BONUS: AI Tool Roundup: What’s Actually Worth Trying?

40:00Some of them require coding. Some of them are right there and ready. But this just makes it native. It's right there. It's where you expect it. It looks like a little globe. You just click the button, and off you go. Oh, Corey, before I move on, Jamie Sepulveda asked a relevant question. When you connect something like Unslock to Claude or Chat, does that cost against your subscription when you need to purchase tokens? No, these are free. These are local. It's running on your computer. Unsloth is completely free. These local models we're talking about are completely free. And what you will do is use these so it doesn't burn your tokens.

40:41So not every task requires the same model. So you can use it to, you know, maybe it's, like I said, maybe it's web surfing tasks. Maybe it's information gathering. Maybe you're using it for bigger stuff than that, but you could from Codex call this, and this model will run for free instead of one of the OpenAI models that is pulling from your subscription quota. I hope that makes sense. I can do that again if that's not clear. No, that makes sense. I want to add something to it. So one thing you could do is, say, for example, you use Codex, which is OpenA as coding agent. You could say to Codex in its rules, hey, make sure that whenever you have a task that's pretty simple and well specced out, that you assign that to Quen27B to go do for me.

41:34And then that way it will send a request to Quen. when we'll go off and do it, and then you're not costing your OpenAI credits on that. Exactly. I want to real quick address this question as well, because it's a good question. How does Unsloth compare to OpenClaw? A thing to understand is that they're actually very different things and could be used together. OpenClaw is a harness in the same way that Codex and Claude Code are. Could you explain that? Yes. Essentially what it is, is a place that houses is your model and a bunch of tools so your model can think of it like giving your model arms and legs to go do stuff like a spider and and uh and it sits here you know in this little nest and it's sending calls out it's doing different things like it's going to go check your email it's going to write you a google doc it's going to open up a spreadsheet from your boss that kind of stuff now the thing to know about an open clause you can set that up to where it runs on your codex subscription.

42:33I don't know about your clawed subscription. There was a big fuss there about that at one point, but it may have, it's probably fine now. Uh, yeah, we'll circle back to that. Cause I, I don't know. Uh, but the deal is you can also connect your open claw to local models on your existing computer. So you can use both of them. You can use all of them. That's one of the things that made open claw so cool when it first came out was just the, the diversity of what it allowed like it would that wasn't a thing we had had before where you could just connect every app you own to one harness and uh you know and maybe harness is the wrong word for anyone who's not technical hub think hub like it's like your ai hub yeah yeah yeah i'll allow it thank you i think harness is perhaps a little bit more on point just because it's like it controls it like If you think about it like a harness for a horse, you're steering the horse.

43:32That's fair. You're pulling it this way, pulling it that way. It's a similar thing here. It helps you sit in the driver's seat to make sure that the agent doesn't go off the rails. Exactly. Exactly. Well, let's hop back over here real quick. Models that actually run code, of course. That's a given. The latest models. One of the nice things is it has this discover feature where you can go in. And it's going to immediately recognize your system. It's going to know you have this much VRAM, this much RAM, this processor. It's going to know your machine immediately when it gets there. And you're able to click on device, which you see right here.

44:12And on device, we'll immediately grab any model you've ever downloaded before, no matter what app it's in. Like it shows all of my LM Studio downloads just immediately. But in Discover, you can look at other models and see what else is out there. And you can usually filter these by, can I run it? What you'll notice is these little, there's going to be these little check marks. You see right here, you'll see this little green circle with a little letter I in it. That means you can run it. Sometimes they'll be yellow, and that'll mean, like, you might be able to run it. And sometimes they'll be red, which means, give it up, bucko, there's no chance at hell.

44:52You better pull up OpenRouter if you want to use that Exactly But at least you know And once you come in this section right here Where you see this UDIQ4XS That is one of the quantizations And there might be For example for this model For Quinn 3.6 Or 3.627B that we've been talking about That's not right is it? No 3.827B is that there'll be, when you click this, there'll be a dropdown and there'll be 25 different codes like this that don't mean much of anything to you, but it'll tell you - The number matters. So if it's two, if it's four, if it's eight, that just shows you like how much smaller it is.

45:36Yes. So two is the smallest, four is probably what you want most of the time. And each one will have this. Each one will have this and you'll be able to tell really simply by just a glance. I can run that one. I can run that one. And, you know, you want to pick one of those and go from there. What else do we have here? Access your models anywhere. Yeah. Remote access is a function, which is delightful. I have not tried that yet. It's it's intimidating, but it's awesome. Have you done it? Not in unsloth. I mean, I do it with my codex constantly. Exactly. Same. I was getting tires fixed a few couple of weeks ago.

46:15It's sitting at Walmart waiting on tires. And I had like four agents going at home through my ChatGPT app on my phone. I was like, this is wild. This is the future, man. This is the future. Yeah, for people who don't know, basically remote access lets you say you're using codecs with files on your local computer, which you can do. You might not know this. when you use code work or cloud code or chat to these codecs or chat to put you work it can actually access files on your computer and via sandbox and then you can actually turn on remote control and then that means you can access it on another computer or on your phone and so that's how cory is able to on his phone talk to the agents that are running on his home computer now the agents aren't necessarily running there they're still over the cloud but for example if you're using Unslock, the agents would actually be running on his computer physically on there.

47:09This is one feature they don't call out on the front page. And I want to call it out because if you've ever played with a local model, you know what a problem this can be. And that is self-healing tool call. Please explain it. Yes, I'm going to. When you are using especially smaller local AI models like 4B, 9B, getting into The smaller it is, the more you'll notice this. What happens is those tool calls that run when you're like, I want web search. I want you to go grab a file. I want you to do this. Sometimes those tool calls get corrupted.

47:47What happens is there's a glitch or there's a something, and basically it's coming through as code, and it just goes blah and messes up. And what this does is this detects it, repairs it, and sends it on through like nothing ever happened, which is really, really sick. And a thing that I'm really excited about. Okay. One question from Paul Clay Design. So for people who don't know, there's another tool similar to OpenClaw, which we mentioned earlier, called Hermes. And so Paul Clay Design asks, can you use Hermes to connect directly to local models like Quen 3.827B? Yes, you can. Absolutely. You could either do it this way directly through like an Unsloth or an LM Studio or an Olama, whatever local AI tool you use.

