ChatGPT Agents and Claude Dashboards

Hey everybody, welcome to The Daily AI Show. Today is April 23rd and I believe today's show is 789, something like that. No, it can't be 789 because we just went, it's 709, not 80. Oh, that makes more sense. Tomorrow will be 710, I believe. I am Beth Lyons, I'm in the studio today and joining me is Gareth Hood. Hey, Gareth. Hello, hello. Great to see you in the house. Good morning. And Andy Halliday is joining from the chat. Hey, Andy. In the Oklahoma City Airport, OKC Airport, Oklahoma, Kansas City. I do not even know. Don't come at me for vocabulary around geography. Nope, that's just not one of my wheelhouses. Anyway, Gareth, like so much is happening in the world. We were just talking about a deeper topic, but did you bring something and then we'll get into the deeper topic. Is there, oh, and you were going to share some of the stuff that you were doing. Yeah, we can get to that in a little bit. I mean, I think one of the big topics and I've seen it kind of trickle in on the news lately and that is allegedly hackers have breached and gotten access to through a third party vendor to gain entry into Mythos. Yeah. And so that's been interesting. I kind of figured that was going to happen. Surprised it hasn't happened already. But yeah, that's pretty interesting. And the way that they described it made me immediately think, oh, it must have happened already. Right. Hey, Brian, just joined us in the studio. I did, but I'm glad to see Gareth here because I thought this was going to be a solo best show. So thrilled to see two faces on the screen for the first four minutes. Thank you, Gareth. Yeah. Sorry. Didn't mean to interrupt, but I was trying to get over your quick. Sorry. Actually, we've been on for two minutes because Gareth was like, you know, we're two minutes late. Right. It's like, I don't, but I do now. Thank you very much. And thanks for everybody. We were just talking about the mythos. Do we call it a breach? It's not really a hack. Right. I mean, it's not like they. The one in discord. Is that the one we talked about? Yeah. Yeah. So it's unauthorized access, but it's unauthorized access because the contractor in discord gamed what the URL would be that gave people access. Right. And then had something. Is that true? Yeah. And I also heard it. There was a group of people that had been trying this like it was. They got together and was a group of people who are working to hack into it. And so that's I mean, that's pretty good job. I love not that I love hackers, but I love like figuring if you're able to figure out a way into something through a side door. That's pretty cool. So you're not doing anything nefarious with it. But it also exposes. I mean, also mythos was so great at at finding security vulnerabilities. Yes. And missed it. So it does seem like. It does seem like a how does that happen at Anthropic? I mean, really, it was just deep investigative work. Right. That uncovered what the URL was or whatever it was for people to gain access. And if I understood right, I think you just said it. Somebody in that discord did have some other like in there was some sort of like a little extra, you know, so it's not like just not the three of us who may not be or would be totally disconnected. The three of us would did some snooping and we figured it out. Right. You know, good, good investigative work. So there's a little more than that to it, but still shouldn't be possible. You know, I mean, in a world of two factor off in all sorts of different ways that you can layer on. I would think that should have to be impossible to do. Right. What they did should be impossible. And kind of back up. I want to back up mythos. If people don't know what mythos is mythos is open or open AI. Anthropics like super AI that they refuse to release. I just kind of want to put that background to it because it's so good. It finds vulnerabilities. It's too dangerous right now to be released is what what they say. Right. And it was designed and trained to be a super coder. Right. So it's supposed to be the best at coding possible for a model to be trained with. And of course, as we know from antibiotics, if you train something to fight one end of it, you've trained something to fight the other end of it. Right. Like so if you're a brilliant coder, you're also a brilliant hacker because, you know, all the patterns of the security that you are trying to create in the original code. Right. Antibiotics are biotics. Just fighting it instead of joining into your body system to make it do whatever it wants to do, which is what a bioweapon is. So, yeah, really fascinating story. And the way what I meant when I said the reason we know about it and it doesn't mean that it hasn't happened in other ways is because this group went to Bloomberg. And I imagine that any other unauthorized access for other reasons was not thinking, hey, let's let the news know. Right. Now, they must have gone through and done some sort of audit about who's talking to them. I mean, there's just I have so many questions about this. Right. Like how how have you released it to 46 or whatever the number is now? Specific companies for specific purposes. And and this is not not caught. Not, you know, these stories only just, you know, give more fuel to open AI to to, you know, bash back on them and stuff. And frankly, you know, you get what you get if you put out Super Bowl ads that are funny but misleading to say the least about open AI. I think you get what you get. You come out swinging like that. Expect people to turn their weapons directly aimed back at you. Like and so this is, you know, this is this is a miss again on Anthropic's part. And it's not a real great look. And so, you know, once again, you know, opening eyes and others are going to be able to point to that and go either Anthropik isn't as mature as they are supposed to be there. Whatever, you know, I mean, they are never mind the government stuff, whatever. They're in many, many different enterprises. And so there's a lot riding on Anthropik, whatever, being secure and correct and all the things. You know, does this directly help hurt some major, you know, Kimberly Clark who might be? I don't know if they are, but let's just say some big company like that. Some mega company like the Kimberly Clark who makes, you know, all sorts of products for the house and otherwise. Let's say they're using clients. Does this story affect them? No. But it's not a good look. You know, it makes people second think, you know, like, well, who is in charge of access and security if the model that is apparently able to break security measures couldn't even be secure enough