Techstrong Gang – May 30, 2024

Hello, everybody. I'm Alan Shimmel. Just kidding.

I'm Mike Biard here. We're Textron Gang, and we're talking today about the FCC is mulling some disclosure rules for political ads. We're gonna be talking about AI in the call center.

And then finally, we're gonna talk about gen AI insecurity. Apparently it's pretty easy to tease out some secrets from those Gen AI models. We'll be back in a minute.

You're watching Textron Deck. All right, folks, and we're back with an all star lineup. A lot of folks today happen to be in California.

We have a guest, Lisa Martin, who's with CMO Advisor, an arm of Char Group, and she's outta San Jose. And then Alan and Mitch are traveling. They're at a conference in Santa Clara.

So we have the three folks who got up pretty early this morning for this one. And then finally we have Bonnie Schneider. Yes.

And myself, of course, Mike Beard. And we're going to start talking about this first topic, which is the FCC is mowing AI disclosure Rules for Political Ads. This may not come as much of a surprise, but Lisa, I'd like to get your sense of what is going on here with AI and advertising and videos.

Yeah, we're gonna need to label all these things and say, this one was created with AI and this one wasn't, because frankly, I'm not entirely sure they're, most ads are kinda lying anyway. So what difference does it make you create? That's true.

I think when we see it come into the political arena, it gets really scary there. I think transparency, accountability is absolutely necessary. Now, what the FCC is proposing is not a prohibition of using ai.

They simply want, um, it to be disclosed. And we know that just a few months ago in New Hampshire, there was about 25,000 people that received robocalls sounding like President Biden telling them not to vote in the primary to save their votes. So that's something that fooled a lot of people, and that accountability and transparency needs to be there.

I support it. I'm a fan of ai, especially Gen AI as a marketer. Uh, I host a, a webcast series called Marketing, art and Science.

And I get the opportunity to speak with CMOs every week, and they're all supporters of it as well, as long as it's done purposefully, uh, with responsibility and use for things like first drafts of content to really kind of get the creative juices flowing. So I think that it's an important tool that can really be very helpful, but it really needs transparency and accountability. Yeah.

So Mike, you know, I, I don't disagree with everything Lisa said, but in, in regard to the instant case with the FTC and political ads and saying that everything that's AI needs to be labeled as ai, I'm reminded of the old gun rights kind of thing. If you outlaw guns, only outlaws will have guns, right? I I if you, if you're gonna say, Hey, election rules say that you've gotta label something ai, there will be a certain percentage of people that will follow that.

But the dirty money, the dark money, the unregistered pacs, the foreign entities, they don't play by the rules. What's the sense? And so if they're not gonna play by the rules, should anybody be forced then to play by the rules?

I have a pet theory on this, and I'll run it by Bonnie here for a minute. Okay. Well, eventually, when we not get to a point where there's so much content that is loaded with so much BS that nobody's gonna believe anything, and maybe it would be good for us to be inundated with all this stuff so that we get to the point where everybody just kind of has a cognitive moment and says, all this stuff is nonsense, and just don't trust it from the get go.

'cause I think today we trust too much content. Yeah, I think that's true. And I think with this situation, especially when AI is able to do a deep fake of a political candidate, that's when you're running into trouble.

Uh, I would imagine whoever's getting, uh, uh, imitated. So there could be a solution of doing, I, I agree with Alan. It would be very tough to enforce, and some people will just ignore any rules that are out there.

But it, there could be a way per perhaps, to do a hybrid solution where some AI is labeled and others isn't. Or perhaps there's a way for more education to be out there for people that are consuming content. For example, the more of us that use chat GBT, we kind of recognize when we see it.

So if you see an article that, um, looks like you, you know what I mean, you can sort of get that sense. Not not everyone, and not all the time, but as more people get familiar with, um, ai, it's per, perhaps it won't be, like you were saying, as shocking or, um, will as fool as many people. Mm-Hmm.

Strange adverbs was a dead giveaway for chat GPT, right? Um, Mitchell, is there a technology solution to this problem? Is there things that we can do to kind of, uh, maybe make it easier to identify AI content without labeling it, but, you know, just give people some kind of capability to, to discern if they're interested.

So you're wondering, is there, is there a Shazam of, of AI generated content, right? Like, what is this, right, where did this come from? I'm serious actually.