48:39Or you could go to Open Router and just get an API connection to that model and use it through there that way. But absolutely doable. Yeah. Yeah, the one thing to keep in mind about OpenRouter is it's still over the cloud, right? So you're still submitting your data, and it aggregates different cloud providers serving the models. So you don't technically know where you're sending that, like what provider you're sending the data to. So just keep that in mind. And, you know, if you're using it for non-sensitive stuff, I'm sure it's fine. but if you want exactly you're using it for a sensitive topic you might want or sensitive information you might want to use something like unsloth for that for sure and i'm dropping in i did a full written review of unsloth desktop this week and i'm going to drop that into the chat right here for anyone who wants to check it out later on um yeah i've had a lot of fun with this tool it's been really cool um the two things i want to flag the two things i want to flag about this before we move on so number one in my own experience testing it the biggest problem with a local version of quen 3.8 27b is memory and this is something that you're going to come up against so they've solved tool calls they've solved a lot of other things they have not solved memory so we just want to talk about tokens roughly equivalent to words you can fit inside your chat before it loses.

50:06And unlike OpenAI and Anthropic, who do this thing called compacting, which helps it compact the information so you can continue talking to it even after you hit that limit, Onslaught just kills it. You hit the memory limit and you can't talk anymore because it just can't handle it. And part of that is just simply a limit of the way these work right now. I run into it occasionally in LM Studio as well, especially with depends on the model some models run heavier if that makes sense and are just a little more difficult and Quinn 27B kind of does because it's dense because it's good the flip side is I can't run it well but man it's good another thing the second thing is reasoning so everyone says like you need to run this thing at the lowest reasoning effort possible because it thinks a lot and that will then lead to the memory problem which means it will spend a ton of time thinking about the answer and then it'll burn through your tokens and then you'll hit your limit and now you have to start a new chat because it can't handle anymore so those are my buy an rtx 6000 right now which by the way you should know grant remember the other day i told you it was 13 000 oh no don't tell me it's more like 17 now oh my by by the end of the week uh i I was looking at them last night for$16 ,900,$17 ,500,$15 ,000.

51:34It's getting to the point where buying a local machine that you can run is like buying a car. And it's going to help you with business. You can at least data center space for less than what it costs to buy an RTX 6 ,000. Yeah, but in the same way that you buy a car to go to work, so there's a benefit to the cost, right? You know, there is a benefit to buying a home data center or home server, let's say, like with a really nice graphic card. And it will help you with work, especially if you can use a lot of these local models and do tasks that you would need to do. Otherwise, spending money with OpenAI, you know, you could save some of that money.

52:13But dust off your checkbook, buddy, because it ain't cheap. But it's like buying a car. It's like$70 ,000,$15 ,000. Yeah. By the way, just a quick note, if you haven't yet, please take just a moment to like and subscribe. We really appreciate it. We love doing those. We love being able to interact with you all, answer questions as we go. And that helps us continue to do it. And also, just another quick shout out to A-Refs. Go check out their new tool. We got the link here in the chat already for you. We really appreciate them sponsoring today's video. All right. What do we have next, Grant? What do we have next?

52:54Let me pull down my screen. Sorry, I've got it. The next thing on the list is, so we're going to talk about some of these that came out, but we're just going to roughly overview them, because I don't think you've used either of these. I haven't got a chance to use these much yet, but they are Cursor Origin and DeepSeek Harness, which do you want to start with? We've been talking about harnesses, so we can talk about DeepSeek if you want. That's fine. That's fine. I have not touched either of these two tools yet, So I'm anxious to to hear a little more of your take. OK, so let me make sure I'm sharing the right thing here.

53:33So big news.

53:38You're cutting out pretty much, Grant, but it might be me. Oh, let me know. OK, I will. Which one of us is choppy to you all? If one of us is choppy, let us know and we'll know whose internet is messed up. It's me. I don't think I can share right now. I got really slow. I think you can share this link.

54:09Corey, you think you can share that link? Screen share that link? Yeah, the cursor. Oh, DeepSeek. I sure can. One minute here. Okay, it's Grant that's choppy. Good to know. Thank you, Rodwin. I appreciate that. It's hard to tell when it's us because if his screen's chopping up, it could be my internet. I never know.

54:34Okay, I'm getting the screen up here. And welcome to DeepSeek. By the way, the new DeepSeek's really good. I used it last night and was quite impressed. Yeah, what did you think? uh it was good it was reasonably quick uh i say that i threw a couple bigger things and it took its time a little but uh i was really happy with the answers uh honestly it was as good as anything else i touch it seems like like i really don't have any specific i know this is a problem yeah like i wasn't doing anything this is why we're not supposed to have this yeah yeah we're really we're really reaching the point we're like yep it's another good one yeah it's one gooder than the other one was

55:24yeah it's starting to get to the point where you know just what do you like better like what do you like using and that may be a field thing i expect it'll be a field yeah like models attraction was feel yeah like like i think a lot of what drew people to claude initially was was was the vibes. Well, anyway, I think I'm kind of cutting in and out, so I'll keep this brief. Basically, DeepSeek Harness, the thing that you need to know is if you're even remotely technical or interested in it, essentially what you could do is you could take the open repo, which is the code base, and you could use this to run a model like Gwen.

56:06You could customize it, or you could use it with DeepSeek, for example, and you can use this on your, you know, for whatever your coding purposes would be. It would be like replacing Codex or Cloud Code with this model. And this whole idea here that everything is a plugin is like the biggest idea from this. And it basically just makes it super easy to use, to connect this thing to anything else is the simplest way to explain that. Yeah, yeah, that everything is a plug-in thing is really cool. Models, tools, skills, sessions, sandboxes. Like, people are used to working in that way. In terms of, like, connecting things via a quick plug-in, via, you know, OAuth or something to hook something up.

