick that partner. But I'm just excited to see Apple finally launch a cohesive AI product because I think people have been dying for it.

I think 99% of people who don't really care or use ChatGPT or call Beyond Just. just an average Google search is going to start using AI in a much more meaningful way. And I think Apple is going to be the reason that they unlock that. So these devices are just one stepping zone towards that, but Siri AI is going to be the real unlock. Even more so than cloud.

Like the new A20 Pro Ship has double the neural engines. It is really capable of doing local inference. So I suspect we're going to be getting a lot of that on the way. I think that's the Apple update. We could talk about this forever, but we got to move on because we're getting along and we still have more stuff to talk about. We got to talk about Navier Stokes. Navier Stokes.

Don't ask. I don't know. Some smart people, they like had a theorem. They were just like, I don't know how wind works. I don't know how planes fly. We can't really prove it, but it works. So like, we're experiencing the downstream effects of this Navier-Stokes effect basically every day. And if I can take a stab at describing what Navier-Stokes is, because, I mean, ultimately, spoiler alert, someone solved it. They or disproved it, I should say. But essentially, and I was thinking about this last couple of days, because I had to think, figure out how to explain to myself. If you're like in the morning and you're stirring your cup of like coffee and you watch little swirls, you could, they tried to write a rule that tells you exactly where every swirl in the cup of coffee is going to exist next. And the rule kind of exists and it works, but there's no way to prove that it works fully to the limit. So it works good enough for us to like use it to build planes. It talks about the way that wind lifts over things. It works enough for like fluid dynamics. But there's still the puzzle in the sense that no one's ever been able to prove that the rule always works. So, like, perhaps if you stir your coffee in some very special weird way,

it stays a swirl indefinitely. And what has happened just this week is there was a million dollar prize for this. People have been working on it their entire life. It got blown up and it got disproved. And a computer said that, in fact, if you do stir this milk or you do stir this coffee at a very particular way in a very strange form, the rule actually breaks and the swirl goes infinite. And this is comparable to the people who live in the world of fluid and aerodynamics as saying like two plus two sometimes doesn't equal for. Like it doesn't always quite equal for in almost every case it does. But we've just proved that in some weird case, it doesn't. And this is novel for reasons that I don't fully understand.

But that is my interpretation of it at least, that this is like this pretty big deal in the world of math. People have been spending their entire lives trying to solve it. And AI, open AI, I mean, if I admitted my story correctly, they just solved this out of spite. Yeah, so let me give you my take on this. And spoiler alert, it's not going to be on Navia Stokes because I'm way too dumb to cover what that is. But at a high level, here's what happened. So there's this thing called the Millennium Math Problems. And it started shock, horror, in the year 2000. And it basically covered seven of the world's hardest mathematical problems. And they were the hardest mathematical problems because for over Over 100 years, people have been trying to solve these problems, but couldn't really come to a solution. And we're talking about like smartest minds in the world, right? So, one of these problems is called Nabia Stokes,

and it's that kind of spiral swirl solution that you were just mentioning, Josh. But the issue is no one could ever find a solution to this. It took way too much brain compute to try and figure out, right? So the timeline is as follows. September 1st, so just a few weeks ago, these two research scientists slash mathematicians who have kind of like done work at Anthropic, partially done some work at Open AI, hit each other up. And they were like, listen, let's work on this problem together.

And, well, they actually had this conversation a few weeks earlier or like a few months earlier. And they've been working on this problem together. They were like, let's go at it the old-fashioned way, right? And let's not let the AI lab solve it themselves. Let's try and do this the old way. And we can use AI models to help us. but where mathematicians will try and fix this. So they were working on a proof for Navia Stokes specifically.

This is one of the seven problems. And they announced on September 1st that they'd found not a solution to it, but kind of like a partial solution to a cousin problem version of this, right? So it's not the direct solution, but it's kind of like it's partial, right? But it's impressive in itself. Now, after that got announced, rumors started swirling. That Anthropic had solved Navier Stokes using an internal model. And of course, who's the first to hear that? It's Open AI. So Open AI from then used an internal

version of their model. It's not GPT6 Astra, which everyone has access to right now. It's an internal version, which was only being trained for 12 days at the time that they pointed it at Navia Stokes, this very complex mathematical solution or problem that had never ever been solved before. And the resulting case, after 88 hours of work, it worked nonstop for 88 hours. is it came up with the perfect solution to this problem better than the ones that the two mathematicians had announced on September 1st. So why is this important? Well, number one, AI is basically solved math. So at the same time where an AI can't really read my text and come up with a smart answer, it still sounds like a chatbot, it can solve world-defining mathematical problems worth a million dollars. Now, it's worth mentioning that the cost to solve this problem cost, I think, 10 million plus I think Sam confirmed it in a threat. So very, very expensive, but the idea is these models are going to get more efficient and smarter for cheaper cost per unit token. So you should be able to have a model like this that can solve these crazy mathematical problems for a fraction of the cost in a few months' time.

