Luke: you’re listening to a new episode of The Brave Technologist, and this one features Jimmy Secam CTO of Just Win, which is an AI infrastructure that empowers small businesses, defined and win government contracts. Jimmy holds a PhD from the Machine Learning Lab at the University of Central Florida where he specialized in large scale privacy, safe neural networks, and previously served as the VP of Ads and Premium Services here at Brave.

Luke: During this time on our team, Jimmy helped launch Brave’s global private ad platform, led the development of premium functionality for the web browser and managed initiatives like Brave News and Brave Talk. In this episode, we discussed building solutions to help small businesses win government contracts his approach to open-endedness in AI and why that matters.

Luke: Balancing innovation and privacy as a builder and G PT five, and its predictions for its future impact. And now for this week’s episode of the Brave Technologist.

Luke: [00:01:00] Jimmy, welcome to The Brave Technologist. As a brave technologist or formerly.

Jimmy: Thanks for having me. I’m super happy to be here.

Luke: Yeah, likewise man. It’s, it’s great to catch up

Luke: let’s kind of level set a little bit for the audience, maybe folks who haven’t, who aren’t as familiar with you.

Luke: Like what kind of drew you into machine learning and, and how has your perspective evolved since your PhD days? in privacy, safe neural network.

Jimmy: Yeah, so I mean, I’ve always been, you know, pretty interested. Like I was this kid, you know, ai, you know, robotics kind of thing. And loved also sort of the high performance computing aspects.

Jimmy: I built like clusters and cluster computers and high school, and I just, I like, I loved it. And that was, you know, one of my first jobs at when I went school here at UCF. one of my first jobs was build a, you know, high performance cluster of computers and we called them beowulf’s in those days.

Jimmy: It’s like very, that’s awesome. Ancient [00:02:00] terminology. But, I love that I used to say we would build them to I faster than any other computer on campus. ‘cause that would be like, okay, what do you do with these, you know, super cool computers now and took that. You know, even interested in the undergrad.

Jimmy: Like took that and turned it into, okay, yeah, well what if we took these, you know, super high performance computers and we’re running. Neural networks on these, you know, had a deep interest in neural networks. This was the early aughts. and then, yeah, and that actually sort of went into my, my grad work where we focused on these yeah, like large scale, parallel neural networks and I mean, it’s, and it’s very common now.

Jimmy: but kind of, kind of back then it was sort of. I remember telling some people sometimes like, oh yeah, we’re gonna take these neural networks or we’re gonna run ’em on, you know, these hundreds of cluster nodes. And sometimes people would ask that, well, why, why would you need to do that? Right, right, right.

Jimmy: It’s like most neural networks at the time [00:03:00] just didn’t really need that and it wasn’t, super fashionable or anything. And so I would be like, oh no, actually you can do these great things. Like you can, if you add lots more data, you can really s squeak out performance and. All these interesting things and, uh, um, and so loved all that stuff and got into, I ended up doing my dissertation around, large scale privacy preserving neural networks.

Jimmy: So, you know, my advisor brought to me one day, oh yeah, this is, hey, did you know you can take all these cool neural networks and lots of individual organizations can kind of run them together? They can keep their data private and compute. Interesting and useful things on the output of the data, that can really help without sort of compromising everybody’s privacy.

Jimmy: And it was just super interesting to me and all the creative ways that you had to, you know, hash certain things and combine certain things and not share certain data and do this analysis. And so that sort of combined with the, [00:04:00] oh, okay, we’re gonna wanna make really large scale neural networks and do super interesting phase.

Jimmy: that was super core interest of mine. And, and it is, I mean, it is so cool that, you know, I remember we, I remember just being a, you know, grad student trying out, some neural network on, funny art map neural networks that are called. And it was sort of, you know, it ran a little faster. It is kind of like, well, would anybody really need to do this at this scale?

Jimmy: And it is, it’s just super cool to see. Yeah, basically 20 years later. Oh, this is, this is like super important. This is super vital battle. It’s wild. like runs a lot of the kind of new interesting piece of the God. It’s wild thinking about

Luke: it. Like, yeah, it’s almost like 20 years ahead of the curve on that.