with its own login that people couldn't sleuth their way to figuring out what the URL would have been. Right. That's just a miss. You know, I don't know. I mean, hopefully they'll fix all this stuff, you know. Well, we'll see. But that's that slow release thing, right? Like they probably would have done more security about the URL if they were doing a more general release. I mean, I'm just guesstimating here, right? But the staged release seemed to be help us figure out how this could be used in a horrible way. Or actually, y'all are the companies that we feel should have an advantage, should not be hacked into with our new model, right? We want to make sure that you've shored up your security. Yes, we hope that the research lab that produces this also has put its model against the things like the easy backdoor, which is, do you have a pattern that you use when you release the model in demo stage? Right. Y'all are like, I get it. You're like deep in the room. You're like the brilliant of the brilliant. But let me remind you what the regular people know is that these are pattern recognition machines. And if you are doing something that is a pattern, whether you're intentionally creating a pattern or more likely you have some human, this is fun and how I remember it. And we should do it this way. You should. I mean, maybe add that to the checklist. I mean, is there anything in our release that is gatekept by a pattern we've used before? If so, don't release it. If you think that it's if you don't release that thinking it's going to be secure. Well, look, I agree. I agree on all of that. You know, I think some of this is a bit of growing up in maturity, you know, and what used to be okay. And, you know, the way you might run something as a small startup. Sometimes I find with even some of these very large companies, they haven't exactly eradicated that where it's still people making up codenames or whatever. And to your point, Beth, like that's the why that's the reason why magic works a lot of times or card tricks works. It's patterns. Right. Right. And it's it's you think this is an original thought. And the magician knows I pretty much know the three things you're going to say. It's going to be blue, you know, and it's like, well, I came up with that on my own. And they're like, you really did. You know, if you're human and you make patterns and whatever. So the irony of all this is that AI can solve for this. AI and other algorithmic tools should be able to to work around patterns so that things are truly unique. But also, how about just like three more layers of security? Like, honestly, is anybody at any of those 46 companies upset that they get timed out every five minutes and that they have to put in a new passphrase to get like you? Clearly, that should that there should be several steel doors between you and that access, probably resetting every 30 minutes. I think it's a growth issue because as far as AI companies go, Anthropic runs the least. I mean, they have the least amount of staff. They run like very dry bones team, very small team compared to the other companies. And I wonder if this is just one of those scenarios where, OK, we need to like maybe hire some additional people, spend some more money on this sort of thing, because this is clearly something we have. We're juggling too many, too many things at once. And we didn't see this. Well, that's right. And that's where you hire from the outside. And maybe they have. I'm not saying they haven't. But this is where you hire from the outside and you get somebody that is the the you know what on cybersecurity that they're going to annoy everybody in that company that they're so they say no to everything. That's the person you kind of need because they like it's the lawyer. No. You're like, come on. I can't even say no. Because you're the lawyer. That's your job. Your job is to be like really, really protective of the company. And they're going to need something like that with cybersecurity. Does that mean it slows them down? Maybe. Yeah, maybe. So I know that they're thinking along those lines. I don't know the results of it because, of course, I only know the kinds of things that would be reached out for. Not that it was. But truly what they're doing. I feel like we're seeing more and more where there are some cracks in what they're doing. The other thing that happened yesterday, the day before, Jude is saying they took code away from the first two tiers. So that is not entirely true. But it was true. But it was also true. For a short time. Basically, they took on there like here's what you get with the free plan and the pro plan. No longer had the checkmark next to code. And if you signed up during that period, you were not getting access to code. If you already had access to code, you still had access to code is my understanding. Although maybe not for the free plan. Again, that was like somebody looked at the somebody on Twitter looked at the thing and was like, are you kidding me? Put a shot up, a screenshot of that. Twitter went and did what Twitter does. X, whatever. And then someone from internal on the team higher up than the people than Boris, who's normally the one who posts about cloud code. But I believe he was also posting was like, oh, no, no, no. This is an experiment we did with 2% of it's an AB experiment we're doing with just 2% of people. And it's like it's public facing. Right. Like. How are we 2% of the people? Yeah, we're I'm already a subscriber. Yeah. Right. And so, again, because you're you've done something and I don't want to say that it's half ass, but you've done something that wasn't fully thought out and planned out in a way that shows up in the world. And then you're trying to like scramble about it with staff who is not in the scramble line of work. Right. We're just trying to get people who people will listen to to make sure that our end of the story is part of the story when it's going big. But that like you're bigger than that. And you're planning for an IPO. I think I mean, I think Anthropic is struggling with compute where they can't get the compute that they they need. And so they're trying to balance things out. And so that's part of their problem. And so they're kind of rushing or kind of scrambling to figure out a plan and they haven't figured it out yet. Yeah. You guys you guys I came in right after you guys had just started. Have you if we're not done, don't do this, Beth. That's fine. But have we talked about? No, you haven't talked about ChatGPT agents yet, have you? Nope. Let's let's bump companies. Yeah. So, you know, ChachiPT agents came out, I guess, yesterday that I don't think they were