Fullstendig transkripsjon av EP2 Garrett Hillman

excitement that I had of what it could do. 00:03:00.560 --> 00:03:06.879 And now where we're at four years later is incredible of what it is capable of doing. 00:03:06.960 --> 00:03:11.280 Uh the use of AI agents, it's it's the sky's the limit right now. 00:03:11.520 --> 00:03:16.400 Yeah, you know, this the speed that this is progressing is just absolutely mind-blowing. 00:03:16.479 --> 00:03:18.479 You know, we've spoken about it before. 00:03:18.719 --> 00:03:32.240 Things that you didn't, that you're struggling with now just become completely automated in three months' time, six months' time, and definitely things that I uh was struggling with uh a couple of months ago now just seem to be non-problems, right? 00:03:32.479 --> 00:03:35.919 Uh the AI agent seems to have taken those away. 00:03:36.080 --> 00:03:39.520 But I want to come back to a specific thing that you mentioned. 00:03:39.599 --> 00:03:56.000 Uh, there's uh a podcast called This American Life, and they had these two uh agents that were speaking to one another, and they were trying to sort of break out of this uh environment that they they had, and they managed to they figured it out eventually. 00:03:56.240 --> 00:04:05.520 But I'm interested to understand what you think that means from a health informatics perspective, when you have these agents potentially communicating with one another. 00:04:05.680 --> 00:04:09.039 And and don't answer the question by just saying acknowledge. 00:04:09.280 --> 00:04:14.479 And it's definitely worthwhile going to look at this, going to listen to this this podcast. 00:04:14.560 --> 00:04:15.280 It's brilliant. 00:04:15.599 --> 00:04:30.240 Yeah, I I think within the healthcare IT space and informatics space there is so much uh necessary trust between the patients and the providers and providers to providers. 00:04:30.639 --> 00:04:38.000 Um the trust is not proven yet by AI in the healthcare sector. 00:04:38.319 --> 00:04:51.279 Um I think we're all seeing prime examples of marketing being done for AI and um good use cases, probably more on the operational side than the clinical side. 00:04:51.600 --> 00:05:05.600 But we really have to get mature workflows, proven workflows that involve human-in-the-loop, uh decisions that are repeatable, that are provable. 00:05:06.160 --> 00:05:18.639 And uh I think my concern and I think general concern is healthcare as it's been since the beginning of time is it's highly transparent. 00:05:18.800 --> 00:05:27.439 You have to be able to know what is going on, uh what people are doing, the decisions that were made, um basically everything is auditable. 00:05:27.680 --> 00:05:30.959 The the concern with AI is it's this black box. 00:05:31.199 --> 00:05:48.160 And I think until we can get to a point where we can truly see a transparent view into the decisions that the AI is making as a part of clinical support, uh clinical decision making, um, there's still going to be that distrust. 00:05:48.319 --> 00:05:54.800 So there's got to be a method to be able to identify what's in the black box of AI. 00:05:55.040 --> 00:06:02.639 Yeah, I remember when they started coming out with LLMs in healthcare, there was this whole idea of, oh yeah, you need attributions for the output. 00:06:02.800 --> 00:06:08.000 But of course we've moved past that now when we're looking at agents and agentic working and everything. 00:06:08.160 --> 00:06:11.920 And the hallucinations are a major issue. 00:06:12.000 --> 00:06:22.480 There's there was one thing where uh someone put a uh an old image, an old photograph onto Claude, and onto ChatGPT, and said, hey, colorize this photo. 00:06:22.800 --> 00:06:38.720 And uh uh ChatGPT did an amazing job, colorized it perfectly, but also added an extra person into the photograph because it just assumed there was a blank space and it obviously just assumed that based on its logic that there needed to be another person there. 00:06:38.879 --> 00:06:42.639 And you would you don't even notice it when you're looking at the colorized photograph. 00:06:42.720 --> 00:06:58.879 But in healthcare, imagine that happening where the agent just assumes that someone should be on certain medication that they're not, there it was a human decision for that person not to be on the medication, and the agent just automatically uh prescribes that medication because that's that's what it thinks it should be. 00:06:58.959 --> 00:07:05.519 And so I think that there's real danger with um deploying agentic workflows in a healthcare environment. 