All episodesCurated · With Hat and Cattle AI Literacy Series · 3 · 48 min
Vibe Coding Gone Wrong, All About AI Agents, and Rules for Your Data with Dr. Yeshwant Muthusamy with Dr. Yeshwant Muthusamy
Dr. Yeshwant Muthusamy separates useful AI-agent concepts from the hype, then walks through what can go wrong when we build by vibes alone. The result is a grounded guide to experimenting without handing over the keys to your data.
Play episode
Transcript Follow along0:00 There was this couple of founders who built this Tea app, T-E-A. And all of those photographs and driver's licenses ended up on the dark web within a week. 0:10 If you don't have the capability to go read the code and understand the code, then you're basically abdicating your responsibility, and you're hoping that the AI has done the code right. And you can't build strategy on hope. 0:25 Welcome back to With Hat and Cattle, where we bring you stories from people who walk the walk and leave you better off for listening. I'm the host, Dani Raschel Chou, and today we have in studio Yeshwant of Yeshvik Solutions. 0:40 He is someone whose career has spanned over 30 years in the technology field. And he is also a chief fractional AI officer, and that's a role I'm sure we're gonna be seeing more and more of. Thank you for having me, Dani. 0:55 This- it's a pleasure to be here. Yeah. So we are going to get right into it. When you started out in speech recognition at Texas Instruments, and I'm sure we're gonna span all your roles as we go through the conversation- Okay. AI was something that you almost didn't put at the front of your resume, right? 1:15 People didn't know about it unless you were in the know. You spent decades in this field, and now everyone acts like it just showed up in 2022. Does having lived through that make you trust this moment more or less? 1:30 I am always been very sk- skeptical whenever I see hype. Uh-huh. And the- these days there is no dearth of hype in the f- in the field of AI. Right. I cannot... 1:40 I've lost count of the n- number of times I've opened up my l- my LinkedIn feed to see people making all these tall claims. But I've been working in this field long enough to be able to kind of cut through the hype- and try to understand what really is happening, and, uh, and, and yes, my, and the number of years s- spent in this field has definitely helped in this, in this regard. 2:07 And for me personally, it's very gratifying to see the arc of development- Mm ... of AI. Mm-hmm. When, when I took my first AI class way back in, in, in graduate school- It was, Yeah, which was just last year. We know. Yes, it was several, several decades ago. 2:27 AI was the pur- purview of books and re- and, and schools and re- and research institutions. It was not something that people in the mainstream were really interested in or were really t- talking about. So it's great that the a- advent of, of something like ChatGPT has had the effect of throwing the spotlight on all of AI. Yeah. And that can be, at a very fundamental level, only helpful to the field. Yeah. 2:59 But also it has a, it has a, it has a flip side, which is that you have all of this hype- Yeah ... and this m- gross misinformation about the technology itself. 3:10 So I'm curious, what is the latest hype that you've set your eyes on that's made you kinda roll your eyes? This thing, I mean, the one that is pretty consistent is one which I hear is that AI is out to take away all of our jobs. We hear that a lot. 3:28 Uh, and that has been the refrain for a few years now, and I don't frankly subscribe to that. I think that the guy or gal who knows how to use AI effectively- is definitely going to take your job if you don't, if you don't upskill yourself, and that I think is true. But AI, at the end of the day, it represents a set of very powerful labor compression tools, the way I see it. And learning how to use those tools effectively is the key. Before you got to be an industry expert, 4:02 you were new in the field. You did not have, I hear now people are going to be able to know how to use technology and not understand the why behind it. 4:12 Basically, I can go and ask Claude to give me something, and I don't know the code behind it. It just- Right ... gives it to me. Right. Right. You start, and correct me if I'm wrong, you started off, you had to know all the why in order to get the result. Yes, I did. I mean, I trained my first artificial neural network way back in the fall of 1989, and I am dating myself here, but that's okay. 4:38 Uh, uh, But then in those days, I... you did not have the advantage of downloading a black box Python package- Right ... and just to execute that pa- pa- package and say, "I'm doing AI." You actually had to write the code yourself. Mm. And so if something didn't work or it s- or something gave you screwy results, guess what? 5:00 You had to go and walk through the lines of code to fix what was wrong. Yeah. And so that, that, that kind of helped me to make sure that I understood at a very fundamental level what anything was doing before I accepted it at fa- face value. Mm-hmm. But that, that's not to say that someone who's be- beginning to use a large language model these days should know how large language mo- models are built. N- no, you don't, u- u- unless you want to. 5:28 Uh-huh. But, but, but you don't have to. I mean, you, you don't have to. But what you do need to know is you need to know the extent of the capabilities- Mm-hmm ... 