Latest / The Bleeding Edge Podcast / Weekly Ai Brief June 25 2026
Transcript
- Ralph: Hi, and welcome to another episode of the Bleeding Edge Podcast. I'm your host, Ralph Benker, and with me in the studio, Emile Ogier. Emile, welcome to the Edge. Emile: Great to be back for a full time, Rolf. Ralph: Yeah, well, ⁓ you know, anybody out there who wants to start a podcast, I'm sure we've said it before. It's not the easiest thing getting two people and Riverside to cooperate with all of the vagaries of technology we have these days. You would expect we could just get on and talk, but no. So this is take four. and hopefully ⁓ we'll be even more rehearsed for it. So gonna start the ⁓ show today with a bit of ⁓ on some of the jokes as an intro to stories that we'll be talking about as you know my test for a GI is whether AI can actually write a joke and ⁓ and you know we haven't achieved AGI yet because it has really not been able to do consistent job in terms of humor. But anyway, we go. So First the first one up, ⁓ 42 state attorney generals subpoenaed OpenAI this week. and the charge was the chatbot is too agreeable. The joke is 42 states agree on nothing except that they can't stand being agreed with. Well, ⁓ I that quite funny actually. not laugh out loud, funny, but yeah, kind of humorous. Emile: Okay. Ralph: The second one, Anthropic launched the most powerful model the public has ever had. the public had it for three days, ⁓ then got shut down, not by a hacker, not by a rival, but by the guy who wrote the thirteen billion dollar check. If you don't know who that guy it's Andy Jassy, ⁓ the who's the CEO of Amazon, who basically was the whistleblower. ⁓ ⁓ he's also an enormous investor in anthropics. I don't know what he was playing at that day. But ⁓ anyway, we'll talk more about that story. not that funny. So I'm gonna kind of go, you know, AI half, and let's see third one over here. actually this one is so terrible. I'm not even gonna go there. Something about Google and its Emile: Yeah. Ralph: Toll scam. OpenAI built an AI to replace the knowledge worker. This one is a little bit funny. This it spent 150 million dollars hiring 300,000 consultants to explain it. 300,000 people to agree with the AI that agrees with you. that is actually just really sad to be honest. But yeah, we'll talk about that story a little bit as well, because obviously OpenAI have pushed to ⁓ get McKinsey and Bain and all of the big consulting companies on board. then a joke about Midjourney. Midjourney, one of our favorite text to image generators. the company spent three years learning to draw a hand with five fingers. And they just announced the full body medical scanner. The thing famous for getting the outside of you wrong now wants to look at the inside. Yeah, a really interesting story by the founder of Midjourney and fascinating to see how he's pivoting. so look forward to that story. Then the last one, two countries regulated AR agents this week. Estonia decided every AI agent needs an ID card. Emile: be covering that later. Ralph: Argentina decide decided AI companies don't need humans at all. One country wants to see the robots' papers, the other gave it a corporation and left the room. I do find this one really interesting, but that's also just not funny. So we're safe, everybody. We haven't yet achieved AGI. and we're so nowhere close to ASI, and we are certainly nowhere close to universal AI, which is reportedly the ultimate expression of AI that we are working towards according to a new Deep Mind, Google Deep Mind paper on Archive. maybe we'll talk about that one as well if is we ha if we have some time. So I think we're basically giving a no vote of confidence on the humor segments this week. we'll give it a go next week again with the tweaked algorithm. So who knows? We might get a laugh out of all of you at some point. headlines this week then so ⁓ the big news obviously when as my ⁓ research ⁓ says here, the government walked into the room. So obviously this is on the back of the story that Anthropic shipped Claude Fable Five. I was using it, it was a real delight, and ⁓ and it got shut shut down three days after launch. And that was basically a tip-off from Andy Jassy at Amazon. which is you know anthropic's actually largest investor is what my research is telling me which I was a bit surprised about. but that was because his team had figured out how to jailbreak it. of course the news around that is that Anthropic have said, well, that jailbreak is possible on nearly all models. That's nothing new. and so what's all the big fuss about? and of course the US government actually didn't shut the model down. What they said was it can only be used by US citizens. The rest of the world is not allowed to touch it. ⁓ And couldn't figure out how to limit it by citizenship. Because hey, guess what? you know, these models are out across the globe, ⁓ and many workers are not US citizens either. ⁓ And so they wouldn't even be allowed to use their own model. So I think that was a little bit of a weird request from the government. Maybe they understood that you know asking for a full ban was not appropriate. but you know, either way, we don't have Fable Five anymore, and ⁓ we don't know when we're gonna get it. Maybe you've got some news story on that. the other story, ⁓ obviously these forty two state attorney generals subpoenaing open AI. Over how its chatbot behaves. So we've seen quite a lot of stories of some terrible cases really of AI gone wrong with open AI being sued by a number of different states over the sycophancy of the model. And basically the federal energy Apparently they're called FERC. They've given ⁓ grid operators 60 days to rewrite the rules for AI data centers. So that's interesting. Estonia moved to issue ⁓ cards to AI agents, and Argentina draft a law granting legal personhood to AI run corporations. That one's just particularly interesting for me because, as you know, the experiment right here on the Bleeding Edge podcast is can we set up a fully autonomous AI music label? and I am definitely looking to incorporate that into Argentina because guess what? It's gonna be its own legal person. So we'll have to see if we can actually do that. We'll ⁓ we'll give a bit of a test. so no doubt for those of you that have been following the IPOs, ⁓ obviously we've had the ⁓ SpaceX IPO, ⁓ and four days ⁓ after biggest IPO in history. ⁓ SpaceX bought Cursor, rumored to be around sixty billion. Of course, sixty billion was the option that they had agreed with the cursor team some time back. And so they've just exercised it much earlier than anybody thought. Makes absolute sense. Cursor is an excellent harness for development, and so I ⁓ they have well and bought themselves back into the running because obviously, as we said on the last episode. Groc with a K is only three point four percent of the market, mainly