56:59And I think that this is definitely the way those things will go. and like it may be that and maybe that answer is is like some kind of an mcp thing i mean i'm sure the plugins are mcp but maybe the answer is that that's that's how we connect all of these things in the future is just just a quick like you know plugging up your tesla at the charger it's just going back to everything is just code that talks everything is good yeah because the agents know how to do it you just say hey you know and that's how i recommend normal people use this like if you actually want to run something like this you just take the link that i just shared in the chat and you say you know set this up for me and you know take that to codex or cloud code or whatever quentin 27b might not be able to do it but you could use one of those other ones i just mentioned and just say like hey help me set this up so i can run my own version of this like help me clone it and, you know, try to pick it out for a spin.

58:01This is such a great explanation. The model is the soul of an agent. A harness lets an agent understand its environment, use tools, and keep working in the real world. I like that. The model is the soul. I think that's a really good way to put it. Also, just a heads up, this is in developer preview right now. so it's it's not like a final version um but it's it's getting it out there to get people using it so they can uh figure out ways to make it better there's a certain point when you're building where you've got to get people to use it so you can figure out where it's broke uh cory would you be able to share uh origin really fast as well we won't spend much time i sure could i sure cover this for normal people i sure can one minute uh let's see here I think I shared the link in the live chat.

58:57Yeah, I've got it here in the in our doc as well. Perfect, perfect.

59:07Okay. This is not it. Origin. Where is it? Where is it? This is like the... Okay, hang on. We're going to do this the old fashion way, guys. Brace yourself. Google. I never use Google anymore. I hear this. It's on their landing page, but they just auto-redirect you there. Yeah. There we go. GitForge for the agentic area. This is Cursor's GitHub killer is what they'd like it to be. From what I've read, it may or may not be, but go ahead. Go ahead. Yeah. Yeah. No, GitHub is basically like Google Drive for your code. It's a lot more powerful than that, but it's where you host your code on the cloud so that you can work with it with your agent.

1:00:01So, you know, traditionally you can write files on your computer. You can still do that, but then to save a copy of it and then continuously make edits and changes to that code and keep track of all of that and be able to reverse if you break something. you can do all of that on GitHub using this tool called Git. And so this is Cursor's version of that. So if you're familiar with Cursor, if you use Cursor, you can basically use them to now host your code instead of GitHub. And that's what manages like your software versioning and things like that. It's a good off-site place to store your code as well.

1:00:39It allows multiple people to work in and out of it at a time. You can serve it publicly. you can do all of the neat things. I don't have a lot more information on it other than that it came and it got people excited. Yeah, I would just say if you want to learn more about how Git works and GitHub, we're going to have a whole live stream about that next week, so tune in for that. Yeah, and one second. Here, we're going to bring up the one of these that I have been having just an obscene amount of fun with. And that.

1:01:18Is going to be this one buzz can you see the screen is it coming through yeah sorry i've got i've got my video blocked so i wanted to peek over and make sure i'm not on the wrong tab this is introducing buzz buzz has been out well close to a month now i'll be dang it doesn't seem like that long and uh this is a really really cool tool so you're all familiar with probably i assume slack teams google messenger or a variety of things like that, ways that people in the workplace keep in touch with one another and communicate back and forth. Well, Block, which is Jack Dorsey's company, has built one of those for agents, and it is sick.

1:02:04It's literally Slack for agents, so you can go in there and you can build agents. It talks here that it's a free, open-source collaboration platform where humans and agents work together in a shared workspace. You can connect your Claude code. You can connect your Codex, your DeepSea harness. You could connect anything you want. And if it doesn't offer it yet, mark my words, they're going to add it. I've been watching these guys build this in public for the last month. And they're shipping updates like day after day after day after day. As people come back with complaints and concerns and bugs and ideas, they ship them as fast as the ideas come in.

1:02:43And this is really, really neat. I'm looking for some screenshots here because I wanted to show you what it looks like. Actually, let's go over here to the GitHub screen because I think there are... When you get down here... I've got to scroll past all this stuff. Yeah, here's what it looks like. It looks like anything else you've ever used. Can you zoom in on that image a little bit? I'm going to see. Yeah, there we go. Yeah, there we go. There we go. And it's just like anything. Now, in this version, you know, you've got humans working in here, but your agents, which you build in here, operate as first class citizens the same way you do.

1:03:26So they can go and work in the channels you allow them to work in. They can communicate with one another. You can at them, as you'll see right here where it says at Fizz. Can you turn that into a clean three beat capture plan? And it's like, absolutely. And here it is. uh so there is there's just a ton of let me go back because there's more pictures you can it has everything a slack type tool has you can do channels you can do announcements you can do dms you can you have an inbox and you can go into the agents tool itself and uh and set them up like it'll have a few dummies in there you can use and i just picked i i i started with three connected to each of the three GBT 5.6 models.

1:04:13So I think they're named Honey, Fizz, and Buzz are my three. But I then created a fourth that runs on Soul, and its name is John the Delegator. Hat tip if you get that, if you've ever watched Sons of Anarchy or happen to know good old music, because that's an old song. And so he basically bosses them around I message him and I'm like Hey John, I want you to do this I want you to go build a landing page It needs copy, it's going to need all the things Plan it out and then get to work on it And then you see a message pop up right below you Like he's typing, just like in Slack And all of a sudden there's a message come out And he's tagging Honey And it's like, Honey, I need you to go make a plan Fizz, I need you to go do this this is very similar to using Telegram.

1:05:11More organized. It feels more organized. Yeah, so if you use Slack, you can organize into threads and channels. Telegram, the interface is just messages. It's not very neat and tidy. It's not as easy to basically compartmentalize different things that you're working on. Or catch the volume of spam that Telegram has a way of doing. That's my biggest issue with it. I get constant spam. I love it. Messages from random people. Yeah. But this, there's, as with that though, there's obviously a mobile version that's available for every operating system already. And free and open. And the way that you use connector agents is through the API thing we were talking about earlier.

1:05:57Yep. So you can connect to OpenAI, you can You can connect Open Router. You can connect to Quend via Unsloth. You can put it all in one place. And if you do it on your desktop, it's just a process. It's just a click-through process. It just takes a minute. Yeah. Excuse me. I think you could probably even have your Open Claw and Hermes agent chat with you in there. No? Here's the other thing. Check this out. Every message, reaction, workflow step, review approval, and get event is a signed event in one log. Same shape, same identity model, same audit trail, and whether the author is a person or a process.

