How AI Is Transforming the Federal Government

[SPEAKER_01]: What's the point of who cares about this AI? [SPEAKER_01]: All it can do is paper work. [SPEAKER_01]: And I was like, but the government is mostly paper work. [SPEAKER_01]: So to be able to do that, 30% faster, that can actually have a huge impact. [SPEAKER_01]: Even if we wanted it to take your job, I don't think it can. [SPEAKER_01]: It gets just not good enough. [SPEAKER_01]: Can we really trust AI?

[SPEAKER_01]: Right now, no, and the reason why is because we don't have any way to test it, and that's a problem. [SPEAKER_01]: Because if we don't have a way to test it, then you can't guarantee you won't do all sorts of really bad behaviors. [SPEAKER_03]: Look, you probably never have a security plus maybe even a security clearance. [SPEAKER_03]: So nobody taught you how to write poems or how to test a security control or submit a T.O. [SPEAKER_03]: package.

[SPEAKER_03]: I'm Chris Arkpala, I fear years ago, I was in your shoes, all the five on paper, but completely lost when it came to our map. [SPEAKER_03]: I had a degree, I had to serve, so I had to drive, and what somebody said, how to test an A.T.2 control or got a date stick by, so I had no clue what that actually looked like. [SPEAKER_03]: Fast 4 or 5 years, I worked across DOD and federal agencies, led control assessments, run ATO package, and pass orders.

[SPEAKER_03]: That's why I built our Neficatomy to teach you the real-world execution. [SPEAKER_03]: They don't cover in certification books. [SPEAKER_03]: Inside I'll show you how to write a poem. [SPEAKER_03]: And don't get bad as bad. [SPEAKER_03]: Test and validate security controls, Translate tech jardan, Navigating this 853 and horror map with confidence. [SPEAKER_03]: If you're in IT support in the Garmin systems or stuck on edge of sub security, this is your way.

[SPEAKER_03]: The people who go through my training don't just be hired, they hit the ground running, because they practice the work before they win. [SPEAKER_03]: Go to horrormathacademy.io and let's get the work. [SPEAKER_04]: Welcome everybody to another edition of the Tech World Podcast. [SPEAKER_04]: I'm your host, Chris, Information Systems Security Officer inside the GuvTech space. [SPEAKER_04]: And in today's podcast, we have a great guest. [SPEAKER_04]: We have Glenn Parham.

[SPEAKER_04]: He is a person that has me history in a DOD space. [SPEAKER_04]: He's the first dinner of AI Tech Lead. [SPEAKER_04]: He's also a tech founder of Pundit Analytics. [SPEAKER_04]: He's also entrepreneur. [SPEAKER_04]: He would account Berkeley and without further ado, Glenn, how you doing? [SPEAKER_01]: Thanks for having me, man. [SPEAKER_01]: Appreciate it. [SPEAKER_04]: Yeah, I'm glad we was able to get you on a podcast. [SPEAKER_04]: I know who's just talking nerdy stuff.

[SPEAKER_04]: Yeah, I cast the things. [SPEAKER_04]: Yeah, I'm glad to have you on. [SPEAKER_01]: It's great. [SPEAKER_01]: I love podcasts. [SPEAKER_01]: It's you're able to, you know, get into depth about these topics and communicate it. [SPEAKER_01]: So it's pretty cool. [SPEAKER_04]: Yeah, and so my boy introduced me, I was just following you for a long time. [SPEAKER_04]: I was like, I want to reach out to you for a long time. [SPEAKER_01]: Yeah, I've seen your stuff, yeah.

[SPEAKER_04]: Well, when you say you started your podcast, I was like, yeah, it's my opportunity. [SPEAKER_04]: Yeah, so we linked up, so I'm glad we had a, we, we were able to, we were able to make it happen. [SPEAKER_01]: So yeah, this is my first main interview. [SPEAKER_01]: I think I've ever had, so I, you know, I'm excited. [SPEAKER_01]: I'm honored to be here.

[SPEAKER_04]: So we're going to make some magic work so so tell me about your early experience in generative AI so you start at which you're an analytics I mean funding analytics right so can you kind of break down what it is and how did you get into it?

[SPEAKER_01]: Yeah, so once you see Berkeley, we're studying data science and computer science, context is like, went to Berkeley a couple of weeks after I get there, you know, President Trump's first firm, he was elected, so the campus is very politically charged. [SPEAKER_01]: And I wasn't passionate necessarily about advocacy, but just like analyzing politics, like the discourse, how polarized things could be, things like that. [SPEAKER_01]: And this was like almost a decade ago now.

[SPEAKER_01]: And so, at that time, I was technical, I was data scientists, computer scientists, started a research organization at Berkeley called Political Computer Science. [SPEAKER_01]: It's actually now one of the largest clubs, research organizations at Berkeley, which is super cool. [SPEAKER_01]: They even have it like a whole major around, it's a whole thing.

[SPEAKER_01]: I spun off a startup from that, just focused on using [SPEAKER_01]: Technically, AI to analyze politics, and then I started using the first version of GPT, so in ChaggbT, the AI capability back in 2018, 2019, it was trash, so like, yeah, it was not helpful at all, but I saw where things were like, directionally going, and that's what started to get me nervous over time as I got better and better, I was like, what

[SPEAKER_01]: You know, I was using in politics, but what is the government doing about it? [SPEAKER_01]: And I went to AI.gov at the time, back in like 2020 or 2021, and the website was down. [SPEAKER_01]: Like, there was no website. [SPEAKER_01]: And I was like, oh, I don't know if the government knows what it's doing with this AI stuff. [SPEAKER_01]: And yeah, that's what kind of convinced me to want to go and get into government, at least for a few years of the Pentagon.