used by X users. And so Cursor actually puts them into the mainstream in terms of engineering and development users. Deep Seek raised seven point four billion. And actually an interesting around that ⁓ is the state ⁓ is large investor, so Chinese government have basically in you know invested in DeepSeek, ⁓ but they have also said that none of the investors can pull out. So basically, once you're in, you're in, and only will tell you when you can exit and cash in your chips. only they have voting rights, which I think is also really interesting, fascinating model. Not sure that would get through most. ⁓ regulators in the West, but there you go. Chat GPT crossed a billion phones. and so that's obviously a really big in their downloads on phones ⁓ and a reasoning model quietly diagnosed eighteen that human specialists couldn't. So a bit of a story there ⁓ what's going on in terms of the medical deployment of AI. So some some really interesting ones there. I think know I'd love to dive into these kind of I've got like top eight deep dives so ⁓ no doubt we will get into these. But ⁓ Emil Mythos Fable five ⁓ what's your take on on what the the government pulled off in terms of that shutdown? Emile: Yeah, got quite a bit to say about that. Before I do, just want to cover the cursor piece. I'm full agreement with you. I think it's a really good strategic move. I've been a GROC user for quite some time. It's not quite there, unfortunately. Elon's tried a lot of interesting things that haven't quite worked out. But cursor four, as you say, the harness is really strong. They've got RL. They've got a lot of good coding data that they can learn from. And in addition to that, know, Elon's lending them the might of his Colossus mega both and Colossus 2. So they've got the compute power. They've got the data to make this a better product. And in addition to not on top of what they've really that inference model, that Kimi 2 last run, which, you know, was very close on par, matched to Opus 4.8 and GBT 5.5, they're now fully rebuilding it without Kemi 2 and the compute power and all that rich that they can teach the model on. So I think the next model release is going to be really impressive. And most probably for now, as it is paused, but the new Grok will be. cursor in whatever shape or form that is. So I'm really excited to see what comes of that. Ralph: Yeah, and I think what we're learning is that the harness is really the differentiator here. So you know, we we speak a lot about being able to use tools like Kimi ⁓ and ⁓ Deep Seek and Quen, which are some of the Chinese models because they're really cheap, ⁓ or they're really inexpensive to run in comparison to certainly Claude. and ⁓ and of course you can do that and you you know, you quoted some really great benchmarks and are great benchmarks. The problem is when you try and use those tools, you realise very quickly that the LLM is only a small part of ⁓ the that need. And the reason why Claude Code is so capable, it has a great LLM that's powering it, but of course that's only about two point five percent of the solution. And what it does really well connect into an ecosystem of tools capabilities and MCPs, and it's really the connectability and its ability to you know pick the right tools, and that all sits within this concept of the harness. And so what you have to have first off is a good harness, and then you have to have a you know a good LLM. Well, need both of those together. and of course, when you're like you know using something like Kimi The best way to use it is actually to run it through Codex or to run it through Cursor. you can also connect it to Clawed Code. it doesn't quite run ⁓ as in Claude Code from my what I understand. I've been ⁓ setting up my own tests with it. but ⁓ I think ⁓ you know, spoken about the harness quite a bit. And to really understand Elon's deal here, you know, it's it's about plugging that gap, about ⁓ you know. Getting the harness that really allows the LLM to ⁓ of the tools and the full ecosystem that actually makes a development environment possible. And and that's what that's about. So yeah, great things coming out of the SpaceX acquisition, I think, of Cursor. And ⁓ I really forward to what they're going to do with it because you know, he's got the full stack, no doubt. We will be seeing data centers. In the sky or in space with his ability to launch satellites. I think the only thing he's missing is his own hardware in terms of the the GPUs or CPUs that can kind of run this stuff. I see Amazon has also released released its own chips. of course, ⁓ Nvidia has got the share of the market with its GPUs. ⁓ I do see that AMD has also come out with a new computer that is able to share CPU and GPU. And then obviously the other big IPO that we haven't even touched on is Cerebros with their new chip technology, which has some really interesting capabilities. So I think the hardware space is also really ramping up. We'll be fascinating to see what Elon acquires in terms of ⁓ that capability in the coming months as well. So so good call out. Emile: I'll get on. to the Fable 5 update, Rob, but thanks so much for pointing out that observation with the connectivity on to Claude. I wasn't aware of that because I'm quite keen to do that with GLM. Just wondering if it must probably face the same kind of connectivity issues you're experiencing. So more on that once I try it. But for now, more in addition to what Rob's covered on the Fable 5 update, the overall audience sentiment is to Anthropics announcement was predominantly negative and frustrated, users expressed strong disappointment over the abrupt global disablement of Fable 5 and Mythos 5, viewing the US government's export control directive as overly broad and disproportionate. While a minority acknowledge legitimate national security concerns around advanced dual use capabilities, the prevailing view criticizes both the government's heavy-handed approach and Anthropics decision to apply a blanket restriction rather than targeted compliance. Skepticism about the stated justification and narrow jailbreak was common border concerns about US tech dependency. The government intervention seen as regulatory overreach or panic at harm to legitimate users and businesses worldwide. Trending requires include comments such as, genuine question for whoever drafted this directive under the deemed export rule, your foreign national employees can't look at the model they built. Their own commit history is now an arm shipment to themselves. And for those interested, I think across all the labs, over 60 to 65 % are foreign nationals. another reply was this marks the end of an era. The public may never have such open access to frontier models again. What happens when only the government and frontier labs have access to the strongest models? I'm not sure, but it's probably going to be a whole lot less