Transcript
Speaker: Hey everybody, welcome to Grounded Truth, a podcast where we gather some of the world's most influential data scientists, machine learning practitioners, and innovation leaders for conversation in the most relevant topics in AI today.
Speaker: As always, I'm your host, John Singleton, co-founder and head of success here at Watchful.
Speaker: Check us out, watchful.io.
Speaker: If you like this podcast and like the content and the people that you listen to, please give us a follow everywhere that you can find podcasts today.
Speaker: And also with me is my co-founder and CEO of here at Watchful, Cheyenne Mahante.
Speaker: How's it going, Cheyenne?
Speaker: Hey man, good, good to be back.
Speaker: Yeah, really excited today.
Speaker: We're joined today by Joe, Andy and Kieran of the AI FYI podcast.
Speaker: We're super excited to have them.
Speaker: How are you doing today, guys?
Speaker: Great, great to be here.
Speaker: Wonderful.
Speaker: Thank you.
Speaker: So maybe a good place to start is how did you guys meet?
Speaker: How did you guys come up with AI FYI?
Speaker: What crazy journey brought you to where you are today?
Speaker: Yeah, maybe I'll start.
Speaker: So I got my first product management job at Figure 8 about five or six years ago.
Speaker: I'd been in AI and label training data for five years at that point.
Speaker: And I got assigned to work on human data labeling annotation tools.
Speaker: I'm sure you two know those very well.
Speaker: And I got given a tech lead and a product designer to help me in my first ever PM job.
Speaker: And they were Kieran and Joe.
Speaker: And yeah, the three of us started building stuff together.
Speaker: Yeah, and I guess my journey there was, you know, started in engineering.
Speaker: For me, my focus was always front-end engineering and really product experience, which is why I think, you know, the three of us worked really well together because we cared a lot about, you know, what is this doing for the customer and, you know, how are they going to actually be impacted by what we're building?
Speaker: So that was a large focus.
Speaker: And, you know, like Andy said, we built a bunch of cool things, got a couple of patents and then, you know, eventually scattered into our own our own places.
Speaker: And now we're back together creating a podcast.
Speaker: So it's really fun.
Speaker: Yeah.
Speaker: As Andy said, I'm the product designer of the group, which I think is probably the least technical position represented here.
Speaker: But I like to think that it brings a lot of heart into how we understand and work with ML.
Speaker: So I've been in that space for just about as long as these two.
Speaker: And yeah, as Kieran said, we started a podcast and we're using it to demystify AI, to explain AI and machine learning to folks who are maybe not so familiar and really tackling sort of subject and domain by subject and domain on how AI is impacting that space.
Speaker: So I'm really curious, like why now?
Speaker: Like why did you choose now to create a podcast?
Speaker: So a couple of things all happened at the same time.
Speaker: One, the three of us missed each other and we were on a group text being like, we should build something.
Speaker: We're bored.
Speaker: That's a good reason.
Speaker: Very reasonable.
Speaker: And then AI kind of hit the cultural zeitgeist in mass last year.
Speaker: So many people who didn't understand what our jobs were at all were suddenly very interested in what we do.
Speaker: And my mom went from saying something like, my daughter works in tech to look at chat GPT.
Speaker: I think my kid helped make this.
Speaker: And we were originally thinking about writing a book.
Speaker: But in the process of researching that, we realized that by the time we got it written, it would already be out of date.
Speaker: This technology moves really fast.
Speaker: And we aligned on our goals that we wanted to help
Speaker: arm people with information and that the best medium for that was a podcast because it could be quick from event to content being released.
Speaker: It was conversational.
Speaker: We could try a lot of different formats.
Speaker: And I think that was basically it.
Speaker: Did I miss anything, boys?
Speaker: I'll also say it was, it's really a great opportunity for us to learn the area, like Andy said, is so fast moving.
Speaker: It's great just to like be able to check in with these two every week and understand what's happening in the space and the new technologies that there are and how it's being applied.
Speaker: Yeah.
Speaker: Yeah.
Speaker: And I think that's really been my favorite thing about the podcast is we don't have any particular topic that we focus on.
Speaker: So we'll kind of just pick a topic and we'll say, all right, let's go research this for the next couple of days and then talk about it on a podcast.
Speaker: And so, you know, 95 percent of the time I'm talking about something I have no idea about.
Speaker: But, you know, I read some things and learned some stuff.
Speaker: And based on, you know, my technical experience, you know, I'm starting to connect dots and feel like I'm learning a lot, you know, in that process, which is cool.
Speaker: Yeah, I think that's something that's sorely needed.
Speaker: When ChatGPT first came out and after the dust started settling, I finally realized, like, why this chat interface?
Speaker: What's the real brilliance here?