Ai, I think something like that is probably what we're headed to very quickly. And, uh, I think we're, I don't know, I read too many sci-fi books. Maybe we need, uh, our own detectors for content to say bing, bing, bing.

But then that's all that we'll get is warning labels that this was AI generated. I, I think, well, first of all, what's the, what's the, uh, the rule gonna be if you violate this? Right?

You know, hand slap, we can't even keep dark money out of politics and a lot of other, you know, campaign finance rules, uh, enforced. So I'm not too optimistic about whatever regulation this might involve. I I think one of the tricky parts about it is you may be watching something that who knows what put who put it together that someone else tweeted or a political candidate shared on X or whatever platform, and there may be AI generated in one part of it.

What, how do you know what part of it? Are we gonna flash up a little sign that says this part's the AI part? Or was it all, uh, I, I think it's just, I, I'm, I fell more into the camp of, I think it's all gonna be not trusted and we're gonna rely on people that we, we trust assuming that we know that it's actually them.

So, so first of all, what you guys are talking about, it kind of sounds like it's something outta Blade Runner, right? If you remember the, the Blade Runner movies, and they had that thing that you could look in the retina of a person and see if they were a replicant or whatever they called them. Yeah.

Yeah. It, it's impractical at this point of the game. But, you know, Mitch, what you are describing is, well, I'm just gonna believe who I trust is exactly where we are now as a country and look at the mess we're in as a result of it.

Because what happens, we tend to trust people who say things we agree with. And so it reinforces our tribal differences are tribal, you know, allegiances and, and it, and it tends to rip apart the fabric of the country or of a society. What's liberalism for, for that manner, right?

And we need to figure out how to get people to trust media, to trust sources. You know, I I I, I was at a, uh, a lecture, uh, tuck Chad, who used to, you know, host Meet the Press a couple years ago, this is when he was still at Meet the Press. And he said that, you know, we're going through a period and he thought it was gonna take 10 to 12 years where people were going to have to learn to discern quality from nonsense on social media.

I think AI muddies the waters even more. But the interesting thing he said is that all of us on this show will never learn. It's the next generation.

It's the natives, the digital natives, the AI native generation now who grew up, who grow up with this, that will be better at, I don't know if they'll be, I be able to identify ai, but they'll have better mechanisms to filter and, and, you know, authenticity. And I see it, I see it with young people who have been turned off to the Fox and CNN and M-S-N-B-C and, and Newsmax conundrum. They go on YouTube and TikTok and, and listen to people you probably never heard of.

Because those people tend, they think at least not to politicize news, just to give it straight news with no bias one way or the other. And that's what they're craving. That's what they're looking for.

So I think that's That general generational awareness, because the younger generation will automatically know there's a possibility this is ai. Whereas once you start getting to older generation, they, they would be shocked. Like, what?

This is ai, how can that be? Yeah. So I think there is gonna be an awareness that comes up with, with younger people.

That's a great point, Bonnie, on the awareness piece, I think that's absolutely critical, um, for more folks to start understanding that the SEC is only talking about TV and radio, not talking about digital content, streaming content. Um, so I, I, I like it. I I bet your point is, all of your points are so valid in terms of the trust.

And I think that's currency these days. And that's something that I think the more awareness that can happen across all generations in the working world today will be, uh, table stakes, right? It's conceivable that I could develop more trust for some sort of AI bot than I do for a person, right?

I can think of a lot of people that I don't trust, but I might have interactions with an AI bot that's better. And I might ultimately trust that a little bit more, assuming I know how it was trained, or I have some confidence in how it was trained. So just because it's artificial doesn't necessarily mean that it's not necessarily better.

I know a lot of kids, Hey Mike, are you saying love is love here? What? What's going on?

I think we should end Blade Runner. You talking about Blade Runner again? I, I'm just saying, man, we should not judge No judging from me, Mike.

All Right folks, we're gonna close this here 'cause I'm not entirely sure that Alan isn't an AI thing 'cause we're about 15 minutes into this and he hasn't waved his glasses once yet. So I'm not sure that that's happening, But we'll Be back in a minute. Cloud native now is the web's leading resource for the growing cloud native ecosystem.

com is your destination for news, thought leadership, features and webinars on cloud native architecture, Kubernetes, serverless, cloud native application development, microservices, service mesh, cloud native security, and more. Stay on the cutting edge of modern application development at Cloud Native now. Alright, folks, we're back in.

