← Back to search
Microsoft's Hybrid AI Windows Event and the Surface Laptop Ultra
Double Tap · 2026-10-08 · 59 min
Show full episode description
Microsoft's latest Windows event went all in on hybrid intelligence, which means AI that runs on your own PC when it can and in the cloud when it has to. Steven Scott and Shaun Preece break down what the agentic OS, a smarter Windows search and the new Copilot could mean for blind and screen reader users. They also look at the $2,599 Surface Laptop Ultra. Windows will decide where each AI task runs, using small local models or the cloud depending on the job and your privacy. Microsoft named DeepSeek, Perplexity, Hermes Agent and OpenClaw on stage, and showed a guided setup for agents that needs no terminal or config files. Steven and Shaun explain agents in plain terms: picture a thermostat that keeps working toward your goal without waiting to be told. The new search lets you type what you want, such as changing the volume or a setting. That raises an old question: why are Windows settings still so hard to read with a screen reader? The new Copilot can work with your files on your machine, and Autopilot chains tasks together, like bundling tax documents and drafting the email to your accountant. On the hardware side, the 15-inch Surface Laptop Ultra runs on NVIDIA RTX Spark with up to 128GB of unified memory. Along with the Surface Dev Box, it starts a wider conversation about data centres, more efficient models and why private, on-device AI matters for NDAs, transcription and business data. Steven's tip is to ask your AI assistant which local models your computer can actually run. The hosts also ask whether leaning on AI costs us skills, and remember the time an accidental prompt injection deleted 21 minutes of a show. Chapters
✨ Episode Outline — click any point to jump to it in the episode
Problem solved
Making sense of Microsoft's Windows/Surface event: hybrid intelligence, the agentic OS, local models, and what it means for blind users.
Benefits
- Plain-English explanation of hybrid intelligence (local vs cloud orchestration)
- Clear definition of an agentic OS using thermostat and Echo analogies
- Windows will pick and install the right local model for you
- Coverage of Surface Laptop Ultra and NVIDIA RTX Spark hardware
- Accessibility angle on new Windows search, settings and Copilot
Use cases
- On-stage demo: laptop in airplane mode coding fully on-device
- Agents resizing photos locally while cloud agents write an app
- Thermostat example: agent keeps house warm without repeated commands
- Windows quantizing models like DeepSeek, Hermes and OpenClaw to run locally
Tools / build
- Windows hybrid intelligence
- Copilot and Autopilot
- Surface Laptop Ultra
- NVIDIA RTX Spark
- Copilot+ PCs and Dev Box
📑 Chapters — tap a time to jump there
0:00
Intro
- Double Tap hosts Steven Scott and Sean Preece react to Microsoft's event
0:56
Microsoft's Windows event and hybrid intelligence
- Hybrid intelligence: use local hardware when possible, cloud when needed
- Windows orchestrates which tasks run locally vs in the cloud
- Microsoft name-checked Perplexity, DeepSeek, Hermes and OpenClaw as quantizable local models
9:28
What an agentic OS really means
- Agentic OS: send agents off to achieve goals unsupervised
- Thermostat analogy: agent maintains the goal, not just one command
- Skepticism: Copilot dark-mode promises never shipped; MCP talk is old
16:01
New Windows search and settings accessibility
- New Windows search and settings features from an accessibility viewpoint
21:18
The new Copilot and Autopilot
- What the new Copilot and Autopilot are meant to do
30:47
Surface Laptop Ultra and NVIDIA RTX Spark
- Surface Laptop Ultra hardware and NVIDIA RTX Spark
37:31
Which local AI models can your computer run?
- Which local AI models a typical PC can realistically run
41:40
Coming up: the NASA accessibility webinar
- Preview of the upcoming NASA accessibility webinar
43:15
Copilot+ PCs, the Dev Box and data centres
- Copilot+ PCs, the Dev Box and Microsoft's data centre plans
[SPEAKER_07] This is the story of the one. As a shipping manager, he's running the dock, stocking supplies, and making sure every order goes out on time. And Grainger helps him stay prepared. From shipping supplies, to PPE, to hand tools, he counts on Grainger to keep products moving and trucks rolling. So his operation runs smoothly, no matter the shift. Call 1-800-GRAINGER, click grainger.com, or just stop by Grainger for the ones who get it done. [SPEAKER_00] Hey y'all, Kelly Clarkson here with Wayfair. I know holiday magic starts at home, so I tend to go a little overboard. The twinkling lights, the trees, the cozy blankets, but the thing I love even more than decorating is having everyone over and knowing my home is ready. With Wayfair, it's easier than ever to make holiday magic at home. Ooh, I think I found room for one more wreath. Don't wait, head to Wayfair.com and get your home holiday ready today. [SPEAKER_03] Well, Microsoft just held their Surface event and I'm confused. This is Double Tap, where blind people talk tech. Here's Steven Scott and Sean Preece. Yeah, welcome to it. Shaun Preece, ready to do this a whole hour on the Surface event? Are you excited? [SPEAKER_04] Yes, of course. And I know exactly what happened. And firstly, it's not the Surface event, Steven. Stop that. It wasn't the... Yeah, I'm not admitting anything. I didn't doze off in the middle of it. Yes, you did. No, no, no. [SPEAKER_03] I tried talking to you and there was nothing. [SPEAKER_04] I was resting my eyes. [SPEAKER_03] I asked, be my AI, what is Sean doing on screen? And it said, it looks like Sean has gone to sleep. [SPEAKER_04] Yeah, okay. So, well, see, a Surface event. I'm all in. I'm excited. Ooh, shiny, shiny toys. But, okay, so this was the Windows event more, really. There were some devices we'll get to, but Windows. And it kicked off with Enterprise, with lots of terms that I have no idea what they meant. So that wasn't a great start for me. [SPEAKER_03] Yeah, I must admit, a lot of this event I just spent going, um, okay, sounds great. I have no idea what any of this is or means. um, we're going to try and break it down for you best we can. Ultimately, I think what I got out of all of that was that ultimately what they are announcing today, or yesterday as it is now, um, is essentially a way of something we've talked about for a long time. [SPEAKER_03] Using local hardware when it can be used and using the cloud when needed. Um, they call it hybrid intelligence. Hi! Um, and that is what essentially we're going to be talking about. [SPEAKER_04] That's what we got out of it. [SPEAKER_03] That's it. That's what we got out of it. Yeah, that's it. So, yeah, good luck with that. Thank you. We'll catch you again next time for another great double tap. No, I mean, look, essentially, this is something we've talked about, right? The power of local machines versus the cloud. Yes. And it's getting to a stage, from my understanding anyway, of what I was hearing there, that it seems to me that they're getting to the ability to be able to really shrink down these models so they can run on more powerful hardware. [SPEAKER_03] Of course, we're not talking about a five-year-old Dell laptop here. We're going to be talking about something powerful, something new, and we're going to talk about the new hardware. But, you know, it does seem as if they're trying to get to a point where they can start to rely less on data centers. Hmm, I can't think why. [SPEAKER_04] Well, see, that was the interesting thing. Basically, the whole motto of this was, it's more reinforcing that Windows is ready for AI. It was all the agentic OS, right? Windows is capable for running those local models without you having to install third-party things. This is going to be built in. And as part of that, it's going to be able to, what's the key word when it comes to AI? Orchestrate. That's the thing. [SPEAKER_04] So it decides, okay, with that task, we can run that locally. No need to farm out to, you know, your chat GPT-6 or whatever it is, run it out in the cloud. We can run that locally. And for other tasks, it says, well, yeah, that's going to struggle. Let's run that out to the cloud. Or maybe there's a privacy concern where it stays on the computer. All that is going to be built in to Windows, which is actually a huge deal, right? It did send me to sleep, partly. [SPEAKER_04] But when you break it down to that, it does