AI PC Buyer’s Guide: Specs That Actually Matter (ft. Dell’s Logan Lawler)

11:54Logan Lawler:well you know i wasn't in those conversations um but i was tangentially kind of around the edge of those conversations and i it's an interesting question because at dell we i mean but previously let me talk about previously previously if you bought a dell you know any client device from us we had very different and separate brands we had inspiran we had alienware for gaming but we also had g-series for gaming we had vostro which was an inspron but that was designed for business we had precision which was a workstation but we had an xps that also had a gp but it wasn't for gaming and we had latitude which are covered we had all these disparate brands and when you know when we did when dell did kind of the study and market research when people ask like hey what model do you like what what system do you have they always said dell that very few knew like oh i had a precision even though precision was one of the higher recalls as long along with the xps and latitude but is that everyone remembered dell but they didn't necessarily know the brand yeah and the problem with brands when you have separate brands you have separate teams you have separate people leading them and there's going to be bleed over right like for example we had g series which is a gaming had a you know a g-force card but we also had alienware but those kind of creep in like what do you buy to the point where you are going to dell.com and you're like hey i want something with an i7 32 gigs and you could have bought a computer with every single one of those options from every brand and like at that point you're like well what what am i buying i don't know and it was just yeah it was too much so when we did the rebrand i mean it was announced at ces we kind of broke it down into well alienware remains because i mean obviously everyone knows the alienware for gaming but we have dell which is just dell and that is for kind of homeschool play right that is you know something you'd have as a personal device at home something your kid takes to school, just something my parents would use, et cetera.

13:51Logan Lawler:You have Dell Pro. So think Dell Pro as in professional, that is your traditional, both desktop and laptop, your managed IT device inside a corporate environment, a business, the security, you know, reliability, et cetera. And then you have Dell Pro Max, which is me. Um, so max performance and basically, yeah, design, you know, very different. Um, and then the other thing that they did as we're, we're still kind of going through the rebrand because we have, you know, different development cycles for products and stuff but yeah um the other thing that was interesting is that that problem i told you where you could go get in this you know every single spec for all the brands it's not the case like previously in precision you could get our our mobiles were the 357 series well you could get this 5000 ada card in both the five and the seven series now you can't in our on our premium line you can only go up to the 3000 blackwell but if you want like the 5000 blackwell 24 gigs in a mobile gpu which is insane you have to go to the plus so it really based on what you need it will very quickly direct you into what we need and so far you know some customers have been buying forever been like i love my latitude or i love my precision it's all the same stuff it's just designed to make it a little easier to source what you need that's awesome gotcha well specifically

15:04Grant:about like the future of pcs right like uh are we seeing a true shift in how computing power is used for AI workflows with what's going on right now. And to put a larger framing on it, there was a great interview with Mark Andreessen on the Cheeky Pint podcast. And he basically said, like, comparing AI to the dot-com boom is probably the wrong comparison. You should be comparing it to the PC era. And like when PCs first hit the market and really started to take off because he calls this computing V2. And that sounded right to me, but I'd like to take on that.

15:39Logan Lawler:I think, and I didn't listen to that, But I mean, I think it's right because the dot com boom is kind of implying a bust. And I don't think within AI and compute, there'll ever be a bust. I think there might be, you know, a slowdown. But compute. And if you think about like in your life, tell me one day that you did not use some sort of computing power. Think about it. Your microwave. OK, well, depending on which microwave you have, like it also has a chip. So does your car. like all these things like and people just don't really know it or see it unless you like read or like you're exposed to it but yeah there's computing and everything to the point where if there's a microburst we are all in serious trouble like you're like you know at emp like we're in serious trouble everything would shut down everything runs off some sort of chip some sort of silicon you know and when you coming back to your original country kind of in the cloud right is i and nothing wrong with the cloud like i mean we you know work with aws and azure and all this kind of stuff but focusing at local here's kind of what i've seen the arc be is at first when it came first off i'm here to say if you're out there doing anything with that if you want to run comfy ui you want to run anything that first time you install it i'm not super technical i've become so you're going to want to rip your brains you're gonna rip your eyeballs out yes like the first time you ever try to launch something from command line and like you're like you're gonna be like i I don't even know what this is.

17:07Logan Lawler:What have I done? Clicking on something and going right. And it's facts. It becomes easier. So I'm here to tell you, I don't stop. Keep going because like the world is pretty cool once you kind of figure some of this stuff out. But for the compute is when I look back, say maybe two years ago, I would have to say a lot of these like cloud GPU farms, they got it right. They made it really easy. Like it made it super simple because everyone kind of knows well not everyone but a lot of people have experience in you know aws and being able to do it and just spin up a virtual machine they know how to do it it's simple it's there and i think that made it very easy which kind of like i would say kind of accelerated kind of a pc 2.0 but now people are kind of starting at first people didn't really ask questions and i think the questions are starting to come is like and nothing is not that it's not secure or whatever, but whoa, I wasn't expecting that bill or be, Ooh, where's my data going?

18:04Logan Lawler:And from a corporate standpoint, which is mostly what I work with, you know, within, within Dell technologies is those are the questions now being asked. And I don't think it's a bust on that front, but I have seen, I can't tell you how many conversations I've seen or different, you know, um, AI companies, you know, that are like started natively in the cloud. And because they're working with big companies, they've had to move either on-prem local, whether it's a workstation or a data center or whatever. So I truly think kind of what the arc is, is like cloud really accelerated. I'm not going to say we're in a slowdown because we're not, but now we're kind of figuring out what local looks like.

18:42Logan Lawler:And when we figure out what that looks like, I think there'll be kind of another rise. So like, I think it's an upward to the right, but I'm not sure is that 20%, 30 % or what it is. And it makes it more complicated, honestly in my humble opinion um and not complicated this is a good thing but i'll give you a perfect example i you know chat gpt's oss model right like if you go back to if you go back six months ago the best desk side gpu that you could get from nvidia was the nvidia rtx pro or the nvidia rtx pro 6008 it had 48 gigs of vram you could not load that on there completely unquantified it wouldn't fit now right now with the 96 gig blackwell now i haven't tested it but it's pretty close from what i've seen that was talking about the 120 billion parameter version yeah exactly you're pretty close like you're pretty close and wow now that you're almost pulling and then that's going to be kind of the next wave i think i know this is a weird answer

19:44Grant:convoluted but i think no this is great and i have follow-ups but you follow okay yeah i think

19:48Logan Lawler:the next wave is you know you kind of had cloud now we're kind of like data center and i think it's like truly local device because i mean yeah that 96 geek blackball i mean it's basically an h100 yeah in a desktop under your feet so that's crazy as that continues to accelerate i think that that'll be kind of the next wave but yeah i mean local compute is doing stuff that i mean six months ago was not possible. You're just like, nope, got to go to the data center.

20:18Grant:And I think the same is true on the model side, right? Like not only did OpenAI release the 120 billion version, they also released the 20 billion version. And Google has been releasing a lot of small or that level of stuff or smaller. Gemma 3N, for example, yeah. Gemma 3N, like really, really small models that allegedly, some of which you could even run on a phone. I mean, how you do that, I think, is still a little bit, People are like, yeah, you can in theory, but we don't know how. But there was like two things that I wanted to follow up on that. First is the cloud point. There was a great take.