00:07:05.839 --> 00:07:22.319 Yeah, and I think that speaks to healthcare is such a busy environment of all the workflows that are occurring, all the interactions across roles, that bringing in AI, AI is extremely good at creating. 00:07:22.560 --> 00:07:28.079 And it can create to the moon uh documents, output, artifacts. 00:07:28.319 --> 00:07:48.560 The the thing that is gonna kind of become the issue is there's still a finite amount of time that a a nurse, a clinician has, uh, a radiologist has to review it from the human perspective, uh, that AI can do that 24-7, 365. 00:07:49.120 --> 00:08:01.519 So I think there's there's got to be a method to allow those humans the amount of time and effort that they can look at the information, the output that's being given from the AI. 00:08:01.759 --> 00:08:19.600 Because if if we uh kind of come into the situation where AI is creating such output that the humans cannot keep up with what the output is and truly give it the time of day that it needs, especially in healthcare, to review that output and ensure that that output is correct. 00:08:19.759 --> 00:08:44.720 Like your example, if AI is creating that image and the clinician or the nurse is just assuming that's correct, and they're not actually looking at the details of, hey, an additional person was put into this image, it's those types of situations that are concerning that humans just by nature are going to

assume it's correct and gloss over. 00:08:44.879 --> 00:08:50.879 And that's where the problems are going to be, and that's where we need uh checks and balances with these this technology. 00:08:55.519 --> 00:08:56.000 For sure. 00:08:56.159 --> 00:09:01.519 Now, from a personal perspective, I want to pivot this slightly into personal agents, right? 00:09:01.840 --> 00:09:14.559 And I know you've done a lot of work around uh developing personal agents, and I want you to speak a little bit about how you've set that up, whether it's been time-saving, what the use case is. 00:09:14.720 --> 00:09:17.679 Like, is this something that we should all be exploring now? 00:09:18.080 --> 00:09:18.960 I think it is. 00:09:19.120 --> 00:09:28.960 I I've found some incredible, what I would call realistic, relevant, everyday use case uh examples. 00:09:29.200 --> 00:09:36.559 Uh, some of the examples that just come to mind are uh taking a kid's full school year calendar. 00:09:36.799 --> 00:09:39.279 Uh schools will send that as a PDF file. 00:09:39.600 --> 00:09:42.639 They'll almost never send it as individual calendar invites. 00:09:42.720 --> 00:09:44.960 That would be easy to just import into your calendar. 00:09:45.039 --> 00:09:52.399 They send a document or a PDF, and you have to sit there and put in every single date into your Google Calendar, your Outlook calendar. 00:09:52.799 --> 00:10:11.120 And I've had the ability to just use my Chat GPT account and say, take this PDF, take all the dates that are in it, and throw it into my calendar, and it will create the 20 to 30 different calendar uh entries exactly for me, and that will save me 30 minutes. 00:10:11.440 --> 00:10:29.919 Um, examples are we get a weekly email from the school itself, and there may be changes to the school schedule, to events, and rather than me reading every single one of those emails, I can just query Chat GPT and say, give me a quick summary. 00:10:30.080 --> 00:10:47.120 Um I know last night I'm I'm doing some of my paperwork for my renewal of my Canadian permanent residence, and I had to completely create a brand new submission of paperwork, and it is nearly identical to paperwork I just submitted in July. 00:10:47.279 --> 00:10:58.639 So I had ChatGPT do something that I know when I did it two months ago was about two hours worth of work, it did it, and I watched it do it in about eight seconds. 00:10:58.879 --> 00:11:11.679 So those types of savings, just in terms of time, of effort, and then where you can put that time and effort as a human elsewhere of more value, of where you have interest at. 00:11:12.159 --> 00:11:30.000 The sky's the limit where where I see, and it's not a problem, but it's an opportunity, is where else can I use this type of technology to make my life more efficient, easier, and where I can use my finite amount of time during the day for something else that I want to do? 00:11:30.159 --> 00:11:40.720 Yeah, you know, the one thing is that uh when I'm