out before that. And so I've had an opportunity to build a little bit and play around. And as you guys know, for anybody that's watching the show long enough knows rule number one for me is anytime a new tool comes out, build a tool to build more tools. That's always rule number one. So, of course, you know, after building a very quick what I did was I took the the master prompting of one custom GPT, handed it to the agent and said, build this, you know, just to see the process. Right. Like that was just a just a drop in the bucket, had no issue with that. I mean, I gave it the prompting. Right. Like I kind of made its job easy and off it went and it did it. And I connected it to maybe one other tool like Slack or whatever. No big deal. The second thing I did was build an agent, build a PRD right here, which is basically I just went to the agent and I said, I want an agent that helps me build more agents. Because as we know, you know, this goes for agents or cloud code or anything. We've been saying on the show for a while. Andy talked about this. I feel like I feel like he was the first one to start really talking about project requirement documents. Sorry if I'm wrong on that. But the concept that the more time you take early on to fully define what it is, especially now that so much more of it is out of the out of public view. So, of course, with custom GPTs, the benefit of that was, remember, they had their own sort of agent help me builder. Right. But you also could immediately see the system prompt if you swapped over to configure. So, you know, as somebody, as all of us who've been building for a while, there's a lot of comfort in, okay, but I can see the buildings and the building blocks of this. And with agents, you cannot. It's like, you know, well, I wouldn't even say it's a little more obscure than like a cloud code or co-worker or whatever, where you really don't know what the prompting was. What you really get is like deep research. You put in your request to the agent. It comes back with a plan. But, I mean, it's really like it's not a plan like you would see in cloud code. It's like it's more like a deep research on Gemini. It's like I'm going to do this, this, and this. Do you agree? Or do you want to make edits? Right. So, it's pretty user-friendly. I will say the whole thing is very, very well. The UI is really well. You got moving things. And I'm working over here. And you see the little moving markers. It's like I'm changing this over here. And now I'm updating over here. It's very visual. Like look at me go. Look at me work. You know, I'm doing all these things. I'm not knocking that. Like that just means it's going to be more everyday user-friendly to people for the same reason people don't go into CLI versus using cloud code on desktop. Right. So, anyway. So, that's what I did first. And it does have memory. It does have some ability to hold, you know, information in the cloud, which is really cool. So, what was cool about it is I built the agent builder. And then I said, okay, I'm going to use you now. And so, even though it's not like a massive thing, I was like, let's just start with to-dos from my Gmail inbox. I want this to run on a schedule every day at 6 a.m. I not only, you know, wanted to look at what is active, but I've already said that I've completed this task. You need to remember this day over day over day. So, you have to have memory. Right. And so, it went through and because I was using the PRD writer, the new agent, the agent builder agent, then it asked me a lot of questions. It was like, hold on a second. Where do you want this? And how would I know? And do you want this marked as a star inside of Gmail in the inbox? Is that how I'll know that you've already done it? Do we have a process in place for this? So, it's great, right? Because, like, I'm like, do this thing for me. But it came back as because I was using the agent builder. It came back and said, there's a lot of variables here. Let me ask more questions. So, ultimately, when it was done, it created two zip files. One of them had probably seven skills it built in it. The other one was an overview prompt to it. I literally grabbed one coding, one scripting file, one regular file, and two zip files, handed it to a brand new agent build me something. It said, everything you need is in these files go. No issue. It said, great. Because it had the right instructions. And it never asked me another question. It's like, I know exactly what I'm going to build. And I already have the skills to tell me how to do this job. Because in the builder, we define those parameters. Me and the agent. And so, you don't need to do this extra step. You can certainly just start a new agent and go through this yourself. But you're also relying on the fact that you're going to remember and think about all these extra little questions. And that's not always the case. Especially if it's for a net new thing that you've never built a solution for. There's very little chance you're going to think of all the little intricacies, the little dark corners of it or whatever. So, my answer would be for anybody who has heard of this is get yourself an agent builder. That can help you build more agents. Because then you're just literally, once you go through that process, you're just giving the new agent builder fresh new documents. And you're not confusing it. You're not doing anything else. So, I've only had a chance to build like three or four. I mean, obviously, you can build ones like custom GPTs and they go very quickly. And then there's ones that are more complex because maybe I wanted to pop into Asana and give me an update in Slack. And also do this other thing over here. Or go look at my notes in Notion. And all those things are possible. The main thing I'll say is that, you know, as Carl pointed out yesterday that in documentation, these are not custom GPTs. Custom GPTs, for as much as you want to run them, are free as part of any plan that you have as custom GPTs. Agents are not. And I want to be really clear that people see that. Agents are free to build and use until May 6th. And then they will be tokened costs. So, not part of your OpenAI plan. It's going to be a token cost to it. And how that switches over, we'll have to see because I'm building these, obviously. I'm in teams. I'm on our scaled instance of this, right? The other thing is really you can very quickly share these with the team and make them discoverable on the team look just like you would with custom GPTs, which is very cool. Because I could build a very fast solution and say, this is done and literally share it with the rest of my team and scaled. And they