Now we're talking about AI in the call center and there's an article up on text drawing AI talking about how these AI models are starting to find their way into this use case. I think it's a very good use case for AI initially, but I'll bring up this point too. My sister-in-Law worked in a call center, just quit that job about a month and a half ago and just said it was too much stress.

It was just constant people, um, you know, having issues that she didn't have a playbook for or, and the people measuring her performance were kind of like, you know, getting it down and each call had to be flipped within a minute. Yeah, I mean, it's just, you know, a little bit on the insane side. So maybe we've finally reached a point in the call center where, um, we absolutely do need ai.

I mean, Lisa, what are your thoughts? I agree. I think there, I think it's a fantastic use case.

It's funny because when I talk with, with folks in, in tech, which I do regularly as part of the media, people think chatbots. I actually don't think that what I see it as is an opportunity to help train models and help it learn. Uh, and so maybe the next experience is better, but we're seeing on the customer experience front, um, definitely progress in that area.

I think we're seeing shifting priorities. Um, I think it can be an advantage to have virtual AI power assistance, customer experience, employee experience closing calls faster. To your point, Mike, with the turn that your sister-in-law had to do, um, I think there's a, it's a good ease case as long as it's done purposefully and responsibly and transparently.

Mitch, I have a, Oh, go Ahead. Sorry, what? Can I go?

Yep. So I took off my glasses for you. Um, So, you know, I put something up on my Facebook page yesterday.

I know I'm an old person, so I use Facebook. My kids tell me only people my age use Facebook, but it was a Pooh and piglet thing. 'cause I'm a big Winnie the Pooh fan.

And, and you know, poo is very upset with the state of the world and what's going on and, and he was looking for con consolation from Piglet, and Piglet said, Pooh, I wish I could tell you everything's gonna be okay, but it's not. But I'll be here with you every step of the way and we'll get through this together the best we can. But I can't promise you anything.

Well, this is the same thing here with these ais with this, in this instant case for, for too many people, we've been blowing smoke up their butt saying, don't worry about AI taking your job. AI is gonna make you 10 x more effective. It's gonna free you up to do higher level things.

AI is gonna make you more effective, more efficient, and more valuable. Well, for a lot of people it is going to do that. But there are gonna be some jobs like a call center kind of job where quite, quite frankly, AI does it better, faster, cheaper.

They never take a sick day or a vacation. And their body of knowledge is much, the body of knowledge in the AI is much greater than the individual who sits, you know, behind a computer or on a, a head microphone. And you know what?

Those jobs are toast. Uh, you know, I'll be here every step of the way with you poo, but you're not gonna be a call center operator anymore. You should look for something else.

Because I think that is a job, as Lisa mentioned, where given the right training and, and body of knowledge, it's a, it's a natural to do that, right? And to say, oh, we're not gonna putis in there 'cause we want to preserve these people's jobs. I I think you have to look hard at who is actually saying that about ai because it's always invariably the tech company that doesn't want to be perceived as the job killer.

So they go outta a way to say, you know, these awesome things about these new roles and capabilities, and there will be some of that, and there might even be more net new jobs at the end of the day. But it's not gonna talent's point gonna happen without a certain amount of trauma into the system. It's just the way things are gonna go, and it's just gonna be part of the issue.

And it may affect older people more than younger people, but it may be harder to get into a new job area for younger people because there is no entry level job. You gotta have five years of experience doing stuff, and that's gonna be harder to come by. Mitch, um, what, you know, in the IT set, what are you guys seeing in terms of, you know, what do you think the expectations for skills are gonna be?

Well, I think one of the benefits of using AI in this use case is we at least won't have ha listen to the message of, by the way, did you know you can do all this on our website? Well, not, I can just do it here with a, with a bot. I don't have to go to the website.

I, I think it's an inevitable application of it. Call centers are not notoriously expensive and, you know, they, they're highly scripted, uh, closely managed and measured, but it doesn't mean that they're successful a lot. Uh, I think the current crop of bots, for the most part are very limited in what they can actually do.

And there's potential for having more helpful results from that experience. There's always gonna have to be someone or maybe a higher level bot or a person that you can escalate to to that. The question is, where does all this training come from and how do we, uh, train domain specific information, um, about our business?