Speaker: And I think the first thing I said was, well, for my mom.
Speaker: My mom now kind of understood in the same way Andy was like, oh, yeah, ChatGPT, that's something like that, right?
Speaker: And, like, it finally clicked.
Speaker: For the average person, we now had something that looked, felt, and smelt like the AI we were promised or saw in science fiction as children.
Speaker: And now this thing's here throwing recipes at me and maybe lying and changing my political views, but we can get to that later.
Speaker: But it's a pretty wild time.
Speaker: Well, what's crazy to me and what was partly motivating for us is when we were thinking about writing a book, I went into Amazon.
Speaker: I'm a product manager, right?
Speaker: I like to go look at the competitors and the data.
Speaker: Went to Amazon, looked at bestselling books and technology.
Speaker: And the top selling book in the space Q3 of last year was called chat GPT millionaire.
Speaker: And it was some guy in the Netherlands who'd published it in 36 languages and it was being sold for 24 99 or something.
Speaker: And it was telling lay people they were going to become millionaires with chat GPT.
Speaker: And I think that is at best misleading and probably downright abusive if we're being honest.
Speaker: And this guy probably used chat GPT to publish it.
Speaker: And he sold tens of thousands of copies.
Speaker: And I just hate the idea of like people being taken advantage of and having their first interaction with AI being, oh, I got ripped off in some way.
Speaker: I hate that.
Speaker: I totally agree.
Speaker: Actually, recently, my grandmother was the target of like an AI phishing attack.
Speaker: I don't know if you guys have experienced this yet with your relatives, but they're now like fairly sophisticated attacks going on.
Speaker: Like they called my grandmother.
Speaker: There's enough like voice clips of me on the internet now to where apparently they've cloned, at least for that call, they cloned my voice.
Speaker: Uh, and they had my grandmother, uh, send something like $5,000 to quote unquote bail me out of jail.
Speaker: Uh, and not to tell my parents or anything like that.
Speaker: We only found out because my grandmother kind of shakily asked my dad, Hey, did Cheyenne get out yet?
Speaker: Uh,
Speaker: And my dad had no idea what she was talking about.
Speaker: But it just goes to show that, like, my grandmother is certainly not equipped to handle that type of sophisticated attack.
Speaker: We're only going to see more of them.
Speaker: I'm curious, like, is there anything that can be done about that?
Speaker: Yeah, I mean...
Speaker: Luckily, we got it all figured out at the end.
Speaker: But, you know, I don't want that to happen to anyone else.
Speaker: So I'm curious about your take.
Speaker: Like, have you seen this before?
Speaker: Is this something that is only going to get more prevalent?
Speaker: How do you guys think about it?
Speaker: Well, it's interesting that you say that because the exact same thing, actually, my friend was just telling about their neighbor, the same exact thing happened.
Speaker: And basically, you know, they got a whole bunch of texts and it was like, are you this person?
Speaker: Do you have, you know, this relation to these people and blah, blah, blah.
Speaker: And eventually, you know, he ended up sending hundreds of dollars, fortunately, only hundreds of dollars.
Speaker: But yeah.
Speaker: You know, it's still sending money to some random account because they're just scared.
Speaker: Like people are just in fear.
Speaker: And so they're trying to buy their way out of something.
Speaker: And yeah, it's, you know, how do you get rid of that or prevent that?
Speaker: Yeah.
Speaker: That's absolutely crazy.
Speaker: I haven't heard of that before, but something we regularly talk about is how we are like on the precipice of this boom of LLMs being applied across so many different things, you know, in existing apps, new apps.
Speaker: And then unfortunately, it sounds like scams and things on that end.
Speaker: Yeah.
Speaker: Now that we have the capability to sort of pretend to be human even more, the yeah, the opportunities to scam people are going to be popping up.
Speaker: And it's going to be an escalating arms race.
Speaker: I was looking at papers today and there was a paper that came out, I think last week.
Speaker: I have the notes pulled up in front of me.
Speaker: I'm just looking at it right now.
Speaker: Okay, yeah.
Speaker: So somebody published a paper that using zero-shot predictions was able to detect bot-generated text with 90% recall and less than a tenth of a percent false positive rate.
Speaker: And I was like, that's really good.
Speaker: So I write a work letter, a newsletter for work, and I wrote, quick, turn in your homework before someone adds this to the plagiarism tools teachers use.
Speaker: But like, yeah, there's this technological arms race happening between
Speaker: AI generated content and detecting AI generated content and caught in the middle is a whole bunch of kids, teachers, elderly, like a whole, you know, the vast majority of the population.
Speaker: It's a scary time.
Speaker: Maybe scary is not the right word.
Speaker: It's dramatic.
Speaker: It's a lot of fast changing technology.