20:53Grant:I'm forgetting who it was, but I'll credit them in the comments. Who was basically saying like we spent all this time putting all of our computing to the cloud. But then it sort of like defeats the purpose of the Internet being like decentralized. Right. And that like now all of a sudden everything is like, you know, going off to like these, you know, couple providers and it doesn't create that resiliency to your point. If like something gets hit, like does the whole eastern seaboard go down, you know, depending on where the data centers are? We don't know. So there is a very good reason to try and have as much localized compute power as possible.

21:31Grant:And to my earlier point, the models are also getting smaller. They're getting not only smaller, but better when they're smaller. which is a very exciting thing to to see and i think that the more that we specialize into niche models the more that we'll see smaller models get used even specifically for various

21:49Logan Lawler:fields in the industries yes agreed yeah and i and i think that's kind of another wave that is coming right is that we we talk about hey these models right we talk about you know what whatever it is name your model whatever but i think the age that we're starting to see and what's cool about local compute and what it can do, specifically in the Pro Max and others, right, is that people don't really talk, and it's kind of surprising me. People don't talk about, hey, this really small model from Google or this, but hey, how many of those small models can I load on one GPU that's under my feet? And can I do vision?

22:22Logan Lawler:Can I do generative? Can I do this? Can I do research? Where I built this, not a gentic workflow, but this multimodal workflow from smaller models that was purely only possible in the cloud, and no one ever talks about that. And I think it's so interesting because it's with model size, it doesn't matter to me. It can be big. It can be small, but it's really you have to have the right model for the right purpose. It's like I always make the joke when I'm talking. I'm like, hey, if I'm going to go, you know, cut my steak, I'm not going to take a katana and cut my steak in half. Right. But like that would be sick if you did.

22:57Logan Lawler:It would be sick. I mean, it would it would be awesome and it would probably get a lot of laughs. But I use a steak knife. Right. So use the right purpose built thing for what you're trying to solve. And then the other thing is the people, I hear about this all the time. And this might not be the right answer, but like a corporate audience there was like, should we train my own model? And I was like, absolutely not. No. Absolutely not.

23:20Grant:Do you have a nuclear power plant at your disposal?

23:24Logan Lawler:Like, are you going to be able to hook up a server farm to a nuclear power plant? Because that's what Microsoft's doing at Three Mile Island to train models. And the answer is no. like yeah think about rag think about potentially fine-tuning something like do that but there's other ways it's like building your own model like listen like meta and open air they are so far ahead

23:47Corey:that just no nothing you build because it'll be as good as the worst thing they have yeah oh yeah

23:53Logan Lawler:like the very first thing they probably built yours will be junk and you have to go hire a bunch of people are contracted out and there's just not enough data scientists in the world for everyone to start being a model builder like it's just right there's just not enough like i can't go build a foundational model i i guess the counterpoint though is like what what if you do

24:13Grant:have a really niche use case that you think you know you don't need like incredible data you know gigawatt level data centers to do you can technically train your own model on like a workstation or a series of workstations and what great question so when i say train i'm talking

24:30Logan Lawler:like foundational model you know uh open ai meta google level what you're kind of referring to is fine tuning and i'll give you a perfect example of something i did on a promax and it's a stupid use case but i think it's so cool because it it highlights the fact of what you can do on a workstation right as i was working with all my partners can't really name them um but they were working with basically a gravel quarry. And the way the gravel quarry operates is they sell rock by size, right? So it's three inch limestone or pebble or whatever. And the way that their model worked with their contract with this other company was, hey, we're going to buy a load of rock and it has to be within 10%.

25:10I mean, it has to be 90 % three inch rock approximately, or we get that load

25:17Logan Lawler:for free and if it is above 97 they pay the full thing of the load right so very old school industry right like not a lot of ai in it yeah and they were losing a lot of money because this other company was like well this is i mean i don't know if they were going in and like picking rocks out of the dump truck to validate that it was under 90 or whatever i don't know but right The use case was, hey, can we take a Pro Max, put it at the edge out in the quarry? Can we hook up a Broadcom camera to it? And the question was, is that we use, you know, Yolo and some other models and stuff, but it did not know what a rock was.

25:58Logan Lawler:It did not know what I mean, it knew what a rock was, but it didn't know what a three inch rock was. So what we did is we fine tuned it to learn what is a three inch rock and what is not. so that was my part that was the a part that was me but then there's some software stuff that went in but moral of the story is it was able to start and then run and it was taking i can't remember the broken camera it was like i don't know so many pictures a second like it was like hundreds and moral of the story is at the end um this is a software component it would tell the driver hey this road this load is good to go or it's not and if it wasn't they went and dumped it into like a mixed pile that they got 50 cents on the dollar for.

26:40Logan Lawler:So very cool use case you wouldn't have ever thought of, but that's what I mean. You can absolutely fine tune a model and where you're not teaching it everything. You're teaching it something very specific to understand not what a rock is, but what is a three inch piece of limestone rock look like? Yeah, right.

26:57Grant:Because you only need it for that very specific use case. Very specific purpose, correct. Yeah, we're not talking about science labs here.

27:03Corey:That's the rock industry. Yeah.

27:06Logan Lawler:No, I mean, exactly.

27:08Grant:Wow. Exactly. Quick tangent about that. I wrote for a company once that was doing basically the equivalent of that, but for drones. And so they have these like automated drone swarms that basically can survey different, like, let's say like in the transport industry, for example, like, you know, basically figuring out inventory levels, you know, checking grain, all that sort of stuff. They run on this sort of autonomous loop and they basically are trying to figure out all of these things. Like how much inventory do we have? Like, is there any theft going on? Really sick. There's so many cool things you can do with that.

27:40Logan Lawler:And then that's just training whatever model, vision model it is to learn what is a theft? What is an inventory mean? Like, is it, can I see this? Can I see that? I'm just making stuff up. Is it across all different form factors? Yeah, exactly. Exactly. Right. Exactly.

27:56Grant:Yeah.

27:57Logan Lawler:And that's not hard. I mean, you're you're feeding in, you know, representative examples of that. And you have very much a human in the loop to say, yes, no, maybe so. And you're ultimately training it. And no one is going to use or even know what I'm talking about. But the very first machine learning that I use personally was back. It would have been 2009. There was this company at Dell I was leading. So weird. Social media listening back when like it wasn't a thing where we were going to like boardreader.com and looking for people complaining about Dell. Well, we worked with a company which was called Radiant 6.

28:31Logan Lawler:It ultimately got bought by Salesforce where they would scrape every social media platform. So it's much like Google, but it would scrape social media to be able to then query against, you know, like, hey, someone said Dell. We could respond to them or whatever. But there was another tool called Crimson Hexagon where the biggest issue we always had with Radiant 6 was sentiment. It would be like, what is positive? What is negative? And there was no algorithmic thing to it. It was like, here's the words that are bad. here's the words that are good and if they use good or bad we know if it's going to be good or bad right well someone would say man that's a bad pc that thing's cooking probably positive right but we use the bad bad word it was negative anyways so i'll tell you is that crimson hexagon the way that they did it and this is much like ai and i use this example all the time is it would show you an example and say hey this pc is you know bad man it's cooking it's great or whatever, you would then say, is this positive?

