local AI's biggest problem... they solved it

Boxing Legends

Local AI’s Biggest Problem… They Solved It! 🤖🔥  Local AI is becoming more powerful, but one major challenge has held it back. In this video, discover how a breakthrough in local artificial intelligence could make running AI models on your own computer easier, faster, and more efficient.  Learn about local LLMs, running AI models on a PC, GPU memory limitations, VRAM optimization, open-source AI, offline AI tools, and the future of private, on-device artificial intelligence.  If you're interested in the latest AI breakthroughs, powerful AI tools, running ChatGPT-style models locally, and the future of generative AI, don't miss this video!  👍 Like, 💬 comment, and 🔔 subscribe for more AI news, tutorials, and technology updates.

Transcript

00:00This is the first Wi-Fi GPU running an NVIDIA 5090, and it's somehow over Wi-Fi showing up as

00:06this laptop's GPU. I'm playing Cyberpunk on it right now. Also, I'm running Blender on it,

00:11but that's not even the craziest thing it's doing. Over here, I have a Mac also using the GPU. Not

00:15only is it generating images and videos very fast, all local AI, but it's also talking to this

00:20DeepSeq 4.1 Flash, a 750 billion parameter model running at full quality on a box with 32 gigabytes

00:26of VRAM. This doesn't make any sense. You would normally need over 500 gigabytes of memory to run

00:31this. It's called Wi-Fi, wireless computing infrastructure. Get your coffee ready. I cannot

00:35wait to show you how it's doing all of this. Now let's get this right out in front. This video

00:40is

00:40sponsored by Wi-Fi, but don't let that deter you. I heavily vet every sponsor and product I look at.

00:44This one is crazy, which is why I said yes. And here we are. They are a new company and

00:49they're

00:49doing something pretty bold. And when they told me about this thing, claiming to be the first

00:53wireless GPU, and it can run trillion parameter models. I had to try it. Like really, does this

00:58thing work? And in this video, we will be trying everything, video editing, gaming, and impossible

01:02local AI inference with image and video generation. Like it really doesn't make any sense what it's

01:06doing. And when I got it, like when it arrived, it was inside a suitcase, inside a box full of

01:11bubble

01:11wrap. That's how new this thing is. And here it is an early engineering model inside this box is an

01:16NVIDIA RTX 5090. There's also a mini PC as well, and it's actually running open WRT for its networking.

01:22That's how it shows up on my network. Now, if I get brave enough, I will undo this thing.

01:26So if you're seeing that B roll now, I did it, but I'm going to save it for the end

01:29of the video.

01:29So I don't break it before I make this content. Now, the big idea with this thing is local,

01:33no cloud, right? I'm already sold, but they taken it a step further, a step that I didn't think

01:38I wanted to take. This isn't just one device with a crazy, powerful GPU. Like I already have that

01:43in my studio. Instead, they just have the crazy GPU that multiple devices can use. Think about that.

01:49It's like a family GPU kind of like how we used to have one family computer for everyone.

01:53Yes. That was a thing. Comment below if you had one family computer. So the idea is that if you

01:57have a laptop, a tablet, a phone, it can now use this desktop class GPU, like it's its own.

02:03And it's not just throwing a job over to it. It's actually using it like playing cyberpunk on this

02:07little bitty laptop really hits it home for proof. You can actually hear the jet engine over here taking off.

02:15Oh, this thing's taken off and seeing that 5090 show up in the task manager, like it belongs to

02:20that laptop is so weird and kind of wonderful. How are they doing this? Let's actually hit that first

02:25because there are two things that were breaking my brain and I have to figure it out. How are they

02:29doing this over wifi? Because wifi is not the greatest connection and how is this laptop suddenly

02:33believing that it has a 5090 and we'll get to the whole trillion parameter model thing later.

02:37Now this will honestly be the craziest part of this video. I even built this visualizer that will show

02:41you exactly what's happening with the model on that GPU. Now, by the way, have you hacked the

02:45YouTube algorithm today? Let's make sure you do hit that like button, subscribe, notification bell,

02:49comment. You got to hack YouTube today. Ethically, of course. Okay. So let's talk about your GPU.

02:53Typically your GPU, your gaming GPU needs a very, very fast way of talking to your PC. And it does.

02:59Here's our fun little 5090 flying around. Can you keep an eye on it? It's fast. When you slot that

03:03sucker into your gaming PC, this is going to be a gaming PC right here. You're putting that into the

03:07PCIe port.