Luke: And then at the same time too, just thinking about, Sam Alman on Theon telling people not to make, uh, open ai, their therapist, you know, because of privacy. Like the fact that you guys were thinking about privacy even back then was pretty awesome too.

Jimmy: yeah, because it was very clear that to do really interesting stuff with [00:05:00] it. You needed to have access to a lot of the data, but of course it was like, okay, you’re actually trying to use technology to solve this social problem too, of saying, yeah, people aren’t gonna do that. They’re not gonna wanna do that for a lot of reasons.

Jimmy: Okay. What can you still do in that case?

Luke: Yeah. Well, what kind of inspired you to, take on this, uh, the role of CTO at Just Win and, what problem are you most excited to solve, when you joined and, uh, and how much of that revolves around, user first thing?

Jimmy: Yeah, so what I was doing before this, it was a social media startup called Maven.

Jimmy: And we, you, you guys know, we, we advertised on Brave a Lot. Yeah, yeah. It was awesome. And, uh, very interesting and it was, it was trying to be this, a social network where AI was very core to it and help people sort of discover interesting things in ways that the traditional sort of machine learning recommenders couldn’t, and find all these great serendipitous things.

Jimmy: And, you know, I learned the valuable lesson that social media [00:06:00] is hard. It’s sometimes you have the juice, sometimes you don’t. And so as that was winding down, I was trying to figure out, okay, well what, what could I do next? And met the, co-founders at, Just Win. We, we were called Breeze, RFP at the time.

Jimmy: And they, and, Ted and Connor, and First they were super great guys. Just super interesting to talk to, incredibly proficient at kind of marketing slash operations. So just first it’s just exciting to work with people you feel are just, oh, these, these guys are good, or what they do. Yeah. Yeah.

Jimmy: Totally. And second, I had talked to a lot of people, you know, in, in that kind of year of just getting ideas from people. It’s like, okay, what, would make sense for you to do next kind of in, in this environment? And. One of the threads that started to come together was, you know, what of all the cool, fun, flashy looking things, you know, a lot of times we were doing a maven with, okay, as you’re [00:07:00] generating images and you know, you’re, you’re talking and you’re chatting.

Jimmy: a lot. I think of the, a the, the value that AI is gonna bring is on those tasks that are just, they don’t seem as exciting. It’s like, and it really comes down to, hey, out there in the world, there’s a ton of documentation and stuff in PDFs and things that people have to pour through and figure out and cross reference and digest and ultimately action on, whether it’s, commercial real estate contracts or government RFPs. ‘cause I had a, you know, I, I didn’t have any background really in, government RFPs. Uh, you know, I, I responded to a couple in the past and been a, you know, part of a government contractor. But it was super interesting to me that like, oh, actually this is sort of very what, you know, just one does is very core to, a lot of, you know, interesting use cases out there that.

Jimmy: Hey, there’s all this, sort of semi-structured things that you have to make sense of. And in some ways, each [00:08:00] of those government RFPs is a beautiful and unique snowflake, but in some ways, you sort of want a lot of the same interesting information out of them. Hmm. That’s actually a really hard problem to, to solve and to, you know, and to figure out how, you know, how do you ingest those, how you break those up, how you make sense of those, in a way that kind of makes sense and structured and, and, and it was exciting in that, like said, after, after a couple years of really trying to present people with things that they weren’t sure that they wanted and different, and David, you know, oh, how about we do, you know, these social things, these fundamentally different ways? it was something that, it was very clear that if you could solve this pain point for people, and a lot of, a lot of these people are, small businesses who just.

Jimmy: Honestly might not even be able to afford to regularly respond to these RFPs without this kind of technology and help because it is immensely labor intensive to come in. Take an RFP, first, find it from anyone of tens of [00:09:00] thousands of state local education, uh, providers or federal. sort of find it, figure out if you can even win it.