Speaker: I think it was this week, too, that there was the deepfake of President Biden calling voters to not come out.
Speaker: So these things are going to be in the headlines a lot.
Speaker: Yeah.
Speaker: Yeah.
Speaker: I've always thought I think we're going to have to start having like personal key phrases or some sort of like, you know, you hey, we're.
Speaker: It's like we have an evacuation plan for fire, right?
Speaker: Well, we need to have a deep fake plan.
Speaker: What's your favorite pizza sauce?
Speaker: Sassafras or something crazy, something just wild.
Speaker: And we have this secret that only we share to validate if we ever need to know.
Speaker: And these things like...
Speaker: I think that the thing that is scariest for me for the average person is like I am, and we all are in AI.
Speaker: We are as close to the problem and the developments as anybody else.
Speaker: And the pace at which innovation developments, the new paper, whether it's some new Mambas and the new hotness, whatever the new architecture is, look at the new smallest model, the newest leaderboard being top, we see all of these happening.
Speaker: If it's difficult for us to keep up with what is current state of the art and what is even possible, the trickle down to someone who has maybe the most exposure with, A, I made a cool list of everything in my fridge with ChatGPT and it helped me out what I need to make is a massive gap.
Speaker: And I think that being able, like, I think we need better tools and I think we need to see an evolution of the ecosystem to one, be prescient of these things.
Speaker: Like we need to be able to validate these results.
Speaker: We need to be able to have better detection tools.
Speaker: And two, we need to be able to know when are these things happening, you know?
Speaker: And so I'm curious, like,
Speaker: Where do you think Gen.ai is really being applied today?
Speaker: A little doom and gloom with some of the scams, which we're monkeys and tool users.
Speaker: That's what we're best at, honestly, that and talking shit on the internet.
Speaker: But we are going to use tools just like you can use a calculator to solve a math problem or add up how much money you made selling illicit substances.
Speaker: They're both tools at the end of the day.
Speaker: Where do you think it's really being applied?
Speaker: Where are we actually seeing the shifts in Gen.AI for the average person?
Speaker: Or is it really even the companies that are building the tools and we're now consuming them?
Speaker: Is it just, we're just going to log into X or our social media platform?
Speaker: That's kind of our primary entry point to what Gen.AI is.
Speaker: So Joe worked at Meta and Facebook, so he probably has some good stories, but I'll say as a product manager,
Speaker: Look at products like Canva.
Speaker: Canva made everybody who was interested into a marketing designer.
Speaker: And I'm a Canva user.
Speaker: I pay their subscription.
Speaker: And it's got some incredible AI tools.
Speaker: And they did an incredible job incorporating it into their existing user flow, where if you had already taken the time to learn how to use Canva,
Speaker: then learning how to add some AI-generated info icons and things like that to your designs is a huge value add for your workflow.
Speaker: Same thing for like Grammarly.
Speaker: People already brought adoption there.
Speaker: They were using deep learning before LLMs, but I'm sure they have generative tools now.
Speaker: You've seen Notion adopt AI to help people write documentation, proofread, things like that.
Speaker: So we're definitely seeing like
Speaker: these intelligence and skilled work platforms drastically adopt AI and make it really widespread.
Speaker: Yeah, I think another big area, which we're already seeing is in actually creating applications.
Speaker: So, you know, writing code, so much of this can be automated at this point that, you know, I don't need to write functions for 95% of the things that I'm creating.
Speaker: And that other 5% is somewhere in there.
Speaker: It's just like, you know, a matter of reading through it and like reorganizing things and
Speaker: That to me is both exciting and scary because how much free reign do we start to give like a machine to go create these applications, right?
Speaker: Because again, you could do good things with it or you could do some really scary things.
Speaker: And I think, you know, John, what you were kind of mentioning is how do we start to prevent this or how do we equip people
Speaker: you know, lay people with the tools they need to identify what is good and what is bad.
Speaker: And I think, you know, there's very tiny examples of that.
Speaker: And like, you know, Gmail will say, hey, this is potentially a phishing scam or on my iPhone, it says like scam likely call.
Speaker: But there's so much more out there that we have haven't even, you know, reached the tip of the iceberg of
Speaker: what we need to do and how we need to build these behaviors in for the general public to be able to be well equipped to deal with AI and all the challenges.
Speaker: Yeah, I've been searching for the article that one of my coworkers posted on Slack today and I'm having trouble finding it.
Speaker: But the summation was they did a study of co-pilot submitted code.
Speaker: It's not just laypeople.
Speaker: These are engineers that are really, really smart.
Speaker: Also really, really lazy, no offense anybody, always work smart, not hard.
Speaker: But code quality and bugs introduced was found, I really wish I could cite some numbers here, I'm just man waving hands here without a citation.