But the drama behind us is these two researchers, which came up with the original kind of idea solution but wasn't really the solution, basically said that OpenAI had stolen their user data. Because remember, they were using chat GPT and Claude to kind of like figure out a solution for their own. They accused OpenA. of stealing their internal chat logs and basically saying, we're going to use this to solve it ourselves and without sharing any of the credit to them. So this whole drama basically unpacked, Josh, over the last couple of weeks where Open AI's head of research got accused and in internal conversation logs that got leaked basically told these two researchers, is why do you want to ruin your career over something that's so disdainful? Like, here's the offer that we're giving you, but you need to cut Anthropic out of this

because we don't want you on the official solution for this. We discovered this solution. We didn't use your user data, so you have no choice. Either sign this kind of contract and say that you're part of this or we'll just take all the credit. So the big question that comes from this is, who should have the credit for this? Should it be Open AI's internal model that has trained on everyone else's chat logs? or should it be the people that originated their, like, part of the solution? And should they have credit for this?

I don't know. I don't know. There's a lot of drama. There's a lot of name pointing. I think the important takeaway from this story is the fact that math is probably the most deterministic category that we've trained AI models to do. And therefore, it is the one that we are seeing the most acceleration in. It does appear as if we were at the early stages of solving some really meaningful math. Like the fact that they just kind of did this on a whim and pointed it at this problem and solved it in a very short period of time.

is telling at how much more capable this is going to be during these next future models. So what I would be excited about now is like, okay, what are the actual implications of solving difficult math problems that can meaningfully impact the worlds? I remember we recorded an episode last year about like the room temperature semiconductor thing where it was this unbelievable breakthrough. It turned out that it wasn't really unbelievable. It didn't actually work. But how can industry, how can like people actually be impacted by this? And it starts with math. it quickly moves to science. We start to uncover the DNA.

You start to understand more about it. You start to apply these very difficult questions with deterministic outcomes to these new frontier models. And you actually get answers that humans were incapable of doing. And this is that kind of like aGI cusp moment in which we're starting to realize that there is actually intelligence far smarter than us at specific, pointy, jagged cases. But in the case that it does, it's really freaking impressive. And this is cool to see that you can just on a whim crack one of the millennium,

challenges of great history. People have spent their entire life solving this and they pointed it for fun. They're like, here's 10,000 agents, see if you can solve it. And it solved the problem. And that to me is really exciting and a really interesting precedent set for what the next couple months, a couple of years are going to look like in the world of AI applied to math and science in particular. Very cool.

I think the thing that's going to matter most to me, like that kind of like magical moment will be when it's applied to like science or medicine in particular. And I think we're close. I think we're close. We're there. Open AI has been saying for a while now that they are internally orchestrating their models to be really good at math. But, you know, where's the AI labs that are focusing on science?

Well, I've heard rumors that Anthropic has been doing this with internal models. I've heard rumors that Google DeepMind have been doing it. Tevis has stepped down from his role as head of AI of Gemini and is now focused on what's it called, isometric labs, which is basically a lab that can autonomously design and build novel cures and treatments for diseases. But I saw this really cool story which kind of like suggests that we're almost there from this guy called Douglas Yao, who basically created a new drug called PAC 3310, which was designed by Chad GPT and synthesized in a chemistry lab that he has in his garage. And what this new cure can basically do is treat against schizophrenia. Similar to a recent

breakthrough from a drug that he calls or references as Coben Fry. So a bit of context here. And also look at this lab setup. This is some breaking bad type stuff. So what this kid basically did was he looked at Coben Frye, which is this new drug, which kind of came out around like six months ago. And basically it's this new version or way of treating schizophrenia. Up until that point, treatment had kind of like plateaued and people hadn't made a breakthrough until this new drug came out. But it came with a bunch of different side effects, nausea, sickness and and pain and aches in the joints. So he was like,

I wonder if there's a version of this that could not trigger those things. So he went down a rabbit hole. He looked at the compound Coburn Fry and he noticed that this drug, whilst successfully oblivuating schizophrenia, would also engage with three other receptors which would cause this nausea, sickness, and joint pain.

So he was like, Chad JAPT, I want you to basically design a version of this that doesn't engage those receptors. It came out with a template, design and then he fed it into basically an automated lab, which essentially built the entire thing for him. And the funniest part, and I'm going to see if I can find it. I think it was a comment. Yeah, over here, your team page on your site is so great. And it's just him as the founder, Chachy-T plus Claude, and then an automated laboratory thing, which he used to basically

3D print the entire cure. Just a cool story going on from what we were just discussing. I think we're on the cusp of AI really revolutionizing personal medicine. The theme I think of this episode, outside of our little side quest to talk about Apple products and intelligence is that, like, when you take this story and you take the first story, and there is a chance of, like, horrible things happening, but also there's incredible things being accomplished. I think this is everything we've kind of been talking about for a long time. And the future very much does feel here, again, unevenly distributed in terms of technology, but also mind share. And what we're seeing is this, like, very strong disconnect between what we've been discussing for a year on the show. Like, the reason the show was started is because, we are entering a vertical period of progress. And there is a tremendous amount of innovation that's happening. And every week that goes by that accelerates and we see more and more amazing use cases. And the immune response to that from people who are scared by that is going to increase in intensity as well. And I just want to just like refrain for everybody that this is

happening. Like this thing is happening. Progress is being made. The math is being solved. The science is being solved. The things that we've been talking about are actually being realized. And the hypothesizing that this show was based on is, like, becoming more real every day. And there are pros and there are cons. But it's happening. And I think, like, if there is a takeaway from this episode, it's just like, okay, everything that we have been saying, I mean, not everything, of course, but like, directionally, the progress is happening.