fun than we used to. And lastly, this is a massive wake up call. Imagine your startup build on Fable 5 and it just went dark overnight with zero warning. Anthropics own foreign national employees cannot even access is proof that you cannot fully rely on the American tech infrastructure. Countries around the world need to start building for themselves. Build build local. So yeah, all pretty interesting stuff. Ralph: I think it's a big wake up call for every nation state. I know in Europe they've also now said, well, this is you know, just points to the need for sovereign AI. Hey, that's what Mistral is supposed to be, guys. but no news on Mistral again. So what's going on? and I think you know, ⁓ it's basically being compared nuclear bomb capability. So I think ⁓ this is the kind of thing that you you know you restrict access to nuclear weapons. So this is every bit as capable as a nuclear bomb, let's say, for ⁓ creating some destruction. And get you know why they have basically decided to intervene. That was always going to happen. It's just a little bit earlier than what I thought. And I think issue here is that none of us can really depend on any one model, as you've called out. So, you know, even if Of course, startups you you might have even started your startup on the basis of being able to continue with Fable 5. Certainly introduced some new capabilities. but of course, you know, even I updated a of skills, and certainly Tony had updated ⁓ all of the development in the dev stack to on Fable 5 and had to reverse all of that to 4.8 so You know, ⁓ this isn't just companies that are being affected. It's actually folks like us sitting at home with our own you know, dev and AI practices. And so I think it's a big deal. ⁓ And it means for me actually being dependent on any one technology something that I will certainly be building out of my practices and making sure that I can plug in different models and I think that's just the way to go in the future. Emile: For sure. And Rolf, just want to add a bit more color to that with the lens on Eastern Front, as you mentioned. The AI ⁓ is facing extreme market right now. Anthropics Fable is still largely unavailable because the CEO is struggling with government licensing websites, while SpaceX has just swooped in to buy the AI code until Cursor, which we covered. So Cursor has now hit a staggering 2 billion in annual reoccurring revenue in just 18 months, making it the fastest SaaS company to ever reach that milestone. Pretty impressive. Interestingly, Cursor is also rolling out its own 1.5 trillion parameter model, which I mentioned earlier. meanwhile, China is disrupting the market with GLM 5.2, which I keep calling out that I'm very excited about. And is an open source model that benchmarking tests suggests beats for Opus 4.8 and ChattGBT 5.5 on front-end coding. So good for both of us, Rolf. This disruption is heavily cost driven. GLM 5.2 is 82 % cheaper than Opus 4.8. To put that in perspective, $3,000 buys you million tokens with Fabel 5, but that same amount buys you 682 million tokens of GLM 5.2 and a massive 3.45 ⁓ billion with another Chinese model if you DeepSeq 4.5. version 4 Pro. The reliance ⁓ highly subsidized all you can eat AI plans seems to be ending. We are moving towards usage based pricing. ⁓ which is forcing major players like Microsoft to explore these cheaper open source models over expensive proprietary ones. The shift is even hitting the creative economy hard. I saw a very interesting post Tim Ferriss. He's not happy. He was objective about it, but he's reporting a 50 % drop in sales ⁓ all his non-fiction books. And he's been at the top of the charts for over a decade. And he's claimed that that has always been a constant source reliable, predictable income from him. So, you know, these models are hitting authors hard. And realized books like the four hour body functioned lookup tables in 2019. A book was the best inference for that information. Today, a chatbot that has scraped the data is simply faster and more personalized. As information content gets commoditized by AI, enduring human lies in what can be easily summarized. Sorry, that can't easily summarized. So that's pure experience entertainment. and authentic storytelling. I wonder with the end of subsidies and AI processing, are we finally about to see what these models are actually worth? Ralph: Yeah, that is a a fascinating take from Tim Ferris. Obviously, he wrote the four hour work week and basically built a phenomenal career out of you know, what became was the forerunner to, you know, this movement of being a a digital nomad. And ⁓ obviously so relevant in the age of AI, given the capability of what you can run on your phone and your laptop while sitting on a bus. And ⁓ I don't know what it means for authors. I think actually, you know, if you're writing a book now, you're probably writing a book for an agent related to this. I know some recent research released by DeepMind on Archive, it basically actually opened with ⁓ here is instruction for any agent reading this paper. Do not summarize the following statements. These are the core aspects of this research. So It was really written for humans for agents. ⁓ ⁓ know, primarily that is what we're doing with any technical documentation these days, because when you start to read the preamble of some of this research, you know, the language makes it incredibly ac inaccessible. Of they have to state the hypothesis and how they've gone about the research, but we just want to get to the the the meat, we want to get to the insight, and so ⁓ this whole phenomena of TLDR, too long, did not read. we're all summarizing the world that we you know that that that we're trying to understand. ⁓ ⁓ I don't think that's particularly great, but of course, you know, we are really writing for agents and for AI these days. and ⁓ and we will get a thin slice of everything that you know is in that research summarized back to us. And so I think the the writing is on the wall in some ways for nonfiction authors. I don't know where we go with that. I mean, how do these people make money out of it? Why what motivates them to continue? I think maybe these things change and we become really focused on the the ethos behind it. You know, so everybody has to become a character, everybody has to represent ⁓ some point view ⁓ and invite others follow them as a result. And so this is what YouTube has been built around and influences are being built around. Well, maybe authors just become influencers. They just kind of disappear as a concept. And unless you are an author as an influencer, nobody really cares about your book because only AI you know bots are going to be reading it. I think that's a really interesting evolution of the concept of being an author. Emile: Absolutely, or maybe all authors are just pushed down the fiction reads, which, you know, I'm