Speaker: But ultimately these are people that have all of the wherewithal of hopefully being able to submit clean, good bug-free code at some point after good review.
Speaker: And we're seeing these tools be just
Speaker: do what they're supposed to, increase efficiency.
Speaker: This is supposed to be easier.
Speaker: I now have to spend less mental energy to check this thing.
Speaker: But now we have this weird problem where now I can very, very easily with just a simple set of words or copying over my code base, get a almost arbitrary amount of text back, whatever the kind of context window might be for any given model.
Speaker: it now creates a problem of having to validate.
Speaker: I have to now observe and look and see, well, crap, 10,000 words eventually got written or 1,000 words, whatever it might be.
Speaker: Now do I have to think the equivalent of making 2,000 words or something like that and thought units, whatever you want to break it down to now make sure that it's correct.
Speaker: And if we're seeing that with people using, I think- So I want to share-
Speaker: Two quick stories and then a book that Joe and I read that I think would be interesting.
Speaker: So Manu Sada is an economist from France, and he wrote a book in 2018 called Treconomics.
Speaker: And it looks at the economics of Star Trek, primarily Star Trek, the next generation, and talks about how realistic or unrealistic is this post-scarcity world.
Speaker: And he does a bunch of analysis in that.
Speaker: And it's really wild to me because he describes in that book constantly how
Speaker: We hand off work to robots in Star Trek, and it frees up humans for all kinds of things.
Speaker: And we're seeing that left and right right now.
Speaker: So the two stories I wanted to share is a paper drop from DeepMind this week or last week.
Speaker: They made a chat agent that's like a diagnostic doctor, and they tested it on 150-ish real use cases, and it beat doctors in conversational differential diagnosis.
Speaker: And I mean, is anybody busier on earth than primary care providers in North America?
Speaker: I don't think so.
Speaker: They don't have time to make small talk with every single patient.
Speaker: Imagine if you could have your intake be with a conversational doctor and then you get a detailed report.
Speaker: The doctor could just walk in, read what the LLM did.
Speaker: And on the flip side, you've got teens today who are 12, 13, 14 using TikTok.
Speaker: If you type AI diet into TikTok, you see a whole bunch of those kids hijacking like
Speaker: For example, there's a company that does AI-generated professional headshots, and they figured out that the professional headshots make you a little bit skinnier and a little bit lighter.
Speaker: And that's not great.
Speaker: But now they're using that as diet inspiration.
Speaker: And these kids are using AI to hack their homework, using AI to make their fitness photos look really good.
Speaker: All throughout, from specialized diagnosticians to tweens, there's generative AI applications happening right now.
Speaker: And I don't think like Pandora's box has been opened.
Speaker: It's going to keep growing.
Speaker: And I think we need to start like Manu Sadia says in that book, we have to start learning how to work with robots because they're going to take over all the work for us.
Speaker: Yeah, I think that's a super interesting point.
Speaker: I mean, you know, we obviously find ourselves at an inflection point in many markets at the moment.
Speaker: Like this happened to land at the same time that we had like a pretty giant macroeconomic shift that was like,
Speaker: outsized even past just tech.
Speaker: And kind of like the messaging that is pushed by a lot of these companies and their leaders that are creating these like Gen AI applications is, don't worry, we're not coming for your jobs.
Speaker: We're here to help, right?
Speaker: And I think that messaging landed for a little while, but now there's a little bit of fervor about the whole thing, given a whole bunch of layoffs that are starting to happen.
Speaker: you know, the SAG, AFTA stuff that occurred fairly recently.
Speaker: There's a lot of, you know, pardon the technical term, but like sweatiness that's happening in the market right now.
Speaker: So I'm curious, like, you know, to your point, Pandora's box is open.
Speaker: Like, there's nothing that we can do to really prevent any of this stuff from happening.
Speaker: I'm thinking like, you know, if this is T equals zero, what happens when we click and drag T out?
Speaker: Like,
Speaker: How far into the future do we have to wait before things stabilize?
Speaker: Like, what does that future look like?
Speaker: What needs to happen between now and then in order for people to feel comfortable working, you know, kind of alongside robots, if you will?
Speaker: Joe, you can go first if you want.
Speaker: Otherwise, I'll give him a whole ADHD rant about this.
Speaker: Yeah.
Speaker: Yeah.
Speaker: We talk about this a lot on the podcast.
Speaker: I think short answers we don't really know.
Speaker: But we end up always just talking about the breakdown of capitalism, I would say.
Speaker: No, I love it.
Speaker: Yeah.
Speaker: And that's just to say that, I mean, like you said, we have...
Speaker: big systems and institutions in place that are creaking under the weight of what we think we're seeing and like the Gen AI stuff and the whisper or like the promise of it.
