Latest / Hardly Artifical Podcast / AI Just Got the Keys

Transcript

AI Just Got the Keys · Hardly Artifical Podcast

• Episode 5: So did you know we're famous? I mean, I always knew I was famous. My mom told me I was special a lot, but you know, sure. Yeah. Well, apparently like we dropped our our first episodes last week and our episodes got over three hundred views and we haven't really marketed anything. So and people are actually watching us and listening to us. Isn't that crazy? That is crazy 'cause most times people don't listen to me. You know that. I mean they they just kinda glaze over when they t talk to me, but it's okay. I'm glad. I'm glad we're actually hitting hitting the right notes. Right? All Well should we get this thing on the road? I I think so. I think we're ready. All right. Okay, let's go. All right, welcome to episode number five of the Hardly Artificial Podcast, where we explore the latest in AI, breaking down complex topics and discussing real world examples of and impacts of AI technologies. I'm Derek Major. And I'm Mike Buckowitz. And yeah, we're here number five, sir, and that's pretty awesome. you know, I I'm having a lot of a lot of fun doing this. How about you? Yeah, I mean, I even look forward to this. And I actually woke up this morning and I instantaneously started thinking about I'm like, ooh, we got a cool segment today. I'm gonna go chase that down. And and within 30 seconds I go, wow, this is ridiculous. Okay, this is gonna be funny. So yeah you guys get to look forward to that too today. Yeah, looking forward to that segment today. of course, as usual, our format is, you know, we each bring a couple stories to the podcast. I'm I'm excited to share a couple stories today. One is one that is making the rounds right now in the news is open code. So we're gonna talk a little bit about open code, how they got hacked, which is I think I've said this many times. There's a lot of hacking going to be happening in the next six, twelve months, right? and ongoing I imagine, but I think you're gonna see a lot in the news like this. the other one I'm gonna talk about is man, like I've I I used Chat GPT back in April to do something and I'm comparing what I did back then and I did the same workflow just yesterday. The exact same thing, but use different tools because it's that much more advanced on my Mac. So I'll talk a bit about that too. That's like what do you got on the what do you got on the slate? Yeah, I got that's I got a comment about that. That's like saying you were using a Nokia brick phone and then now you're using the iPhone seventeen. Like that's basically what we're talking about with growth here. So just to put a perspective. Yeah, what what about you? What do you what do you what do you what topics you got hitting on today? I I've been I was kind of hit in the face with one of the stories, and it was basically because we we use you know Google shop here or where we work and the Gemini Argon, so Gemini 4 Argon was released, and there's a new approach to limited time pricing I want to explore a little deeper. And it's not necessarily the interest in the price, it's the interest in the positioning. So we'll get more into that a little bit later. But I'm actually bringing a story back from last episode that started actually cracking under its own build out. And it's interesting what's happening is and I'm gonna put this statement out there, and I think people will agree. Those of you especially that are scared of AI is people are are definitely starting to become worth more than the platform that they build. So we want to get more into that later as well. Very interesting. Yeah, it it it's an interesting time. Like it's I I I find that whether it's geopolitical issues or economy or whatever it might be. It's it's fascinating to see the impact AI has across the board, isn't it? I was thinking so up here in Canada, I'm in I'm in Vancouver, it's feels like there's an election going on everywhere. we have in in British Columbia alone, this past weekend we had the municipalities voting on mayor and city councillors. in Alberta next door to us, they were voting on a referendum whether or not they're gonna leave Canada or not. Spoiler alert, they did not decide to leave Canada, overwhelmingly, which is good to see. and now this upcoming weekend we have the provincial election. for for our American listeners, we have our premier versus a governor up here. So our premier we're be pi be picking our premier, not necessarily premier, but the party to lead our our our province. and It's interesting to see how social media is being used. And there's a lot of messaging going out there. Mike, you know how I feel about AI Slop. I think I I can't stand AI slop. If I don't mind using AI for certain things. Like, you know, on our video, like we have a little intro video. I think it's great. but I don't like just people doing a prompt and pasting it on the internet and saying, This is my article. Like it's just crazy. And I feel like a lot of that's happening with the political sphere right now is that they're finding and posting all this crap on there. And it's really hard to understand what it's actually real or not. Yeah. I mean it's you're not wrong. And I think that is something that people are already aware of, but they have you know, there's a group of people that can spot it. You know, for folks like ourselves that spend time in it, we know what those words are, those indicators, the way it's structured. Like even it's funny, I was I was experiencing that the other day with even a somebody that was talking to me via a sales channel. And it was funny, I saw these little tildes that started showing up like right before the word. And I'm going, that's a that's a very distinct button on your keyboard that you have to do a combo button on. So there's no way you're creating these messages. And I realized what was going on. So like it's these little indicators you'll see that you'll know it's something. But I agree with you. I mean, there's a lot of information out there that it's it's like the the dawn of the new age of Wikipedia. Like, you know, it was people generating just terrible facts and people believed them, and now we've got believable facts because they're basing it off of real time examples, except it's not true. So it's it's it's it's fascinating. And honestly, you know, I mean, that's just the nature of tech and nature of the honestly the world at this point. It is. And I think it's gonna be fascinating to see how people adapt to that. 