Welcome to this episode of the justice team podcast on the justice team network. And today we're talking about artificial intelligence and creating your own large language model or your own chat pod. Some use cases for it. And I'm sitting with the person, Alec Lowi from personal AI that actually built my chat bot. So tell, tell everybody a little bit about yourself and how you got to build AI.
Yeah, sure. Um, well, I think the answer is you. I think, I think that's how I got here was you. Um, when I started working for you, I think it was, it's been almost a year. I did a little bit of work for you and. You know, you were involved with personally I at the time and through that process, I kind of, um, you know, being a math major and going through a lot of computer science stuff at UCLA and I had a very good understanding of how data works and how I worked.
And when I got involved in that project with you, it was kind of like a, uh, an awesome opportunity with this company for them to bring me on and start. Doing some more development work for them.
Yeah, and you know, just to give everybody a snapshot of what we're going to be talking about here today is the use of essentially creating your own internal chat GBT based on your mindset, the way that you work.
And one of the reasons I, you know, partner with personally, I had to build my chat bot for this use case is to give what I think mentorship on scale, but the ability for to have a bot be able to answer questions very quickly, smoothly with your own intelligence, but built off of your own data set, right? Your own, not everything on the Internet, not read it, not, you know, Pornhub or whatever, but like I don't think you use Pornhub for mine. Did you we could okay.
Well, maybe okay That's a little much maybe wiki feet So anyway, we're here from Alex So whatever we went through this process and we'll talk about how we built mine and how other people you you've helped other people build Or theirs is you know, whether they're big lawyers that have a lot of Dad out there on them or big people in the industry. I know there's a lot of famous folks that are creating their owns to help their, you know, help fans on scale some stuff.
So kind of walk us through how, how you built the AI chatbot.
Yeah, of course. Um, so for you, I, I mean, we can talk about your use case among a lot of the other legal use cases that we have. Cause, um, we did started dipping into the enterprise space, I think a little more about four months ago and we've had a lot of success there, but.
As far as building out the models for, uh, the legal industry, uh, we on our platform, cause we have our own UI and everything we, we, I can say kind of has successfully designed a way for people to create these things on their own. And for people to, um, I guess maintain these models from a user friendly perspective. Cause when we talk about fine tuning and training AI's, um, It gets very confusing. It's very hard.
There's a lot of developers that involves and we're kind of trying to present a way for people to just be easily say, I have all these documents. I have all these case files, emails, zoom transcripts, whatever it might be. Um, upload to the model very easily as if you're uploading to Google drive and everything kind of does the rest on its own.
Yeah. And like one of the things, you know, your company is SOC 2 compliant, which is the highest level to be able to have all your, your, your data, private things like this, but whenever you have to be able to essentially get words onto text, to be able to use it, text words to put into the
model.
Really?
Not anymore.
What do you do now?
Um, we do a lot of stuff with image recognition, image in image out. Um, we also do natural language plus image with image out so we can Uh, we're actually doing a lot of stuff for a sports retail business right now. Um, creating product mock ups. So you could think of, you have a 2023 shoe model or something. And you want to create more product iterations based off of that same shoe model. You can upload a picture of it. Um, a prompt describing what you want the model to look like.
And then develop as many iterations You can also do, um, there was a marketing campaign they did for Chinese New Year with a new tennis racket they just came out with. And we built all these images with that tennis racket of all these different people, um, holding it and, and like little teddy bears holding it in like a Chinese New Year background. It's like, it's, it's really cool.
Yeah, and I know you and the team, Suman, the, the, the CEO founder, are going to be at La De Gra this year actually helping people, lawyers if you're there or whoever's attending. Yeah. If they want to build their own, their own personal AI, they could do that at
the workshop at Lauderdale this year. Yeah, exactly. So we're, we're going to put on display a little bit of AI orchestration. And what I mean by that is, um, you have senior partners that exist in a law firm. You have paralegals, you have researchers, you have all these people that work together. Right? Um, and we're going to put on display a way for these AIs to be able to work together.
So that the paralegal has the expertise of not only just their own AI, but also they can refer to Bob's AI, to another senior partner's AI, they can get opinions from multiple people at once, take those responses and develop opinions based off of other AI opinions as well. Just like the
AIs are collaborating to try to get the best answer. Yeah, with
all, with all, um, with all like a human in the loop, obviously, but, um, We'll put it on display there. It's going to be fun. I love it. Yeah. And you
know, just to show some of the, if you're building something based off of you, like I know when you build ours, we had to transcribe every episode I've done a bourbon of proof of justice team podcast, um, you know, books that I've written, um, published articles, all the trial transcripts, deposition transcripts, and to sanitize all those to get them into the model. How, how large of a lift is that for you?
