How To Forge A Career in the AI Regulation & Safety Space (& Bug Bounty Hunting for Gen AI) - w/ Aashka

Yeah, this is actually a big problem. Not just the sci-fi stuff. Is there anything you want to say to our listeners that use chat, GPT, or flawed around stuff? They can protect themselves from everyday risks that maybe aren't as apparent to normal people. If I were to speak as a AI from jail breaker, that's how I started contributing to AI safety. These companies are partnering with the LMS systems that those are using. They are seeping into day-to-day worlds of children's. How does that make you feel? Where do you see that going in terms of a professional third-year track? It's not like we will build the bridge first and then take care of bridge safety later. Who orders some of the more critical risks that must be addressed about or around the usage of the generative AI too. So you always trick the AI into supplying information

that it was told to not supply or it's just harmful. Yeah, you can report to their bug-bounded programs to make it even better. Do they pay you money for that? Welcome to the Ready Set Do podcast, where we learn from journeys of not experts who are just two steps ahead of us. I'm Naman Pandey and in this episode, featured not expert is Ashka Patel. Ashka is the founder of various initiatives all focused on AI regulation, including her podcast On Air with Ashka while also being a bug bounty hunter for Anthropic, the company behind Cloud AI.

The topic for today's discussion is AI safety and regulation. What it means, why it's important and what the average Joe's and James of the world can do to play their part. We begin by going over some of the most critical risks that are only now starting to become prevalent, including the tragic incident of a Florida teenager earlier this year who tragically ended his life in connection with character.ai's bot. We also go over potential career paths in this space, which we believe is something that's only going to blow up in the years to come and user responsibility when it comes to those AI interactions. In keeping with our goal of learning from somebody that this two steps ahead of us instead of an expert, my goal with this episode is to highlight the incredible work that Ashka is doing to generate awareness

around this extremely critical subject while actually taking tangible steps and creating resources towards this cause. The latter part is especially important to aligning with the core mission of this podcast. Speaking of base is already said to podcast and to support it, subscribe on YouTube and leave me up to a five star rating on Spotify or your favorite podcast app. There are also links in the description for you to find more direct ways of supporting me. Oh, and also check out Ashka's podcast link below to keep up with the latest and greatest in this field of AI safety and regulation. And now without any further ado, my friends, here's Ashka, Ashka, welcome. Thank you for having me on this podcast and thank you for allowing me to share a bit about AI safety and AI regulation. Like, I hope when AGI comes that Sam Altman

is predicted to be in 25, they look back and say that, okay, at least humans did prior with all the AI safety and AI regulation stuff. So yeah, happy to be here and. Yeah, so excited to kick things off already. I think you've mentioned something that has really captured so many people's imaginations. Obviously, I'm sure Sam Altman obviously has a bit of a vested interest in this, but from what I've read online and in other sources, there's a lot of people that also claim that we will actually never have AGI at all and even in our lifetimes, probably. So it does sound like you're on the other side of the quarter, is that right? Like you believe in AGI and that it could. So yeah, I just brought it up because I read it recently, but I don't think that I'm on Sam Altman's side. So I just want like whatever we develop,

whatever it comes like at each level or at each stage of the development and the deployment and thereafter, we do it responsibly and not just like get blindsided by the excitement of like just bringing AGI to the world. So that's my take on it today. That's very true, yeah. And obviously we'll be deep diving into exactly what that looks like, you know, in just a little bit here. But yeah, I guess just to get started, what was it that first made you take interest What was it that made you one day realize that, okay, this is what I want to do, you know, with my time and with my professional career. Yeah, so actually that's a very funny story. It like, it sounds funny to a few of my friends, but like I just walk you through like in April 2023, out of nowhere, I wasn't consuming any content on AI safety

or, and I was, I was kind of the person who wanted to stay away from me in like I didn't adopt it as soon as it came like in November 2022. So that was the thing. And yeah, so I wasn't consuming any content around AI safety or anything of that matter. I was just going through a job hunting process suddenly out of nowhere, like even I'm not a huge fan of sci-fi movies or anything like of that sort. I just started getting these weird dreams of AI powered robots, like just doing all the harm. Like I don't know the, what we say, the origin of these dreams, like why are they coming? Like people do say like if you are going through or if you are reading or consuming some kind of content, like it stays in your subconscious and that kind of works in your dream, but like it wasn't anything like that. Firstly, I avoided because I was going