Speaker: And I guess the short answer is like systemic changes.
Speaker: Andi, do you want to elaborate?
Speaker: So 10,000 years ago, 99.5% of humans worked in growing food.
Speaker: About 150 years ago, about 95% of humans worked in growing food.
Speaker: Today, that's less than half a percent of all of us.
Speaker: Technology to thank for that, number one is a piece of technology called a combine harvester.
Speaker: The chance of all of you having heard about this is almost zero.
Speaker: I highly doubt you've ever read it, right?
Speaker: So humans have gone through this before where some of our labor becomes defunct and few can do the work of many.
Speaker: In economics, this is called creative destruction.
Speaker: And it's not a pretty process.
Speaker: Like everybody will have the story of like, oh, when we replaced horses with cars, you suddenly had a lot more mechanics jobs.
Speaker: But like that doesn't mean the farrier who made horseshoes suddenly knows how to be a car mechanic.
Speaker: People's lives are inevitably changed and hurt in that transition.
Speaker: So we're going to see a drastic change in that, I think, in the coming decades.
Speaker: When our society has gone through these changes previously, it's government regulations that have helped prevent people from starving to death.
Speaker: Do I trust our current government to understand AI and save people and provide universal basic income when their jobs disappear?
Speaker: We shall see.
Speaker: Maybe not.
Speaker: And I think at that point, the thing that scares me a little is the people creating these tools don't even know the extent of things that those tools can do.
Speaker: Right.
Speaker: So the people that created GPT-4, they don't know every single thing that it can do.
Speaker: Right.
Speaker: They know that the concepts behind it and how to create it and how to extend it.
Speaker: But, you know, there's a prompt that it has never seen before that one day it's going to see and provide some output that no one has ever realized.
Speaker: And I think that's the difference when I think about this shift versus that agricultural shift is like the agricultural shift.
Speaker: you can understand those tools and understand the impact of it.
Speaker: I think here, what's different is like, there's so much that we just don't know how it's going to impact.
Speaker: And, you know, the thing that was surprising for me, right, I mentioned that some things aren't, you know, using as much AI as I thought is, you know, we, we explored an episode on diet and fitness and,
Speaker: And, you know, Andy mentioned that there's these people that are using these AI applications to generate images of themselves for that diet motivation.
Speaker: But when it comes to actual fitness and, you know, physically improving your body.
Speaker: there was very little impactful AI that I was able to see.
Speaker: And to me, that's kind of worrying because that's the actual thing that you should be doing is like, you know, let me go out and like move my body and improve my physical health.
Speaker: But there actually aren't a lot of AI tools that are having an impactful difference there.
Speaker: So I think that's going to be the shift is like, where is it really having an impact?
Speaker: Yeah.
Speaker: Yeah, that makes a ton of sense.
Speaker: And I
Speaker: I think I completely agree that this is all leading up to systemic shifts and it is unlikely to be pretty.
Speaker: Uh, we can only hope that, uh, the powers that be can help smooth things over, but this is also a global phenomenon, right?
Speaker: We have to extend past just kind of like the myopic view of, of the U S like this is something that's happening everywhere now.
Speaker: Uh, because again, Pandora's box is open.
Speaker: Um,
Speaker: I also want to like zoom in a little bit because I think we could talk forever about like the long-term implications of this, like of the shift, you know, tectonic plates are going to collide and, you know, it's going to be doom and gloom on one side and you know, happy go lucky roses on the other.
Speaker: I think, I think we all like kind of can see both sides of this.
Speaker: I'm really curious about, you know, bringing this almost to something that's like as concrete as we can possibly get it.
Speaker: Like to your earlier examples, like we've got,
Speaker: primary care practitioners, you know, like physicians essentially using Gen AI for intake, right?
Speaker: And then you've got on the other side of the coin, this almost like, you know, fitness toxic, like this toxic fitness type application.
Speaker: There's this like underlying thread across both of those, which is trust, right?
Speaker: Doctors have to trust that the Gen AI application is going to work the way they expect it to.
Speaker: with a fairly high degree, a high bar of sophistication, right?
Speaker: It needs to do at least as good as the physician.
Speaker: Otherwise, it's just not worth it, right?
Speaker: You are then, by definition, providing worse care to your patients, which I would hope no doctor would do knowingly.
Speaker: So that's one side.
Speaker: But on the exact opposite end, you've got a bunch of teenagers and so on that
Speaker: implicitly trusting this application that it knows what the kind of idealized form of fitness looks like.
Speaker: But as we all know here, AI is inherently biased and it's a fundamental problem.
Speaker: It's very difficult to diagnose, let alone debug.
Speaker: Ultimately, we need to provide interfaces that provide practitioners, users, some degree of confidence and trust in the system.