mean that, hey, AI is here to stay, and it's going to be a major part of the next Windows. Windows and AI agents seems to be the thing. Because you talked about, you know, making the models smaller, and that was something that jumped out as well. The amount of talk about all the models that are out there, they mentioned perplexity, they mentioned DeepSeek, Hermes, and OpenClaw, [SPEAKER_04] all of those models that usually you would think they would sort of push away a little bit and just concentrate on co-pilot. But they were really bringing them in, and they were talking about, you know, we can now, what's the term, quantize these models to get them really small to run on, you know, your local PC, and Windows will know exactly when and where to use them. That was actually really cool. [SPEAKER_03] Yeah, I like that. I really like that a lot. And, you know, we've talked about this before, the challenge of, you know, continually using data centers, the continuing need for data centers. It doesn't sound like that need is going away anytime soon, to be honest. But, you know, if we can build better hardware and get that into the hands of people at a good price, that's not going to happen immediately. But when it does happen, it does mean that we're going to be able to do a lot of this AI on device itself. [SPEAKER_03] We know at the moment, for example, with our smartphones, that's limited to some degree. Don't have a go at Siri again. Leave it alone for one episode. Well, there's no Windows phone to shout at, so I guess I'll just shout at Siri. Any excuse. But no, I mean, it isn't powerful enough. Even Gemini is not powerful enough to run locally for everything yet, right? And nothing else is. There are certain tasks. We've talked about this before, very specific things that AI can do locally. And that's a good way to think about it. [SPEAKER_03] But the challenge for us as consumers is knowing the right choice, knowing what to do. Like people will say, oh, have you downloaded this model? Or have you tried this? And I'm like, I don't know. Or, you know, oh, I've tried locally. It doesn't work for me. Oh, well, you used the wrong model. Yeah. And that's a big problem, right? Because people who are dipping their toes into this and who are trying to understand, okay, so what is it I need? What is the product essentially that I need to do the task? [SPEAKER_03] You know, that's where this can be helpful because this is going to know the best tool and point you towards it. And I guess in the background, go off and install whatever it needs as well. So it knows what it needs. It knows what it can do. It knows what it can do locally. So, for example, if you're writing an email, I imagine it can do more on device with that than perhaps it could do sending it to the cloud. And why? What's the need for that? I mean, they were even talking about coding. [SPEAKER_03] And one of the examples given was where someone actually, the girl on stage, put her laptop into airplane mode. So to prove the point, this is all running on device. This is all running here. You know, what that means is the nerds are going to be much quieter at airports because they'll all be sitting on their laptops. You know, you don't have to worry about Wi-Fi or anything like that. No, I have no Wi-Fi. It doesn't matter now. I can just turn off my airplane mode and I can get on with my coding and leave everyone else sitting quietly. [SPEAKER_03] But it is exciting to see where this is going. It does feel, and I have to say, a lot of what they announced and what they talked about felt a bit vague, as often these things do with Microsoft. Because you said, you know, it's coming in the next Windows. There isn't really a next Windows. I keep waiting to hear about, like, Windows 12. And it never seems to happen. There never seems to be a talk of a Windows 12. It's just this continual iteration. It's as if Windows 11 is just this rolling thing now. But we'll just get these features as they come along. [SPEAKER_04] That was the idea. It was meant to be. This is the final Windows. Now it's just going to be updates. Update, you know, Tuesday. It's just going to evolve. There's not going to be, hey, now it's time for Windows 18 or whatever. [SPEAKER_03] Yeah. [SPEAKER_04] So that was the idea. And I don't know how practical that actually is. Once the core code starts to get to a certain level, isn't it easier just to say, right, it's a fresh one now. But we'll see. But, yeah, you say it was vague. I'm not sure it was vague because they did bring out all the terms there. Hybrid intelligence and MXCs and I don't know, whatever else. FinOps and SynOps and whatever else. Yes. [SPEAKER_04] Lolly Ops. Pops. Yeah. There was so much I didn't understand. But to me, the main problem is that we've kind of heard this before. Now, this may be because, you know, they've been sort of in blog posts and things. They've been talking about Windows becoming the agentic OS operating system. And, you know, the MCP, the ability that AIs can just sort of plug in as a module and control, you know, aspects of Windows. [SPEAKER_04] They've been talking about that for a long time. And, you know, as we often say, it's almost, okay, I'm tired of hearing about it. I actually want a product or I want something that actually shows what it can do. I still remember Copilot right at the very beginning and how excited we got about, well, you can tell it to turn on dark mode or high contrast mode. Never really worked, though, but it was coming soon. And how many years ago was that? You know? [SPEAKER_04] It's fine to be, to talk about this and be excited about it. But it seems to be, okay, now just deliver on these promises almost. But the amount of partners they had there, you know, when they were bringing on all the different companies and the thanks they kept giving out to the partners. More talking about the hardware side of it. But the very fact they were mentioning all these AI models out there, like DeepSeq, I think does show. I think we moved along quite far here and something will actually appear. [SPEAKER_03] Yeah, well, there were some tangible examples, real examples of how this will be put into action. I think for people who don't follow this closely and who wonder what we're talking about when we talk about agentic OS. I mean, tell me if I'm wrong here, Sean, but I think the best example of this is the Amazon Echo. That's agentic OS, isn't it? I mean, essentially you're asking it something and it's doing something. So let's say you have a smart thermostat and you've asked it to turn on the heating. That's agentic OS, isn't it? I mean, it's going off, it's turning on your heating. [SPEAKER_03] You don't have to go to an app, you don't have to turn something on. It's doing it for you. It's being the switch on. It's being the finger that says on. [SPEAKER_04] Well, it is. It's an aspect of it. But I think the key thing when it comes to agents and agentic is the very fact that you send it to do something and you don't have to supervise it and it carries on doing it in the background. You don't have to wait around. It just goes off and does it. Now, the latest ones and OpenAI just revealed their dots, as they call them, their agents. They're very much, it's kind of changed from go and do A, B and C. [SPEAKER_04] It's now I want to achieve this goal. Go off, figure out how to do it. And then don't stop working when I've gone to bed or whatever, or even when the computer turns off if it's running in the cloud. Don't stop working until you've completed that task. It goes off and it's its own agent. It's got its own agency. So I think that's where the term comes from rather than us supervising it all the time. So when they say agent. [SPEAKER_03] So in essence, if we're using the Amazon Echo and the thermostat examples, I'm trying to keep this on the level here so people can follow it. [SPEAKER_04] Yes. [SPEAKER_03] Essentially, it's gone from me asking it to set the thermostat to 19 degrees and it now saying, well, actually, the house has reached 18 degrees, but I know you still want it warm. So I'll maybe set it to 18 now for you. And actually, as the temperature dips, I'll bring it back up to 19. It's got that kind of brain behind it to know kind of what you aim to achieve and it's going to work out how to do that. [SPEAKER_04] Yeah, that's a good example. Yeah, exactly. And it's constantly running and doing that. It's not, oh, he's gone out now, so I'll just stop and have