5:38 of that large language model. You need to know what information to give it and what you should expect from it. Mm. And most importantly, you need to know what not to expect from it. Be- be- which is often- That's- ... 5:51 a lot harder- Uh-huh ... to do, be- because you... it is not going to be the, it's not gonna be the silver bullet. It's not gonna necessarily think through every problem for you. Right. It can get you 80% of the way. 6:05 It can save you a tremendous amount of time, as I'm sure you have be- be- you have f- found out, and I use AI every day. Yes. And so it is a tremendous time saver, which is fantastic. But you need to be at the helm of everything that you do with it. 6:22 100% correct. If someone's listening to this and the, you are here as part of the AI literacy, literacy series that I'm running. I started that a few months ago, and AI has progressed by leaps and bounds. Yes. So beyond just learning how to prompt, which we don't hear anymore. Right. 6:42 And now is, are agents still something that the community should get to know, or at this point, should we already know how to deploy our own agents, and there's something new that we need to be focusing on? No, I mean, with agents, you have to be very careful because talking about agentic AI and AI agents sounds really cool. Uh-huh. 7:04 But the q- question is, do you have the proper AI governance framework in place- Mm ... to make sure that those agents are reined in, and they are doing the right thing at all times? You, you are getting into it immediately. 7:19 Which sh- goes to show that you are just in the know about this, and it's- Right things that you talk about. So I am delving into using prompts to do videos in Seedance- Mm-hmm ... and trying to get better at it, trying to get the consistent image and all that. Right. 7:36 And so I take my video, put it into Descript, which is an editing software. Yeah. And a couple of days ago, my husband asked, he saw me copy, paste, copy, paste, copy, paste. He goes, "What are you doing?" 7:49 I go, "Well, I'm trying to make this an hour long, but with the credits and everything, I could only, like, afford it the 10 seconds." And he said, "Can't you do that faster?" And I said, "I'm sure- But I haven't figured it out yet." So the next video I'm working on, Descript's made a suggestion- Mm ... on how to improve the video, and I just typed, "Hey, can you actually loop this to be an hour and extract the audio that Seedance put in, and so we can do another audio?" What took me maybe about 8:19 an hour and a half, it did in seven seconds. Yep. And it just- I'm not surprised. I just stumbled across it, and again, this is on me because I didn't do my research. But now that I know that, oh my gosh, all the time that it's gonna save. Yes. 8:34 It's gonna- Which goes to what you're saying. Yeah. Yeah. I mean, it's a tremendous time saver. But I'm sure you made sure that the f- that the final output was in your voice. Yes. 8:45 It, it was Dani who was, who, who was speaking. Well, I made sure the final output... It's funny you say that. I didn't quite check it. The vi- I didn't quite check the video thoroughly, and it's of my Pomeranian. It's cartoon. Okay. It had two butts. 9:00 It had two butts. And so I then figured out how to replace the video, and again, something that would've taken me so much time took- Yes ... seconds. Yes. Yes. 9:12 It was mind bo- like- Yeah ... mind-blowing. So Can you explain, maybe Give us, like, a two to three-minute summary what vibe coding is? 'Cause now we hear, "Oh, you should be vibe coding." Yeah. Yeah. 9:27 Which you, I have not done it yet because I'm just like, "Oh, my gosh," like, what does that even mean? Well, w- w- vibe coding is basically where you give the large language model prompts in English. Uh-huh. 9:42 And it builds the code base for you. Oh, okay. Uh, going back again to you won't need to know the why or the foundation. Right. Right. Okay. Right. So for example, you can vibe code an app. 9:57 Mm-hmm But, and here is where I have serious problems with vibe coding, okay? Okay. Uh, for me, I am completely okay with having the large language model do the heavy lifting of doing the c- actual coding of the app. 10:12 But here's the k- kicker. Before I show the app to anybody, I would want to read through every line of code myself. And make sure that it is doing the right thing. 