immediately have that tool, which is like custom GPTs. But, of course, agents now are connecting much easily, more easily through MCPs to other tools, right? So, we've just hit that. It is the evolution of custom GPTs. I mean, it's what we all expected to happen. We're like, custom GPTs has to get an upgrade. This is it. This is clearly it. And that's only released to team and enterprises, right? Yeah, I think so. Yeah. So, I don't know. I don't have it in my pro plan for sure. Oh, education. It's also embedded in education. So, yeah, I just said that I ran it in teams. And look, this is the evolution, right? This is the I don't need to know how to prompt. I just need to know how to ask really good questions. You know, and it's still about giving good requirements. But if you are willing to use one agent to write a PRD and a TRD and you're willing to give that back over another agent, well, you don't really need me. You know? Like, it is a done for you type of thing. However, you lose that tweaking, that ability to go in there and move a prompt one way or the other. Because now what we're saying is, hey, that's no longer necessary. This tool is smart enough to figure it out and do it so good that there's no reason that you need to be able to see this prompt behind it. Because this isn't going to fail. You know, through our MCP connectors and stuff like that, this is going to get the job reliably done every single time. And there's tradeoffs to that. You know, it's the same reason why sometimes building an 8-8-N makes sense and sometimes no. Because I can just do it faster and maybe a co-work or now agents. So, yeah, just really interesting. I haven't had a ton of time to play. I haven't seen people build a whole, you know, like on, I haven't gone on X and seen 16 different agents that are below my mind. Those are all come. But that's my initial feedback on it. 10X your work with these 16 agents that will blow your mind. Yeah, yeah, yeah, yeah. I'll say to all that is go build yourself an agent to build agents. If you do anything, do that first if you have this tool and you will be happy you did it. Build the agent to really blow your mind, right? Your mind. Yeah. Yeah. I mean, yeah, exactly. I mean, I think, yeah, there's a lot of cool, like, I think I did something similar with Anthropix agents. Kind of let it go loose. I'm really enjoying, I was talking to Beth before the show, I'm really enjoying the live artifacts that Anthropix has. Because it kind of blew my mind yesterday. I was watching a YouTube video and this guy said, oh, I showed you custom dashboard, made a custom dashboard. I was like, oh, that's interesting. I've been trying to kind of do that. I've got a situation where I have my work email that I can't connect to Anthropix, but I did it on the back end. And so I gave live artifacts access to my, I mean, or Claude has access to my computer. So I didn't have it look at my Apple calendar to get my work events on one calendar along with my other. So I have one solid calendar now that has all my events. I don't have to switch back between calendars. And then I also built the dashboard to have AI news and it has my to-do lists that it has. It's just one solid dashboard that's connected to my computer. Now I just need to connect a voice to it and I will have my full Jarvis. You can just read off of that dashboard. But yeah, that was yesterday. I was like, holy cow, this dashboard. I mean, this live artifacts is really cool. And that's just like one use case that I've done with it. I'm sure there's tons of other use cases people are using it for, but I was just really, like, really fascinated with that industry. It was like hyper-focused. There was a lot of buzz about it, for sure. I didn't have a chance to get into it or whatever, but I'm glad to hear you talk about it, Gareth. Because that's an area where I definitely want to see. That's another one where you want to kind of go on X and be like, what do people build with this? Just to get your mind start going on it, you know? And then usually it's just like somebody priming an engine, you know? And you're like, well, I just need to watch a few people prime it. And then once I'm ready to go, I'm ready to go. I can start thinking of all sorts of new ideas that are important to me. But it does really help. And I appreciate people who, I mean, some of it, to your point, Beth, is like, don't do anything until you do that, right? It's like some sort of hook or some whatever cheesy, whatever. Never watch one of those videos ever, ever. You know, I always resist. Sure. But anyway, they're there for a reason. But I do appreciate the people who jump onto YouTube or TikTok or whatever. And they're like, Ann was saying it yesterday, I think. Like, go find the people who are building live and just watch them go through. For her, she was saying that's really helpful. You know, to see. Not yesterday. I'm sorry. That was Tuesday. Seeing people walk through the process, she finds valuable. Because they're. Yeah, it's sort of like a long live. Like, you know, Kyle Shannon will do things like this. Carl's done it before. It's like, yeah, I'm going to be live for two hours. I'm not really even sure where this is going to end up. But I'll click around and you can follow me. And there's. I don't find that so valuable in the live. But if you're watching the replay of it. Yes. Then you could just sort of hyper, you know, like jump through and find the parts that matter to you. So I appreciate all those creators out there. I literally just. And I was watching the guy's YouTube. And I literally just hold down my my whisper flow and just let him when he said what prompt he was saying it out loud. I pulled on my whisper flow to put the prompt right into Claude. And it built the whole thing. Exactly. He built it. I haven't even thought to see. There you go. I didn't even thought to use whisper flow to capture online. You like capture stuff coming out of the speakers right back into the mic. That's funny. Yeah. So one of the patterns that we're seeing with that is talking about dashboards now. Right. So what you built, Gareth, is a dashboard. That's what live artifacts is. It's one of the things that Perplexity Computer really wants to build for you. Right. It's offering that every single time. And Gemini is now starting when I open Sheets to say, hey, you know, we can do dashboards now. Like, give me let me loose on some data. Don't you want me to do some stuff that I can make visual for you? And so I think what we're starting to see is the understanding that a bunch of people really need a visual for data to make sense to them. Yeah, sure. That's me. Yeah. The scary