a cup of tea. It's constantly. [SPEAKER_03] It's not waiting for a command. Yeah. [SPEAKER_04] Yes, exactly. It doesn't just turn off when you're not there. So you think of, you know, a Gentic OS. So basically, you've got as many agents as you want and you just send them off to do things, you know, so you don't have to have 7,000 file explorers open and browsers and whatever else. You've got these agents working in the background. [SPEAKER_04] And again, tying it back to the hybrid intelligence, you know, some of them are running just locally on the machine, maybe going through your photos and resizing them while others are, you know, trying to write an app for you which are out in the cloud. So it sounds very cool. I just want it now. Yeah, no, I agree. [SPEAKER_03] And it does feel more realistic because we've seen it in other ways. I mean, we've heard about OpenClaw. We've heard about HermesAge. Yes. We've heard about all these other ones. And, you know, we've seen it to some degree with assistants where we're starting to see it. I mean, I joke around a lot with Siri, but you're starting to see, you know, that capability coming in where, you know, you are able to say, okay, go through my, I don't know, my events that are coming up in the calendar. [SPEAKER_03] And perhaps, you know, just make sure that, let's say, you know, someone invites you to dinner and by text, you can send or set up a reminder, essentially. Those kind of things, those kind of agentic tasks that you're asking it to do. I think the word agent is the problem for me. I think if you called it an assistant or, you know, or a helper, you know, that kind of thing, because it seems agent just feels the wrong word for this. Seems sneaky. Well, there's so many terms coming. Assistant is more appropriate, right? [SPEAKER_03] Because you're asking it to do something for you. You're asking it to assist you in doing a task. [SPEAKER_04] Yeah, I know. And that, what's different between them is kind of, it just feels more like, is this just tech jargon, just trying to almost confuse things, right? Does it really matter? That's why my nerds. Well, and the point is that most people just want to ask, you know, their technology to do something and for it to be done. That's all we care about. Now, whether that's an assistant that just does a, you know, we invoke it, it does the task, and then it goes back to sleep. [SPEAKER_04] Or it's an agent where we send it off and it goes off all over the web or all over your computer and doesn't stop until it's done what it's done. I don't think it really matters. The point here is, you know, what is Windows going to become and how are we going to interact with it? And isn't going to make it more accessible, of course. That's the cool thing. [SPEAKER_03] Yeah, well, we'll come to that. But I will say that I think this is where the confusion comes for a lot of people. Because we have friends who have been in touch with us regularly. And what's the thing that we always go back to them with? Every time they come to us and say, okay, but how did you do this? Or how did you do that? And we go back and say, just ask. And there's a kind of a hesitance to say, yeah, but how did you do the thing? Because I'm asking and it's saying no. And it's like, yeah, but you've got to ask it again. You've got to ask it, okay, but why not? [SPEAKER_03] Why are you not able to do this? Do you need permission to do it? And, you know, if you do need permission to do it, and I'm happy to give you that permission, then tell me how to give you that permission. And then you can go and do it. You know, whether it be, you know, attaching a file to an email to send, or even just sending the email, rather than asking me to constantly check it. That kind of thing. You know, so it's like, you've got to learn to ask. And I think that's where, if you understood it as an assistant, you'd go on better. [SPEAKER_03] Because then it's an understandable, okay, I'm asking you to do a task, and I want you to go and complete that task. And that's kind of what they're building in here. We've seen this with Claude. We've seen it with others. But they're all cloud-based. They're not what you would call operating system level. Siri, of course, I haven't really got into it in macOS 27. I did laugh at one article I read, which I think was called Remove Mac AI. It was like a tool that came out. Oh, yeah, they've done that. [SPEAKER_04] 12 gig, apparently, it saves if you remove the local AI. [SPEAKER_03] So you can get rid of all that space, or claim back all that space on your Mac, and get rid of Apple Intelligence if you don't want it. I mean, look, I've not gone that far. I don't think I will. I would rather keep trying it. I mean, that's if I'm even getting any benefit out of it at all. I have no idea. It's only if it starts getting in the way I would want rid of it. Yeah. But I kind of like where Microsoft's going here. So they talked about the new search. Now, I have to say, every time they talk about new search, we all get a little bit terrified because it's like, here we go again. [SPEAKER_03] They're going to break search for us because search was never exactly ideal. It used to be good. It was simple. Yeah. Because I remember the Windows 7, you know, which was just to be ideal, right? Yeah. The Windows 98 was great. You know? [SPEAKER_04] Well, you try doing a search now. You open a file manager and you do Control-E or, you know, navigate up to the search field. You type something. It seems to take ages and it's like, it's just not good at all. And you think a search is like a basic function of Windows, of an operating system. But yeah, it's definitely broken, I would say, currently. [SPEAKER_03] I would say it's no different to Spotlight in Mac. You know, it's the same kind of problem you have. I think it's just doing so much that the screen reader struggles to get focus on what it's actually coming up with in terms of a result. [SPEAKER_04] I don't even know if it's screen reader. I just think it's the way they've implemented it. It just doesn't work very well in Windows. [SPEAKER_03] Well, it's changing. Hooray! The difference is going to be better, better than before. What they're saying is now you can basically just type in what you want to do. So if you want to open Outlook or whatever, then you can do that. But you can even do other things. You can invoke actions. What does that sound like? Hmm. What does that sound like? It sounds like Spotlight. It does sound like Spotlight. So you can go in, you can say things like, you can, and this was interesting. This is when we talk about accessibility. You can change the volume on your computer. [SPEAKER_03] You can change a microphone volume. You can, you know, make changes to settings all within search. I mean, that's exactly where you might want to say turn on dark mode or turn on narrator, although obviously you can do that other ways. But, you know, you get the point, right? For people who don't know that, you can just turn it on directly in that search bar. [SPEAKER_04] Yeah. [SPEAKER_03] I'll be intrigued to see how that works. [SPEAKER_04] Well, you think how Windows settings, you know, the new Windows settings rather than the control panel. It's not great. I mean, it has improved a lot. But it's still, it's so easy to miss controls and settings in Windows settings when you're using a screen reader. I'm forever juggling between, you know, NVDA, object navigation, narrator, and just to make sure I'm covering everything. [SPEAKER_03] Can I be controversial on that one? [SPEAKER_04] No. So the ability to do a search and just bring that setting up to the search area, that could be really cool as long as it's accessible. [SPEAKER_03] I think my name's on the door as well here. Can I be controversial on this one? Of course you can, sir. Yes. Thank you, Sean. So is this a screen reader problem or is this a Windows problem? What? Settings? Yeah. If you go into settings, right, and you're trying to navigate around, and I have this thing where I'm constantly having to, you know, move around with caps lock and arrow keys to try and read all the text around about it. You know, I tab straight into an on-off switch. I've no idea what the on-off switch is for, right? [SPEAKER_03] So I'm having to caps lock left and right and all the rest. Yeah, exactly. What on earth is this for? Is that a screen reader thing or is that a Windows thing? [SPEAKER_04] That's beyond my pay grade, I'm afraid. I