So it's becoming less and less over time. That's a big part of our development process is creating a lot of new training, optimization methods and stuff like that. Um, now I would say when we trained your model, I think the process took a couple of months for us to get all of those files in there, um, and optimize them. And I can happily say we've for the same kind of volume as what you had given us, we've shrunk that time period down to about a week. Um, and now we're able to kind of.
produce these models for these people, um, very, very quickly and even set up real time training automation.
Yeah. And one of the other things that's interesting with this is you can set, um, like a credibility score, like a veracity score to see, will you allow your thing to try to guess the answer or will you set a higher percentage to 15, 20, 50 percent to make sure that it's, it's grabbing it to get the right answer.
Yeah, yeah, exactly. That's our, uh, our personal score threshold. So we have Um, a feature on our platform because we we try our best to Fully prioritize accuracy and put that on the forefront Um, so our personal score threshold Is a way for us to kind of measure The amount of data that was used to create the AI's response.
And if there wasn't enough of your data that was used to create the response, then the message doesn't go through and it'll say, let me sync up with Bob on that or something. Let me ask Bob.
Yeah. And it'll tell you, I could see when people were chatting it, if I, if Bob doesn't know the answer, it's like, well, let me ask the real Bob Simon that one. So, which is, which is good because you don't want it to guess. Also one thing on there is you can, As a user go in and edit the response to edit the memory. Can you explain us what that is and how it works?
Yeah. So, um, the way that, uh, I guess our real time messaging training works is if you're having conversations with people or people are having conversations with your A. I. Um, we have this little plus button next to each, uh, message, uh, bubble. And you can say, I like this message. I want this message to be trained, or I like this message.
You can kind of go through the whole conversation thread and save them to the AI's memory so that if that same question is asked before or asked in the past, then you can have the same response. And you can also make sure that response is what you want it to be in.
Yeah. And it's an ongoing process because you still have to load more documents into its data set as it grows. So like every new show I do, we try to get in a cadence of putting it in. So it continues to learn. Yeah. You know, my use case, I want people to not only be able to listen to our podcasts and shows, be able to just to query the bot for some quick advice. And we see it because I've said, Oh, you're monitoring ours. I'm always very impressed at some of the answers it comes up with.
It's like, wow, that actually nailed it. It did better than I would. And yeah, somebody asked me who was a tech company and he's like, Hey, can you write like a testimonial for me? And I was like, just ask my bot. Cause I want to know what it says. And it nailed it. I love this. I was the one last night with nailed it. I love it. Like I just asked it. Um, Like who was Morrow Fiore? If you don't know Morrow, he's well, let's see what the answer it comes up with. But we have my voice on it.
We also changed it to an avatar. So we'll talk like, I mean, it's here. I apologize, but I cannot reproduce. Oh, we can't do that one. Let's do that one. Let's, we got to do a shorter one. Who is, so we'll ask, what is your favorite whiskey? It should know that one. What is your favorite whiskey? That should know. Yeah. But then the idea is if it gets it wrong, you can edit it, go back and get, by the way, if anybody's listening, ask my chatbot, which my favorite whiskey is.
You can send it to me for Christmas. I'll give my home address out to you at any time. Um, it should be able to get the right one, but yeah, it doesn't, let's
see if It's a long one again. You can just play the messages for a moment. You can hear the voice avatar.
Yeah.
Let me see if I can
do that. I'm just waiting for the voice activation to pop up.
Yeah, yeah.
Let's see. Let's go. See? Summarizing key points discusses the author's love for rye whiskey , particularly whistle pig boss haw. I think it's 'cause it's mine. As well as their appreciation for good bourbons like Weller Booker . It also mentions a legal talk show where the author enjoys sipping fine whiskey while discussing legal topics. So this is, this is my testing my own internal one. So it's summarizing it for me. But if a user's using, it's a different experience. Right? Right.
But you can see it has my voice. Yeah. And it got it right. Um, but you know, the interesting thing with this is what are some of the use cases that, that you're seeing people use to create their own personal ai? We talked about law firms having that super user regardless of your position at the firm, but how are other people doing it now?
Yeah, so that's a great question actually. Um, so, uh, in the legal space, um, one of our customers, uh, an advocate's ed at dia. Yeah, I love that. Yeah, yeah. He, um, he's another good example of a legal use case right now. He's, um, because he works in Hawaii and California, right? So he has two completely different codes of evidence. And he has a persona that's trained on, um, the California code of evidence as well as the Hawaii code of evidence.
