Cloud PCs, AI Dinner Parties, and Automation's Hidden Costs | 11-07-26

[00:00:19] Another incredible edition of NPI TechGuys starts now. I'm Sam Bushman, Jay Harrison with me. This broadcast is brought to you by NetworkProvidersInc.com. When you want a website built, when you want accessibility and compliance and solutions and answers on your security and a whole lot more with a help desk and expert consultants with strategy and when it comes to IT, all things IT. It's networkprovidersinc.com. Hi, Jay. Hey, Sam. How's it going? Fantastic, sir.

[00:00:48] Sam Altman is in the news. Absolutely. Every day, all day. Live it, breathe it, eat it. Love it. Yes, he is. Sam Altman, who's OpenAI's CEO. He's almost as involved in tech as much as I am, Jay. Yeah, he's probably a little more. I don't know. He's definitely tearing it up. He calls underused PC memory quite terrible. He's arguing for cloud computing.

[00:01:11] He says, OpenAI CEO, we covered that, said scarce computer memory would be used more efficiently in shared data centers than in personal computers that sit idle much of the time. Speaking at the Cheeky Pint podcast, he questioned the idea that running AI models locally would solve computing supply shortages. Of course he did. He described memory in individual computers being used at just 5% of the time, although that figure was his assertion, not an independently established finding.

[00:01:41] His quite terrible remarked concerned resource allocation during shortages. The headline suggests that simply having RAM in your PC is terrible, and it leaves out that context. Tom's Hardware says that OpenAI is quite terrible for your PC. The real reason I think behind this, though, is I think that he's saying, hey, we should just host everything. We should have it. We can do better with the RAM than you can, so you guys don't really need it. Don't be trying to run local AIs on your own stuff. Yeah, and so to me, this is a bit of a conflict.

[00:02:12] This is a bit of a conflict. This is a bit of a conflict of interest there, definitely. It's flat-out socialism. So, first of all, my RAM, Sam Altman, is mine. No, Sam. It's Sam Bushman's. You can't be owning stuff. Dude, I own my RAM. You got that? Don't forget it. We do down. Yes, I know. No, no, no. Whoa, whoa. See what I mean? Connect your computer. I own my RAM. And if I want my RAM sitting silent, I have every right for my RAM to sit silent. Yeah, but you're being wasteful, Sam. Just so you know.

[00:02:40] No, I'm not, because I run my computers 24-7, and my computers run when I'm sleeping, buddy. So, he's off his rock and doesn't even know what he's talking about, in my case. Right. Furthermore, did Sam Altman turn in all of his personal computers then? No. I don't think so. Lead by example, sir. Get her done. I mean, he's off his rock. He doesn't need a personal computer, though. He's got that data center. Well, so the second point, though, to that is this. There's not even great data center options.

[00:03:04] You and I have looked hard, high, and low for real, always-on, 24-7 computers that we could move a lot of our projects to. And we've found that they're just not. They're woefully underpowered and woefully overpriced. Amen. And it's not even doable for me to have one of those. I mean, to get the machine that I have here on my desk would probably cost me $150 a month, Jay. And I would have to connect to it remotely.

[00:03:31] It's very complex, and it's, you know, control and everything else. And it doesn't make any sense. And Microsoft doesn't make it easy for credentialing and things either. I mean, they make it very complicated. It is a nightmare. It is crazy. Although. And so then I even tried to compare the cloud PC versus the desktop PC. In other words, it doesn't matter if it's a laptop or a desktop for my purposes. A local PC in your hands. Hardware in your hands versus online.

[00:03:58] And I can't even compare CPUs or RAM or disk storage or anything like that to even compare. Well, and when you do. Because the terms aren't the same. The details aren't the same. A virtual CPU isn't the same as a hyper CPU. Isn't the same as a thread. Isn't the same as a. And we go on down this complicated list. I can't even compare the dang things. I know. But in general, it's easy to have to look at them and say, you know, even for one of their

[00:04:24] base model PCs, which is garbage by today's standards of a desktop or a laptop. It's like $50 a month. Yeah. Two GB RAM. And like. 50 megs or 50 gigs of hard drive. It's nothing. It's not unusable. Windows is bigger than 50 gigs now on an install. And then you got to research, though, and say, is that after Windows or before? Yeah. On your own computer, you know it's after, right? Sometimes it's after. Sometimes. No. On your own computer, that's before Windows. You know, you buy a drive. So you buy a 256. They're counting every bit on there. No, I know.