29:27Logan Lawler:Is this neutral? Is this negative? And you would tune about 20, 30 examples. And then it would apply kind of machine learning across that to be able to say, hey, here is what we believe based on what you said is this. And that's really fundamentally what AI is. It is you basically having someone data science, the background basically say, this is what this is. and then you're looking at the output of that and you're kind of validating. It's kind of the same thing when you use Copilot. It says, was this correct? Was this a good generation? Every time you say that, that's just kind of a feedback in the loop to your question that ultimately goes back to the data scientists to make AI better.

30:06Corey:You know how AI coding agents feel fast until you're stuck fixing their code and realize it would have been quicker to just do it yourself? Warp changes that. It's an agentic development environment, a new kind of tool that makes working with coding agents effortless. Warp connects to your codebase, understands your contacts, and uses the best models, from GPT-5 to Claude Sonnet 4.5, to generate production-ready code from the jump. And when you do need to step in, Warp's built-in code review and editing tools make it seamless. You can see diffs, reprompt, or make quick changes right alongside the agent.

30:42Corey:No tool switching, no wasted time. That's why over 700 ,000 engineers from companies like Netflix, Ramp, and Amazon are already using Warp to save hours each week. So if you're ready to go from prompt to production faster, check it out at warp.dev. You know, something I've been mulling for a while, you know, with regard to buying a new computer is what is the bigger risk right now? Is it underbuying? Is it overbuying?

31:16Logan Lawler:underbying underbying i mean with without a doubt um i mean without a doubt and uh i mean i'll give you a good example right is that uh copilot for example um you know microsoft copilot is there's kind of the cloud version and then we have the the local kind of running version right but that requires an mpu i think and i'm going to butcher it's either 40 tops or 30 tops something it's more than just what was coming out earlier which was like 10 or 12 or 20 or whatever and if you kind of bought that system and you didn't get a gpu to supplement yeah i mean what are you going to do right like and i think the it's a thing where i mean i'm trying to like i'll give you kind of a good example is that when i i bought my last car and i don't buy cars i buy them new and then i drive them to the wheels fall off but i always buy the very best model that i can so 2018 this will seem normal to all of you now but it had parallel parking heck yeah it had lane assist it had crash detection now that is standard on all of today's cars where when i tell people oh i got that they're like logan i bought a car like two years ago that was all standard that's like free stuff yeah but in 2018 that was really but we didn't know right and it's kind of like with the pcs that i'm not telling you to go out and buy the top of the stack i'm telling you is if If you've never bought a GPU ever, go and make that one small couple hundred dollar upgrade to an RTX Pro Blackwell and get yourself four or eight gigs of VRAM.

32:53And I promise you, you're going to really be glad that you did, because with that, you're now going to be able to do things that one, you've never been able to do.

33:02Logan Lawler:But two, three or four years down the road, your computer will still be very functional, like from a hardware perspective, especially with models getting small. and things like this, you'll be able to use that for a very long time. And it's much like your phone, like with your phone, same thing. It's like, I want to get the best phone I can because I know that I'm going to get a bigger hard drive because I know that my daughter is going to get all my phone. She's going to take a thousand pictures and it's going to be full. I'm going to be pissed off about it. And so that's why I get something big to protect myself down the road.

33:33And it's like, that's what I'd say.

33:35Logan Lawler:Overbuying is rare. and but i would i'd also say that the risk i mean yeah if you've got the money over by and i'm not just saying that because it it's you're okay like you're not like in an ep like if you were saying hey like i want to go buy this you know i don't know f1 car and if you take on an f1 track it's going to go 200 miles an hour however fast they go but it doesn't really make sense to go in and buy it for a street legal vehicle because you're only going to go 35 miles an hour. But with AI specifically, the more VRAM you go, there's not a limit on it. Like there's not a limit. Having more means one more you can do the faster that you can go, the more unquantitized your models are, you know, all the type of things.

34:21Logan Lawler:So overbuying is not a risk. I don't see that very often. But I guess you're essentially just buying a major problem.

34:27Corey:Yeah, and I guess you're essentially just buying additional years of relevancy for the machine you're spending your money on.

34:35Grant:Exactly. Correct. Exactly. Yeah, I think about this a lot. So, like, okay, traditionally these days I have been a Mac user, right? Like, I have an M2, and every time a new M series comes out, I'm always like, man, do I just shell the$5 ,000 and get that, like, you know, 96 gigabytes unified memory or whatever. and every time I'm like no because next year there's going to be another one that comes out that's going to be even better and so I've been delaying and delaying and delaying but then now that I have a Dell Pro Max that has a 24 gigabyte dedicated GPU and I see all the models I can actually run on it right I like a lot of the stuff that I would want to do I could probably do on that right I could probably do into the future to your point about on models getting smaller exactly I guess the question is what like what is the real value I have a I have an answer to this but what is the real benefit of being able to run models on your own pc besides like come we've talked about comfy companies great you can make images you can make video but unless your workflow has to deal with that or you're just doing it for fun like like there's also lm studio which is right Which I use all the time.

35:50Grant:Yeah, you can run like a GPT-level AI, even one from OpenAI, the GPT-LSS, directly on your computer. And that's like you can run it offline. You don't have any data going over the cloud. And you own it in perpetuity. So, I mean, is that the ultimate use case here for local AI?

36:09Logan Lawler:I mean, here's what I'll say. And I'm going to give you kind of a two-part answer. is that I'm going to break it down for the business slash data scientist, then I'll give you kind of like the normal person, knowledge worker, right? The real value, and I see this all the time, is with companies that have a data center, have some GPUs in their data center. First off, as much as people might like to think, money is not unlimited at corporations. Meaning like, yeah, we might be worth billions of dollars, or we did, you know, like there is money. there's people that have to be paid like there's all these things the money is not unlimited and and i'll be honest most of the time generally is when you see things like i won't name the company but maybe a year and a half ago you saw maybe an airline um have some issues we won't say the name i fly them all the time but that's because they they did not update their hardware infrastructure and that's kind of the same thing that i've seen with the eyes they'll buy something it won't update but there's more and more people that are getting excited and want to access it the value of local is I cannot tell you how many people I've seen that are data science developer or whatever who are like man I want to get some time in the data center and they're backed out three months or they're backed out two weeks or they're like hey we can't afford you to run this experimentation for two days we will give you one day to the point where I know I'm not this not an open AI comment is not a meta comment but just general companies in general I think AI from their development and kind of some of the stuff they're trying to do is being impacted because the access to the data center and those shared resources are limited and it's stifling creativity like i work with this company to the point where they were using cloud and they were like oh my god like i i don't want to develop my tool anymore because i cannot afford another bill like i just can't to the point where their creativity and their thought and like it was shut down so the local ai value of that, let's use, you know, grant the Dell Pro Max plus that you have all of that experimentation.