don't know if it's to do with MSAA or UIA or the, sorry, the exit. [SPEAKER_03] UFC. I knew you were going to do that as soon as I said it. [SPEAKER_04] I don't know. I don't know if it's to do with the accessibility layers, right? Because they use different ones and screen readers and whatever else. It should all be invisible to us in the background. But as we all know as screen reader users, actually it's not. And there is a difference there. And you do miss things. Now, yeah, honestly, I don't know. I don't think it's a Windows thing. I think it goes a little bit deeper than that. I do think it's partly screen readers. [SPEAKER_04] But still, you just kind of think, well, shouldn't this be covered by Microsoft and Windows until it is totally fixed? [SPEAKER_03] Well, arguably then it's a Windows problem, right? Because it's not being picked up properly by the screen reader. There's an issue there. I mean, for example, if I'm going through the Mac settings. Oh, yeah. I mean, I'm just comparing. Look, I've been down on Apple for a while. So, you know, come on. I have to bring a little bit of fanboyism in here. Otherwise, you know, the Apple circle will throw me out again. As I think they probably have already, to be honest. Yes, I think so. But, you know, when you go into Mac settings and you go through the settings, it will tell you the text. [SPEAKER_03] So if you go to a drop-down box or you go to a radio button, it will read out the entire sentence. You don't have to be constantly navigating back and forward to read it. So obviously it's possible to be done. That's what you're saying. It's better. Well, at least you can read the information. Yes. [SPEAKER_04] Yes. [SPEAKER_03] Okay. I agree. I often sometimes wonder if it's the positioning of the items. You know, because I find if you have a radio box or a checkbox before the text, you get on a lot better. It seems that when it's after, that's when some of the problems seem to arise. So maybe it's just a layout issue. But I don't know how these things work. We should probably get someone on who knows about this stuff to talk about. [SPEAKER_04] We should, yeah. Because actually it's an interesting, it is an interesting story about the accessibility layers and, you know, the cross support across screen readers. What supports what? And what do you miss on some? It is very interesting, but it's way beyond me, I'll be honest. [SPEAKER_03] So, yeah. So they're going to, and they talked about the search bar. They kind of talked about that briefly, but then they moved on to talk about the new Copilot. I think this is like Copilot version 10. Yes. Because they're always bringing out a new Copilot. It's like, yeah, that old Copilot. Yeah, it was terrible, wasn't it? Oh, why would anybody use that? When was the last time you opened Copilot? I did it by accident once. [SPEAKER_04] Once? Okay. I used to love Copilot. I remember talking about Copilot on here. You were all in, you were like, oh, no, Copilot, my Copilot. Well, not quite, but yeah, I used to love Copilot. You know, the conversational. [SPEAKER_03] I think that's exactly, I'm fairly sure that's exactly how you sounded. Yes, okay. [SPEAKER_04] I used to say the conversational mode was absolutely great. I used to love it. Mickey Mouse over here. [SPEAKER_03] Or Minnie Mouse. [SPEAKER_04] The conversation, the chat mode was fantastic when it first started. But since then, again, a lot of it is to do with navigating around the app. It's just, it's doable, but why was it clunky and not intuitive at all? I opened Copilot just this very moron, sir, because I just found out there's a Copilot key on this keyboard. I didn't know it was there and pressed it by accident. [SPEAKER_03] So you didn't intend to open it. I did not. You did what I did. You just hit the key and found it just opened by itself. [SPEAKER_04] But it did get pretty, yeah, it wasn't great. So the Copilot demo apparently was quite impressive though. Well, it was. I said apparently. [SPEAKER_03] Yes. That's where I was. Sean was gone. [SPEAKER_04] I've had a long day. Sorry. Cool. Yeah. [SPEAKER_03] I mean, yeah, it was interesting because what it's doing is it's kind of what they were talking about was getting into that whole point about agentic OS. So you were able to ask it to do certain tasks and it had all the access to the files on the computer. The things we've been saying for a long time, you've got access to everything right there. So can you use all of that? And so what it's doing is now with using all these local models, it's able to do more locally on your machine, which means stuff is not going to the cloud. [SPEAKER_03] So if you're asking it to alt tag, say, personal images or relabel images for you, for example, they're not going to the cloud for that to be done. So that's staying on your device. Now, I will caveat all of this by saying that there was no mention of which computers can actually do this. Right? Because I'm thinking, well, can mine do this? I mean, I've got a fairly new Asus ProArt machine, which I quite like. Can it do this? [SPEAKER_03] Yes. I don't know. Maybe not. Maybe they'll say, no, no, it has to be this new machine that's going to cost an absolute fortune, which we will talk about in a minute. Because that was kind of what it was all leading to. Yes. But ultimately, you know, what they're saying here is you're going to have this co-pilot experience where you can ask it to do things. It will know you can use Office in co-pilot. So instead of co-pilot in office, it's Office in co-pilot. So you can literally just chat to a document and it will just make the changes and adjust things [SPEAKER_03] as need be. Accessibility, of course. Interesting question. As always with that, how that will work. I would like to think it'll be accessible. I never know with any of these. I have to say. Well, I mean, I remember the first days of this. They were all pretty terrible, to be honest. I remember the first apps that came along for all of these. I remember ChatGPT on the Mac was one of the worst experiences I ever had. [SPEAKER_04] And do you think they've really improved that much? I mean, they've improved. [SPEAKER_03] I haven't used ChatGPT, but from people who use it, they tell me it's way better than it used to be. ChatGPT in particular. I don't use that one at the moment. I'm using Claude. And I don't think Claude's great. I think Claude's okay. Certainly better than it was when it first came out. But, you know, I think ChatGPT is better because it speaks more. It's more verbose. You can hear the responses. It's much better in that sense. But anyway, with co-pilot, yeah, I mean, look, I kind of want co-pilot to be good because [SPEAKER_03] it's a bit like I want Apple Intelligence on the iPhone and on the Mac to be good because it's got access to all information. It can go off and it can do the tasks. I have a feeling this might actually be pretty good. What? Yeah, no, I do. I think this would be pretty good because it seems to talk more. I would say it talks more to the average person. The kind of things that people would do. You know, it talks about, well, I mean, boring stuff. Let's be honest. Boring things. He was doing his tax returns, wasn't he? [SPEAKER_03] Exactly. Boring stuff. You've got to do it. But, you know, it's like he can go in and he can do all that. That's great. You know, boring stuff. And it's local. [SPEAKER_04] Nothing going to the cloud. Yeah, which is exactly the point. What was the autopilot part of this? Because when I hear autopilot, which is a term they banded around, I'm thinking, OK, so now it can control. I can say go into settings and do this or that. This is co-work or what's the chat GPT one called? Computer use, right? Is that what this is? [SPEAKER_03] No, I don't think so. I think this is more. No, I think this. No, the autopilot is essentially chaining tasks together. So I think that was the tax return example that you heard. So he goes off and he says, OK, bundle all the information together for my accountant. And then it says, OK, so it's now going to do that. It's going to zip it all up. It's going to write the email to the accountant. It's going to attach the zip file and it's going to draft it ready to send. But that is computer use, isn't it? [SPEAKER_04] It's a different set. [SPEAKER_03] Well, not quite, because computer use would be, well, I'm going to go to a website, get information about, say, a product or tickets or something. And then, you know, I'll