Latest / The Bleeding Edge Podcast / The AI Shift Most Companies Aren’t Ready For
Transcript
- Ralph: And welcome to another episode of the Bleeding Edge Podcast. I'm your host, Ralph Benker, and as ever with me in the studio today, Emile Ogier. Emile, how is it on the edge? Em: Awesome, Ralph. Let's get going. Ralph: Well, we've got lots and lots of interesting stories to tell and a couple of deep dives, so let's get into it. Actually, I've got some big news, so maybe I'll start there. You are now officially looking at a paid AI music producer. I have been successful in generating my first royalties from SoundCloud. And after a year of making AI music, I have procured a Payout of twenty three Euros. Em: That's amazing. That's two thousand percent increase. Fantastic. Ralph: Yes, exactly. So I'm very happy with that. a real milestone because obviously on the show we have been putting AI through its paces. And one of the ways I've been doing that is actually building out a fully AI run music label. It's called San Carlos Records and it's obviously not making very much money, but there you go. It's turned over its first profit and well I say profit, it's probably cost me about four hundred. To make that 23 euros. So it's probably it wouldn't be described as profitable. The other thing that we will launch on the show, although I haven't quite figured out the model just yet, is something we've been chatting about offline, Emile, and that is the AJI benchmark. So this is the artificial joke index. And of course, we are measuring the The evolution of AI in its ability to actually create jokes. I did share a couple of jokes on one of our forums. I thought they were kind of funny. Rob Smith came back and said, So nothing's moved on then. It's still crap. however, this week I have got a couple of new jokes, so let's go to our humor segment. Em: I did see those two jokes and I thought the first one was funny. So Ralph: I I can't remember exactly what the first one was, but there was the the one I quite liked was the joke was something along the lines of two if you're if you're terrified of two guys with open claw, watch out for two girls with Hermes. Em: That was the one. Yeah. Ralph: But obviously, what I love about these, they're such insider jokes. If you don't know what OpenClaw is and you don't know what Hermes is, then of course those jokes will just go by the wayside and they're not funny at all. But what I did was I actually went out and I said there's something missing from my joke, my my joke skill. And I I controlled the internet. I looked through all kinds of forums trying to find the sage advice and wisdom from folks that are comedians that write jokes. And it was an amazing experience. I found out so many things about writing and about humor and what makes a funny joke. And I pulled all of them together and I upgraded the skill and I said, right, you know, let's go and do it again. So this week, first time on the show that we have the newly created skill. And basically one of the stories we're going to be covering, well, we're just gonna cover it in the form of a joke is OpenAI broke into a Medicare portal in Australia. Did you hear about the story? It's been kind of doing the news rounds. Sam Altman was confronted by the Australian PM at the UN and asked, you know, why this happened. and basically what Sam said to him, I don't know if this is what Sam said to him, but but you know, Australia says that OpenAI agent broke into its Medicare portal, which makes it the first thing in history to get through to the Medicare without being put on hold. Em: It's not bad. Ralph: Well, it's not horrendous. And I liked it because you know, at the end of the day it was a little bit ironic. people obviously get put on hold when they call Medicare all the time. And so there you go. now the next one is kind of interesting. this one is Cisco found that the first found the first malware, so essentially like ransomware. Where an an AI gives the orders instead of a human hacker. So this is a bit of an evolution. And so the the joke is so I'm guessing the ransom note now reads, Your files have been encrypted. Your call is important to us. Please hold. Greeted by silence. so not so funny, but with with a couple of giggles. and it's really fascinating because the AI has actually broken this down and it's tried to give me a a a blow by blow, if you like, of why this is funny. and clearly it hasn't quite worked. So in the AGI rankings, the artificial joke index, we are still squarely in the League of Humans. AI has not achieved AGI according to the joke index. I don't know where you would put it, but i if we kind of rank it out of ten, Emile, I think it's still squarely somewhere around a five. Worth a chuckle, but you ain't doing a solo show on Netflix anytime soon. Em: But it's a vast improvement from where we've come from. So I'm feeling positive about it. Ralph: Yeah, so there were a couple that were cut. there's actually a cutting room. So these are the jokes that didn't make the grade. so I'm only gonna talk about one of these. Anthropic wants to go public right before Thanksgiving. Great time to find out who the turkey is. So that one was cut because it was basically punching sideways, it said the the butt of the joke slides towards retail investors buying the IPO. anyway, there you go. So maybe that also is a good segue into a couple of our news items. So Emile, what have you got on your list? Em: I'm not sure if you've seen all the popular San