10:23 But then that's just me. I do realize that, that, that that's just how I'm wired. I mean, for any software dev- development or AI development pr- pr- project that my firm takes on, even though I have my team back in India do the work- Mm-hmm I insist on doing the code reviews myself. 10:43 Yeah. So I walk through every line of code that they have written, because I realize at the end of the day, the client is gonna get something with my company's name on it. And if it's not doing the right thing or if it is doing something wrongheaded, guess whose head is on the line? Yep. It's mine. So I insist on walking through every line of code. So, but that, that's just me, but that has actually stood me in very good stead, because I've found out about bugs, and I've s- I've found out about gaps 11:16 that those guys missed. Because they are using AI too. I mean- Right. Yeah ... a- and, and that, that's fine. If you're using A- AI to code, I'm completely okay with it, but you need to know what the output is. 11:29 You need to be able to read the output at some level Which means you're, you're bringing your own so- software engineering expertise to bear to make sure that you are s- so that you are in a position to find out if AI has made a mistake. And trust me, AI makes mistakes all the time. One of my favorite quotes from Paul Byrne, who was a guest previously, and he's been to Mind & Machine, he's been a speaker, he said, "ChatGPT's a, a little liar." 11:57 And, you know, like, you are... Yes. You know, it's gotten better. But oh my goodness. Yeah, so but vi- vibe coding is basically where you g- you talk to it in English and it produces the code for an app or whatever it is. Yeah. 12:13 And I, I am okay with that I talk to it English and it producing code part. Uh-huh. But what most people don't tell you is that, uh, do you have the capability to go and read the code? Right. 12:26 If you don't have the capability to go read the code and understand the code, then you're basically abdicating your responsibility, and you're hoping that the AI has done the code right. And you can't build strategy on hope. Yeah. 12:42 I, I, i- in this regard, it might be worth mentioning this story really quick. Please. Uh, there was this couple of founders who built this Tea app, T-E-A. Okay. 12:53 I don't know if you've heard about this story. No. Tea app is meant, as the name indicates, tea, it is an app where young women can go to spill the tea- Okay about their most recent relationships. Ah, yes. O- okay. 13:08 So if you have just gotten you've just gone through a, a, a breakup or you've had a divorce, it is meant only for women, and the way you got onboarded onto that app is you have to take a selfie with your driver's license- Okay ... visible to the camera. Uh-huh. 13:27 What can go wrong, right? And so these guys went around bragging about how they used this vibe coding platform, and they went from ideation to the app being available in the App Store really fast, right? Yes. You, from the time they got the i- i- idea to the time that the app was available and was being downloaded. Great advertisement for vibe coding. 13:53 What these guys failed to do is check the security features of the database. Yeah. And all of those photographs and driver's licenses ended up on the dark web within a week. 14:06 It was a PR disaster of unimaginable magnitude for, for them- and for the vibe coding platform. Oh my gosh. Just imagine all these, all these women who had trusted this platform, their information, their driver's license, addresses, everything was on the dark web. And that is the classic horror story of why vibe coding, if not properly used, can do serious harm. Yeah. 14:36 Yeah. And so I use that example. I mean, it's an unfortunate example to use, but it is real. It actually happened. Mm-hmm. And it was all over the news. I mean, it was all over the, all over the tech news. Right. Right. Right? So, and but, uh, and wh- when I read that, I was like, "Really? 15:00 You guys, nobody in your startup company took the time to read the code?" If you had taken the time, you would have noticed the obvious security gaps. So that's why I say that by all means use vibe coding. Uh-huh. 15:14 But you better have someone take a look at the code that is output. Yeah. Because otherwise you're flying blind, so. Thank you for sharing that. You know, I, I wanna vibe code- ... a video game literally for me where I go to the Rainbow Bridge to see my Pomeranian who passed last November. Yeah, yeah. And I'm scared to do it because I'm like, "I have no idea what I'm doing. 15:39 I'm gonna release something that it's gonna give people just a bad user experience." But this has real world consequences. Right. Where, you know, and being a woman myself, you know, if there was someone on there on Spill The Tea who, you know, God forbid, there was domestic violence or something- Yeah and the ex didn't know where she was, and now it's out there with the photograph and the address on her driver's license. 