Jimmy: We, we have people describe to us that. Yeah, it used to be that sometimes, you know, they’d read through it for a day and it get to the bottom, oh crap, we can’t win this. This is, this is this, this is actually, you know, not for us. After, pouring through 70 or, you know, pages of it.

Jimmy: And so, you know, it was something that could provide very clear, if you could help people, hey. Help me make sense of all these thousands of things coming out every day. Get a very deep understanding of what my business does and what is actually interesting to us versus what’s not. and then help me break this down.

Jimmy: Figure out what I actually need to do. Do I even qualify? Okay, what do I actually even need to do to win this? And then ultimately help me get together this response so I can, I can go win this thing. Because you know, the tragedy I think, was that. [00:10:00] Without some of this, this technology, it would sort of be the usual, you know, all the same usual companies sort of, applying, you know, responding to these RFPs, because they were, they’re just larger and they could afford to, and they had people dedicated and they had a mature process, versus somebody saying, Hey, look, we do great work.

Jimmy: but we just don’t have time to set somebody full-time on this. If you can help me actually, you know, let these places know that I exist and I, and we do good work and we’ve done good work in the past, it makes a difference to, to these, you know, great small businesses who are just, they’re just trying to do, I need great work out there.

Jimmy: And, and we, well, it feels like we need a lot of that.

Luke: I mean, the government too, like I think people don’t really understand or have a appreciation for just how, how huge of a, employment government is with these RFPs, right? Like it’s, it aren’t, the numbers just like kind of massive with, with the scale of like available opportunities to go after across all the different, you know, areas that they, [00:11:00] issue.

Jimmy: Yeah, I mean, it’s, it’s enormous. It’s, a huge piece. I, I won’t, won’t, I won’t quote things I, I don’t know, well, off the top of my head, it’s an enormous piece of the, the economy and again, it’s, when they’re looking for specific things.

Jimmy: And could just be easily lost in that sea of, of like real contracts and real opportunities to be done. So, you know, a lot of these things aren’t, aren’t small, they’re real, they’re large continuing projects, and across every level of government it’s, it’s just, you know, there’s things needed federal level, like states have their, you know, sometimes their own purchasing.

Jimmy: The counties do it. Cities do it. So, yeah, when you add all that up, it makes a pretty significant, measure of, of yeah. Economic activity that you can, you can be involved in, the government can be, you know, an enormous customer.

Luke: How is just win, uh, leveraging AI like to, uh, you know, help, empower small businesses in this.

Luke: I know you kind of mentioned a few things, but, um, you know, maybe we can unpack that a little more.

Jimmy: Yeah, [00:12:00] sure. So there’s a couple of different parts of it, and I’ve come to appreciate how hard each of each one of these, in the whole process. So, you know, first there’s, there’s finding these things, right?

Jimmy: So, because, the federal database, the SAM database is a little bit more centralized, but when you start to get to state and local and all of these things, It’s just spread across tens of thousands of entities. And so, you know, the way you might have to, initially, the way you might have to find ’em, might be subscribed to, you know, a dozen of these things.

Jimmy: And like I said, cumulatively, thousands of these things can be coming out every day. So sort of the first part is, you know, we have our, we have our customer base that come in. They through a, a pretty easy process. They can sort of generally describe what they do. Like, okay, well we’re, you know, we’re a marketing business.

Jimmy: We’re kind of focused in these areas, and this is what makes a really good RFP for us, and this is what doesn’t. And so. Instead of having [00:13:00] to go just search and track these now or just scan the list, we can, we’re taking each one of those things and, and we can now practically at scale, sort of pick them apart.

Jimmy: Figure out, okay, who’s the customer for this? what is it? if we need to go out and, give more information about it. Okay. Break, you know, break down the, the actual, you know, again, could be a hundred page PDF document. On, Hey, is this something that could fit for them?

Jimmy: So, you know, without that AI element, again, picking it apart, it can be a very hard search, a very hard and noisy search problem. Otherwise, you know where in, but instead it’s like, okay, it’s, the AI is, you know, scanning sort of each one of these and trying to make sense of it for the customers.