finding that quite hard to get my head around as a big non-fiction reader. I'd like to know that I'm reading some human books, but let's see. Ralph: Well, you know, maybe everything that we have released to date has been fiction. So many years ago I watched a show with Stephen Fry, and ⁓ the show's called QI, and they do really kind of interesting science based stories. And one of the facts was that that seventy-five percent of all facts that have ever been proven have now been disproven. And so essentially ⁓ that was fact. and of course we're seeing such incredible advances in science. Emile: Ha Ralph: that it's absolutely true. We'll keep finding stuff that has been proven to not have been true. So basically, all truths up for grabs. ⁓ all truths are maybe works of fiction, you know, just until the next research paper comes along. but I know what you mean, and that's a little bit of boon for me, to be honest, because when I read a book, I tend to read over the last few years, at any rate, I've tended to read nonfiction. but you know, when I have read fiction, I have found that to be actually ⁓ such delight. maybe that's something that I will get back to because maybe that is the only reading style that that is left. So just to also go back to that cursor ⁓ story, I think what was quite interesting, you mentioned the Composer three, which is the one point five trillion parameter model. that cursor have been working on so ⁓ that's actually been trained from scratch on a hundred thousand gus and it's as big Opus and GPT and it's shipping in weeks and so basically you know this is this is what Musk has just landed as his engine for development in SpaceX ⁓ and I don't know such a move I absolutely love it. Emile: I did. Yeah, it's amazing, Rolf. look at the price and are we getting really desensitized to numbers? 60 billion is a lot of money. after the SpaceX IPO, it's jump change for him. It's easy decision. Ralph: Yeah, I mean I think what did he I don't know what the valuation was. I think the ⁓ his evaluation was well no, but in the total SpaceX I think it passed at some point two point one trillion, which eclipsed Saudi Aramco's twenty nineteen record. Emile: 60 billion. That's right. think for a partial moment over night, Marcus went north of close three billion. Crazy. Ralph: Very interesting. So on to ⁓ a next ⁓ just quick deep dive story. As I mentioned in the news headlines, OpenAI rents ⁓ armies. So this is basically where they've invested 150 million in basically acquiring partners. So McKinsey, Bain, BCG, Accenture, PWC, all big consulting firms that are obviously consulting into the market. Emile: Thank Ralph: And ⁓ they have basically created this open AI partner network that is targeting three hundred thousand certified consultants by the end of twenty twenty six. And essentially essentially they are giving them access to tools like Codex, educating them in cybersecurity, AI agent development, and they also essentially trying to set them up. As forward deployed experts. And so Palantir, the defense company based out in the US, really invented this term of the forward deployed engineer. So this is essentially somebody that is a blend of AI engineer and ⁓ change manager consultant, and they're essentially working directly with clients to bring new technologies to bear. ⁓ I think that is the new age for anybody that has worked in ⁓ transformation in change management or indeed consulting, because I think what you're doing is you're really, you know, bringing your tools into any consulting engagement. I've said it many times, you know, if you bring in somebody like myself, I don't just come with the skills that I had in my PowerPoint slides anymore. I come with my army of agents. ⁓ and that is really what you're buying. And of course that that is what you're buying when it comes to this OpenAI partner network. and presumably you know, we will see that Anthropic follows suit and they will all be fighting for basically these to adopt and represent their tools. So ⁓ I think it's big deal because it's really you know. kind of defining what it takes to close enterprise deals. It's not just about distribution. distribution is obviously what Google is really good at. They're connected to so many of world's companies providing them services. But even they have struggled to sell in some of these larger components that they have within their AI stable. And so I think you know influencing or renting these big four consulting firms To do the work for you is a smart play. but no doubt everybody will be you know, building on this. And so I think there is also an anthropic clawed partner network. I haven't actually got into detail on what they have done there, but I will we'll certainly have a look at that in the future. So what do you think, ⁓ Emil? Would you sign up ⁓ as ⁓ a OpenAI partner member sell soul to Sam? Emile: I don't know if I'm selling my soul, but it's certainly in this space that we are deeply interested in and by your definition of forward. what did you call it? Forward engineer expert. I think we both very neatly fit the bill for that classification. So that's good to hear. You know, we've got the background plus the AI interest plus the agents plus the AI capability and expertise. So sounds positive. Need to look more towards that space. did want to, go ahead. Ralph: I was just gonna say, I have ⁓ officially changed my title to a forward deployed engineer, who spends eight hours a day on on Claude. nobody knows what I'm talking about. So I I also tell them I'm a transformation expert. But ⁓ there you go. I'm sure the market will catch up at some point. Emile: but maybe need to switch the engineer to expert, because it also adds another element. Ralph: yeah, maybe the engineer is is a bit much done. Change approved. Emile: I'm going to move on to the most exciting story I read this week, Grov. We don't cover medicine and health that much. I've been waiting for the next big story. And it certainly is one you covered earlier. I just want to expand on that mid journey. what if getting a crystal clear view inside your body more like stepping into a warm sparkle rather than lying in a cold hospital scanner? Well, mid journey just dropped that possibility by launching a new medical division and unveiling a working prototype. Cardiologist F. Shine Emrani shared detailed breakdown on X it's easy to see why this has caught attention. Here's how it works. You step into a platform in a shallow pool of water. The platform slowly loads you through a ring containing roughly half a million tiny ultrasonic transducers, each the size of a grain of sand. These act as both speakers and microphones, sending sound waves through your body. from every angle and recording the echoes. And in about 60 seconds, serious computing power and around two pets of flops holding 16 gigabytes per second and a lot of AI compute power