[SPEAKER_04]: And that's amazing because I'm like, I'm like, time back into your startup because when politics, we don't know what's real, we don't know what's fake. [SPEAKER_01]: Yeah. [SPEAKER_04]: So that's a great company. [SPEAKER_04]: And I heard with GPT-1, you could even use it. [SPEAKER_04]: Like you had to actually code to be able to use. [SPEAKER_04]: That's what I've heard. [SPEAKER_01]: Yeah, absolutely. [SPEAKER_01]: There was no, like, chatchipy-t, nice website.

[SPEAKER_01]: You had to code everything up yourself, basically. [SPEAKER_01]: It could literally barely put two words together that were concurrent. [SPEAKER_01]: So it was like, it was cool kind of, [SPEAKER_01]: I do wish that back when I was doing that, like we had the AI models that we have today, I'm like, oh my God, I could have done so much cooler stuff. [SPEAKER_01]: But it's just funny to see how bad those models were, even just a few years ago, compared to what they are now.

[SPEAKER_04]: Yeah, now it's creating books, deals, and it's a secret amount. [SPEAKER_04]: I create images with Chet's C now. [SPEAKER_01]: Yeah, it's crazy. [SPEAKER_01]: It's insane. [SPEAKER_04]: Yeah. [SPEAKER_04]: So, okay, so you're working, you did just start up.

[SPEAKER_01]: Yeah, we got a choir, so I was in Berkeley with my two co-founders, and we got a choir by a one of the DCs, like largest public affairs firms, and these firms, they like to analyze and monitor politics and stuff like that, so that's why they acquired us, and that's what brought me out here to DC, so over the DMV.

[SPEAKER_04]: Okay, so you out here, you know, from Cal, you from Cal, you hear, and this is when you start making the transition to the government, so you got into the government, um, [SPEAKER_04]: Of course, you already been working with AI, and that was a good mix. [SPEAKER_04]: Can you kind of explain like, ANI, what is ANI? [SPEAKER_01]: So ANI is, there's different types of artificial intelligence. [SPEAKER_01]: There's like artificial narrow intelligence.

[SPEAKER_01]: So that narrow AI, there's a general artificial intelligence. [SPEAKER_01]: And then there's what people are talking about like artificial super intelligence. [SPEAKER_01]: and that's where things are kind of going. [SPEAKER_01]: On the narrowest side, that's what's been really the state of the art for the past few decades or especially the past decade in national security, where you would be able to use AI for very specific tasks.

[SPEAKER_01]: Like, you know, Wally, the movie Wally is one of my favorite movies. [SPEAKER_01]: Like, the way I describe it is like the, you know, Wally, the robot, like making that, [SPEAKER_01]: cube of trash, does that really, really well? [SPEAKER_01]: You can do it like 10 million times. [SPEAKER_01]: That's like narrow AI. [SPEAKER_01]: And that's just security.

[SPEAKER_01]: That means like using computer vision to analyze satellite imagery, pick out like a truck from a photo, things like that. [SPEAKER_04]: Okay, so that's where we're at right now. [SPEAKER_04]: And I think is it the same thing as general general AI? [SPEAKER_04]: Or is it two different things?

[SPEAKER_01]: So there's general AI in generative AI. [SPEAKER_01]: They all sound the same, so they need some to do a better job with branding it. [SPEAKER_01]: But generative AI just literally refers for the ability for AI to generate text, or code, or images, things like that. [SPEAKER_01]: When we say general AI, what that refers to in Google's definition of it is AI that is able to do any kind of intellectual or cognitive tasks that a human can do.

[SPEAKER_01]: So if you're sitting at your computer and doing paperwork all day or translating stuff, [SPEAKER_01]: That all of these different tasks, if that would be considered AGI, and there's a big debate right now as to whether we are either about to achieve AGI tomorrow or in like 50 years. [SPEAKER_01]: I'm more, I think we're nearing the corner, we're going to hit that like in the next couple years, but some people are like, no, it's 100 years away.

[SPEAKER_04]: Okay, so okay, so we got that understanding now. [SPEAKER_04]: So an I specific, GGI is almost smart in humans, got ASI, which is super intelligent, so we're going to be in sci-fi. [SPEAKER_04]: Got you. [SPEAKER_01]: Yes. [SPEAKER_04]: Okay. [SPEAKER_04]: So the way the government is working, we are in an I. [SPEAKER_04]: How do you think the government can utilize it specifically or where you think it can go with that right now?

[SPEAKER_01]: So, on the narrow side, I would say, up until over the past decade, there's this really big, probably one of the largest AI projects in, at least in the department of defense, something called Project Maven, and that's been using computer vision to analyze, [SPEAKER_01]: satellite imagery and drones in the context of Ukraine conflict, that's been really huge, drones have been a really, really important thing in Ukraine.

[SPEAKER_01]: So to be able to use computer vision to detect different objects and tanks and when a bridge is blown up and things like that, that's been huge. [SPEAKER_01]: that and that's still big, but I would say probably over the past two years, now we're rolling in generative artificial intelligence and we're kind of nearing AGI. [SPEAKER_04]: So we're going to learn that.

[SPEAKER_04]: Okay, so so with that, um, [SPEAKER_04]: How do you think it would impact the federal government with introducing it?

[SPEAKER_01]: Like, how is that affecting you? [SPEAKER_01]: I think, I mean, I did a kid in my whole work towards bringing in these generative AI models, which are, I think, going to lead to AGI. [SPEAKER_01]: And it's kind of funny because I think a lot of people at first, like around chat should be T times. [SPEAKER_01]: So I was like 2022. [SPEAKER_01]: They were like, what's the point of, this, like, who cares about this AI? [SPEAKER_01]: All it can do is like paperwork.

[SPEAKER_01]: But the government is mostly paperwork, so to be able to do that 30% faster, that can actually have a huge impact. [SPEAKER_01]: I've been working with DOD, but I also consult different governors and stuff like that.