16:08 I mean, those founders are very fortunate something, it didn't go to that extent. Yes. I mean, it is... I mean, I was shocked. I mean, th- things like this, I mean, it takes a lot to, to kind of shock me. Uh-huh. But I was like, "This is beyond." Yeah. 16:25 I mean, this is just bad coding. In your opinion, is AI governance moving faster or slower? Yeah. H- AI g- governance, depending on whom you're t- talking to and which part of the world you're in- Mm-hmm ... 16:39 can be a prickly to- topic. Right. Like here, uh, people equate governance with- overly strict restrictions on what you can do with AI. Whereas in Europe, with the EU A- AI Act- they've taken a more pragmatic, more measured approach. And the focus of the EU AI Act is that you should not... AI shouldn't be used to harm people. 17:02 And so I, I think the happy m- m- medium is somewhere in between the some of the overly restrictive aspects of the EU AI Act- Mm-hmm and what certain people want here in the US, which is the complete open wild, wild west kind of a- Yeah ... thing. Uh, but, but I would lean more towards what the EU AI Act is saying, because at the end of the day, all this talk of unfettered innovation, which you hear a lot of here from the tech bros, when you kind of peel back the layers of the 17:35 o- onion- Mm-hmm it's n- there is n- the welfare of the people isn't very high up on that list. And because, and you, and, and, and you, and you cannot a- adopt this, "Let's keep doing it until somebody gets hurt, and then we'll see what needs to be done." 17:53 And so what the guys in the EU AI Act have done is they've kind of, uh, they, they are being a bit more proactive in that respect. And so I think the happy m- medium is probably somewhere in between, but it probably leans more, much closer to the EU AI Act than- Are you speaking to the recent, and I'm speaking recent like days ago, where the, that clause came into effect, and now the EU's requiring that if something was made by AI, it needs to be labeled, um, deepfakes need to be labeled, which I 18:24 found interesting. My husband and I had an interesting conversation on that, 'cause he said, "Well, if it's fake, it's a deepfake." I go, "I understand that, but in my experience I've only ever heard of deepfake in regards to something that was a threat." Right. And so if a criminal is using a deepfake, it's, they're not gonna tell you- Yeah ... 18:45 "this was made by AI." Right. Right. But also other standards like biometrics and, and- Yes ... things of that nature. Is that what you're referring to? Yes. 18:54 Also what the EU AI Act is saying is that you want to make sure that the AI models are not are, are not e- exhibiting any kind of inherent bias- Mm-hmm or they don't end up doing things. Like for example, if you have an AI be in charge of, uh, be in charge of re- re- of recruitment. You want, you, you don't want it to show any kind of inherent bias towards certain ethnic groups. Right. Or, and, and those kinds of biases can creep up in the data even without you knowing about it. 19:29 And what that means is you have to constantly monitor and audit your data that's going into the model. You have to test your model, and you have to conduct periodic au- audits about how the model is doing. 19:44 All that is onerous. But, but it is needed. Mm-hmm. So. You know, even, so I do have people from different walks of life who listen to the podcast, you know, some who are supportive, they're in executive roles already. 20:00 For those that are listening that are, are listening to grow, you know, business leadership, personal development, they're thinking that's great, you know, but how does that affect the everyday person? You know, I understand business-wise we're not wanting to put all our customers' information into an LLM that doesn't have guardrails. But for me, what about if I put-- You know, I download my information from my Garmin, right? 20:29 All my health and all my recent runs and everything. Does that expose me to any danger if I put that into an LLM? Uh, it depends. I would say in general, a good, a good rule of thumb would be to not put any personally identifiable information into a large language model. 20:48 Now, most of the s- most of the subscriptions, even the paid subscriptions- Mm-hmm ... like ChatGPT Plus, Claude Pro, they make no representation of keeping your data secure. 21:00 I mean, th-there is a, there is a s- switch in both ChatGPT Plus and Claude Pro where you can say, "I don't want my data to train an AI model." You kind of flip that switch to off, and you think you're safe, but you're never really safe. 21:17 You have to go to a ChatGPT team or a ChatGPT business before OpenAI will make a legally binding representation that they will not train on your data. Interesting. 21:28 So in general, for you and me, a good rule of thumb is to not put any personally identifiable information. So, for example, to go, going back to your example of the, your health data. Mm-hmm. If there is anything in that data that links, links it back to you as a person, you should either remove your name or any other information that links it. But just pure health data, it could belong to any, any person. So the, so you just have to be care- the, the key is personally identifiable information. 