1:06:33In practice, it feels like a team workspace. Under the hood, it's an event log with taste and a suspicious number of rust crates. This is so AI-generated. Yes, it's another AI-adjacent developer tool. We're sorry. The difference is what agents can actually do once they're inside. They can open repos and patches. I mean, that's all the stuff. I will say they beat me to this. I was trying to build this so now I don't have to now I don't have to it's alright, of course he needed more money he's not making money off of this it's open source no he's not he's giving back to the people but you can go and have conversations with your projects you can go have triage a bug without giving it the keys to the kingdom have their own keys oh yeah, agents have their own keys as well.

1:07:26So not just the model they're using. And that's a really big deal. Scoped by identity, not permission flags. So yeah, you can also give them varying levels of access, just like you can with a human user. Yeah. So let me give you a concrete example there. Let's say you're using GPT and Quinn together in the same workspace. Maybe there are certain things that you want Quinn to have access to and not chat GPT because it's like on your computer local, you're working on your trademark project, you don't want to give that information to OpenAI over the cloud. And then maybe there's some things that you don't want Quinn to access because Quinn's not smart enough and it might F it out.

1:08:08The crassness. And then maybe you want to keep Quinn off that project and you want to put OpenAI give OpenAI access to that. So there's lots of ways you could do the permissioning. I like that. That's something I missed the first time around with this. Yeah, it's really cool. and the truth is they made it fun to use is the truth of the matter like your agents use emojis like you'll come through and you'll notice that hang on, let me go back do you ever catch them talking to each other? yeah, you can watch them I mean without you unprompted I have not caught that but what you'll notice every time you tag one the first thing that happens is the two eyes emoji pops up which means I see you Maya and typing you a message is what the other one means.

1:08:56Like, I'm on it. And it's just, it's a cool layout. It's easy to use. It's a good approach. It's also not the only thing that Block has cooked up that I'm going to talk about today. You've got to explain this other thing to me because I still don't get it. Okay. Now, before I show you this, I want you to know that it's among the most traumatic homepages I've ever visited. And it's a lot to take in. But it's a really cool tool. And I'll show you first. That way you can get past the, oh my God, what is that? And then we'll talk about how cool it is and why it matters. So, here we are. Welcome, my dear friends, to Bird.

1:09:47this is another tool by Block that just dropped this is at the fun level what you're looking at are you're able to give your agents bodies you're able to give them little bodies you can use them as pets on your desktop you can do all kinds of things you can make trading cards with them this is cool yeah it's such a fun tool here's a number of little things they've used them for but we'll talk about what this is in a minute but like here you go prairie mountain man's mind is blown yeah yeah it's insane but it's absolutely you know what we don't have enough fun in these tools man this is this is really cool stuff we're alive at a fascinating time with some abilities that our grandparents would have never dreamed were possible and and dang it if that means i want to talk and dance and swiss army knife to do my legwork, then so be it.

1:10:45I like the typewriter. I'm a big fan of copycat. It's the future. It's the reincarnation of Clippy made manifest, but now if anything it could be Clippy. Yeah, honestly, they need a Clippy, if I'm going to be honest. Look at this. Pushback. It does feel a little bit Black Mirror. That's funny. It very much is. The eyes have a real Black Mirror vibe to them. Just a concept, maybe. never be alone again click me never be but can you can you show the interface though like like what what is actually what is it like to actually use this whoa uh yeah let's go back how did they make this this is so funny it's amazing isn't it it's it's just it's hilarious like you could absolutely do it without all of that but i mean if it's an option i would use it uh here's what you should know here is the interface this is it you got widgets and various things you can stick here what this is is like a home base for your agents your skills your automations your projects on your computer so just like with with the other one you connect your cloud code you connect your codecs you connect whatever the heck you want to connect here and and you can operate all of those agents in here but you can also give them a body now why this is cool is if if you lose a computer if if for some reason you just lose a subscription cancel subscription you have access to this where your agents live they don't live in chat gpt they don't live in claude they live in here and as do your skills.

1:12:25So here's another cool thing. It's like, you know, if you use all of these tools or some of these tools like Grant and I do, like I have skills that exist on three different computers that exist in Claude, that exist in Gemini, that exist in ChatGPT, and they're all some level of different because I've updated them periodically and maybe not gotten all of them. This is where your skills live and you can call them from any of your connected harnesses. so if you come in here and make a change that means when you go to claude code when you go to codex wherever you go it's using this version and that is huge it's still hard for me to wrap my mind around that and perhaps we can try and like take it even one step um like simpler for normal people like what when when would you act what's a real world world use case when you would actually use this okay i'll tell you what skills first yeah here's the deal okay let's say i use i'm going to say codex because that's kind of my main i use i have codex downloaded on three different computers those skills don't talk to each other i mean like if i load it on my mac it doesn't work on my pc i have to go and either download and send that skill over there this is going to be fixed soon.

1:13:45But I have to take it over there and do that, or I have to create another skill similar to it, and they're not matched. They don't match. They're not identical. They're going to perform a little bit different. If I want to make a change, it means I have to go to three different computers to make that change to ensure they're all right. So if I say, I want you to end every sentence with, I have to go do that on my Mac. I have to go do that on my PC. I have to go do that on the web interface but with this i can go into that specific scale right here and make that change and that change will be immediately reflected in all of those other apps it's it's skill versioning and think of this as the hub on a wheel and your agents as and your harnesses as spokes.

1:14:36If that makes sense. They all feed right back to here. This is the hub of your skills, the hub of your agents, the hub of your automations. And then you can from Claude call the same automation you could Claude call from over in Codex. Okay, wait. So with these little guys here, are they technically agents that work in all of these platforms or are they skills that work? Okay. Yeah, these are agents. when you make an a don't click here. Hang on, I've got to click there.

1:15:08Oh, wow.

1:15:13Turn your sound down. Oh, the internet's being stupid. It's like a rave, but the music's not loading. Oh my God. We keep getting a glimpse. Yeah, maybe switch off of this. Yeah, I think you're right. We're not prepared for this.