go and do that. Or fill in a form online. Yeah, exactly. [SPEAKER_04] I would assume that's the same thing, to be honest. If it's going off and it's going to, you know, your folder and it's finding your accounts and it's zipping them up, then opening your emails or whatever, doing it in the background. Isn't that just the same as driving a website? [SPEAKER_03] I suppose one thing I'd say is, I mean, it is all within the office world or the PC world. [SPEAKER_05] Yes. [SPEAKER_03] Yeah. In the PC world. That's a good name for a store. But yeah, I mean, look, I'm excited by it. I think it's certainly, you know, it will jazz up the PC a little bit. It'll make it do more. I also thought it was quite interesting that in Copilot, you can build apps, native Windows apps directly in Copilot as well. Copilot. [SPEAKER_04] Yes. Well, there was a lot of mention of GitHub in there as well, right? [SPEAKER_03] Separately. But yes. Yeah. Yeah. I've never used GitHub Copilot. I really should because I know a lot of great work's been done on it, but never used it. You're an absolute disc... [SPEAKER_04] No, neither have I. I mean, the trouble is it's just we stick with what we know, right? Because at the minute it's I'm sticking with Claude. But I know people are trying different things now. So, and maybe this is, you know, this is the way it makes it easier. It's so funny talking about this event because it almost feels like, do we really understand what's going on here? You know, at a surface... No, absolutely not. No, we don't. And I'm going to be honest about that. At a surface level, absolutely. [SPEAKER_04] Okay. So this is all about... Oh, very good. [SPEAKER_03] I like that. Surface level. That was good. Thank you. [SPEAKER_04] I didn't understand. Oh, yeah. Yeah. Windows is all about AI. This is just sort of putting the flag in it and saying, yes, we are all in on AI. We are going this way. And hence all the partnerships and things like that and the hardware, which we're going to get to. But when it comes down to the terms, nitty gritty and how it's actually going to work, I'm still none the wiser than I was, you know, the last event they had really. [SPEAKER_03] No, I think Microsoft do have this tendency of... Some companies have a tendency of just announcing things that never will come to be, right? And that happens. Microsoft will bring this all out. It's just we have no timescale. No timescale at all. And of course, it won't just come out as one big release. That's the thing. It's not like this is a new operating system. It is ironic, though. We sit here every year and we shout at Apple because, here we go, another operating system, all these new features. Here we go again. [SPEAKER_03] Why don't you just wait a year? And there's Microsoft saying, well, we're not going to do that. We're just going to keep it as is, but we're going to update it as we go. And we go, no, I know. I want to know more. And I want a big release with all this stuff in it. I want Windows 12. Yeah. And we're not getting that. So, you know, all these things will come down the line. Arguably, you know, that's not a bad thing, right? Because they'll be tested. They'll be checked rather than just dumped out. [SPEAKER_04] I mean, have you seen the Windows updates? The last four have broken something. [SPEAKER_03] Yes, I know. I know. Well, I mean, I will say this about Microsoft. The one challenge they have is the amount of, well, legacy. Legacy. Yeah, absolutely. That they have. That's the biggest issue Microsoft have. The amount of machines they support. [SPEAKER_04] Yeah. They should do the Apple thing. Just say, right, from next month, all those machines that are dead to us. Don't even know what they are. Headphone jack, never heard of it. [SPEAKER_03] Yeah. But I'm fairly certain that the Trident nuclear missile system, which is based in the UK, is run off a floppy disk system. Yeah, it is. Like Windows 3 or I can't know what it is. 3.1. Yeah. Can you imagine? I mean, yeah, no more support for that. I'm pushing the go button going, nothing's happening. You're always pushing the button, Stephen Scott. Stop that. Do you know, that's how we could solve peace in the world tomorrow. [SPEAKER_03] Microsoft could create world peace by just ending support for Windows 3.1. Oh. [SPEAKER_04] And we could live in nirvana. I'm glad you clarified because you first said solve peace, which isn't the right term. [SPEAKER_03] Right. Do we solve peace? I don't know. I'm too tired. [SPEAKER_04] To start war. Yeah, it's awful. But talking of pushing buttons, let's talk about the hardware. [SPEAKER_03] Yes. Well, we knew it was coming. The Microsoft Surface Ultra laptop was coming. What we didn't know was price. Now, I have to say, trying to find information at the time of recording is challenging. There's not a lot of info out there. Lots of talk about what is coming down the line. I went to the website to find out more about it and it said, this has not been regulated by the FCC. Checked in the UK. [SPEAKER_03] It's not been approved by the UK Electromagnetic Society, whoever it was, that say, you know, it's okay to exist. It's not safe to use. I don't know what that means. So, clearly, that website. [SPEAKER_04] It hasn't been certified yet. Let me just say that before people say it is safe. Of course, it's safe to use. It hasn't been certified here in the UK, apparently. Or in the States. Said it on the website for the States as well. [SPEAKER_03] Hang on. They mustn't have pushed the button yet. Sure. Do you know what it is? Do you know what? They need some kind of agent to look after their website. [SPEAKER_01] It's not a good look. [SPEAKER_03] Send all the specs off. It's not a good look when you can't get the Microsoft website up. It's like they're sitting there talking about building all these amazing apps. It's like, yeah, but your website's wrong. So I'm imagining by the time you're listening to this that more information will be out. What we do know is a starting price. And Sean Preece, I don't know if you caught this because you were sleeping. [SPEAKER_04] Yes. [SPEAKER_03] So, here we go. [SPEAKER_04] Did they say the starting price? [SPEAKER_03] I was asleep. Obviously, I missed that. You absolutely were. They did say a starting price. What was the price of the Microsoft Surface Ultra with its RTX Spark processor? Not Intel. Not AMD. Not Qualcomm. It's running the NVIDIA RTX Spark. Which could have up to, this is not in the entry level, shall we say, but can run up to 128 gig of RAM. It can have a petroflop of AI stuff. [SPEAKER_04] I don't know what that means. Okay. I'm going to say, okay, this is a laptop, right, though? Yes. Okay. So this is 15 inch. This isn't a server. I'm going to say $5,000. [SPEAKER_03] I kind of thought the same. But the starting price here in US dollars, $2,599. It's a bargain. Let's buy two. [SPEAKER_04] All right. Okay. Is that with like four gig of memory or what? No. What is the starting point? [SPEAKER_03] I'd be amazed if it isn't 32 gig because that is a powerful machine. It's supposed to be a powerful machine. I'd be amazed if that starts at 16. We shall see. It didn't say I couldn't find out. I was hoping to find out before we could start recording. [SPEAKER_04] What you're paying for is that processor. I mean, it is cutting edge AI processors. [SPEAKER_03] So this is a new category in a way of processors. So we've had Intel. We've had AMD. Obviously now we have Qualcomm and ARM processors. And now the new player on the game and the block or the new game on the whatever is NVIDIA. Retired. With RTX Spark. NVIDIA, of course, no graphics cards. Now they're also doing the CPUs as well as the GPUs all on one chip. What does that sound like again? What is it? Almost like a... What was it called again? [SPEAKER_04] Apple MacBook. They did say unified memory. So it is very much like the Apple Silicon, right? The way they're doing it. Incredibly fast memory. But of course, with the memory prices now. I would... To be honest. At that price, I'd be surprised if it came with 32 gig. [SPEAKER_03] Yeah, no, you're right. I mean, it may well be 16. But you know what? This is not for us. This is not for the likes of the new guys. It's not for us. [SPEAKER_04] If you're listening, Microsoft, please. [SPEAKER_03] Oh, I mean, if they want to send us one, that's fine. But that's never going to happen in a month of Sundays. [SPEAKER_04] I will fight you for it if they do. [SPEAKER_03] Yeah. You know