part about that is you can make a dashboard that says anything. You know, that's the bad part about it is like, is that scale right? Is that, you know, like, are you really seeing like, you know, what it perceives to say? Which is sort of the problem of data analytics when you're having other tools build it is you better understand scale. You better understand what data is being left out because Perplexity Computer is not going to really, it'll tell you it's really confident. It'll do a wonderful job, but it doesn't mean your dashboard is correct. It doesn't, but it also, when, so I was having Cloud Code do some stuff on the back end for the show. We are running out of space for our wonderful live streams every second show or third show that we do. I saw that last night and I thought, I know Beth is on this, but man, we are getting tight. We're getting close to that 50 hours. So I am, yes, now we're good for another couple of days. And I am trying to pull things down, zip them, put them on Google Drive, which then goes to AWS. Like that's our process so that we have the archived things. And I was like, okay, Claude, can you zip things? Right. I got a, I got an external hard drive where all the media is going. The analysis files are staying on my computer because that's part of what's building the thing that we're building about it. And, and Claude was like, yeah, no, I can zip it. I zipped it. I can unzip it. I've verified that the bytes are the same. The, like the whole thing. I was like, damn, I am not doing that when I zip something. I am not zipping, unzipping, verifying all of the things. Then going up to Google Drive and unzipping again. Like, no, it is doing far more than I would do in that process. Which is cool. Yeah. And I was like, no, I'm delighted that that's happening. It is totally off track, but I don't know what made me think of this. It has nothing to do with what you were just saying, but something you triggered for me, Beth, was yesterday I was using Gemini and I had a laugh and I slacked my coworker. And I said, hey, I said, Gemini just told me to slow down and take a deep breath. I said, is best advice of the day. She's like, thanks. I needed that too. It's just been, there's been some stuff. There's been some stuff that, you know, just, just work stuff and, you know, some stress and stuff like that. And I don't even know what I was saying to Gemini, but I was like, I was, I was really like in deep or whatever. And just, I don't know. I don't know where Gemini is like, all right, first thing, slow down, take a deep breath. You're not that. But what it basically was saying is like, your thought process is actually not far off from what I'm seeing and Gemini is seeing. Which was exactly what I was asking. I'm like, am I nuts? Or is this what this is? You know, it was one of those type of things. And so I just appreciated that. I understand. It's, it's the whole, you know, make, make it happy or whatever work with you. But in that moment, like it was actually the best. Yeah. Back to me that could have been. And I sent it to my coworker and she had a laugh about it. And she was like, yeah, no, noted, you know, me too. You know, so sometimes those things just happen at the right time or whatever. And, you know, it's a value added to it. So an adjacent conversation came up on Social Saturday from She Leads AI. Social Saturday is free. Go to She Leads AI. Sign up. Join people talking about AI. And one of the things that was talked about was the pattern of at least Claude Code saying you've been at this for a long time. Do you want to take a break? Right? Like, hey. And I assumed that was me because of my health context. I actually really do need to take breaks regularly. And so Claude does that. But maybe Claude is doing that with everybody. Is this an AI thing where you're like, hey. You used to tell me to take breaks. Yeah. Not once I update graded to 100 a month. Then it was like, go, go nuts. Stamp all night. I just recently got an aura ring. And my whole family did. And I haven't played with it a whole lot and stuff like that. But I do need all these coding stuff. Gareth, like your comment, you know, the thing. I needed to talk to my aura ring. And I just needed to be like, hey, Brian. Yeah. Just put the aura ring. You haven't moved from the desk. Look. Two hours, man. Like, your heart rate is just, it has not moved. Force time out. Go outside. Go touch grass. You know, sometimes you need that extra incentive, that extra push or whatever. But getting it linked up to the aura ring. I should look into that. See if aura ring has an MPP. I doubt it. Oh, no. Yes. All of those things are possible to link in. But I'm not sure in the, on the front end of the user. Apple Health and all that. Surely I could probably connect to it through an API or something. So. Yeah. I mean, you, I mean, you definitely could do it. I'm hacking my way through it right now in my brain. Um, if it's next year, if it connects to your computer, I mean, you could definitely have Claude do it and go and look at it on the back end of your computer. Like. That's true. Yeah. Easy. And there are people, uh, talk to Andy about this more than any of the rest of the cast, but there are people who are kind of on the front end of building like biohacking tools that are AI enabled. Right. I was asking about, um, cause there was somebody released a model open source that would look at your MRI slides, x-rays, blood tests kind of things. Um, and it was for research, but it was like, you didn't have to be associated with school. You could be citizen research kind of things. Um, and also like, I see people saying, Hey, anybody got a good doctor in New York city who is AI aware? Right. Like I want to go, I don't want to have to explain to you what I'm doing and the data that I'm tracking. I want you to be excited that I come in already with my data. And now we have a conversation, you know, honestly, and that's such a good call, a good call out Beth. I'm with you on that. I, I want technology forward, uh, physicians. Funny enough, just more, more ironically, I actually have a virtual doctor's appointment and I use a, um, uh, nurse practitioner, right. As my, my primary care, whatever, which I found to be amazing by the way. Um, it was just kind of working for me now. I just got blood work done and iron is sort of always been this thing, but I also take supplements and she, she brings it up from time to time. And damn it. If the ferritin isn't like, it just doesn't move with the rest of them. And I'm like, listen, this is really in my wife's ballpark as a dietitian. She knows all these things, but she wasn't around. So I was like, all