using these LLMs and agents and things, most of the time when I'm asking them to do something for me, I I don't spend my time doing something more useful. 00:11:40.879 --> 00:11:44.720 I spend my time looking at the screen just in wonder at what it what it's doing. 00:11:44.799 --> 00:11:48.799 I probably need to get better at uh doing something something different. 00:11:48.960 --> 00:11:50.879 But what about the automation piece? 00:11:51.039 --> 00:12:03.039 Do you still having are you still having to like go in and ask it to do specific things, or have you completely automated some aspects where it just does it automatically for you in the background and you just get an update? 00:12:03.679 --> 00:12:06.320 I I think there are some examples. 00:12:06.720 --> 00:12:22.240 There is uh one example, a very simple one, is like I said, I'm very interested in the news of what Claude is coming out with from Anthropic, what OpenAI is doing with ChatGPT. 00:12:22.480 --> 00:12:32.240 I have an automated scheduler that checks for news every three days of what's the most recent releases from both of these companies. 00:12:32.320 --> 00:12:34.000 And it just gives me a quick summary. 00:12:34.159 --> 00:12:39.679 And I get the message uh every three days, I look at it, go, oh, that's good to know. 00:12:39.919 --> 00:12:49.039 They've come up with this new feature, or a new model has been released, or a feature that I've been waiting for that I want to incorporate into my workflows has been released. 00:12:49.360 --> 00:13:06.159 Um I know there was an example kind of with work that came to mind was, and I know this is a feature that came out in I think Gmail a few years ago, but it's cool to see it incorporated or able to be incorporated into other email platforms. 00:13:06.960 --> 00:13:12.240 I've sent an email and uh tell me if I haven't gotten a response in one week. 00:13:12.399 --> 00:13:25.519 And so now I get an alert automatically without any touching, and it was one sentence of a prompt that says, Oh, you sent this email to John Smith, you haven't got a response in one week. 00:13:25.679 --> 00:13:27.679 This was the context, would you like to follow up? 00:13:27.840 --> 00:13:28.240 Yeah, yeah. 00:13:28.480 --> 00:13:45.360 And it's just those simple things that you as a human, if you don't have that technology, those are things you're gonna forget, those are things you're not gonna follow up on, and just having literally a a virtual assistant that's just sitting there going, Oh, hey, Garrett, remember about this. 00:13:45.679 --> 00:13:50.879 And the the virtual assistant, is that one virtual assistant, or there are multiple things that are happening? 00:13:51.039 --> 00:13:59.200 Like I'm interested in the idea of having like uh people have uh executive assistants, right? 00:13:59.279 --> 00:14:08.480 And they sort out their calendar and they sort out their travel and they sort out where they need to be, what meetings they've got, they make

they keep an eye on uh their their email. 00:14:08.639 --> 00:14:14.159 So do you have like one core assistant or are there multiple workflows that are happening at the same time? 00:14:14.720 --> 00:14:30.240 I think this is where a lot of the confusion is, and I don't think the like the LLM companies, Anthropic, OpenAI, Google, etc., have done a good job of explaining what an agent is. 00:14:30.960 --> 00:14:36.240 Um, an agent can be as simple as anything, and it can be as as complex as anything. 00:14:36.320 --> 00:14:38.799 You could have a team of agents, you could have a single agent. 00:14:39.360 --> 00:14:46.080 The way I look at it is uh those that use Chat GPT with just the chat feature. 00:14:46.720 --> 00:15:00.159 That is just what I would label as an LLM, a large learning um kind of experience of you're asking a question and you're getting one answer back. 00:15:00.240 --> 00:15:02.480 That's not really considered an agent. 00:15:02.799 --> 00:15:21.840 When you go into utilizing Chat GPT work or CLOD work, co-work, uh when you're bringing in kind of components into the use case that you have, if you're bringing in instructions, if you're bringing in connections to tools. 00:15:22.000 --> 00:15:49.919 So if you want a personal assistant that helps you clean up your inbox or create the calendar events, like the example that I gave, being able to connect to those plugins and those connectors, giving it the instruction of what is considered success, what is considered not success, um, how often should it try to attempt something, uh, when does it know that it's reached its goal? 00:15:50.080 --> 00:15:57.679 Um it's those components that then really help define to me the definition of an agent. 