what? You can have it for a week. And I'll take it back. But no, honestly, I'm not overly fussed. I'm kind of keen to see what my little machine can do with all of this. And it's not going to happen. The thing is, everything they talked about is going to come out over the next while. Some features may be reserved entirely for this particular machine because it's going to have that capability. And, you know, this whole talk of local models, again, it depends on the hardware. So let's not get too excited. I mean, I remember negative Julian saying, you know, [SPEAKER_03] well, it's all very well saying that, you know, I can cram all this information onto my phone. Is that likely? Probably not. And I think that's the case for a lot of us in the machines we currently have. If you've got a five-year-old laptop sitting there, even a three-year-old laptop sitting there, that's not a MacBook, then I don't know what you're going to be able to run on that. I mean, if it's a gaming machine and you've got a gaming desktop with a decent graphics card at 40 or 50 series, then you're going to do okay. You might be able to run some good models. [SPEAKER_03] If you've got a high-end Mac Studio, for example, or even a decent Mac Mini, you might be able to run some models. And I will say that I think this is where Apple can feel like they've fallen behind, although they haven't. But it feels that way because they don't talk about all of this the way that Microsoft does. They're more open in talking about the fact they're working with all these different partners. Microsoft, or Apple, I should say, are quite restricted in who they work with. But you can use all of these models on these machines. They just don't tell you that. [SPEAKER_04] There is kind of a reason, though. Apple very much, they own the entire stack, as it were, the software, the hardware. Everything is controlled by them. Apple Silicon, for example. Everything is controlled by them. They really don't want to advertise. You know, the whole Gemini, you know, using the Gemini code to help with their Apple intelligence. I'm sure that's stung. They want complete control. [SPEAKER_04] Whereas Microsoft have always sort of been a little bit, yeah, the other, what's the word I'm looking for? Manufacturers for their hardware, for example. You know what I mean? It's always been the Microsoft way. [SPEAKER_03] It's funny, though. I saw a story the other day. I don't know if you saw this about Claude now available to be used alongside Google Docs and basically Google Workspace. It kind of made me laugh because I saw one of the comments immediately was, oh, well, there's absolutely no reason for Gemini to exist now. Because there you go. It's all in. It's, you know, Claude could just get into everything with Google. So what's the point? Why do we need Gemini? [SPEAKER_04] Well, that's kind of the point. [SPEAKER_03] It's a little bit dismissive. [SPEAKER_04] It's kind of the point, though, right? It shouldn't matter about the model. If Windows works as it's meant to work, then it doesn't matter what model you use. It all does the same thing. We shouldn't have to worry about that. And we shouldn't have to worry about, well, I've got a subscription to this model, so I'm stuck with this. It should be easier for the end user. I'm not entirely sure it is at the moment, but we will see. [SPEAKER_03] I would propose this to anyone, you as well, because you've got a decent machine there with your Lenovo Legion. You've got a graphics card in it. It's a gaming laptop. You know, what I would say to anybody who's got a machine like that, or any machine, frankly, just get your Claude or your ChatGPT or whatever and ask it the question. Okay, look at the specs of my computer. What local models could I run? [SPEAKER_03] And ask yourself the question, what is it you're doing with AI and other tasks that you would like to do privately? So, you know, a transcription is one that comes up a lot. Not everybody wants to transcribe or have every conversation that they've had transcribed in a place where it ends up public. True. Or at least is sent to AI to be mined and used. So, you know, if you wanted transcription locally, there are ways you can do that. [SPEAKER_03] Again, the trick is to ask. Because the thing that blows my mind every time I use any of these assistants is that a year ago I couldn't have done this. One year ago there's no way. But now I'm able to say to my computer, okay, look, I want to transcribe audio locally on my machine. What do I need to do that? And can it be done? Can it be done well? You know, because, okay, it can always be done. All right. There's always a model that can do it. [SPEAKER_03] But, you know, it's like the dollar store version versus the, you know, $10 store version. Put it that way. And so, you know, you're going to get, you know, your mileage may vary depending on the machine, but also the model as well. And, you know, there's times when I've tried, for example, the Gemma 4 model, which, you know, was infinitely better than the Gemma 3 model type thing. And you've got to check these things and ask the question, you know, can my machine run this? And you don't even have to know what the models are. [SPEAKER_03] You just have to know what you're looking to achieve. And then ask it, you know, can this be done? And it does mean you might be able to avoid huge costs. Because, and the reason I mention that is because I think for blind people, this is a huge moment with AI. The capability we now have at our fingertips is way beyond what we had a year ago. And I think it's so important that we start to use it more and we start to find ways to use it in our lives. [SPEAKER_03] I do want to talk about, I'm not going to do it today, but I do want to discuss. But the amount of AI I use and whether or not that's a good thing or a bad thing. That is an episode in itself. It is. I think there's a great conversation here. And we may even do this as part of our Double Tap Live conversations, because I think it'd be great to get you calling in on this. In fact, let's do that. Let's make that our next topic when we get together on Double Tap Live to discuss this, [SPEAKER_03] because it'd be a great conversation to have and get you guys involved as well. Well, you know, if we're using AI and we're using it to do lots of things, change settings on the computer, I'm using it to download apps. I had to download Adobe Acrobat today, Adobe Acrobat Reader. I mean, I could have fought the website. Is that still going? Yes, still going. And, you know, I wanted to check a PDF. Oof. Yes. [SPEAKER_03] And, you know, there I was. I had to download Adobe Acrobat Reader. I could have gone and fought with the Adobe website. I thought, you know what? I've got Claude sitting here. It's doing nothing. Well, that's a lie for a start. It's a lie. Two other machines were working away as we spoke. But there I said, just go and download it for me. And it goes off and it downloads it. And the next minute I said, you know, open this PDF when you're done. And then it takes the PDF and opens up for me and job done. [SPEAKER_03] And within like one minute, the file was open and ready to go. You can't beat that, right? But does that come at a cost? I don't mean AI cost. A monthly cost. Yeah. Well, it does come at an AI, a monthly cost. But does it come at a cost of skill? So that's the question I'm asking you. So you can get in touch on that one and we will get a debate on that one going. I'm certain that'll get a lot of interest there. Yeah. [SPEAKER_04] Yeah. [SPEAKER_03] I kind of want to start the conversation now. I know. It's a good conversation to have. Just a quick note to say tomorrow on the show, we're going to be wrapping up our thoughts on the NASA webinar, which took place. We're going to have a number of experts joining us on the show. Dr. Robert Shelton and Dr. Craig Moore joining us from NASA. Can't believe I'm saying that. Wow. [SPEAKER_03] But they're joining us along with Christine Malek, who of course is and has been a major part of bringing this together with NASA. So they're going to be joining us tomorrow on the show to talk about what the NASA webinar was like. If you've got a chance to go along, do tell us what you think. And apparently it was a fantastic event. And one question I did ask, which I thought was interesting, was, you know, what kind of questions did you get? And of course, you would imagine tons and tons of very scientific and smart people