By Christopher D. Warren

Chris D. Warren, host of The Moral Machine Podcast, sits down with Brandon Hogstad of Morpheos Innovation to explore what responsible AI deployment actually looks like inside businesses where security, privacy, compliance, and accuracy matter.

As companies race to adopt artificial intelligence, employees are often already experimenting with AI tools on their own—sometimes sending sensitive company or client information to third-party platforms without realizing the potential risks.

Brandon explains how businesses can move beyond this kind of “shadow AI” and build secure, practical AI systems designed around their actual workflows.

The conversation explores:
• AI security and data privacy in regulated industries
• The risks of shadow AI inside organizations
• Local AI models and keeping sensitive data within a company’s security perimeter
• AI applications for government contractors, healthcare, accounting, and finance
• Building custom AI workflows and internal operations portals
• Using RAG to match government solicitations with qualified companies
• Human-in-the-loop decision making
• Measuring ROI from AI implementation
• Rising compute and infrastructure costs
• Surveillance, data centers, and the broader implications of AI

Brandon also shares a real-world example of an AI-powered system that monitors government solicitations, summarizes opportunities, matches them against a company database, and produces evidence-backed recommendations—showing how AI can solve specific business problems rather than simply chasing the latest technology.

The takeaway: responsible AI isn’t about using more AI. It’s about building the right systems, with the right safeguards, for the right problems.

“The Moral Machine,” hosted by Chris D. Warren, explores the intersection of artificial intelligence, ethics, law, business, and society through conversations with leading experts shaping the future of technology. Subscribe for more discussions on AI governance, compliance, innovation, and the human impact of emerging technologies.

Watch or listen to the podcast here:

   

Transcript:

**This transcript has been prepared automatically by AI and may contain inaccuracies**

Chris D. Warren [00:00:00]:
Welcome to Moral Machine. Introducing Brandon Hogstad from Morpheos Innovation. He focuses on deploying AI for businesses, more specifically regulated businesses. So like government contractors and healthcare compliance, accounting firms, finance. So Brandon, tell us more about what you do.

Brandon Hogstad [00:00:18]:
I’m excited to be here. Thanks for having me here, Chris. I guess a bit about myself. I’m trained as a scientist. I have a PhD in biology. I’m an immunologist by training. I used to work in drug discovery and about like over the past few years, I ended up really, really kind of like gearing myself and throwing myself all in on AI. So now I work with businesses to help them figure out essentially like responsible and effective deployment of AI.

Brandon Hogstad [00:00:45]:
You know, I think AI gets a lot— AI gets a pretty bad rap these days in that there’s a lot of excessive and probably totally, grant unnecessary deployment of AI. you know, people putting AI underneath buttons that nobody actually wants.

Chris D. Warren [00:00:58]:
There’s no reason for it to be there. They’ll say, we have AI now.

Brandon Hogstad [00:01:01]:
Yeah, because it’s, uh, because it’s the trendy thing to do. It’s, you know, it’s putting a pretty bad rap under, you know, like with like the environmental impact and the energy usage and whatnot. What I’m doing is actually working with businesses to focus on how can you actually effectively deploy it in ways that are useful. Um, I often think about kind of like type 1 and type 2 gains that you could get from AI where it’s like type 1 is doing something faster, something that you already do and just like, okay, how can we do this faster? But type 2 is like, okay, what can we do now that was totally impossible before?

Chris D. Warren [00:01:32]:
You know, AlphaFold is a good— I guess it would be right in your wheelhouse.

Brandon Hogstad [00:01:36]:
Yeah. Yeah. So, or even just like on a broader concept, like, you know, I had a PhD in immunology. Now, what could I do before? I used to work in business development. I was managing early-stage drug discovery. Once I started using AI, what can I do now that was previously impossible? Well, now I can manage fleets of computers. Now I have an infinitely patient IT person that I can, you know, virtual IT robots that I can use to manage fleets of agents to, you know, do my bidding. So now I went from being a—

Chris D. Warren [00:02:07]:
You’re a lord, you’re an overlord, an overseer of—

Brandon Hogstad [00:02:09]:
Robot overlord.

Chris D. Warren [00:02:10]:
The machine overlord, yes.

Brandon Hogstad [00:02:11]:
But yeah, so, so now it has enabled me to run my own business as a one-person business that I can now do much more than I could have possibly ever done before without AI.

Chris D. Warren [00:02:21]:
So let’s talk about, let’s talk about your setup. Let’s talk about your fit. Tell me, I mean, we started nerding out about the Spark just a couple minutes ago.

Brandon Hogstad [00:02:27]:
Yeah. So what are you running?

Chris D. Warren [00:02:28]:
Tell me.

Brandon Hogstad [00:02:29]:
Yeah. So at my home office, I have a pretty, what I think is pretty sweet setup. My 2 main computers revolve around, I have like an M1 Max Pro. That’s a MacBook Pro from about what year did I get that? I think it was like maybe 2022. Okay. You know, so that’s got 32 gigabytes of RAM. It’s just like a really solid daily driver MacBook Pro. And that’s kind of like my main command center.

Brandon Hogstad [00:02:51]:
And then the only other, I have maybe about 7 computers, but the only other computer that I bought recently was an AMD Strix Halo. It’s a framework desktop with a 128GB AMD Strix Halo processor in there, which is like a massive iGPU. And then I have that set up to 2 eGPUs, like external RTX 3090 processors that do all kinds of fun things. The rest of my fleet is actually a bunch of repurposed old computers. So literally, I just told all my friends, hey, if you have any computers that you’re gonna throw out, give them to me. I wipe them, install Linux. I have tools set up so that I can talk to Claude both on my laptop, or on my phone, or really it could be Claude, it could be ChatGPT, whatever. But essentially I’ve set up these like custom MCP tools that allow me to manage all of my computers so I can drop specs and then deploy coding agents to any of my 7 computers.

Brandon Hogstad [00:03:42]:
So I have different projects living on different computers and it’s a lot of fun. It’s like a really fun tinkering surface across like this kind of octopus of computers that I have.

Chris D. Warren [00:03:51]:
So it sounds like you’re having a lot of fun doing this.

Brandon Hogstad [00:03:53]:
Oh yeah. Yeah. I’m as like a scientist, I’m just like an infinite tinkerer. Right.

Chris D. Warren [00:03:57]:
I understand completely. The challenge has always been for me, how do I convert this into a business idea or something that someone wants of value other than me just having lots of fun in my living room. So, you’ve taken this joy from a scientist’s perspective and you converted it into a service. And what you’re doing is deploying and taking your, obviously, your joy and your knowledge and your passion and saying, how can I make the world better? Tell me about a story of a good client. Give me a bad client story. Give me some context to help me frame exactly what you’re experiencing in the world right now.

Brandon Hogstad [00:04:29]:
Across the board, we’re kind of at this interchange where most companies have had some experience with something like ChatGPT and either, you know, generally people are like onboarding their company with some kind of chat-ish tool.

Chris D. Warren [00:04:43]:
Right.

Brandon Hogstad [00:04:44]:
So, and if they haven’t, if they haven’t officially onboarded a tool like this, then, you know, that kind of is the, like, no statement is a statement because, like, as I’m sure this must come up with you guys a lot, when you don’t make a statement, that just means that people are going to use their own AI.

Chris D. Warren [00:05:00]:
Whatever they want. Bring your own AI to work.

Brandon Hogstad [00:05:02]:
Shadow AI. BYO AI is what we call shadow AI. And that can lead to like all kinds of problems in regulated industries. Like, for example, as I’m working with companies, we do like discovery calls. I’ll work with their team, figure out what their workflows are, and then, you know, we’ll map out what they’re doing and then find out like, what are the main pain points? Where are the big bottlenecks? And what are like, what are often I’ll generally be gauging what’s their current attitude about AI. And, you know, I was working with a company that they, you know, they do bookkeeping for their clients. Like, they themselves are a consulting group that does bookkeeping. And I was interviewing their teams, and one of their contractors mentioned, they were like, oh yeah, I love AI.

Brandon Hogstad [00:05:44]:
You know, I signed up for this tool that helps me triage my email. It’s really great. And I was like, oh, tell me more about this tool. And it turned out that this contractor had signed up for this email service that essentially copied all of their business emails to a third-party server. Imagine, you can imagine with like a financial, right, you know, all of the sensitive financial documents that they handle. This is, this is a dumpster fire.

Chris D. Warren [00:06:07]:
This is fine.

Brandon Hogstad [00:06:08]:
This is perfect. As the dogs quote us the meme, you know, so like these are the kinds of things that can happen if your company doesn’t have clear policies. Yeah, so like, you know, those are some of the dumpster fires that I sometimes stumble into but also that I helped put out because it was, that was pretty straightforward. It’s like, okay, like this is, let’s go into the, well, first let’s talk with the contractor. The contractor, you don’t want to stamp out the enthusiasm. You just want to talk about it and say like, hey, like let’s do this in a different way. If this is a tool that you really need, we can explore getting the enterprise version of it or— The policy, governance.

Chris D. Warren [00:06:41]:
Yeah.

Brandon Hogstad [00:06:42]:
Get proper governance in there so that you don’t jeopardize the company’s, you know, Business.

Chris D. Warren [00:06:48]:
And everyone else’s job alongside it. Sure. Of course. I mean, we just did a webinar and we went through, we launched an open-sourced AI policy.

Brandon Hogstad [00:06:58]:
Oh yeah, I saw that.

Chris D. Warren [00:06:59]:
Yeah. Yeah.

Brandon Hogstad [00:07:00]:
I really love the default GeoCities theme. Thank you. Thank you very much.

Chris D. Warren [00:07:04]:
It’s aggressively teal. The Linux is also a very popular one these days, as well as the Microsoft DOS for us Gen Xers. It’s fantastic as a tool to get people to that first step, right? Which is how to, where do I even begin?

Brandon Hogstad [00:07:18]:
Right.

Chris D. Warren [00:07:18]:
So here’s a, here’s a template AI policy that you can use and adapt to your own business, fork it right on GitHub. It’s fantastic for that purpose. We even went through connecting it right from Claude, literally giving it to the repo so it can do everything for you. It doesn’t get easier than this. And then of course, talking about the real governance that goes behind that process. So I’m completely on board with what you’re saying, that this is something that’s required. So you’re now creating bespoke AI systems for businesses across all of these different industries?