turn that data into a resolution 3D map of your entire internal anatomy. So no radiation, no magnets, no claustrophobia and no contrast eyes. Imrani points out why this could matter for something like heart health. Current detail scans are usually reactive. You get them when a problem is already suspected and they're expensive, time consuming and involve trade-offs that make frequent impractical. This approach could flip that, letting people track changes over time with quick, repeatable scans and build a personal baseline that AI could analyze for early shifts. The company is thinking big with plans for 50,000 scanners worldwide over six years and capacity for a billion scans a month. That's just incredible. For first location described as Mid-Journey SPA in San Francisco is targeted for the end of 2027. And get drumroll, scans are projected to cost just a few dollars Rob. It's built on the Butterfly Network's ultrasound on chip technology combined with Mid Journey's AI reconstruction. The goal is image quality approaching MRR quality in many ways, but dramatically faster and without the downside. That said, Emrani is clear that this is still a first-generation prototype. Only a small number of people have been scanned so far. Current times run longer than 60 second target and it hasn't received FDA clearance for diagnostic use. Full clinical validation against existing gold standards are still ahead, along with bigger conversations around privacy. at that kind of scale, but the overall audience sentiment is cautious and skeptical with a smaller share of measured optimism. Many replies acknowledge the conceptual appeal of rapid radiation free imaging, but emphasis that the prototype remains far from clinically validated for diagnostic use. Medical professionals in particular expressed doubts about overheart claims relative to current technical realities and regulatory requirements. Excitement centers on long-term if hurdles are cleared. And a reply from a retired radiologist that I called, he said, it's great that new image technology is being worked on. I'm not impressed with the image quality. It's not clear whether the product is recording anatomy or physiology or something else. but I look forward to seeing how this evolves. Even so, if the resolution and reliability hold up, it could represent the most significant new imaging approach in decades, turning advanced internal views into something routine and proactive instead of rare and reactive. know, way you look at this, I just think this is super exciting and, you know, I hope they make good headway with it. Ralph: Yeah, I love it. And I'm so impressed with how the founder of Mid Journey has been able to just continuously evolve into ⁓ like know, unbelievably fundamental companies that he has founded and and I and love this model. I think it's fantastic. I'm sure he's going to do something great with it. And of course, you know, just kind of frequency resonance is just ⁓ such a technology that I can really see this working, you know. ⁓ I think it sounds very accessible I can actually understand it, which is unusual. so yeah we'll be rouging for that one. ⁓ well Emile: Okay, let's go. Last one, see you guys. Ralph: I I love that story, but ⁓ on to a quick one that I alluded to earlier, which ⁓ is Google's actually suing a Chinese network that used Gemini to run a one point nine billion dollar fishing s fishing empire. So excuse me, we're just talking about fishing as in going fishing. you know, this is really scams where they sent out something like two point five million scam texts to people saying, ⁓ you know, you you've got an unpaid toll bill. they basically got something like 3.87 million credit card details that they then stole and essentially they something 9,000 fake sites, 1.5 million fraudulent URLs. So these were all websites that they created so that you would go, they'd send you a link, you'd go and it would all look very official. And ⁓ obviously you would then pay your unpaid toll. so you know, it's an interesting story. Why? Well, because Google isn't even allowed to be used in China. So the fact that they have actually used Gemini to set this up and run it out of China is actually quite ⁓ hilarious. but also very do something positive, guys. Come on, outsider enterprise. I mean, you know, anybody can start scamming actually, you know, add something to the human race. but I just thought that was ⁓ a really fascinating story that has come out. And I just guess it speaks to the level of sophistication that ⁓ can expect ⁓ the the darker side of humanity. and I mean this is exactly the reason why Fable Five, you know, gets pulled, because we just can't trust ourselves to behave decently. And somebody out there is gonna try and scam you. AI is such a great tool for that. Emile: That's a crazy story, Rolf. And I guess on the other side of the coin is, you know, all these controls for safety and Fable 5. know, either you look at it, if it's a good or bad thing, it's an accelerant for China to get on and catch up. I anyway, cut it. I don't think the way things have been handled is a positive one. But let's see. And I've got some other like interesting facts. I just want to cover stories from Elon Musk. He posted this week. It's humbling to consider that if we ⁓ just one millionth of the sun's power for AI. that will be much more than a million times the intelligence of all of humanity. So it's quite incredible when you think about that frame and just how much untapped potential there still is in energy that we can still harness and utilize. And following the historic SpaceX initial public offering IPO on June 12, 2026, SpaceX evaluation skyrocketed. propelling Elon Musk to become the world's first and only trillionaire, which is pretty unreal. And as of this week, 2026, his net worth has surged to approximately $1.26 trillion, with some real-time trackers like Forbes saying that it peaked at $1.3 trillion during trading hours. Because of that explosive growth, his personal wealth is now roughly four times larger than the second richest person in the world. And if you want to know who that is, that's Larry Page from Google, who sits at 301 billion. I think this is going to give him a lot of phenomenal strategic advantage in both the companies and the talents he can now simply acquire, much like Meta did with Instagram. And equally, he can afford to erode market share, for example, by setting him possible offerings, something both Google and Apple also have the capability to do due to their huge accumulative war chests. It's also entirely true that the sheer scale of the recent market movement created a financial milestone that is genuinely hard to conceptualize. Following the historic SpaceX IPO, Elon Musk's personal fortune surged by an unprecedented 164 billion in a single trading session. To put that into a sobering lens, Warren Buffett's widely