1:15:37Just hit the wrong button. Oh, well. But it's – did you hear the – it's talking still. Can you hear it? No, I'm not hearing that because you're not sharing your screen. It sounds like Teletubbies. Okay, so if you're a normal person, you probably don't want this tool. But if you're a freak – If you're a freak, that one's for you. It's for me. It's so for me.

1:16:11It's so fun. I love that they made that. Yeah. Yeah, that's kind of what I thought. I was like, I love that. You know, like everything else is like, oh, good. Another dark mode interface. Yay. It looks just like every other coding harness. And this looks like every other coding harness if you just threw some monsters from Digital Circus into it. Yeah. no for real and you know what we need more of this i agree i agree it can absolutely be work and be fun uh it does not have to be miserable as long as it does the job like uh who cares i love the idea of skill versioning yeah skill versioning excites me a lot just because of my specific situation i i just i use a number of different machines and and things and uh that's a thing that's been on my God, I need this list forever.

1:17:04What do we have left? Grant, anything? Or was that our last one? You know, we could talk briefly about GrokBot and all that stuff. But yeah, maybe just briefly touch on GrokBot. And if there's anything else you all are curious about, let us know. I remembered him and Eric's show great job very well. but yeah if there's anything you all are curious about or wondering about go ahead and let us know but while we're finishing up here we'll be another yeah 10-15 minutes probably so 10-15 minutes yeah yeah i guess like one other topic i think i would like to talk about and maybe just like save this for when we ramp is like there's this idea that i've heard a couple times recently that I want to discuss, which is consumers don't care about productivity.

1:17:55And all of the AI apps have been focused on productivity. And so far, there hasn't really been a huge breakout AI consumer app. And, you know, lots of people don't like AI because of the data centers and, you know, copyright and all sorts of stuff, like legitimate reasons to dislike it. And so I wonder to myself What would people want Out of these tools What would you actually want As a consumer But what's a normal thing What's a normal thing that your phone can't do for you So you want a robot I use no I use voice mode as like a Cooking helper regularly Things like you know For instance I make my wife a specific milk.

1:18:44Like it's a blend of half and half and 2 % that she uses for her espresso. And I'm always like, you know, it's doing my math. I'm like, hey, I'm making that milk again. And he's like, all right, man, it's a 60-40 blend. How much half and half do you have? And I'm like, I have five-eighths of a cup of half and half. And he's like, then you need three-eighths of a cup of 2%. That's it. But see, that's handled. That's handled by, you know, Chachi B.T. or Gemini. like what's something that they can't do or don't do that people don't do or can't do yeah like what's something like because i think like what what apple is about to do with siri is going to make your phone so much easier to use in theory that that's going to blow a lot of people's minds like you don't have to click into all these different apps to be able to use them after four weeks of having apple intelligence what you should know is i thought it was really cool for about four days okay and then you got over it and then i just went back to using chat gpt voice mode because i is there a consumer use case for this stuff outside of work i think entertainment i think entertainment like you know video image text that'd be good you ever read a book and no one wants to talk to you about it that you just you just watch something you're just Dying to talk to someone about nobody else will.

1:20:07That's a thing. Ooh, nutrition tracking is a good call because I'm kind of doing a thing like that right now. Right. With ChatGPT Health where I'm, you know, it's kind of helping me through making better dietary decisions. And, you know, kind of helping manage my weight, keeping an eye on my health, keeping tabs on things. And that's it's a really good application. Like there are a lot of applications that I think exist inside the tools that exist right now that people just don't think of repairing your car. Like I've literally diagnosed a sound in an automobile or the mechanic. Yeah, I think all of that makes sense to me as like chat.

1:20:56I think what I'm thinking about is more on like the user experience, like the interface side. and I think actually it's basically just abstracting away as much interface as possible. And I think I'm thinking of it as talking is why I'm thinking. I'm not thinking of sitting in a terminal. I'm thinking of I do all of this with voice mode just walking around. Yeah, I think that's right. I think basically like voice replaces you staring and pressing buttons and eventually you're just talking like, hey, can you do this for me? And then it actually does it and you don't have to go in and babysit it and check on it.

1:21:30I think that's probably that version of devices, which seems like what OpenAI is building towards. That is going to be the thing that most people like. That's very much the thing we've entered into over the first half of 2026. The era where that kind of stuff is doable. Like last night, I had ChatGPT complain and demand a refund from a video model company last night. who I guess I shouldn't put on blast until I see if they're going to refund me. You can check in your email right now. Yeah, literally, I'm pulling up my email. Like, let's see, make sure I haven't gotten it. Tough luck. Yeah, that's hilarious.

1:22:13No, I haven't. Okay. Yeah, I think... Go ahead. No, I think, like, yeah, being able to have it be a true personal assistant for you, I think that is probably a use case that the consumers could get behind. But maybe it's not framing it as productivity. It's framing it as like a helper, like a help buddy. Oh, I'm going to blast him. I just got the email. Hang on. Oh, okay.

1:22:43Cut that.

1:22:47Problem with lives. I'm talking about Hey Jen. I don't know if you're familiar with Hey Jen. But their tool has some behaviors that are absolutely not okay. Like, you can use plan mode, but after you plan a video, if you're not staring at it while it works for 10 minutes to prepare the plan, you've got less than one minute before it just assumes you approve and goes ahead and auto-generates a hunk of garbage and eats 10 % of your tokens every time it hits send. And this has happened. I've had it use the wrong avatar. I've had it do all kinds of things. and last night I had ChatGPT reach out and request a refund.

1:23:28It wrote a very nice but terse email explaining why, offering details, and I just got, hey there, thank you so much for reaching out. Unfortunately, we don't issue refunds on successfully generated content. We understand this may have been generated by mistake, but keep in mind that each generation consumes significant processing resources. As a courtesy, we're going to give you 100 credits of my 600 to support my work. So as far as I'm concerned, I'm going to cut that subscription and never, ever work with or recommend you again. See, this is the problem. Dark patterns, as Grant said yesterday.