You know, a call center for a telecommunications service to, to business B2B, it's gonna be much different than, um, how do I pay my credit card bill and I wanna dispute this service, or something like that. So I, I think that's where the domain of like, how do we learn to do that and how we do, do we do it effectively? Do we have to figure out ourselves or are they gonna be a crop of companies that are really good at training models for, um, business call centers?

And we'll effectively outsource the whole thing into some digital world with some level of escalation to a human. I agree. I think that, uh, one of this is gonna be a true test of how does AI compare to real human emotion?

Because when people are calling a call center, typically they're upset about something and they wanna speak to the supervisor, they wanna escalate as, as Mitch mentioned. So how do you, you have to train the AI to deal with any potential, um, emotion that the, the caller might have and, and whatever they're saying different responses now. I see.

I think we're seeing that now where they do have like canned responses, but this is where it's gonna be really tricky because if somebody's calling and they're upset and suddenly they say, wait, are you a real person? You know, I think that's what they're trying to avoid. Open the call center door.

Hal. Um, hey Lisa, what is your advice to tech companies in terms of how to position ai? Because as we said earlier, it seems like it's a little, uh, there's a disconnect, shall we say, as Alan pointed out, between reality and what the, the, the tenor and the tone of the message is from some of the larger tech companies.

Is there work that needs to be done there? Definitely, you know, as a marketer and a glass capital person, I see things like AI in the call center, um, or even AI applications across, um, organizations for faster processes, better processes, more creative processes, as a lot of advantages in there. And when I think of the call center, I think of well, improved customer experience, reduced wait times.

Obviously there's a lot of frustration. It's very controversial. Um, I think the opportunity there for tech companies is to be honest about what they're using, how they're using it.

Wine brings up a great point about the emotion, and that's where I think humans still have to be in the mix to be able to deliver that. Because you're right, people are calling in the call center, they're upset about something and they want it resolved as quickly as painlessly as possible, especially like credit card dispute or something. But I think tech companies need to be, um, really honest, upfront, transparent, like I keep saying authentic and, and, and explaining what they're, how they're using it, why they're using it, and what's in it for me as a customer.

How is it gonna make my life better? That's the want, want tech companies to be able to deliver. All right, Alan, honesty, do we need a little more honesty in the system?

I mean, you kind of started at this conversation, so how does, in your ideal world, what does that honesty look like? It's common sense. I I think honesty is common sense.

You know, I, I had this conversation once with a professor at Indiana University, um, I forgot the business school there. It's a really fine business school at Indiana. I forget the name of it.

My son was accepted and it was around the minimum wage question, right? He said, you, we don't want minimum wage uh, laws in this country because it takes jobs away, right? Because, um, if, if, if we raise the minimum wage where it's cheaper to automate something, employers will automate it rather than hiring people.

And so it's gonna cost people's jobs. If we, at that time we were talking about the $15 an hour minimum wage, and my retort was, if you are doing a job that can be done, can be replaced. It was the Kelly School, by the way.

Mitch. Mitch put it in there. And it is, it's an excellent business.

Call one of the top in the country. But if you're doing a job that can be replaced through automation, and as an employer, I could do it cheaper, faster, better with, in this case, ai, I'd have rocks in my head not to do that, right? It, it's, it's called capitalism, right?

And you, you don't wanna make, make work jobs. This isn't the great depression in the Tennessee Valley authority. If, if, if AI could do a job better than a person can, the market's gonna dictate that we use AI for it.

So what I say is, it's common sense. You gotta look at what, what is it I do? And I, I had this conversation with my own children.

I have it with every intern we hire here at Techstar. What effect do you think AI is gonna have on your chosen career path in the next five years? If what you were planning to do can be easily done by an AI chat bot or, you know, whatever it morphs into, you know, talking to it.

But if AI can do what you are gonna do, 'cause it's basically repetitive, kinda work like that, think of something else. Think about how you could leverage that AI to do something else, maybe similar related, but you know, if you are just handling customer complaint calls at a call center, I wouldn't be aspiring to collecting a pension on that job at this point, right? Because I, I think, you know, common sense dictates that that's going to be something that will be handled there.