Speaker: And it needs to scale from the layest layperson all the way up to kind of like a domain expert.
Speaker: I'm curious about how you all think about that.
Speaker: Like, have we seen any of those interfaces yet?
Speaker: Do we think that they need to still be built?
Speaker: Who should be building them?
Speaker: Like, how can we trust these interfaces?
Speaker: Like,
Speaker: Yeah.
Speaker: So I think it's super important.
Speaker: And particularly from like a design perspective, I'm always trying to think about trust and safety and also explainability and that extends to governance.
Speaker: I think that's maybe the first step we take to sort of remedy this is to move slow, make sure it's safe, make sure people understand what they're using.
Speaker: I think, unfortunately, it doesn't always happen or it isn't happening, you know, when there's the prospect to move fast and make some money.
Speaker: I don't know if I have a good example of...
Speaker: anyone who's doing this very well.
Speaker: I have seen like open AI learn and do everything they can to eliminate the bad, you know, sometimes at the sacrifice of like proper information.
Speaker: But we talked a little bit in an episode about mental health and we sort of went into it expecting to hear about how mental health companies, tech companies,
Speaker: we were expecting to hear that maybe they jumped headfirst into LLMs and had basically established like robo therapists, because that seems like a very obvious use case.
Speaker: Like, like we have people who are chatting and want good mental health and Hey, there's this LLM that does an okay job.
Speaker: Let's have them therapize.
Speaker: And we actually learned that there may be moving a little slowly on that because they respect that mental health is something that requires great safety.
Speaker: And there was a quote from, um,
Speaker: one of the big players in like, I have this like better health or talk space that basically said, like, we feel the urge in this industry to jump on LLMs and really like get this into the product, but we're just sort of waiting to understand it and make sure it's safe.
Speaker: And that was very heartening to hear about.
Speaker: So I hope other companies are doing that.
Speaker: And sorry, I, I, I'm so curious, like the cynic in me wants to say,
Speaker: Like, yeah, maybe it's out of like, you know, good faith towards the mental health industry that they are not deploying LLMs or it's possible that government regulations are working as they should be and almost preventing that type of action where they might be deficient in other areas.
Speaker: I'm curious, like personal take, are these companies acting as like good citizens in the world or are these regulations acting as they should?
Speaker: I think- I'm gonna pop in there.
Speaker: Making stuff with AI can be hard and messy.
Speaker: We've all done it.
Speaker: I think one of the ways you cover your behind, if your investors are like, why haven't you shipped yet?
Speaker: Is you say, we're being responsible.
Speaker: So I'm being the deep cynic here.
Speaker: And also I think some stance that we've taken in our podcast is that like, it's on the individual AI practitioners
Speaker: the heck you're building because our government can't even keep kids in America fed.
Speaker: They're not going to catch up on regulations fast enough for how fast this technology is moving.
Speaker: So, you know, when you have jobs where you are injecting people with needles or you are being given a gun or you are the one that answers when you call 911, there's regulations, there's guidance, there's oaths you must take in order to be in that role.
Speaker: I think people in AI should treat their jobs similarly.
Speaker: You've been given a powerful tool
Speaker: The first time I was able to control the lives of hundreds of thousands of people with AI, I was 23 years old, fresh out of college, had no idea what I was doing.
Speaker: I really wish somebody would have made me take an oath about what to do with that.
Speaker: I'll also say that I think regulation is absolutely at play.
Speaker: Like even this year, the FTC actually find better help for a different issue involving patient data and advertising.
Speaker: But with that in mind, that probably discourages other like mental health tech companies to
Speaker: from being irresponsible.
Speaker: Sorry, I don't know if I said it, but that was the first time that the FTC actually penalized a tech mental health company this year.
Speaker: Yeah, and I think the other thing beyond regulations is something I'm seeing at my company.
Speaker: So I work at Primer.
Speaker: We're basically building applications for national security, government intelligence, stuff like that.
Speaker: And the thing that is really holding us from, you know, going crazy is actually the tolerance of our users.
Speaker: And in that field, there's essentially zero tolerance for any sort of hallucination or anything like that.
Speaker: And so for them, they're like, don't ship this thing, even though there's a cool demo and it could make our lives easier.
Speaker: Like, don't ship this thing until you're 100 percent confident.
Speaker: that it's not going to make something up and cause me to make a decision that could potentially, you know, change the lives of, you know, actual humans.
Speaker: So I think it is going to really be a combination of like, you know, this government regulations, but then also who are the users?
Speaker: And again, that kind of goes back to the doom and gloom is when you're putting something out to regular consumers, you
Speaker: you almost don't care about their tolerance as long as they're going to give you money or give you their eyeballs.