And he's able to kind of draw connections between those two codes of evidence in the sense of, he always talks about how he knows California better. So he'll ask his Hawaii one to say, Hey, are there any statutes in here that are similar to what is in California that I can use for X, Y, and Z? Um, so he's kind of able to draw comparisons for that. And, and, uh, he does use it for a lot of like drafting emails. So we can do. Um, you know, live integrations with the email.
So if you receive emails, you can say, Oh, I want my AI to respond to this one. You just hit the star button or a flag and then a draft really up in your inbox with the, with the email response, you
can plug that in now. Wow.
And then we also do he, a lot of, um, legal use cases do involve like the, uh, zoom meeting transcripts as well. So, um, We also have mechanisms where I know a lot of zoom meeting transcripts are titled the same. We can dump all those in there and you can recall them by date. What was, what did I talk about in this meeting last week on September? Whatever, what were the action items? What were the summaries? That's a good call.
You said what you can record all of your meetings and then dump into the AI summaries. Yeah, we have one company that's uses it for the same reason, but for earnings call reports and QB ours. And they're actually able to draw trends because they have so many of these zoom transcripts from over the past from their business and they're able to draw all these analytical trends from years worth of meeting transcripts just from quarterly business reviews. Wow. And data.
Yeah, and then, you know, going back to the, a lot of our listeners are lawyers or aspiring lawyers to people in the legal space, you know, one of the visions we had is imagining having access to great strategy from great trial lawyers or lawyers in their space. Right. Right. And then putting them together in this lounge that you talked about. So you have like this beehive mentality.
It's like, imagine asking like the 20 best trial lawyers a question about strategy, letting them collaborate and come up with the answer.
Yeah. We're going to show that a lot. Oh really?
Oh, I love it.
Yeah. Yeah. We'll show that at Lottie Gras. So we'll have, um, a couple of different AI personas we'll put on display for that. And then. We also have the, the Bob Simon avatar as well. You got it ready. Yeah. Yeah. Well, can you tell you the story about that? No. Okay. So during the, so we set it up where, you know, you have to record like the two minute long training video.
So, um, the responses, we can connect them to the avatar so that if you were to ask it a question, the response from the AI can be the same response that the avatar gives in the video. Right. So it would be you. Um, And the, the, they didn't accept our training video. The first one that, uh, that you sent them, they're like, it has too much, what did they say? Too much profanity. Oh shit.
Well, I did. They asked me to talk for five minutes. So what the fuck am I going to say if I can't swear?
So I had to call the fricking CTO and they're like, Oh, it's fine. I'm like, just push it through. Like, it's fine. This isn't going to come back through. So we made it, it's done now. Um, so we'll be able to show that at
the, at Laudigrass. I'm excited. You know, I always say one of my goals is to be able to do a podcast or an interview and not actually be there. And be in fricking Hawaii or something. Be in the Maldives in my underwear. I always say, you know, well, can you show it? Cause I know you built a fun, like a funny Bob, like with my sarcasm, cause personas, the ways that I answer things on social media, the way that I talk. Do you have some examples? You can open your computer.
Cause we, we had to, I have had some people testing it and it was becoming so sarcastic. It was a little mean. I was like, we got to go back to the professional one for a little while, but it was really funny and really good. Like I know people were chatting. It was like, I can't believe it's coming up with these answers on me. I'll
ask it. Uh, can you tell me about Brad Simon? Brad Simon? Oh, you mean my little brother? Yeah. I can tell you about him. He's the co founder of our firm, the Simon law group. He's also got his hands in justice HQ. Smart kid I'll give him that Graduated from George Washington University in Southern Law School. He's got a knack for real estate litigation rarely loses emotion But don't tell him I said that his ego is big enough already Brad's got this whole tech thing going on too.
He's working on some virtual tabletop role playing game called Ravel Wow, sounds like nerd stuff to me, but hey, whatever floats his boat
That's good. I mean, that's how I would probably answer that question. Yeah. Yeah, we teach little memories tweak it a little He's technically my little brother by six minutes, but we would put twin or womb mate things like this. Yeah Oh, man, like we asked it about. Um Maria monroy one time she does all the you know, law rank seo and I always tease her that she's mean I think it's funny, but it gave a really mean response. We were laughing because we were friends were testing it at the same
time Yeah, yeah we had the uh Because we have your sms number right we're able to connect the phone numbers to these personas and You Yeah, so how do people, how do people test the Bob bot? It's bob. persona. Yeah, yeah, yeah. So you can go to, um, bob. personal. ai and I actually just turned on your public chat page so people can go on there um, and chat with your, uh, your AI bot uh, without having to make an account on our platform.