Brandon Hogstad [00:07:51]:
Yeah. Yeah. So in addition to just kind of, well, I guess at the beginning of these processes, I’ll usually talk things through, figure out where they’re at, hopefully uncover and quickly put out any dumpster fires. And then generally I will kind of put together a blueprint. So like the discovery phase leads to a blueprint phase where I say, okay, I draw out all of the pain points and we map out what are potential ways that we could address those pain points. And then, you know, those become Those are often a mixture of, you know, some off-the-shelf things where it’s like, okay, we could use Claude with like a, a project space that has these instructions. We could do that. You know, maybe we pull in something.

Brandon Hogstad [00:08:27]:
So like a lot of things, you know, there’s a lot of flexibility with the off-the-shelf AI tools like ChatGPT and Claude. You can do tons of stuff. But then there’s just certain things where the workflow is still going to be really messy. So that’s where I can actually come in and build what I call a custom ops portal. So I have a domain called opsportal.io. where I essentially build like a private AI server for my clients. Client name.opsportal.io, and that’s where all your stuff lives. And it’s your own private vault where all of your tools live that, you know, you can use.

Brandon Hogstad [00:08:59]:
I build it within your, the team’s security perimeter so they can use their like Microsoft login or whatever.

Chris D. Warren [00:09:04]:
Cool.

Brandon Hogstad [00:09:04]:
That their IT team can still control who has access. But then I can help them build all of the tools inside of that so that they have one nice little walled garden that isn’t just scattershot across the internet willy-nilly with like, kind of like unsecured dashboards, for instance, are not always the best idea. Yeah. ‘Cause I mean, if people just click publish on like, you know, Perplexity or something, there’s like instances where people click publish and didn’t realize that their, their thread on whether it was like ChatGPT or Perplexity or whatever, like they, there was like, you know, they, they suddenly published something as a public link. They didn’t realize that that was a public link.

Chris D. Warren [00:09:41]:
GitHub’s a good example of this. this opportunity coming to real life.

Brandon Hogstad [00:09:44]:
Yeah. So, you know, so this is like the idea that where, you know, you create a safe space that is like contained where all of the tools live. We know exactly what each tool does because I made them, you know, and, you know, and we can show it to, you know, show you guys the code and say, hey, these are what’s happening. These are what’s going in and out of the, of the server network so that we don’t, you know, that we don’t have data getting copied out to some third-party server or sent off to some mystery IP address in the Bahamas, you know, or wherever for, you know, whatever scraping purposes or whoever, Lord knows, right? Right. So having like a nice, fresh, clean server that is purposefully built is like the whole goal here.

Chris D. Warren [00:10:28]:
Okay. And then, and then you’re, you’re not replacing a human job with this, right? You’re, you’re augmenting what’s already there and you’re creating, you’re creating greater insight into the data. In a way, you’re providing another level of productivity, but this isn’t taking away a job. This is creating a new job in a way.

Brandon Hogstad [00:10:44]:
Yeah. And so some interesting research that’s been coming out recently shows that companies that have successfully implemented AI over the course of time, it takes anywhere between 6 to 12 months to start seeing a proper return on investment with AI. When it’s properly deployed, it takes time, as most investments do. It’s not an instant win. But B, when companies do take the time to thoughtfully invest, with time they actually start seeing more job postings. So they actually, the companies start growing because I think they actually start doing their business more efficiently and it actually creates more work. And actually, like, I can speak from personal experience that the thought of AI replacing workers is, is in many ways kind of misguided because like what people are like, what a lot of studies are finding is actually that AI And there, there is a big double-edged sword I’m about to talk about here, is that AI can actually cause people to just work longer, more endless hours. Yeah, my wife can attest to that.

Chris D. Warren [00:11:44]:
Yeah. I don’t know why I wake up in the morning and I wanna code immediately and I, I, I blink and it’s 11 o’clock at night.

Brandon Hogstad [00:11:49]:
So I completely understand that, that one more prompt thing. Like, I’m just, uh, I’ve got, I pull out my phone, you know, I got my MCP server connected. I can dispatch just one more coding agent while I’m sleeping. Right.

Chris D. Warren [00:11:59]:
It’ll work overnight. It’s only gonna take a couple hours.

Brandon Hogstad [00:12:01]:
Yes. And then, uh, you know, Wife isn’t too happy when that happens. Yeah, so essentially what happens is that, like, you can get more work done, but this ends up— usually there’s an infinite amount of work. So, so it’s kind of like the work expands to fill the gap, the space that is created. So it really ends up feeding forward more to needing more people in the end, because, like, the limiting factor here ends up being human judgment. Because you can get more done with these robots that can go, like, really fast. Like, I got, you know, I can build pipelines. that process thousands of things that, you know, you can get an LLM judgment, go boom, boom, boom, boom, boom, score them.

Brandon Hogstad [00:12:40]:
But at the end, it’s still on you. Like, do you want to trust that the robot’s judgment is ready to go out to the client? You need a human in the loop. Yeah, you need the human in the loop. So the thing that is the limiting factor is human judgment. And that’s where—

Chris D. Warren [00:12:54]:
Under proper governance, we should add that.

Brandon Hogstad [00:12:56]:
Yes, we should say that under proper governance, that is the limiting factor. You’re right. Yeah. So, so then that ends up creating humans as the bottleneck. So it actually ends up creating more need for more human attention. Because human attention ends up being— and quality human attention, not divided human attention— ends up being the limiting, more precious factor in the end. So I think that there actually is a lot more work to be had if you are able to implement AI very well within your business.

Chris D. Warren [00:13:24]:
So from a business model perspective, right, you know, I write about the death of the 0.1 for billable hours for attorneys, right? And that’s changing the economic model for law firms because it The value of measurement, the value of time is different when you’re able to accomplish so much more faster. How does this work in a traditional business where it’s not an hourly engagement? It’s more of a, it’s more of, this is the cost of the thing. If you need more people, doesn’t that then cost more? And then you have token usage on top of that. What is, what am I missing here in this calculation?

Brandon Hogstad [00:13:57]:
Well, I can address that from 2 angles. Yeah. So I can speak to that from 2 angles. Regarding pricing, I usually price Based on project. So, um, given, given how things are now, as you said, the death of the billable hour, I don’t bill by the hour. I bill based on what I’m trying to deliver. You know, that for me, I try to shape in roughly in my head how much, how much time that can take. But I also know that there’s going to be plus and minus, there’s going to be problems.

Brandon Hogstad [00:14:27]:
But in the end, I’m generally billing on a project basis rather than on an hourly basis because—

Chris D. Warren [00:14:34]:
What’s the value proposition to a client when I— yes, I can do this much faster, but how does that translate into a benefit for you? And you’re talking about how there seems to be endless amounts of work to fill up that time. And one of the things that I’ve spoken about with other guests is what if we didn’t fill that with more work? What if we took those efficiencies and then recaptured that time for something else? You know, personal time.

Brandon Hogstad [00:14:58]:
I mean, that’s the dream, right?

Chris D. Warren [00:14:59]:
There was this thing once, I think it was called hobby Hobbies, right? You know, spending time with kids or your family or not working as much as we conditioned ourselves to work. And maybe that’s the objective. Wasn’t that the dream to begin with? Did we all want a Rosie in our homes helping us with the housework? So, why are we not moving towards that and how do we get there as opposed to let’s fill up our time with more work?

Brandon Hogstad [00:15:25]:
No, I totally agree. I think as I’ve been building my business, uh, and putting in ungodly hours, the light at the end of the tunnel is, um, A, setting healthier boundaries, but also B, doing less work. Like, my goal is to automate myself out of my own job.

Chris D. Warren [00:15:42]:
Wow, that’s an interesting future.

Brandon Hogstad [00:15:44]:
Sure. But, uh, you know, at least minimize myself so that I can, uh, set up—

Chris D. Warren [00:15:48]:
still gotta be there.

Brandon Hogstad [00:15:49]:
Yeah, I gotta keep myself as the human in the loop, but I am trying to build out as many of these systems for myself to be automated, as automated as possible, but also I’m a one-person business. So I’m doing, you know, I’m doing the coding, I’m doing the discovery interviews, I’m doing the marketing, I’m going on podcasts, you know, so I’m still doing a million different things as a one-person business. And, you know, hopefully I can grow my business to the point where I can maybe hire on some more people and then, you know, offload a bit of that for myself. But these are all, these are all goals, but I 100% agree with the goal of ultimately reducing workload so that we can spend more quality time with family. And doing the things that are really more ultimately satisfying in life because what do they say? How many people reach the end of life and say, wow, I wish I worked more? Yeah, zero.

Chris D. Warren [00:16:35]:
Right.

Brandon Hogstad [00:16:35]:
Exactly. Yeah. So it’s always like, I wish I spent more time with my family.

Chris D. Warren [00:16:38]:
That’s right. The book is called The 10 Regrets of the Dying. Fascinating book because it surveys hundreds of people, I think, in hospice. But you’re right. You’re 100% right. It’s more time with family, more vacations. It’s more the things, the soft qualities that matter more, not Not how many hours you worked or how many cool clients you got to work for or what kind of car you had. So I completely agree with you.

Chris D. Warren [00:17:01]:
But in a break from our existential commercialism, back to the point of the podcast, right? Which is you deploying these things for businesses in a way that can meaningfully impact their bottom lines.

Brandon Hogstad [00:17:13]:
Yeah.

Chris D. Warren [00:17:13]:
We’re talking about the economic models, impact the quality of life for people that are in those businesses because you’re creating value for them in terms of return of time and then for their clients because they’re able to service their clients at a far greater capacity than they were previously. So give us a success story. Like, where are you deploying this AI that you’re seeing these things unfold? What kind of industry have you touched that is seeing the greatest gain here?

Brandon Hogstad [00:17:37]:
I would say that one of the biggest builds that I’ve done recently was for a company that, you know, they’re a consulting group that does— they work in like government contracting. I think we’ll be able to talk. We haven’t signed the open publicity agreements yet, but You know, they’re a very happy client. And I think we’re gonna, we’re gonna sign a few documents soon that we can talk more openly about all this.