[SPEAKER_01]: And just wild use cases, like helping people working in jails and correctional facilities, [SPEAKER_01]: out and release it on good behavior and all sorts of things that you just never think of they're able to do so and so you're able to just like make government employees that much more productive. [SPEAKER_04]: Yeah, and then another thing is like being in the government, I hear about this AI and also some, I don't know if you familiar with RPA too, so robotic process.

[SPEAKER_04]: So with those tools, I've been seeing the culture has been changing. [SPEAKER_04]: Like, people are scared they're going to lose their jobs. [SPEAKER_04]: I mean, social perspective on the cultural impact, huh? [SPEAKER_01]: Yeah, I think that's definitely, it's funny because I think a lot of people have that opinion, they're like, oh my god, this is going to take my job.

[SPEAKER_01]: But then for me, I've used these models so much, and I'm like, even if we wanted it to take your job, I don't think it can. [SPEAKER_01]: It really gets just not good enough. [SPEAKER_01]: And so I think my approach has been not just throwing AI at a problem and throwing it in a black box and hoping it's going to work.

[SPEAKER_01]: but really working with the analyst and some, I did a project with security operation centers, actually getting them to use these chat bots and leverage them instead of it just fully replacing them. [SPEAKER_01]: Because that's really, I don't think that's ever going to happen. [SPEAKER_04]: Yeah, and then two of them can tell create opportunities for people to create a startup. [SPEAKER_01]: Yeah, exactly. [SPEAKER_04]: Exactly.

[SPEAKER_04]: But what doing all that, it can create some technical challenges also like, you know, with the ATO process. [SPEAKER_04]: And this is what we're being in now. [SPEAKER_01]: We got the trigger me. [SPEAKER_04]: Yeah, you got the old days that's coming for AI soon. [SPEAKER_04]: Yeah. [SPEAKER_04]: We got all these things like what are some of the technical issues we bring in this in the federal. [SPEAKER_01]: Yeah, you talk about the 800. [SPEAKER_01]: Yeah, yeah.

[SPEAKER_01]: That's exactly right. [SPEAKER_01]: On the technical side, man, ATOs, your audience probably knows this. [SPEAKER_01]: These authorities, authority to operate, are really challenging. [SPEAKER_01]: You really have to prove towards the government, these authorizing officials, or these chief information officers, that your system is really secure. [SPEAKER_01]: So with AI, they're hardware problems that are really difficult.

[SPEAKER_01]: AI runs on [SPEAKER_01]: GPUs you may be your audience may know like in video. [SPEAKER_01]: It's now the largest company in the world bigger than Apple Everyone's yeah or billionaire. [SPEAKER_01]: It's just it's hard out even their worth four trillion dollars. [SPEAKER_01]: It's like It's more than I think Apple and Amazon combined.

[SPEAKER_01]: I don't know if it's combined, but still anyways, it's ridiculous [SPEAKER_01]: But to bring those GPUs into, like, as you know, the different national security, you know, in pack level 5, 6 above that, like all these different places, just logistically as hard. [SPEAKER_01]: So that's a hardware difficulty.

[SPEAKER_01]: And then on the software side, these AOs and CIOs are like, you want to take this AI and just let it go rogue in some national security system, like, [SPEAKER_01]: no way, right? [SPEAKER_01]: So it's, and part of the work I'm doing now with what I launched is figuring out what that path to getting an ATO for some sort of AI system or people call them AI agents, what that can actually look like.

[SPEAKER_01]: Because CIOs and AIOs are authorizing officials, they're going to have to be like very intimately, you know, part of the solution. [SPEAKER_04]: Yeah, because you can, yeah, that's a good idea, actually, to help create AI just to help running a lot of this stuff. [SPEAKER_01]: Yeah. [SPEAKER_04]: But it's just the trust factor. [SPEAKER_04]: It's like, can we really trust AI?

[SPEAKER_01]: Right now, no. [SPEAKER_01]: And the reason why is because we don't have any way to test it. [SPEAKER_01]: And that's a problem. [SPEAKER_01]: Because if we don't have a way to test it, then you can't guarantee, you won't do, and all sorts of really bad behaviors. [SPEAKER_01]: So with what I launched recently, we're calling it Gov Bench stands for Benchmark, like a test, we're literally creating mechanisms to test AI in national security domains.

[SPEAKER_01]: Hey, in this scenario, what would this AI model do? [SPEAKER_01]: And that's kind of just the beginning. [SPEAKER_01]: It's going to take us a long time to get to that full, like ATO, where everyone feels comfortable, but this is kind of like the building foundation. [SPEAKER_04]: I'll see you building it over time. [SPEAKER_01]: That's right. [SPEAKER_04]: Yeah, okay.

[SPEAKER_04]: Yeah, because and then also to one that kind of talk about like even with a generative, uh, you know, it goes off the LMMs and large Yes models, um, I've been introducing someone's like nipper GPT and all these things and they're not good And I'm predictable. [SPEAKER_04]: They don't know what you're going to get like what I think was the issue is to fix those That's a issue people have with those right now [SPEAKER_01]: Yeah.

[SPEAKER_01]: I'll say these large language models are very powerful, but they weren't intended exclusively for like military use, right? [SPEAKER_01]: So if you're using it to generate a poem or, you know, you know, beauty less passive aggressive in an email, it's great, which maybe that's a lot of government work.

[SPEAKER_01]: But [SPEAKER_01]: If you're trying to do other things like supply chain, optimization, and the South China Sea, like the Large Language Models weren't trained for and on that, and so, to your point, like in order to make the models better, with Gug Bench, we call it AI Government alignment, making the AI models work better on like these really random, [SPEAKER_01]: government tasks, military stuff, even things like tariffs, right?