38:10Logan Lawler:You can run on a Pro Max. Now, is it going to run as fast as the data center? Absolutely not. Because you're not talking hundreds and hundreds of gigs of VRAM. You're talking 24, but you can piece and parse out and do smaller experiments to test. And then instead of being stifled, I mean, dude, I mean, I don't know how long you work. I only try to work eight to 10 hours a day. I let that stuff run on my computer overnight. It's done. And then I kind of have my answer right that's the value of local because you can test and experiment it makes you faster it's safer all that um you don't need to use that to like justify getting the time at the big data

38:45Grant:centers you know correct you're like hey approve this it's impressive yeah correct and you're

38:48Logan Lawler:talking you know a five thousand dollar system versus you know you get like one of our you know like i'm not a server guy but an xc 96 70 or 60 or 80 or whatever the model is i mean it's hundreds and hundreds and hundreds of thousands of dollars because each one of those gpus is thirty thousand dollars like or more so i think that's kind of what i would say about the real value of local ai for like a business data scientist company the the real value of local ai and i'm gonna give you and there's so many cool things and like with comfy for a marketer dude it's great like it's great to use but honestly it's just fun and that's what i think of comfy ui is like it's just kind of There's this chintzy fun tool that like I can turn a Dell logo into a fire with a zombie in the background eating it.

39:37Logan Lawler:Like, you know what I mean? It's just it's just cool. But there's practical use cases. But the real value of AI and I don't think I mean, I think there's been a couple of seminal moments. Like I think when, you know, the original chat GPT open and people, you know, it was how many ever downloaded how many are used how many times in the first month or whatever. That was kind of a seminal moment. I I think the real seminal moment will become and I'm gonna be a perfect example is every use case for like a normal person or knowledge workers a little different right is like I I use comfy I like I said only because of I want to set things on fire and make cool little images because they're fun I'm the same and I and I just I don't have much value it during the day because that's not the job I do yeah but I also do you know the reshaping workflows podcast and someone in marketing was like hey Logan What you're talking about and the partners you have and video is great.

40:31Logan Lawler:We want to turn this into a blog post. I was like, dear God, because there's nothing more than I hate than to one, hear the sound of my own voice. And two is to set and listen to a 30, 40 minute hour long episode and type out notes. So I was like, how do I solve this problem? Because I don't want to spend an hour doing this. It's a waste of my time. Well, because I had a GPU, I was able to go find a free model from NVIDIA. It's called NVIDIA Parakeet was two dot. It was V2, but now there's V3. It does a lot more European languages, but you're able to very simply so easy. Well, maybe not. My mom could do it or my dad, but like most people can read instructions or read Reddit or can like piece things together.

41:12Logan Lawler:It was very simple, you know, set up a virtual environment on my PC, downloaded Nemo, instantiated the models, et cetera, to where I loaded a 38 minute episode and it transcribed that audio file and punctuated it and time stamped it in about 15 seconds and then i did exactly what you said grant is i put in lm studio he used you tried gpt's oss model asked hey what's a good blog post structure what's it look like hey here's the transcription which was 75 pages long and said hey write me a blog post that you know follows your advice and there's my blog post literally took me from downloading it to maybe three or four minutes versus an hour.

41:52And that, so it's hard.

41:54Logan Lawler:I think it's hard to describe the value of AI for kind of a knowledge worker or someone at home until they find or see that use case that really like speaks to them and solves a problem. And I don't think we've really gotten to that place. I think AI is really focused on a lot of corporations and stuff like that. I'll give you another example. it was at IBC which is the international broadcasters something something something it's in Amsterdam and I one of my favorite even though it is AI it's more computational but is Gaussian splats and nerfs right where you basically take a 2d image you turn it into a 3d reconstruction you can like go inside it visualize it so a lot of companies use it for like hey I'm going to build a 3d world of my factory or this part or surgeons will take a 3d scan of a patient to practice surgery.

42:42Logan Lawler:So we were like, whoops, I cut the wrong spot. Oh, well, that was the digital person, not the real person. So there's a lot of that. But I was at this and we had a really cool setup in our booth. And it was my buddy, Mike Rubloff, who's like, I'm not going to say the foremost world's expert on it, but he's pretty damn good, man. Like he knows what he's done. And he'd sent me a bunch of pictures for this demo, whereas him sitting on a bench in a park. And I was showing the demo and it just comes to life and everyone sees and you can go in and navigate around but the person was watching was a wedding photographer and they their mind was blown like you could just see it on their face i was talking and they're like so you mean to tell me that this tool which is called post shot and i can send you all the link to put it down below but i can take pictures because i'm a wedding photographer so you're saying i could recreate and capture these images of a couple's first kiss where they could be able to relive that in 3D for the rest of their life.

43:38Logan Lawler:And I was like, that's exactly what I'm telling you. And I'm going to sell that for$1 billion. It's more money. Oh, I know. I know. And it's all free tools, but you just need a GPU, right? And there's additional revenue add on packages. It's just a cool factor. And I mean, to be honest, like my wife will never listen to this. So I'm going to say it is we got married in 2011. I don't remember the ceremony at all, man. I was so freaking nervous the whole time. I know we got married on a beach. I can tell you who was there. But I don't remember our first kiss. I guess not that I don't love her. It's just it's been time.

44:12Logan Lawler:And how cool would it be to be able to relive that moment and show that to our daughter where you could go in and actually from every angle see that.

44:20Grant:It's so cool. And I think here's where the local part comes in because some of this stuff you were saying you could totally do on the cloud with OpenAI and Claude. You have to pay them. On your computer you don't have to pay them. But the key thing here is You might not want to upload your wedding photos to the cloud to OpenAI and Anthropic where they're going to keep it indefinitely. Yeah. We still don't know what happens through data when it goes in there. Correct. But you can do that on your computer with your own photos or like if you get permission to a professional, they could do it for you and give it to you.

44:53Grant:And it's done locally on their computer and then they can delete it. Correct. So you have control over your data when you do it locally, which I think is a huge thing that people need to remember.

45:01Logan Lawler:I agree with that. Let me make another point. it kind of tags onto that is that yeah in a business setting exactly what you're saying but let's just use the wedding photographer i'm going to go on a limb and i'm not trying to stereotype wedding photographers i'm going to go out on a limb and say not very many of them i think know how to go spin up an aws instance and like use s3 probably not right so for certain people in certain use cases hey cloud is an option but they it's complicated enough on a pc to do it and it's local and it's there and all your instructions there's no instructions on how to spin up pair because i look in the cloud i'll tell you that right because i looked that's true and it's a tiny model it's a tiny model i mean it's like i think yeah 500 million parameters 400 it's not very big it's it's tiny um but yeah no i mean i think that that is the real value because just i will never forget this i mean i remember this more than my my my kiss on my you know but uh it was just her face and just like the oh my god like moment and until we have either a seminal moment where it's something like chat gpt where everyone uses or people have these own kind of real epiphany moments that is when i believe ai is just going to be just up and to the right in the high hockey stick the ability to save an avatar of your lost loved one maybe correct you know that i mean yeah reconstruct an image of our dog who's aging who's 15 that hopefully will never die it'll be a very sad day but could you imagine being able to reconstruct

46:32Corey:him swimming in the pool when he was three years old yeah you know oh that's cool man like it's so

46:37Logan Lawler:cool yeah it's awesome and it and it goes beyond just hey cool little image it's like something from your life like it's awesome and that's where i think local gives anyone the ability but it just takes a little bit of like research or understanding to do that and that's why i love going to events and talking because just seeing those people's faces you know and like like that epiphany moment

46:58Grant:real quick what tool do you use to spin up like let's say a local model like parakeet or uh let's say like a what did you say gashi and blotty i've never said that word allowed i've read it okay

47:09Logan Lawler:so well two different things i mean it'll depend on what i'm doing right so there's i would and i would echo your sentiment like hey you're coming in you're listening right now to this conversation you've done less than one hour's worth of ai stuff in your life please do me a solid go to i think it's lmstudio.ai or something like that i can't remember the rl that is the that i i would say that that is like like we have a live tutorial on that we can link as well like okay perfect

47:44Grant:Yeah, link that. Yeah, it's so awesome. It's so awesome.