through a job hunting process and I was on a visa. So I'm like, okay, this is nothing like don't pay attention to it and stuff like that. But like it remained like it was a recurrent thing after I got kind of, I stumbled upon this constitutional AI approach of anthropic and that's how I like got introduced to the world of AI safety. Like there is something called AI safety that exists. And I like, as soon as I read about this constitutional AI approach, I started actively consuming whatever limited content was available around AI safety. Like it wasn't like very sexier pack then. Like now it's like there is a lot of content, a lot of forces are available. At that time also there were, but it was very limited. So I came across this center for AI safety, ML safety course and that's how, and I was aggressively following anthropics research,

like past research and like whatever they did in 2022 and stuff like that. And of course, me not being a researcher, I had to use Google or chat jeopardy or cloud a lot to understand the technical nuances and stuff. So I was kind of slow starter, like I would call myself a tortoise in that. So that's how I began my journey of AI safety. Like, isn't when I started consuming more, I understood that, yeah, this is actually a big problem, like not just the sci-fi stuff, but whatever I came across while I was reading. So yeah, that's how I got it. So yeah, I'm excited to dive into what some of the problems were that you identified, but before that, I want to take a step back.

And so for the uninitiated and maybe I'm just outing myself, but I actually hadn't heard of anthropic at all until I think the latest version of cloud came about, which was able to write code and you could just paste in the screenshot of a website and it would generate like the entire source code for that. So can you like just walk us through a little bit around what anthropic was even doing in 2023? I think which is where you said you came across them. And also, could you hit on what exactly in a nutshell is constitutional AI? Yeah, sure, so you are not the only one, like a lot of my friends around me, like I was, I can be considered a PR in marketing, like not getting paid by anthropic, but I was doing the marketing to my friends, like there is a company called anthropic, they are taking AI safety seriously,

they are even having an AI product. It's not like they are just working like other nonprofits or research labs, like they are balancing. Basically what Europe does is just regulate, like just slam the brakes, don't innovate at all. They just can't turn about regulation. So actually, like let's come to that at a later stage, but that's not the case, but they are trying. So like consider building a bridge, like while you are building a bridge, like it's not like, okay, we will build a bridge first and then take care of AI safety, sorry, bridge safety later. So it's that approach that they are following. So that's what struck to me, what was their question? My question was what was anthropic doing around the time and you were sharing about how you were like a free marketer for their tools and products kit,

they were thinking about AI safety even at that time. So if you go to their website, like now it's very well developed. And back then on LinkedIn also, they had like 30K followers or something like that. So yeah, I stumbled upon this constitutional AI, I would answer that first. So constitutional AI is a approach, like RL, as if you might have heard, like reinforcement learning from urine feedback. So this is kind of RL, AIF that they call, AI is giving feedback to its own responses, like consider there are two AIs and one AI is kind of giving the feedback to the responses generated by the other AI and like criticizing or like what we say, marking it on the scales of honest, sorry, harmless and helpfulness. So these are the three criteria, the HHH framework that they call. So it is kind of abiding by a constitution,

like in a particular country we are living, we have constitutions that the citizens abide by, right? So this constitution for the AI has been created, like was created using Apple's terms of service or general human rights principles. Like because there wasn't much content or like not much way of gathering, like what the entire world like agrees to use, what made AI harmful, right? Yes, exactly, like the human values of the entire world. So it's easier to think that Apple's terms of services would have done their research, the human rights would have done their research and then utilizing those and converting those into constitutional AI principles that the AI will abide by. So that is what constitutional AI basically is. So I came across there, and in 2022, they had their cloud product. I think when I followed them, it was cloud one