[SPEAKER_02]: If you were good to IRS, yeah, exactly. [SPEAKER_01]: And it's like the most boring stuff that no one would ever care about, but like that the government runs on. [SPEAKER_01]: So. [SPEAKER_04]: Okay. [SPEAKER_04]: That's it. [SPEAKER_04]: Okay. [SPEAKER_04]: So we're, we're going to figure that out. [SPEAKER_01]: Uh, and so hopefully. [SPEAKER_04]: This is like, it's all over the place right now.

[SPEAKER_04]: And as she was startups there, a lot of VC companies are really trying to get into this space. [SPEAKER_04]: And they don't understand the government, man. [SPEAKER_01]: That's exactly right. [SPEAKER_01]: That's kind of my biggest frustration is people either know AI and they're kind of AI nerds. [SPEAKER_01]: I come from Silicon Valley, so I know a lot of them, or their government nerds out here in DC, but very few people know both.

[SPEAKER_01]: And so I think that's what we're trying to do with Gav Bench is bridge that gap to them. [SPEAKER_04]: So, so I know you worked at a deafcon. [SPEAKER_04]: You helped out with red teaming and I know you super proud. [SPEAKER_01]: Yeah, yeah, yeah. [SPEAKER_01]: I'm like, okay. [SPEAKER_04]: So, so yeah, you work with, um, we do some rain too. [SPEAKER_04]: Like, what type of vulnerabilities you see when you help out deafcon? [SPEAKER_04]: If you're able to say that.

[SPEAKER_01]: Yeah, um, yeah, it's a public event and a lot of the vulnerabilities have been disclosed by now. [SPEAKER_01]: So, red teaming AI, red teaming, we've helped over the past couple of years, basically trying to figure out, like, are there instances where these things like chat should be tea, when you use them, can you inject things into that prompt? [SPEAKER_01]: that will make them do bad behaviors. [SPEAKER_01]: So things like that is really what has been explored.

[SPEAKER_01]: And the idea is, as you know with red teaming, you identify these exploits and then you go and you work with the companies and they try to go patch them. [SPEAKER_01]: So that's been great. [SPEAKER_01]: My thing has been, and we're working on with GovBand. [SPEAKER_01]: We actually discovered a vulnerability that we're disclosing this week to the National Security before we go public next week, but with large-ing which models and Chinese models.

[SPEAKER_01]: But the hard part is you need to have AI red teaming for like military and government use cases specifically. [SPEAKER_01]: Like it's not good enough to just be like, oh, you can inject a prompt. [SPEAKER_01]: In our case, what we identified were like, [SPEAKER_01]: Catch you VT for Military Use. [SPEAKER_01]: And if you add certain words, it can go really south very quickly.

[SPEAKER_01]: And so that's what we're trying to advocate for, is like red teaming these systems specifically on like military domains. [SPEAKER_01]: Because that's nipper-tpt, you know, one of my best friends is who founded it calling out the Air Force Research Lab. [SPEAKER_01]: They now have, I think they're approaching 700,000 users, which is I think a third of the whole US military. [SPEAKER_01]: I mean, it's just a lot of people.

[SPEAKER_04]: Yeah, because the biggest thing I'm worried about is injections, and are those somethings that can injections, or... Yeah. [SPEAKER_04]: Because people say they can mimic AI, like, say, sometimes, they can get an AI pretend to be somebody, but you don't know. [SPEAKER_04]: Is that a thing? [SPEAKER_01]: Yeah, and there's something people call them sleeper agents, which is this idea.

[SPEAKER_01]: It's kind of like those, you know, CIA, spy movies, where, like, you say a word, and they, [SPEAKER_01]: come alive or something. [SPEAKER_01]: I guess like in those Avengers movies, it's which sounds crazy, but like there's a bit of a lot of research that like the AI labs have published about being able to say certain phrases to these chat bots and then all of a sudden they [SPEAKER_01]: are nefarious or trying to do bad things.

[SPEAKER_01]: And right now, they don't have access to anything, so it's just research. [SPEAKER_01]: But you can imagine, imagine we have that connected to a military system. [SPEAKER_01]: And you activate a secret spy, that could go really pad really quick. [SPEAKER_04]: Yeah, then too, even with these AIs in the government, that's why they're not full on working property because they don't want to put the data out into the wrong place exactly. [SPEAKER_04]: It's very basic.

[SPEAKER_04]: Yeah, very many. [SPEAKER_04]: They can barely write a email, so. [SPEAKER_01]: Yeah. [SPEAKER_01]: No, I know. [SPEAKER_01]: Very kind of email. [SPEAKER_01]: That's why when people ask me, like, I go on talks or whatever, and they're like, [SPEAKER_01]: The first thing people are like, oh, how are you using AI in the military? [SPEAKER_01]: Everyone thinks we're talking about like lethal systems, you know, generative AI, chatGBT connected to like some kinetic drone or whatever.

[SPEAKER_01]: I'm like, [SPEAKER_01]: people are using it to write emails, and they're barely doing that, and so we have a long way to go.

[SPEAKER_04]: Do you think, like, AI, like, get to the point, because this is what a lot of people include in myself is worried about, you think it can take, do you think it can take, like, it's so job, like, all the jobs, you think at some point when you get to the ASI or super tells the level, do you think it is possible, or do you think you always need somebody there?

[SPEAKER_01]: I think we're always going to need somebody there. [SPEAKER_01]: I just think there will be more tasks that, so I look at it like, um, [SPEAKER_01]: Artificial general generative intelligence is largely being helping software engineers and people said, oh, software engineering is going to be an outdated profession, but over the past I would say three years now engineers move quicker, but it's not like they're being replaced.