Jimmy: So, that’s first party, that’s kind of the, the sourcing. Then, you know, we’re, we’re, we’re sending those along, we’re scoring those for users. ‘cause like in any, you know, cm like environment, you might [00:14:00] have a lot of these, opportunities coming in every day, right? So, a lot of times there’s.

Jimmy: You know, unless you’re in a little niche kind of area. And that can be its own problem of, of, okay, well whenever somebody talks about this, I really want to want to know. But that can be its own problem of, okay, well cool, there’s probably, a hundred of these that I could do what is really in my wheelhouse or not?

Jimmy: And that mm-hmm. And again, that’s something that the AI becomes very good at. Sort of going through and scoring these and saying, yeah, this is, the a customer who’s really in your wheelhouse. This really aligns with the things you want to do. it feels like you can, you should be interested.

Jimmy: You can win this. Mm-hmm. So that’s sort of the second part where AI is really important of focusing that attention. And then the third part, sort of in, in responding, right. Okay. When you have those giant documents, a lot of this is making sure you’re just, you’re compliance, right? That [00:15:00] when you write your response, which itself might be 50 pages, right?

Jimmy: Yeah. Right, right. All of the things that they laid out that you know you’re supposed to do, that you’re doing, and again, through having the AI sort of scan all of that and make sense of it and say, Hey, you know, careful for you, you missed this, or this is an important piece. Then users are able to say, okay, great.

Jimmy: We’re, you know, Hey, I have something similar I’ve responded to in the past. I can do this. Here’s how I make sure that I am, I’m responding to the correct things. And so like finally making that response, whether it’s. Making sure you hit all the items that it said for, or actually helping you to, you know, write some of that content based on great things you’ve done in the past.

Jimmy: that the AI is pretty, fundamental to, to being able to, to respond to that, you know, in an efficient way for you Small.

Luke: That’s awesome, man. It sounds like you guys are kind of like combining like a lot of useful use cases into like a [00:16:00] com comprehensive solution in a Goliath of things, right?

Luke: It’s wild, man. yeah. I mean, and and you’ve also mentioned too, like the concept of, open-endedness and ai. can you define what that is and share why that matters?

Jimmy: Yeah, while I was doing my dissertation. My hobby, the thing, you know, that sort of kept me distracted, was working, There was a project, called Pic Breeder that I worked on with a professor there, Kenneth? Kenneth Stanley. and it became a big part of sort of my broader interest in ai and it intersects in, in sort of interesting ways. Like I said, it, it was the ideas in, you know, sort of more colloquially, it’s kind of, you know, open-endedness is when.

Jimmy: These sort of complex systems can kind of grow in complexity and interestingness without bounds. So we talk about, you know, the, the usual examples are like, oh, okay. Sort of evolution to [00:17:00] overall, you know? Mm-hmm. Is this cool open-ended system where interesting things always come out of it, or, sort of, you know, the, the human economy and creativity and scientific endeavor.

Jimmy: Of, well actually, you know, unlike a lot of things in AI that are very constrained, or, Hey, this is, this is the task at hand, and, and these are the parameters you have to solve, you can get very good at these things, but they never become more interesting over time. open-endedness sort of instead deals with, okay, well what about systems that can become more interesting over time?

Jimmy: Hmm. And so, you know, and through these unexpected, stepping stones, you know, Ken Stanley, you know, went for a co-founder on, on Maven. He, he loves to talk about stepping stones where. When you really are trying to do, and he wrote, wrote a whole book about it of, why greatness can’t be planned, the myth of the objective.

Jimmy: And in it he talks about actually the research he and [00:18:00] I and, and, and along other people at UCF had had done together where we were working on this pick breeder system and. We would find that when you do, it was a sweat based system where you would select images that were interesting to you, and these images were drawn by neural networks and you would made mutate kind of the more interesting ones and when you found somebody else’s that they had done, you were able to branch that one off and take it in a whole different direction.