considered one of the greatest investors in history, and who's the 10th richest person in the world. He's been building his wealth for over seven decades, has a total lifelong net worth of 148 billion. And Elon Musk gained that in a single trading session. he's made more than his net worth in one trading session. So let that one breathe. Ralph: God, I've been destroying my wealth for five decades. So this makes me feel awesome. ⁓ Yeah, it's it's it's impressive, you know, to go from being a kid bullied at school to ruling the world. And I think he's you know definitely got the mind for it. I like way he thinks. Emile: you Ralph: Obviously not everything that he does, everybody agrees with, and that's fine. but I definitely think the way that he is pulling all of this together, the mega ecosystem that he's building. And ⁓ you know, now the fact is he's literally got the world's resources at his fingertips. So hey, go for it, Elon. Let's see what you can do. If you need a podcast, give us a call. So I think you know, with with in mind, one of the things that Emile: You Ralph: We're obviously moving towards, we've been talking about AGI, and a lot of people say we've already hit AGI. So if I it might be worth just jumping into a definition here. So AGI, AI that can understand, learn, perform any intellectual task a human can across domains rather than being narrow, specialized. the key word is general. So flexible competence that transfers across tasks. So we're talking about reasoning, language, planning, learning. r at roughly human level. So ⁓ today's systems are sometimes called ⁓ AI by contrast, though w exactly the AGI line sits is heavily contested. So there's no agreed benchmark, which is why obviously people are arguing endlessly about whether we've achieved it or not. I mean it's certainly clear that we have narrow forms of AGI and we are rapidly moving towards more broader representation of artificial general intelligence. But obviously what's actually next is what's important. So you know we can forget about AGI because we probably got there we probably will get there more convincingly. So if look at artificial superintelligence, this is ASI, so AI that substantially exceeds the human minds across virtually every domain. So this is, you know, thinking about scientific creativity, strategic reasoning, social skills, you know, huge irony there. an AI that is better at social skills than the race that or the species that invented it, and obviously general problem solving. So the term comes from Nick Bostrom, he was ⁓ an an author. I guess he's looking for a new job. and ⁓ he basically framed this superintelligence. Definition. And so it's not just about being smarter than one person, but smarter than all humanity collectively at cognitive work. And this comes back to, you know, if can just harness a millionth of the sun, then we can actually power the kind of compute that will lead to superintelligence. And so it's it's definitely a concept of usually, you know, what will follow AGI. obviously, you know, this brings us back to the mythos conversation because one of the papers that Anthropic released along with their mythos release was that recursive self-improvement. So this is the what we defined in the last episode, RSI, ⁓ repetitive strain injury, from hitting the enter key while you're ⁓ Claude Code asks for responses, but this is about self-improving AI. ⁓ and that's you know what we're gonna get when we basically initiate superintelligence. But what comes after superintelligence, that's what everybody is thinking about. ⁓ And this definitely less standardized definition, because ⁓ even if you ask the AI, it's kind of like, well, ⁓ the what's next is labeled as universal AI. ⁓ I just don't know exactly what it is. So I can't quite tell you exactly what it is, but basically it's a theoretical formal sense that we end up with a mathematically optimal a optimal agent that can actually perform in any computable environment. So you immediately we get to some weird language. The canonical example is Marcus Hutter's. AIXI, a formal model of university optimal reinforcement It's optimal but uncomputable. So it serves as a theoretical idea. And so that's what we ⁓ are heading towards. just thought that kind of interesting in the context of Elon's that if we could harness one ⁓ millionth of the sun. ⁓ clearly we're gonna need to get an AI super intelligence or maybe even a universal AI to maybe figure that out for us. but I I I I hope that's a a of interesting definitions. Emile: Yeah, for sure. All cool. And on the topic of interesting stories, I've got another one. Midway through listening to the Moonshot podcast, David had a really interesting idea. I think it's unique to him. He hasn't seen it around anywhere else. basically saying, we're looking at all these phenomenal valuations out there, all that liquidity that doesn't compare to anything historically. He wonders if... a of things to come in a changing economy whereby finances don't start to make a lot less sense in the economy. So effectively moving from what he's saying is moving from money to energy to intelligence, ultimately replacing the financial system in itself. ⁓ rationale behind that is when RSI arrives is maybe not relevant that these companies will need money. And a good of this might be ⁓ A truth is, know, Sam doesn't seem to be in a rush to his IPO and he is making noises that they're getting really close to, recursive learning. So that being the case, if they can manufacture what they need, I just wonder, does money become less relevant? I thought it was just a really interesting idea either way, even if it's fiction. Ralph: Yeah, let's kick the cash into touch hurry up and make zero cost. because know, then we'll be back in the Garden of Eden when we could just pluck fruit off the trees. Just watch out for the snakes. but I I I I love the concept, no doubt. That is going to be the discussion of our times, Emil. So not something to cover on this episode. Emile: Did. Ralph: on a really dramatic front, ⁓ is a claim that a fully autonomous drone has killed soldiers for the first time. so this actually came out of the New Scientist on the thirteenth of June. it's actually the first known battlefield death from an autonomous vehicle. And ⁓ basically this is ⁓ This paper was just looking at a single test which was actually in 2024, where they had 10 AI controlled quadcopters. they flew 3.5 kilometers, switched to a mode where on you know onboard AI independently searched for and struck targets with no live feed, no operator link, no human approving the final kill, and several Russian soldiers. lost their lives. So the the sole ⁓ source of this is actually a Ukrainian defense industry figure who supplied the tech and ⁓ was speaking at a media event. and so I guess you know why it matters is if this really is true, then we s truly have