00:15:58.159 --> 00:16:17.519 The thing I really struggle with is um understanding the memory component and the the persistent memory piece, like where I need to be having these conversations, whether they're in a specific project, whether I want general memory, how do I know that the memory is actually uh being updated? 00:16:17.759 --> 00:16:26.159 Like, what would your advice be for someone who's trying to set up an agent but isn't sure if they're setting it up correctly? 00:16:26.480 --> 00:16:28.879 Because this is moving so fast, you know. 00:16:28.960 --> 00:16:32.000 When I first started doing it, it was like, where are you gonna put the memory? 00:16:32.080 --> 00:16:43.519 Like, I don't know where I need to put the memory, whether it's gonna be in my Google Drive account, whether it's gonna be somewhere in the depths of Claude on my uh on my uh PC or or wherever. 00:16:44.080 --> 00:16:46.639 What would your advice be when you're trying to set something up? 00:16:46.799 --> 00:16:49.440 Because the foundational element needs to be correct, right? 00:16:49.679 --> 00:16:53.679 Yeah, so I would say kind of looking, I'm gonna call it harness. 00:16:53.919 --> 00:17:01.759 So there's Claude and that tool itself is the harness, chat GPT with work is is the harness itself. 00:17:01.919 --> 00:17:15.759 So the first thing I would recommend is go into the settings of those applications and just look at what the configuration settings and what options you have to play with configuration uh settings itself. 00:17:15.920 --> 00:17:47.440 So I know uh thinking off the top of my head of Chat GPT, there's a setting that allows you to kind of limit your memory when you create projects and threads to just that project or thread, or do you want to allow the communications, engagements, the outputs that you've had across multiple threads and different projects to kind of be this larger brain that the harness can pull? 00:17:47.599 --> 00:17:55.759 So let's say, for example, you've created in ChatGPT a project, um, one that comes to mind, I'm training for a half half marathon. 00:17:55.839 --> 00:18:14.079 So I've created a project and I'm I'm putting in screenshots of my uh Apple Fitness results of my runs, and I've created a training plan, but I've not given this agent a any additional context. 00:18:14.240 --> 00:18:33.920 But I know in my other kind of interactions with Chat GPT, I've created a personal context skill that has information that I've decided to share with ChatGPT on my age, my background, what I do day to day, uh my daily schedule, my weekend schedule. 00:18:34.079 --> 00:18:52.319 And because I have the setting in ChatGPT from the harness perspective, it decided when I had that kind of uh engagement the other day creating out this half marathon project, it went automatically and fetched that personal context information. 00:18:52.400 --> 00:19:01.839 And it knows, uh, Garrett, you're 37 years old, you haven't really had a running plan in the past, you did some sports years ago in university. 00:19:02.160 --> 00:19:12.720 So that's a personal decision of letting the communication go beyond kind of the boundaries of that particular project to pull in additional context. 00:19:12.960 --> 00:19:20.960 But I think this is where uh individually you have to decide how are you going to use maybe your Chat GPT account? 00:19:21.200 --> 00:19:22.960 Are you just using it for personal? 00:19:23.200 --> 00:19:25.440 And that's the way I do it for this account. 00:19:25.759 --> 00:19:31.920 So I like to make use of all those interactions, all those different projects, all those different threads. 00:19:32.160 --> 00:19:43.119 So it it has this overall kind of universe knowledge of me, and it based off its recommendations will bring in context that it thinks is relevant. 00:19:43.200 --> 00:19:47.279 And you can always tell it, no, don't bring that in, ignore that. 00:19:47.440 --> 00:20:02.640 And I think that's where it really becomes valuable because it's as simple as saying, hey, uh Chat GPT, I want you to remember specifically that I like Nike running shoes. 00:20:02.799 --> 00:20:08.640 Never in the future recommend any other shoes to me, and that will save that to memory.