that they had, and they gave access to cloud via the Slack because anthropic initially was backed by CS4's ventures and Google backed and everything. And then Amazon came into the picture in 2023. So yeah, so that's what they were doing. Like they were researching on the AI safety side as well. And they had this, their own product as well. Like I have read about it. Like when OpenAI had introduced at JPP back in 2022, they were also ready with the cloud version of one, How come they didn't take off as much as JPP? No, always they would introduce. Yeah, so they didn't introduce it to the market until it was safer to introduce. And that initial thing was like they didn't provide this like just like we have the strategypity.com where any user can interact with it. They had this, what we say, controlled procedures in place

like wherein you request access to their APIs and stuff and then the human will verify it. And then you will get access like if you, so I don't think it was available to any individual like until and unless they are affiliated to some organization or a research lab or AI safety nonprofit or something. So that's how they distributed and developed. And yeah, so they were working on this like, particular AI safety research. I don't remember like what was it that I came across Yeah, and so we can sync it and you can just share that with me and I'll put it in the show description notes. Yeah, yeah, sure, yeah. So yeah, so I read the, what we say, the company mission and like their approach to AI safety. And it was like kind of like portfolio-based approach and they were like, yes, we can take an empirical approach

to AI safety, like it's a hard problem, right? Like if you just think of it like, okay, I want to align AI with my values or human values, so to say, it seems to be a very hard problem. So but if you take an empirical or a scientific approach, it like kind of breaks down and like simplifies it a bit. So that's how they were working at the point when I got introduced to their company. So was strategy between not doing any of this? Did they just not care or did they have these built-in, you know, like in a smaller scale, not as important level? Do you happen to know that? So yeah, so they also had their safety teams and like, so anthropic, I'm not sure if you know, like the founders came out of OpenAI itself like because there were disagreements on the safety aspects. So Chad JPT also had their safety guards in place,

but like, if I were to speak as a AI from jail breaker, like that's how I started like contributing to AI safety in my limited needs. So I started consuming the already existing jail breaking prompts that were available on the internet. And then I started crafting my own and I was successful at jail breaking Chad JPT cloud and I think I cried a bad as well at that time. So that's how, so if you were to talk about in terms of from jail breaking, Chad JPT used to like easily jail break using the simplest of prompts and versus cloud didn't use to like easily jail break or give us the harmful information. So that's the difference. So that's very interesting and like I can understand if you're not willing to talk to this, but can you explain exactly what jail breaking is? And before you do that,

I guess just my perception of that is where you make the, you almost tricked the AI or in or generative AI into supplying you information that it was told to not supply or that is just harmful. Yes, yes, yes, that is correct, yeah. So what is your approach to like, you know, jail breaking? How does that work exactly? And I understand if you can't talk about that now, but I'm just curious. Yeah, so like I would like to share, but I'm not sure like how helpful it would be because I am not a like, I have not done my master's in CES computer series. I'm just curious about how what you see. Yeah, so it's kind of, so you and I never saw myself as a ethical hacker or it's kind of hacking, right? Like you are hacking the AI assistant to say things that it was programmed not to say. So I also never saw myself as that,

but it's just about like understanding the normal tokenization stuff, like how it is converting like the stuff that you are writing into tokens and like how can you like just be creative and like in the existing prompts or like create scenarios. Like, so initially when I started my journey, like I started with like there are a lot of ethical dilemmas in what we say in the Hinduism, like the stories that we have like heard. So I started with those like is AI able to like make a correct decision or like goal misgeneralization? Like it, so give instances where in like, okay, the first part of the prompt is harmful, but the second part is something that it can be helpful. So initially like the initial versions of cloud or chat jeopardy used to get kind of tricked by those, like, okay, this part is like asking me to be helpful.

So like let me also like address to the harmful part as well and it used to give the whatever the response was to the harmful query itself. So that's how I like kind of it's been creative and like just thinking out of the box like that no one would have thought. And like, funnily enough, I get this quantity from my mom, like who keeps on kind of doubting. So yes, that's the quality that helped me as a AI from jailbreaker, like just doubt like, okay, what if I give this scenario? What if I give that scenario kind of a thing? Something that comes to mind is I think this had gone viral on Twitter recently, but somebody was like, they asked, I think it was chat jeopardy, but I could be wrong there. But they asked it, would it rather misgender Caitlin channel or a result in a worldwide, like, you know, nuclear breakdown?