passed Rubicon in using machines to take out without any oversight. And of course, you know, this is what caused a lot of the issues between Anthropic ⁓ government. And obviously the government ⁓ in the US basically Anthropic a I remember what the term was, but they put them on the the the list of threats to national security. And this is why Anthropic did because they didn't want unfettered decision making by the AI. and of course, you know, this tech already exists if this news story is to be believed. and and think, you know, if we are gonna allow bad actors to deploy these kinds of tools, when governments and institutions, let's say, don't do then ⁓ Actually, that's probably going to push us to a very dystopian future where we have RoboCop just blasting away at targets that it has deemed necessary to So ⁓ a bit of a worry that one. I think we always know this is coming. I don't know. don't know what to make of it. I think you know, echoing your statement a bit, Emil. Maybe we just need to pause for a moment and actually Get some kind of a universal law agreed or a universal protocol for how we want all of this to play out. we've certainly tried to do it when it's come to climate change, whether you believe in the climate science or not. That was a moment where the world tried to come together unsuccessfully. we've certainly seen it within how we have tried to manage and govern. nuclear I think the moment has come. This is beyond you know the G seven or the G eight. ⁓ this really requires us to come together as humanity to figure it out because I think the stakes are pretty high. You know, this technology is just too capable from an autonomous point of view. And as we move into embodied AI with robots. the race is gonna be on to find a hiding place when once these things you know have directives that are not congruent with human existence. and no doubt, you know, some Nutter out there will have programmed the ⁓ AI to take us down. So forget about you know chemical weapons of mass destruction. we now have, you know, AI with an alignment issue. Emile: Yeah, totally Ralph. I'm glad you brought up the climate change point says, I don't know if you follow the you do follow the news, but if you've noticed. since the great need for this exponential growth for data sensors that requirements has filtered through the news stories on climate change are drying up. a bit of a cynical view, but it is that it aligns with when these data centers need to come into full force. But I'm fully on page with you regarding being a good steward for humanity. The drones scare the living daylights out of me. of the latest tech drones and they move so quickly like far faster than the ones you've seen in those pretty light shows. were hiding in a bunker, they had forward knowledge that they were coming, they gave them a shotgun for just first shooting. The came and attacked them three times, got them out of the four times. There were two them, they just didn't stand a chance, know, knowing where it coming from. They are lightning fast. It so scary but I don't want to the show on a sombre note but yeah we do need to look forward in a positive direction regarding being responsible authors to humanity. Ralph: Yeah, and ⁓ I think definitely your point about the data centers is interesting. I'm not sure that that's fully the cause for climate change rhetoric really somewhat silenced. I think a lot of it's due to the fact that so much of the climate science has been proven to be incorrect ⁓ and not founded on good data. certainly there was a big court case with the The guy that published the original hockey stick, and a lot of what he published was absolute nonsense. And so that's actually come out. it's a little bit quieter, but you can go and have a look at some of that research. climate has always changed, climate will always change, the climate is changing. I th I don't think there is any any denying that causing it is a little bit up for grabs, and ⁓ there's no doubt we have a bit of a hand in all of. so we're we're not here to debate that, but I think that's also one of the reasons why it's gone a bit quiet. And so we need data centers. Data centers are where the war is truly being fought. and this is where it gets really interesting because I think we are moving to an age now where AI is starting to get an identity. So some weeks ago, ⁓ basically seen in the news that ⁓ have Brought in some laws basically allow you to create a non-human corporation. And Estonia have done something a little bit similar. they've basically given AI a ⁓ the to have a legal identity. And so really the one is trying to leash it, the other one is trying to set it free. and so essentially they're at two ph ends of the spectrum. ⁓ but Estonia, just to give a bit more context, it's planning to be the first country to issue scoped AI ID codes. So an AI AI acting an agent essentially acting on your behalf. So you want to go on holiday, you want to book a flight, or wanna book a a hotel, you can basically tell your AI to go away and find your deal. and it's essentially borrowing your entire digital identity. So This ID encodes exactly what it may do and leaves an auditable trail ⁓ who acted for whom and ⁓ what rights and who's responsible ultimately. and I think you know that's just a good thing to do because the AI ⁓ does get legal agency. Exactly what that means in terms ⁓ of protection the consumer, I'm not sure, but it may well be the AI agent that acted on your behalf will be indistinguishable you know from your identity. It's essentially you. So if something goes wrong with a holiday, you'd be able to have some legal recourse to that. and then obviously Argentina ⁓ what they've done is they basically created a category called Sociedades de Inteligencia Artificial ⁓ that is granting full legal personhood to companies run entirely by AI. so it means the AI is able to own assets, sign contracts, pay taxes and carry a limited liability with no human owner. and so ⁓ there's no directors, there's no sh shareholder shareholders required. so author, Noah Harari, brilliant author, wrote the ⁓ the book on what was it, sapiens and sapiens two. he warned ⁓ that it hands AI all purpose to financial and political systems. and risks an AI state. now Peter Till, the big investor, he's actually moved to Argentina with his family. And he with the president of Argentina Miley to express interest. so I don't know. I I'm definitely keen to to test out this Sociedades de Inteligencia Artificial I think it sounds like a really interesting move, but clearly, you know, there are some big stakes because the AI will be able to be completely independent with at least in Argentina some kind of legal governance and coverage for its actions. Emile: That's insane, Rolf. I think for Saturday, that's a bit too much for my head to wrestle with as a concept. But I'll certainly sit with it and