[SPEAKER_01]: Now they can just write more code and develop more products. [SPEAKER_01]: I think it's going to be a pretty similar roll out for like, like isos or is like all [SPEAKER_01]: they'll realize, okay, well now I can just do more ATOs or I don't know, you know, I can just move faster, but not like fully replace them. [SPEAKER_04]: Yeah, like who is that when an RMS patient hurts, I'm here at some point E-master or like you have a parmified.

[SPEAKER_02]: Yeah. [SPEAKER_04]: Like, if I could get something to write documentation, man, it took me six months to write SSP. [SPEAKER_04]: Yeah. [SPEAKER_04]: So, yeah, I think you can use it more as a tool. [SPEAKER_04]: I do think if you're not adapting to these tools, you might be left behind. [SPEAKER_01]: That's, the someone who has a great phrase is, you won't be replaced by AI, you'll be replaced by someone who knows how to use AI.

[SPEAKER_01]: And I think that's perfectly summarizes it. [SPEAKER_01]: Like, you should be leveraging these tools. [SPEAKER_01]: You're going to be more productive. [SPEAKER_01]: It's kind of funny, some software engineers. [SPEAKER_01]: For some reason, refuse to use AI chape bots to help generate code. [SPEAKER_01]: And I'm just like, okay, then you're just going to be slower. [SPEAKER_01]: Like, I don't, you know, you're kind of left behind.

[SPEAKER_04]: Okay. [SPEAKER_04]: So Glenn, how do you like, if somebody's in the government right now, what are some things you suggested to kind of learn to stay on top of things? [SPEAKER_04]: Did you learn the general VA or should you learn code? [SPEAKER_04]: What would somebody do? [SPEAKER_01]: I would say, and I always recommend people talk about AI a lot and it's a scary thing.

[SPEAKER_01]: I'm like, just go use it, just try it out, go to chatchvt.com right now, experiment with it, see what it's limits are, see what it's really good at. [SPEAKER_01]: The DOD has things like nipperGPT and there's other AI chat bots that are coming online, like on the different. [SPEAKER_01]: a nipper and separately all the different DOD networks.

[SPEAKER_01]: So first I would just say that go use his chatbots and experiment with it. [SPEAKER_01]: What we also published a couple of days ago, I literally just published a like seven steps to prepare your organization for it. [SPEAKER_01]: Yeah, I'm gut bench. [SPEAKER_01]: Yeah. [SPEAKER_01]: And so it's actually like, I made it into a worksheet so people can like fill it out.

[SPEAKER_01]: And the seven steps, you know, largely touch on like go talk to your AO or your CIO and understand what their concerns are. [SPEAKER_01]: Because you're going to have to get to authorize. [SPEAKER_01]: So you need to know that upfront.

[SPEAKER_01]: Go figure out where your data lives, is it a Microsoft SharePoint, is it a some Palantir thing, go right down the different tasks that your government organization does, and break it down because you're going to want to try to use AI and these different things.

[SPEAKER_01]: So these different, like, I think what I'm trying to do with Govbench, [SPEAKER_01]: is make it less intimidating to start using AI, not viewing it as some like crazy foreign thing, but as something that you could actually integrate into the your day-to-day work. [SPEAKER_04]: And that's a good business too, so we're going to touch on it one more death later, but that's a good business.

[SPEAKER_04]: Just giving an information, just that can decent a time, because like for example, like an exchange, you familiar with them? [SPEAKER_00]: Yes. [SPEAKER_04]: I want to, I'm a client around with XAs, and I'm like, how do I get them to buy this? [SPEAKER_01]: Yeah. [SPEAKER_04]: You can say that's more time or money. [SPEAKER_04]: So I'm like, that's the thing is just talking to them, getting them educated, helps lessen them, and make them understand it more. [SPEAKER_01]: Exactly.

[SPEAKER_01]: It makes it less of a crazy AI, scary thing, and it actually, you know, it's like the rubber meets the road. [SPEAKER_01]: It really makes it real for them. [SPEAKER_04]: Okay, okay, so okay, so do you do you think there's gonna be more pushback on the government because like for example I could talk to certain people and they keep pushing back.

[SPEAKER_04]: I don't know what you use But it's gonna yeah, or you think AI's gonna eventually take over the whole government because you know with software in the government right now is very outdated Do you think don't read read update everything?

[SPEAKER_01]: That's what I'm hoping for and we're already starting to see different parts, the different research labs in the DOD, like Air Force Research Lab Army, DARPA, like they're using AI to actually upgrade essentially their software bases or their software, the code that's written from older programming codes, like Copa and like these very old things. [SPEAKER_04]: goodness. [SPEAKER_01]: I think that was from before I was born.

[SPEAKER_01]: Yeah, but now we can use LLums in generative AI to convert that to C or Python or like whatever really more modern language is, and that's a huge thing. [SPEAKER_01]: It's going to take time and it's not like a one-click solution where translates everything perfectly, but it can make that translation process like that much quicker. [SPEAKER_04]: Okay, and also, and I'm just a question for myself, I'm trying to get into certain AI contracts.

[SPEAKER_04]: So, what would be the best thing to go about getting onto those contracts? [SPEAKER_04]: Like, do you want to break it? [SPEAKER_04]: Do you want to break it to your team? [SPEAKER_04]: Do you want to go look who's hiring? [SPEAKER_04]: Or how would I know how to get into that space so you can be up to date on what's going on? [SPEAKER_01]: That's a great question, actually.

[SPEAKER_01]: the different ways that you could almost inject AI into people's workflows and like the day-to-day work that they do, and then work your way back from that and be like, okay, let's go figure out the contracts that surround that. [SPEAKER_01]: Maybe you could be a subcontractor on some like larger system and be like, hey, the value I bring is I'm going to add layer in AI to make this super annoying. [SPEAKER_01]: What's a good example?