questions. [SPEAKER_03] But then also just lots of people asking, like, you know, how would blind people survive living in space? Which is a great question. And we'll get the answer to that tomorrow. Because, you know. Is it? Is that a great question? Other than the obvious answer, which is, well, you're never going to find anything ever again. It just floats off. Yeah, that's right. Where's my mug? I have absolutely no idea. It's on Jupiter. [SPEAKER_04] It's kind of like what happens now, to be honest. Just everything just floats away if it's not in the hand. [SPEAKER_03] Well, that's what they say, right? If you move, if you take my mug and you move it away from me, you may as well put it on the surface of the moon. It's the same thing. [SPEAKER_04] What's it with you and mugs? You're obsessed, sir. [SPEAKER_03] It's a real thing. We all use mugs. I don't. Some of us. Some of us work with them. Talking of hardware, I mean, it just rolled him back. I love how you missed that one entirely. You are tired today, aren't you? [SPEAKER_04] From a blind man in space, I just sort of tuned out. The Copilot Plus PC was mentioned a lot. I thought they dropped that. I thought I read a lot of news articles saying, yeah, Copilot Plus, yeah, they've dropped that. It's no longer called that anymore. Yet, in this event, they talked about it quite often. [SPEAKER_03] Yeah, they did. They did. I thought that was going to be a dismissed term. I also kind of expected more Surface laptops, like, you know, Surface Pro. This wasn't a Surface event, Stephen. [SPEAKER_04] No, I thought it was. Well, it had a Surface in it. I mean, that's better than nothing, right? I mean, they also mentioned Dell, HP, Asus. What else was there? Oh, yeah. [SPEAKER_03] So, that's right. Because they brought out other variants of this Surface. [SPEAKER_04] All running the RTX Spark. [SPEAKER_03] Yeah. So, lots of different partners on that. And, of course, different prices, different touch points. You might have a cheap one for me. Or beige boring machines. Nice. [SPEAKER_04] Also known as PCs. They've also got a big box, the DevBox, haven't they? [SPEAKER_03] Oh, that's ridiculous. [SPEAKER_04] I don't know how much that's going to be, but they were talking about it running, you know, almost. It's a supercomputer under your desk. [SPEAKER_03] No, no, there was two. There were two machines. So, there's one called the Surface Dev Box, I think. Which was, like, you can develop applications and stuff on it. That was $6,000. Oh, you got a price for that. $6,000. $6,000 on that one. But then there was this, what they called the data center in your home. Yes. Which is like, what? I mean, that was part of me going, that'd be quite cool. But, you know. It's like, I absolutely should not be. I'm like, yeah, that's quite interesting. But, no, definitely not. This is not a consumer device. [SPEAKER_03] This is for companies, right? Companies who want to. And there's a lot of companies like this. They want to be able to do this. They want the information local. They do not want to go out to the cloud. You know? I mean, it's funny because in this world where we think, well, you know, you could just develop any application and any idea could just come to life. And you can just do it all on AI. That's true. But there's still companies and people out there who kind of want to do stuff privately and not have it, you know, just exposed to the world. Like, I might have a great idea and I don't want the entire world to get to it before I do. [SPEAKER_03] And that's the people who buy this. [SPEAKER_04] Not only that, there's the enterprise level of it where they just can't, you know, if they're dealing with client information. [SPEAKER_03] Yeah. [SPEAKER_04] Legally, they can't let that slip out of their hands. They've got to keep it all in-house. [SPEAKER_03] Well, that's kind of how I learned about local models because I kind of thought, what was the point? But then I was watching a YouTube video of a guy who was saying, you know, I do a lot of work. I do under NDAI. So I can't, not NDAI. NDA. And it's a video format. We mentioned that. So non-disclosure agreement, right? So that work has to be private. [SPEAKER_03] He cannot use the cloud for this because that would be infringing on all kinds of issues and copyrights and whatever else. So for that reason, they have to run things locally. At that time, he was using a Mac Studio for that. But, you know, and of course, people still will. But this is kind of where this technology is important. And companies will, I think we're going to get to a point where companies will in certain places will have their own AI. [SPEAKER_03] I can even see a point coming, if I'm honest, in the future where, you know, we'll have home AI servers. And that'll be it. We'll just have a home AI server, which is kind of, people are talking about that already. They're talking about, what was that? Minilabs or something is the new thing everyone's talking about online. [SPEAKER_04] Yeah. I mean, it's how much, it's always cost, isn't it? I mean, would you rather pay a monthly subscription, which you're probably going to do what you already do for Lady A+, for example, if you like that? Or would you rather just do it yourself? But that's a big cost at the beginning, right? [SPEAKER_03] But that cost is going to come down. And I think that cost is more likely to come down quicker and is a better way to go than continually building data centers. Oh, it's got to be. Look, from the nerd point of view, oh, the RAM prices have gone up. Yeah. And also villages are disappearing and built, places are burning up because of the environmental impact these data centers have. I mean, where I live, there's talk of a data center being built and everyone's terrified. [SPEAKER_03] What's going to happen to local power, water? What does it do for the environment? Because they heat up environments. Now, in Scotland, that's not necessarily a bad thing. You could do some heat around here, so that would, you know, we'll be opening the windows in winter to say, get that heat in here. But honestly, we are worried about all this stuff. And the thing is that in order for this to be successful and give us all the things we want, like we were talking yesterday about live AI. Yes. [SPEAKER_03] You need to double, if not triple, the amount of data centers just to be able to achieve that. And these data centers have to be bigger and bigger and bigger every single time. It's not sustainable. [SPEAKER_04] Oh, absolutely not. Just the energy that it needs is not sustainable. So, well, maybe it's the, you know, something that will push the technology on, you know. But the trouble is you make better, faster processors that can handle so many, you know, petroflops of calculations or whatever. And then, well, that just ends up in a server farm and it gets more powerful, right? [SPEAKER_04] There are always going to be a need for those because of enterprise usually as well. As much as enterprise, you know, wants to handle their data locally, they also want the fast service of things, you know, running a customer care line through a server farm. I don't know what the answer is. It is really tricky. And I say all that and I recognize all those negatives and the impact that data farms, data servers have. [SPEAKER_04] At the same time, I'm still using AI a lot. [SPEAKER_03] Oh, yeah, me too. Yeah, absolutely. And that's the quandary. I think the word is quandary. I like that word, Stephen Scott. It's also a laundrette in the corner. Well done. But honestly, I mean, when I read about what's happening in China. China. Well done. With all their models that they're making, which can apparently run on local systems and are far more efficient. [SPEAKER_03] And you kind of think, so it's possible to do it, right? It seems to be possible to make models that are more efficient, that don't need all of this power. And so you kind of wonder what's going on here. Is this a money grab? Is this an opportunity for just massive companies like NVIDIA to make huge amounts of money? Oh, yeah, you'll need that. Yeah, you'll need that graphics card. Yeah, you better buy 20 million just to be sure. [SPEAKER_04] Yeah, but there's a demand from us. We want this AI. [SPEAKER_03] No, but the point is that if there are models out there that are running, that are actually capable of doing the same job better, but more efficiently. And, you know, because they come from China, we're tending to avoid