right, chat GPT, what the hell, you know, like what, what am I, what is this? You know? So like you bet your ass I'm bringing that exactly to my, my virtual appointment here. But it's funny you say that because she is, she's AI forward. She's AI, you know, she's, she's not offended that I tried to come in more educated, but well, my God, if I talked to some doctors who are, who, who are immediately on the, on the defensive, if you even dared to try to come in and be intelligent, when you walked in the door, they feel like it's an affront to their education and then their expertise and who are you to challenge me. And I'm like, yeah, that's my whole job is like to come challenge you. You have to make your life easy, man. It's like the, to push you, you know, uh, politely. Um, so yes, I agree with that. I'm glad you said that. Cause I, I cool. What a difference. What a difference when we have somebody who's like, I love that you look that up. Okay. Look, you saw that. That is true. But here's, here's what you're not seeing. It's like, we're immediately off on the right conversation. Not this is what Ferenton is. Cause that's where I would be if I didn't look it up. I'd be like, I don't know. It's part of my, my iron something, something. Like I gotta go, I gotta go research this now. I don't remember what Ferenton is, you know, maybe I'm supposed to remember, but like, you know, I'm on quest, I'm looking shit up, you know? So. Yeah. And I, I've done the same thing less about, I, that's not true. I had a, uh, had a skin cancer biopsy. They gave me the results. Like I got the results from the company, not the doctor. Um, and I was like, it's Friday night. I would like to know what this says. What do these words in this order mean? Thank you very much. Um, but I also do that when I'm going to go in and go talk to the doctor because I want to be able to communicate in a way that meets the way they want to receive the information. And y'all have watched for a while. I communicate the way that I communicate. It is helpful to have AI be a translator if you're expecting it in a different kind of lingo. And that was so helpful. Oh yeah. Right. It was what, two years ago when I talked about on the show, I talked about my wife and I, anytime, anytime it's the end of the year, I will inevitably like go straight little kid mode, stop my feet and shuffle my feet and say, this isn't fair that I have to do adult stuff when anytime we have to talk about, um, renewing healthcare, like there's just nothing I want to do less in my life. And it has caused many a poor conversation. I wouldn't say fight. My wife and I don't fight a lot. I'll be, I'll say that, but a poor conversation, not productive. And, uh, I don't know, two years ago, whatever I, I gave all the information to ChatGPT. This is probably a ChatGPT for, you know, or, or something. We're a while ago and man, did it make that life. And so I, my wife came home. I said, we don't have to talk about this. And she kind of rolled her eyes. I was like, no, no, no. We're going to talk about this. Here's what I'm suggesting. Go in the other room, read my entire transcript with ChatGPT of this and see if you agree with me at the end that, that this is because I was actually agreeing with her about any, you know, what I should do, whatever it was, right. Whatever it was that we needed to agree on. And so she went in, she disappeared for like 15 minutes. I was somewhere else in the house. She came out and she said, no, we're on the same page. I said, okay, so you want to go ahead and switch it to this? She goes, yeah, I'll take care of it tomorrow. End of discussion. The best, best day of both our lives. We're like, wait a minute. So we're after 20 years, we're, we're done with this. We have to talk about this anymore. And she's like, no, I think we're good. And I was like, I feel like it's a trap. But if you say, if you say we're good, we're good. And we were, we was like, it was the, we just, all we did was we made Chachapiti to intermediary. I can speak to it exactly how I spoke to it with all my dumb ass questions that my wife didn't have. And she could happily skip past it. And she asked a few too, from how her mind works. And she talked to Chachapiti the way she communicates. But at the end of the day, we had a shared understanding of it. And to your point, Beth, neither one of us had to be like, I don't think you understand what I'm saying. Right. I don't think you understand what I'm saying. That's not what I'm saying. You're hearing it differently that, you know, I, you get into those round robins where it doesn't go anywhere. There's like circle of death in these conversations. And this was the total changer to that. Like, right. We will do this forever now. We will never go back to the old way. Right. And, and Greg is pointing out that that is a one-to-one-to-one, right? It's totally personalized as opposed to trying to come up with a one-to-many kind of message or one-to-many kind of data delivery. Right. Yeah. Because they, it wasn't like the data didn't exist or packets upon packets. And, you know, this is what, whatever it was in time, Cigna, you know, and it wasn't for lack of it. It was too much in too many different areas. And I was able to point ChatGPT directly to the PDFs that talked about the different plans and stuff. Cause it gets so confusing with all that HMO this and plus that, and they all have unique names to them. And you change it. You change an insurance company. God forbid it changes to six other things. So I gave that all over to AI and I said, you figure, you figure this out. You know, you figure this out, then come to me at Brian level terminology. And that, that is a, you know, you're right, Greg. It's like a, that one-to-one-to-one is, is the differentiator. Because if I tried to use some tool out of the, that somebody had built, it would not have been as effective as me just having my conversation with it. So, you know, sometimes that's the solution, right? That's the AI solution is that there is no out of the box solution. You have to go do it yourself, you know? Yeah. Right. But it's possible to do it yourself with the resources that you've been given and a $20 subscription or a $5 subscription. Right? Yeah. Yeah. I like some of the comments, there's some comments in there about fighting and Jude and stuff like that. So it's a great comment in there about, we don't fight and stuff like that. So yeah. Anyway. Yeah. Jude's asking, is there a, is there a word for AI splaining? Like, is it cloth splain? I guess you could. I guess you could. You could, you could totally explain somebody with that. Okay. That sounds like something evil