[SPEAKER_01]: Maybe it's something in the [SPEAKER_01]: I don't, we, okay, you know the 28.75 form, can someone please like solve that because that has to be the most annoying thing. [SPEAKER_01]: I've probably sent a thousand, 20.75s and I still don't have access to anything. [SPEAKER_01]: So that's like a perfect example of like someone could use AI maybe to streamline that and you know, get a contract from that.

[SPEAKER_04]: So when you go on that there's a lot of, because even my exact year, like somebody can integrate that in the government, or create a government version of that. [SPEAKER_01]: Yeah. [SPEAKER_04]: I'll help you manage. [SPEAKER_04]: That's a great idea. [SPEAKER_01]: Okay, you know what I mean? [SPEAKER_04]: Yeah, you can actually go ahead, yeah. [SPEAKER_04]: So for more of you, there's a lot of opportunities. [SPEAKER_04]: There's a lot of jobs. [SPEAKER_02]: So much.

[SPEAKER_04]: You start your own company. [SPEAKER_04]: Yeah. [SPEAKER_04]: You can start your own company. [SPEAKER_04]: Maybe you can educate them all, and yeah. [SPEAKER_04]: So what you're kind of explaining is, this is kind of a boom right now where everybody needs to get on, right? [SPEAKER_01]: Exactly, I think it's the reason why I'm so excited about it. [SPEAKER_01]: I've dedicated my whole career to it.

[SPEAKER_01]: AI can really, when we talk about artificial general intelligence, it can be applied in all of these different domains. [SPEAKER_01]: It's not going to solve everything. [SPEAKER_01]: But from a business perspective, it's like, you know what I mean? [SPEAKER_01]: It can be layered on top of like every single type of task, whether you're cybersecurity, you do logistics or whatever it may be.

[SPEAKER_01]: So, [SPEAKER_04]: Yeah, because even I was looking at why a commentator, they're trying to get to the point where a lot of these consulting companies, they're looking for companies that literally just consult, to have an AI consulting company out to them. [SPEAKER_01]: Yeah. [SPEAKER_04]: Which will save a lot of, but I still think you need somebody in person. [SPEAKER_04]: I don't know, yeah.

[SPEAKER_01]: Yeah, I think that's exactly right. [SPEAKER_01]: And I actually think people say the consultants are going to be the first job that's like automated, but I actually disagree because I think with like these kind [SPEAKER_01]: They kind of bring a, they have a lot of different experiences and can bring context that like AI is just not exposed to, especially in government, like government or national security, they don't make this stuff public, right, you know?

[SPEAKER_01]: That's so. [SPEAKER_01]: Very few things. [SPEAKER_01]: Exactly. [SPEAKER_01]: And so naturally, if these AI models, we talked earlier about how they're trained on public information, if that information is not included in their training, then they're not going to be good at it. [SPEAKER_04]: Okay, so what are you doing to take advantage of this? [SPEAKER_04]: Kind of you got glove bench. [SPEAKER_01]: Yeah. [SPEAKER_04]: Whatever things you're doing.

[SPEAKER_01]: So yeah, we're primarily focused on glove bench. [SPEAKER_01]: So we brand it. [SPEAKER_01]: We call AI government alignment. [SPEAKER_01]: The idea of taking AI and making it work for like, [SPEAKER_01]: either really boring or just like general government use cases, because like I said, these AI models aren't trained or optimized for governments. [SPEAKER_01]: We're trying to use it like Asksage and NipperGBT. [SPEAKER_01]: They use the same models as like chatGBT and so forth.

[SPEAKER_01]: But so many times people use those chat bots, and it'll say, oh, as an AI model, I'm not authorized to do this military thing, because these AI companies are kind of nervous to engage in that. [SPEAKER_01]: So what we're doing with Gov Bench is like trying to mend that gap, sort of, bring the AIs, as smart as AIs talent with people in government. [SPEAKER_01]: Oh, yes. [SPEAKER_04]: So you're basically finding the talent and you're bringing them to the government.

[SPEAKER_01]: Yeah, is it like is it like a talent-based network or I would say we're more like a research Lab sort of things research. [SPEAKER_04]: Yeah. [SPEAKER_04]: Yeah, you're kind of like you're doing studies yourself and you're bringing it's your suddenness of the government

[SPEAKER_01]: Yes, so we're technically actually a non-profit just because it makes it easier for us to work with vendors and the government side. [SPEAKER_01]: Honestly, what I'm calling myself, so I work for the Chief AI Office, like under the Secretary of Defense, up until I left. [SPEAKER_01]: And what I found, that was probably the biggest Chief AI Office in the world. [SPEAKER_01]: I mean, in terms of the budget, and it's like a thousand people.

[SPEAKER_01]: But AI moves so quickly that these Chief AI offices in government, they just cannot keep up. [SPEAKER_01]: So that's what I view God bench as, is almost like the chief AI officer, chief AI officers have actually told me this, they're like, you're probably going to do the work that we want to do, but just can because we're moved to slow. [SPEAKER_04]: It moves like molasses, you know, so.

[SPEAKER_01]: It's like I feel bad for them because at the end of the day, their government employees and like just because they're doing AI or they have that title doesn't mean they can just like do things overnight. [SPEAKER_01]: And so with what's been off awesome with Gavbench is we just [SPEAKER_01]: We think of an idea, we put it out and it's immediate. [SPEAKER_01]: We don't have to spend years like going through the processes.

[SPEAKER_04]: So it ain't even published a lot of stuff on your website. [SPEAKER_04]: So some people are reading it. [SPEAKER_01]: Exactly. [SPEAKER_01]: If you go to go to go to bench.au, that's our website. [SPEAKER_01]: The next 30 days are going to be insane for us. [SPEAKER_01]: Like we're doing a lot, but yeah, that's where people can learn more. [SPEAKER_04]: So I'm going to take a good bench, I'm going to go in there some more time. [SPEAKER_01]: I appreciate it.