1:24:08Yeah, you can't be a jerk in this world with these tools. Like, people will replace you. They will find a way to replace you. Absolutely. Absolutely. There are other tools that don't behave in ways that are shady. It's not about the refund. It's about the fact that it's set up to burn your credits whether you need them or not. I wrote about this in the newsletter a couple of weeks ago, but basically I think the world we're entering into is, if you're going to be successful in the software business, you have to ask yourself, do people like you? because that's the thing that's going to matter. Like when it comes to human customers, agent customers, they're just looking for the most straightforward way to get a task done.

1:24:56What's the easiest, what's the best recommended and they're going to just go do that. And a lot of software companies will just build only for agents. I think that's a lot of what's happening now. But for the consumer business, which still exists, you really have to ask yourself, do people like you? Because otherwise they are going to replace you either with their own custom solution or an open source equivalent that realize that a lot of people hate this company and they want to go and make their own version that isn't as dark in terms of dark patterns. Yeah. And I'm just, I'm done with that.

1:25:29Frankly, credits are a dark pattern in themselves. I went on a Twitter tirade about this at like midnight last night. The fact that credits are an untransparent way of, it gives them the ability to just change what a credit is without me ever knowing and change the value of it without knowing. And I very, very much support a company that's willing to charge me a flat rate and then let me pay API fees. I don't care if it costs more that way. I really don't. I understand that you have to keep your tool alive. I'm happy to spend money to ensure that that is the case, if it is a good tool. But I do absolutely, I'm just tired of credits.

1:26:22I think it's sneaky. Yeah, I agree. I was thinking, what makes credits different than paying per token? And I think it's because per token is very cut and dry. It's like everyone knows, it's like 1 million tokens. credits is much more opaque as you said but if you're charging for an image or a video well you know you can't really charge per token for that so no it's like you charge per generation and you just say x amount of generations is yeah yeah yeah something that gives us an idea of what it is uh but um just credits sounds like you're taking nickel tokens and selling them for 50 cents is what it feels like is happening.

1:27:09Well, a lot of times the companies that are offering this are like pass-through, you know, they call them layer companies or wrapper companies. Wrapper was the popular term where it's like someone else is actually creating the model and providing that. Someone else is providing the data center. You're just providing the interface and then you charge a fee on top of that. That's why for all the tools that I'm building for myself and others, I just do everything as bring your own API key because it's like you can decide what you want to pay. You know, one of the problems is like this is a great tool.

1:27:43The truth of the matter is the layout of it, the setup, it's a great tool. It has issues, though. It's not super consistent. Like its whole thing is you create your avatar, but it uses just random stuff that's not my avatar and calls it a successful generation there. And, yeah, I'm not paying for that. I'll go find a better tool I'll figure out how to make it work with Firefly and be way happy I'm also working on something I want you to test I'll share it with you today I appreciate that we'll maybe do another stream on that well everybody thank you do you have any hot takes on it I know neither of us have tried it yet because it's$200 because it's$300 Well, mine was asking for$300.

1:28:35Yeah. Yeah. And I was like, I want to try this, but I don't want to try it$300 bad. I've heard a lot of people like it. I've used the model. The model's really good. Yeah, I use it every day for research. Because with my Twitter Premium Plus thing, I get a decent amount of usage of Grok. And it's a good model. um but yeah firefly is shiny and should give you some serenity as well so i can stop raging against the machine i think he's talking about the the the sci-fi show and movie oh the movie yes and the show i need to watch that again it's been ages man yeah ages and ages and ages that's a good idea yeah grokbot would you say grokbot is most equivalent to Open Claw or Hermes or Perplexity Computer?

1:29:28Is it? Or does it do something that's even more new than that? Yeah. Yeah. It does? Yeah. It's apparently considerably better. Do you want to bring up the... Yeah, let's look at it. Let's read the notes. I like notes. Okay. Like and subscribe if you haven't yet, please. We appreciate you. It's been a lot of fun. I really appreciate when the chat has questions and thoughts and ideas. It makes these a lot more fun. So check this out. This is like a mix of computer. You're cutting out really, really bad.

1:30:19Yeah. Yeah. Hang on. I'll bring it up.

1:30:25Okay. Yeah, you can talk now. You sound okay, I think.

1:30:35All right. Let me bring up my screen.

1:30:42All right. GrockBot. AI teammates you can give real work to. Bots can sign in to your tools, use them just like you do, and come back with finished work. Very cool.

1:30:59Very cool. Oh, I like this. I like the setup here. Account manager, challenge account. You know, this is kind of like what my agents and scheduled tasks are doing, I guess. In workspace? Chattw2 workspace? Yeah. I mean, even just in codecs with scheduled tasks. a lot of a lot of like you know okay I've put this stuff in there like my whole my whole neuron workflow my email workflow is just scheduled tasks it's not like a workspace agent though I would argue it is an agent I think I think my main problem with this is the price point like both chat2bt and anthropic have$100 tiers and$20 tiers. And yes, of course, Grok has$10 tiers and all this sort of stuff, but this is not available at a price for people to test it.

1:32:02So you're either all in and giving it a shot or...

1:32:08I say that. Oh, I'm on the XAI page, not... Yeah. We need the Grok bot pricing page. We got this wrong in the newsletter the other day because it's really confusing. to try to find this information. Okay, let's just see that in here. Yeah. There it is. 200, 300, 200, 300, or... They have 120 for Workspace. Okay, so that's probably the best way to try it. Yeah. Yeah, that's for Teams, so you'd have to buy more than one. Yeah, but like we said, if you wanted to do Workspace Agents with ChatGBT, you just find a friend to go in on. Get a business account. Get like two or three people together on the same business account.

1:32:50and rock it that way. Hey, Paul, thanks. If you're still here, appreciate you being here and for chatting with us. Get Ultra. GrockBot's own computer signs into your tools, routines on a schedule. Yeah, you're like technically kind of supposed to put this on a computer, aren't you, by itself? Like its own machine? Yeah, I think there's a cloud one. There's a cloud one as well, I think. Cool. They host it for you. It looks neat. I'd love to try it if there was a, you know, a little less out of pocket because I've got so many AI subs already. Yeah. I'm sort of the same where I can't throw off my current, yeah, workflow.