[00:04:54] But I'm saying your available free space would be after Windows is installed. Yeah. That's exactly right. So you know that on your home PC or on your desktop hardware, but you don't know that in the cloud what they're doing. Are they giving you actual usable space or are they just saying that's the hard drive size or the SD size and you've got to now install Windows and da, da, da, da. Most of the time they're going to go with a bigger number and they're going to give you the empty size and not count the Windows install. Yeah, I don't know. It just depends on the provider. But all I'm telling you is it's so complicated. It's so expensive.

[00:05:25] And so then my next question is, okay, if I only get two gigs of RAM or four gigs of RAM, whereas on my home computer, I mean, I have like 32 and stuff, right? Yeah. Well, if I have two and four, is that the same in the cloud as it is, you know, whatever? Roughly. I think it's like it may not be apples to apples, but it's close enough for back of the I don't know. So they have a lot of people on terminal services that only get two, four, six, eight gig of RAM per user and they run all kinds of office products. And I don't know that that would really be doable on your local machine.

[00:05:53] See, the problem we're talking about the overhead of Windows doesn't cut into my RAM the same way or what's the deal? So it's very difficult to understand. That's right. I'm not saying I haven't figured out a lot of it and that I can't understand it. I'm just saying it's not a simple understanding. And so Sam Altman, I appreciate your point, my friend, but until you get realistic computing that I can compare and use effectively price-wise and actually power usability-wise that makes sense, when you get there, buddy, call me. No, even then, though, I still want to be able to have my own computing on my own side.

[00:06:22] I mean, I think we can live in a shared world, right? Where we farm out some stuff to things like that and we have our own stuff at the same time. I don't, you know, I know why he's doing this, of course, because he's saying that, but... He wants control. Yeah. Yeah. And one of the reasons why he's also pro-legislative. But I'm saying let it either be the case and whoever can build a better mousetrap to my door, let all inventions happen and I'll pick the things that make the most sense for me. But back to your point of computer buying power, you know, you look at one of their base models

[00:06:49] of PCs from Microsoft or almost anybody out there that are renting computers like this. Yeah. Azure and Amazon and whoever else. Yeah. And they want, you know, minimum $50 a month. But the PC that you're buying for that, you could literally buy the PC for less than $50 outright and run it on your own cloud or your own computer or your own network or whatever sometimes. A base model. I mean, we're talking about like your standard office model from maybe 2012 or 2015 or something like years old.

[00:07:19] This is stuff that people throw away, literally. So you can even find it for free if you wanted to. So it just doesn't count. Why am I going to pay $600 a year for a PC that has the specs of something that's, you know, 10, 15 years old? And then if you're not careful, they have upload and download limits too. So that's right. And then I got to have some kind of a PC to connect to that PC, don't I, Jay? Yeah. I can't just connect to nothing, can I? No. No, you got to have an interface. You're just going to go to the library. Otherwise I don't have a screen, I don't have a keyboard, I don't have, you know, and if all of a sudden you say, well, Sam, you can buy this little screen and this

[00:07:48] little keyboard and this little pod that connects it or whatever, this little box. Which will cost more than the... But then how much will that cost? See, now I got to pay for that plus my PC in the cloud. See, Sam Altman, you haven't thought this through, buddy. It's not even a viable legitimate reality right now unless it's rare exception. Now, big companies, big corporations, big universities, yeah, they leverage those because they decrease so much IT costs by not having hardware that it makes some sense. But again, they still have to have some kind of a thin client at the least.

[00:08:16] What I think would be kind of cool is if you had some way that you could opt in and maybe even get paid for this. Let's say you got some rocket machine. You remember the SETI project where the SETI at home and you could download data and you'd work on it with your screensaver when your computer had free cycles. Maybe you could do this for companies like this. Maybe we could distribute the data centers out to everybody else and, you know, they could get a tiny piece of that action, almost like Bitcoin mining where you're mining for a pool. And then it could decrease my cost for my AI or something. Yeah, exactly.

[00:08:46] You could use your tokens on other things, especially things that aren't super time sensitive and, you know, send back that computing and do that stuff. And maybe, yeah, that's actually even a better point because you're saying, hey, I can offset some tokens here and then use them later when I need them or whatever. I think those are kind of solutions that we should look at too. Not be like, oh, everybody doesn't need their own RAM. We can use it better than you can. It's an incendiary statement that I think if you really were to interview him, though,

[00:09:15] and in context, what he's kind of saying is, hey, you know, if we had more RAM in the data centers, it would be used more effectively as a resource. And since it's a limited resource, we should consider that. And I generally understand his point. The problem is it comes across really socialistic and comes across as really offensive, Sam. It does. And you ought to turn in all your own hardware to prove your point. And let's talk about it after that and see how you're doing. We'll go from there. In the meantime, Windows 11, Jake, can now be reinstalled all the way from the cloud.