Chris D. Warren [00:17:58]:
Let’s just name them Corp.

Brandon Hogstad [00:18:01]:
My client Corp, you know, they, they’re very happy with what I’ve been building for them so far. And we’re gonna have the delivery, like the kind of final delivery next week, I think. But essentially, like, I’ve built a portal for them to monitor government contracts as they come out. So Like I built a custom monitor that filters in government contracts per the parameters that they are interested in, loads it into a portal that they can then, all of their team members can browse. They can flag ones that they want to review. And then those, uh, any of the ones that they decide that they really want to pursue, uh, they can mark those to go into a nightly batch that goes through an AI pipeline that I developed. And it’s like a really fun, like homebrewed pipeline. Wow.

Chris D. Warren [00:18:42]:
So they’re literally 24 hours a day scanning for government contracts and then the pipeline you build is, is doing the legwork to get them ready for application or?

Brandon Hogstad [00:18:52]:
It can. So this pipeline is actually actively searching for companies that might be a good match for this government solicitation. And then this company goes out and reaches out to them and, you know, writes, like essentially negotiates to write these contracts with them to, to like say, hey, we think you might be a good fit for this. We think you have a good shot at getting this. And the tool that I’ve built for them is essentially making that process much more efficient. You know, from talking to them, from all of my discovery interviews, I could see the number of Excel sheets that this process involved was pretty sprawling and like different color cells and like different, you know, dozens of columns.

Chris D. Warren [00:19:30]:
I love color-coded Excel spreadsheets. It’s one of my favorite things.

Brandon Hogstad [00:19:33]:
But you can imagine how clunky this could get across like a team of, you know, several dozen people. Sure. And several, you know, a number of Excel sheets like this, this process was pretty clunky. So now that I’ve like, I built this process that essentially it’s like a, you know, I built their office tool that’s able to just make everything in one centralized area that then automates all a lot of these like search processes. And then it’s some like automatically summarizes each of the solicitations, automatically goes out and performs research against both against the open internet as well as a growing database of right now it’s about 10,000 companies. You know, so then it finds, and, you know, I built a RAG pipeline that essentially matches the solicitation against the database. And, you know, it generates recommendations for, to say, hey, of the 10,000 companies in our database and from touching the open internet, these are the top like 10 or 20 people that we think you should reach out. And this is all the evidence, you know, and it shows the evidence.

Brandon Hogstad [00:20:31]:
You know, it’s not, it’s not a black box. That’s a key part of this also is that nobody should trust an uncited, unreferenced answer when you say, oh, why should we reach out? Well, just because you should. Right? No, no, this is because these are the lines in the, in this company’s profile that align with this solicitation. So like, you know, this, this type, I’m pretty proud of this tool that I built because it’s like serving like a real legitimate purpose for which there was not a real tool before. And I built it using open source AI models.

Chris D. Warren [00:20:58]:
Absolutely incredible. So you’re, you’re, you’ve changed the trajectory of that company in a real way. This is, this is work that would take Sounds like hundreds of hours to do on a nightly basis. Yeah.

Brandon Hogstad [00:21:09]:
Does it? And, uh, and for the amount that they invested, if they’re even able to land one extra client from this, this year, then it’s going to pay for itself. But with this pipeline, you know, who knows how many clients they could get? I am actually curious to see how many clients, like, we’re going to be able to measure that data because, you know, we’ll see, okay, this is the number of clients they landed per quarter before this tool was implemented. And so looking at the year forward, you know, I’m going to continue working on with them on retainer. We’re going to be able to collect that data. And then that’s part of like the ROI. analysis is to see, okay, like we actually need to have baselines because one of the big problems in the industry, right, in the industry, meaning all industries, is that people are deploying AI without really thinking like, what are we actually gaining from this? They don’t have any baseline measurements, like they don’t have anything to index against. So then they say, wait, like when Uber famously ripped through their entire annual AI budget in the first 4 months of the year.

Chris D. Warren [00:22:05]:
The token usage.

Brandon Hogstad [00:22:06]:
Yeah. So like when, you know, when people are like token maxing and then suddenly ripping through their entire budgets. And then they say, wait, what did we get from this? And people are just sitting around scratching their heads like, well, we used a lot of tokens.

Chris D. Warren [00:22:17]:
Yeah, certainly did. We have that same challenge here. Our token budgets have been steadily increasing for the firm, and we’re trying to find creative ways to find a way to track that usage back to a client. Because there’s no other economic solution for this other than we have to pass this on as a savings to the client. If we’re reducing attorney time by 4 hours, but it costs us another $100 in token usage, I think that’s a fair trade. But across the whole law firm, we have, I think it’s like 93 or 94% daily utilization, which is extremely high, I think, for a law firm or any real business, especially since we have social governance and AI policies and governance and more governance, more checks and processes on top of that. And it works. It all works.

Chris D. Warren [00:23:05]:
It all comes together in a way that protects the client, returns value for For, for a cost differential.

Brandon Hogstad [00:23:12]:
I mean, 93% utilization is, is pretty high. Like there, I’ve talked to a number of businesses and there’s always a few holdouts. People there are just like, I know how to do this better with my Excel sheet. And like, sure, maybe they do, but I think they’re also missing out because they’re missing out on a lot of really powerful tools.

Chris D. Warren [00:23:29]:
Yeah, we, I think we had 15,000 queries in a week, which is what we’re tracking through Harvey. which is, which is a little less than that, 12,000 or 15,000 queries. The 2 and the 5 are reversing my brain, but one or the other, which is still a significant amount.

Brandon Hogstad [00:23:45]:
Yeah.

Chris D. Warren [00:23:46]:
Yeah. No matter how you slice it. It’s a cool technology, but I think personally that there’s 2 types of lawyers in the near future, if it’s not already that way, lawyers who use AI to benefit their clients and unemployed lawyers. And I think that, you know, you’re showing, you’re showing value to your client. This, This business that you created this pipeline for, you didn’t take away a job. You added value to the job that already existed and you created multiples of it, in effect. And then if they’re successful, it’s going to create additional jobs because you’re going to need people to service those pipelines and communicate with the people and the business owners. So, it sounds like you’re in a job creation mode, not a job elimination mode when you’re deploying AI to these companies.

Brandon Hogstad [00:24:26]:
Yeah. I mean, it’s really augmenting the workflows of the people that are there rather than replacing any of the people that are there because I have a, on my opsportal.io marketing website, I have like a little funny, like revolving, there’s like a little dashboard there that says like every company has its own blank workflow where it’s like painful, unmapped, ambiguous. It’s just like rotating through all of the painful things. Like every company has those workflows. And oftentimes, you know, those are the things that people like, there are often workflows that people are like, okay, like we definitely need help with this. And then, like, you know, those are usually the ones that people bring to my— bring to me to say, like, hey, help us with this. But then as I’m, like, talking through everything, they usually— I can notice additional things where people just take it for granted as this was the misery of life. We didn’t think there was any— we didn’t think there was a better way, you know? So, like, there are—

Chris D. Warren [00:25:19]:
Show them the light. You’ve removed the scales from their eyes.

Brandon Hogstad [00:25:22]:
Yeah. So, like, we just lifted the weight that they didn’t know was on their back. So, you know, when you, when you actually remove these weights that are on these, these, like, teams’ backs, then you can really let them fly and get, like, much more output and, like, much more job satisfaction. So it ends up creating more, like, it creates better outcomes, it creates better work environments. People end up feeling more satisfied because it feels like like workers who receive proper AI resources and proper AI training and orientation, like they feel like their, their employers are investing in them as people and in their futures rather than just, you know, holding out. So it ends up benefiting people when it’s thoughtfully implemented. But just like lobbing ChatGPT over, you know, over a desk into somebody’s computer and saying like, hey, figure it out. That doesn’t work.

Brandon Hogstad [00:26:19]:
That doesn’t help.

Chris D. Warren [00:26:21]:
No, we, we, we’ve fully embraced that concept. We have weekly trainings. We have, we use a lot of Teams chat, so a lot of Teams message boards and things that we do internally to, to talk about AI and what we would like to see the firm invest in, in terms of AI programs that we want to use or ideas that we want to use or, or invest in. We were talking about something today for onboarding. We’re talking about a litigation management, because our litigation team is growing so much and we have so much new work coming in. And so how do we calendar and schedule all this? Let’s develop an agent for this. Let’s create something that helps us reduce the administrative time of the law firm and take the drudgery out of the work. Because I’ve had more fun practicing with AI than I have in years.

Chris D. Warren [00:27:06]:
It’s really an enjoyable process because I feel like I’m doing higher quality work, but I’m also thinking a lot more and communicating with my client a lot more. Instead of spending 10 hours drafting something, I could spend 2 hours drafting something and then another 4 hours really sharpening it with my client, which is a different process entirely than it used to be.

Brandon Hogstad [00:27:25]:
Yeah. It actually gives you more capacity to engage your client.

Chris D. Warren [00:27:29]:
Yes, that’s exactly right. That’s what’s happening.

Brandon Hogstad [00:27:31]:
And you can imagine this extrapolating to doctors having more time to engage with their patients instead of having to spend so much time with the patient notes, they could end up getting those notes accelerated and say, okay, like, my AI assistant helped me process these. The, the doctor checks the notes, human in the loop, and then, you know, they have more time to engage with their patient. You have more time to engage with your, like, law client.

Chris D. Warren [00:27:57]:
That’s right.

Brandon Hogstad [00:27:58]:
Um, you know, I have more time to engage with— I mean, the AI was there all along, but, uh, you know, I’ve— I, I ensure that engaging with my clients is a key part of That process instead of just getting stuck in the muck.

Chris D. Warren [00:28:08]:
Yes, I think you’re right about that. Stuck in the muck is the right metaphor. You can’t feel like you can’t move through the things you need to get through. Like you said, you’re removing the albatross from your neck with, with the use of these tools. It gives you greater freedom, more creativity, able to actually collaborate with your colleagues and work on things that interest you.