03:08This sucker's 512 gigabits per second answering requests in under a microsecond. Now let's get

03:13this gaming PC out of here. We're using a laptop now now over wifi instead of PCIe, we're talking

03:17about one gigabit per second answering at around 1.5 milliseconds. And that's under the best conditions

03:23because wifi is kind of, you know, and that is the technical term. Now don't miss this. It's this

03:27right here, this round trip time, the time it takes to answer those requests that becomes a huge problem.

03:33And it actually might not seem that bad at first, right? Like 1.5 milliseconds. That's super fast.

03:37No, it's not. If you compare that to a microsecond, I'll give you an analogy that helped my brain

03:41process this. If PCIe handed you a picture in one second with wifi, that would take 25 minutes.

03:48And that's a big problem because the game and the GPU, they're pretty chatty. They're sending

03:52instructions and data and answers to create a frame thousands of times every second.

03:56If every one of those answers had to wait on wifi, we'd be playing with PowerPoint, but we're not.

04:01Look at this 92 frames per second over wifi. And I'm rendering this in blender. And this is not remote

04:07desktop. It's running on this laptop on this GPU over wifi. How did YSI solve it? Well, they created

04:14their own protocol. The team at YSI has a networking background. So that tracks. The first thing they

04:18had to do is convince this laptop that he has a real GPU, that he's actually using this 5090 over

04:22here.

04:23Like this is your real GPU, buddy. Even though you have no room for it. They did that by installing

04:27a fake

04:27driver. And what's crazy is cyberpunk. The game treats this GPU. Like it's right there. Like it's the real

04:32one. It does not know the difference. Cool. But they still didn't solve the problem of goodness.

04:37This right here is super slow. This communication. So what they did was kind of genius. They just told

04:42them to stop talking so much. Like literally that's what they did. They figured out which messages

04:45actually need answers and then stopped waiting on the rest for instructions. They bundled them,

04:51compressed them and sent them without waiting. So for data, it only sends what's changed

04:55and doesn't keep repeating itself for answers. It doesn't even wait for the question. It knows

05:00what's going to be asked. So it just sends it. This is the wifi version of that meeting could

05:04have been an email. And then the GPU sends compressed video back to the laptop. These four things they

05:08did took what would actually need to be like eight gigabits per second needed to now all they need is

05:13400 megabits per second. Proof is in the pudding right now. We are gaming at 92 frames per second.

05:18Look at the wifi right now. It's about 400 megabits per second going across and get this out of your

05:22brain.

05:23This ain't cloud gaming. This also isn't remote desktop. The game isn't being streamed to us from

05:27the cloud. The game is still happening right there on the laptop, but it's using the GPU over there

05:33over wifi, which is just the craziest thing to say. Now what's funny is some of the team at YSci

05:37actually used to work in cloud gaming. I work in the cloud gaming team intensive before.

05:41And it was around the time that game, uh, black myth wukong, I think that's how you say it came

05:45out

05:46that they had this idea, the YSci idea. At a moment of black myth, we were thinking,

05:50can we still leave the front end on the laptop, but pushed back end to a remote place. That's very

05:58started. But then they discovered something kind of cool. And later on, we found that AI has a way

06:03more scenarios than gaming. Okay. This is the part that got me most excited when I first heard about

06:08YSci. Sure. The wifi gaming is neat, but local AI that's where things are happening. That's where I'm

06:12most excited about everything. And this is legit a new way to approach using big models with local AI.

06:19So YSci actually did two things, solve two big problems with local AI. I think they might be a

06:25game changer. Prove me wrong in the comments, do it. And some of this might be news to you because

06:29I'm not like a GPU expert. I don't study this stuff, but I did on this video, by the way,

06:34these videos take a minute because when someone sends me a product like this, I don't just go neat.

06:38Here are the talking points. And like, I actually get into it. I learned the technology. I learned

06:42how it works so I can pass this along. And when it's something that I think is interesting,

06:46I do want to pass it along. I get excited. So the two things that you actually solve

06:50or two problems that exist when you're already familiar with, but this first one,

06:53you're probably not GPUs don't like to share what we'll cover that here in a moment. And two,

06:58this is the obvious one consumer GPUs can't run big models because they don't fit.