1:33:32For additional hundreds of dollars. Yeah. Not worth it. It's cool, though. And X really stepped up. You know, the truth is, and the thing I credit it to is the same thing I credit OpenAI's success right now to, which is smart compute decisions early. I thought you were going to say... Not having Dario as a CEO? No, but both are valid. Yeah, yeah, no, he built a gigantic data center, and it's paid off for him. Yeah, yeah, and like two years ago. Like, it was very much the same time OpenAI was buying compute, so was Musk. And while that may have seemed crazy to anyone who's not staring at it through a microscope, here they are and uh and and sam and elon were right that demand would go insane and uh it did we're going to build the biggest data centers we've ever seen and it's still not going to be enough i do like their their idea to just put them in space where nobody has a problem with them i'm i'm fully down with that i see i have i have mixed emotions about space i i creep out a little over aliens being able to access your yeah maybe maybe what they are is nice little alien blockers we could attach some like i'm not even going to say those words or we'll get kicked off youtube uh but i don't even know what you're saying thankfully yeah but what i what i was gonna say is um i think like you know obviously people in memphis are really upset with the amount of natural gas plants that he's using, you know, generators that he's using to power this data center.

1:35:15And, you know, he's a solar, he's like the proponent of solar and batteries. And he knows that batteries are the way to actually protect these data centers from what's called micro fluctuations, which is during a training run, there's like this really bad spike and then it throws off the whole thing and it can be really costly. So you need to have extra load that you're able to, you know, basically power on exactly when you need it. So just need to switch all of those generators over to solar panels and battery and i think a lot more panels take time because it means real estate yeah i know there's always a trade-off with everything nuclear takes a lot of time because you know it's very dangerous and very expensive to build yeah yeah there's always a trade-off i i i frankly i'm i'm fully pro solar data centers i think solar data centers make a lot of sense i think there's lots of rural area where you can put massive ones there's one not far from me that is is it's five miles long and like one to two miles off the interstate in each direction and it's just miles and miles of like 12 foot square solar panels back to back to back that follow the sun all day long if the sun's up it just keeps following it to maximize what it can do and like it's it's amazing it's it's amazing it takes up you know a few square miles and I assume that one's going to a data center, but I haven't.

1:36:41I heard rumors it was either going to potentially be for Musk's Memphis data center, which could be very well true because it's 150 miles. So there's definitely some getting it wherever it's going to be that has to be done. But the other assumption was that it might have been Zuckerberg's. Oh. Have you read this story recently? It also has paid attention to compute, by the way. Yeah, that's true. But have you read the story how a lot of politicians are now basically putting the kibosh on data centers and not just liberal ones like Republican ones? All of them. Yeah. Yeah. Because everyone realizes that like outside of our bubble that we live in, people are not into this stuff.

1:37:28You know, and what's funny is the states who do take it, like Texas is going to make a lot of money. Texas will rebuild its entire electrical infrastructure that's been a god-awful disaster for most of our lives. The money is going to be ridiculous. but you know in my opinion if you have areas that are you know removed from people like I don't care it's in a data center I mean in a in an industrial park or something like for god's sakes factories are loud they're pumping all kinds of terrible stuff out into the air around you they're they're noisy there's a ton of traffic like to me it's it's more become it's less about the reality of data centers and more about this is how we rage at AI.

1:38:23Yeah. Well, I mean, I think this is their, their, their, here's how we fight back against it. It's a physical manifestation. I agree. But yeah, I also think that there's a lot of people who like chat GPT and like using it and don't want a data center in their backyard. That's fair too. This is based on NIMBY and Sons reporting. Everybody wants more prisons, but nobody wants it to be near them. This all comes full circle to my whole point where it's like, we need to stop thinking about scale. I think scale is the original sin of the AI industry. And it's why we're burning billions and billions of dollars on this.

1:39:00It's why we're building so many data centers so quickly, getting ahead of ourselves to some degree, but then also being behind because you see what scale brings. But like at a certain point, you need to scale down in order to scale up. Because think about how much more you can do if you fix architecture to be much more efficient, to be much more memory efficient, to be able to work. Like you get so much more capacity out of what you've already built. And not only that, if you can make it so that everyone's computer can run these things, then people will have access to it locally. But that's exactly what's happening already, I would say.

1:39:34Like now, an H100 runs these new, more efficient models that have continued to get better and better and better. An H100 runs GPT, I think this statement goes back to 5.4. I want to say it runs GPT 5.4 faster and better than it ran 3.5. I believe it. Which is getting more out of those. My problem is I don't think the idea that U.S. companies are going to just take a break and hand over the global lead in the AI races is a thing that is realistic. Like I get the idea. I don't disagree with it as a concept. I just can't see a world where it happens. I mean, I will say there's a lot of attention being given to efficiency in these as well.

1:40:39When I was at GTC, I went to Startup Open Mic. And there may have been another name for it, but that's what it was. It's like Startup Karaoke. Were they doing stand-up? It was more like a pitch. It was just like, you got three minutes, come make your pitch. and they just got in line and came up and made their pitch. And almost all of them were data center efficiency, data center cooling, data center distribution, all of these big problems around making data centers do what they do at a much broader scale without getting larger, doing more environmental harm, requiring additional electricity. Just like with the H-100s where it's all about getting more out of what you have, there's a lot of interest in that.

1:41:36A lot of interest. I think people would be more on board with data centers in the abstract if companies were saying, here's all of the powerful algorithms that we're running in order to make your lives better instead of replace you at your job. Yeah. And do the tasks that you like to do. Like we need to be using it, this stuff to be doing more things that we couldn't do before, not just figure out how to do all the things that we can currently do. Because that's not how we roll. Oh, sorry. Go ahead. I didn't mean to. No, no. I'm about to agree 100 percent of what you're saying. The problem with Americans is not Americans, humans, is that we have our ideas and they are what they are.

1:42:21And come hell or high water, that's what we're going to do. We're going to go and we're going to chase ideas that validate what we already think. And people just don't change their minds, not as a rule. Like, you know, it's not a situation where they're going to sit down with the AI companies. The AI companies are going to be like, oh, my God, this is terrible. Why would we have ever suggested this? Or where the AI companies are going to say, like, here's this. Where, in my opinion, what I would hope people would see is like the Moderna announcement yesterday and understand that what's happening right now is what's required for that to happen, not just in melanoma, but in lung cancer, in pancreatic cancer, in stomach cancer, in bone cancer.