It's the same thing about, we probably all had friends who were bank tellers growing up, right? Bank teller was a decent lot of married women, you know, back then, because, you know, it was a decent job for a married woman who was thought you go to work in a bank who, who has bank tellers. Now even when you, those rare occasions when you walk into a bank, you're there to either see someone at like the desk, a manager or someone, or you go to the machine instead of the teller.

So I, I think you gotta use common sense here. It's not, it's not some evil thing, it's not some evil plot to destroy the proletariat or something. It's just this is, this is technology and AI is just the latest, you know, phase of it.

Lisa, flip this on its head a little bit. Do you think a lot of the people who may find that they're being automated out of a job will then just take AI and go launch their own company to provide the capability that they were previously employed for? And we're just gonna see this explosion of entrepreneurs who were gonna run around with AI and just say, yeah, I don't need a big company to do that for you.

I'll do it myself. And here's the fee. I, I wish I would say yes, but I don't think that's reality.

I think that the folks who aren't gonna take the opportunity to be upskilled, who are going to be phased out, um, in part by what AI can do better, faster, cheaper, I don't think that's the mindset of a person that is going to have their own ideas and take it off and create something. Um, I wish that was the case, but I think if we just look at the general population, um, I, I think that's not, People are not that motivated or that entrepreneurial, but let's look at media for instance, right? Look, I grew up, I grew up in New York City, so we were very lucky, right?

We had, uh, Mike, you did too from the Bronx, right? We had, we had channel 13, 11, 9, 7, 5, 4, and two. So we had seven TV channels.

A lot of people, Mitchell grew up in Nebraska, he probably only had three or four channels, right? And then cable came, and I don't mean anything wrong by it, Mitchell, you know, Hey, we knew more About Musk now we got, We got WGN and the world, the world was our oyster, you know, Right? Yeah.

No, yes. So, but, you know, then cable came and all of a sudden we had a hundreds or a hundred channels, or 72 channels or whatever you paid for. And then the internet came, and now everybody's a broadcaster and everybody's a media person and everybody has their own, you know, channel, right?

So can you do a similar thing with ai? Yes. If, but to Lisa's point, you gotta be motivated and entrepreneurial to want to do that.

Generally speaking, the person who wants to work, let's say at a call center or a bank teller job and that, you know, is what they're looking for, they're not going to go do what we want, you know, what we're talking about here. They, and that it's the way of the world. Unfortunately, No, Alan, there's a whole nother factor here.

I think that that is possible. And that's the data. So the people who have all sitting on all this data from all these recordings, from call center, the, the data, the interactions and what happened that, that's a huge plus for someone else who's brand new to that space, right?

I'm gonna go do my own thing and just create a chat bot to do this myself. Well, if I have all this data from years of operating a call center and now I have the tools to analyze it, not just analytically, but you know, turn voice into text and do the text analysis with Gen ai, I can, I can train models much more quickly to do more things and analyze success outcomes versus non-successful or desired or non desired, whatever it might be. Versus starting up from scratch and just having a model that, you know, functionally works.

But I don't really have the expen experiential data to train it. I've gotta go find some way to train that. I think that's a huge advantage for people who are already doing this function.

The business, not necessarily the workers, but can't forget about the data. All right folks, we're gonna be back in a minute. I'm gonna parse some, uh, lucky to grow up in the Bronx in the seventies and the eighties thing with Alan there.

'cause you know, I'm having all kinds of flashbacks right now, but we'll be back in a minute. Hello everybody. We're back and we're talking about gen AI insecurity.

It turns out that folks at immersive labs ran a test and they had like 31,000 people participating in this thing, and they hit a secret. And then the LLM, it was basically, uh, some social security number or, and a couple other things that was like 10 different levels. And almost 80% of the time the LLM was tricked, shall we say, into coughing up the secret.

So it appears to turn out that, Mitch, maybe you a thought here that LLMs can keep a secret about as well as a five-year-old. So can we business around that? Well, this is ACT actually something that came up.

We did a, uh, a day at, uh, to, at, uh, RSAC recently, and we had the CISOs from GNA, or sorry, open AI and a number of different organizations. And it came up about this prompt injection of not only getting data out of the model, like things, things like social security number, and there's ways that they can, they can filter that through the, through the prompt engineering or the training of, of the model itself, but also disclosing what the model knows. If you query it enough, you can, and you built enough, uh, you know, prompts that you've engineered to tease out what the model knows, you can actually start to disclose things that might be considered IP about the organization, how it functions.