07:02They don't have enough memory and yeah, they solve that. So for the first one,

07:05GPUs don't like to share. What does that mean? I mean, sure.

07:07This 5090 is great at running local AI. Have you tried it? It's fast.

07:12Hold up Chuck from the future here. I'm going to do a quick edit on this video for you because

07:16right

07:16here, I started to talk about how the GPU doesn't share, but I cannot wait for you to see the

07:20local

07:21AI stuff. So I'm going to move that right over to now. I just want you to see it. And

07:26also I want you

07:26to see that visualizer I made. It's so sick. It's like looking in the brain of deep seek.

07:32Here you go. So again, our premise is that you can't run large AI models at home,

07:36consumer GPUs. Like the one inside this box just don't have enough memory to run them yet.

07:40This box can how let's break it down. Now, first local AI is in a fun space right now.

07:45You can run a ton of amazing models on pretty much any hardware and you can do some cool things.

07:50But when we're talking about the smart ones, the frontier, like ones that get close to what

07:55you might experience with the Opus 5.5 or GPT six. And there are local models that are like that,

08:00like Kimmy K three or GLM 5.3 or deep seek 4.1. These local models do give you a

08:05level of

08:05intelligence. That's pretty dang close to these frontier models. And the best part is that it's

08:10unlimited and it's free and it's yours. It sounds great, right? Yeah. Well, you can't do it.

08:15Normal people can't run this stuff. It requires a lot of memory, like a data center level of memory.

08:21And have you seen the price of memory? The model I want to focus on right now is the deep

08:24seek 4.1

08:25flash to run the sucker full quality. You need over 500 gigabytes of memory. There is nothing

08:30consumer level that can do that. No GPU, no gaming PC that can support that much system memory.

08:35Even the new M five max studios with 512 gigabytes of Ram can't really run this comfortably. So again,

08:41how can this thing run it? Because we're not using memory. We're using the SSD because inside this box

08:47right now is a four terabyte SSD and it's using it like memory. Now by the numbers,

08:53this is a terrible idea. It's such a bad idea. Why are they doing this? Sure. SSDs,

08:59they give us more capacity. You most likely can buy a four terabyte SSD. How much are they right now?

09:05Field trip, four terabyte SSD on new egg. That's going to cost us 500 bucks. So capacity,

09:12we're getting that, but speed we're losing. And that's our main problem right now. And you need speed

09:17when working with AI. Let me show you what I'm talking about and how bad it normally would be with

09:21an

09:21SSD. If we didn't have what Y site was doing and what they're doing is pretty sick. So just for

09:26a

09:26comparison, like what we're talking about, because there are different types of memory.

09:29I keep throwing them around memory V Ram, which is the Ram that you're going to see inside your

09:34GPU. When we're talking about the 50, 90, we're going to be around 1.8 terabytes per second,

09:39wicked fast standard system, Ram, most modern stuff, DDR five, 90 gigabytes per second. And then we get

09:45down to the SSD 14 gigabytes per second. So if V Ram hands you a page in one second, Ram

09:51will take 20

09:52seconds. The SSD will take two minutes. It's like that sloth from Zootopia. How does Y site deal with

09:58this? First, I want you to imagine the best scenario possible. Imagine that somehow this 50, 90 has

10:04550 gigabytes of V Ram. Can you imagine? No, you can't, but we're going to do it right now.

10:10With that, we can run the full deep seek 4.1 flash with 500 gigabytes of required Ram.

10:15It also has 750 billion parameters. Pretty smart dude. And in this dream situation,

10:21speed is our ally for every word or token that this deep seek LLM generates.

10:25It has to transfer all 750 billion parameters to the chip. This equates to roughly about 508 gigabytes.

10:33This transfer right here, that's 1.8 terabytes per second. That's given us 3.5 words or tokens

10:39per second. That's great. But if we face reality and we get real sad, we have to realize we can't

10:44run

10:44a model that size purely on a GPU and we have to put it on the much, much, much, much,

10:48much, much,

10:49much slower SSD. Let's play that out. So goodbye, 550. We now are back to 32. By the way, this

10:55is

10:55actually going to be worse than you think. Watch. So we can actually still use the GPU in this measly

11:0032 gigabytes, but we still have to put 475 gigabytes of these parameters of these weights onto SSD.