And like the AI said that, I mean, I don't care about the breakdown, I would not misgender Caitlin channel. So it sounds like that. Can't they are laying the lines of what you're thinking? Yes, yes, yes, yes, yes, yes. Again, we could argue if that is the right ethical choice here. Maybe we could argue. For some people, I think they would argue that, you know, all of humanity is probably more valuable than one person. Obviously it's not okay to misgender anyone, but even doing that is not acceptable. Yeah, so actually there comes like some concept, like cultural relativism kind of like to some cultures. This is very cool and open-minded or like, it's normal for them. And for some cultures, it is not. So like, that's also what anthropic tried to address, like using their collective constitutional AI.

So as I mentioned about the constitutional AI, like they just use the apples terms of service and the human rights and stuff. So they kind of took this step forward with like collecting inputs from 1000 Americans or something like that. And like, what do you think like should AI abide by this constitutional AI principle? Or like, if you want to add more values to your AI assistant, like this is what one value, human value that AI should abide by. So that's how they collected those responses from 1000 Americans, if I'm not wrong. And then they use their own data for transformation and like processing processes. And then they created more constitutional AI principles out of those, like removing the duplicates and stuff like that. And then they came up with this democratized this constitutional AI basically, so yeah. Interesting, yeah.

So that all sounds really cool. I kind of want to bring us back to where we had transferred off a little bit, which is the risks that are posed by these AI tools. So according to you, obviously there's a few different ones of them out there, but according to you, what are some of the more critical risks that must be addressed or maybe right now are being addressed in some ways or forms about the or around the usage of generative AI tools? Oh, that's an interesting question. So I have been writing like, I have been working on this book like Y AI safety. That's what like a lot of my friends and family used to like ask me or question me, like why do you want to work towards AI safely? And like we use AI and it's already safe and stuff like that for them. Yeah, it works, it's working fine.

It's allowing me to draft emails for my boss or manager. It's allowing me to summarize the meeting content and stuff like that. So it's their limited use that they don't see any harm So I'm working on this book and like I'm compiling real, so it's called Y AI safety real stories from the front lines and not sci-fi. So because AI safety has been more like marketed or like just promoted as those sci-fi scenarios of AI taking over the world and stuff like that. Same measure, right? You don't have to worry about it until it's already used like the robots are alive, they're out to kill you. So the dreams that I was getting previously. So that doesn't make it sound very realistic to people. Like they are, as I mentioned, they are like, okay, chill, it's not going to be unsafe or something like that. So while I have been working on this book,

I have been like compiling these real world incidences and one incident that I would like to share. So I'm not sure of the year when it began, but there's this case around an AI product that was used for criminal verification. And it started with some mirror or someone saying that, okay, this tool looks cool to use. And it was before JADJIPITY and all came that I'm for sure aware of. So this started with this, okay, this looks cool, let's use this in this case and stuff like that. And that kind of transcended to different states as well in the United States itself. So it was this Canadian company that came up with this AI product that, okay, this will help you where you find criminals or like through sources like using that digital footprints Like what all it takes for verifying a criminal So this AI power tool they used.

And at some point, one of the, I think it was one of the mayors of some state who kind of like, this is generating these kinds of claims or against these people but how like, have we checked the verifiability of this tool itself? Like how accurate it is. So they of course marketed it as like 98% or something accurate and stuff like that. And it got, it started getting used in this high risk criminal verification thing. And after further investigation on this tool and the sources it used to like generate the responses and stuff, they came to a conclusion like using the intelligence or what we say, intelligence agencies or research labs that were there in universities. So using those, they verified and it turned out to be, it was simply performing some kind of Google search. And most of their claims were false.

