Latest / The Bleeding Edge Podcast / Weekly AI Brief Sep 16 2026
Transcript
- Ralph: Hi, and welcome to the Bleeding Edge Podcast. My name is Ralph Benker, and with me in the studio, as ever, Emile welcome to the Edge. Emile: Great to be back, like the energy, Ralph Ralph: So it's been a few weeks, not because we haven't been recording, but because our recordings have failed. we've had internet issues and performance issues and all the kind of issues that podcasters around the world experience regularly. If you want to start a podcast, you need to have the patience of a saint and probably a whole bunch of episodes in the bag so that unlike us. You can keep publishing every week. But we're back and we're committed as ever to bring you the AI news. And there are lots of wonderful stories, so let's dive straight into it. I got a bit of a news roundup Emile so I'll kick off with that, and then there's a couple of deep dive stories that I'd like to get into. and so no doubt you've got some items to delight us as well. Emile: Yeah. Ralph: but Without further ado, let's just jump into those headline stories. So the big one over the last few days has obviously been NVIDIA that has spent the week buying up the rest of the stack. So NVIDIA obviously makers of some of our favorite GPUs and hardware that actually runs the AI era, but what they've actually gone and done. this week is they have bought Hugging Face. Hugging Face is obviously the platform that distributes open source and open weight models and that apparently is a twelve point nine billion dollar deal which is pretty good money considering you know Hugging Face is essentially a catalogue of AI models and AI tools that you can download. so they've done really well out of that deal, I think. and so news of all of this has actually caused their shares to jump eight point seven percent. and a share will now cost you two hundred and twenty-seven dollars and eighty-four cents. that was as per this news scan running. and basically one of the things that's happened there is they're estimates of a they beat their estimates of a estimated 45 percent revenue jump and so obviously the market really likes that and 8.7 percent on their share valuation so big news there a couple of other things they're also doing is they have funded a political action committee so they are going to be lobbying Washington, I'm not entirely sure what this means, but they filed paperwork for an employee-funded political action committee, which is essentially a PAC and that is obviously a lobbying movement. So exactly what they're gonna lobby for, we're gonna do some investigation. but what's really interesting about this news story is just the full stack nature of where Nvidia's going. So they obviously have the hardware. they also have open weights and open source models, so they're Nimotron family. and in addition to that, they now have the distribution layer in the form of Hugging Face. And so this is truly one of the most integrated full stack AI companies that we have out there. So it's it's a pretty big deal, and I think you know, no doubt Nvidia is positioning themselves as the future. of AI infrastructure. So in other news, AWS and NVIDIA commit to another two million additional GPUs. There's some next generation infrastructure going in there. So this is the capacity that has to be physically constructed. So powered and cooled somewhere in the next 18 months. Obviously we've been tracking some of the news items around the sentiment towards data centers, whether data centers can actually pay off and the big story around that is that data centers are really defining I think the next era of energy generation. So we are talking about sub gigahertz nuclear energy plants to actually power some of these things. I think it's going to be driving a whole new era of small-scale modular nuclear energy, which I I think I I think personally is quite exciting actually. and there has been some interesting advances to what they call clean nuclear energy. You know, just exactly how clean it is, how safe it is, we will have to take a look at. But it's still been one of the most efficient energy production you know technologies of our time. despite fact that Germany has basically closed all of its nuclear plants and France has something like 21. So then we also have perplexity in the news so they've actually shipped their new portable computer and it's a local agent harness and it's kind of interesting that perplexity is not a tool that I've been using a huge amount more recently it's obviously a favorite amongst a lot of people just because it's a great research tool. and a really effective everyday model for for basically you know using it as a chat agent. but this starts to really build on their ability to drive a bit more kind of let's say agentic AI because obviously it basically can do lots of things on your computer, which means it can automate your work. so I think it's definitely one for us to try. The other one that I'm really keen to try, and hopefully you're gonna cover this. I don't know if you've got the story, Emile, but you're not one to miss a beat on what's going