Oh, awesome. And then that's the one that actually has my voice on it, right? Correct. Yeah.
Yeah, yeah. So you'll be able to get those, um, those voice transcripts. Look, I'm
making it so easy to get deep faked. Awesome. I'm
fucked. I'm fucked. Did you see the deepfake I sent you?
Yeah, I used it at the Cala presentation. Yeah, yeah. He's gonna remove all
my
tattoos or something. Now I'm a Cowboys fan and I'm gonna switch to the defense. Yeah, deepfaking myself. No, but I think this is like, the way the world is going is, imagine being able to replicate yourself, your brain. And one thing, the idea that we had, I think we talked about is what I would love to be able to do. Imagine Loading this up with all of the posts persona of like my dad, right.
And then, so when my, you know, he's not going to live forever, I hope he does, but I would love to have my grandkids have the ability to ask pap, pap, a question and have an answer as him. Right. Um, and have an avatar of him to be able to ask him those questions. I mean, imagine how amazing would be able to talk to your, Your grandpa for your grandma right now. Yeah, you know,
yeah knowledge preservation. That's kind of what our value proposition is, right? I mean, you have a lot of senior partners that retire from law firms every day And that's kind of the whole thing imagine having the ability to still Harvest the knowledge of a senior partner that retires and being able to keep their philosophies flowing by the law firm
Yeah. And, but in order for you to do that, you have to have uncorrupted memories. I always think about Harry Potter when they take the memories out and distort them, right? You got to be able to have somebody in there that's actually keeping true. The, the archive of the memories and getting them in there. But yeah, if you want to do with it for your, your parents or grandparents, you would just need. I'll want them to either talk a lot or type a lot or interview a lot to be able to get that.
Yeah,
yeah. I mean, you can, you can also, um, get, you know, whether it's emails, just things that they've written. Emails, books, um, posts, whatever it might be. I mean, that's really all we need. We, it's, it's getting to the point where we need less and less training data to be able to talk like you.
So can it extract
all of my social media posts? Yeah.
Oh,
that's,
we should add that to mine because it'd be We have, we have a lot of things we've got to add to yours. I'm excited. I mean, I would love to create this for like my parents. I think it'd be awesome. Or you imagine being able to talk to your mom when she was 25 or 30, right? If you're doing that, right. It'd just be awesome. Can I see mom at 40 years old and ask her a question? You know, that'd be cool.
And then we, uh, we, we made you like the, we have like the professional, um, Bob bot persona. And then we have like the, The fun Bob, like alcoholic slash unhinged persona that's unfiltered. Yeah, yeah. So we can, um, I was thinking like while I'm at Lautograph, you know, I can have the computer set up and, you know, people come ask me questions. I'll just type it into fun Bob and he can answer some of the questions. I love it. People are going to die, man.
People are going to be like, like this one, uh, I asked it who, who your other brother was, Brandon Simon. Yeah. And he was like, oh you mean my little brother he thinks he's hot shit because he's a partner at our firm He's a diss.
He's a decent lawyer, I guess and graduated from Thomas Jefferson with some fancy honors, but it's not Harvard He's got a thing for bourbon too, but I doubt he can hold his liquor like I can Spends his free time playing board games with his life and losing from what I hear from it with his wife and losing This is true actually Typical Brandon, at least he got good taste in dogs. Those Bernese Mountain dogs are probably smarter than half our associates.
Oh man, Bob Bot. Bad Bob Bot. But that was pretty funny, that's how I would answer it to his face. Bad Bob Bot. Bad Bob Bot. I've been screenshotting on my Instagram, if you go to PlanetFunBob and see my stories. I've been screenshotting some of the funny answers and when it nails it. Yeah. Mike, something about my wife, whether she likes snakes and it gave a very funny answer. You should look at it. It was really funny. Um, but yeah, so, you know, how do people find you, Alec?
If people want to, you know, are interested in building their own internal personal AI
bot, what do they do? Yeah. Yeah. So, um, anybody can go to personal. ai. That's our website. Um, you can also reach out to us on LinkedIn. We're on LinkedIn. And, uh, you know, if you go to our website, there's personal. ai. There's a page where you can schedule a demo on there. It's personal. ai slash pricing. Um, you can go on there and schedule a demo with us.