So imagine you're an attacker and saying, where, where are, where are the vulnerabilities? Where's kinda the low hanging fruit to go after these people? And the ability to slow roll, prompt engineer, uh, engineer prompts into a model, into a service or a bot or whatever it might be.

So it actually is a, a new attack vector from a security standpoint. I think it's, uh, a one we're early in understanding 'cause we barely know how to engineer prompts, um, and more or less protect them. So I think we're gonna see a lot of discussion about this is being a big, not only data leakage, but also attack surface.

Alan, what's your take on this? 'cause I think, you know, it's, it's cause for pause, at least for a second, because we gotta think about what data's actually going into these LLMs. And we historically have never thought about what data goes into what application.

We just kind of went with it. You know, this, I don't think this is anything different than many, many other technology waves we've seen. When people are deciding, you know, designing their ais and how they're gonna utilize the LLMs, one of the top three things they're not thinking about is, well, how secure is this?

Right? We will get better security in our LLM data when the consumers and the users of the LLMs demand better security right now. They just, you know, they're, they're enamored with the parlor trick, with the, you know, magical, you know, auto magical nature of these things.

And I, I don't think security is getting the attention it may get later on. And, um, it will, security will get important eventually. Like I said, when people say it, they, they need it and they want it and they demand it, then they'll do it.

But it Could be the dots. I I think that that's right. Like for security to be, uh, uh, at the forefront, something has to happen.

But maybe because it's coming from an LLM then to the user, we weren't gonna see it right away. It could, it could be a subtler thing where the, the user then takes an action which makes them less secure and they don't put two and two together and realize, wait a minute, where did I get that advice to do this? That came from an LLM.

Lisa, do you think that this is a, a marketing nightmare in the brewing here? 'cause, you know, in about four or five months, there might be an issue, and the first thing we're all gonna agree on is that it's the tech company's fault, Right? Right.

Finger pointing, it's gonna happen for sure. You know, I think it, it highlights the importance of rigorous testing, validation, and the development of AI powered systems for any company. I also think though, that there's opportunity for innovation.

What we're seeing here now with the flaws, and Mitch brought up a great point about being another attack surface. Um, you know, ransomware attacks happen in every 11 seconds, but I think that discovery of these flaws that immersive cloud has uncovered is really preventing opportunities for organizations to be more innovative, more aware. I think that awareness piece is so critical, especially securities concern because people are often the weak link.

And now we're having, uh, demonstrative ai, uh, secrets being revealed. But I think that for organizations, the companies, I think, and this will spread to other industries as well, leveraging AI powered security testing, monitoring, trading consulting is really where this can go to be a positive thing to acknowledge, yes, we have an issue here. This is what we're doing, this is how we're going to innovate, learn from it, and start reducing security risks.

I don't think they're gonna go away. They can be hopefully dialed down. You on that testing front that, that you're talking about, Lisa, um, certainly something that will emerge is pen testing for models, right?

The prompt testing, just like we do pen testing for networks. Um, I'm kinda curious about tying a thread back to our previous con conversation about AI generated or misinformation, disinformation, you know, we could actually have, um, offensive responses to attacks that we, we can confirm they're actually an attack. Alan and I have been in the industry where it was like, do we, do we block this attempt to, to attack our network?

Or do we block this ip? Do we set it on auto? Do we just kinda let a person deal with it?

Imagine a day where the model's responding in a different way because it, it thinks this is an attacker and it's gonna send it off into a honeypot or into other, other directions, maybe towards my competitor. I'm not sure what, but there could be some interesting, kind offensive measures that we could have with, uh, AI and Gen ai. That's a great point.

Alan, do you think this is gonna wind up being like a separate category of products? And I'm asking this question because, you know, I was talking to somebody about this the other day and they were like, Hey, let me get this straight. You're gonna sell me this thing and this thing isn't gonna really completely do the job or work, so I gotta buy this other thing.

And in their analogy was it's like, you wanna sell me a car, but you want me to pay extra for the locks and the keys. So, you know, is this just a feature of something in the LLM or do you think we're gonna see like this whole kind of, uh, third wave of data, LLM security products that people will pay extra for? Again, this is the natural way, I think first security's bolted on, then it's built in, so it'll be bolted on for a while and it's built in.