them. And I mean, look, I can totally probably get behind the reasons why. But then also I'm like, yeah, but it kind of shows the point that you don't need to have all this data center power in order to do the job. [SPEAKER_03] I'm just kind of confused, I suppose, by what I'm hearing from, you know, these people are saying, well, I've tried this model from China. DeepSeek is, of course, the one we hear a lot about. It was interesting to actually hear it mentioned by Microsoft. It almost feels like it's validated when you hear it at a stage like that. I thought that as well. [SPEAKER_04] Yeah, because you hear so many negatives about, you know, these certain open source or open weight. Yeah, happy. Open source or open weight models that you think, OK, it almost makes them feel sort of shady almost, right? [SPEAKER_03] Yeah. Yeah, it does. It does. I mean, it's the same with Hermes and OpenClaw. They mentioned that on stage. Wow. [SPEAKER_04] Yeah. [SPEAKER_03] And actually, the one thing that also was quite interesting is that they're building in some kind of getting started guide, actual process to setting up something like an OpenClaw or Hermes on the computer. I don't know if you fell asleep during that section as well. [SPEAKER_04] No, that is so cool. Because, you know, no need for terminal, it mentioned, or config files. You know what I mean? It's just a, you know, which one do you want? Click next and it will sort it out and get it all ready to go for you. That's exactly what it needs. It still feels nerdy at the minute. [SPEAKER_03] Yeah, it does. When it comes to AI. And that's why people are hesitant. I'm still confused as to why I would need it versus what I'm doing just now. But that's because I've given God so much access. I think, well, do I need to give it any more? Sure. And I'm kind of nervous too. I think it's because at the moment, and the reason I haven't, I suppose, is because ultimately I feel with Claude, despite all the horror stories I read, you know, I have not had any issues with things going missing, deleted. In fact, if anything else, it's quite the opposite. I asked it to delete some files from my computer. [SPEAKER_03] It wouldn't do it. It was like, no, definitely not. I will put them in a file on your desktop called Check This First. That's right. And then you will do it yourself. Yeah. It's like it's the opposite problem for me. I can't get it to do half the things I want it to do. It's like, no, you can't. I can't do that. It's like, come on. Go on. Go on. Come on. Please. I'm blind. I've thrown everyone at it. Trying that. I even throw the I'm blind regularly. I have my blind card stamped. And does it ever work with AI? [SPEAKER_03] Sometimes, sometimes it'll say, oh, well, let me try. Or it'll say, nope, I'm sorry. I can't do this. I'm not going to tell you what I was asking you to do at the time. But let's just say. You're trying. Yes. It was. [SPEAKER_04] You're trying to jailbreak AI. You're like one of those hacker people. I'm like, I'm going to start wearing hoodies. [SPEAKER_03] We injected a prompt in one of our shows, didn't we? We did. Yesterday on the show, you said, I'm not even going to say the phrase again, because I can't go through this again. But you said a phrase which basically suggested that we should start again, right? Yes. I've said it. I bet that'll probably. Yeah, that's enough, apparently. And with AI helping us build the show, as it does, it assumed that we meant we wanted the whole thing to start again. And it actually deleted 21 minutes of the episode. And started again. [SPEAKER_03] And we've said that so many times now. That's definitely going to happen with this one. It'll happen all over again. But it was so interesting that I actually took that as an instruction. Yeah. [SPEAKER_04] I wasn't talking to it. Yeah. No. Again, an example of prompt injection. So far, it's just, you know, the security experts say that this is a problem. And you go, really? Is it? And then you realize, oh, it's actually listening to what I said and thought that was a command and did it. It could have gone worse, considering what I said. It could have gone a lot worse. [SPEAKER_03] Well, that's right. Absolutely. I said at the time, I hope you never use that phrase. It's not a good phrase to use in case someone takes us an order. And guess what happened? So, yeah, this is the world we're in. But it's fascinating, isn't it? I mean, it is confusing. And I know there's many of you who will be sitting here listening to this saying, why did I listen to this for the last hour? I have no idea what these two are on about. Yes. And I get it. It is so confusing. And all the language of it is confusing. It still feels nerdy. I will say, having been in this, I mean, I remember when I first started watching the [SPEAKER_03] Google events and I kept thinking, what on earth is everyone talking about here? This is like a whole foreign language I don't get. And I still don't quite fully get all of it. But I'm starting to understand the language a bit. And part of that is because I'm using it. Because I'm using it each day, I'm kind of starting to learn and asking it as well. Okay, what does that mean? Break that down. Because sometimes it will come up with an answer and I think, what on earth are you talking about? And so I just stop and say, okay, what do you mean by this? Yeah. What are you talking about? [SPEAKER_03] And then it'll come back and say, right, okay, so here's what I'm talking about. And if I don't understand it, I'll go back and say, still don't get it. And then try again. I'm not a smart person. I keep telling people this. No one listens. Don't make time for us. Thank you. Make it sensible to me. I am an idiot. Listen, that's it from us today. But I hope you enjoyed it. It was a lot of fun. We are going to be, as I say, talking about that NASA event tomorrow. The one that just happened last week. [SPEAKER_03] I think it happened. So we're going to be reminiscing on that. And also coming up on Saturday and Sunday, our weekend specials, Remembering Brett Halley. And great conversations with Brett that we had. We'll be playing them over Saturday and Sunday. And more details of how you can listen to that on Double Tap Live. We're back tomorrow, though, with plenty more. Join us then if you can. [SPEAKER_05] Bye-bye. [SPEAKER_03] Bye-bye. [SPEAKER_07] This is the story of The One. As a shipping manager, he's running the dock, stocking supplies, and making sure every order goes out on time. And Grainger helps him stay prepared. From shipping supplies to PPE to hand tools, he counts on Grainger to keep products moving and trucks rolling. So his operation runs smoothly no matter the shift. Call 1-800-GRAINGER. Click Grainger.com or just stop by Grainger for the ones who get it done. [SPEAKER_06] Rubrik is the security and AI operations company. Built for what happens after an attack hits, not just the moments before. AI has turned the threat landscape into quicksand, moving too fast for any human to fully predict. That's why an agentic cyber resilience platform matters. Automated recovery. Clean data. A business that keeps moving no matter what hits. One platform. Not a patchwork of stitched together tools and gaps. Don't wait for the next attack. Secure and accelerate your business at rubrik.com. Again, rubrik.com. [SPEAKER_02] Observe. The North American dad. 50s. Great polo collection. He enters his vehicle information on Carvana. An offer appears in minutes. Remarkable. He squints, reads it again, smiles and says, Dang, that's a good offer. The highest form of approval. No haggling. No hassle. No one called him boss. He's still on the couch. Astonishing. Sell your car today on Carvana.com. Pickup fees may apply. [SPEAKER_00] Kelly Clarkson here with Wayfair. I know holiday magic starts at home and Wayfair makes it easy. They have everything from Christmas trees and mantle decor to outdoor lighting and inflatables that I'll use year after year. Head to Wayfair.com today. [SPEAKER_08] Wayfair. Every style. Every home. [SPEAKER_01] Don't miss a minute of United Soccer League action. The USL Championship and USL League One are on CBS Sports and ESPN platforms all season long. With additional coverage on SiriusXMFC channel 188 and select matches available in Spanish on 2DNA. It's your time to take part in building what's next. Visit USLChampionship.com or USLLeagueOne.com for complete listings. [SPEAKER_01] Find your local club and be part of what's next.