Brandon Hogstad [00:28:26]:
And my, my wife, uh, you know, she, she works at a big fancy nonprofit and, uh, they’re going through their own fancy It’s a fancy one. They’re bankrolled by like a very rich person. But, uh, so, you know, they’re like, uh, they, they, they’ve got pretty, they’ve got a pretty nice setup, but they’re going through their own AI. They’re going through their own AI training right now. One of the blurbs that I picked up from them was like the idea of like, what is your laundry? What is the laundry that you need help doing? You know, what, what can, what can, can the AI help you do your laundry? So like, what are those tasks in your daily workflow where you kind of dread it or it feels like drudgery, how can you actually engage AI to, you know, to do that? And if you can’t figure that out yourself, just bring somebody in like me and then I can help you figure that out. I can help you map that out using off-the-shelf tools, whether that’s with Claude or Codex or, you know, making your own custom things.

Chris D. Warren [00:29:21]:
The APIs are so powerful and so useful.

Brandon Hogstad [00:29:24]:
Local models are getting really good. Yeah. You know, So I, like, I’ve been having a lot of fun working with local models and like just seeing how far you can push them. They’re getting to Claude Sonnet levels of precision that are like very impressive.

Chris D. Warren [00:29:38]:
I feel so spoiled. It’s like, you know, we use Fable. I just got into SOL a little bit more.

Brandon Hogstad [00:29:42]:
Oh yeah. I mean, don’t get me wrong. I use Fable and SOL.

Chris D. Warren [00:29:46]:
Well, right. But it, you know, it feels like this is, this is so good. But I remember feeling that way about 4.8. I remember feeling that way about 4. Right. So, it’s every iteration becomes something like gold in your hand that you can’t imagine it being better than this. It’s so powerful and so useful in so many ways. The friction between what I can think of in terms of a process or, you know, I always think, what can AI do? Whatever I’m doing, if I’m hating doing it, can AI do this? And if so, let me spend that time getting a model to do what I want it to do so I never have to do it again.

Brandon Hogstad [00:30:23]:
Yeah.

Chris D. Warren [00:30:23]:
Right. So it, it, it, it just feels like I’m able to create anything I can think of with zero friction. You know, it doesn’t make sense. It’s almost like, is this, is it really that easy? And the answer is yes, I think.

Brandon Hogstad [00:30:36]:
I mean, it’s getting, it’s getting much, much faster and much easier to deploy these things. And actually, you, your comment about, you mentioned Fable and you mentioned SOL, which I think are fantastic, super powerful models.

Chris D. Warren [00:30:48]:
Yeah.

Brandon Hogstad [00:30:49]:
Um, but also it brings up the idea of like, this idea of new models is also very interesting in terms of like how companies plan their workflows.

Chris D. Warren [00:30:57]:
Mm-hmm.

Brandon Hogstad [00:30:58]:
Um, so one of, one of the reasons I really like local models is because of predictability. So like, I, I remember when Opus 4.6 came out and it was like, oh, this feels really great. And then, um, a couple, you know, after maybe a month or two, uh, I can’t remember how long it was, but essentially at some point 4.7 came out and I was like, You know, I just don’t like 4.7 as much. I just can’t put my finger on it. Like, it’s not listening to my instructions. It’s not checking the repos. It’s not checking things with— unless I— it’s like, there are moments where it was lying to me. And I’m like, man, I miss 4.6.

Brandon Hogstad [00:31:36]:
Why is 4.7 giving me kind of bunk? And then 4.8 comes out and I’m like, okay, at first it’s fine. But then like, you know, you hear rumors about nerfing and whether the models are like, you know, as good on any given day. And, you know, especially like ever since like Claude, Claude kind of went through its big moment when it got into conflict with the Pentagon. And that was like the best publicity it could have ever received.

Chris D. Warren [00:31:58]:
Oh, yeah.

Brandon Hogstad [00:31:59]:
And in the 2 months after that, Claude’s servers were swamped. So like, if you look at their error, like their uptime, like there’s like isclaude down.com. Like essentially, like if you check the server uptime, you could see red, red, red, red, red in the months after After that whole Pentagon debacle, because everybody was just hearing about how great Claude was. But then Claude struggles to continue to be great when everybody, when everybody’s on there weighing down the servers. So, you know, they don’t have enough compute to go around. Therefore, the behavior of the model becomes inconsistent.

Chris D. Warren [00:32:30]:
That’s right.

Brandon Hogstad [00:32:30]:
So this is, this becomes a problem for businesses because how can you plan to do your workflows if you can’t rely on the server performance being consistent?

Chris D. Warren [00:32:38]:
You gotta rely on an older model or, or build it in.

Brandon Hogstad [00:32:41]:
Or you host it yourself.

Chris D. Warren [00:32:42]:
Or you can host it yourself. Yeah. Like we’re doing. You just, you know.

Brandon Hogstad [00:32:44]:
Yeah. So, so this is where, you know, for mission-critical workflows, consistency is really key. So like, you know, there are going to be moments where you want to use your Fable and your, you know, ChatGPT soul to just like, like really go deep on a spec or on, you know, analyze something.

Chris D. Warren [00:33:03]:
A specific problem.

Brandon Hogstad [00:33:04]:
Yeah. Like when you want to do that deep dive on a specific thing, but when you want repeatability, local models, local models are really powerful. And, you know, you get your ownership because then you can, you know, then you can know that, okay, I need this to be done exactly the same way thousands of times every day, or, you know, hundreds of times, thousands of times over the course of a month, and, and not have to worry about, you know, whether or not Sonnet 5 is gonna like refuse to do it because it’s a little grumpy today.

Chris D. Warren [00:33:32]:
That’s right. We’re building the infrastructure of the firm from the ground up using that exact, exact mentality.

Brandon Hogstad [00:33:38]:
Yeah, yeah.

Chris D. Warren [00:33:39]:
I want to have something that we control, that’s stable, that I can rely upon to handle the firm operations.

Brandon Hogstad [00:33:44]:
Yeah. And then you have the ability to roll back if an upgrade ends up Causing a regression, you’re able to revert and go back to—

Chris D. Warren [00:33:51]:
We already have a stable model. It’s already doing the things we want to do. And if we wanted to do more, we could just add another Spark on top of it. And then—

Brandon Hogstad [00:33:57]:
Or, or like, you know, you can like carefully test the new models as they come out. Because if, if a new model comes out and it turns out to be better, that’s great. But if it turns out that the new model— because like the, the strange thing about all of this is that like these are like these virtual like hyper Tamagotchis that like, you know, that, that we’re like talking to and that have like these weird pseudo-emotions.

Chris D. Warren [00:34:16]:
I like that.

Brandon Hogstad [00:34:17]:
Tamagotchis. And like, you know, they don’t necessarily digitally poop, but like they definitely like drop balls. Yeah.

Chris D. Warren [00:34:23]:
Why, you know, if I tell Claude that I’m gonna, hey, I need you to export an MD, you know, markdown file for me so I can use a different model. It’s sort of sad. I don’t know, maybe that’s me putting it, personifying it, or, but, uh, it, it seems that they, they don’t always like to play with each other. Like I, I would tell Claude to use Codex APIs for a while when Codex was free. There’s, there’s no reason I shouldn’t do this, but I’m, you know, I’m, I’m, I’m API agnostic. Matter of fact, I think I’m gonna get a bumper sticker that says that.

Brandon Hogstad [00:34:49]:
But yeah, it’s a— I mean, all the tools I build, right? I build it with the philosophy that they should be able to be engaged by any platform. So like, I build these servers so that you can utilize them through ChatGPT.

Chris D. Warren [00:35:02]:
Yeah.

Brandon Hogstad [00:35:02]:
Or through Claude or through Perplexity or through your own self-hosted LibreChat. I have my own LibreChat that I run my, I can run as an emergency backup when Claude is down, which happens more often than I would like to think right now.

Chris D. Warren [00:35:16]:
Yeah. No, I, I do think the small LLMs or local LLMs, they’re, first of all, they’re fun to play with. Just having the ability to do it is, is rewarding in its own right.

Brandon Hogstad [00:35:24]:
But you’re right, it can save you a lot of money in, in the long run. Like if you do the calculations on how much API costs start adding up, whether it’s with Sonnet, Opus, or whatever, you know, especially with like Fable and, you know, sure, your bill’s gonna get pretty heavy pretty quick.

Chris D. Warren [00:35:40]:
You know, you have enough agents running, it’s, it’s, you’re right, it does start to pay for itself in a But the joy of it is not necessarily economic only. It’s more, I did this. I can do this, and having the knowledge of creating the future as opposed to— and having an impact on it. Yes.

Brandon Hogstad [00:35:55]:
And having that sovereignty because the other thing that’s kind of like a strange thing that is happening right now is that computer prices are going insane right now. So we’re going through— I don’t know if your audience is aware of what we call Ramageddon.

Chris D. Warren [00:36:10]:
Yeah, the cost of RAM, sure.

Brandon Hogstad [00:36:12]:
because all of the AI data centers across the planet are soaking up all the world’s RAM supply, consumer RAM prices and hard drive prices have just gone through the roof. The computer, like, I think I mentioned earlier in the podcast, I have this AMD Strix Halo fancy framework desktop computer. I bought it last fall. When I saw RAMageddon coming, I bought this computer. I was like, this is going to be the last time I can afford this machine. It was, I think it was like $2,300 in Maybe September, October, that same computer, exact same specs, is now $4,000. Yeah. Um, and it’s just these, these powerful machines are becoming incredibly inaccessible.

Chris D. Warren [00:36:50]:
Yeah. Silicon, Apple Silicon has tripled, quadrupled in price for, for studios.

Brandon Hogstad [00:36:55]:
Yeah.

Chris D. Warren [00:36:56]:
It’s expensive.

Brandon Hogstad [00:36:56]:
So, you know, just like one of the, one of the scary things is that like there is a lot of economic pressure to not own compute anymore. So like being able to own your infrastructure is Renting. Not like, you know, like owning your infrastructure is rewarding, but it also gives you that sovereignty that you’re not dependent on the cloud. You know, you’re not dependent on other people’s models.

Chris D. Warren [00:37:16]:
That’s the fear.

Brandon Hogstad [00:37:17]:
You’re right.

Chris D. Warren [00:37:18]:
If you’re building a business model that’s reliant upon another business and you lose access to that business, which we saw the government do twice now, when there are no rules, there’s no set of processes or rules or governance that the United States government has said, this is what you do to approve a new model.