Interestingly, I'm already starting to get soli unsolicited spams about let us help you build your own LLMs, right? Because right now I think a lot of organizations don't have the wherewithal, the know how to build their own LLMs. And, and so there's a cottage industry of let us build your LLM for you.

I think eventually custom LLMs and s SLMs will become much more the norm than, than, you know, just using these massive LLMs trained by open AI or, or what have you. Um, but right now that's what we're stuck with. And, and security is gonna need to be bolted on.

Once you can control your own LLM, you'll also be able to control a lot more about the security of, and the data in there, obviously, but it's a process, right? Like everything else in technology. I met friend Damon Edwards, one of the founders of the DevOps movement.

Damon always tells me it's a process. And, and that's what we're going through here with ai. You, you know, it's a process.

It it's an evolution. We'll get there eventually. It's gonna be a little wild west, you know, for a time, but, you know, so is the US So, and, uh, it happens.

That's just the way of things. Lisa, do you agree with that? Is that kind of the arc of things and how does a tech vendor position something that is, shall we say, inherently insecure in a way that somebody wants to buy?

Yeah, that's, it's a, it's a great conundrum that we're living in right now. I do agree with what Ellen is saying. Um, I think that we're going to have to see, um, the transparency, the honesty, the accountability, and the acceptance that until John's point, we can get control over the data in LLMs.

We have to acknowledge that there are inherent security risks that probably, I mean, we're talking about what immersive labs found. There's probably so many more that we haven't found yet. And so the, the control factor is key, but also so is the acknowledgement that, hey, we're all still learning here together.

Um, it is gonna be the wild wild west for, uh, for probably quite some time. But acknowledging that, and, and I think being honest and truthful about that is a, a great thing that tech companies and I think companies in every industry can do to say, we understand that there are inherent flaws here. We're working to identify as many as we can and, and, and dial it down and we eradicate them.

Mitchell, am I gonna need an LLM to monitor and secure my LLMs? Well, I think there's very much possibility. I mean, think of this just like the way we have specialized humans in, in techniques like penetration testing and evaluating security of systems.

The same thing will happen with AI. And, and, uh, I'm reminding of reminder of the movie a creator on, uh, on Hulu, you know, and not to give away the, the, the, uh, the end of the story, but, you know, be careful what you ask for when you meet who the creator actually is. Well, in this case, I think we're gonna be using AI to test, to create and test ai just like we're gonna do for software.

We're doing gonna do the same thing for security. Is it the panacea that's gonna solve all the problems? No, it's not going to.

But I think it's inevitable. We, you know, to Alan's point, this is the way of things or, or, uh, to, uh, to the Mandalorian, this is the way, Speaking of the way you Way finish strong Mitchell, strong Boy, but Mm-Hmm. Bonnie is usually at the forefront of using a lot of these tools.

So as we discussed any of this Yeah, did it give you any sense of, uh, concern or what's your Thoughts? I, I, I think that, yeah, it does, because there's the thread of we're still figuring out how do we catch these, these cybersecurity breaches makes you wonder the cybersecurity person, that's the, the bad guy thinking of, um, they don't know yet how they're gonna do it, and they're still figuring it out. So it's hard to be one step ahead of what they're doing.

But, um, yeah, I think it makes you more aware. But every large learning model, it definitely is influenced by, like we were saying earlier, who's programming it. So you have to kind of look at the sources, I think, too.

But, um, it's something to be aware of as, as you go forward, because as, as, as everyone's been saying, Alice was saying, it is the wild West, we don't know. There you go. You're gonna need pictures of the humans who created the AI so we can trust the humans and then therefore extend the trust to the ai.

Yeah. I'm gonna end this conversation here, but I am gonna pass it over to Alan who's gonna explain what he's doing in Santa Clara. Sure.

Well, thanks Mike. Actually, it's not just me, but it's Mitchell and we're joined by, uh, we're on a few, uh, research mission here. You know, one of the sister companies we're working with at FU is, uh, tech Field Day.

Uh, a Tech Field Day. For those who don't know, it's been around about 15 years. It's a great program where they, you get like about a dozen delegates, delegates in a room, and companies come in and, and explain their, their company and their products to the delegates and the delegates question and, you know, and, and, uh, find out about them and, and probe in to see.