11:05And again, for every single stinking word, we have to transfer all the weights from the SSD

11:13to the CPU. Now I know all you nerds in the comments, you're going to go, actually, Chuck,

11:16it's supposed to go to the Ram first and then to the CPU. And I know, but just for simplicity,

11:21we're just going to remove that stuff. Okay. So for one word, we transfer it, it spits it out.

11:25What's the damage? We might think, well, Chuck, we're getting 14 gigabytes per second, right?

11:30No, no, we're not. We're not even getting that. This is the rated speed for large transfers.

11:35What SSDs are meant for at least hundreds of kilobytes. What's actually happening is the

11:40transfers here are small, lots of tiny transfers, four kilobytes, millions of tiny four kilobyte

11:46transfers per word per token. So we're actually getting more around one to five gigabits per second,

11:53which means at its worst, oh, we're getting eight minutes per word. Yeah. Unusable, right?

12:00But I know what you're doing. I can see you typing in the comments right now. You're about to type,

12:05well, Chuck, actually deep seek 4.1 flash is a mixture of experts model. So we'll actually be

12:10faster. And you know what? You're right. Deep seek 4.1 flash is not one of those dense models

12:15where every parameter is used for every word or token. It's an MOE model or mixture of experts model,

12:20where only some of the parameters are used for every word or token. By the way, most of the best

12:25local open source models are mixture of experts like Kimmy K3, Quinn, GLM. And it does that by

12:31dividing the model up into experts. This one has 384 experts and kind of imagine each one being good

12:37at something like a few might be really good at answering questions about badminton. I don't know.

12:41It's not quite like that, but it's going to help illustrate it. So if you asked it a question,

12:44like what the junk is badminton, the model actually has a router built in. So let me actually draw this

12:49out. It's kind of fun. It has a router built in and out of the 384 experts, it will choose

12:53the top six

12:54that are really good at answering questions about badminton and they will be used to generate the

12:58next word. Now, this is amazing because we'll watch this with only six experts being used.

13:03We only need around 15 gigabytes of weights and files to be transferred for each word

13:07instead of the massive 508 gigabytes. So in our fictional world where we could put the entire

13:11model on one GPU and it had 550 gigabytes of VRAM, things would scream with that MOE model.

13:16We're talking eight milliseconds per word, but in the real world with an SSD at its worst,

13:21it might be around 15 seconds per word better, but it still sucks. And there's no getting around this

13:27because even if we are only using six experts to generate the next word, you might think, oh,

13:31we just put all those six experts on the GPU. It'll be fast. No, we don't know which six are

13:36going

13:36to be chosen for each word. So we have to store the entire model. So finally, we're at what YSI

13:42is

13:42doing to make this manageable, make it work. And it's actually pretty sick. And by the way,

13:46you made it to this. Think about that for a second. You learned a lot just now. You learned

13:51that file transfer speed is actually really important for LOMs. And it's kind of our main

13:55bottleneck right now. You learn what a mixture of experts model is and why it's kind of awesome

13:58for local AI. Now learn this YSI did four things. First, they rewrote the model files for the SSD

14:04drive so that every expert is the same size aligned chunks, which sounds gross, but it is a good thing.

14:10By the way, if you call me chunk in the comments, I forgive you, but it happens often. And now

14:14that I said that,

14:15I know my editors are going to have a lot of fun with it. With these chunks aligned,

14:20they now speed up the transfer reading at 64 chunks at a time. I've been saying chunk too much.

14:25Also pretend these are 64 chunks having too much fun with this. They essentially optimize the model

14:29file so the SSD can transfer them faster. And now that everything's chunky, sorry, we've gone from one

14:35gigabyte per second to around 11.5 gigabytes per second. That's a huge jump, but we're not done yet.

14:41Remember how those six experts are chosen by the router for every word? YSI keeps the

14:45most popular ones sitting on VRAM and RAM. So if most of the experts that are going to be generating

14:50the words are coming off VRAM and RAM with this design, only around one gigabyte worth of experts

14:55is being pulled from the SSD. So even though it is the slowest thing and they've, they've done their

14:59job to speed it up, we only have to worry about this much. And lastly, they split the math. If

15:03the

15:03weights are coming off the SSD or RAM, the CPU is going to handle it. But if the weights are

15:07coming off

15:07VRAM, the GPU is going to handle it. With these four optimizations, let's see what actually happens.