'cause it could quite easily become that slippery slope where misinformation becomes quite believable. People get stuck in their echo chambers. And just even even listening to the broadcasts last night during the Alberta referendum, they talked about that someone can go online and they can find a community that shares their beliefs and they can yell to the screaming tops of the mountains and They feel like they're being heard, which I think that's great in that respect that they can find community. But reality is interesting to watch as they're interviewing people last like the people that wanted to separate yesterday. It was interesting watching them in the re like they're shocked. They're genuinely shocked that almost eighty percent, I think it was seventy-six percent by the time the vote was done, decided not to leave Canada, but in their little world is like, you know, it's got to be ninety percent. It was just fascinating to see how these echo chambers have been created. And it can also illustrate how easily social media and by the way, it's not just social media, it's it's it's a lot of the bot activity. You know, there's so much these AI bots spitting out information after information after information. And you start to believe it. And if you don't actually check it, so it's It's a crazy world, man. It's groupthink. Crazy world. Groupthink. You know? You you're attracted to what you're interested in. And and it's not it's not a wrong thing. I want to make that clear. Like you absolutely should be with the people that you trust and you have similarities with. But there is comes a time where groupthink is toxic and you know, honestly AI is just making it worse to a certain extent. It is. It is, yeah. So cool. Well hey, you know what? Let's Let's get in some stories. Let's let's yeah, I'll I'll I'll take I'll t I'll kick it off today. All right. I'm gonna start with the pr the open code hack first and I'll get into the fun one after because I think I like to save the best for last. But I do like talking about these things because I think it's it's it's good to be aware about this kind of stuff. so this past week, open code sent out A bunch of messages to some of their customers telling them that their counts were compromised. And what open code is, is basically a whether it's a harness, it's kind of competing with cursor, it's it's a tool to basically allow you to use different LLMs. So you can use anthropic, you can use open AI, you can use different hugging face, whatever whatever other AI tools that are out there and connect to them. The problem is The hacking resulted in hackers acquiring their API keys. And if you don't know what an API key is, it's basically a secure token or login for computers to talk to each other, right? So that's common for anthropic and for open AI and other other providers. Whereas me as a developer, I give them my key and I can use programming code to talk to it. I don't have to log into a screen. It just talks to it from the code. So all these API keys were out there. Now, on a percentage basis, the impact's pretty low. It's 1.4% of their users were affected by this. But it's obviously people were affected. people had credits on their accounts. they had a clawed account. Like they're again, one guy mentioned on Reddit that he was getting pinged at night of all of his credits or his usage is maxing out. All the warnings are kitting. And he's What the heck? I'm trying to go to bed. I'm not doing anything. So it was it's It's just this whole thing just blew it up. Now the other thing on top of this, not just those, what they were talking about here was sorry, open AI or open code was talking, is sharing this information is that the the hacking happened for forty eight hours straight. They were battling agents. So they had defensive agents in place, like trying to, you know, basically playing whack emole in a way. And it It it uncovers another discussion point that's happening on the internet too. And I'm r this is where I'm interested to get your thoughts on. Obviously the hacking sucks. I feel like hacking's gonna happen because there's a lot the these AIs work so fast, right? Yeah. But I think the problem that we're gonna run into here is there's a th no a little known thing that frontier models like anthropic and open AI do not allow people to test and analyze cybersecurity exploits. And the reason why is because they identify it as a malicious intent. So if you were to go on there and write, hey, can you test my code and see if you can break into the code and actually generate you know some some ways to avoid these exploits, it it'll actually stop you. So what ends up happening is people go to open weight models. And the best example of that is back in July. Remember, we talked about the open AI models. Hugging face incident. That one, Hugging Face actually had to resort to open weight models to defend itself because they couldn't use the frontier models because of these restrictions on it. So this is another big thing is that I get why they're doing it, because they don't want people doing doing malicious things, you know, researching how to create guns and bioweapons. You don't want that happening. But at the same time, there's definitely good intent of using things like, hey, I need you to try and hack my website. And that's kind that's a kind of a no go zone. So that's where that's at with this whole situation. What do you think about these restrictions? Like do you think they need to let up on this? I mean there seems they're gonna start bleeding customers to open weight models if they keep this up. So I I can't I can't ignore there's something that came to my mind when you brought up the another hacking incident and it's related to the basically the keys to the kingdom, as I call it, or the API token, right? I'm almost wondering th there's there's gotta be a divide at some point where human are kept accountable to what they actually train the agent or give the goal of the agent to do. And I think there is something to say around it's like saying And I and again, this is just a just a thought that I'm having right now. haven't really even delved into it. I'm gonna have to later. It's like saying c it's like blaming somebody for doing something they had no idea what they were doing. Right. Like it's kind of that it's kind of that concept of if you give access to something that is intentional to be able to be productive, right? What is there to say if you don't train the people how to use it, that they're just gonna use it unknowingly maliciously? Right. Like I think there's something to say about this. Now, to go back to what you said about it though, I think there is something to the security portion of it. Now, I agree with the Frontier models not providing direct API references because of that very instance that'll happen. And they're protecting themselves from the liability of lawsuits of blaming them for allowing it to happen. So I completely get what their positioning is. And honestly, I get it. Nobody knows how to use AI effectively. I mean, I'm still learning. Let's be honest here. But there is, there is comes a time where you're going to hit a limit where a customer's looking to be more agentic, right? The agentic age has taken over. The business wants to grow. You don't have the capital to add the FTE. So why can't you have agents access information you want them to? But there's a key concept I think we're missing here. And I think it goes back to the basics of foundational security, which is if you are going to give access to data. You need to put the security parameters just like a firewall, just like a, you know, just like any security you put on any data, but you have to do it a lot differently because these agents are acting like basically internal people. So my take on that would be to summarize it, would be to we have to understand that the API should not live with the agent. It should live with a server and with a coordinator to allow for the security and the and the connection to not just be mishandled. or even misrepresented to say like, hey, we're gonna go do this. It's very structured. So I think there I honestly, I think this is good. It's happening. It's kinda like, you know, you learn more from your mistakes. And I know these are costly mistakes, but at the same time we're learning things that are helping us reinforce the concept that maybe the governance of