Jimmy: And he found, interestingly enough that. That actually got you to more interesting places faster than starting with this objective of what you wanted to be. Hmm. And in the book, which is great, I highly recommend it. you know, that, he sort of goes on to say, actually not only is that really important for these sort of algorithms, uh, but it’s really an important principle that kind of governs our lives too.

Jimmy: Mm-hmm. Where, Life is [00:19:00] this very interesting kind of open-ended system and, you know, can be enormously complex. And when you are trying to solve really, ambitious problems and you know, you’re bee lying straight for that problem, you know it, it’s not always obvious. And a lot of times you sort of grind against that and say, okay, well we’re go, you know, we’re slight.

Jimmy: We’re making this thing slightly, slightly better. And instead when you go to these sort of interesting stepping stones that kind of feel like they don’t have anything to do with where you wanna go, you actually end up where you want it to go faster. So you discover these things, you know, it’s like, oh, and the, you know, the transistor was made, the intent wasn’t to make digital computers.

Jimmy: It was, to do sort of, faster, better, and more power efficient radio things than people discovered, oh, actually you can do all these cool, interesting things for them. So. It’s something that, I have, you know, a, a huge kind of amateur interest in, [00:20:00] and, you know, also, when at the, um, Maven, that four startup, I mean, that was sort of core to how we wanted to design that social network.

Jimmy: But, you know, even, you know, even though that, that that sort of wound out, it’s sort of, It’s a principle, you know, it’s both a principle and an interest that I keep, in the rest of my life, where a lot of times, you know, I find that the solutions to these like really hard problems, and I think a lot in AI and will find lots of these interesting things in ai, you know, will be able to be used in these unexpected ways and get us places that we just, could not have imagined, you know, beforehand.

Luke: No, it’s interesting. I mean, like, and you think about how many times like. You know, like Einstein talks about taking a walk and, and ideas come to mind or, or the shower thoughts. Right. Or, or things where like, you know, throwing up at things in the mix and not being so overly focused on one simple thing all the time can actually lead to like, and knocking some things around, right?

Luke: Like, getting some ideas for things. It’s awesome man. Um, you know, and kind of, um, [00:21:00] not totally related, but, but somewhat, what are your thoughts? I mean, we’re seeing. Kind of a mix of like agents and copilot things, like, what are your thoughts on AI as a copilot versus AI as an agent, and where should the boundary lie from your point of view, in giving systems more autonomy?

Jimmy: Yeah. Yeah. It’s an interesting question and I think it’s been coming up a lot lately. I saw, I saw this great post on X by, uh, Carpe, and, you know, he was, I agree with the sentiment that he would say. Actually, it feels like maybe in the last, even a couple of months, because again, months, months can be lifetimes in sort of new, AI land, you know, that things have become almost overly agentic too fast, you know, he talks about, yeah, yeah.

Jimmy: When you, many of us are using cursor to write code or other, you know, a ai, you know, assistance and, um. It can be, you may want it to, you know, like a lot of people yolo with Claude [00:22:00] code and sort of say, okay, yeah, no, keep going. Like keep pushing this forward. but you know, you made the good point.

Jimmy: It’s like actually sometimes you don’t want that the agent is going too far and you really want to be able to just say, no, no, no. I need, a quick check, or I need a quick answer on this. Like this, this is as much as I need. And I think the. I think, you know, and he talks about like, well, you know, someday these systems probably will, you know, sort of offer that as a parameter itself, right.

Jimmy: Of, of being able to, say, well, you know, how autonomous, how far do you want me to, to, to go with this? Um, you know, because I think it can be, uh. Again, in this sort of effort. I mean, the agenda things coming are very interesting. You know, let me say that. Right. You know, sort of the amount of time, I think it’s very cool.

Jimmy: For instance, the thought of, well, yeah, what if you gave an ai, you know, sort of both interest in things and a very long time [00:23:00] horizon to work on stuff, like if you mm-hmm. If you said, okay, well it’s interested in this and. You know, you gave it six months to build things and build things on top of those things and came back.