22:04 Any information that can be tracked back to a specific person is a g- good idea to keep out of it. Thank you. We're gonna go to your Nokia years. Okay. 22:14 A lot of your Nokia years were spent inside standards bodies- Right Negotiating with competitors over things that customers wouldn't think to notice. Now that you're chairing an AI governance working group doing a version of the same thing, what do people get wrong about how these frameworks actually get written? 22:36 Yeah, that, that's an interesting, uh, q- question. I mean, in general, if your, if your company has, has said it is okay for you to go and participate in a standards group, it usually means that there is something that the company as a whole wants to get out of it. 22:55 And during my no- Nokia years, I was, uh, leading Nokia's team in the OpenMAX AL and OpenMAX IL working groups. And those, and that, that was part of this standards c- consortium that was kind of different in the sense that the whole point of why some co- company would join that standards group called Khronos was that you wanted to build a software ecosystem- - around these APIs. So IP was not the major, uh, uh, l- l- lookout. Mm. But most standards groups that 23:28 I've been a part of, IP is the thing. You go in there because you're trying to push your technology into the standard s- so that when that standard actually gets released, your IP is now part of the standard, and anybody who then implements that standard has to pay a small fee to your company as part of the p- patent pool. 23:52 I mean, I know I'm kind of g- going into the weeds here, but that, that's the reason most people join a s- standards group. Mm. And even though during my Nokia years, I had the a- advantage of working in a standards group where building the API ecosystem was the main goal and not the IP part, which was quite, quite refreshing because you're not trying to push your own IP in- into the standard. 24:18 But once I left Nokia and went to Samsung- Mm-hmm ... it was back to going into a standards group where IP was the big deal. And so I've, the point being I've sat on bo- both sides of the fence- Uh-huh ... 24:32 where IP is the big deal and where IP is not. So that gives you a really good insight into how people think in the room and what they say in the room, and you kind of learn to decode and read between the lines of what somebody from a different company is saying. Mm. 24:49 Because you know what that company is doing. Mm-hmm. And so then you kind of... It is a, the, the very, a very simple way that I like to de- describe working in any kind of standards group is it's like a heady mix of politics and technology- Ah. Yes ... 25:05 where you have to be on your feet. It is not enough for you to just know about the technology that, uh, that is the main, that is the focus of the standards group. You need to understand the, all the nuances of the politics in the room- Mm-hmm ... 25:22 as to who's present, why are they present, what are they trying to do, why did he say what he said. All those things you have to pay very close attention to. 25:32 Can, can you tell us how you learned that lesson? Ah, by trial and error. You just, the first few meetings you just, you just shut your mouth and you listen, and you watch, and you watch people. Mm-hmm. 25:45 And then you realize, "Okay, n- now I get it." And so then I found that I took to that whole environment l- like a fish to water. Uh-huh. I mean, I enjoy being in standards group. 25:57 I mean, I love that, I love the fact that we meet in these various locations around the world, but most people have a very warped sense of why I'm going to Nice, France, or why- I'm going to Seoul, South Korea. Uh-huh. 26:10 They all sound like great places to go to, but the fact of the matter is I spend five days of those e- of the, of that seven-day trip cooped up in a room with no windows- Yep ... arguing with grown men and women about, about te- about technology for eight hours a day. 26:27 Yeah. And at the end of it, we all go to dinner with the same group. So I get about half a day of actual sightseeing time. Uh-huh. But most of it is spent w- working, so. Right. But, but, but it's fun. 26:39 It's a, it's a different kind of- Yeah ... fun, so. So my husband listens, and he's in, in cybersecurity, so he's going to know all the acronyms that you give us. 26:49 What is IP? IP means in- intellectual property. I'm sorry, I, I, I should have me- mentioned that. You're fine. You're fine. He actually... There was someone who talked about S-A-A-S, and so when I first came across it, that's how I said it, and then the, the- It's SaaS. 27:07 The guest was so kind. He didn't correct me, but he did say SaaS. Yeah, sorry. I'll, I'll try not to speak