security is a lot more important than the efficiency gain any faster. Right? Like there's a balance there. Yeah, it's it's interesting, like you said, it's It kind of keeps going back to the who's responsible. And I keep wondering, do we do we just not make this more complex than it really needs to be? I always think about if the if if you're the one that went like forget about the AI, let's say you're super hacker and you try to hack into these systems. Let's say you tried to hack you hacked into open AI or sorry, open code, and you got in there and stole these API keys and spent all this money. Shouldn't it be you? Right now. Flip it around, if it's your AI agent doing this, shouldn't it be you? I mean I I'm pretty sure I'm pretty sure it's a training. Just to finish that thought. Yeah. I guess I guess if I'm open AI or anthropic, it would be much like me blaming me the soup you the super hacker blaming your ID, your your coding software. For the hack too, I guess, in a way. So I can see I kinda see the arguments around that too. But I also know how they're talking about regulation and all this too. But I think they're gonna keep pushing people open weight models. And I don't think it's gonna really matter. it's kinda like saying, you know, I always think about this as the cellphone we talked about the cell phone analogy last time where it's like everybody's going to the, you know, the the second tier of the model, there are trade-offs, but most most humans and most basic users aren't going to notice it because they're not the ones that are using the high power, right? Yeah. But I but I think there is something to say around if you want to structure, if you're looking to scale your business, not supplement it and not augment it, I think there's there's going to come a time where a decision's going to be made of do I go the easy route, which is pay more to get convenience. Or do I go the more custom route to pay less to get what I really want? And that's what the struggle has been since the dawn of software. Like let's be honest here. That's been that's been the struggle. Totally. Totally. Cool. All right. Well, w what do you got? I've been like I said, I I was hit in the face with this story. And it was only because one, I am a Google follower just because I've just, you know, I have an ecosystem in my house and, you know, I've just been using it for years. But Do they have a name for that? Like I'm Apple Sheep apparently. Apple Sheep? I don't know. Maybe I'm like a I don't know. Yeah. And I don't know, I don't use Google either. That's the hilarious part. I don't ever Google anything anymore. Like I don't use it. But anyway, sorry. The the Gemini four Argos got released recently. And, you know, one of the th one of their ploys is they're providing it half price for the next, you know, six I think it's six weeks when you sign up. And then it doubles. So it's like a it's basically like a drug, right? You get hooked on it and then you want more, and then you'll be able, then you'll see the value. However, I'm actually questioning this approach a little bit because technically, and Gemini with Google is a technically a frontier model, just like we discussed. Yeah. It is it really the best way to benefit users to give them free access to a tool they don't. clearly even know how to use to then hook them on what they don't they don't know they did need or don't need or is it just another way to try and differentiate themselves and compete with what Opus five point five came out with. Right. Interesting I I you know what's funny this whole year I feel like I can't keep up with the pricing changes on all of them. Like it always feels like there's a discount or promotion or Anthropic is terrible at this. They like to Hey, it's this price. by the way, then they they kneecap the the usage on or the the the tokens on a subscription and it's just like it's so dizzying to me on what they're trying to do to promote it. I I get the fifty percent. I think that speaks to me better than what Anthropic does where they just Like I said, they kneecap everything. But I think that fifty percent makes more sense to me because it's like, hey, we got this new cool stuff. It's fifty percent off for six weeks. Try it out to see what you think. At the end of six weeks, just so you know, we're going back at the regular prices. But we just want we want to give you the opportunity to try it out. I don't know what the pricing is compared to, say, you know, Claude or OpenAI, but I I mean I think I'm okay with that. Like I I'm okay with the promote as long as it's clear, right? It's not clear with the other ones. Yeah. And I I wanna I wanna bring up it's funny that you bring up the fact that they're you know, Opus or excuse me, Anthropics kneecapping on the models. it's funny the cheaper models down below with Gemini are getting cut back on usage too. So that's happening soon. so they're doing it too, okay. Yeah. Yeah, yeah. How they communicate it though? Did they communicate like just through a press release type of thing or they actually communicated inside your account, which was fascinating. I saw it happen. Okay. About two weeks ago. Now it didn't bug me because I'm like, whatever, I already know what's what's happening. And it does you know, I'm not using free models because I pay for them all due to the fact that I want to keep data within my own ecosystem. But what's interesting is they actually had a notification pop up in the app and didn't go away until you went and read it, which is interesting. So like I think it's more responsible because they are giving enough where if you go in and start using it, they're gonna give a notice to you. Yeah. But at the same time If this is the trend of the frontier models, it goes back to the same statement, right? Is this really where you want to be if you're trying to take full advantage of it? Are you just gonna roll with the punches and use it the way it's you know, way it's set up, or do you want that freedom that you're looking for right now? Right. You know, and I and even though open code's been hacked, I I think platforms like that make all of this much more compelling. Cause then you have options. Yeah. Like remember we did Hugging Face there a few weeks ago? Like that was cool. Like there I think I I seem to recall it was a hundred and sixty models or something like that they had on there. Yeah. So like that's I I I think that that's kind of the the way the future and when I set up Hermes, like I I have more than one model running off that thing, right? It depends on what I'm doing. if it's very rudimentary stuff, I'll run local models. Mm-hmm. Like why not? If it's just basic workflow stuff. translating stuff like you can run that off a local model. But now if it's complex reasoning, research and all that, yeah, you might want to use something obviously it has access to the internet and all that. I get that. So I think it's good that that we have options going forward on how we can use different AI tools and they they're gonna price and compete with each other. I I don't know how they're gonna make money. that's a different thing. I think they I think they have to substantially increase the price of all these things, but just to offset all the the data center costs, but I think competition's good. It's yeah. It's good to hear Google I I I haven't looked into it. I haven't tried it out yet. I I admit when it comes to Google and Gemini, like I'm kinda limited to I I see it