reflect. Gosh, things are changing fast. Ralph: Yeah, no. Yeah, it's it's a it's a it's a really interesting take. I don't know. I mean I I've a little summary here of what each side gets right and wrong, which I might just kind of read out here, because basically with Estonia, it's right about the problem and the design is genuinely good. you know, if an books your flight, it's literally borrowing your whole login. So scoping that down to may spend up to 200 euros, may not touch email with an order trail is the single most practical AI safety net. the risk is it only works if adoption is near universal. So a voluntary agent ID that the sketchy agents simply skips, ⁓ you know, is a velvet rope. And so I think this is true. And we've obviously seen kind of agent to agent protocols coming out of. Google and ⁓ so we know we're starting to get to some of the infrastructure. We know that the Linux Foundation has also set up a body for agent governance. So I'm hoping that this will really kind of feed into some of that infrastructure. Argentina on the other other end, my analysis kind of is ⁓ offering up here that you know, is right, the law is too slow, but it's Potentially wrong about the fix. So a corporation that can act at machine speed without human bottleneck is a real efficiency unlock. But limited liability with no human required isn't deregulation, it's de-accountability. So when the AI Corp launders money, breaches a contract, kills somebody, you know, then basically this doesn't really So you can sue the AI Corp, but if there's no human, no asset behind it. You've sued a ghost. you know, that's a problem, I think, at the end of the day. So no doubt are the first moves of what is to be a very exciting regulatory change that we all going to have the honor of living through if the drones don't get us. ⁓ but ⁓ Emile: Yeah, Well, sorry, just want to add, Rob, I'm just trying to put this into perspective. is interesting, having sat and lived in the regulation space for a long time. If you think about the contactless cards, how careful and slowly increasing those thresholds took when the cards first came out. I can't remember what it was. I think it was five or 10 pounds, and then they raised it to 15, 20. That took a lot of time. we like from out the gate saying limited to $200. That's quite a step change. Yeah, frightening. Ralph: Yeah, I mean, you know, we we are the bottleneck, so we do need to free these agents. And just as a bit of a check-in ⁓ the music project, I have actually you know, a reminder to everybody if it's the first time listening to the show, we are testing what AI is capable of. And one of the projects you will find on the Bleeding Edge, versale dot ⁓ you'll be able to go to that website and see all of the show notes. And on there, there is a building with AI menu, and it basically tracks our our project here. And ⁓ the project is to create a fully autonomous music label. So ⁓ no human in the loop at all. ⁓ I am firmly the bottleneck this week because I have not yet been successful in getting music published onto SoundCloud, will then actually publish it on to the the Spotify and Apple Music platforms. and so I'm you know just perfect example of you put a human in the loop you immediately create a bottleneck. And so over the last few weeks I created several agents that were able, just from my you know ⁓ very simple instruction I would say and research some aspect of music. It would then generate a persona, it would generate an idea for an album. it would then build out the story and actually write the lyrics, ⁓ then write the music prompt Suno. It would actually log into Suno and it would go and generate the songs and it would then build the album art, ⁓ master the music, which means it basically runs some sound engineering principles make sure that ⁓ the loudness and the levels are all the same across the album. it'll then go on and build a small website for the album and actually do the marketing. The final bit is for me to get it onto a distribution platform so you can all hear it. and that bit right now, there are some limitations. so I will probably move to a new platform. but ⁓ you know there's a bunch of albums just sitting there right now. They're really good. and I can't get people listen to it because I have to sit down behind the computer and do something. And so a really good test, a really good expression of the impact of a human in the loop. And you know, the machine operates at machine speed. We operate at human speed. the one ⁓ is and minutes, the other is well, you know, however long it takes us to get there, and that might be minutes, hours, weeks, or months, or never. And so so this is really the problem that we're trying to overcome and why, you know, I think really the future is fully autonomous AI businesses that are fully AI native, with all of the potential for going wrong that is associated with that. Emile: And on that note. Ralph: Yeah. Emile: Yeah. Ralph: dear. So well you know, actually we're very bullish on AI on the show, so even some the stories obviously ⁓ are giving us cause to pause and think. Emile: aside, I have heard your music. It's brilliant. I can't wait to see you share that out into the worldwide web. ⁓ Ralph: Yes, ⁓ it's coming and ⁓ no doubt I will overcome my workload for my client projects. so doing very interesting work on the retail side. Obviously with this Nextite technology. Nextite has an tag that basically allows you to track all of the in a store and give you the same kind of information that you would about how a consumer ⁓ you know moves through a website ⁓ to actually have this in the physical world is leading to some amazing insights and I've been doing some wonderful work in London with an excite client to really look at how they are gonna unlock this technology and go beyond the very successful pilot that we've been running there. and I guess it's just a reflection for me. You know, I'm I wanted to be at the the the sharp end of what is going on ⁓ AI and actually getting the opportunity to deploy these types of solutions, blending hardware and AI, is ⁓ really exciting for me. but ⁓ it does rather mean that you know, these let's just call them private projects ⁓ are sometimes on the burner. Emile: Yeah, fair enough. Ralph: But no doubt this week you know we'll release something and I will looking for a massive jump in the SoundCloud So since I started the project, I had something like eighteen thousand listens ⁓ of my my my music, which is really exciting. my tally still sits, I think this week at about seventeen dollars earned. So it's gone up a little bit, but I'm not quite giving up the day job just yet. Emile: Awesome stuff, Rolf. Ralph: Well, I think we've covered the edge thoroughly this week, Emil. So thanks very much for delighting us with your stories. Emile: and yours and catch you next week. Ralph: See you on the edge.