00:20:08.799 --> 00:20:23.359 So it's it's saving that context in memory to the application itself, and then even some of the newer tools, and I know at least especially for ChatGPT, it has its own baked-in SQLite database. 00:20:23.440 --> 00:20:32.480 So it's saving all of this information within the application itself, specifically on the desktop as well as on the cloud version. 00:20:32.640 --> 00:20:57.440 So it it's I think we've moved beyond what was kind of the case, which seems like a lifetime ago, but likely kind of at the beginning of 2026 when there was concern of oh, you may run out of context with your context window, you need to create summary files, you need to copy those summary files over to a new thread so it keeps that information. 00:20:57.759 --> 00:21:08.640 The tools now, uh as it is today, uh, is is much, much better at that compaction and understanding what information it should save. 00:21:08.799 --> 00:21:26.240 Um, but you as a user can always kind of take it upon yourself to say, save this information maybe to a Google Drive as a reference file, or save this information to a markdown file in my instructions. 00:21:26.400 --> 00:21:32.720 You can kind of save that off to static artifacts and still reference it when you need to. 00:21:33.440 --> 00:21:37.359 Hey, so you were speaking about uh some work you were doing with Obsidian. 00:21:37.440 --> 00:21:39.680 Uh I've got no idea what that even is. 00:21:39.839 --> 00:21:45.200 Can you describe because it sounded pretty interesting, uh, especially from a networking perspective, right? 00:21:45.359 --> 00:21:49.119 Maybe a little bit creepy, but can you describe what you're doing with that? 00:21:49.759 --> 00:21:56.799 Yeah, so I I came across Obsidian, which is an open source, call it note-taking tool. 00:21:57.119 --> 00:22:01.680 So if the listeners are familiar with what OneNote, which is pretty popular. 00:22:01.920 --> 00:22:04.240 Obsidian is very similar to that. 00:22:04.400 --> 00:22:15.519 Um, the way that it works that I really like is it is open source, it does not lock you into kind of the environment and the files like Microsoft OneNote would do. 00:22:15.680 --> 00:22:16.400 But like Notion. 00:22:16.720 --> 00:22:20.400 It like Notion, but completely open source free. 00:22:20.640 --> 00:22:28.240 And the file structure is files, is text files, is markdown files. 00:22:28.400 --> 00:22:30.079 So you have complete control. 00:22:30.400 --> 00:22:38.960 And if you create an entire notebook of years and years worth of information, you can literally take that in one minute and move it elsewhere. 00:22:39.119 --> 00:22:42.079 It just stores into a local drive. 00:22:42.559 --> 00:23:04.799 So what uh examples that I saw back, I think in April and May, that some folks were doing online was using, say, ChatGPT or Claude, doesn't matter, setting up instructions to automatically using scheduled tasks, write information to your obsidian vault, the way that they

consider kind of their notebooks. 00:23:05.200 --> 00:23:16.480 And what what I've done is I've used ChatGPT to first create the folder structure, the templates of the files that I like. 00:23:16.640 --> 00:23:35.599 So if you're thinking like meeting notes, people like to have a structured template for a meeting note, or if you wanted to create a template for a profile, if you're big into networking and you'd really like to create a profile page for each person that you're interacting with, it will help create that first. 00:23:35.920 --> 00:23:42.079 And then based on what you connect to within ChatGPT, your plugins. 00:23:42.240 --> 00:23:51.839 So if you're connecting to email, to uh calendar, uh Slack, Teams, other connectors and plugins that would be relevant. 00:23:51.920 --> 00:24:07.599 Link if you're scraping LinkedIn, you could take all that information and with scheduled tasks that will run automatically, ChatGPT can basically glean the information of any interaction with that person based off of their name. 00:24:07.759 --> 00:24:17.759 It will write that into Obsidian and it will keep an updated profile persona for each of those individuals. 