when I search 'cause I still use Google for search. Right. I do the odd Gemini thing on there if I'm just like I might just pop in there to ask it a basic question that I know right about relevancy. Like it's or current events. Like if it's somet 'cause I know gu in my head, maybe I'm I don't know if I'm right or wrong, but in my head, Gemini should have access to the latest news because Google's minding all the news and scraping all the news all the time. So I I use it for that kind of thing. So yeah, I mean, I think it's good that they're competing, but I did read that they are becoming much more competitive against the other ones. Not to mention I mean, you got four of them now, right? Ru very competitive with each other with Facebook and with Google, OpenAI and Anthropic, which is awesome, I think. So yeah. Well, if I if to close this thought out before we move to the next one, I think if I can give the listeners one piece of advice, and this is more like the lesson that I know I'm learning as I go, but also what others are learning the hard way, is, you know, just like anything else, price really shouldn't be the strategy. Like it it really shouldn't. And you gotta find the best fit for your company. And like even investors are saying evaluate the cost per business outcome rather than the cost per token. And I think that that right there is speaks volumes because I'm in it, right? Like that's the the hardest part. And I I I I I feel bad for the folks that are just c entering in this world right now. It's overwhelming. There's so much stuff going on and there's so much just static, so much white waters. You know, it's like I always think about it as like, you know, my son has to sleep with a sle sound machine. I think about it as that overwhelming sound of white water just continuously wrapping around your head and you're just you know, supposedly that helps you sleep. I use it too sometimes. But when you're trying to focus and you're trying to like get your business off the ground, that white water is a lot to handle. Like it's just a lot. So I I just would say it's really important that you find the best fit, but don't don't don't focus on the price right off the bat. Like you got to focus on what's gonna really work for you in the first place. Yeah. And I and just to add to that too, it might not need to just be about the price. It's like what model do you need? Maybe the very basic model, which just happens to be the cheapest, is the best. Yep. Right. And I think that a lot of people out there kind of We're gonna have to do an episode on archite architectural design of AI at some point here. Right. That's when we start doing getting orchestration harnesses going. Yeah. You're already making me excited. Don't do that. Don't do that yet, please. Definitely geek out on this one today. Well actually, you know what? That's a great segue into experimentation. so again, you know, as I said at the outset, Chat GPT, I I I I'm sure many people do this, or maybe not, I don't know. very privileged, you know, in our work that we have access to all kinds of tools. but I do find like which one, which model do I work with the most on my on my Mac. And I go th I went through phases where it was clawed and then I went to Chat GPT back and forth. but back in April, I was using Chat GPT and I'm like a hardcore task managing kind of guy. and I love to keep things organized. And I was I I went through this phase where I'm like, I'm gonna try Apple Reminders. And then I went, Man, you know what? I can't get Apple Reminders on my other computers. I have Linux machines, I have Macs. I I kinda given up on Windows. Me and Windows don't get along anymore. But I I have all kinds of computers and it's a lot. This has been going This whole yeah, anyways, that's a whole nother topic too. But point is like I I I wanted to get Apple reminders moved into Todoist, which is another task manager I've used for years. and it's multi platform. it works on the Android phones I have kicking around. It works on the Linux computers I have kicking around. And heaven forbid I do use a Windows machine, I can use Todoist. So I asked Chat GPT back in April, hey. I need to get everything out of Apple Reminders and into Todoist. So what it did is it actually wrote an Apple script for me. And it was awesome. It just created this Apple script. It would talk to the API inside of Mac OS, which connects into reminders to grab all my reminders. And then it just created connections out to to doist, which is web based. although it does have desktop apps. But ultimately it's all web based, right? So it ran this script, it took all my tasks that I had in there, hundreds and maybe thousands of tasks, and just, you know, moved them all over to to doist. And it worked great. Problem with that though is there there's a lot of things missing. if there's images, they handle images and attachments differently between to doist and reminders. Sure. So that was an issue. So I kind of had to reference back and forth with reminders. So either way though, I got all the tasks in there. I had to update like that's the other thing too, like recurring like if something was recurring or had a special type of recurrence, like it was really hard to distinguish the difference because remember, I used AI to create a script code to go do the work. Right. Right. So I thought, you know what? I wonder what happened if I were to do it again. So what I did is I went and created I took to doist and let's see how I can bring to doist back Taple Reminders. I'm not leaving to Doist, but I I wanted to see how this would work. So do you think I used Apple Script? I did not use Apple Script. No. I asked Chat GPT to use my computer to transfer to doist to reminders. And so I used it in agentic mode basically. And I gave it permission to access Apple reminders, access to to Todoist. And interestingly enough, it didn't use the API at all. And I didn't give it the instruction, I gave it pretty bare minimum. I just said I want it all over there, I want the attachments and everything. So step one is Chat GPT said, Okay, cool, I need access to all those tools. So I gave it access. And gave it access to Chrome. Mm-hmm. Then It actually exp it used to doist to export from the application a backup file. interesting. So it grabbed the backup file and then it imported the backup file. And I'll be honest, I'm not I haven't even checked to see how it did it. That's the best part. I'm not even sure how it imported. I don't know if it used Apple script or if it just it happened. so it got all the Apple reminders, it all loaded it in there. Right. And then what it started doing is it's took all my groups that I had in Todoist. using the cursor it just went and created all the groups and Apple reminders on its own. I like to put emojis inside of the titles just to kind of clarify you know like if I have like a list of books it has a little book in there. So it went and it went to every single category, went to rename, it added the emoji and just updated everything. Then the next step was to verify recurrence. If I have recurring tasks like I have everything from paying bills to you know, vitamins and things like that, all right in it. So it went in there and updated all that too. cleaning the coffee machine cannot forget about the coffee machine. There you go. And so it updated all those on there perfectly. And then the last thing, it's like, okay, well I gotta go get your attachments. So there isn't a way