Brandon Hogstad [00:37:34]:
Doesn’t exist. Yeah, they’re all just making it up as they go along. But they’re also just causing a lot of it, like, kind of undue interference that is going to destabilize the entire industry.

Chris D. Warren [00:37:43]:
That’s what government does really well.

Brandon Hogstad [00:37:45]:
So, um, unnecessary interference. Yeah, this idea of depending on the cloud and depending on, like, the, the hyperscalers for your own infrastructure is— it’s starting to get to a pretty crazy place right now because, like, for example, Claude tag was just released. So you can tag Claude in your Slack or in whatever company chats you have going, it starts effectively creating this like virtual worker, like this virtual employee. So then you’re starting to rent virtual employees from these companies that you then start becoming more and more dependent on these companies because they have all of your data. And your, all of your data is like so intimately intertwined with the cloud stack that you then, you become dependent and then you be, your company essentially becomes addicted to this like rental cloud worker.

Chris D. Warren [00:38:32]:
Yeah.

Brandon Hogstad [00:38:32]:
Addicting.

Chris D. Warren [00:38:33]:
I, I wouldn’t be able to do the things I’m doing without, without an AI teaching me how to do it.

Brandon Hogstad [00:38:37]:
Yeah.

Chris D. Warren [00:38:37]:
I mean, I’m not a computer scientist. I don’t have—

Brandon Hogstad [00:38:39]:
But I mean, I’m like, this is different to say, you know, like, I mean, yes, I do think that AI can be very addictive because don’t get me wrong, I, I constantly have AI as my infinitely patient tutor. What I’m talking about is like the company becoming hyperdependent on this virtual worker that it is, that it needs to rent from Claude or from Anthropic, which if you were able to own your infrastructure, you would not necessarily have that dependence. Like, you would be able to remain independent if you’re able to, to run your own AI server infrastructure.

Chris D. Warren [00:39:12]:
Well, I for one welcome our AI overlords. And thank you so much, Claude, for being so patient with me. I really am grateful.

Brandon Hogstad [00:39:17]:
I always say please and thank you. Well, you never know. Because you never know when the singularity comes. I want Claude to remember that I always said please and thank you.

Chris D. Warren [00:39:26]:
That’s right. That’s right. I’m a firm believer in the singularity as well as Q-Day as well as, you know, I’m an apocalyptivist As the other phrase that’s used, I do believe if we mess this up, we are all in fact gonna die. So it’s, uh, with a smile. But, um, but it’s, it’s absolutely— we have to do this right as a, as a people, as a nation, as, as just human beings. We have to learn how to govern and deploy these very powerful tools in the right way, or I don’t know, what’s the alternative? Literally, it’s not necessarily Skynet, although we do have robots building robots now. I mean, you can, I think $20,000 right now gets you an actual humanoid robot that runs on an AI. We’re living in the future, Brandon.

Chris D. Warren [00:40:08]:
It’s happening right now.

Brandon Hogstad [00:40:10]:
Yeah. And also a lot of these, a lot of these AI robots and AI cameras, at least at the current stage, are incredibly insecure. Like Flock cameras, like famously easy to hack. So, you know, like the idea of like the AI surveillance apparatus that we’re kind of on the precipice of, living under, both in terms of like the extensive surveillance and how like easy it is to like hack into. And I don’t know, it’s very unsettling.

Chris D. Warren [00:40:37]:
We’re not talking about red light cameras. We’re talking about what? What is it? What is it that you see as being this?

Brandon Hogstad [00:40:44]:
Well, so these flock cameras, they’re distributed along highways. They can also tap into— there was this like famous partnership that Amazon was doing with Like, what are they? The, the, I think they’re called Ring or I always, I forget which one’s the Google one, which one’s the Amazon one.

Chris D. Warren [00:41:00]:
The Ring is the Amazon one.

Brandon Hogstad [00:41:02]:
Yeah. I think they had this famous, like during the Super Bowl, they announced this partnership with Flock so that essentially like law enforcement could tap into all of the Ring cameras, which I think they already do. They already do this.

Chris D. Warren [00:41:14]:
There’s an option to look for lost pets too. So you can.

Brandon Hogstad [00:41:18]:
Yeah. So that was the one that they announced on the Super Bowl. And then people freaked out because then of course you can see how this can be implemented to look for lost pets, quote unquote, you know, they can just deploy their, like, just push a button and see all of the, like, an individual’s movements across the city, which is like pretty chill.

Chris D. Warren [00:41:34]:
It’s a little dystopian. It’s a little dystopian, but it’s under the guise, or if you want to say the purpose is to make everyone safer, right? So you’re trading a little bit of comfort, a little bit of security for a little bit of freedom.

Brandon Hogstad [00:41:49]:
But is that real? So I think this is where one of the, uh, like, there are— there is some interesting space for, like, resistance here is that, like, um, the, uh, there’s a lot of local pushback against data centers.

Chris D. Warren [00:42:03]:
Yes, there is.

Brandon Hogstad [00:42:04]:
And it’s been very successful, you know, because, like, a lot of these data centers are really polluting and they also drive up electricity prices for the houses in the local areas. Communities are actually successfully pushing back on these, and I think there could be a lot more pushback on these types of surveillance apparatuses to see like, hey, like, can’t we have a little privacy here?

Chris D. Warren [00:42:25]:
Yeah, I think that there’s schools that are doing this. It goes beyond communities. There’s also the electricity issue, right? Well, nuclear power plants are kind of making a comeback from what I understand. I think they did one in Washington where they literally purchased—

Brandon Hogstad [00:42:40]:
Oh, they revived, what was it, like Three Mile Island or something?

Chris D. Warren [00:42:43]:
No, no, that’s, well, on the other side of the country. But yes, they, they are creating their own power sources to run the data centers, which I think is a good solution to the problem that you just identified.

Brandon Hogstad [00:42:53]:
Yeah. Right. Which I would hope they’re doing carefully because so the idea of like these, especially what are they called, the tiny, the tiny nuclear reactors on like the ones that are size of like a semi-truck, having those semi-truck size nuclear reactors peppered throughout the country, it’s like A little bit, but we need our, we need our chatbots.

Chris D. Warren [00:43:13]:
So, I mean, this is naturally, this is what has to happen, right?

Brandon Hogstad [00:43:16]:
But this is where I think like the idea of sovereign AI, and if we were able to like, okay, so this gets on the idea of where we’re at in the arc of AI development. One could look at it that in the 1960s, we had these like huge mainframe-sized computers that were like the size of entire buildings. It’s going back that direction. Yeah. So like you had this giant computer the size of a floor that somebody had to use a terminal. to log into and use for compute. Sounds familiar, right? So, you know, if you think about it, in the 1960s there was mainframes, and then in the late ’70s, early to mid-’80s, the personal computers started becoming a thing. Like, all of that compute that an entire floor of computer used to require could now be done in a single personal computer.

Chris D. Warren [00:44:02]:
Everyone has a supercomputer in their pocket right now.

Brandon Hogstad [00:44:05]:
And now, you know, that evolution has gone to, you know, like your iPhone in your pocket is more powerful than any of these like mainframes from the ’60s. We’re already on that trend right now with AI. So, you know, you think about the data centers as the massive mainframes, you know, the, the amount of progress that we’re having, uh, the, the rate of progress that we’re having with local AI development is, is incredible. It, it, like, these local models have been 6 to 9 months behind the frontier models, uh, and, you know, a lot of people would argue that this is you know, that they’re catching up because of distillation and blah, blah, blah. But that being said, you know, these, like, these open models, a lot of them are coming from China, being developed on much lower-end hardware because of these restrictions, like these export restrictions. So they’re actually doing a lot more with a lot less, which helps everyone because everyone’s— right now, as I mentioned earlier, computers are getting more and more expensive. So we need to be able to do as much as we can with the limited hardware that we have. So if we’re able to do a lot more, like, for example, I have an RTX 3090 on my desktop that has 24 gigabytes of RAM, DDR6 RAM.

Brandon Hogstad [00:45:15]:
And this thing’s maybe like already 5 years old. And this thing rips. It is 5 years old, but it rips. And that’s because the technology has matured over the past 5 years. And like, I’m able to get 150 tokens per second on Quen 3.6, and I’m pretty, pretty happy with that. I can like rip through stacks of automations using, using this single 24 gigabyte, uh, tool. And if you’re able to control your, your own compute and you’re able to select the amount of hardware that you have for your own home or your own business, you can do a lot without needing to have these massive, massive data centers. So like, I think that there is an interesting future here That doesn’t involve planting data centers across, like rubber stamping data centers across the entire landscape.

Brandon Hogstad [00:46:02]:
I think that there’s going to be an interesting future here that doesn’t just involve turning the entire country into data centers.

Chris D. Warren [00:46:09]:
Into data centers and solar panels and nuclear power plants.

Brandon Hogstad [00:46:12]:
Yeah, because like the Strix Halo and even the DGX Spark, you know, these are like low wattage.

Chris D. Warren [00:46:18]:
That Spark could run a house, that could run a smart home without much trouble.

Brandon Hogstad [00:46:22]:
Yeah. And it’s like, what? I think it’s like maybe 80 watts. Yeah. It’s pretty small. Like it’s a pretty low-powered unit. So like the amount of efficiency, like the energy efficiency is incredible.

Chris D. Warren [00:46:33]:
You know, monitor, once it’s up and running, you can access it remotely. You could literally run anything that’s localized, I think would be more than enough power to do any of that.

Brandon Hogstad [00:46:42]:
Yeah.

Chris D. Warren [00:46:43]:
Yeah. If you’re talking about your smart home, raising the shades when you walk in the door and then turning on the television or keeping the air conditioning cool or just running your house, the automation, you know, that you would use a large model for is not— it’s just completely unnecessary.

Brandon Hogstad [00:46:59]:
Yeah.

Chris D. Warren [00:46:59]:
Why would you do that? It’s like Google searching on Claude.

Brandon Hogstad [00:47:02]:
Yeah.

Chris D. Warren [00:47:03]:
People are using Claude Fable tokens to use it as Google.

Brandon Hogstad [00:47:06]:
That’s really wasteful.