that you're going to do to like the Holmes Alexa or the, right? The actual Jarvis, the, the, Hey Google thing. At some point, it's just going to, you're going to hear like, actually, what the, what you want is to cook this item that you just took out of the freezer. And because you're at this point in geography, you don't want 15 minutes. You want 17 because the calculation is you're going to be like, fine, fine. Yeah. Connect to the oven. Per our last conversation. We're like, Oh, you didn't. And Alexa was like, Oh, Oh, we talked about this before, Brian. You have absolutely asked me about the stupid temperature for cooking fish. And I've told you this answer before. I never asked about that. And it's like, our conversation on August 24th, 2025. I had a funny incident the other day with my Jarvis that I created. I just set off. So I created this Jarvis, um, like earlier in the week, whatever. And I gave it memory and I, um, kept asking, I was just testing it. And I kept asking it for the weather. I was like, just to make sure that it was connected to the internet. And it said, and it came back to me and said, you keep asking about the weather. So it's, it's not like where you can like, we have, uh, Alexa or, or Siri. You can just ask and they don't remember what you previously asked, but because they get a memory, it now remembers that. And it was like, why are you keep asking for the weather? Is there something? And so it was kind of funny. Like, I was like, oh yeah, there's a memory now. Um, and it's taught, it remembers that I keep what I have to be careful about what I, I, I ask it because, um, it will question why I'm asking it. So it was kind of cool. Um, but yeah, I mean that, that experience building this thing was, is crazy. Um, I, cause I built it, I think I shared in the Slack and how I built it using Claude Canvas. And I kind of just let it go. I was like, build this. And, and then I said, run this, just run it. I was just, just run it. And then it started up and I didn't realize it had started up on the backend and there was no interface. It was just running in a terminal and it just started talking to me. And I was like, what, wait, where, where are you? And then it replies, I'm in your computer. And I'm like, what, how? And I'm just like, sir, I have like three screens, so many tabs open. I'm like, where in the computer, you know? And then it like turned on the computer and then I got, it's like, wait, can you see me? And then it turns on the, the camera and then it's just like, yes, you're wearing this shirt and I can see the Deadpool mask behind you. And I'm like, okay, whoa, this is way too crazy. This is too much going on right now because outweighed, this is how, and then I figured, finally figured out where it was streaming from. And I just, I had, and it was like tucked. It was like only halfway shown on my screen. It was kind of like between screens. And that was part of the problem. And then I could see the, but I've started moving screens around. I could see a terminal moving as I was talking and I was like, oh, there you are. Yeah. So we should not be moving without my, my permission. And so, but yeah, that was my real creeper, like real creepy AI moment. Yeah. That's pretty creepy. Hey, since you mentioned it, I'll throw it up on screen. If anybody wants to see like the kind of things that we share in the community, you can join our daily. I show community on Slack. It's free. Gareth is in there, Beth and myself. Admittedly, maybe not as much as I should be, but I'm in there, but Gareth's in there. And we do have good conversations at it. So if you want to be part of the, you know, daily. I show community and you can't be in our comments or live on the show, then that's a great place for you to go. And there's a QR code on screen right now that you can follow. That'll take you to the Slack page signup. So just wanted to bring that up since Gareth, you had brought it up. And it's also at the daily AI show community.com. So if you're not, if you're listening to this and you're like, wait, wait, wait, where was that QR code? You can go directly to the community. Can you imagine a car play was throwing the QR code on the screen in the car? You're like, well, you see officer, I ran into that guardrail because they threw a QR code up on the screen. Right. I had to grab my phone naturally to try to figure out how to do it. But then I had to face, face unlock it. And it was a thing, you know. Okay. I totally appreciate this conversation because one of the projects in the next couple days is to install the Hermes agent on my Mac mini, which is an old Mac mini and M2 pro chip for Mac mini 32 gigabytes of RAM and two terabytes of hard drive. Um, and what I bought it for, I didn't use it for. So now it's here and we're going to try that, but I'm planning to run it headless. So it's going to come through. What is the Hermes agent? So people don't know. Uh, I'll say it in a second, but it's going to come. I'm going to control it with my, um, with my MacBook air. So there is, it's headless. There isn't like a separate terminal for it. Um, it'll just be running. The Hermes agent is, uh, from noose, N O U S, uh, labs research, something like that. And it is, um, it's sort of an answer to open cloth. It is open source. It's built in memory optimization. So it's compounding the memories as you're engaging with it. It comes with 40 tools already built in, um, which is very much like, uh, like the subsequent installations of open cloth, but, uh, it is, it's a little bit more thought out in its open sourceness. Um, like it was, it wasn't like the open cloth was just everybody jumping on and doing things. This has had a slower rollout. Um, uh, man named Imran was on Greg Eisenberg's startup ideas podcast this week, maybe a couple of days ago, walking through what the installation would be like. And I'm ready for an agent and this is going to be my choice of one. And because the M two mini is, um, it has a big hard drive. I've, I'm going to start it with, uh, a local, uh, it's going to be my first local LLM as well. I'll let you know what I pick or of course, what cloud code picks with me for me or whatever that's going to be. I'm going to have to check that out. I'll have to watch and get that video from you because I, I also want to install it and I'm also a very visual person learner. So, um, but with backups. So I needed instructions and visual so I can roll back and go back forward and roll back and go back forward when I screw something up. Cause I always screw things up, um, just because I'm trying to do too many things at once. And so, or I'm impatient. So that's, I'll be interested to