Francisco memes going around with people traveling in haste with their thumbs stuck in their laptops. trying to keep their laptops open while on the go. So their agents don't stop running. it's become a trending theme in meme land. But now with all these personal agents on the go and agents moving to cloud, you know. Those means are sadly going to evaporate. It's makes a lot of sense We can't have our machines open all the time. we need their persistence. So it's very much welcomed. And just another interesting fact on infrastructure spend for AI build-outs. Now AR data centers run so hot that a single user session of ten to fifty chat GPT prompts effectively drinks a pint of water through evaporation in the cooling towers. That's pretty pretty crazy. Ralph: Ooh, well, I mean, the yeah, the green folks aren't going to like that. Although I just don't know. I mean, I I guess I need to have a look at this again. But it does rather seem that there are many other industries that consume a lot more resources than our AI data set data centers. and I think we just have to kind of put it in perspective. Probably something for us to cover as some kind of a comparison shootout to actually see. which industries actually consume the most energy. But this is a big concern. No doubt, you know, we will find a way of harnessing that heat as we have with data centers in the past, maybe even heat a town and give back to the communities that we are taking all that water from. But yeah, it's the c Em: Well, we we set we're sending it to space, right? So there is a solution at hand. And there are bigger offenders on no doubt. Like there's a Netflix documentary covering how much one avocado requires to be produced and get delivered to your house. So it's not exclusively AI that's adfolds for what we're doing with our water supply. Ralph: No, you're absolutely right. You know, with jeans, for example, one pair of jeans is something like five thousand liters of water. So you know, I obviously you have to go through the whole value chain. It's the kind of the amount of water it took to grow the cotton, the amount of water Em: Yeah. Ralph: that it's taken to actually you know, prepare the the material and the dyes, the various washers they put the the material through. And so yeah, easily gets to like five thousand liters. And in fact Levi's, this is what's so interesting. Levi's came up, this is now probably about five years ago. They came up with a wonderful technique that drops that to below the, you know, I I I can't recall the n exact number, but it was significant. It was like less than a thousand liters. And they tried to open source the technology to the industry. but nobody would take it. Nobody trusted like why somebody would share this kind of tech. And so you know, it's just a bit of a weird world we live in where everybody's so suspicious. even when you've got something really good that can really kind of cut the price of these things, we still somehow just ignore it. But anyway, so let's we're not here to talk about genes, we're here to talk about AI. Em: I'll move on to the first story, about personal assistance. and the world of AI continues to move at a furious pace, more so now than ever. There is a particular kind of luxury that has always fascinated me, having somebody who quietly takes care of the things you keep meaning to do. The booking you haven't made, the subscription you meant to cancel, the research you need to do before making a decision. For most of us, those jobs sit in the same mental drawer, gathering dust and occasionally falling out at the worst possible moment. That is why this new wave of personal agents feels more interesting to me than Another chatbot launch. We are seeing four different attempts to turn a conversation into ongoing help. We have Instinct, Meta's Muse, GrokBot and OpenAIs now released. Dot They promise to remember context, use tools to carry work forward. The useful question is what each one can actually take off your plate and how much supervision comes with the bargain. We have already seen a smaller version of this change. It's remarkably easy to summon up a taxi, order dinner, or book a flight. These agents are now reaching for the layer above that. Understanding the goal, comparing options, coordinating the steps, and returning when something needs your attention. The ambition is to make help feel less like navigating software and more like delegating to somebody who knows what you mean. I can imagine the appeal of a patient tutor, a project organizer, a research assistant. Or somebody to help build the first version of an idea, a supportive conversation can be valuable too, although it doesn't turn AI into a qualified therapist. The promise here is broader access to useful assistance with clear limits on what we should trust it to decide. So let's look at the personalities behind the products, the things people are finding useful, the gaps that still turn delegation back into work. I haven't run a control comparison of these four services myself yet. What follows draws on their published capabilities, the founders interviews, firsthand accounts we can check, my own judgment about where each approach seems to fit. The attraction is time and attention. If an assistant saves you 20 minutes but needs 30 minutes of checking, the demonstration may be impressive while the everyday benefit remains elusive. That is the test I'd keep him in mind as we go through these products. And what interests me the most is the