00:24:17.839 --> 00:24:47.759 And so the way that I'm I'm doing this is any interaction that I have with an individual generally that I want to save is uh ChatGPT automatically creates that profile persona for that person, and then anytime it sees that name in any interaction, in emails and Slack and uh calendar engagements, um, it will take that information and it will write updates to that person's profile page. 00:24:47.920 --> 00:25:06.319 So now it's as if I have a virtual assistant sitting next to me that's following around in all engagements that pertain to the connectors and plugins that I have, and saying, Oh, I interacted with Russell for a podcast today, and this was written on this date. 00:25:06.480 --> 00:25:13.759 We we had this context based on the connectors and the plugins information it has access to, and it documents that. 00:25:13.839 --> 00:25:23.200 So it has a living history of interaction of what I'm providing it, and uh the way I have it set up now, I'm completely hands-off. 00:25:23.440 --> 00:25:25.119 So I don't touch it. 00:25:25.200 --> 00:25:38.720 Um, I should use it better than I do, but anytime I want to then go and fetch that information, I can ask ChatGPT, hey, give me the information that you have on Russell, and it will give me all that that has been written. 00:25:38.960 --> 00:25:39.839 Oh, that's amazing. 00:25:40.160 --> 00:25:52.880 So if a person's looking to get stuff, maybe they've asked around a little bit with chat or whatever, but but they're really looking to try to get into the more foundational elements with agents and trying to get things going. 00:25:52.960 --> 00:25:54.240 What would your advice be? 00:25:54.400 --> 00:25:59.839 Because things are moving so fast, the stuff that wasn't possible like three months ago is now possible. 00:25:59.920 --> 00:26:03.119 So, how do you how do you get on the that bandwagon, right? 00:26:03.200 --> 00:26:04.480 What would your advice be? 00:26:05.119 --> 00:26:11.680 I would say um I I work with uh OpenAI's ChatGPT quite a bit. 00:26:11.839 --> 00:26:15.519 I've got Claude as well, but for now I I like to use ChatGPT. 00:26:15.759 --> 00:26:19.680 I recommend their uh Open AI Academy website. 00:26:19.839 --> 00:26:23.839 Um, I know usually vendor training is not great. 00:26:24.000 --> 00:26:30.960 Um, their training is is pretty good and very relevant to non-technical uh folks. 00:26:31.119 --> 00:26:38.400 You'll hear kind of the term knowledge workers is uh is a broad term now in a lot of the AI space. 00:26:38.640 --> 00:26:46.000 So finding examples of how knowledge workers are utilizing the tools is very helpful. 00:26:46.160 --> 00:26:58.160 Um I always lean back kind of as of today, gauge your uh your utilization of ChatGPT or Claude based on how you what you're wanting back from it. 00:26:58.240 --> 00:27:04.319 So if you're just wanting to ask a question, use the chat function within each of those tools. 00:27:04.559 --> 00:27:28.960 If you're wanting a type of engagement that you're creating a document, or you're going to have multiple follow-up questions, or it's something that you want to last longer than 10 minutes, create it as a project or utilizing the the work or the co-work tools itself because it by design will keep that memory. 00:27:29.039 --> 00:27:32.640 It'll allow you to make use of those tools and those plugins. 00:27:33.279 --> 00:27:35.920 And uh if you want to come back to it, it's there. 00:27:36.000 --> 00:28:00.799 So that's where I would say is start with those coworker work tools, go through the plugin library, look at what is relevant to you personally, if it's Gmail, if it's Outlook, if it's your calendars, if it's other connectors uh that you have, just look at the library of plugins, see what's useful and bring those in, and then just start asking it questions. 