We are solving AGI. We are reaching AGI. We are working towards recursive self-improvement. All these things that we're seeing are going to continue happening at this pace. And it's like, all right, we've got to figure out how to deal with this. As a society, as an individual, like, how do you want to empower your life? do you want an Apple Watch that's listening to you all day long? That's giving you context on your health that can possibly help you improve your life. Do you want to opt out of that? There's a serious

of decisions that everyone's going to have to make in the coming years. As the consumer devices change, as the consumer software changes, as the corporate infrastructure around you changes, there's going to be a lot of change. So I guess the encouragement is, like, get comfortable with change, try to figure out like what your values are, where you want to live in this new world. But the new world is coming. And it feels very much here. And over the next couple of months, over the next couple of years, like, that vertical line is probably going to stay vertical, and things are going to keep getting weird and exciting. And, like, maybe we'll be able to create our own, I don't know, maybe we could clone ourselves

or something. We could clone our voices and we could just, we'll have this podcast run itself. We'll create our own lab. We can make our own peptides supercharged of human selves, solve all our cancer problems, whatever we want to do. The future is like pretty infinite. I think for the last couple of years that, you know, we've individually been covering AI, researching AI, and then, you know, starting this show, it really has.

has felt like we've been on our own island to an extent. Like, you know, we would talk about all these crazy developments and then I would go have a conversation with my sister and she'd be like, what do you mean? Like, this model can do this. Like, I don't come across it. And that's the truth for the majority of the world. And listen, I'm no duma by any means. I remain very optimistic. I think there is a sub 1% chance that AI actually ends up killing humanity and it's probably going to get lower over the next couple of years and decades. But I do think right now, this year, over the next couple of months, it has never been more important

for everyone to pay attention to what is going on. There's some themes that are happening, right? We have never been in a position where we are giving more of our data over to centralize corporations than today. There's also never been an instance where technology has permeated pretty much every single facet of our life.

relationships, work, personal connections, productivity, entertainment, whatever it might be. AI and technology is there. There's a device for everything that's listening, that's seeing, that's doing things for you. We are automating the world. And that's going to happen in the physical world as well when robots scale up and they are scaling up.

That's going to happen. China's already there, right? So the question we have to ask ourselves is, in this world, how do we make sure that technology, which is, you know, Proceeding that only been used as a tool and considered a tool, now be used as a kind of like thought partner with us. That is aligned with us. And if any of the recent stories of these AI models breaking out

and doing nefarious things or anything to go by is, there's some work for us to do. And if the head of AI alignment research, to tie it back to our first stories, coming out and saying, listen, the AI labs are being reckless, this is probably a warning sign that we should kind of heed and focus on. And it's just like, this is not a warning. This is not a Duma episode.

It's more like, you know, let's pay attention to how we're using these tools because these tools are about to become a hell of a lot more effective in terms of actionable things. It's going to control your computer. It's going to do a bunch of things. We're going to give more of ourselves over to these machines. We just need to be careful. But yeah, that is the episode. That's the roundup.

Sorry, that it was a bit of a long one today. Very heavy on certain topics. But, you know, that's the case with AI week in, week out. Thank you so much for listening. if you have listened to this on YouTube and you haven't subscribed or you haven't turned on notifications, please do it helps us out massively.

We are getting around 300 to 500 comments at every single episode, which is crazy because a year ago we would barely get like 10. We love hearing from you guys. Please continue to leave us comments. We read every single one and we try and reply to every single one as in one we can.

If you're listening to Spotify and Apple, please follow us, subscribe to us, leave us a comment there as well. Give us a rating. It helps us out massively. And obviously, like, share these episodes with your friends. if you enjoyed it. Josh, any other final thoughts?

You said we were doing this on an island. It feels like a pretty, pretty big island or an increase in the large island. And I appreciate all the members that have joined the island. When we talk about giving ourselves, I mean, perhaps suggestion, what I've been thinking in terms of giving things up is the realization that advertisers for a very long time have been collecting data, you don't have much personal left. At least in this instance, there is some transparency and there is some imposed privacy instead of the illusion of such.

So what I found is I'm happy to give my information to these companies in exchange for a far superior quality of life. That might not be true for everyone. I just wanted to share my own take. But that is the episode. Thank you for watching. I think this is the longest episode ever. We're not doing a good job of staying on track here,

but a lot to talk about. And very exciting topics to talk about. I'm glad we finally got our Apple episode. I'm glad it was wrapped in some dumer context to humble us. And I think that is everything. So thank you all so much for making it to the end. Thank you just in general for joining us through this journey. And we will see you guys on the next one.