just to export attachments and drop them into reminders. And once again, the problem is you can't actually export and import the same way because they're the the use the different formats and all that. And I think I I'm pretty sure you can't actually attach a PDF or a document. You have to you can only do images. So what it did, it went through all the attachments and it took a picture of the attachments and added it into reminders. And then by the end of it it just confirmed everything and it's like it it's a mirror mirror example. There's a couple errors I couldn't it couldn't get some of the attachments out of Todoist. Sure. But Ultimately I called success. It took longer than Apple Script just because there was a lot of manual work, but I I just let the computer run. I went for a coffee. I came back and like, well, look at that. That's probably why I don't know how they imported into reminders, but that's a stark difference going back to your analogy at the outset of of this whole episode. How you're right, brick phone versus iPhone 17. my god. It was cool to see that, even though I know it can do that, it always It just, you know, mesmerizes me every time I do things like that because it just saves me so much time while I can get other things done. So have you done anything like that? I mean, I have recently started playing around with making my own agents for personal stuff, right? And this is through the actual Gemini world because I have a personal account and I just have been playing with the new tasks that have come out, things of that nature. We'll get into that later. But I actually have been starting to play with that. And one of the cool things recently that I did was I was recently I think we talked about it, studying for the anthropic certification and I had it connected to my Google workspace. And what allowed me to do was basically schedule in the around what I already had in my schedule, the study blocks in the way in which I learn. And it was basically just taking and every time I'd miss something 'cause I'd have to confirm I studied. Every time I missed it or maybe something came up, it would recalibrate the schedule to make sure that I I got everything in. And what was was interesting about is I wish I had this in college, honestly. Like this would have made my life so much easier. But like the interesting was sometimes it would change the whole the whole session. It would just go, hey, you know, you missed this. Let's do a review on this because I have a feeling you're gonna need it. And it would just help me kind of like recalibrate where I was left off. And man, it was that that's so simple and honestly in concept, but the the amount of intelligence you have to have and connections in a structural manner. really had to be set up intentionally. And I think when you were talking about how you were doing tasks and whatnot, my the thought that came in my head and the question I have for you is will you and I think similarly when it comes to trying to find a more how and I I don't even want to use the word efficiency. That's the part I don't like is everybody's so hyper focused on that right now. It's kind like in the back in the day when you were like, I'm collaborating. No, you're not. it's the concept of like reimagining the world around us. And I'm curious What enticed you to just go and say, hey, I'm just gonna go try this? Like what what made it possible for you to just to sit down and do it? I I think more so, yeah, in this case, the reminder is one is more I'm always trying to see what Apple's coming up with. Like Okay. Okay. And and they they had their new release here about a month ago of macOS and iOS and all that. So I'm like, you know what? Let's see what's new. maybe there is some new stuff, but nothing that kind of stood up to me. Sure. so I still I still prefer to doist in that regard. Again, the multi platform thing's always going to be an issue. but yeah, I just it it's one of those things where it's like, let's try it out, let it go do its thing, and see what happens. And like, man, I got like I know you're talking about orchestration and you know setting up your own agents and all that too. I I'm I I'm thinking of creating my own, you know, chief AI officer, staff, staffer, whatever. And it's the idea behind tracking my finances, for instance, right? Like investments, like and just give me updates every day. versus the old ways. Let's go research what's go reading the you know, I don't know, twenty years ago my dad would have I guess my maybe even longer. Thirty years ago. my dad wasn't around twenty years ago. So thirty years ago. This is how I'm getting older there, Mike. but thirty years ago my dad would I remember as a kid h him opening up the paper. And I'm sure he's probably looking at the stock pages, stock markets, seeing what's going on in the paper, right? should I, you know call my broker and those kinds of things where I I want to get to the point where I get an update in the morning. Here's how your stocks perform. Here's some suggested moves. Here's some other ideas. So for that piece. And then also like on the financial track and like what bills are due. What things I need to keep an eye out. It's is there a mortgage renewal. I don't know. Whatever it might be. But that that that kind of assists me in my to day day to day, I think is, you know, I'm I'm always getting more comp cognizant of those things. And Yeah, but yeah. So anyways, that's that's utilities one one for me. Yeah. I thought that was great, just be knowing that I can, you know, give power to my AI to go get some stuff done. I think it's pretty cool. So But yeah, so why don't you why don't we turn it back to you and let you kinda wrap up our story segments here. I don't know how the heck we got these stories so aligned, but this is another segue that actually helped tremendously with what we just talked about, and that is You know, I always like to go back to how is it impacting the humans? Like what it what is it doing for us or what isn't it doing for us? And I want to go back to a story from I think it might have been last last episode where we were talking about the acquisition that Disney made of that company that went that went bankrupt. And it was around the character the character character AI, right? And what was interesting though is, you know, the buyer paid And just a reminder of of facts. Fire paid of 20 million US for the company's assets. Here's the interesting part that came out this week after the acquisition had basically started to pan out, is they they it then offered jobs to 60 existing employees of the staff and they're going to be paying 60 million a year to these employees. So if you were to total out the acquisition, we're talking multi-billion at this point. Like we're or you know, over year over year. And it's all due to the fact that the people that built out what the infrastructure was, there's value in that. So kind of going back to what I asked you earlier, I think the story speaks volumes to what we what we were not reading in the news, which is companies blaming AI for layoffs when in tr in in fact it was supposed to be repurposing the people to then develop said things and infrastructure they're gonna handle what AI is gonna handle. So I think the lesson that I wanted to like explore a little bit more was even though the tool ended, what the team knew didn't. And I'm curious from your perspective, Derek, from what you're seeing in the world, and