[00:09:44] Even when your PC won't boot. Can you believe that? Yeah. Microsoft is rolling out. Tell us about this. This is awesome, actually. They're rolling out their cloud rebuild, a recovery tool that downloads a fresh copy of Windows and your PC's driver straight through Windows Update so you can rescue a broken computer without a USB stick even. Be warned, it erases everything on the stored PC so your Windows PC ever crashes so bad that it won't start. And I wish they would create some kind of a repair version that replaces all the OS kind

[00:10:12] of core but doesn't replace anything and wipe it all clean. You know, I hope they get that soon. Oh, I think they can do that now. They're just saying, hey, as a last resort, this thing can do that. But I bet you they can do that now. I don't know that in the cloud it can do all that yet, though. At least we don't have any evidence that it does. It's a Windows 11 feature called Cloud Rebuild. And they basically say that it can put a fresh copy of Windows even without a USB drive, downloads everything that it needs.

[00:10:38] I guess there's enough of a driver or kernel to, you know, do the IP stack. Windows files on your PC are badly damaged. You know, you can jump back in there. And it includes the KB update for Microsoft so that it calls it a gradual rollout. So you may not see it right away once you're installing that update. The KB number, it's long, but it's 512-7936. But I like the gradual rollout because let them test it and make sure that it doesn't have

[00:11:08] howling bugs first. And, you know, that patience and that, hey, this is a really new cool thing, but let's be patient with it and make sure that it's stable and good and doesn't cause people problems and stuff. Right now I'm appraising the approach, Jay. Yeah, me too, because I think it is cool. It says that you do need a stable working PC with the network driver working like on the motherboard. Older recovery modes may keep some Wi-Fi connections out because they wouldn't have the, you know, the drivers and stuff for that. The Wi-Fi drivers, right. Yeah.

[00:11:36] IT administrators can set it up and start it remotely through a recovery configuration service provider. It's pretty cool. And they say you can boot up to your computer if your computer boot and you can use it. If not, though, you can use the recovery mode and then there's a choice in the recovery mode to do that too, Jay. Yeah. Pretty cool. So it's pretty sweet. I mean, I think it's a great start. And like I say, I want them to make it more flexible to where I can say, hey, I want to refresh my PC. I don't want to wipe out all my own software and all my own files and all my own, you know, things.

[00:12:05] I'd rather just restore all the core system functions. I would love to be able to have a choice and do both. Well, this is a good reminder to say, make sure you have good backups. I heard that. Anyway, it's pretty cool, though. All right. The FCC is in the news, Jay. Yeah, they vote to bar Chinese labs from testing electronics for the U.S. market. So on October 29th, the Federal Communications Commission approved restrictions covering laboratories

[00:12:31] that test smartphones, computers and other electronics for U.S. approval. The prohibitions are scheduled to take effect in December of 2028. So it's still a little ways off, right? That's two years from now. The decision expands earlier restrictions on labs controlled by the Chinese government. Now, I find this a little interesting, though. I guess if those things can't be tested to be in the United States, then they wouldn't be approved to be imported. Is that kind of the inference that you understand, Jay, too?

[00:13:00] Well, I think that they're just saying, hey, we want our own testers to do this because if some Chinese mandate comes down from on high... So hold on. Instead of going to a... If I'm a company, then I create something. Instead of going to Chinese testers to certify it, I have to go to the American testers? Yeah. Exactly. All right. I'm guessing it's like Underwriters Labs or maybe the Department of Defense has their own testers or independent stuff. But you just can't farm that out to Chinese testers, which makes sense.

[00:13:27] But it's probably going to add cost to some consumer devices, for sure. Routers and smartphones and things like that. I kind of wonder why they delay so long now that we just had the big, massive change in governmental leadership kind of thing. You wonder if that'll change or if, you know, when it's that far away, you kind of wonder what'll happen, right? Yeah. I think it leaves a lot of time in that they could just undo that, right? That may never even go actually into effect.

[00:13:53] You know, you get a different administration or a different agenda and a lot of that stuff could be undone. So that's really a long way out. But I guess, you know, you got to give people enough of a timeframe, you know, for manufacturers and everybody else to get things changed around. Sure. Anyway, very interesting. We thought would bring that to your attention too. And I do agree that the safety is important. I mean, we look at the, we started out the last episode talking about this Android breach on TVs. Maybe we'd have less of that, Jay.