Speaker: And that's where it becomes really scary because these companies and these engineers could put out things that they think are really cool, but aren't actually spending the time to think about the impact that it could have
Speaker: If you suddenly put this out to a billion people and and there isn't the opportunity for consumers to necessarily push back as easily as someone like a government agency who's paying us, you know, millions of dollars to build something for them.
Speaker: So I see this, there's kind of like, there's two sides, you know, that we have to take into account.
Speaker: One, social.
Speaker: We all have to come together and understand these tools are going to be proliferating through pretty much all of our lives and touching almost every domain.
Speaker: We need to understand and build laws around regulation, around how to consume them, what they can be applied for.
Speaker: And on the other side, we have to be able to, and on the implementation of the engineering, we need to be...
Speaker: able to provide better answers when on the social side as a user, as an individual, I have to be like with my doctor, I can ask why.
Speaker: Might be annoying, he might bullshit me, he might lie, you know, that's humans, but we have a court system and we have laws and we have things, we have, you know, this whole entire system too, that's regulated around me being able to ask why and in those situations that are deemed, you know, critical or important or worthy of regulating, like healthcare decisions,
Speaker: There is a process that has to be able to be reproduced and said, yes, this is why I said based on these numbers, we have malpractice, the doctors have malpractice insurance for this very reason.
Speaker: Lawyers have an entire sector of society making sure that everybody's being honest.
Speaker: We don't have that today for LLMs in particular.
Speaker: And I think that the gap for the average person in particular is
Speaker: can I ask this model why?
Speaker: What does that mean when it gives me an answer?
Speaker: And that's a lot more subtle today due to architectural, you know, limitations.
Speaker: The transformer architecture that we have today is stochastic ultimately.
Speaker: We have no idea.
Speaker: And, you know, you said, you know, the creators of ChatGPT4, they have no idea what everything it can do.
Speaker: And I think, you know, the kind of the
Speaker: technologist in me finds that a deeply unsatisfying answer, honestly.
Speaker: And that makes it unfair for the other side of the social, which is required where I should be able to ask why.
Speaker: And if you as a practitioner can't give me a reasonable answer on something that I deem important or the government deems important via regulation, we need to have better tools.
Speaker: We need to have a better answer for that.
Speaker: And we will not see, I think, penetration
Speaker: of these large language models and quote unquote modern AI in those domains until the tech, the practitioners have a better answer for that.
Speaker: Yeah.
Speaker: And something that we do to actually get around this is
Speaker: force the model in its outputs to provide references.
Speaker: So we're always giving it input content and saying generate something based off of this and anything that you put in that output needs to have a direct reference back.
Speaker: And so again, within this small space of national security, there is a way that we're kind of working around that.
Speaker: But as you scale it out, how feasible is that to always do in every field?
Speaker: It's going to be really challenging.
Speaker: And for now, I'm really happy about the fact that almost every AI tool or feature has like a, this is AI generated, like caution.
Speaker: And I really hope that sticks around for a while.
Speaker: It took people 30 years to get used to like what UX is currently.
Speaker: And if all of a sudden everything gets really simplified and then there's a little sparkle icon that we now all know means AI and it means caution.
Speaker: It might be a little Delulu.
Speaker: Like that I think is a really good pattern to teach people.
Speaker: At my company, Virta AI, we help a lot of people, you know, integrate AI into their products and platforms.
Speaker: And like the standard best practice we give them is
Speaker: Be explicit with people that this is AI generated.
Speaker: Give them the little thumbs up, thumbs down icon, always right there so they know they can complain and you can keep an eye on the complaints.
Speaker: Don't try to sneak in any AI work.
Speaker: It's not going to work.
Speaker: And do you think that there's like, where's, where's Gen AI being applied or not being applied today where you think it should be applied?
Speaker: So for me, I think it's the childcare,
Speaker: scholastic applications there are still some industries that are heavily dependent on human labor that is deeply undervalued like healthcare in America would absolutely collapse if there wasn't a whole bunch of first generation Filipino immigrants which is what it is so why there are not tools being built to help our crowded classrooms our elder care setups I really don't understand
Speaker: Yeah, I'll let go of that.
Speaker: I just say more broadly, my favorite thing about the internet and data is how it brings more information to more people.
Speaker: Sometimes good, sometimes bad, but I think on the whole, very exciting and going to be a positive thing.
Speaker: And yeah, anywhere these LLMs
Speaker: and Gen.AI can bring information to people, particularly if it's going to help with healthcare or education.
Speaker: That's really awesome, as long as it is safe and good information and not tainted in any way.
Speaker: Yeah, and something I've been thinking of is like almost the like meta concept of using Gen AI to actually help build the government regulations.
Speaker: Because I feel like, you know, at the current pace, you know, like you said, of the development of AI itself.