Jimmy: That would be fascinating, right? To, to say, okay, well this is like fully agentic. Fully agentic to the point where, it is not really even trying to do my work. It is trying to figure out what it wants and do cool things, you know, that it finds interesting. I think this, it’ll be absolutely fascinating to see what comes out of stuff like that at the same time.

Jimmy: A lot of, you know, I think a lot of success of applying AI over, you know, in real products over the last couple of years has been trying to constrain the process. Right, right, right. Knowing, okay, the thing will do really well on this and different models form differently and you know, whatever. But it will do really well on this.

Jimmy: Okay, let’s get it going on this thing and let’s give these guardrails to it and you know, and I think sometimes in those [00:24:00] contexts, agent-based approaches can sort of just kind of degrade to, well, okay, is this really just, you know, it becomes almost about code organization more than anything versus okay, right.

Jimmy: I truly want this thing to be agentic. I want it to, you know? Yeah. You kind of claw code. It’s like it’s going, you search my code or means to search the web. And, and I think, yeah, so I think the, you know, the line for me is just. It will be different in all those applications and it’s about me as the user being able to express, yeah, that’s.

Jimmy: That’s as far as I need it to go on this right now, or No, I need it. It’s, it’s fine. This, this should go on, you know, for a long time and do something very interesting here.

Luke: Yeah. It’s funny too because, uh, I feel like there’s kind of a perception that, people have around this stuff where, like I’ve been talking to people that have been trying out different genic things in the browser, right?

Luke: And they’re like. Yeah, it’s cool. I can see that it’s doing things, but I could have actually just gone and done it faster myself, you know? And you’re like, yeah, you know, it’s like, [00:25:00] it feels like we’re not quite there yet with a lot of the fit on this. It, it feels like, um, you know, oh, we’re, we’re trying to see what the thing can do, but like, not necessarily that useful yet.

Luke: I, I don’t know. We’ll see how it shakes out. Yeah.

Jimmy: I, I think you’re right because it’s, it’s, it’s sort of the example of, okay, you know, everybody seems to have this dream of. I just tell, you know, I tell the agent to go out and you know, we’re going on a vacation, so we’re traveling and what people seem to try to struggle to present kind of is the like, oh.

Jimmy: And then I came back and an hour later and tickets were booked and things were done. And, And well, not only is that not comfortable for a lot of people, but it’s, it’s just, okay, that’s not, it’s just not what you a lot of people are looking for right now, versus, hey, it may actually be a lot more useful.

Jimmy: It’s like, oh, I’m on a giant, page of tours. oh, okay. Yeah. Hey, can you tell me which one, you know, starts in the af? Can you gimme the one that starts in the afternoon and hits both, you [00:26:00] know. big Bend in bath, England or whatever as okay. Mm-hmm. I can scan through and I can give that to you.

Jimmy: And I didn’t need to, I didn’t need to go off and book in or whatever. It’s just like I bring, you know, I answered a useful question and then I was able to, to put that together. And I think again, it does come back to like. I don’t want, I don’t, as a user, I don’t want to feel like I’m panicked and pushing the stop button.

Luke: Fair enough, fair enough. I think in that vein too, like, how do you kind of weigh the trade offs right now between innovation and privacy? it does seem like there’s some interest coming around on privacy on this stuff, but it does feel like it’s kind of an afterthought still, but I’m just kind of curious your take on it.

Jimmy: Yeah. Well, yeah, I mean, it’s, it’s been pretty core to everything that I, I try to do, all the, you know, throughout my career. So I think that. You know, one, I sort of, I sort of try to, live by that principle of like, okay, if I don’t really know why we need this information, just like, just don’t collect it.

Jimmy: Right. And I think that’s, you know, it’s [00:27:00] been helpful in that because it felt like for a while the really, you know, the running sort of, of, of wisdom around, you know, failing wisdom around this was, was. Hey, you need to instrument a system and you need to, you know, have data collected so that you have everything, you know, you wanna be able to, to track things, make recommendations, and, do all this stuff.