on with Grok and Grokbot. I think we mentioned it in the show a couple of weeks back, but it's definitely become the favorite personal AI assistant of so many out there. and for a mere $200, you too can own a piece of the pie. and Grockbot is definitely on the list for us to give it a go at the show. Emile: My main story actually. Ralph: perfect. Well, we'll shut up then and let you get on with it. Emile: then. Video story is super exciting. with one small reservation just because I'm protective on open products I'd say, you know, if any company is gonna buy a company like Hugging Face because it's always been free, I think they in the best position to do so. because they seem to have a very determined path to keep models open and accessible to audiences without you know trying to pigeonhole us into one particular model. in terms of context because we know that SpaceX IPO went for a gazillion dollars. And just as relative context, if you think about when Cursor was bought out, they were bought out at a 15x multiple. Now Nvidia has bought out Hugging Face at an 80x multiple. So let that sink in. and they certainly are positioning themselves to get more into the AR market. This gives them distribution via Hugging Face. And in addition to that, the harness via strategic play with perplexity is going to add to that stack. So NVIDIA now becoming a full stack business model, they don't only want to sell picks and shovels. it's exciting space, and I look forward to seeing where that's going. I will move on to my first story, which is all about automation. There's an interesting new development from XAI, and I think it points towards a much bigger change in how we're going to work with AI. It's called Grokbot, as I'm sure you're all familiar with. And what's important here is less to do with the word grok and more to do with bot. Because it isn't just another chatbot sitting in the window or waiting for you to ask questions. AI is describing Grokbot as always on AI teammate that you give it a job, it gets access to its own persistent cloud computer. This is quite an important aspect to it, signs into the applications and website it needs, and then actually works through the task, potentially while you're off doing something completely different. So imagine saying something along the lines of every morning, check my inbox calendar and project updates, work out what actually needs my attention. Prepare everything you can and only interrupt me when there's a decision I need to really make. That's much closer to the idea, and it gets more interesting because you don't necessarily have to stop at just one bot. XAI says people can run multiple bots simultaneously, perhaps one handling research, another dealing with operations, another looking after engineering, and the bots can communicate and hand work between one another. You can even have something resembling an AI chief of staff I know you're doing good work on this front Ralph and that starts to feel considerably less like using software and considerably more like managing a tiny digital workforce. There's another feature I think that's particularly important. You can teach Grockbot repeatable processes. Complete a task successfully, turn that process into a reusable skill. then creates a routine that runs automatically on a schedule or in some cases when an event occurs. XAI is also gradually introducing a demonstration feature where you can perform a browser workflow and have that bot turn what it watched into a draft skill. I think this is pretty exciting stuff. Of course this immediately raises the obvious question how much control are we comfortable in giving these things? XAI has an approval mechanism for consequential actions such as sending messages, making purchases, deleting information, And that distinction is going to become increasingly important. An AI that can suggest something that is useful, an AI that can actually click the button and make it happen is fundamentally a very different category of technology. Grok Bot launched early last week initially for Super Groc Heavy. Cursor Ultra and Cursor Teams, premium subscribers, with support including desktop and iPhone. But the bigger story isn't really whether GrokBot itself wins. It's that AI companies are rapidly moving beyond the question, how intelligent is this model towards far more consequential one, which is how much useful work can I hand over and come back later to find it actually finished for me. And if that transition works, the next productivity revolution may not come from giving everybody a better chatbot or a better model. It might just come from giving everybody a better team. I will pick up on more automation later because OpenAI is doing something similar, but back over to you, Ralph Ralph: Yeah, I think this is definitely the evolution of what's coming next. So chatbots have been great at disseminating the world's information. And obviously, you know, we've been doing our research and answering a lot of questions, searching, and chat has really kind of changed the game in terms of how we gather and browse information. however it's not Being great at actually