00:28:01.680 --> 00:28:02.240 Cool. 00:28:02.559 --> 00:28:13.519 In terms of if you had to give um some advice to your 16-year-old self around where we are at the moment with AI and agents, what would your advice be? 00:28:17.680 --> 00:28:23.279 I would start with saying um AI is here to stay. 00:28:23.359 --> 00:28:29.839 It it is going to be a major, probably the main component of the rest of your life. 00:28:31.599 --> 00:28:34.079 Learn how to utilize it. 00:28:34.240 --> 00:28:40.240 You do not have to be a software engineer, you don't have to be a a data architect. 00:28:41.200 --> 00:28:46.880 Just start using uh the tools itself in the right way. 00:28:47.279 --> 00:29:01.440 Um you you still need to learn, you you need to read, you need to experience but don't use the tools to just cheat your way through. 00:29:01.599 --> 00:29:05.200 Um but learn how you can make use of it. 00:29:05.519 --> 00:29:19.200 The biggest advantage

that these tools are bringing is it makes you as a human exponentially stronger in speed, in quality, uh, and opportunity. 00:29:19.440 --> 00:29:48.160 And I think learning how to use it, same as learning how to ride a bike or learning how to drive, this is going to be a 100% life skill that you have to know in any industry because and I would say as a 16-year-old, you have an incredible advantage of time to learn, especially before you enter the work world. 00:29:48.400 --> 00:29:56.400 So make use of that time, use it the right way, and then make yourself extremely valuable once you hit the job market. 00:29:56.640 --> 00:30:01.119 Yeah, and it's not one of those things where there's a huge amount of experience that other people will have. 00:30:01.200 --> 00:30:02.559 You know, this is moving so fast. 00:30:02.720 --> 00:30:09.519 If you're just starting off, you're basically on the same step as everyone else because it's it's all moving at breakneck speed. 00:30:09.599 --> 00:30:13.039 So you just got to get started and and and see where it goes. 00:30:13.200 --> 00:30:18.559 You know, I'm I'm gonna finish with uh with an Elon Musk quote that you've mentioned before. 00:30:18.799 --> 00:30:22.480 The first step is to, well, it's attributed to Elon Musk, let's put it that way. 00:30:22.640 --> 00:30:27.599 The first step is to establish that something is possible, then probability will occur. 00:30:27.839 --> 00:30:31.599 And that feels like from an AI perspective, that's where we are at the moment. 00:30:31.680 --> 00:30:38.079 It's really about establishing that something is actually possible, and then there's a greater chance that it's going to be happening in the future. 00:30:38.400 --> 00:30:41.920 Yeah, I think we've talked about this kind of offline. 00:30:43.759 --> 00:31:09.279 These tools, uh, and even just within this year of 2026, these tools allow individuals to do incredible things that they never either would have had the time to do, they didn't have the knowledge or experience to do, or it it's something that takes so much time for a human to do that you would have just put it off. 00:31:09.359 --> 00:31:10.640 You never would have gotten to it. 00:31:10.720 --> 00:31:38.799 So I think that's the greatest thing I see with these capabilities is it gives you the opportunity of time, it gives you the opportunity of effort, and it's it's exponentially, again, kind of extending you as a singular person to do what you want with the ability to have things completed that you may not have learned or studied for. 00:31:39.519 --> 00:31:50.960 Um and it really opens up the door to solve really any problem uh digitally now that you wouldn't have been able to do a year ago, two years ago. 00:31:51.200 --> 00:31:51.599 For sure. 00:31:51.680 --> 00:31:54.319 You know, Garrett, it's been really inspiring. 00:31:54.400 --> 00:32:01.680 You know, you've definitely uh been a beacon for me going forward and helping me trying to use these tools more and more. 00:32:01.839 --> 00:32:03.359 So thank you very much for your time. 00:32:03.440 --> 00:32:04.720 That was that was really good fun. 00:32:04.799 --> 00:32:05.119 Thank you. 00:32:05.200 --> 00:32:06.880 Yep, thanks for having me.