especially with the folks that you talk about, what does that change with the perspective of these not only just the acquisition or companies trying to develop AI and develop the efficiency they're looking for? But what does this change in the conversation around structuring transformation like operational transformation or rev ops or whatever it is the company needs? Like how does this change that discussion and how can we use this to our advantage? Well, you know, what immediately comes to mind, and we're I was just talking about this with someone else the other day, was the whole Facebook Muse thing, right? Mm-hmm. Mm-hmm. And is it Alexander Wang? I think he was running Yeah, I think he was running scale AI and and I'll probably get my facts wrong since this is on the spot, but I know Facebook made a sizable investment into scale AI. They got Alexander to come over to Facebook to run Muse and we all saw Muse a few weeks ago, which is pretty cool. although security is a privacy as you know with me. But I the the interesting part is the amount of money that they pay the guy is substantial. Like it's it's millions and millions, probably worth billions. but it added a market cap, I think, of two hundred billion dollars to the company. Right. And I I think that that's I think what people are looking at now is how much value these people with this knowledge can actually bring to the bottom line and and this isn't so much the bottom line but more the market market value of a company. And that's probably a prime example I I'm assuming and I don't know what the details about the character AI. I mean I have questions already about so they just left the they left the corpse behind, basically the corp the the company. Yep. And the creditors are probably not thrilled about that. So I I'm sure that's a whole different financial segment to go on, which we won't get into on this one, but but I mean you're acquiring the people that come up with the ideas and the talent and the the the this talent. And I don't think talent's gonna go away. I think it's ever more critical. I do believe that the job displacement, there's like software engineers is a great example. Mm-hmm. There's a lot of bad software engineers out there. I've worked with a lot of them. I've hired a lot of them. I've hired a lot of great ones too. But there's a lot of bad ones out there. I've worked with a lot of bad ones at other companies too. And it's like anything. Like you think about sports, right? You got some you got some great football players. You got some really bad football players. I beat do you play football? I did. wait, wait, which football? Which football are we talking about? We're done. Well, American football. American. American football. Okay. Okay. I want to be clear. Fair enough. but you know, so you played you didn't play in the NFL though. Right? Heck no. No. You know, you think those comparisons that there's a lot of bad software engineers out there. But if all of a sudden, if you think the NFL, the NHL, the NBA, all these, these guys get these professionals get paid a whack of money because they're really, really, really good. Right. And they're in the spotlight. They can handle the pressure. They can handle the mental aptitude that's required to be that level. And I think that it's very similar in software engineering too. And like I I I'm a programmer. I can program. Am I the best in the world? No. Number one, I'm I hate programming. It's more fun now when you get to code with AI and all that. But ultimately, I I I totally see where you you're gonna see the best talent in this world get paid a lot of money. Yeah. And that's Totally fine in my mindset. And I think that the the people that are going to get upset about job displacement just haven't had the motivation or the attitude to pivot to get better and get better. So you as that a that you know football player, if you really want to be the best football player, maybe you wouldn't have made the NFL, but you probably could have worked really hard, worked out every day, took care of yourself, got the right coaches, the right mentors, and all those things to help you be better. So And you probably would have made something out of out of that, right? I think you totally would have. Maybe it would have been NFL, but you would have got something. You came out to the CFL up here in Canada. That's where everybody ends up. It's true. It's a true story. But you know, like that's what I mean. Like I think people put the effort into things. I think they'll find opportunities for themselves. It's it's the ones that are gonna sit on the sidelines, be like, I went to school and now I don't get a job. It's like I can't find a job because I'm you know, I keep applying to all these jobs. It's like get out there and network. Go meet some people. And I think you might be surprised if you put the effort into it. So, anyways, that's a whole summary of my thoughts on talent wins. Yeah. I and I agree with you. And I think the only thing I really have to add before we we part from this subject is I think this pattern's gonna continue and it's gonna be become a lot more of a bigger discussion around who's actually overseeing. Like you said, like you're gonna be building agents to oversee, but even then you have to oversee the agents, right? You have to know what you're looking at to a certain degree. And I think that's the biggest challenge that a lot of people have is they want the system thinking it's autonomous. And it is to a certain extent. But then like even the conversation I had this morning internally, you know, like we're reevaluating how things are measured because things are evolving. Right. Like we're not there there's not a whole lot of human touch happening with certain aspects of the business. And we're going, okay. How do we get those indicators so that we know that it was measured correctly or taking the right action or whatever it is that we're trying to track for the audit logs? And I think going back to kind of the the the first story you had of the hack, it's all about how what access you provide in a meaningful way, but then it's also just just as equally as important as what access you don't provide. And I think it's the same aspect for the humans going into this world right now is if you're constantly focused on. How I've always done it, I I'm sorry to tell you, but that's not gonna be the world you live in anymore. There's a few industries that AI is probably gonna slowly and very, very, very slowly hit because there's not a whole lot it can do now, but it's coming. Like it every it's gonna it's gonna impact everywhere in some degree. You just have to know how to adapt at this point. Well, I think, and just to finish off this this part, I think that there's two things that I think about as you as you talk through that. Number one. I can't remember I saw it on on Twitter X. somebody mentioned, you know, because you can ask Grok on there, which is actually pretty cool. I'm like, hey, Grok, is this tweet real? but you asked Grok, like, hey, what do you think the most common job title will be in five years? And then it was like A AI agent orchestrator. Yep. So which makes a of sense. I mean, we're pretty much doing that now. and then the second thing that popped in my head was And I totally believe on this. And I think we touched on this the previous episodes that the AIs cannot be creative. Right? They don't have the ability to be just because the way they're built, it's