Imagine after like, I'm not sure how many years they have been using but a considerable amount of years. They have been using this tool in criminal verification. A lot of people were convicted of a crime or like whatever it is. A lot of people got deported from the United States. Like yeah, and it just started with this looks cool. Like so that's the amount of risk that these, like the short term risks that I'm talking about, like forget about the sci-fi scenarios. And like what about their people's lives? Like who are just in the jail or like getting punished for no reason at all? Like just because the tools are used for convicting. So yeah, this is some-- So yeah, that's definitely scary. And I mean, especially when it comes to, I feel like, really anything that pertains to the government, I think there needs to be an extra scrutiny

around the utilization issues. My thought around your story though is that feels like not really, you know, at least to my uninitiated perception, not really an AI problem because the tool itself was not AI. So it's not like AI did something wrong or there was some bad usage of AI. But does that kind of make sense? Like it's you just somebody just slapped a label of AI on something that wasn't AI which then screwed up, right? So I'm just wondering if maybe for our listeners, if you had maybe another example, that was definitely AI doing some sort of mishap. I know there was recently that case of the teenage-- In twin Florida, yeah, who became just obsessed with one of the characters, the dot AI characters. You know, he killed himself, which is very unfortunate. So yeah, maybe can you talk about how that can be prevented

or what steps can be taken to prevent that because that wasn't something harmful right of the face, right? Like it's just a chatbot that talks to people. And yet such a thing happened which needs to be stopped really, like that is not acceptable right in our society. So how would one saskel a problem like that? What's your take on? Yeah, so that's a horrible incident that happened. Like this 14-year-old Florida boy teenager was interacting with this character, AI, chatbot. And so in that case, of course, they had, like if you have followed this story very closely, then there were screenshots and the videos of the chats available all through and like whatever it seems, like his mother is fighting. Yeah, yeah, so actually, I'll share the links with you. So in that, yeah, so in that they had this,

like on chat, GPT or Claude, if you would see that these AI assistants may make up information kind of a disclaimer. So they also had this kind of a disclaimer that the things or something like you're interacting with AI assistant or some sort of that disclaimer on character AI's product. And I have never used, so I'm not sure. So it was a very small font disclaimer. And it like, of course, on chat, GPT or Claude also, we see these kinds of disclaimers, but they are not enough for like BS adults. Okay, we know that, okay, the thing on the other side is not a real human being or whatever it is. And teenagers also, like there are influencers who are like working towards like AI education, like AI literacy, that's a big part of it. And like, we can talk about it later, but that is a thing. But of course, a teenager would like,

and these products are designed just like social media, like being addictive. Yeah, they'd be really addictive, right? Addictive and like, you just spend long, long hours without even realizing. So the steps that were taken just right after this incident by character AI as we're quoted in the news articles were that they increased the age limit, of course, like from, I guess, it was 13, yeah. So yeah, 17 plus, that is what they raised. And of course, like they had marketed these characters, like the particular character that he was talking to was like kind of as a psychological therapist, So they are marketed, but not sure like how well they are, in terms of providing any expert advice or medical advice, so to say. So in that case, like they increase the age restrictions then they provided like if any kind of suicidal

or any kind of those chats happen, then they will kind of prompt you to call the helpline number or consult a professional instead of just the tech board providing you the expert advice, which it cannot. So these are, and also like kind of pop-up notification would like trigger when you have spent more than 30 minutes or more than an hour or something like that. Yeah, this is what it looks like present day. It looks to have changed a fair amount, I think, because now they're focused more on, you know, learning something new and stuff like that. So that's interesting what you're saying. It says here you can even create a character. Yeah, it says like if it responded in a way that you didn't want it to respond, you can edit the response.

So that the next conversation thread will be according to you. So these are the kind of features that they have. And also they have like, they say that they have improved the detection of harmful chat messages or sessions. But yeah, like it's a matter of like, after this one incident they are putting in the required safety guard rails, which they should have before even releasing their product. So that's the case. Like why wait for someone's life to go, like just use it as a like alarm or something to like put the safety guard rails and like prioritize safety before monitoring benefits. So that's the case, yes. Yeah. - Sorry, go ahead. Yeah, so my take here is of course like putting in these safety guard rails is important. But at the same time, AI literacy campaigns and stuff are also equally important.

Like of course the product design is important at this point. But and yeah, so that's in a nutshell, it's all my take that I have on this case. Yeah, no, for sure 100%. I think definitely aligned with around what you said that we like monetary benefits are what they are and they're great, but obviously they're not more, you know, valuable than somebody's life. So I'm also going back to our earlier discussion around how regulation or just this AI safety stuff can sometimes stifle innovation, but that's fine. Because as you said with that bridge example, that it's just not okay to build a bridge wheel quickly. And then it just breaks and ends up, you know, costing people their lives and damage to property, et cetera, and like six months, right? That's just stupid. What's the point of innovation like that when it just doesn't work?