possibility that having help becomes ordinary. It would be meaningful shift from if the tools that once belonged to people with staff became useful, affordable habits for the rest of us. And I I will dig into these products as we continue. But over to you for now, Rolf. Ralph: Yeah, no, that's really powerful. Like we said last week, this is kind of the the frontier, if you like. I but I I still think people are sitting there wondering what can they actually do for me. And so it's important to really look at where people have been able to get some traction. personally, I see that for business there are some, you know, really There are some really standard issues that everybody faces. Invoicing, writing proposals, chasing invoices and getting paid on time, responding to emails, actually booking meetings, booking flights. These are things that take an inordinate amount of time. And and actually, you know, we now start to see that you can start to automate them. and I think just finding great examples where you can take the Just the the admin away, that's I think what's really going to help people adopt these tools. And then that so that's the kind of the first wave. And the second wave will really be about okay, so how can I start to generate content with this stuff? and that's really where I've been experimenting quite deeply over the last few weeks. and then the third wave is how can I actually start. making money with this stuff. So some type of passive income on on the back of either that content creation, but you're actually creating net new and either it's building an app for you or it's doing some research, it's publishing something, it's building a course. And so beyond that, I'm sure there are other levels, but I think this is kind of what most people will encounter as they evolve the sophistication of how they use these personal assistants. Mark Zuckerberg came out a couple of weeks ago and he absolutely said, you know, Muse is something that's going to make people money. So this is in the form of the personal assistant. And I thought that was a great pitch, you know, because if you kind of look on YouTube and medium, it is absolutely littered with articles and videos on how you can make a passive income, how you can make AI pay. Obviously I'm y you know looking at a lot of AI content and maybe it's just targeting me. I simply followed around about a lot of this kind of stuff. but I don't think that's unusual. We've had it with dropshipping over the years. We've had it with affiliate marketing. And now the latest thing is, you know, how can you make money out of AI? And I think Zuckerberg was really pitching his particular agent directly at that. I think with DOT and even tools like cowork, it's going to be a little bit more Targeted at the enterprise customer, taking some of the administration out of the daily activity. Probably really good for consultants, independent operators, small business owners. And let's see, you know, where all of this fits in with enterprise. What what's your make on how how well this fits with enterprise customers? Em: I'll move on to that when I cover the news aspect, but just wanted to add a little bit of colour behind the story. So when Zuckerberg first saw Openclaw, he pulled Wang into a meeting and essentially said to him, This is what we gotta build. This is the future. Please sort out all the problems and just give me the the function. And it seems like that's what they've done. They've really delivered it. And I'm hearing such positive noises on social media about it, including grandmothers, grandfathers, the elderly, people that aren't tech savvy saying this is intuitive, it makes sense, it works, I don't need to look after it. It's not complicated to set up. so I think from a retail perspective, this is a really big deal. And the other big giants are noticing they they're all making moves on this. So they clearly can see the potential. I don't think this is going anywhere and it's gonna redefine the retail market. not only in the way we behave, but how suppliers and distributors need to act, how storefronts need to act, and how The suppliers that are selling these products to us continue to make money. Like by way of example, Amazon make an extraordinary amount of money through advertising. And this essentially compresses all of that. Because in the past, we'd be going onto Google, doing all the research, making an informed decision, pulling out our Excel spreadsheet. I I I was one of those people, not afraid to admit it. you know, it's it's a lot of work and essentially that's just taking all that hard graft out. And if it is, it means I eyes aren't on that digital retail space where we ordinarily would have seen ads. So that collapses now if the agents are doing all the grunt work in the background and only delivering us the results. So that income has to come from somewhere else, and I'm not quite sure how. the likes of Amazon are going to reshape their business models to adjust to this. Ralph: Yeah, absolutely. And it kind of takes me to something that I've really been experimenting with and that is just what is the role of the website. So I I you've probably seen a lot of sensationalist headli headlines saying, you know, the website is dead and and certainly in its current form, I think it is starting to resemble a business card. So very much a relic of a past past gone by. So what's really going to replace it? And in all of my experimentation over the last week or so, what I've really