47:47Logan Lawler:So awesome. It is a meal delivery service. You just have to warm up the meal. You select your models. It does everything for you. That's kind of like level one. You run Parakeet through there? No, that's level one. Level two is what you're talking about with Parakeet, where you have to have a little bit of technical knowledge, where you're going to have to, for example, go to GitHub, clone a model. You're going to have to then set up a virtual environment within, I mean, I like Windows. I'm becoming more of an Ubuntu fan. That's neither here or there, but running either in WSL or Ubuntu or whatever.

48:20Logan Lawler:And you're going to have to, you know, download, start Nemo, you know, install your sub dependencies, instantiate the model. That's kind of the second level. It's still not hard. I compare that to kind of like a food, like a meal delivery service, kind of like HelloFresh, where all the things are out there and all the instructions are there. You just kind of have to follow it. yeah that's where i will do the parakeet or if i'll do like allura training or something like that that's kind of where i'm at that's middle layer that's kind of middle technical and then what i call like advanced is there are no instructions you have zero idea and you just go figure it out yeah and that is not a level that i would say many people are writing python in this case you're actually writing you're doing all that you're underlying you're modifying underlying codes like you're changing weights and biases you're changing the data orchestration layer it's stuff i don't think either any of us three could do and that's where you're talking about true data scientist machine learning engineer you know algorithm developer that's

49:21Grant:why they get paid so much but yeah wait until somebody makes a like desktop app that can do that yeah i know well i mean that's what i think those seminal moments are it's like the first time

49:33Logan Lawler:i use lm studio i'm like wait this is free you have all these models and literally all i have to do is navigate to this button and click which one i want to download and then i just kind of talk to it i was like yeah this is and everything i'm doing lives right here yeah and it's free yeah that's the crazy part's free i don't know yeah it's amazing forever you know like let's

49:55Grant:say there is an ai crash and all of a sudden you know let's god forbid one of these companies goes down or like the whole economy goes down or whatever, you will still have AI artificial intelligence on your computer no matter what. Like it is an investment in your future, which I which I that's why I think we talk about it here so much, Corey. Yeah, because it's something that like no matter what happens with the rest of the world, you can teach yourself how to do and you can have it as long as you want. Yeah, I agree. Yeah, I agree.

50:22Corey:It's your it's your apocalypse proof prepper tool.

50:27Grant:Yeah, no joke. people are downloading uh the entire wikipedia like to their computer like there's a tool that lets you do this and then they can spin it up and talk to it with rag honestly i'm gonna do that yeah it's a great idea honestly everyone should i'll include the link to the tool that lets you do this yeah i would i want that because that i mean and that's who would have ever thought of

50:49Logan Lawler:that use case right but it's a great yeah it's a great use case like it's like prepper guide 101 It's like you think food and like, yeah, you need those.

50:56Corey:But like, yeah, I didn't know how to make a make a meal out of an elm tree and pond.

51:01Logan Lawler:You know, do you know how to start a fire? Do you know without using gas or something? Like, do you know these things? Like, I don't know.

51:08Grant:Granted, if you're in that situation, you probably don't have a power to power your phone. Probably true.

51:14Logan Lawler:I mean, probably true.

51:16Grant:But you never know.

51:17Logan Lawler:You might have a decent battery life on your Pro Max or a solar rig or something. And I love doing that too, is like the thing about connected AI versus disconnected that I always say is, and I'll usually do this in any demo or talk I'll do is, um, to, for people to understand is to get those models. You have to be on the internet, but then once you've got it locally, I could literally, I could disable my wireless card. I could disable my Bluetooth card and it will all run. Now it will never get any better or worse. It's the same. It's like a balance sheet. it's locked in but it's pretty cool to see what it can do just by having that on there and i think that's a great use case i mean i would tell everyone regardless of what pc you have go get lm studio play around with it but we talked about this and i want to talk about this is like the prepping and like the future is and kind of a word of caution because i i thought about it after i had you guys you know on reshaping workflows is like the like it taking away our cognitive ability a little bit and and i and i thought about that a lot last night and and i was just like god and i'll give you the example that i gave you know when you guys were on the show was a mom my daughter plays competitive softball was like previously to you know chat gpt on your phone it was all the team wants to eat okay well we know we're in houston we know we're in baytown uh we know we want mexican food so then it's a quick google but outside of that you got to make the decision how How many there's all these inputs, right?

52:49Logan Lawler:How many reviews? What's the wait time? How many tables? Like, you know, is there bad review? Whatever. You have all these inputs. And then you have to use your brain to say which one fits the best situation for us. And it might take you calling or it might take reading a few reviews. And I think that's very good and healthy for our brains to have to make those little micro decisions. And but the mom, in this case, went to chat GPT and basically asked, hey, we're in Baytown, Texas. We want to eat Mexican. We have 20 people in our group. We want to make sure it has margaritas because our team's sucking right now.

53:20Logan Lawler:We need to drink. And, you know, we want a fun, fun atmosphere or whatever. Wow. And it was like, boom. Just one. Boom. Decision made. Now, there are times I understand that we all get very busy with life and we're very busy with life. But I'm I am concerned generally after I've spent a little time thinking about this is that it it's kind of like with my daughter. I've always tried to get her. I've never tried to give her the answer. I've always said, well, what do you think? Or, hey, what's your opinion? Or, hey, like, here's two different points. Which one do you align with? And I think that's very healthy for us as humans.

54:09Logan Lawler:Because to get to this point through evolution, it is a cumulative bunch of decisions that we have made as humanity that have gotten us here. But with AI, that decision making power is a little not there. Like, yeah, you can make the decision to use AI or not, but we will all eventually do the same thing. And I don't know. I'd love to hear your thoughts. I am worried that is it like an idiocracy type situation? Like, do you know what I mean? And that's what I thought about. And I was like, you're just not even thinking. Like, I don't know. It's scary now that I think about it. And you guys put that thought in my head.

54:47My first thought is that with every innovation, they've told us it was going to make us dumb.

54:54Corey:With the television, with the personal computer, with the mobile phone, with video games, with movies. We've always heard that. Now, that's not a perfect comparison to AI, which actually does the thing. However, we've come out of all of those, in my opinion, a smarter human race, for lack of a better phrase. And I really think there are absolutely people who are going to use this in a way that does that. but my hope is that with good education and teaching people how to use this to make yourself better I think there is opportunity there for people to be even more intelligent than they've ever been before as far as the idea of how to learn with the tool, how to use the tool to learn but like you know I did think very similar to what you're talking about last Friday night we've been trying to get my wife a real ID because we're trying to travel

55:55Logan Lawler:okay that's the one with the star on it right yeah that's the one with the star that you can't fly

56:00Corey:in america without my god so to get one i checked i was checking we live in suburbanst louis and i was checking dmvs and and like you know some of them they're all on different website systems oh yeah none of them book the same not all of them even have booking and i got so mad and i went to chat gpt agent and i finally said hey uh i need an appointment to get a real id on tuesday morning at a Missouri DMV within 150 miles of my house. Please go search them all. Come back with an appointment. And it came back with 120 minutes from my door. And we walked in at nine o 'clock in the morning and we were out in 20.