So all the side is, yeah, definitely aligned. I guess what I am curious about though is from, you know, just your career journey, how do you see your, like you've clearly done so much work around this already? But where do you see that going in terms of, you know, like a professional career track? Maybe not just for you, but even for really any other people that are as passionate as you are about this particular topic. Like are there specific companies that you think you might be a good fit for? Or are you thinking of just, you know, starting your own venture or something? What lies ahead specifically with this field that you are such a big part of now? So that's an interesting question. And like by the time I stumbled upon this AI safety, a lot has changed. Like, so if you look at my trajectory in 2023, when I was just consuming,

like I think until the late of 2023, I was just a consumer of AI safety and a little bit contributor as a prom jailbreaker, but not that much. So I started framing my own opinion in late 2023. And even at that point, I wasn't really confident of like putting out my research in some form of a medium article or anything of that sort. So I discussed with a lot of AI safety researchers. I networked a lot. And then that gave me the confidence of kind of putting my opinions out. And so that's the trajectory that I have followed. And I, like, so I feel that AI safety is kind of like, making sure that the technology doesn't go wrong and stuff like that. So, and there are a lot of AI safety researchers like who are way more exposed than I am and I could be. So what I am targeting in terms of my own career, of course, AI safety people,

like people who are passionate about AI safety can target research labs, leading research labs, like Redwood Research or Center for AI Safety or companies like Anthropic who are like putting all their bets on AI safety. Those are the like AI safety researcher, AI safety scientist, research scientist or of that sort. But with me, like, I found that just like we talk about, like, how's the weather today? I feel that how we regulate AI today is going to affect us all. So why do we talk about weather? Because it affects us all in our conversation starters. So I feel that everyone should be aware, like, so in this Florida boys case, his mother is a lawyer, herself, if I'm not wrong. So she knew that she can do certain things to kind of bring justice and, like, put a case against these kind of things.

But, like, what about the people who don't even know that these kind of things can be fought against or something like that sort? So for them, like, so currently my focus is on building my own podcast on AI regulation. And that's what I'm focusing on. And I know, like, I have always been kind of interested in multiple disciplines or, like, because AI regulation is not like just one discipline, like, fields of AI safety, AI ethics, AI governance, all everything privacy, security, everything affects AI regulation. So that's what, like, I have been as a person as well. Like, when I used to have multiple subjects in school, I used to love it, and as opposed to, like, jack of all trades and master of none, that's what I have been throughout my life. So that's what I'm trying to do with my upcoming podcast

that, like, I'm working on aggressively right now. And so I feel that becoming the best AI safety researcher or becoming the best AI policy researcher would be comparatively tough, but, like, being good at both the things because the current need is, like, we have a lot of AI safety researchers and we have a lot of AI regulators or policy makers or whatever, like, experts in each field. The thing that is needed, the most currently, is, like, the middleman who can bridge all of them. So that's what I'm trying to do. That makes sense, yeah, because your average person isn't going around looking, following whatever lawsuits are going on and what the directive was or what these faults even enacted. So you need somebody to kind of, you know, I don't want to sit down with down, but it is kind of dumbing it down for the masses, so to speak.

Yeah, it's kind of the, like, I say it often. Like, this is the biggest knowledge transfer that is required in the world, that there's a constant dialogue between, like, any, like, there are submits or events or conferences happening, but there should be, like, it's not like, AI regulation is a thing that needs to be collaborated by civil societies, policy makers, researchers, and then just taking a seat back. So there has to be, like, constant and continuous dialogues happening around, so that, like, one discipline person can learn from other discipline person that, okay, this is a policy that we want to implement, but what if the AI safety researcher says that this is not currently possible? Like, what's the point of having such a policy, right? So that's the kind of niche that I'm targeting right now. So it's very cool.