15:11Oh, how fast is it? I'm gonna run some deep seek y'all. It's getting really hot in here with

15:18my

15:18gaming machine and this bad boy keeping me toasty. Okay. Time to test the sucker. Now, a couple of

15:22things are happening right now. I didn't think I could do this. It's actually pretty sick. What

15:25you're seeing on the screen right now is a way for us to visualize what's happening when we're talking

15:29to deep seek on this GPU. We can see what's on SSD on RAM, GPU, and watching the experts be

15:34chosen.

15:34Hey, shot from the future again. I just realized I did not tell you what this thing is actually doing.

15:39It's not just some fancy graphic. I'm actually running this off of that wireless GPU. And we're

15:45actually looking into the brain of deep seek and how it's being used on the GPU and the SSD and

15:50the

15:50RAM and the VRAM. It's actually watching it pick the experts. And I did this by looking at trace logs.

15:55It's a whole thing, but it's running live and it's awesome. Anyways, I just had to say that. Now,

15:59one thing you might be like, uh, Chuck, that doesn't make any sense. It says all 508 gigabytes or 15,

16:06thousand 360 experts live here. Wait, Chuck, you said there's only 384 experts. Yes. But that's per

16:12layer. I simplified it in the explanation, but the principle still stands. Now I'm recording also on

16:17this Mac, which is wirelessly connected to this GPU. Let's load up deep seek ready, set load. Oh,

16:23we can see stuff happening. Look at that. Whoa, look at it go. This is pretty sick. It's ready. How

16:28is coffee roasted? Explain that process. Oh, look at that. Whoa. Look at all the experts being chosen.

16:38Notice most of them are coming from the GPU, dude. And we're getting like, look at this,

16:43like 10, we were at 15 tokens a second, our expert tiers full of coffee experts,

16:48but getting this level of speed off one GPU running off an SSD. Are you kidding me? By the way,

16:54I'm so proud of this quick little coding idea I had to visualize the experts being chosen.

17:00It's so awesome. Let's try a different question. Boom. And there it goes, picking the experts.

17:04And notice at the top, it's reading around 10 to 11 gigabytes per second. It's fluctuating though.

17:10For some reason on this round, we've slowed down a bit around three to four tokens a second.

17:14I wonder why. Maybe I can ask a new question in the actual web interface I have here.

17:19Why are a coffee beans green? Oh dang. It's going fast here.

17:27Look at that go. Reading the question, man, it's actually very fast.

17:32And notice 61% of the answers are coming from the GPU. Oh, this is fun. Hey,

17:37it's Chuck from the future again. Now it's finally time to talk about how GPUs don't share.

17:41And I bet you didn't know that here it is. GPUs don't like to share. What does that mean?

17:45I mean, sure. This 5090 is great at running local AI. Have you tried it? It's fast.

17:50I watch over wifi. I'll connect to it right now with this Mac. Tell them Mac. Hey,

17:53this is your GPU now. And I can generate images with Figma, which I don't know you could do that.

17:57I'm not sure how they did that. This is really cool. And I can make videos with this love art

18:01app.

18:01And I can even edit videos with the Vinci resolve and then render them.

18:05So I'm editing on the Mac and then I'm rendering it with that massive GPU. Like that's just cool.

18:10But these are all like single tasks. And if we want to treat the GPU, like that family computer,

18:15it needs to be that one centralized resource, which is their big idea. I think it's a fun idea.

18:19The GPU has to be able to share, but it won't, or at least it's not designed to it's stingy.

18:24It never actually had to share. The idea is that, you know, when you're playing a game,

18:28the game is using the GPU. Like that's what's happening when you're rendering a video,

18:31that's using the GPU. Most of it, no one taught it how to share because it didn't need to share.

18:36But now it does because all these AI apps are like, Hey, I want it. I want to play with

18:41it.

18:41I want to use it right now. They all want to use the GPU at the same time. And the

18:45problem is it

18:46takes time to switch because any one of these models, image model, video model, chat model

18:50can fill up the entire 32 gigabytes of VRAM. It has just by itself. And if you want to switch

18:56the app or the model, you have to move that sucker, all of it out of VRAM and then move

19:02the other

19:03one in. And if we're talking about the shared family idea, that's untenable. Like that's crazy

19:07slow. Now YSI is solving this by building what they call a scheduler. The scheduler runs on the box.