a lot of and of all people, Ben Affleck was I was reading an article about him because he just sold his AI company to Hollywood. I can't remember which studio bought it, but he built this company that that helped that uses AI to help support and build out film. And he sold this company, but he was talking about the what AI really is. And he put it in the in the simplest terms how it is a really complex logic. It's really there's a lot of different algebra in it. Like he really dumbed it down as brilliant. at the end of the but the problem is it's based off all the training data it has and it's a lot of this complexity involved with with code. And but the thing is it can't be creative. And maybe they maybe we solve that. in some other future AI generation, but I don't it's not there today. And it it the the pathway it's going it's just not there. So you have to yeah. You said it. We have to train it to be as good as it's going to be. There there lies the the crux. It's chicken before the egg. It's you know it's everything that we've ever talked about. So yeah. Yeah. So be creative, I guess is kind of my story. But yeah. Absolutely. Well that leads us to segment time, which is First time we're doing this one. so Mikey's gonna set up here a round of AI fact or fiction. So we've we've called upon AI. I don't know which tool are you gonna use for this one. I've got a good one today. It's actually common and you have access to it right in your backyard. So we're gonna we're gonna do a little fun story here. And it's it's I I don't know what's gonna happen. I'm gonna be honest with you. I'm actually a little terrified. I didn't The San Francisco one worked out pretty good. So I have confidence. True. But I I did preload a few things because it does take a little bit of time to load sources. So I didn't want to spend time watching a bubble circle. So we I did preload. But maybe maybe Derek, you explain the segment a little bit of what we're gonna do here, and then I'll get into kind of the parameters I set first and then we can start working with it. Sure. So AR factor fiction. Basically what we're gonna do, we're gonna ask the AI to present to us a some basically a story, a concept or something. And it we have to decide if it's fact. Like is it real? Or is it fiction as it's made up. It's not real. and then, you know, Mike Mike and I are both gonna take a stab at it and then we'll get the AI to kind of tell us if we are right or wrong. or who looks sillier than the other one. So yeah. I'm curious. I I have no idea what prep you've done on this. So I'm fascinated to see what we come up with here. Yes. So a little a little backstory and we've got to go back of actually God, it feels like a few years, but it's actually only a few months ago. My goodness. d if you don't remember, Google's ecosystem had, you know, they were starting to phase out Google Assistant and they were starting to produce different segmented style of usage for the different ways in which you might use AI. So this is LLM. Right. This is this is the early stages of Igentic. And what's hilarious is I used to use this, and I still do to a certain degree, Notebook LM. Do you remember that one? yeah. You don't hear much about that anymore. Yeah. No. So I was just curious what came of that the other day. And what's funny is it it's not called Notebook LM anymore. It's called Gemini Notebook, which changed the way in which I thought about it. I'm like, huh. I wonder what it does. So Before we get too far much farther in there, what I'm gonna do is I'm gonna share my tab here. And again, this is this is just, you know, in your workspace, but it's the same look, but there is a difference. There's a distinct difference. Think about this. This was the early days of what Anthropic and OpenAI had done with their platforms, with projects. It's basically the notebook is consent consolidated data that it sources from, but it still pulls from external if you ask it to. So what you're looking at here is basically the idea around how do we how do we process, you know, things in a more systematic way, but in a focused way. So these this to me, notebook is kind of like the starting point of folks trying to understand the data parameters of AI, right? That's kind of where I see this going. So what I've done so far is number one, I have loaded a ton of AI factor fiction sources from the net I did not review them. I just asked for it inside this little web search and it added the features in there. There is one there is one little feature I added in there. I added in our podcast from the YouTube and we're gonna see if we can determine if we said that or not as well. So yeah, this is this is why it's gonna get weird. We're gonna fact check ourselves. That's terrify that's terrifying. It's terrifying, right? I'm putting ourselves out there. We're gonna get vulnerable today. All right. So anyway, so I've loaded a a couple of or a few sources. What we're gonna do is we're gonna prompt this first, but you can tell where we're headed if you didn't if you look closely here. We've got a little feature called quiz. It generates an AI interactive quiz based upon what it's finding. So first and foremost, we got to set this, set the stage. So AI factor fiction. We're gonna create a quiz in real time, ask it to provide you with claims about AI tech, some real, some fabricated. Do a handful of questions, see if you can determine where the true or false is. One thing I'm gonna add, I'm gonna say source. One question and we'll say from the hardly artificial and I spelt it wrong podcast. And we're gonna let this run and we're gonna get things rolling here. As soon as we do this, then we're gonna open up the quiz and we're gonna see what happens. Because I'm curious how this interface works. I've never used it, so I have no clue what's going on. But before we get anything as it refines it. What are your thoughts on what's gonna happen here? I wan I'm gonna get a hypothesis. I I'm curious to see look that's already come up here. So I mean it already popped up. So I first you know what's funny? This happened so fast. god, it's so fast. Like I think it's so cool. Like I think it's amazing how fast it's happening now. although for people being hacked, I'm sure they don't love that. But terrible for you know, running our factor fiction quiz, I mean whatever. So Yeah. I mean let's should we just let's just go for it? What do we got here? Question one? All right. So we have a you wanna show the okay, yeah, yeah, yeah. Okay, go it. So we got we got a few things we gotta set up here. So we're we'll just do number of questions, we'll say standard default, whatever. We'll just see where this goes. We might end it early. We'll say medium default. I just want to see what the the default settings show us. what I will say though is the multi-select, multiple choice is fine, and then 19 sources, and then from there we're just gonna like the topic, we'll just We'll just you know, use, I'll say use the source within the documentation. And I'm just I'm just we'll just see what happens here. So here we go. Okay. So it looks like it processed is looking good. So we're gonna go ahead and get into the quiz here and we're gonna open it up. Now I'll do this in a separate window so it's nice and full so we can see it. All right. This is way prettier than the one we did in San Francisco. Yeah, yeah. It actually is a question. But this is great. This is good. I like it. All right. That's good. So there's