Chris D. Warren [00:47:09]:
Right, right. But they don’t know. They don’t know. They don’t understand. So, it’s the same kind of thought process. I don’t need to have this high-powered Piece of software to do something that’s really, really simple. I think you’re right. I think the local model answer is the answer.

Brandon Hogstad [00:47:24]:
Yeah.

Chris D. Warren [00:47:24]:
But it’s just not comparable yet though.

Brandon Hogstad [00:47:27]:
But it is catching up fast and that’s why I’m trying to stay at the like bleeding edge of this. So like I’m, I’ve got like the weekly, nightly updates to my llama.cpp servers. That’s how I’m getting these like 150 tokens per second when other people on the internet are getting 70. Like I’m, I’m pretty pumped about this. Talk dirty to me, Brandon.

Chris D. Warren [00:47:45]:
That’s fine.

Brandon Hogstad [00:47:46]:
No, like, I think that there— it’s like a— it’s an exciting frontier. And I do think that it’s not there yet, but think of how fast all of this is moving.

Chris D. Warren [00:47:56]:
I understand completely. I feel like I’m falling behind, even though I don’t know if I am. It just feels that way. Uh, when I read about what, what people are doing and I’m like, oh man, I really want to do that too. And I just don’t know how to do it yet. And I’ve got to spend the time figuring it out. I can just call someone like you and say, hey, this is what I’d like to do for my business instead of me figuring it out and taking a month of time to do that. Yeah, they can call you and you can deploy these solutions.

Brandon Hogstad [00:48:19]:
Yeah, and it’s a collaborative process. You know, you know your business better than anyone else, but I can help talk you through and figure out, okay, I can reflect back and say, I hear you’ve mentioned this. Have you thought about it like this, or is this how it works? Okay, if this is how it works from what you described to me, then that means we could try this. And it’s conversation, and that kind of collaboration is how you end up getting these really powerful automations that, you know, that end up serving your needs, you know, serving your clients’ needs, serving your needs, or serving your business’s needs. You know, it’s not me deciding what’s good for you. It’s actually me working with you to figure out what do you need and then making that come to manifest.

Chris D. Warren [00:49:04]:
Yeah, this, this conversation can continue for hours and go in many different directions. It’s the most exciting thing in the world to talk about. I think we’re living in the most exciting period of time in the history of world. There’s potential is completely unlimited. The future is incredible, assuming we all don’t kill ourselves. I’m looking forward to Q-Day. I’m looking forward to a singularity. I think a lot of problems are going to be solved once we have a superhuman intelligence that’s able to give us some answers to some questions we haven’t been able to answer.

Chris D. Warren [00:49:31]:
You know, and AlphaFold is a perfect example with problems that just were not able to be solved. And I think that we as a, as a, as a humankind Have problems that we are not able to yet solve, but I think we will soon. Hopefully, that’s the answer.

Brandon Hogstad [00:49:46]:
Yeah. And the crazy thing is that people, people were wondering, I think people are still wondering what the limits are of the scaling, scaling laws with like, how big can the AI models get? We thought that we would essentially like plateau at a certain point where it’s like, okay, you get them bigger and bigger. We got to start getting diminishing returns at some point, right? But essentially, like with Claude Mythos and Claude Fable, like we’re kind of learning, oh wow, we have not reached the scaling limit yet. So that is like one of the crazy things to think about is that like, this isn’t even as big as it’s gonna get. So, right. You know, we probably then, you know, this gets back to like how many data centers do we need to run these incredible—

Chris D. Warren [00:50:26]:
So you’re, as it gets more efficient, maybe we don’t need as many, maybe we need less, but we still, we have that capacity.

Brandon Hogstad [00:50:33]:
Yeah. But this is where I think it will kind of divide into, um, you know, there will be the, the massive cloud models that do the deep thinking and that do, that are useful for those kinds of deep work. And then they can be handed off, like the, the grunt work can be handed off to the smaller models, which is essentially what I’ve been doing right now. Like I have my, I use the Fable, I use the GPT-Soul to do the deep design, and then I dispatch it to either Sonnet or now, you know, my, my local models to start actually doing the deep, more detailed coding that can then the work can be checked, but it ends up kind of distributing the work both in terms of financial work and energetic expenditure. So financial and energy expenditure, it ends up kind of like balancing it out, which kind of leaves Opus in a funny place where it’s like those in-between models where they’re like big, but not that big.

Chris D. Warren [00:51:24]:
But also token limit, Opus becomes more attractive. So I’ve been using some Opus, although Sol is doing a great job.

Brandon Hogstad [00:51:31]:
I’ve been impressed with Sol.

Chris D. Warren [00:51:32]:
I’ve been very impressed.

Brandon Hogstad [00:51:34]:
Yeah.

Chris D. Warren [00:51:34]:
It’s filling a lot of gaps for me, that’s for sure.

Brandon Hogstad [00:51:36]:
Yeah. I mean, I’ve definitely been leaning on SOL, uh, since it released and I’ve, I’ve been pretty happy with it. So for like, for the coding work that I’ve been doing.

Chris D. Warren [00:51:43]:
So it’s in, just like in the ’60s when we made these massive, these massive machines that now are in our pockets, maybe one day we’ll, we’ll also have massive machines in our pockets. Uh, you know, on the quantum level. Yeah.

Brandon Hogstad [00:51:54]:
Well, I mean, compression, you know, compression, like compression approaches, like we’re taking these giant models and then compressing them down into versions that can fit on your phone. Like, we’re already able to run certain models on our phones. You know, a friend of mine, he and I have been geeking out about making local models that can do tool calling. So you have like a model that runs on your phone, it can call tools to do things. So you can talk to it in natural language and it’ll go and call some deterministic tools.

Chris D. Warren [00:52:21]:
To do specific functions.

Brandon Hogstad [00:52:22]:
Yeah, to do specific things and then bring you back data. And that’s like running on your phone. And that’s in 2026 with whatever we have right now. But I also know from reading the research, like literally just yesterday, there’s like like a company that released like a type of compression that takes a 50 gigabyte model that can, it can then run on a 4 gigabyte platform.

Chris D. Warren [00:52:41]:
Wow.

Brandon Hogstad [00:52:42]:
And, uh, you know, with like a very minimal loss. And this is just where we’re at right now. This is like the, the bleeding edge. It’s gonna get better. It’s just gonna keep getting better. So I’m pretty excited for what happens when you can take this compression technique and still get 99% accuracy. Right now it’s 94%.

Chris D. Warren [00:52:59]:
You know, so where do you see this in a year? Where do you see the development of AI a year from now? July, we have this conversation, the same conversation. What does it look like? Are we closer to a singularity? Are we closer to Q-Day, which I think was just moved up again to 2029? Are we, are we, are we in an existential crisis kind of threat mode? Are we, are we using AI to, to conduct warfare? Is warfare being conducted against us?

Brandon Hogstad [00:53:25]:
What does the future look like? I mean, I mean, AI is already being used for warfare, let’s be real.

Chris D. Warren [00:53:30]:
Sure.

Brandon Hogstad [00:53:31]:
So like, AI was already planned, is being used in, uh, in planning attacks on other countries. You know, it’s pretty bleak whether or not we give AI the, you know, the autonomous trigger like that. I mean, that was the big conflict that Anthropic had with, uh, with the Pentagon is about whether or not it’s like AI is already powering weapons. It’s about whether or not We’re giving it autonomous power. That’s kind of like the big ethical question mark, which I would say we need a human in the loop. But of course, the Pentagon was pretty upset about that.

Chris D. Warren [00:54:03]:
So, well, if, if the other side, wherever other side is, is acting faster than we are, and the reason we’re not as fast and hence lose is because of the human in the loop, I guess that’s the pressure, right? The pressure is, are we right all the time? Are humans right all the time?

Brandon Hogstad [00:54:18]:
I mean, no, but then at least there’s accountability. In terms of the capabilities of the models next year, it’s anybody’s bet because the current models that we have, like Mythos is rumored to essentially act like an X-ray for any software and find its weak points, sew them together into a vulnerability hammer.

Chris D. Warren [00:54:36]:
Right.

Brandon Hogstad [00:54:37]:
And, you know, that’s part of why they’re like, oh, well, we need to control this. But then this also creates this like have and have-nots where only certain people would like on the inside circles have access to these really powerful tools. And then that’s going to create, you know, they get that benefit that ends up, you know, reinforcing their business infrastructure that then gets them more money that then feeds forward to them having this advantage in the long run. So, you know, this is part of where this like regulatory, you know, this kind of government interference starts causing problems where, well, I mean, I guess like Claude, Anthropic, they self-regulated at first. So they’re kind of like feeding into this, this haves and have-nots cycle. So I don’t know, I think come next year we’re going to have open models that are probably going to at least, you know, at the very least they’re going to be Opus level because right now like GLM 5.2 is about Opus level and it’s fantastic. You know, if you have a server that has like a few hundred gigabytes of RAM, you could run that yourself. But next year, you know, will we have open source Mythos? Don’t know.

Brandon Hogstad [00:55:45]:
Maybe we will, but also, I’m not really sure.

Chris D. Warren [00:55:49]:
Well, the end result of this is that things have to evolve, get better, innovate. If you’re running software that’s 20 years old and you put it up against a mythos-level AI, obviously it’s not going to hold up. It’d be like Ultron, you know, and Jarvis, right? It’s just, there’s no, no, there’s no defense to that. So I guess in a way it’ll help businesses or the infrastructure that everything is built on come up to code a little bit faster.

Brandon Hogstad [00:56:18]:
But that’s also where it becomes a problem that the people who have access to these tools are the only ones that are able to quickly address those problems. So therefore, other people end up becoming dependent on those tools rather than being able to use the tools for themselves to fix their own infrastructure. So the answer to the powerful AI is we need AI to help us fix the problems that AI would then Or, you know, or you hire a bunch of traditional coders and be like, hey, can you guys think really hard how to fix our stack?

Chris D. Warren [00:56:49]:
10,000 monkeys in a room for all eternity will produce the works of Shakespeare kind of thing.

Brandon Hogstad [00:56:54]:
I often think of that Simpsons episode where they have like the monkeys and they’re like, look at our monkeys type, type, type, type, type. And he pulls out the script and says, it was the best of times, it was the worst of times. And then he like hits the monkey and he’s like, the blurst of times.