Jimmy: And I think the reality of it is, is like, well, doing correct analytics, tracking and, understanding use behaviors and all that stuff, it’s actually very hard. Yeah. And it’s easy to get steadily wrong. You know, that you could easily look at all that data you collect and it’s like, oh yeah. Now we were supposed to do it like this and now I’m sorry.

Jimmy: It’s like, so I think that it doesn’t really work out as much as people think it should and then there’s a principle I think that’s kind of emerging, but I think is interesting, that I think it’s sort of generally applied to fault before, and it’s now specifically I think applying to LLMs where.[00:28:00]

Jimmy: I think before you could really collect you, you could really think you’re gonna bring a lot more information to bear on, let’s say a recommendation or on a, you know, guy, you know, recommended user behavior or whatever, and think like, okay, well if I knew, if I knew this about them and if I knew, you know, what they were doing here to do that, and then, you know, I, I would be able to make this great recommendation.

Jimmy: A lot of times recommendations can be simple like, so. You know, canonical example on that is like, oh, people are searching for things around noon. A lot of times they’re interested in lunch. Right? Right. We’re able to, it’s an example of a thing where you’re able to use simple and, you know, not very privacy invasive kind of signals to, really do a good job.

Jimmy: And sometimes the simpler recommendation is, the better one. So I’ve always tried to keep that in mind when. Okay. Well even if you had all the information, [00:29:00] what information would you really want to start with on this? Mm-hmm. You know, to, to, you know, a, a, a lot of the way there.

Jimmy: And I think interestingly enough, I’ve sort of seen it mirrored, on LLMs and things like that, where. You know, kind of, you’ll hear people talk about this emerging contextual engineering and how important that is, and it’s like, okay, well that’s providing, the prompt engineering is really important to sort of steering and telling the LM what you want, but this contextual engineering of, hey, you may be doing some rag system or providing them what other contexts around, you know, what they’re doing.

Jimmy: And you can definitely get into, you know, situations where you’re just saying too much. You’re providing more irrelevant stuff and you know, you may not get the same performance that you’re looking, out of from there. So it’s like, okay, just because you have a 400,000 character context window doesn’t mean you should use a token context window.

Jimmy: Doesn’t mean you could use all of it, right? Like Right, right. Oh, sometimes it does pay to sort of carefully choose what [00:30:00] information you had. And so for me, I try to start from those principles anyway of saying, okay, well tho those things will actually steer you to be. Just more private by default and a lot of times come out with a, a better product.

Jimmy: Like why Think back to when we worked at Brave, right. or when I worked at Brave, right? Hey, yeah, yeah. I get it. I get it. Back to when I worked at Brave Worked at Brave and, when we were putting together the ad system and kind of taking that to different, you know, advertisers and marketers and they would ask, okay, well.

Jimmy: Yeah. So you’re gonna do this, this very private ByDesign ad system. Are you gonna be able to give me all these great reports about, the demographics and all this and, and you know, these huge studies that a lot of their, partners had, had typically provided them. And, you know, our line always, I, I believe this was.

Jimmy: Hey, we’re not gonna maybe provide you everything that you used to get, or we’re gonna provide you exactly what you eat. Right? And [00:31:00] in some ways, sometimes that’s just better of like, actually here was the real signal that you needed to make a difference to things. And so, you know, a lot of times that that less is more, that is, you know, kind of aligned with privacy is just better anyway.

Jimmy: Yeah. Um, and I think that again, It’s more important than ever now that people kind of understand like what they shared with who and why and for what. Right? It’s like, well, ‘cause now if you can have agents actually going out and doing your bidding and accessing, you know, your accounts and Gmail and things like that, this is, this is pretty, pretty private territory, right?

Jimmy: And, and mm-hmm. You both want to know, okay, this is exactly what I’ve enabled for everybody. This is exactly what, what we have. And, it’s gonna be into your control to, shut it off if you need to.

Luke: Yeah, totally. if you could make a big bet, what’s one bold prediction you have for the future of [00:32:00] tech that you’re be willing to put a bet on today?