Uh, I do a lot of custom demos for people, so if you do schedule one with us, I'll customize and kind of bring these things to life for them, for them to really be able to see. And, um, we're starting to see a lot of traction from a lot of different types of enterprises and stuff, because, um, the use cases with this thing are kind of endless, you know. It's, it's
Yeah. And what, and what do you think's in the hurdle? Cause so much has changed since we built mine. I mean, I'm just, I'm learning new stuff right here. But where do you, where do you think the future stuff is going to go for this?
So personally, I will be running on device soon. Um, we're currently working with a large chip company to make that happen. What do you mean on device? You won't need internet to talk to the AI really. And it can train on all your local data, not your cloud data on device. That'd be insane. All private, all secure. Um, and then we're also, um, working on, uh, a couple of other things with some large, uh, networking companies, the sports retail business.
We're trying to, um, kind of scale this thing in businesses and in law firms because everybody can have a persona. Every team of every business can use a persona and have that knowledge and be a part of the process in a sense of, Um, efficiency, ROI, all of that stuff. Um, we're starting to see a lot of success with, uh, these larger businesses. Education too. We built a fully asynchronous tutoring classroom, no teachers, first through eighth grade, um, doing a lot of cool stuff there.
So yeah, I mean, um, we're kind of just, Kind of just doing what we can with this. Yeah. I'd love
to see it where, you know, in professional sports, imagine being able to use artificial intelligence to predict what the offensive coordinator is going to play. So you can run a defense on it. It's got to be getting there. I'm trying to train my bot right now on an answer. Yeah, yeah. Yeah. We're going to teach you some information about you. So if you're gonna reach out, if you have questions about Alec.
How much he can squat, what his dating life is, what he likes to eat, look like such a huge, defined definition of a human being. You used to play baseball, right? I did, yeah. I pitched in college. And I got into powerlifting a few years later. Powerlifting and pitching do not equate together. Not really. No, it does not, but if you want to reach out to Alec and go to justiceteamnetwork. com to ping the ping us here and we'll be able to reach out to him through there or go to personal.
ai to reach out to him. But I think your email is just alec at personal. ai. com. Yeah.
Yeah. Uh, you can reach me at alec at personal. ai. Yeah. Yeah. Um, I'm open to any emails too. I, I like to, um,
But will you be answering or will be your AI bot? Yeah.
If my AI bot answers, I do have a signature that says, this is from Alec AI. Oh, okay. Yeah. So at least people will know, like, like the message will all look the same and you'll still probably read through the whole thing thinking it was from me, but at the bottom it'll just say Alec AI.
I love that. I mean, even if it's given somebody, yeah.
Wow. So it's not like, oh, this is just some automated bullshit AI message. Like, no, like it's legitimately from me because, um, it's modeled after my previous emails, but it will say.
Yeah, I asked how much can Alec, Alec low our squad and it says not enough that's the answer Maybe he'll get
791 Are too freaking smart man. There's this one time I built this sales chat bot for our website and we're still kind of working on it because You know, we want it to be cool. We want it to be robust This thing is like spitting out memes and contents and context of his response. Wow GIF memes.
It'll be like Oh yeah, personally, I, uh, you know, we, we do a real live data upload and, and, and, and training, um, come visit our pricing page and it'll show like, you know, but don't be like this guy and it'll show like a mean girls meme. Oh my God. Yeah. That's awesome. Or, uh, and then there was one time where, you know, like Rick being Rick rolled me. Yeah, yeah, yeah. So the AI was like, yeah, you can schedule a demo with us.
Um, You know, just visit this link and you clicked on it and it frickin rickrolled I told it. Yeah, I told it to troll people sometimes so she'll show she might rickroll you a few times,
you know, that's I mean Here's the thing about the AI I know when somebody uses AI to write emails or letters or things and it's always too many adjectives and it sounds like You know, you know what I mean, but this is why I like this. Cause it actually has your voice. You can make it sarcastic. You can make it fun, have a personality. So that's cool. Well, thanks for coming on this episode of the justice team network on the justice team podcast. Again, technology AI is all fascinating.
We're filming this in September of 2024. It'll probably be out very shortly, but I mean, Alec, when you come on the show six months from now, it might be completely different.
Yeah, I'm excited.
We might have our avatars do it.
We should we should we should we should do it with like I'll be in San Diego We'll be filming the podcast up here. Oh, that'd be sick. And I'll be in the Maldives in my underwear. Thank you for listening