56:37Corey:And I was like, I will never go to the DMV without doing this again. I mean, I love that use case. Like that's a use case. It's like, here's a thing I hate. I don't want to deal with that. But I think to your point, there's there's definitely a risk of that. And there will be people that that absolutely happens to. I think the key though, is that we've got to remain curious as people. And I think that means when you're having a conversation with someone and they say something you don't really know about, you know, jot yourself a note and go ask Chachibichi to talk to you about it later. Learn about it.

57:09Corey:Use the tool to help you learn. Uh, even if it's just reading and talking to you, as long as you're still in taking that information in a way that is helpful. I think there are ways to combat this that will leave us better than we were before. That is the techno-optimist in my soul, of course, that really hopes that this is all going to end great and it's going to save the world. But I think we'll see people go both ways. Yeah.

57:40Grant:I'll give the bear case to that, but I'll try to not depress you.

57:46Grant:So I think that it is an industry-wide problem. I think it is a society-wide problem, and it's an individual problem. So each person is going to have to decide for themselves, like, where do I want to offload my thinking to the AI? Like, the DMV example is a great case. Like, yeah, it's not worth my time to do that.

58:08Logan Lawler:Transcribing my podcast that I already recorded, I had to prep for that. Whatever. I'm aligned with that 100%. Yeah, yeah, yeah.

58:16Grant:So there are those use cases that just make common sense, makes perfect sense. But let's say you want to be a software engineer and you're now offloading all of your thinking to Codex or Cloud Code and you're not learning anything that you're doing on the backend. There's a way to do it where you are, where you basically send a request, then you go read the code, then you ask questions about the code. What is this? What does this function do? then you read it yourself and try to say okay i understand you know i can see here to here it does this and that and you're wrestling with it but that still takes time that's not saving you the time that an agent would when you could just ask it to build you software and it builds you software the demo that they gave on monday was just like incredible how good this thing is getting um so the question is well on an individual level do i personally want to learn this or not or do i just personally want to have done this.

59:11Grant:But then when you apply that at the industry-wide level, all the incentives are there for them to create tools to basically do everything for us because they can sell it to companies. And then on a society-wide level, well, we want the GDP growth, well, we want this and that. We want to improve technology. And we're in a global race to create AGI, whether or not you've subscribed to whether we should or shouldn't do that. like so there's all of these incentives that is pushing us to like automate as much as possible so like you know maybe the companies are gonna you know try and automate as much as possible maybe governments are gonna try and automate as much as possible I think it has to be something that we build into the design and we think about ethically and right now I don't see any incentives to do that.

59:57Grant:I agree with you Grant like and

59:58Logan Lawler:this is not meant to you know cast aspersions on us as humans but generally not everyone but a lot of humans will always take the path of least resistance.

1:00:07Grant:Yeah.

1:00:08Logan Lawler:I have done that. We're designed for comfort. That is what, that's where we're at. We're designed for comfort and, you know, safety and all this kind of stuff. And you're right. There's a lot of economic incentives, individual incentives. There's not going to be that person. It's like, Hey, I want to go play college softball, but I'm not going to put in the work. Well, there will be people that will put in the work, but that's why there's only a very small percentage that do. Right.

1:00:30Grant:Yeah.

1:00:30Logan Lawler:I think the, the, the, the solve of the problem. And you bring up an interesting point. I won't talk about the coding agent because that's a whole other thing, but you know, and using the use case of like, you know, answering questions for my daughter. Right. I think it's something very easy where it's instead of, Hey, instead of me giving you the answer, let me give you three alternatives. And you just code that in to the point where, Hey, here's three things. And then I can then ask a question about like, Hey, what's the benefits of each one. Yeah. And like being able to then think through all of that and help your reasoning process and then yeah you could blindly pick one sure but like i think humans natural curiosity would say well i got three like let me ask and see what it says and i think that's something that we should think about and like not and i hate government regulation i'm not saying that but like just ethically something that should be put in because i just i don't know man i and i did watch idiocracy and i'm just like i watched it two are we getting funny enough dude i know i was like i watched it on an airplane i was like oh my god like are we literally getting our law degrees at costco here like because because think about like gbt yeah chat gbt like i mean think about coding right go back i mean a couple years ago you either one had to learn it i mean you may not know what's the school for it but you had to learn it and you had to practice it and you had to do exactly what you said grant you know you had to learn and understand and all that that's what made you great as a software you know dev or engineer or designer or whatever or you went to school and added more traditional path certification whatever but now and this is the scary part for me is i don't have any traditional training in writing python code but with claude and other things that i use i can write enough or i and i can use it enough to fix little gaps and things that i'm doing which would have absolutely roadblocked me versus you imagine giving that to my daughter who knows nothing and can play around with it that is a and it almost i mean it's weird that we haven't seen an acceleration of people who have more of these skills you know what i'm saying that's kind of weird you think like we never have enough engineers we never have enough you know software devs we never have enough this but now you have a tool that can more or less do the damn job for you, but we're not seeing an explosion of those people, unless you tell me I'm wrong.

1:02:52I think you're going to see more people needing to become, I jokingly like to say that I'm

1:02:58Corey:tech literate. I can carry on enough of a conversation. I can tell what I'm mostly looking at when I'm looking at code other people created, but the idea of me just sitting down from scratch and opening up a terminal window to do it is not a thing that's at least currently in my wheelhouse. like, but I can recognize problems. I could change dimensions. I could swap a library out. I could, you know, some of the basics like that. And I think there will come a time where just those simple skills will be enough to carry you a long way. But, you know, essentially somebody's still got to be around to, I guess, fix it.

1:03:40Corey:But, you know, when you think about coding even, and I know this is a rabbit hole, you know, we were talking about how you could go back. You could go back 10 years and that's when you could start going to boot camps and stuff and learn coding, get a real job. And 10 years before that, you were in a university, though, where you were taking the most intensive math courses the college had to offer. And working your way through linear algebra and calculus and a variety of other things, probability and statistics. and they kind of streamlined around a lot of that into people who are working and have been for a number of years.

1:04:16Grant:So I think like to Logan's point about you know you're not seeing a lot of people you know all of a sudden explode and have this skill and get hired. I think what's happened well there's been multiple studies and like think pieces about this that basically says the people who have the skills are getting better. Yes. And people are able to learn new skills faster, but there's not as much of a transition from like people who have new schools skills that are learning them into the roles. It's more so like we can do less with, with, we could do more with less, um, not less people. Although I think there might be some of that happening and there probably will be a lot more, uh, you know, like I'm me as an engineer.

1:04:59Grant:Now, if I'm like a senior dev at some company, I can do a lot more. I don't have to go hire someone to do it. Uh, so what I, My personal stance on this and why I think everyone should have local tools and be teaching themselves as much as possible is I think everyone is going to – like this is going to make a flattening of the job market, if you would. I think there's going to be more small companies. And sooner or later, there will be a tool that can help people go from end to end to launch software. And you will see this explosion of shovelware, as it's called. It's like where you're just shoving it out the door.