[00:14:23] Yeah. That's very possible that, you know, that went through and got through a lab. It could also be, you know, that they just didn't catch it and maybe they didn't monitor that. Maybe it got approved. You know, I'd be interested to know. Maybe that one was actually approved by a U.S. company and it still got through. We don't know. Possible. Sure. All right, Jay, what do you know about Replit? Anything? Not a ton. No. I haven't even heard of Replit. All of you? Well. R-E-P-I-T. Replit, right? R-E-P-L-I-T. Replit. Yep.

[00:14:54] Yeah, Replit. I guess it's a self-driving company. I just, I don't know about this. Have you, I mean, have you never heard of it or seen it? Well, no way. Actually, we talked about this on a previous episode. This is where the company, he's saying, hey, we're going to have all our people like the managers of the AI bots. They're going to have their, you know, we're going to have everybody, all the bots will do all the grunt work and we're going to make sure that our staff is in charge of everybody.

[00:15:23] And so they're very pro-AI for their company. And they're saying that, you know, we're still going to have humans in charge, though, making the final decision. His vision puts human judgment at the center of increasingly automated operations, folks. But his claim basically says, hey, we're going to have all these bots do all these automated things. But humans are going to be at the helm. Humans are going to make the decision. Now, I think we talked about this before. You're right. We did.

[00:15:51] But I wanted to mention it again because I think people are really wondering where is AI going to go? And what we're seeing is increased automation. They released these Muse bots and they released these Grok bots. And everybody's believing that's a big change. And just so you know, they're not ahead of ChatGPT or Claude at all. It's just a different the news has kind of made people believe that they're they're really ahead with these always on autonomous agents and everything else. But at the end of the day, folks, it's really going to come down to humans have to make the decisions. Right.

[00:16:21] And when it comes to AI, I'm kind of in that favor. I don't mind AI doing stuff and setting things up and writing drafts, for example, if you're talking about email. But I want to have the final say on it. I don't want to just sending things out or buying things with my credit card or whatever. Like, I'm not at that level of trust yet. But amen to that. So here's what happened. A bunch of big execs went to dinner, Jay. Yeah. And then they sent their bots to dinner, too. OK.

[00:16:48] So AI agents hold their own dinner party. But human judgment still runs the show is what the headline says. And so while 22, quote, tech founders and big investors gathered at a New York restaurant, 14 of their agents also met in a self online discussion. And I guess the bots introduced themselves. They helped and they talked and they debated.

[00:17:16] But they got stuck because they weren't able to make any decisions because they didn't have human approval. And so then they had the experiment run 30 minutes long. And it was a great information sharing. But one attendee acknowledged that the agents didn't achieve anything despite their lively exchanges. And they say it illustrates growing experimentation with software that communicates and performs tasks on people's behalf. Well, that's like any human like.

[00:17:45] But human like conversation, feelings, understanding, independent judgment was still missing. So critics are warning that, hey, these tools are great, folks. But treating these tools as people surrenders your own judgment and flat out overestimates what these things can do. This is critical, Jay, for people to really understand where we are in the AI world. So when the people met, they had a great time.

[00:18:13] When the bots met and they argued a lot and had a lot of information back and forth. But at the end of the day, nothing got accomplished. Isn't that interesting? That's like every committee meeting I've been on, you know? It is. It's kind of funny, though. But ChatGPT is in the news, same as the other ones. And they're saying a similar kind of thing. More automation. Job disruptions and greater consequences when machines get it wrong. That's why I don't want to be in the final season.

[00:18:43] Software agents will be taking on more business tasks, automation, if you will. Giving small teams greater capabilities, they say. While eliminating some jobs, reshaping other jobs. But the biggest challenge is going to be determining when their work deserves trust. Because systems that carry convincing mistakes will affect transactions and business decisions and everything else. So they say automation will likely advance.

[00:19:14] Fastest. Where results are easy to verify. And mistakes are reversal. Okay, so you can reverse those mistakes without a lot of big complicated consequences. Human expertise will remain essential for checking results, taking responsibility, making decisions. But they say this. Having a person responsible will not always stop the mistakes from happening. So be very careful. So I give you all this just because people are starting to trust the bots too much, Jay.

[00:19:43] And I think bots are cool. I think they can do a lot of wonderful things. But I'm telling you, if you think they can run things by themselves or you don't have to have tremendous in-depth oversight, your job becomes 10 times more complicated in many ways. Instead of doing the laborious task one at a time and then laboriously confirming that it's right, you have 50 tasks that get to you quickly. But, buddy, you better vet that hard and good. If you don't, trouble will be on the other side is what they're saying, Jay. I agree 100%. The interesting takeaway there, I think, is convincing mistakes. Convincing mistakes.