[SPEAKER_01]: Yeah. [SPEAKER_01]: Yeah. [SPEAKER_04]: I need to learn how to prepare, you know, the business person. [SPEAKER_04]: You know, there's always money somewhere. [SPEAKER_01]: Yeah. [SPEAKER_04]: So I got to know what I got to be on top of everything ahead of me. [SPEAKER_01]: Yeah. [SPEAKER_01]: That's actually, now you're bringing up a good idea. [SPEAKER_01]: I'm going to provide more resources on kind of the business opportunities there as well.

[SPEAKER_01]: Just because I know people, [SPEAKER_01]: wants to engage, but it's like, how do you actually is kind of a different story? [SPEAKER_04]: It's like, even again, the government, you have to get an ATO. [SPEAKER_04]: That's one of the reasons. [SPEAKER_01]: Exactly. [SPEAKER_04]: Yeah. [SPEAKER_04]: Yeah. [SPEAKER_04]: Yeah. [SPEAKER_01]: Yeah, original authorization. [SPEAKER_04]: Yeah. [SPEAKER_04]: But a lot of people would know how to even do that. [SPEAKER_01]: That's, I know.

[SPEAKER_01]: Yeah. [SPEAKER_01]: So providing resources, and that's what we're going to be providing over the next month is like a guide, almost like a playbook on. [SPEAKER_01]: Hey, if you want to authorize an HEO for your AI or an AI agent, it's not going to happen overnight, but like here's that path. [SPEAKER_04]: So, and then again, I'm going to give you money too, sometimes, you know, if you present a good idea, they'll give me, I think they need like 250 depending on the agency.

[SPEAKER_04]: I think a 200? [SPEAKER_01]: Yeah, I like through the small business. [SPEAKER_04]: Yeah, yeah, a small business department. [SPEAKER_01]: Yeah, exactly. [SPEAKER_04]: Yeah, yeah. [SPEAKER_04]: I know this, I need to give you some money. [SPEAKER_04]: If you can get through that weightless, you know?

[SPEAKER_01]: Yeah, yeah, there's a great, like, I think, Sever, sorry, the small business, something program, which give out, like $250 to, like, up to a million dollars and, you know, to help you kind of kickstart your stuff, especially if you're doing the government specific work. [SPEAKER_04]: Okay, that's cool, okay. [SPEAKER_01]: I'm on LinkedIn, I guess my website is GlenParm.com, so I usually post everything on there.

[SPEAKER_01]: Post on LinkedIn a lot, maybe a little bit too much, but it's nice to be able to broadcast my ideas. [SPEAKER_01]: And then I actually started a podcast a few weeks ago myself. [SPEAKER_04]: It's a great part. [SPEAKER_04]: It's a great great part. [SPEAKER_01]: I appreciate it. [SPEAKER_01]: It's great. [SPEAKER_04]: Thank you. [SPEAKER_04]: I'm serious about it. [SPEAKER_04]: It was a great part. [SPEAKER_01]: I'm inspired by you. [SPEAKER_01]: And yeah, this production quality.

[SPEAKER_01]: I need to get like this one day. [SPEAKER_01]: But yeah, it's called model citizen kind of blending. [SPEAKER_01]: My background at AI, National Security, trying to make it easier for people to understand. [SPEAKER_01]: And it's been good for me to like start to communicate this stuff and not just like as a nerd, but like to an audience which may not be technical and stuff like that. [SPEAKER_01]: That's super important.

[SPEAKER_04]: So yeah, just just learning about all this stuff. [SPEAKER_04]: Because I was listening to it. [SPEAKER_04]: I was learning a lot just sitting there from the government employee perspective. [SPEAKER_02]: Yeah, exactly. [SPEAKER_04]: Yeah, sometimes contractors view sometimes out of the bag, but you're just doing what you're going to do because you're doing all you can do. [SPEAKER_01]: Exactly.

[SPEAKER_01]: And like the government places so many restrictions on government employees. [SPEAKER_01]: So it's unfortunate. [SPEAKER_01]: Like it's kind of ironic. [SPEAKER_01]: I feel like I have more influence now [SPEAKER_01]: The Pentagon now that I'm outside of it then I was on the outside.

[SPEAKER_01]: I'm like I wish I listened to me when I was there, but okay Like in the Pentagon, yeah, yeah, I mean when you're a government employee I will say so I work for the chief AI office for the secretary defense [SPEAKER_01]: for several years and I think because of how sensitive AI and military is like just the optics around it, at least my organization was very nervous. [SPEAKER_01]: I didn't go speak at all, it was very, very locked down.

[SPEAKER_01]: Um, I think, and I think they would agree with this.

[SPEAKER_01]: That was probably not the right call. [SPEAKER_01]: I think we should be much more transparent about how we're thinking about AI, you know, the future of it rather than just like hiding behind the Pentagon and being like, oh, well, we have some super secret amazing sauce.

[SPEAKER_01]: Well, you realize at least I realized the government employee is, [SPEAKER_01]: like in national security world sometimes they like to project strength and pretend like everything's figured out and you go behind the scenes you're like oh the papers are kept in a coal mine shaft oh okay this is not good literally literally cuts us like in an elevator going down a coal mine I'm like okay all right well

[SPEAKER_04]: you know you never know you might go back might not but what you're doing is great work you know I appreciate it so keep doing it because I'm I'm watching and I've been watching because I was watching you and I was like hey post some yeah talk more about God man I'm like I'm notist I'm like I'm waiting so yeah what is your goals for the next five years That's a good question I think

[SPEAKER_01]: The two main objectives with Gov Bench in particular are we want to maximize AI coverage, AKA like the number of employees who have access. [SPEAKER_01]: And by the way, not just like National Security, I'm talking people who work in correct like jails, you know, which sounds, I know people are like, what, what are they using AI for?