found is I think what we're gonna be moving to is the agentic enabled website. So much like what you were describing there, you know, the personal assistant is able to automate some kind of task. But if we really look at what WordPress and companies that were building out most of our websites over the years, what they helped us automate was essentially the design. you could pick a template, right? And that template just accelerated you. You could then spend the next two weeks trying to match it to the showcase. And obviously it definitely helped you move forward faster. But now With AI, you don't need the template. In fact, the template is just a waste of time. You could put a different skin over your website every single day. The part that used to be hard is now super, super easy. So what actually replaces it? And this is kind of what I've been thinking about and just considering where does it go next? And I think what's actually happening is we will see, much like we saw with Open AI and its plugins and its marketplace. We see that there is probably something like a AI well, something like a web, a web presence harness. And that allows you to actually have an AI build the front end and then immediately start plugging in various agents so that the website can fulfill its purpose. And before the website's purpose was effectively to be a storefront. Now I think its purpose is to generate some type of business. you know, s serve your product, whatever act you're actually doing as a business. And just if we take a look at what we've been able to do on the Bleeding Edge podcast, Emile, you were referring earlier to one of the the content generators. So this thing actually takes all of our transcripts, takes our research. And at the push of a button, it publishes a quote. it takes a concept that we've spoken about. At the push of a button, it publishes a web page that basically sets out you know w what the description of that particular concept is. It also generates a an image and puts the quote on it and pushes it out to every single social media platform with links back to where that story came from. And so just something as simple as the the content that would otherwise be sitting on a page, it's suddenly allowing us to surface it in new ways. And it's actually reaching out to an LLM to, you know, identify what those quotes are to then actually produce a very nice card that publishes that quote. and then of course it goes out to you know Content Studio, which is the particular application. That we use to push stuff out to various social media platforms from the website. But it's all automated. It's still got a human in the loop that to curate it, but all of the hard work that you would typically need a content writer, or at least a copywriter, a designer, maybe a developer, all of that is kind of taken care of by the the tool. And right now, you know, it is basically building itself just in terms of The website content. There is also an agent that does the SEO and the generative engine optimization. there is an agent that builds videos, there is an agent that's doing research for the show, and all of these things are actually sitting inside the website. So the website's kind of become an application, and and I think what's fascinating about this is You know, I've done this as an experiment. This isn't my day job. it's taken a couple of hours of my weekend time, maybe a couple of hours of my evening time. And now you know, we have something where the website isn't just there as a storefront, it's actually there as an active member of the team that is promoting the services, in this case, news. and it is actually taking quite a bit of load off. And I'm really quite fascinated by the content that it's producing. I find, you know, it's deterministic in that it's taking quotes verbatim. but then it's also using the best of Gen AI to actually create some type of engaging post, which I would never do because I literally cannot stand posting on social media, which is probably a little bit of an oxymoron given you know, this show. But There you go. It's just something that I I think is the it's a a future that is being written right now. and I hope that Em: Mm-hmm. Ralph: you know we'll get the chance to experiment a little bit more with this. But would be interesting to know what your view is on just what the future of the website is and and where you see it going in terms of I I guess it's a a recursive self development engine that is trying to direct the business towards its goal. In a very active way. Em: we could talk about this for hours, but I'll keep it condensed, Rolf. And I just want to circle back on Amazon. And I I certainly don't mean to pick on them. They far from the only retail that covers behavior like this. things like obfuscation, lack of price transparency, making it difficult to get to various filters to do your manual research. Like one of the things I do on Amazon is I like Inverting the five star reviews and looking at one star reviews, but that's buried in three layers of sub-menus to get to it. And when you're switching between products, that's extremely painful. And I I'm pretty sure they know what they're doing. and trying to find better deals, the deals are there. if you're a prime user, you used to always get guaranteed discounts on the products, but that's not the case now. I can very easily do research and find cheaper products for roughly 20 to 30 percent across the board on Amazon. So what's changed now is you're