1:43:03And there are nine other ones already active right now. And AI has played a big role in that. And I think the world needs to see more of the promise and less of the flock cameras and less of the, you know, ways websites are taking your data and ways that, you know, companies are looking to manipulate you. manipulate you. I think when that's all they see and that's what they think, that's what they assume all this is, and that's what a data center represents. And that's fair. I mean, like, I don't like that stuff either. But at the same time, things like that Moderna announcement give me a lot of hope for the future of our species.

1:43:53Yeah, I think both perspectives are valid. I think like there is unquestionably going to be amazing things that we can do with math and with computation that will unlock so much more possibility and benefits for us than we could have ever imagined but at the same time you know that we got to solve these problems and that's why we talk about both you know we talk about both the upside and the downside yeah yeah and and and both are are legitimate concerns i don't i don't at all mean to shoot down the concerns of the against data centers. And frankly, it's their yard. It's their right to decide I don't want this here.

1:44:30And they don't have to tell me why. They can just say no. But I do feel like at the same time, I see the value in the data center itself. And I believe that I can see what's on the other end of this. Maybe not the other end, but what's in the future ahead. And I think we find somewhere where it is okay. I think... space. Hostile takeover of North Dakota. You know, I don't know. Just kidding. Yeah, what if, well, no, funny theory, what if we just decided like one state, we're like, okay, we're just going to turn this into, you know, this is where they all are. Everyone here gets a brand new home in some other country, other state.

1:45:15You pick where you want to live. But we're going to need, you know, I don't know. I don't know. I do think, though that a lot of what will help with this is decentralized servers so like okay we're trying to build like massive data centers like on the outskirts of towns they're like you know miles long buildings like yes okay that's good for scale but you know that's good for the models of today we don't know what the models of tomorrow are going to need um why not have everyone just has like like they should have a battery on their house solar panels on their house have a data server on their house and then we distribute that as a grid system and we sell the energy yeah every individual gets their cut my personal belief system is that we should be as self-sufficient as possible at every node of the system so that you know if something goes down then you know something else can cover it and you know if you have these big centralized points of failure like sure it helps you for serving lots of customers but then it also So like, you know, the war could break out and somebody bombs the data center.

1:46:21And then it's like, oh, there goes, you know, half of the country's intelligence overnight like that. Well, you know, there have been companies that wanted to do that with blockchain. Like, I mean, I remember dealing with a gaming company a while back where that's what they were doing was like when you go to bed, you've got this little app on your thing. in and like companies doing 3d rendering for movies and stuff we're able to tap into additional compute resources from your laptop from your computer and bring all of that compute together into one big bundle that could help them generate movies faster in terms of like you know rendering 3d and cgi films uh how'd that go for them i mean it was really interesting it absolutely worked but I don't know that anybody cares enough to try it.

1:47:06I'll have to find that. I interviewed, I did, I think, a video interview with him. I'll find that and share it with you. It was, that's cool. It's from when I was with a different magazine, but it was, it was interesting. And I think kind of supports the idea that you're talking about on maybe a more smaller scale in terms of like not solar panels on homes, but, but at least that. Yeah, I would love the idea to be like, hi, I want solar panels on every inch of real estate I have. I want a patio. And I want a hose that goes two ways. One way takes energy. The other one rains dollar bills into my kitchen.

1:47:45And, you know, honestly. I just think if, yeah, I think if people, if you want people to feel involved in this whole thing, right, you have to involve them. You can't just say we're going to build gigantic things out in the middle of nowhere that's going to replace you. that doesn't make me feel very involved in the process how about you say hey we're gonna we're we have this new system you know it's decentralized everyone's gonna get their own server their own data set their own mini data center in their home you know and we'll all be more self-sufficient we'll all have more intelligence like we will be a better civilization if we do this we'll be more who pays for the data centers just curious what's your what's your thought there is this a thing that like who pays for it yeah companies just give you these to run and pump the energy off of for them or is this a people will go buy their own in-home data centers there is a lot of different ways you can slice up the business model isn't there a company that is doing this i think there's a company that's doing this it's sort of yeah there's a company that will uh stick i think it's a gb 200 nvidia gb gb 200 i believe on the side of your house yeah uh but i don't know the specifics of how it works we've reached out to them about a podcast but i don't think we ever heard back uh we're gonna do that and do something about data centers in general soon because i think i think i would like to pick the brains of someone who knows data centers and uh because like like there's a lot i don't know and would like to better understand and uh i'm curious to just see how they work so we're going to try and get i don't know an electrical engineer or somebody who knows data centers i feel like uh i feel like this is kind of classic you and me where the longer we go on a stream when we're not teaching people something we just eventually start talking about data centers just eventually land at data centers data centers is the hub where all things lead all right guys it's been fun i i really appreciate everyone who's here thank you for staying so long and listening.

1:49:46Thank you to AREFs again for sponsoring today's episode. We really appreciate that and glad to have you. On that note, we'll see you back here next week when we have a cool one scheduled. Is it scheduled already? Oh yeah, it's already up. I'll link to it one more time. So this is a beginner's guide to GitHub next Thursday, same time. We have hijacked a human from GitHub to come and tell us about GitHub and help us understand GitHub and teach us how we can use it better. It's a thing I think that a lot of people, when they dabble in Vibe coding, are, like, really intimidated by. And I'm saying that based on the fact that I was really intimidated by it.

1:50:28Oh, sorry. The GB200, it is... Grace Blackwell. It is, yes. It is the earlier-gen NVIDIA Grace Blackwell.

1:50:43server rack gpu server rack thing that that runs in data centers all over the place um and i it's it's the current is the gb300 and it it just started going out like december or something but the gb200 is still a monster and uh i think that's what this company's putting on houses we'll find out who they are report back but yeah thank you robert robot cat always here always a real Rodwin, thank you both. We really appreciate it. We'll see you back here next time. Farewell for you. Let us know next week or in the comments if you use any of these tools and what you think. Yeah, I'd love to hear your thoughts.

1:51:25You guys get bobe. All right. I'm going to call it a day. See you all.