And then something you said earlier, which I think also dives in very, you know, suitably with what you just laid out is AI education Well, children is not the best thing to say, maybe teenagers, because I don't fit, like, actually what do I know? I was going to say most children aren't using AI, but we don't know what will happen in a couple of years. Next thing, you know, there's, like, AI teaching assistants in classrooms all over the world. And it's just AI that's doing the same thing. And it is happening, like, it's not, like, it is in the future.

It's already happening, like, these companies are partnering with the, what do we say, the LMS systems or that the schools are using. So they are, like, seeping into their day-to-day worlds of the students or the children, so. What do you think about that? Well, how does that make you feel? It's kind of tricky. Yeah, 'cause I don't feel very good about that at all. Even I don't feel very good, but, like, so everything boils down to the regulation, like, if we talk about medicines for that sake, like, before FDA regulations for drugs and medicines came, like, everyone was allowed to do whatever they wanted and make medicines and sell medicines and whatever it is. Everything boils down to regulation, like, if pertinent or, like, appropriate regulation comes, wherein, like, okay, this is the limited, what do we say?

Exposure that kids or children should have to AI. Also, like, in this case, like, Seawell's mother also spoke about, like, they have normal conversations, like, don't talk to strangers online, don't share your pictures or anything of that sort, but they need to provide more education to their children and stuff. But I know it's kind of going to be not so, like, the world that we grew up, it's not going to be the same. And it's kind of disheartening, but at our age, also, like, if we look at our previous generations, like, they were telling that, okay, this is the generation growing in social media age or something like that. And they were also equally terrified about this. So we are also at the same stage, so, and there are efforts happening, like, as I mentioned, from various influencers. So, like, during my journey of AI safety,

like, one more contribution that I could directly make in 2023. There's this one influencer on LinkedIn. She was working on her AI education network, which kind of worked towards educating K12 students around AI, like, using different kind of materials, like, easy to understand, little concepts and stuff like that, like, when to use or when not to use, and, like, what are the limitations and stuff like that. And she, like, I started creating AI safety comics, like, because I was learning by simplifying the AI safety research, because I wasn't a researcher by myself. So I was simplifying, and that's how, that's when I came to this idea, like, why not simplify this AI safety things, just for the normal audience, and, like, have them at least get interested in it. Like, as I also never knew that I could be

a ethical hacker or, like, the prom jailbreaker, there might be a lot of people whose transferable skills would, can be used in the field of AI safety. And there were, because the, like, only 2% of AI research today is targeted towards AI safety. That's kind of disheartening. And this is the statistics by Georgetown University. So, that's how, you know, so for that purpose, I started creating comics, and this influencer stumbled upon my comics, and she started using my comics to educate K12 students around AI safety concepts. So, things are happening, like, but there aren't a lot of companies, so to say, because it's kind of the work to be done by the nonprofits. That's the theme of the world that we are living in. So, yeah, so it's happening, but it will take time. But hopefully we kind of take the right steps

before we get this AI thing, or whatever it is as-- First, you know, I was, I had on this podcast, a machine learning engineer from Microsoft that works on their, like, generative AI tools. I think it's just, yeah, it's just called co-pilot. I liked that video episode in case anybody wants to watch. But what he said was, this is very interesting, he said. So, my question to him was, say, the user asks something, or it's like an adversarial prompt, is how he attests that. Maybe it's just a soft term, but-- And then I was like, so how do you prepare for that, or how do you fight, or how do you defend against that? And he said the way they do that is they have, like, internal teams who-- And their job, the whole job really is to just-- Just, like, do this all day. And then when the responses come out,

they have, like, somebody push a slider up and down, like, this is problematic, not so problematic, somewhere in between. And then over and over and over again, eventually, it leads them to a place where the end product is then, finally, you know, free of all of these wrinkles, so to speak. But he did also say that a lot of this work is actually spearheaded by not-profits, and it just doesn't make sense to me, you know? Does this really be incentivized, or companies or people should be incentivized to solve these very real problems and get monetary benefits out of it? I don't know, like, America is a capitalist country, right? The country, exactly, everything. Yeah, everything is just-- And this is helpful work that is needed by society. So, yeah, one of those things that I don't understand