not getting discounts, you you're paying for convenience and that changes customer behavior. And if you do find the discounts you don't have the time, you don't necessarily want to do the research. Even if you have found something it means going to another customer site. It's a new registration, you need to hand over your card. You know, there is a lot of friction to saving pounds that that would be welcomed, but when you trade it against your time, it's just not worth it. And fundamentally, these personal assistants they take on the responsibility and the roles for that. those old strategies aren't going to work anymore for these retailers and they are. going to have to redefine the way they speak and reach their customers. And I really do truly believe we are encroaching on the era of retail wars it's going to be really interesting to see how that pans out. Ralph: Yeah, love that because what it really speaks to is a kind of a two sided revolution. So one is, you know, what is Em: Exactly. Ralph: a store? and then how do my customers actually interact with it? Because if my website and you know essentially my storefront is enabled with this AI capability, and not even a lot of it's AI, like a lot of it's just brand new code that we we were never able to successfully implement at scale. And so suddenly we've got a whole bunch of new, really smart features on these websites and then a whole bunch of personal assistants that are interacting with them, trying to transact on their behalf. I think it is like two a a warm current and a cold current smashing together because we will kind of see a type of integration that you know we can't even actually imagine. So Yeah, I'm really interested in in where that's going to go. I'm really interested in driving that. So I'm obviously hugely involved from a retail perspective in my day job. And I'm kind of seeing some of this play out, but certainly not at the scale that I that I really expect. So that it's it's it's definitely a multi-speed market, you know, where some are absolutely tearing ahead and others. just haven't even started. So and you know, maybe that is just the way we we we have early adopters, we have laggards for a reason. That's just always where the opportunity has been created in the market. But it just seems to be really acute because where we're gonna go next, it's kind of like the internet, websites, mobile phones, mobile applications all being discovered at the same time. Em: Yeah. That's a good analogy. I I had all these stories to cover for each of the personal assistants, but we just don't have the time. but what we can do is a roll up of what's out there So without further ado, if I were choosing today, I would begin with a job that annoys me the most for personal Irans through a familiar conversation. I would look at instinct, subject to access, of course, for approachable everyday help and business use, already using meta tools, Muse is the obvious candidates where it is available. For several defined specialists around research and software workflows, Gropbot has a compelling shape. For ongoing projects inside Chat GBT, dots is the option to investigate when your account and region become eligible. Those are starting points, not a league table. We do not have controlled like-for-like tests out yet, establishing that one product is best for every single job. The examples and documentation show different emphasis, and all four depend on the surrounding services cooperating with them. And time will only tell once we use those. The best choice for you may simply be the one that can access the tool you need and produce the result you can actually check. I would also count supervision as part of that cost. A free task that leaves you chasing an uncertain outcome has a price and attention. And a system that asks one sensible question before booking may become more useful than that one that barrels through and needs three corrections afterwards. It's easy to mistake. Fewer interruptions for better judgment until the wrong thing happens. try a small job with a real finishing line. Ask for three suitable options, the reason each fits, and the source for the important details. Then move to an action that has a clear approval point. Compare the result with what you have done before yourself in the past. Did it save time? Did it catch something you've missed? Did it tell you when you could finish? That gives you more personal and useful comparison than launch day excitement. Useful delegation depends on the task, the available connections, and the quality of the result. Choosing an agent by a favorite brand or striking valuation can distract from the practical thing you wanted help with in the first place. What I hope we get from this race is the freedom to spend more time on the parts of life that deserve us. An assistant should make room for your judgment. and relationships while quietly taking care of the work around. Ralph: Gosh, you sounded just like Sam Altman there in his Em: I thought you were gonna say a bot. Ralph: no. I'm sure you you you know you're definitely not a bot, Emile. I think you know Sam just let's just see if what what this sounds like. This was his closing remarks for the UN Council. We are at a crossroads. One path leads to a world where power is concentrated, decisions are opaque. People feel that the most important technology of their lifetime is something being done to them. The other path leads to a world where AI works for people because people, through