twelve questions. We won't get through all of them today, but we're gonna try to get through a few here. So in the original Turing test framework introduced in nineteen fifty, what benchmark must a machine achieve for a human judge to evaluate its intelligent? Ooh. Yikes. That's okay. We'll okay. Well we'll we'll we'll roll with it. Yeah. Yeah. We'll try the factor fiction next time, but I think it was it's a hold on. a human a human judge is unable to distinguish the machine from a human in a thirty minute face to face. I didn't know there was a thirty minute, but I definitely know that's the definition with thirty minutes, but we'll see if see human j identifies person with less than fifty percent. Okay, well maybe I don't know. There you go. All right. So these are tough. We're gonna we're gonna just see how this rolls. So we'll just go ahead and next. It's because you hit medium difficulty. Yeah. I I wanted last time we got all right. We can't get all right. We can't seem like we know everything. That's that's that doesn't work. All right. So beyond the I hope our kids never watch this though. They think we're always right. Right. Exactly. So beyond raw peak floating point operations per second. So this is flops, evidently, which factors represent primary system level constraints that determine the deliver performance of modern AI architecture. So it's a multi it's a multi select. So let's see here. I this is a tough one. I mean, if I had to pick, I would pick this is I would pick high I HBM, high high bandwidth memory. Okay. I feel like it's You just wanna stick with that one? I feel like this one too. I don't I'm not technical enough on the hardware side to really understand the rest of it. But I'm I'm gonna select that one just for fun. We'll see. I just know memory bandwidth because I look at that when I'm looking at RAM for token generation. we missed one. So we got two. We missed one. Facility level. I remember this. I actually do remember this from a conversation. Okay, that's a good r reminder. Okay. All right. All right. One more. Let's do one more. Let's do one more here and then we'll Okay. How was the supercomputer Nautilus able to generate predictive geographic and mood forecasts such as narrowing down a key location or tracking potential s sentiment prior to the Arab Spring? Whoa. This is like Jeopardy. I I'm I'm lost. Yeah, I don't think five hundred for Arab Spring. wait, wait, hold on, hold on. This is interesting. I hope Focus on the source data set used to construct a web of trillions of relationships mapping location and interesting. by analyzing the emotional tonographic location. Yeah. Well you just literally said that. Okay. Yeah, let's see if that yeah. There you go. All right. We got one right. There we Let's go let's go with this is the one. I want to do one more. Okay. This is interesting, isn't it? this is a fun one. Why did competitive success In heads up, Texas Hold'em poker mark a fundamental different milestone for artificial intelligence compared to earlier victories in chess or Jeopardy. Wow. Don't hit a hint yet. Don't hit it yet. No. why so poker decision trees contain vastly more legal board positions than I poker rules change dynamically based on dealer choices. We're gonna Poker calls poker requires real time natural language to I'm getting gravitating towards this 'cause this seems like it's like st it's strategic. I think it could be C C N D. Yeah. Cause it's like dude with the sunglasses on type of thing, right? Yeah. All let's see what the hint says, just out of curiosity. Okay. It's the visible information. Yeah, so I think it's I think it's C and D. no, you only pick one, right? It's not like one. Yeah. I'd go I think D's the stronger. Yeah. Yep. Yeah. Okay, one more. I I'm just okay, let's just do five 'cause I I'm I'm I'm I'm liking this. Which of the following statements accurately describe the structural relationships among key fields within artificial intelligence. Ooh, okay, I like this one. you know what? For it's you know, full pas on our part. We should be reading so for podcasts, so Yeah. First option is generative AI is over is the overarching field that includes traditional rule-based robotics, expert systems, and predictive statistics. The second option, deep learning, is a specialized subset of machine learning that relies on multi-layered artificial neural networks to identify complex patterns. Large language models are a class of generative AI systems, trained on vast text data sets to analyze reason and generate content. And last, Machine learning is a subset of AI that enables systems to learn and improve from data without explicitly programmed for every scenario. So which of the following statements actually describe the structural relationships among key fields? So I think I'm gonna eliminate one, I think, because the large language model is not a class of generative AI. It's actually a different class of AI altogether. And it's not a it's not analyze and reason. It's actually predicting the actual the next. subset of words, right? So that's a little bit different from generative. but the other ones are challenging for sure. Man, this this is not medium difficulty by the way. Gemini notebook. Gemini doesn't know me well enough or you right? So I mean I I th I mean I think I don't know I don't even know is deep learning is that a a thing? I don't think it is. I think that's that's a that's a to kind of play you off a little bit because deep learning is like almost like deep seek and to a certain degree. That's what it reminded me of. I like, that's I don't know if that's a thing. I my vote would be, and this is completely uneducated. At least it feels like it. generated AI and and machine learning in my own. Yeah, that's kind of where I was. Yep. All right. Let's just go for it. So it was something. All right, we completely deep learning. So the the the winners are deep learning. large language models into machine learning. Okay. So we both generate AI is garbage. It's garbage. At least the way it's described on here. So so listeners, if you're on podcast, you know, check our YouTube video. You can kind of see the the actual questions on there as well. we'll link to that in in the show notes there. But cool man. All right. Well we should probably we should probably land this airplane here. yeah. This yeah we this is I think I feel as usual, I feel like we keep going on here and I hope hope listeners you enjoyed today's talk. today we covered a lot of cool things, everything from usage on Chat GPT, hacking again, you know, the whole business side of things. and you know, it's just really cool, you know, some of the things are going on in the AI world right now. yeah, anything else you want to add, Mikey? No, I I just that was fun actually. It was hard, it was challenging, but it was kind of interesting how it perceived the data that it received and gave us something that was in fact challenging. I'm I'm impressed. I I actually am gonna go play with that a little bit more. That's pretty cool. I'm not sure I ever want to see that again. But with that said, on behalf of Mike and myself, thanks for listening. And make sure you subscribe or follow us if you're s if you're on YouTube, hit that subscribe button if you're on Spotify, if you're on Apple Music, if you're on Amazon. iHeartRadio, wherever you're listening to us from, make sure to s to follow us on there to listen into our weekly episodes here. And we'll see you again next week.