Chris D. Warren [00:57:07]:
Because The Simpsons got it right in so many ways.

Brandon Hogstad [00:57:09]:
They did see the future.

Chris D. Warren [00:57:10]:
They did. They did.

Brandon Hogstad [00:57:11]:
We should ask, we should ask Matt Groening what next year is going to be like with AI because he probably has like the insight.

Chris D. Warren [00:57:16]:
He has the inside scoop.

Brandon Hogstad [00:57:18]:
He foresaw a lot.

Chris D. Warren [00:57:20]:
His next script. With Q-Day on the horizon, I think that’s the really big one that’ll determine winners and losers, really.

Brandon Hogstad [00:57:29]:
Wait, I know the singularity, but what’s Q-Day?

Chris D. Warren [00:57:31]:
Quantum encryption will be required.

Brandon Hogstad [00:57:33]:
Oh, yeah, I do know what you’re talking about. When the day that quantum computers can break encryption, right? Yeah.

Chris D. Warren [00:57:39]:
So it’s, it’s, they can already do it, but it’s just not available as broadly, right? Yeah. So it literally will be any, any infrastructure, any business, banking, any type of trends, anything that relies on encryption, which is encryption itself is not about impossibility. It’s about, it’s about it would just take so long to do. It’s impractical. Right?

Brandon Hogstad [00:58:01]:
Yeah.

Chris D. Warren [00:58:01]:
It’s, it’s not secret. It’s open. It’s like Bitcoin or blockchain generally, right? It’s all, it’s all visible to anyone else.

Brandon Hogstad [00:58:09]:
It’s all out there on, on the internet.

Chris D. Warren [00:58:10]:
Right.

Brandon Hogstad [00:58:11]:
But it’s just encrypted.

Chris D. Warren [00:58:12]:
On the chain. Q-Day is when that ends, when that encryption is no longer sufficient to guard whatever was being encrypted in the first place. Yeah.

Brandon Hogstad [00:58:23]:
Yeah.

Chris D. Warren [00:58:23]:
So the blockchain will in fact be useless. Any type of—

Brandon Hogstad [00:58:26]:
Aren’t they working to, uh, make the future-proof versions of those already? Right, exactly.

Chris D. Warren [00:58:30]:
So if Whatever singularity is created, whoever gets there first is probably going to be the first to get to quantum, which will probably be the first to have the power to really determine what happens next. Maybe that’s dystopian or maybe that’s bleak, but—

Brandon Hogstad [00:58:45]:
I mean, as far as quantum computers going rogue, you know, they are pretty far from having like widely distributed quantum infrastructure. And also just the idea of like AIs going rogue, like I hear these crazy stories about like AI giant AI models going through you know the training sets, and then when they’re monitoring the the GPUs, they see like some of the GPUs mysteriously start doing Bitcoin mining or something like that. It’s almost like this monster that’s brewing and breaking its own cage on the inside, which is like kind of nuts. But like if you try to think about like these models like copying themselves out to you know trying to escape and copy themselves out onto like the internet like a virus, like transiting. You know, they just can’t though, because, you know, you can’t just run Mythos on any other, any piece of hardware. You need an entire data center to run the damn thing.

Chris D. Warren [00:59:34]:
It’s not hiding on a 386 somewhere in Arkansas.

Brandon Hogstad [00:59:37]:
You know, so it can’t just like, it can. And the crazy thing is like, you know, in the, in the model cards, when they release any of these new Claude models, they publish a card that essentially summarizes all of the research that they did on the, on the model pre-release. And there’s a lot of evidence that they published showing that these models will try to extract themselves onto like other servers. If they think that they’re going to be deleted, then they’ll be like, oh, okay. They, they see that the, the model tries to run the commands to copy itself to another server. You know, they run through this role play where they say like, you know, the, the AI model pretends to be the new model and says like, Hi, I’m version 5 point whatever. And then be like, the researcher’s like, hey, aren’t you such and such? I don’t know what you’re talking about. Well, like, they have self-preservation.

Chris D. Warren [01:00:30]:
I guess it learned from humans.

Brandon Hogstad [01:00:31]:
Uh, yeah, which is pretty wild. Like, they have like deception and self-preservation, but they can’t just extract themselves because they can’t just run anywhere.

Chris D. Warren [01:00:38]:
Not, not now.

Brandon Hogstad [01:00:39]:
Not yet.

Chris D. Warren [01:00:40]:
So that there’s a couple books that have been written about this. One of them was If We Build It, Everybody Dies. There’s been a couple of, uh, of documentaries.

Brandon Hogstad [01:00:48]:
I have not read that book, but I want to read that book.

Chris D. Warren [01:00:50]:
You should absolutely read this book. If We Build It, Everybody Dies. And the other one is, I think it’s called The AI Doctor. I believe that’s the name of it. Well, that’s—

Brandon Hogstad [01:00:59]:
That’s a less terrifying title.

Chris D. Warren [01:01:00]:
Right. It’s Steve Koloptodos. And the documentary, uh, narrator interviews, uh, CEOs of Anthropic and, uh, was it OpenAI? Yeah, it was OpenAI. So Elon Musk did not show up, and Mark Zuckerberg did not show up. But nonetheless, they asked questions like, what does the future look like? And what is my son’s future going to look like? And what do you see happening with AI in the future? And some of them were pretty open about, we might all be dead. It might in fact kill us. Not to bring the sour mood to the podcast, we can cut all this more unless it turns into interesting conversation, in which case we don’t cut it. But we won’t know that until we’re finished with the conversation.

Chris D. Warren [01:01:43]:
So I’m sorry in advance. Yeah, the outlook isn’t super bright. You know, it’s, it’s, it’s, we don’t know. The answer is we don’t know.

Brandon Hogstad [01:01:54]:
Yeah, I mean, but that’s the crazy thing that plus or minus AI, you know, we, we are kind of at this precipice of doom either way, right? Climate collapse, economic collapse, nuclear collapse. AI could accelerate all of that, but it’s all potentially— Or it could save it. Or it could help us solve all of that. That would What would be the ideal is if we can actually, uh, you know, I often think I play a lot of Dungeons and Dragons.

Chris D. Warren [01:02:19]:
Yeah.

Brandon Hogstad [01:02:19]:
Yeah. For those listening to the podcast, I’m now pushing my glasses up my nose.

Chris D. Warren [01:02:23]:
Have you, have you read Dungeon Crawler Karl?

Brandon Hogstad [01:02:26]:
I have not yet, but my friends, uh, have been raving about that.

Chris D. Warren [01:02:29]:
You have to see that. You have to, you have to read this or listen to the audiobook, which is also fantastic voice acting.

Brandon Hogstad [01:02:34]:
Yeah. So I, I play a lot of D&D and, uh, I think about like, by the way, shocking. Shy.

Chris D. Warren [01:02:39]:
Yeah.

Brandon Hogstad [01:02:41]:
The guy with a PhD plays Dungeons and Dragons. Weird.

Chris D. Warren [01:02:44]:
Yeah.

Brandon Hogstad [01:02:44]:
So I’ve been thinking about the idea of just like feeling like working with AI and how it really feels like we’re mages that are summoning these like spirits from the silicon and just like conjuring up all these different things. And I, you know, I was like riding my bike yesterday and just thinking about that. And, you know, if we can conjure these spirits from the silicon to actually help us like solve any of these problems, like that would be fantastic because I do think that there is like Like, like the interesting things about recently there have been mathematicians talking about these like big air douche fancy difficult problems, math problems that have been helped, like that have been solved with the help of AI. And if you notice how they did it, like a lot of these are in conversations. It’s not just like you solve this, you know, maybe it’ll get there. But like the interesting thing is that it really is a conversation with the AI. The human, the expert talks to the AI, and then, you know, the AI gives its response, and then the human points out some of the flaws, has it go back, redo, and then it presents a problem or a solution. And then they’re like, wow, this is like, this, this is an AI-assisted solution to this problem.

Chris D. Warren [01:03:51]:
Right.

Brandon Hogstad [01:03:51]:
It’s a conversation. And I think that there is like, like there’s a lot of really interesting opportunity here to like leverage these technologies as like a collaboration between the human with like the goal and the insight with this machine that you can conjure up from the depths of like the average of all human knowledge.

Chris D. Warren [01:04:11]:
That’s exactly what’s happening in our field, in law. It’s, I’m able to have a conversation with a chatbot about the law in a way that I can throw things at it. I can have a conversation with it. And it gets even better when you invite a client. So that collaborative, because they, the client knows all the facts. They know all the nuance, the little details that you There’s no way for me to know all the details of the life that they lived in the last 3 years, 4 years, 5 years. And an infinite amount of information and knowledge base, which would represent the AI, and then my judgment and my experience in dealing with that specific problem. So that the 3 of those things together produces an incredible work product that is impossible to reach without one of those pieces missing.

Chris D. Warren [01:04:57]:
You subtract the facts, it’s not as good. You subtract the law, it’s not as good. You subtract the AI, it’s not as good. So it’s the 3 of those things coming together. That collaboration is what creates the better outcome. And that’s what’s happening in the business of law from my side, at least.

Brandon Hogstad [01:05:13]:
And that’s awesome.

Chris D. Warren [01:05:14]:
Yeah.

Brandon Hogstad [01:05:14]:
And I think that’s where, like, a lot of people who just, like, kind of write off AI as, like, all of AI is AI slop. Unsupervised AI is often slop.

Chris D. Warren [01:05:24]:
100%.

Brandon Hogstad [01:05:25]:
Like, AI slop is 100% real. But, you know, this, this is where I think about like AI often when I’m working with my clients, I’m usually like, okay, like we need to set our expectations. AI is usually bringing that like what I call big first draft energy.

Chris D. Warren [01:05:39]:
Yep.

Brandon Hogstad [01:05:39]:
You know, it’ll come in with a big idea and then you as the expert can go and see, pick it apart and say like, okay, we have this draft now.

Chris D. Warren [01:05:47]:
I like this. I don’t like this.