1:05:37Grant:But we just haven't got there because it's still, with coding, it's still at the level where you do need to know all of the ins and outs to deploy it and actually get it running and actually share it. And distribution is always going to be the problem with a lot of this stuff. So it might exist, but you might not have access to it.

1:05:56Logan Lawler:Well, I think you made a good point. And, you know, my opinion is, is that, you know, I don't want to eliminate jobs like not down for that. But you're very much right is that some people do take that approach. But you can definitely do more with less less. Yes. Right. And I think that you made a point earlier that I think was really interesting is that and I don't know why, like and I'm not talking about the comfy UI type thing. But if you think about just general AI tools, I read about manure on every day, right? The treats to try, you know, if I'm being honest, like, and you see this stuff, it is very enterprise corporate focused.

1:06:31Logan Lawler:Like I, at least I get that feeling. It's like, if you were to go and survey a lot of the companies in that, it would be like, Hey, we're, we're B2B. There might be a few B2C, right? But most of it's B2B. And I think you're right. And the point you made earlier is that a lot of it is kind of focused on, you know, the churn of businesses, like being able to, like, get yourself in the door and all that because you have money. But I think what's underserved is a lot of B2C use cases, like generally. And I think that that is even, like I said, back to the original point, it was like the value of local compute.

1:07:02Logan Lawler:And I think ChadGPT does both, but it was kind of consumer first, right? But it's these tools that, I mean, being able to put, you know, a sensor in your fridge that does computer vision that ties to your phone to tell you that you're almost out of milk and you need to reorder it.

1:07:18Corey:Or that you could make Hamburger Helper tonight. That's all you're able to make. Yep.

1:07:22Logan Lawler:Exactly. Exactly. Exactly. I love that. I think we'll see a shift towards that, too. Because, yes, business spending power is great, but consumer spending power is much more.

1:07:34Grant:That's true. At least I think. Yeah, I think we'll see. I think it's just, you know, like to your point, people go where the money is and B2B is like where the money is with a lot of that stuff.

1:07:45Logan Lawler:When you're selling to one business versus 100 consumers, right, for the same amount of revenue or whatever.

1:07:50Grant:Exactly, exactly. It's slightly easier in that regard.

1:07:53Corey:Well, Logan, before we let you go, we always like to ask, what's in your personal AI stack right now? What models are you using, tools along those ways?

1:08:04Logan Lawler:um well we kind of talked about parakeet we talked about gossian splatting which i'll send you the link of that post shot is amazing um you know post shot is a free tool it's called joss at post shot um i'll tell you real quick is that it is it is ai-ish because there is the the computational ai to put because you're basically taking let's just say you know you've got my you know my bottle here you're taking images all the way around from different focal points. And even though those are separate images, what happens with a neuroforgasm is finding it all kind of a point cloud is it all puts it together to make a 3d representation.

1:08:38Logan Lawler:Right. And I really love any free tool. I use Ellen studio all the time. I use that. I use parakeet. Um, you know, I, I'm a big, another big one that I use and I don't know if anyone uses it. I feel like it's kind of under the radar and maybe it's not, but I did this project and I can't, well, I'm going to allude to what it was and then I'll give you the link to it but so basically we're going to this animation studio and very popular uh they have shows on Netflix they've been around for 25 years um maybe vampires I don't know you can figure that out and read between the lines but um they wanted what they wanted to do with AI was hey to create a storyboard is like really expensive it takes a lot of time and no one really likes to do it can you take every show that they've done so many shows all of our shows all of this every episode and create me a laura that would and it's kind of artsy right i'm not artsy but um the line density shading color gradient i capture all this where we could use that to impart like hey we're going to do this other show to kind of storyboard out what we wanted to do right yeah and so i use my pro max train to laura but a lot of people that do lauras they do god what is it like the one i can't remember the name of the one tool like everyone seems to use it like um it's like one laura or something it's like what's what everyone uses i use kohe for a while ko h y but i found this one i don't know if anyone uses you should it's called one trainer have you ever used it no no i'm not dude bro download one trainer from github uh it is dope it is this is awesome it's one the the the gui for it way better it allows so much more tweaks and parameters actually gives you because kohe gives you zero instructions like it's kind of like you take the standard you play around with it and if you don't get it right it just doesn't work one trainer's got really good documentation I mean, you're, you're able to change weights.

1:10:42Logan Lawler:You can go precision up, down, multiple GPUs. Like you can actually train instead of training, like in Koei, like I would train, okay, here's all the images, right? You can do concepts. So I could train it by concept and use keywords for concept versus like stock keywords. Cause you don't know you have all these images, like, and you have to tag them and all this, but I knew what they were. So I can, I use the concept feature. It's great. It's free. and the other great part about it, Koei will kind of tell you how much longer is left, like sort of, it'll kind of give you like a range, but this one will give you like, here is exactly where I'm at and it'll let you click and show a test of your training anytime you want.

1:11:23Logan Lawler:So I'll click at 20 % to see the image. Then I'll go at 40, 60. So you can actually see it develop. And as you develop, I've gotten to the part where I burn stuff in, where it's just like gobbledygook because I over-trained it. You can play around and stop it at 60 because you're like, oh, this is perfect. Stop right here.

1:11:39Corey:Yeah.

1:11:40Logan Lawler:Dude, one trainer, man.

1:11:42Corey:I've already pulled it up. I'm going to check it out here later this afternoon. Yeah, one trainer. You got me excited.

1:11:46Logan Lawler:Yeah, one trainer, man. We'll include a link. Yeah, do it, man. One trainer is good. And that's the other thing. What I love about AI is I love the homegrown, open source, free, where you've got some guy or gal who's just really damn smart who just took their time to develop this. and that type of stuff before you ever go out and buy anything if you can find something like that i would much rather use that because you're going to be able to talk to that person you're going to get support you're going to go to the github page ask question they'll respond i would always look for some yeah please help uh yeah exactly always look for the free stuff first and i mean obviously if you're a business and you have security whatever figure that out yourself but for the consumer i'm talking to consumer right now yeah look for the free stuff first because i will almost guarantee you something really good exists, you're just not looking on GitHub or Hugging Face or in the right spot to find it.

1:12:41Grant:All right, let's do a couple of lightning round questions here. Like just off the cuff, just your general thoughts. So what is the next two to three years outlook like for hardware and AI?

1:12:54Logan Lawler:I mean, I'll say this is that we adult typically release systems on a two to three year cycle, right? Right. And if I was to predict, say, three years out and I know NVIDIA is kind of development cycle, I think there's two big things that you're going to see is typically NVIDIA, every other GPU dedicated GP release. And this is not like insider information. You can go research this yourself is most of the time at the top of the stack. There's improvements every time. But like Ampere, six thousand forty eight gigs. ada 6 000 48 gigs now blackwell went up to 96 i think what you're gonna see in the next round of gp release it will not be the top of the stack will not be 96 i i don't say i don't know if it's going to double but i think there is the demand and they have the ability to kind of take that a little out of cycle the other thing that we're going to see three years from now and this is Once again, I'm going to qualify it.