too much of nerd. You... No, please, we welcome it. 27:18 So you were CTO, or chief technology officer, of a company whose core asset was a pat- patent licensing portfolio, and you yourself hold seven patents. Yeah. Amazing. First time I've met someone, I think, who's a patent holder. So I'm sitting here like, "This is cool." Thank you. Then you moved into generative AI, which is built on ingesting other people's work. 27:43 Did those years on the licensing side leave you with a position on that? I am very c- careful about what gen AI, I mean, what can be done with ge- generative AI. 27:55 And of, of course, you might have heard recently that a judge ruled that Anthropic's use of all those books, those mi- m- millions of pa- pages of books, was ruled as fair use. Yeah. But the only reason Anthropic got docked was because they, because how they got those books. 28:14 They actually got the books on a dark website. Ah. So the fact that they used those books to train was considered okay, but because of the sourcing of those books was from a shady dealer, they got docked. Wow. 28:30 So, but, but, but the key part of that, that ruling is that the judge ruled that the use of, of the books and the text in the books f- to train the mo- models was considered fair use. And so, I mean, it is okay f- to use the output of generative AI. Mm-hmm. 28:48 But li- like I said ear- earlier, you have to make sure that the f- final output that g- goes out of your door has your voice in it. Yeah. And is q- is actually representative of your brand. 29:02 So- You now have moved from the corporate world and you started your own business, Yeshvik Solutions. Mm-hmm. Yeah. Can you tell us a little bit about what made you wanna step away from the corporate world and become an entrepreneur? 29:18 Sure. I started Yeshvik Solutions in April of 2012, soon after- Wow ... uh, my stint at Nokia ended. Uh-huh. So I decided that I have enough of a, a bank of technical knowledge that I can kind of, I can then be able to leverage and have my own c- consulting practice. 29:37 And of course, back then it wasn't solely focused on AI. I was trying to le- leverage my expertise in m- media standardization- Yeah ... and offer myself up as s- standards consultant. 29:50 Mm-hmm. Which seemed like the obvious thing to do given the amount of time I had- Right spent doing that. I also said, "Oh, I do s- I do standards consulting and, but I also can consult on speech recognition and things like that." That, that's how it started. 30:07 But then w- within eight months of me starting my consulting firm, I got called by by Samsung, and I went and a- accepted a role in their audio research group. 30:17 But I have been very upfront, uh, du- during the time I've held roles in co- in corporate America, and I've been very upfront in saying that, "Look, I al- also have this, practice, my own, own, own consulting practice, which I'm not about to close down just, just because I've gotten a job here." 30:35 Uh-huh. "But I will make sure that there will be no conflict of interest, uh, and so on and so on." And but part of the reason why I s- started it was because I felt that if you're your own boss, the chances of you getting fired are much lower. Uh-huh. 30:51 And so being able to do your own thing wa- have a, wa- was a big attraction. But you're not only that, an entrepreneur in that regards. You have classes now where you have cohorts and you're teaching. That's how actually we met. Yeah. 31:06 I took, I took your class, and because I did not want to get left behind regarding AI, which... And it was, it was wonderful, and we're gonna talk about the, the cohort in a sec. I think I was... 31:18 Was I cohort seven or eight? Uh, this is, uh, let's see. I think I was- The one that I'm go- that I'm going to teach now, which is the second cohort of the, of the workshop which you took, is going to be cohort number nine. Oh, okay. Yeah. So that was eight. Yeah. 31:34 Yeah. You've taught now over 100 people who've walked in knowing nothing or very little, right? We're not industry- Right ... experts. Set the technology aside for a second. 31:45 How did the people change between that first h- cohort you taught and the eighth? That's an interesting question because my, all of my courses and my workshops are designed specifically for non-technical business professionals. The way I see it is if you have a tech background, if you have a programming back- background, then you don't really need to come take a w- workshop from me. You can, you can use your technical background to go watch a Y- You- YouTube v- video or o- open up 32:17 a book- Uh-huh ... because you have that background, and you can learn the new concepts by yourself. You might be able to understand the terms which are being used thanks to your background, and if there is any kind of programming involved, you are probably, if you already know one language, to try and pick up a new language is not a big deal. But I found that the largest, the