And the whole thing is broadcast live. And as part of our relationship here with future, our growing relationship, we are broadcasting live. And we'll also have it available on demand on Textron tv.

So immediately following Textron Gang today, uh, we will be broadcasting the, uh, tech, it's the app dev field day. So this one's focused on app dev. We had yesterday's show on as well.

And I believe that may be already available on Techstrong tv. So do check that out. And Mitchell and I, this is our first one.

So, uh, Steven Foskett is the, uh, founder of Tech Field Day. He is really fantastic to work with. We've got a few other FU folks here, uh, on the delegate panel, as well as just some awesome smart people in AppDev.

So do check that out along with the rest of techstrong tv. And, uh, we hope you enjoyed this. And Mike and Bonnie, thanks for holding down the fort back in what I assume is Sunny Voca Ratone.

And, uh, we'll be, we'll be home soon. Leave a light on. All right, folks, thanks for watching the latest edition.

We'll see you next time. Uh, this is gonna be a story about the cost of ai, and it goes like this, creating a large language model. That's the thing that drives AI, is really, really expensive.

Um, Dario Ahmadi and the CEO philanthropic, which is a very big player in the ape and AI space, he's saying it's gonna cost, uh, a billion bucks to make a model. Now, to give you an idea of how much a billion dollars is, if I had a million dollars, it would take me a thousand years to spend it. So that means that in 30 24, I'd run outta money.

And today's 2024, that's a long time. Um, I think the long and the short of it is that, you know, the days of somebody sitting down at a personal computer and cooking up some code that made a multimillion dollar business, I, I think those days are over. Uh, it just takes a lot of money to get into the game these days, you know, but it wasn't always that way.

I mean, Microsoft got its start on the personal computer. They made the operating system das, which was for the IBM pc. Uh, another one, uh, al, which was the first spreadsheet for a PC, was written by two programmers who have to be aware of, and Dan Bricklin and, um, Bob Frankston.

I mean, even something as well known as, um, Google search was really created by four programmers. And, you know, what did they need to do there? Well, they needed some PC on their desktop.

They probably needed a couple of servers and a data center, uh, a connection to the internet. But, you know, most importantly, they needed to be creative and they needed to use their brain. Um, and that was, didn't cost a lot of money back then.

Now, you know, I know I'm being a bit nostalgic about the freedom and prosperity, the personal computer. And don't get me wrong, I'm very aware that big business needs big money. Um, bell Telephone, which is sort of by Alexander Grand Bell, the, excuse me, the second person he hired was a guy, uh, named Thomas Sanders.

And Thomas Sanders was a finance guy, the treasurer. And why did he hire a finance guy? Because Bell knew that it would take years to build a telephone system, that there would be more money going out for a very long time then coming in.

So this notion of big business needing lots of money and a really high cost of entry, that's nothing new. But, you know, I I, I'll let you in this secret. You know, I'm self-taught, and all it cost me to get into the game was I had to save up some money for pc.

I had to buy some books on computer programming. And I had had the time, but eventually I learned and I became a, uh, I became a professional programmer. And the professions, um, been very good to me.

I have absolutely no complaints, but, um, without that co low cost of entry, I'd be nowhere. So, you know, it seems to me that today for a person to get good on a technical landscape that sort of is really being driven by ai, they're gonna have to work for a big corporation very much in the same way that, you know, IBM mainframes ruled a programming environment back in the fifties and early sixties. And, um, I, I, I just don't know if another one of me is gonna be able to come along or, you know, a gates or bricklin or, you know, the kind of Google, uh, when I'm not near a league by any means, but without that low cost of entry, um, they're just not gonna be able to make it.

So, I mean, I guess in order to prosper, you know, in the future, you're not only gonna need to know the in and outs of machine learning, but you're also gonna have to know how to get along in a corporate environment, which is, um, not something I've been very good at. Um, but, you know, it didn't really matter because all I never needed to get by is, you know, a pc, you know, a connection to the internet, and, um, my brain and a good amount of creativity. And, um, I've been okay.

But, you know, you know, I'm wondering if this high cost of entry is really gonna prevent some, um, aspiring, uh, technologists to be able to even play. I really wonder about that. And, you know, it doesn't matter that the growing cost of AI is gonna prevent this little guy from getting in.

Oh, I don't know. Um, but is it something we're thinking about? Yeah, I really think it is.