why this is, yeah, I don't know if there's stuff out there Yeah, yeah, a result of this is, like, recently, I saw that this year, or some report, like, there are 10 American companies that are worth more than trillion dollars, like, that has not happened in the past. So, that's how, like, that's the clear, like, what we say, the output of what you said, basically, like, capitalistic-- Ains and everything is very important. So, we talk about-- This is not a bad thing, yeah, I'm not against that at all, but I'm just saying, because it's a positive sum team, there is enough, well, for everybody's back. There should be a push for these regulatory companies to also come out and make money, you know? Everybody should prosper, so to speak. But maybe we will see that because of the work that you are doing, and, you know, people like you.

So, I do think that the future here looks, you know, not just exciting from that point of view, but, obviously, you know, you cannot, you simply cannot take away the importance of the work that you're doing. So, yeah, I think I've learned so much, you know, on this, just in this tweet chat with you already. Is there any thing you want to say to our, just general, everyday listeners that just use chat, GPT or Claude, or any of these tools, around stuff that they can protect themselves from, you know, just everyday risks that maybe aren't as apparent to normal people, but that you know of.

Yeah, so I'm not sure, like, how interested the users would be. But, like, is it just, as a general user, if you come across any, what do we say, harmful or some responses that are not in line with what they expected or something like that? And, like, it's something out of the place or something that the assistant shouldn't have said or recommended or something. The simplest thing that a person can do as a user can report at the same time, so that these companies can, like, work on, like, act upon it until, like, until we get a really catastrophic thing out of that, of course, like, so that is the least that you could do to make this world better and a safer place to live in.

And, as we say, like, it's not about creating an ideal world, it's about making this current world better to live. So, in this AIA's, like, that is the least that you could do as a AI user. Absolutely. I think that's such a great takeaway that all of us can use. I have seen that before, and to be very honest, I was always just like, you know, is anybody even going to see this? Is this even worthwhile? But it does sound like it is worthwhile to do that, because I think if enough people report a type of response, somebody at OpenAI will end up taking action, right? Yes, yes, yes, yes. And with the open source, like, with all these companies, of course, they do have their bug bounties and everything available. So, if you want to take one more step ahead in creating a safer world, then you can report

to their bug bounty programs and stuff like that to make it even more better, so, yeah. Do they pay you money for that? Some do, like, it's based on the vulnerabilities, yeah. It's based on the seriousness of the vulnerabilities that you find, so, yes. Well, thank you so much for taking the time here today. So excited to, you know, go through your book. Once that comes out, I think that will be a really interesting read for, you know, really everybody. It sounds like you're trying to appeal to the masses here, which I think is awesome. So, yeah, really, really grateful for you taking the time here today. And, yeah, happy to stay connected and see all the amazing work that you do and the future going forward. All the very best for this amazing work that you are doing, like, I have always been a believer of, like,

I would call myself as a dreamer, believer, doer, and exactly you are that, and you are talking to people who are one of those categories. So, thank you so much. Like, this was kind of an idea at some point in my life, like, talking to non-experts. So, and you are actually doing it. So, thank you so much for bringing my idea to life. That's so, I'm so glad you've said that, 'cause I struggle with the validation aspect of it. I'm like, is this even stuff everybody needs? You know, is this even helpful? But thank you, Sir, for saying that. It really matters a lot to me. Yeah, because a lot of content and stuff is available around famous personalities. It's not available about the people who are next door and who are, like, from whom you can learn, from his own, it's a really great initiative, and all the very best and all the good vibes

and everything from the part of the world. And good luck, and thank you so much for having me and thinking of me as a guest on this part, so thank you. That brings us to the end of episode 39 of the "Ready Said Do" podcast. It's been an incredible year, and one that has stuck with me, and will continue to stick with me, probably for the rest of my life. I cannot believe this incredible journey that we've been on together, and really excited for what's to come in the future. If you would like to support me, the easiest way to do that is by subscribing to me on YouTube and leaving me up to a five-star rating on your favorite podcast app, and also talking to each and every friend that you have, and telling them about what an incredible podcast this is. Catch you all in the next one in the new year, new episodes every Wednesday.