largest demographic is non-technical business professionals who want to take advantage of 32:50 AI, but who most likely don't know where to start or how to start. When my cohort, there was 70-year-old accountant. Right. Yes. I was that so impressed and so motivated. Yeah. 33:02 I was in the right room. He's a, he's a tax advisor for high net worth individuals, and he's been in his field of work for over 50 years. Oh my gosh. And he took my first class. Okay. 33:15 He took the, my g- Generative AI for the Rest of Us- ... uh, course, w- which was the f- first course I took. He also took my next workshop, the AI for, AI for Business Growth workshop, and he's one of my most enthusiastic students ever. I mean, he is non-technical, about as non-technical as they come. And but he just, he just latched onto the possibilities that AI could bring to his line of work. 33:43 So, he and I talk all the time, and he kind of bounces off, i- ideas off of me on- Uh-huh ... whether Claude can do this, whether ChatGPT can do this. He's, he's pretty much taken all the courses I've offered. 33:57 What's a question that you get now that people weren't asking two years ago? Uh, that's, uh, that, that's interesting. Uh, the q- q- question they ask is what next? Yeah. 34:09 Right? And they want to know whether the job that they're doing now is gonna go away because it's gonna be really, really replaced by a bunch of AI a- agents. 34:20 Because AI is just advancing so fast. Right. Right. Yeah, it, it is, but it, but it is also the case that there is... You have to, you have to look at it in terms of what value is it bringing for you- Mm-hmm ... and not just embrace it just because it's new, right? Yeah. I mean, and that, that's my, the thing with this, all this hype about agentic AI. 34:43 Don't get me wrong, agentic AI is here to stay. Mm-hmm. But it, the, the kind of the well-thought-out, judicious deployment of AI agents is gonna be the key. Mm-hmm. 34:55 I mean, there are, uh, it's, I think of all the c- companies that have embraced agentic AI, there's a very small percentage that have actually dared to put them in production. Uh-huh. 35:08 M- most of them are still in the p- POI or the p- POC stage, pr- proof of concept- Okay ... stage. And so, and that, and that kind of, that, that kind of hesitancy is well-founded because you also hear all of the negative stories of, of what happens when AI agents, uh, e- effectively run amok, and- Yes yeah, so. 35:30 You- So you have to be... I mean, they have their, they have their place, and if they are properly deployed, they can do wonders. Yeah. But, but it all comes down to, at the end of the day, the human, the human design, because we are the orchestrators, right, of all of these AI a- agents. And, and that goes back to what, what I mentioned earlier, humans at the helm, not just in the loop. 35:58 When we hear stories, and there was one where they were testing an AI, right, and then they were asking it to shut itself off, but it had given it, like, fake access to a fake employee's email where the e- the employee was having an affair. Yeah, yeah, yeah. Right? Yes. Yes, yes. 36:15 And so the AI was threatening, was blackmailing- Yes the person. Like, "No, you're not shutting me off," right? Are these... When we hear stories like that, is- are that, is that a one-off or are those things we're maybe gonna see more and more of if people don't- Yeah ... understand how to? 36:33 I- if I, I mean, I don't know all of the details of that, but those I don't think are generally d- deployed models yet. Mm-hmm. Those are kind of artifacts of what's in the lab. Yeah. So I would, I would think of those as the kind of the bleeding edge models. Okay. 36:49 And, uh, it also comes down to what tweaks you have made and what freedoms you're, you're kind of giving those models. Yes. And we also ha- heard of this well-documented story of OpenAI models breaking through their ha- sandbox harness and going and attacking Hu- Hugging Face. Yes. Yeah. And, and, and I actually went through the blog that the guys from Hu- Hugging Face had written as to how exactly it happened, and it was a fascinating read. I mean, of, of course, it was a 37:21 geeky read- ... but it was, it was quite, it was quite fascinating how it chained multiple vulnerabilities- Wow ... and just went in. And so, and they actually put in actually code snippets which it had changed, and so that was actually a f- really fascinating read. Mm-hmm. 37:40 It gives you an insight into what can come down the pike if you don't put proper, guardrails around it. Mm-hmm. And so we are back to this whole thing about the AI governance framework. Right. So we are going to pivot because not only are you, you have all this background in, in the technology field. You hold seven patents, US patents. 