Jimmy: So, let’s see. Um, the thing that’s on my mind lately is, especially with the GBT five release, You saw that happen? they went big and bold on saying they’re gonna deprecate all the other models that the one mall. And, a lot of media pushback from people sort of saying, Hey, no, I, I had actually grown accustomed to the nuance, like I loved 4.5 and I had grown accustomed to it.

Jimmy: Or, you know, what people had, uh, you know, not a lot of people used 4.5, but people who had very particular taste in this. And my prediction is sort of in 18 months, you’ll see sort of more common, consumer level things. You know, maybe from big tech companies or from sort of, very motivated startups and things trying to, you draw the right line around, you know, model ownership, right?

Jimmy: It’s, uh, say saying goes on X you know, not, not your way, it’s not model, you know, kind of [00:33:00] thing, right? Um. And so it’s like, whereas maybe a year ago there wasn’t strong interest in, or, I mean, you know, there, there wasn’t as much interest in sort of device level intelligence, uh, of like, okay, well what, what could I actually run on a phone or run locally or whatever?

Jimmy: And I think I predict you’re gonna start to see some of that converge into. Um, you know, kind of consumer level products where the consumer feels confident that it’s like, okay. And I mean, especially you start seeing interesting things like, uh, well this is, this is an AI companion that is like a core part of my life that I, you know, that I really enjoy being around and, and is important to me.

Jimmy: That people will sort of feel more importance around the ownership of that. And of course that’s aligned with privacy too, you know? Totally. So I think some of those things had become, will become so core to people that in ways we really hadn’t seen over the last 20 [00:34:00] years, you will start to see.

Jimmy: Migration to kind of local device, you know, owned weights, things that you can own and reproduce and, you know, have greater control over as the consumer. Um, where it felt like there were a lot of, you know, there were interest in that over, let’s say the last 20 years. I was like, okay, well I can host my own this and this, or, you know, I can host my own file storage or host my own email or, or whatever.

Jimmy: And it just sort of. Never, you know, there was always an interesting baseline of people who would do it, but you know, wouldn’t go too far and wide. And I think you’ll start to see that become a lot more important to people because like, this is actually quite core to, you know, my life or my business or, or what I do.

Jimmy: And, and you know, I need to be in control of the changes to.

Luke: Totally. No. Yeah, you’re right. It didn’t ever feel like it escaped the garage, but I think now with what we’re seeing, it’s totally like, I, I would totally bet on that. Like, uh, with what we’re seeing now with, with how, just how these things are [00:35:00] operating too.

Luke: I mean, like, so it is so interesting too, like there’s such a scramble to get AI everywhere. Kind of from the C-suites down at, you know, fortune two thousands. I don’t know when the last time where I’ve seen that happen at that kind of a pace. And so now you’re just kind of seeing all this noise, but like we’re also seeing a lot of the, the side effects of that.

Luke: And uh, yeah, I think this local stuff will get a lot of attention, like you’re saying. Yeah. Well, Jimmy, dude, like this has been fantastic, man. I mean, where can people find you online and find, find out more about your work and, and, and follow along?

Jimmy: Oh yeah. So, uh, you can follow me. I, uh, I do converse with people on Blue sky sometimes.

Jimmy: And, uh, at, let’s see, what, what am I? Blue sky. Am am at JSON social. I’m Blue sky. And, uh, excellent. You know, you can, uh, or if, if you’re curious about what Just Win does go to just [00:36:00] win.ai and, uh.

Luke: So Perfect. I’m always happy to talk. Awesome, man. Like we’ll include it in the show notes too, but, um, yeah, dude, thanks Jimmy for coming by.

Luke: This was a great conversation. Love to have you back too. Check back in on things and, and yeah, get more POB. It’s awesome, man.

Jimmy: All right. Thanks Luke for taking the time.

Luke: Awesome. Thanks for listening to the Brave Technologist Podcast. To never miss an episode, make sure you hit follow in your podcast app.

Luke: If you haven’t already made the switch to the Brave Browser, you can download it for free today@brave.com and start using Brave Search, which enables you to search the web privately. Brave also shields you from the ads trackers and other creepy stuff following you across the web.