Speaker: How are humans going to keep up with that?
Speaker: So maybe, you know, the AI can actually help us regulate itself.
Speaker: But then you kind of get into this feedback loop of, okay, wait, are we humans actually checking that work?
Speaker: Yeah.
Speaker: I think we're only weeks away from there being a dramatic story where like a piece of legislation went in front of the Senate and it says, as a large language model, I cannot help you.
Speaker: I'm sure that'll happen anytime soon.
Speaker: Yeah, I think it's a, it's a exciting time.
Speaker: And I think that if we handle it correctly and hopefully with podcasts like y'all's you know, we can weather the transition and be stronger and better for it at the other end.
Speaker: I think it's a long ways.
Speaker: And honestly, I think there's a lot of, a lot of,
Speaker: a lot of hard times and hard questions ahead and not much, and we're probably, you know, we're humans.
Speaker: If there's one thing we are fallible and, you know, we're probably going to make some mistakes, but, you know, as long as we keep, as long as we keep kind of the focus on what the implications in society are, what these things can actually do and making sure as practitioners that we either through hook or by crook via regulation, we're building these tools that are, you
Speaker: meeting the needs of our users and being ethically applied.
Speaker: I think there's a lot.
Speaker: We're going to see a lot, particularly over the next 12 to 24 months.
Speaker: Yeah, I've been the cynic of this episode so far, but technological growth has helped humanity so much.
Speaker: The fact that none of us on this call are farmers,
Speaker: That's really nice.
Speaker: I like that.
Speaker: It's good for humanity.
Speaker: And it's the min-max of this Jenny I boom is really interesting, right?
Speaker: Because could we live like a bad sci-fi movie, post-apocalyptic, there's the Elon Musk and everyone else?
Speaker: Yes, completely possible.
Speaker: On the flip side, could we be living in Star Trek in a couple of decades?
Speaker: Yeah.
Speaker: We're like, we just go up to our replicator and we go, oh, great, hot.
Speaker: And we're like, don't have to do anything else in our day if we don't want to.
Speaker: Also possible.
Speaker: I see these things as equally likely at this point.
Speaker: So like, I'd like to bet on Star Trek, you know, post-scarcity paradise, please.
Speaker: I just have to at least correct you a little bit here, Andy.
Speaker: I do have 10 chickens and take care of six acres of grapes.
Speaker: So technically, I think I could be in a very wishy-washy way, be declared a farmer.
Speaker: So I have to defend myself a little bit there.
Speaker: You should talk to Andy.
Speaker: Yeah.
Speaker: I don't know if you could tell, but I've got my own batch of blueberry wine I'm fermenting, right?
Speaker: Amazing.
Speaker: Oh my God.
Speaker: That's so fun.
Speaker: And specialty or true love is agriculture.
Speaker: So she could go on forever about it.
Speaker: Amazing.
Speaker: Yeah.
Speaker: Just moved out myself to get closer.
Speaker: Combine harvesters to you.
Speaker: I'm sorry.
Speaker: No, by all means, we'll take it.
Speaker: I'd prefer the mansplaining for the rest of the audience.
Speaker: You know, as farmers, we're few and far between, as you said, only 0.5%.
Speaker: So, you know, we are now the elite.
Speaker: But yeah, well, guys, I think this is a great conversation.
Speaker: I had a I had a lot of fun.
Speaker: Maybe as a wrap up, plug the podcast, anything in y'all's lives, you know, fun things you're doing this weekend.
Speaker: You know, if you guys want to take a take a turn.
Speaker: Yeah.
Speaker: Hey, I'll plug the podcast.
Speaker: Check us out.
Speaker: We're AI FYI.
Speaker: We're at AI FYI pod.com or find us on Spotify or wherever you get your podcasts.
Speaker: And we're also ramping up on the socials, TikTok, Instagram.
Speaker: So we'd love to have you.
Speaker: And we're going to be talking about a lot of AI topics.
Speaker: Yeah.
Speaker: And a reminder on our AI FYI podcast, we're really trying to demystify and talk about, you know, in lay terms, what is AI?
Speaker: How is it impacting us in different areas of our lives?
Speaker: So, yeah, come check it out.
Speaker: Awesome.
Speaker: Well, I will forward the podcast to my mom.
Speaker: I look forward to listening.
Speaker: And guys, I really appreciate the time.
Speaker: So just to wrap up, thanks everybody for listening.
Speaker: This is John Singleton and Cheyenne Mahante at the Grounded Truth Podcast.
Speaker: Kieran, Andy, Joe, love the conversation.
Speaker: Look forward to chatting soon.
Speaker: Thanks, guys.
Speaker: Likewise.
Speaker: Thanks for having me.
Speaker: Bye.