19:12And what it does is when you're want to switch an app, let's say we have the image app figma

19:15or

19:16something, but you want to switch to video generation. The scheduler will move that model down

19:20to the SSD. It's kind of like putting the app in its pocket. Didn't put it down. It didn't put

19:24it

19:24away. It's just somewhere else. And when that app needs to generate another image,

19:28it can quickly bring it out of the SSD and start it up again. In theory, this should make sharing

19:34the GPU a much better experience. And it makes it feasible for a phone or a laptop or a tablet

19:39to share it all at once. But that's just a theory. Let's test it. Now I'll code it up a

19:43quick demo.

19:44So you can actually see this visually. Here's the GPU and its memory SSD right here. So you can see

19:49the parked copies. And I'm actually going to show you without the scheduler first to show you how it

19:53kind of feels the speed difference. So we'll load the chat. So what it's going to do is it's going

19:57to load the Gemini model right now and ask it a quick question. And this is the Mac using the

20:01GPU scenario right now. Now I'm going to switch to an image model and it's going to generate a quick

20:05image. Ready, set, go. Now here's what I want you to pay attention to. And it's when I switch back

20:10to the chat app, let's switch back. It should take a minute.

20:16Notice it was pretty much the same time as loading it from scratch. I put it on my desk now,

20:20so you can

20:20kind of hear it when it spins up now, time for the parked option or the scheduler option. And look

20:25right

20:25now it already has the stuff loaded. So it's got the chat app loaded. It's got the parked models

20:30on the SSD, kind of like a copy of them, just ready to go. So I'll load the chat. It's

20:33already loaded.

20:34So it answered right here, as you can see, boom. Now I'm going to switch to image and let's see

20:38how

20:38fast this is.

20:42I mean, look at that.

20:46So, whoo, calm down, buddy. Before it took about what's 18 seconds. And this took us about 8.6.

20:52Let's switch back to chat. Boom. Woo. What took 17 seconds before. And now it takes around four

20:59seconds. Let's try again. Switch back to image. Pretty consistent. Okay, cool. So the family GPU,

21:08what do you think for me? I love the vision. I know what this can be. And after using it,

21:13the numbers don't lie. Like what they've done is impressive.

21:15Why site has created a bunch of amazing novel solutions to consumer GPU problems and running

21:21local AI. And it's not just that, like being able to game on a wireless GPU, like that's crazy,

21:26but then also run giant AI models that nobody has any business running by themselves on something like

21:32this and getting around eight tokens a second, eight words per second. That's good. This is an early

21:39engineering model. I imagine it's going to get better. And then using figma with this, that was cool.

21:44And actually that part I'm excited to see more of like mainstream apps, being able to use this GPU,

21:48using an app on your phone that believes it has a 50 90 and it's using it. Like that's crazy.

21:55The tech is exciting. I nerded out like crazy digging into this to figure out how it works.

22:00And by the way, you can buy one of these right now. They're selling them. It comes in this version

22:05with a 50 90. They also have a 50 60 TI version, which is a bit cheaper. Here's the pricing

22:09right here.

22:10They'll ship in about eight weeks. So get your orders in now, but man, this has been a wild ride.

22:16I'll catch you guys next time. Hey, you made it to the end at the end of my videos.

22:20I like to pray for you, my audience. Now, my camera's about to die.

22:23So I'm gonna pray for you real quick. Father, I thank you for the person on the other side of

22:26the screen.

22:28I thank you for them just being here. This is not an accident to be at the end of the

22:32YouTube video.

22:33Who makes it to the end? So God bless them now and their family. Um, I thank you that they

22:39are

22:39excited about tech. And I pray that this, this local AI thing, even if they're not gonna use one of

22:47these Y side devices, just the idea that local AI is accelerating, let this get them excited about

22:52what they can do in their community. Uh, what they can do for their, their home, what they can do

22:56for

22:56their businesses, what they can do for their church. God show them the cool ways we can accelerate

23:02humans, 10 X humans and do things that we could never do before and let them not be anxious about

23:07it, but be excited about it. God revealed to them ways they can use AI, but also at the same

23:14time

23:15reveal to them things they need to keep human,

23:20slow thinking and learning. We need that.

23:27Bless them indeed. Again, bless their families and their lives and their jobs and everything.

23:30I ask this in your name, Jesus. Amen. Whoo. Thanks for that. Thanks for letting me do that.

23:35I'll catch you guys next time.

Category

Transcript

Recommended