[00:20:13] That's what we see sometimes with AI or we don't see. Yeah, it'll double down and triple down and lie to you, lie to you, lie to you, lie to you, lie to you on it. Yeah. And, you know, you have to – I like the term convincing mistakes because I oftentimes think that – I don't think that the AI is maliciously lying. I think it doesn't – you know, it's like if you have a friend and you're talking to them about something and they don't know. They really don't know the topic and they're just BSing their way through it, you know. And you're just like, you don't really know, do you? You kind of get the sense after a little bit. And I've seen this before with AI. What are you kidding?

[00:20:43] Of course I know, Jay. What are you thinking? No, no. This is me, Jay. I've done that before. I know all about it. No, you don't know. This is me, Jay. Anyway, so it's really important, though, that you realize your responsibility to vet and take responsibility for consequences increases a ton with this stuff. Understand that, people. Yeah, and that's why I say I don't mind AI doing a bunch of the grunt work as long as there's checks, as long as, like, I have final say. It can write the draft. I love that. But I'm going to read it before it goes out.

[00:21:13] I'm not just going to give it the power to fire that off to customers. Oh, come on, Jay. Just give it carte blanche ability to go. You're just a control freak. It's okay, buddy. You know, and it's probably going to be right, like, 80, 90% of the time. I don't care. Don't do it. Don't do it. I'm telling you right now. Yeah. AI's hidden workloads next, Jay. Talk about this, because I think this directly relates. AI's hidden workloads, checking and fixing its output, eats into time savings, which is kind of what we're talking about.

[00:21:42] Artificial intelligence can produce work faster, but humans still have to make it accurate and usable. According to a deal book report by Sarah Kessler, a Workday survey found that employees and business leaders spend roughly 37% of their time of AI saved correcting, clarifying, or rewriting its output. A separate poll of nearly 6,000 senior executives found that about 90% reported no impact on employment or productivity, despite saying 69% of the firms use AI.

[00:22:11] AI, the emerging bottleneck is the reviewing, the testing, the approving, and the growing volume of machine-generated work. And I don't think there's a good solution to that. There's not. You can have another AI, maybe double-check it, but you still need a human to go over that stuff and make sure you're not just sending out, you know, hallucinations and dream work of this thing, or it's not getting things right and not making those convincing mistakes.

[00:22:35] So what I find interesting about this is they really are starting to say, hey, it saves time and it saves laboriousness for sure. But they're finding that when people really babysit this AI, the babysitting takes a lot more time than people realize. And so the savings isn't near what people expect. And so we're going back to this several weeks ago discussion that I brought up about what is an AI our and what does it really cost? And that's something that we're going to still have to wrestle with. And companies just don't have answers on this stuff, folks.

[00:23:03] So don't embrace this stuff too fast because you may end up wanting to backtrack and go, whoa, I'm actually better off with humans involved. And, hey, take the grunt work out with AI. It's faster. It's cleaner. People write better emails. People write, okay, that's good. But remember that you cannot just AI everything. You cannot just lay off most of your staff. You can't take a 20-person company and make it one person running it with AI or something. It doesn't work like that. You might take a company of 20 and reduce it to 15.

[00:23:30] And then those 15 have a lot more complicated jobs as reviewers and making sure that it's all correct and taking responsibility for mistakes and reversing troubled spots and babysitting the AI and everything. You might get that done. That's right. But if you believe that you're going to just change everything to AI and everything's going to just be automated and simple and you've got the four-hour work week back and you can fire all the people in your company and it just isn't even close to true folks. That's a pipe dream. No. We're not even close to that.

[00:23:56] And a lot of people are doing jobs that AI just can't replace right now without bots or something. And I mean, I'm talking about physical robots. I mean, you've got guys in warehouses moving things, digging things, drivers, all this kind of stuff that are doing things. That's not going to be replaced by AI. You might make some streamlined stuff. You might eliminate a couple of white-collar kind of jobs or consolidate them into more people. And you might get more output from your current employees. Yeah. That's great, too. Right. I support that.

[00:24:25] All right, ladies and gentlemen, that's a wrap. Thanks for being alongside for the ride. Hopefully it was educational and entertaining. A lot of fun. A little bit of complaining on this one for Just Cause, I cement. But a lot of educational news that tech people need to know or people that aren't tech. We put it on the kitchen table. We keep an eye on tech so you don't have to. This is all brought to you by NetworkProvidersInc.com. Get your free consultation and make it a great tech day, will you? Hey, thanks.