completing work. So maybe you can prepare a presentation. you can obviously do some deep research in something and publish it. you can get it to crunch spreadsheet and come up with a better model within the spreadsheet. but actually if we look at what Grok and Grok Bot are doing, and you know what we saw with basically the first iteration of Claude Bot Hermes agent. these things are really designed to do work. And obviously this is nothing new what Grokbot is really bringing. you know, this capability he's probably been around for about eight months since we saw Claude Bot and and Hermes agent really being deployed in anger. But Grokbot seems to be the most capable out of the box. So the easiest to set up, the easiest to actually get it to complete work. And it seems to be reasonably accurate as well. So I have been looking at a number of heavy users and some of the feedback. And obviously this thing is probably performs best, you know, when you initially set it up as a personal assistant. And I think this is where everyday people are actually going to get the most benefit. So you don't have to sit there and think about a sophisticated and complicated set of automations. It can be straightforward stuff like organizing your email and generating morning briefings from your email. and I have to say I've recently been pushing the boundaries a little bit in terms of actually getting AI to assist me with a number of things related to tax submissions and just preparing for year end and and just doing those kind of like you know, laborious tasks of managing expenses and invoicing and I have to say security concerns aside Claude has been absolutely phenomenal at helping me with financial reconciliation so hands down this thing has actually been able to save me days worth of work every month. and I'll just give you a bit of an example. You know, one of the the things that has been a bit of a struggle for me is when I receive international payments, often there are hidden charges, and that means that the invoice amount and the paid amount don't actually reconcile. And within the particular platform that I use for my business, it means that those payments and invoices don't automatically reconcile, which Then leaves me with a manual task of having to do that. Anyway, I kind of got sick of all of that, pointed Claude at it, and it actually identified every single scenario that has basically caused those invoices and those payments not to reconcile. And it then did an automatic reconciliation on the back of that. And it's created a ready-made file that I can now hand over to my accountant and say, this is the invoice that matches that payment. and honestly that's been such a game changer because it's been one of the things that's actually been holding me back from submitting some important paperwork. And it's been so effective at doing all of that that I'm actually now looking at moving away from one of my accounting platforms because I've trained Claude on how to manage all of that. and I think one of the things that we can do here is you know really deploy. Grockbot and these type of agents to to do that kind of work. Really just searching through every single email as it relates to expenses, you know, building invoices, actually reconciling payments, and just really making sure that we are on top of our admin. And just in terms of a practical use case, this has been the one that I've been so excited about. Obviously, if you have research and if you have kind of repeatable admin, activities and those are all great targets to actually point Grokbot at. but as I say, I've been so impressed with how Claude has has run this and some of the agents that I've built just to automate that process. but I'm definitely looking at using Grokbot in a similar way. and so unfortunately it means that my test allocation that I've been pushing towards Kimi over the last months. Probably gonna come to an end, Kimi.ai. yep, I'm going to be shifting that spend over to probably Supergroc and just see how that experiment go goes for the next couple of months. I've been using Kimi heavily for deep research, but I think it's time to involve GrocBot in the process. I think they're running at something around $30 a month to get access to Supergroc. a little bit less if you actually take the yearly subscription. Emile: Three hundred, I think. Ralph: 300 for the yearly yeah that makes sense so at thirty a month it's like three hundred and sixty dollars per year Emile: Mont monthly, I think. Ralph: monthly so I've got here supergroc it says upgrade to super groc thirty dollars a month Emile: wow, they must have changed their pricing. That's phenomenal. That sounds like a good deal. Ralph: But get grokbot with super grok is what it says. let's have a look at annual. annual Emile: Yeah, that's been dropped for sure. Ralph: is twenty five dollars a month. I have been seeing a lot of people talk about the two hundred a month. So Emile: think that's something to do with cursor. it's slashed down. So yeah, get get it while it's hot Ralph: Well I I I think there's a couple of different plans. I'm just on Grok dot com and as I say