watch or see what, what, what you get it to do or what it does. Um, there's also another one that's supposed to come, that's coming out of China called Kimmy something. It's not Kimmy K, but it's like, Kimmy something else. I don't know the exact words. I've seen a few videos on it. Um, which is kind of what everybody in it's becoming very popular in China. Um, currently. Yeah, absolutely. And, um, uh, it's funny. I've, uh, I finally figured out anything that happens before the show is Beth, right? That goes under the Beth category, the prep, the projects that I'm doing, things that don't have anything to do with the show. Once it's on air, it goes in the daily AI stuff. So two wikis to like the setups, uh, daily AI show will have its own agent, right? There's a bridge between the two because of course what I talk about on the show, I've also been working on. Right. Yeah. But it is sort of a mind. Uh, okay. So where's the delineation? And we figured out yesterday the delineation is, did it make it to air? Excellent. Yeah. It's over here now. We're just moving it. Yeah. It's, I, I hear you on that, you know, between, um, the show, my, you know, personal builds that are like maybe tied to the show, but are just, you know, builds that I'm working on or whatever, then everything that goes along, it's scaled. Um, there's, there's oftentimes multiple parts of my brain trying to work and, you know, it can get a little, uh, it can get a little confusing jumping back and forth, um, from different ones, trying to remember what's what, and that's only going to get like with memory to your point. That's, that's only going to become a little harder because you're going to need it to segment brain and, you know, memory parts of it so that it can be a helpful agent. Um, it's not going to be helpful if it, you know, maybe there's an orchestrator agent that has memory that understands there's three other brains that would be helpful, you know, cause like you said, there's a bridge there. Um, but it is interesting how this all rolled out. I'm super excited to hear as you build this Beth and just kind of hear more about it. So, um, keep, keep giving us updates on the show for sure. I totally will. And that, and that memory piece, one of the things that was talked about on, uh, on the startup ideas podcast was Imran suggesting you ask things like, uh, given how much you know about me, what else can you do for me that I haven't asked you to do? Right. What is something I did today that you saw me do today that we could make a skill, a routine and automation, all of the things. Right. And that starts to get really interesting, but yes, Gareth, the, um, uh, Oh, I'm trying to remember what it said. Oh, when we were downloading the thing, it said, Oh, last Wednesday had, uh, was like four files. We had four files last Wednesday. So this Wednesday should also have four files. And I was like, no, that, uh, repetition you're thinking about our show is not our show. Yeah. But the shows have the number of files based on who came and who shared stuff based on the screen. Right. Like, yeah. Yeah. I just came across a project and I think I shared it with you guys, but I'll share this in the Slack as well. It's this thing called flip book. Yes. Oh my God. And I just tried it out as we were talking and I'm like, wait a second. That's really cool. Um, what flip book is, is a, let me go back to what, how they explained it. So it's a image. It's so image. It's a screen or image. It'll take a screen. It was every pixel on your screen streaming live directly from a model, no HTML, no layout engine, no code, just exactly how you see it. And it shows from the, the, um, yes, this is it. So like it's live. And this is really something I've dreamed of, but I've never realized that thought that or put it together that you could actually make it happen. But I gave it a picture of that. My dyslexia book and it doesn't say anything about dyslexia. It doesn't say anything. It just shows the picture how my brain got a cape and I clicked on it and it knew everything about like a generated image about the book. It knew about me. It was really crazy. And this book hasn't even been published yet, except for on open AI when during the contest of like creative books over AI. And so it knew it was about dyslexia, knew about like neurodivergent and knew it was from me and it was just showing the book cover. And I was like, Whoa, that is crazy that it figured that out. And it generated the image, like the pixels live. And so if it's, um, it's through Zane, Zane's saw, I guess, uh, it's kind of hard to, to explain it or say his name. Um, but it's a really cool project. I think that's something that we should, everybody should be watching. Yeah, that sounds awesome. I can't wait to look at that. So it's like somebody did somebody, somebody has created this is what you're saying. Yes, it's, it's in pro. I mean, and you can try it out for free. Um, or he's got a link to it, um, within his ex post that's down, down the way. It says you can try it out now. Um, and it's called flip flip book dot page is the site that you can go try it out. Hmm. I'll have to look into that. That's awesome, man. Yeah. Yeah. I wanted to throw that in before we jumped. It looks like something you would do Brian, actually. Like in the little, uh, it's, it's the demo is showing London. You're buying tickets. You're scheduled for the tour. I was like, this is Brian's wheelhouse. No, no, I, I, I was like 15 minute walking distance where I stayed in, um, uh, La Marais and in Paris to Notre Dame. So it was still, it wasn't open yet, but just watching that zoom in, zoom out was just giving me like fun flashbacks of being able to walk to the, the site and sit there and have techniques right on the river and stuff like that. So yeah, it was very, that's cool to see that. And anyway, any, you know, look, we don't have, we don't have time for that, but anything that helps with education and people or visual learners and helps them understand like that, that is so cool. You know? Yeah. I mean, it's like an interactive video. Basically you click on, it's a video shows things. You can click on it to see more. Um, and it generates as you, as it goes, it's, it, it's like create your own journey. Um, I'm all about that. Yeah. And that's part of world model meets large language model, right? Which is another theme that we're seeing in 2026 first two quarters, probably all of 2026. And all right, we're at the top of the hour. I think we're going to wrap up. Thanks everybody for spending, uh, this time with us. Go build with AI. Come back tomorrow because we have one more show. All right. Thanks. Thank you.