1:13:55Logan Lawler:Logan Lawler's opinion, no insider information, is that NVIDIA at GTC announced, you know, Spark, which is their founder's edition. And we will be, I don't know when we're releasing this, but at the time of where we have not released our Dell Pro Max GP10. So what you're going to see, and this is Logan's looking, is traditionally AI has been local, has been very much dedicated GPU. there's a reason why nvidia i believe is going to more of this system on a chip arm design both in their server products and their local workstation products microsoft can google it uh they to power their data center they went out and bought a nuclear power plant at three mile island like and that's not an exaggeration that is the level of power they bought three mile island i mean like literally for the nuclear power plant and dedicated gpu servers cooling all of that ai will only reach a certain space because there's not enough power in the world unless we want to

1:14:59Grant:just drop a nuclear power plant in your backyard however with the system on a chip kind of arm

1:15:04Logan Lawler:design it is not as powerful yet but it is very uh very power efficient yeah what i think you're going to see is three years from now two and a half years from now we launch new dell pro max products instead of having a separate intel or amd processor separate rams well hard drive will always be separate but ram gpu and this you're going to see products that are all integrated in because of the power that's why for example i hate you for this grant but you add your mac right why is it awesome because the battery life's great that's because it's an arm design that's why tried to buy arm like i think that's where they're going because they know for nvidia to scale and ai to continue to grow for from a let me say from a computational standpoint if we want to reach agi we want to do crazy things you you can't build nuclear power plants everywhere like you just can't like how do we get more efficient so two big things is you're gonna i think you'll see an out of cycle dedicated gpu jump vram and i think you're gonna see a lot more nvidia specific i I mean, it's also interesting investment in Intel.

1:16:13Logan Lawler:I don't know. Just throwing it out there. Just throwing it out there. No insider information. Seriously. Just hypothetical guesses like everyone else. But I think that system on a chip you're going to start seeing. Yeah. Become the commonplace.

1:16:28Corey:That whole integrated memory idea where it can be either your RAM, your VRAM. It's all one bank.

1:16:35Logan Lawler:it is one integrated component that has i mean rgb 10 is 128 gigs of dedicated cpu gpu ram and it levers up and down dynamically based on the work and task that you're doing um which is very cool if you're doing ai i mean it loads a 200 billion parameter model on this pretty dope like in terms of what you can do um but you're hey you're doing something that's more memory intensive great are you doing something more cpu that's great too and i there's a lot of advantages with it because you're not and this is what i think people don't understand is and this is not like a rapid fire question i'm giving you like all these technical but um but i know sorry next one i'll make more rapid uh but in a in a system you have your processor your ram your gpu that data has to go bing bing bing bing bing bing that takes time and yes the gpu accelerates that parallel processing all that but that transition it takes time and now it's millisecond it's like no time at all but it still slows things down that integrated chip it's there yeah there's no there's no bouncing there's no delay there's no latency there's no lag yeah and i think that's the other benefit of a system on a chip is not just the power but it's the speed memory bandwidth

1:17:51Corey:and all that kind of stuff so 100 next one i promise will be shorter would you say that this

1:17:56Grant:is them redesigning? Because I've been thinking like computers need to change, like the operating system needs to change, like the hardware needs to change for this new paradigm if this really is computing v2. Do you think the system on the chip is kind of like that hardware, that's how hardware is going to change?

1:18:10Logan Lawler:I think that you're, specifically with AI, you're going to see a lot more system on a chip. There are a few things, like for example, and I'll say broader with our team, you know, GPUs support any parallel computing, CUDA enabled like engineering, ISVs, Media Entertainment, you know, Adobe Premiere, all this kind of stuff, right? That's all Windows-based. None of those run on, you know, none of those really run on Ubuntu, right, or Linux. So I think that what you're going to see is a shift, and I'm going through the shift as well. I was a DOS guy back in 88, and then I've been a Windows... Remember DOS?

1:18:45Logan Lawler:Oh, I remember DOS. You're putting in the five and a half inch floppy and did it, you know? Oh, it is great. But I've been... on the windows journey from kind of the very beginning. And now I'm definitely on the Ubuntu train because I have to be. And I think what you're going to start seeing is that like WSL and other kind of emulators are going to get a lot better where you can do, you know, AI Linux Ubuntu work. But I think you're going to see more people just from either from a development standpoint, develop for windows or people are going to have to start learning Ubuntu. Yeah. Cause that's, that's the only way it's going to work.

1:19:21Grant:I agree. I think at some point, they've got to train an AI model to make it so easier to port to different systems. Honestly, that's a great use case. Yeah. Someone should make this. Why is no one making this? Make it really easy to port from Windows to Linux. Make it really easy to port from Windows to Linux to Swift.

1:19:38Logan Lawler:Honestly, that's a really smart use case because it's not that you know how to do it, but it's defined in the way that you need to do it. And you can start. I mean, that's actually a great idea.

1:19:48Corey:Viewers, go for it, Bill. Hey, somebody listening, go do it.

1:19:52Logan Lawler:And then pay grade for the idea.

1:19:55Grant:Train this. Because you see all these cool tools that will launch on Product Hunt and it'll be someone doing Mac only. We have codecs. We have all these coding agents. Why can't you? A solo developer can only pick one right now. But that should be the most obvious use case is make it so that it can be universally ported to other systems. I think that's genius. And phones. And phones. Exactly. Agreed. 100%.

1:20:19Corey:Well, Logan, thanks so much for joining us today, man. This has been so much fun. Where can people go to find you on Reshaping Workflows?

1:20:27Logan Lawler:So I will give you all the link because I don't, like, it's basically, you can find us really on any podcast platform, you know, Apple, Spotify, and then the usual, like, Pocket Cast. Like, you know, I have an agency that does that.

1:20:40Corey:All other places.

1:20:41Logan Lawler:Yeah. All the, we are everywhere. But personally, and it's up to you, you guys can feel free. I love talking to people and I'm pretty comfortable doing it. You know, if you ever want to talk AI or have questions or need hardware or whatever it is, I'll give you my email. It's just Logan, L-O-G-A-N dot Lawler, L-A-W-L-E-R at Dell.com. And same on LinkedIn, just Logan dot Lawler. I'm the goofy guy that might be catfishing you a little bit with my old picture, but hey, it's, it's representative enough where you'll be able to find me. But yeah, I mean, I'm out there travel a lot. If you ever see me stop and say hi.

1:21:14Logan Lawler:And yeah, this was great, man. I appreciate it. And I, like I said in the beginning, I was trying to blow smoke. Like I love the neuron. Like I love what you're doing and really educating people about AI. And I am not just hyping you up. Like I read the newsletter every single day. I learn a lot from your newsletter in terms of thing. And so I think if you're, if you're listening and at the end of the day, you don't want to jump in and start developing or do whatever, just educate yourself.

1:21:42Grant:Yeah.

1:21:42Logan Lawler:Get the newsletter, sign up, listen, learn. and at some point you'll see something that's going to like that example I gave before about the wedding photography and Gaussian splats it'll be something that's relevant to you and you jump in and then you can kind of start but I would definitely check out the neuron the newsletter

1:21:58Grant:is great listen to the podcast thanks Logan we really appreciate that yeah we love your show too and it was great going on and really appreciate having you absolutely I really appreciate that so we'd like to give a quick thank you to our sponsor of this video whisper flow make sure to go check out their awesome AI transcription tool, whether you have an issue that makes typing difficult or if you just want a great transcription tool so you don't have to type all day. Go check them out.

1:22:21Corey:That's it for today, everyone. Farewell for now, humans.