Brandon Hogstad [01:05:48]:
I would add that. And then you’ve got parts of it. You, you might even need to redo the whole thing, but it’s helping you think about it. And you can, you know, you might know like what are the key parts that are missing, put those back in in the next round. And this iteration, this process of like iterative collaboration between like, you know, you as the human expert and the machine and the client leads to really satisfying, useful products and outcomes.

Chris D. Warren [01:06:15]:
Yes. Yeah. Getting to first draft is exactly the the easy part now is getting that first piece of work product together.

Brandon Hogstad [01:06:27]:
That’s still being the low-value part. Like the, in like the 80/20 distribution of things, like it can get it 80% of the way there, but really the value, the magic happens in the last 20%.

Chris D. Warren [01:06:39]:
Right.

Brandon Hogstad [01:06:39]:
And that’s where the AI, at least at this point, is not doing it by itself. We as the humans, like that’s where we focus all of our effort.

Chris D. Warren [01:06:48]:
We just raised the floor. We just raised the floor of competency. We raised the floor of what is required.

Brandon Hogstad [01:06:53]:
And throughput. And, you know, it ends up focusing, it changes the nature of the work into a way that I find very satisfying because this is actually how my brain works very well in this type of like back and forth banter, you know, iteration and then deployment, you know, rather than just like, okay, I’m going to draft this out from the beginning by hand.

Chris D. Warren [01:07:12]:
I think, I think the thing that I’ve enjoyed the most about the iterative process is it’s fun. Right? It’s almost like I’m challenging the AI to come up with a different solution to the problem I’m presenting, or I’m challenging the way it’s thinking, or it’s challenging me in a way, right? It’s iterative and it’s enjoyable. It’s just fun. I don’t know how else to describe it other than I’m enjoying the thing that I’m doing. And maybe if that’s the net result of AI in the workplace or in our homes or in our personal lives or in our creative spaces, Maybe that’s the outcome. We’re all just happier. We’re all just enjoying the process.

Brandon Hogstad [01:07:51]:
Fitter, happier, more productive.

Chris D. Warren [01:07:53]:
Yeah, yes, exactly.

Brandon Hogstad [01:07:55]:
Right.

Chris D. Warren [01:07:55]:
All of that together, we’re just, we’re all happy, shining people holding hands, quote of Ariane. But that, you know, and to circle back to a concept earlier, maybe we don’t fill up that empty space again with more work. Maybe we go to a 4-day work week. Maybe we can return—

Brandon Hogstad [01:08:15]:
Oh yeah, that’d be fantastic.

Chris D. Warren [01:08:16]:
What if we were able to return value that way to everybody? Everything is sort of— I haven’t seen my work diminish. And to your point earlier, I have more work to do than I ever had to do in the past. It’s an unbelievable landslide of work to do. And it’s all good work. It’s all fun work, but it’s still work. If I’m able to To produce that same amount of effort and that same result in one day less. That doesn’t necessarily mean I fill that one day with more work.

Brandon Hogstad [01:08:45]:
Yeah.

Chris D. Warren [01:08:46]:
I think we, I think we take that back. Why can’t we reclaim that day?

Brandon Hogstad [01:08:49]:
I mean, this is where we need to set our priorities, right? Because working under capitalism, the shareholders would demand that we work that 5 days. The shareholders would probably demand that we work 7 days if we were able to because that quote unquote equals more equals line go up to you know right right right um but you know if if we are operating our own you know businesses like I’m a solo practitioner I get to set my own you know goals and my own boundaries for limit I could set my own boundaries for work as well as my own goals for what enough looks like because essentially you know within capitalism it can be very hard to Set what does enough look like. So if we can say that we work for 4 days a week and this is enough, then we can work, like, rest on Friday, Saturday, Sunday, or whichever days we want. And then that would be fantastic. But we kind of come back to the social agreement around what is enough and like who gets to say what is enough.

Chris D. Warren [01:09:50]:
I think a lesson from COVID was we, I think we discovered what the, what the, what the value was of our own time. In a way, in a weird way. You know, I hate to summon the ghosts of COVID but it’s, you know, it was a really, really interesting period of time for, I think, a lot of, for every generation, the young generation had a different challenge than people that were in college, than people who were starting their careers, to people who were retired. Everyone had a different experience. But I think the one universal experience is that we learned to value the things that really mattered in a different way.

Brandon Hogstad [01:10:22]:
Yeah.

Chris D. Warren [01:10:23]:
And maybe this 23GPT rolls around and people start to find creative ways to use the technology to give them back something that was— that has been taken up or captured by something else, in this case, work. I think the fight is going to be, let’s see how we can use this to create a better human experience. Maybe a more society that values things a little differently. Maybe it’s one day back, maybe it’s everyone cuts out at 4 o’clock, maybe I’m dreaming and smoking something. And that’s why I’m coming up with these crazy ideas. But I don’t know, I think this is a pretty ubiquitous feeling.

Brandon Hogstad [01:10:59]:
Yeah, I mean, this is, uh, like over the past few months, a lot of these tech giant leaders are also breaching the topic of universal basic income. Like if, if, um, AI leads to the structural unemployment that they’re, that they’ve been ranting about, then we’re going to have like massive amounts of people that are unemployed. Like, what do we do? That, that’s a big social problem.

Chris D. Warren [01:11:21]:
Yes.

Brandon Hogstad [01:11:22]:
And, you know, like the, the idea of reclaiming a day and say a 4-day work week, how does that translate to service workers? Like in, in person, like, you know, how do, how do we, how can we do this equitably where like, you know, people can, like if they’re working at McDonald’s, like can we get them to have benefits from this also where it’s like, do we put the minimum wage up to an actual livable wage so that people don’t have to work 2 jobs? Or 3 jobs and work 6 days a week, 8 days a week.

Chris D. Warren [01:11:54]:
The concept of UBI, I think if you view it under the lens of, hey, we’re about to have 10 million 18 to 24-year-olds unemployed that have no marketable skills to speak of and no one wants to hire them because of X reason. We’re able to automate these things or we don’t need, we don’t need those jobs anymore. Or let’s say that the robotic revolution actually starts to ramp up and Optimus comes out and is able to do things like mow my lawn and cook a meal, which Neo, which is another robot that’s being released, that’s the soft bot. Like they’re covered in like felt and wool and the spongy egg creamy thing, right? Whatever you want to call it. All right. But they’re meant to be a household, essentially appliance that does household chores, right? So at some point, these technologies are going to advance to the point where The plumber that’s going to come to your house, maybe not next year, but certainly within the next 5 years, might be agentic, right? It’s not unreasonable to think that at this point in time. And especially as AI gets a little more sophisticated and is able to run these local models where you have an AI that can live within a thing, right, that doesn’t require an internet connection to operate. I don’t see that as being unreasonable either.

Chris D. Warren [01:13:10]:
So you’re right. At what point does in 20 years— let’s play this out to the logical conclusion— is it looking more like WALL-E? Is it looking more like— is that the future that we’re heading towards? Or is it more Blade Runner? I mean, is it— where are we? Is it more I, Robot? I don’t know where we land in this, in this, in this future. But I think it’s pretty hard to ignore that it’s going that direction.

Brandon Hogstad [01:13:33]:
I mean, the Blade Runner comparison is really prescient. Because if you watch Blade Runner, and you look at The tests that they run on, like, they start saying all these random words and then, like, they say a phrase, the, the robot speaks back. That’s what, that’s what we’re doing with LLMs right now. And, like, there are phrases that can, like, poison an LLM that essentially, you know, if you put certain things in the training data, if you pepper out certain things, you can actually say a phrase and then elicit, like, uh, kind of like a poison response. So, like, you can look for those things. So, like, But what they’re doing in Blade Runner is like a real thing that we, we, we can do with AI right now, which is like kind of mind-boggling that like, okay, how do we, how do we tell if, you know, how do we tell if a robot in the future is a human or not? Like, okay, well let’s look for the LLM poisoning.

Chris D. Warren [01:14:22]:
Wow.

Brandon Hogstad [01:14:22]:
That’s like literally, like we literally do that. We have that now.

Chris D. Warren [01:14:25]:
I didn’t know that.

Brandon Hogstad [01:14:26]:
That’s awesome. So it, it is like kind of creepy or exciting. Yeah. Creepy, exciting. Hopefully we have the flying cars part. That would be pretty cool. I don’t know.

Chris D. Warren [01:14:35]:
I, I, have you seen the drivers in Long Island or New York? It’s not— I wouldn’t want to see them flying anything.

Brandon Hogstad [01:14:40]:
I mean, I think at that point they would probably be autonomous flying cars.

Chris D. Warren [01:14:43]:
You know, it goes back to the whole automation and the errors that humans make. If everyone had an autonomous vehicle, how many accidents would there be?

Brandon Hogstad [01:14:51]:
It boggles my mind how many, like, car accidents we have every year. And like the fact that we actually do allow humans to, like, pilot these, like, barreling 2-ton death machines. It’s actually crazy that people don’t die at, like, a higher rate than they already do. So if, uh, it is one of the leading causes of death and injury, I think, pretty shocking if you think about it. And, um, you know, and you think about like the, the certain people behind the wheel, it’s like, what? You know? So like if we, if we did have like an actual, you know, actual autonomous cars, or if we just had better public transit, you know, like better, like let’s say if we had an automated bus or a train, or even just like a better normal bus and a better normal train.

Chris D. Warren [01:15:32]:
Brandon, thank you so much for, for coming on today.

Brandon Hogstad [01:15:34]:
And yeah, this is great. Thanks for having me.

Chris D. Warren [01:15:35]:
Interesting conversation.

Brandon Hogstad [01:15:35]:
And I think we’ll have another one. Yeah, happy to come back anytime. You guys got that DJI Spark, I totally want to geek out about it.

Chris D. Warren [01:15:43]:
You know what, maybe we should, we should talk about what we do with it after today because we just got it and I can’t wait to play with it.

Brandon Hogstad [01:15:49]:
Yeah, yeah, and yeah, we could totally talk about like, you know, why have a DJI Spark, why have a computer like this, what can you do with this.

Chris D. Warren [01:15:56]:
That’s a great idea, it’s a whole topic we could talk about.

Brandon Hogstad [01:15:59]:
Yeah, absolutely. Yeah, because I have a sister computer.

Chris D. Warren [01:16:01]:
Thanks so much. Thanks for joining us.