[SPEAKER_01]: Basically, anywhere a government employee, federal, state, or local, school board, all these places, [SPEAKER_01]: If they have access to email and word documents, I think they should have an AI chapel. [SPEAKER_01]: So there's that coverage. [SPEAKER_01]: But we also want to maximize the value of these chatbots, not just to write an email, but to help you process construction permits faster, which leads to more permits being granted, which leads to cheaper housing.

[SPEAKER_01]: I mean, there's a lot of work to make that happen. [SPEAKER_01]: But I would say in the next five years, [SPEAKER_01]: like maybe get to at least 50% of government employees have at least access to these shop bots and then maybe like realize government employees can move 20% faster. [SPEAKER_01]: I think that would be huge. [SPEAKER_04]: These are prompt engineering.

[SPEAKER_04]: I can't wait to see what you're doing because if you're a prompt engineering that's another space like you're really tapping into something that nobody can really

[SPEAKER_04]: yeah really do so and i think like things like prompt engineering i'm like you can only learn it by doing it you know you just have to experiment with it just play around yep but you know how how we can be sometimes we got to kind of show you you know yeah show the art of the possible and then yeah that's true but what is possible i even open a i didn't even know it can make slides and so recently i was like you can make oh slash shows yeah yeah it's really good yeah so what is one quote you want to live with the audience it can be a quote statement anything

[SPEAKER_01]: That's a great question. [SPEAKER_01]: I think one quote I actually really like is, I have two quotes. [SPEAKER_01]: So one of them is I often like to say progress is not inevitable, it's intentional. [SPEAKER_01]: I think especially in government, people just assume that the government's gonna modernize over time and that we have all these different capabilities [SPEAKER_01]: And that's just not true.

[SPEAKER_01]: We just talked about some pieces of paper or kept in coal mineshafts, and yet we're in the year 2025, right? [SPEAKER_01]: So in order to modernize a government or do AI, it's not just going to magically happen. [SPEAKER_01]: You have to really do the hard work and make a come-to-fruition. [SPEAKER_01]: So that's like one quote.

[SPEAKER_01]: And then the other one is, [SPEAKER_01]: I think I heard this on some other podcasts the other day. [SPEAKER_01]: What's obvious to you is not obvious to others. [SPEAKER_01]: And I don't know, for me, that's been kind of how I viewed my experience with AI. [SPEAKER_01]: We're a lot of the things I'm now talking about publicly were like obvious, or I thought was obvious to everyone. [SPEAKER_01]: But it's not, and like, I should be understanding that.

[SPEAKER_01]: You know, not everyone is going to spend time learning about this stuff, and that's okay. [SPEAKER_01]: We all have our own, you know, specialties. [SPEAKER_01]: But, you know, what I can bring to the table is that I do know this domain, and I can communicate it to a broader audience.

[SPEAKER_04]: And keep knowing you know you'll be surprised like even like when it comes to IT people don't know how to do certain things Yeah, even the ATL process, you know they so some people in the government still don't know what the ATL process is I don't know if anyone knows That's how you can sell to the government so it's always something to sell so exactly what you're doing especially in the AR spaces

[SPEAKER_04]: really like changing especially because the federal government doesn't adapt anything and to see there even willing to do it, I'm kind of like surprised that they're even going this road, but I will say it's going to create a lot of opportunities if you're taking advantage of it.

[SPEAKER_01]: I will say one business opportunity that would probably be relevant to this audience is the DODCIO is they're doing a bunch of, they're trying to basically leverage AI in the ATO process. [SPEAKER_01]: So I think there's, I know they put out a bunch of stuff and people are bidding on it. [SPEAKER_01]: But I think there's a huge potential to make money, but not just make money, but like actually that's probably the biggest bottleneck to modernizing the government.

[SPEAKER_01]: So I would definitely recommend people to look into that. [SPEAKER_04]: Yeah, me too. [SPEAKER_04]: I'm going to look at myself. [SPEAKER_04]: I'm going to take a look. [SPEAKER_04]: Yeah, you see if I can use Chad to be co-sum. [SPEAKER_04]: Yeah, we're just going to get MVP. [SPEAKER_01]: Yeah, I believe it's vibe cut. [SPEAKER_01]: Yeah, yeah, yeah, pretty sure. [SPEAKER_04]: So, so Glenn, I appreciate you for, you know, give me some of your time to get this podcast.

[SPEAKER_04]: I really appreciate you coming on. [SPEAKER_01]: Yeah, thanks so much. [SPEAKER_01]: This has been awesome. [SPEAKER_01]: I was nervous enough for at first because I've never been an interview, but this has been a really like a conversation. [SPEAKER_04]: Yeah, we just, we just chilling just trying to have a conversation and figure it out.

[SPEAKER_04]: You know, in the future, we're definitely going to try to see if we can do it again with the things that are doing so if you're willing, hey, you always come back. [SPEAKER_01]: Absolutely. [SPEAKER_01]: Hopefully, you know, a couple of months or a year from now, we've made a lot of progress. [SPEAKER_01]: Check us out over the next 30 days. [SPEAKER_01]: We have some wild stuff coming out. [SPEAKER_01]: I'm wild. [SPEAKER_01]: I appreciate it.

[SPEAKER_01]: Yeah. [SPEAKER_04]: So with that, thank you everybody for watching the Tech World Podcasts. [SPEAKER_04]: Remember if you're on YouTube, like the video, subscribe to the channel, comment down below and also share the video. [SPEAKER_04]: Check out our MF Academy.io if you're trying to learn about our MF and how the process is.

[SPEAKER_04]: Also go on TechWorldPodcasts.com by some merch, support the channel so we can have some of these great guests keep coming back on the podcast and remember everybody, get 1% better every day, peace out, see you on the next one.