it's advertising Super Groc at thirty dollars a month if you pay monthly and twenty five dollars a month if you pay yearly. but I I think there are some other plans as well, so we'll investigate that a bit deeper. but Emile a question for you. So what use case are you gonna put Grok bot through? Emile: Everything and anything. you know I've been talking about it for about the last two, three weeks. but I was very cognizant of the pricing. I did see something on X mentioning a discount. I didn't appreciate was that significant. So I think that's gonna substantially change my strategic decisions as well as yours. but I just wanted to add, in terms of taxes, that's very welcomed and well-timed. news because I decided as of yesterday to pull the trigger. having spoken to a good friend and a successful lawyer, he's moved everything for his taxes. he's got quite a complicated tax system with an advanced portfolio and he said it's just everything's been seamless. And all he's using is Claude code and using a few connectors such as Clear Books and it's getting everything done. So I'm I'm super keen to try it myself. I'll see how I get on with Claude and regardless either way, I'd love to have a play with Grockbot. And what makes it particularly interesting, you bang on the money, the the big substantial shift is just how easy it is to use because we've had the likes of Hermes and we've had open claw, but one of the biggest issues is it's been very complicated to set up. It's been far from seamless and easy. And there's been security concerns trying to get the harness to set up. And outside of that, even if you had all the right experience to do all of that, just keeping on top of it, the maintenance, all the login admin and friction has been difficult And it seems like Grok Bot has been one of the first iterations that we've heard, at least, all the discourse that I'm hearing on X easy to set up. There's no friction, there's no multiple sign-ins, you just toggle it on. It gets its own virtual machine and all the access points it needs all coexist on that one VM. So, you know, there's no connectivity complexity against all the the the cross tool sections that it needs to use. Ralph: Yeah, I think it's phenomenal in terms of the ease of use and the accessibility. And of course it's able to log into websites on your behalf. and actually, you know, one of the things that I found so incredible is I I've been able to actually use Claude to interrogate tax systems and actually log on to a number of you know, various government portals and and actually it's uncovered multiple different ways that it can save me on various tax liabilities and commitments that I have. And it's even gone back in my history and said, Ralph, here are a couple of things that were incorrectly filed. You know, with you within the six-month window, you can Emile: Right. Ralph: actually challenge these and resubmit them. and there's a substantial saving there. So you know, it's able to interpret and ingest all of the relevant tax law and you can build your files and folders around your specific operation and this is what's just incredible. I mean I I obviously been using tax accounting and various services and they've been okay but this is next level and I've never actually had my accountant proactively come to me with some of these savings. In fact my accountant has caused some of these overspend probably down to my own poor management of some of the administration and last minute requests. but nonetheless, I I think I'm feeling much more comfortable in in managing this and also just dealing with the regulatory compliance. But not only that, ingesting the changes. So we all know that tax law changes every year. and in fact, you know, this is the reason why tax lawyers exist, is so that they can keep you legal and up to date with the latest. But given the fact tools are so capable of ingesting enormous amounts of tax law and you know I gave it four years worth of accounts which it literally analyzed and gave me the results for in about three minutes. so that's you know a real welcome use case I think very practical almost anybody can use it. I think this really changes the The role of accountants, of tax lawyers. I think there's a role for them, there's no doubt. I think if they can actually leverage these tools to make it even easier and safer for everyday people to actually just leverage AI, then that's the benefit. so you know, them essentially training a version of a drop bot to do your local taxes and kind of giving you that access to that model while being able to rub rubber stamp. And do final submissions on your behalf. I think that's just an incredible opportunity. So we'll have to see what others make of it. that kind of brings us to the end of our episode today. It was a bit of a bumper crop probably because we've been dying to talk about all of this and not had a successful recording. so thank you for listening and we will make sure to get the next episode out promptly. So be ready. We'll see you on the air, Emile Thank you so much. Thanks everyone. See Emile: Thanks everyone. See you next week. Ralph: you next week.