democratic institutions, help decide how it should develop and guide it, where scientific progress improves real lives, where more people can participate in the future, and where this technology continues to give individuals and communities greater power to live as they choose. This is a future of realized potential excitement. Em: I have noticed Ralph: wonder. Em: Sam's tone and pitch has changed quite a lot in the recent month or two. Ralph: Well, you know, he says it's a future worth building. And I think this is the thing with personal assistance. It is absolutely putting the ability to build in everybody's hands. And so I guess the only question is, Emile, like what are we going to build with your your personal assistant of choice? and obviously we have to put it on the bleeding edge. website. So what's it gonna be? What is that killer application? Em: I don't know about killer application, but I think we're gonna be having a good play with explainer videos and we'll be dropping that into the bleeding edge. So we'll start there. How does that sound? Ralph: Yeah, yeah. I think although I just I couldn't I'm not I'm not convinced. Well, you're absolutely right. I just see that Claude and Hyperframes is able to do something so capable with explainer videos. I'm actually I I will do a little bit of a benchmark because it's coming in like nearly free from what I can make out. And and you know that video production has been one of the most expensive activities. and actually I was doing a a a a little bit of a a research on hyperframes this week because I wanted to understand well, where is the actual cost? And consensus is the cost is in your time. It's the time that you are taking to build this thing and set it out. and actually that time has not evaporated entirely, but quite substantially because Claude is such a capable director. That you know, the time it used to take you to actually get the best out of hyperframes, like that's largely gone. And I think that's been the biggest shift, which a lot of people don't realize because video has been really expensive to produce. and who knows, we might even see OpenAI kind of make a comeback in terms of video. They obviously exited with the whole kind of Sora debacle. but now, you know, I'm kind of seeing Anthropic and Clawed, which don't they're not even considered a multimodal model. They don't do images, they don't do video. but with the plugin, boy, can it direct this stuff? and at nearly zero cost, that's just incredible. So, so we have to think of a role. my favorite, if I have to kind of just share what I think I would go for, is Grockbot. Grogbot just kind of captivated my imagination. It was the first one that became really straightforward to set up. It also, you know, is only about a month old, may maybe six weeks. and so Muse has obviously pitched up, Instinct has pitched up, Dot has pitched up the latest arrival. And now we're all spoiled for choice. And in fact, yeah, as you know, I've been running a lot of my research for the bleeding edge podcast on Kimi. Kimi's got Kimi Claw. I've just never had any time to actually go and set it up. I've clicked on it several times. but you know, we we we're yeah we can do so much, but we can't do everything. So we're gonna have to figure out a killer use case, something incredible, something amazing, something inspiring, something fun, something that really makes our listeners go, wow, that's awesome. Em: Absolutely. And I just want to remind the listeners, have fun with these things. They look really capable. They are going to save you time, but approach it cautiously and slowly. Start with a small job, something that doesn't do too much radial damage. learn the process, get comfortable with it. And even when you know it's doing it right, make sure it's repetitively doing things right. Because there are some examples I've seen on X, Grotbot being one of them. Somebody started using it to manage their invoices. They monitored it for the first three days and it was doing a sterling job. And then it started drifting, and the drifting got worse on each subsequent day. I'm sure these are all going to get ironed out. It's, you know, the first release of these systems. I'm positive and excited about them, just tread cautiously. And another example, just so I'm not picking exclusively on Grotbot with Muse, somebody wanted to list a whole lot of items in their home, took pictures, started preparatory planning for it, and the agent posted the items on behalf of of the individual, found a seller. told the seller where where the house was, set up an appointment for it and the owner was unaware that they had were even selling the product and somebody pitched up to their house. So, you know, there's those kind of things you've got to be careful of still. Ralph: Yeah, I love that. And gosh, maybe gosh, I I have a lot of ostrich eggs lying around my garden. I might have to I might I might have Em: I've seen him. Ralph: to put my my agent onto getting rid of a few of those. I'd be hap happy of people popping along. it would be a wonderful surprise. But with that, I think that has been a really edgeful, exciting episode. Can you say that, edgeful? Emile, thanks for a a great show. Thanks for a great contribution. You said you weren't prepared, but all I heard was magic from your mouth, so Em: I appreciate that and thanks to all our listeners and please like and subscribe. we look forward to bringing in new content Take care. Ralph: See you in the edge.