38:04 You now go and teach what you know to other people. You have a PhD as well. You also are a gourmet chocolatier. Yes. Tell us a little about that journey. Sure. 38:15 I love chocolate. I mean, for me, when my wife and g- when my wife and I go out to eat, if the dessert course doesn't have chocolate, it doesn't count as dessert for me. So I need to have chocolate. 38:28 Uh, so, and I love to cook. I mean, if you ever take a look at my P- Pinterest feed or my Instagram feed- Uh-huh it has only pictures of the dishes that I have cooked. 38:40 I have nothing else on there. No pe- no p- no pictures of myself at some location or anything. It's just pictures of food. Nice. And it's pictures of food that I've cooked. So, I love cooking. 38:53 I mean, the, the whole act of creating something from scratch which didn't exist before, I find that to be a heady it's a high. Uh-huh. Uh-huh. So, and so, to me, uh, starting this business called The Blissful Gourmet was just a natural extension of those two things. Yeah. I love chocolate and I love to cook, and it was my friend and business partner who actually talked me into starting this. 39:20 And he said, "You are doing these really well. Why don't we make a business out of it?" And then here we are. So being that you make the gourmet chocolate at home, it falls under the, the Texas Cottage Food Law. Right. Uh, which means that The Blissful Gourmet company can't grow past a certain size. But you have a CTO resume, and again, those seven patents. You know exactly how to scale a thing. Yes. Yeah. So you picked a business with a ceiling built into it. So what does that ceiling give 39:53 you? Well, that ceiling hopefully does not last for too much longer. Okay. Because, I mean, I can very easily break that ceiling tomorrow if I move to a, to renting a c- a commercial kitchen- Uh-huh to actually ma- make the ch- chocolates. 40:08 Yeah. That is what distinguishes a, a business that is m- making the food in their home kitchen, and that's why it's called a a Texas Cottage Food Law- Yeah ... compliant business. 40:19 But if I were to move to a commercial kitchen, then that restriction completely goes away. Mm-hmm. And I wanna do that. I mean, we want to d- do that, but we wanna do it organically. Okay. I mean, it's like I don't want to build a four-car garage when I just have a small bicycle, you know? Mm-hmm. Yep. I want to be able to buy the four cars first and then bui- bu- build a- ... build the four-car garage for it. Yep. Yes. 40:45 So that, that, that's the way I lo- look at it. And I want to get the word of mouth spread f- for the time being, and we are actively going after the corporate gifting market. Oh, okay. So you have a stra- of course you have a strategy. Uh, I mean, we are online. 41:01 Uh-huh. We have an online store. And we'll link it in the show notes. And but I also w- but we are also in talks with banks- Mm-hmm ... financial advisors, and with some big law firms. 41:14 And I recently did a w- wine pairing event in, for Fort Worth, where I we- where I actually supplied the boxes with, with, with very specific chocolate truffles, which were placed in the specified order to match the se- sequence in which the wine was being tasted at that event. Oh, that's amazing. 41:35 And- That sounds so good. So, and, and that was a great exposure for me and for my friend who was hosting that event. Yeah. You had every credential to go be part of the Bay Area version of this story, but you built your practice here in North Texas. What does the AI conversation sound like from there or from here that it doesn't sound like in San Francisco? In San Francisco, uh, that's a good I think there is enough AI business here. Uh-huh. 42:05 So I've never felt that I'm missing out on s- s- something. Plus, I like the s- central location of the Dallas-Fort Worth area. I like the low prices. You know? 42:17 I mean, let's be frank here. Don't get me wrong, I love g- g- going to the Bay Area. I love the Golden Gate Bridge. And every time I went there to visit family or friends or for the business, I used to ask myself, "Am I making the wrong decision in not moving here?" And each time the a- a- answer from inside my head was, "No, you're not." 42:40 We have a pretty vibrant A- AI scene here. Before we let you go, I'm curious, and I feel this is going to be asking you like which one is your favorite kid. Out of your, your patents, 'cause we don't have time to talk about all seven. Yeah, yeah, yeah. 42:57 Talk to us about your favorite one. Well, the, the f- favorite one has to be the one that is assigned to my own consulting practice- Okay ... and on which I was the sole author. 43:08 And so that is something that I came up with. It was my idea. And I actually managed to get it done with just one office action. Um, and, and here, once again, I'm go- going into nerd speak, but an office action is the number of times that the US Pa- Patent Office- comes back to you and says that, "Th- this is not right. Go change something." Oh, I just filed the trademark application- Yeah for With Hat and Cattle, so I'm waiting for, to see if I'm- Oh, yeah ... gonna have an office action 43:39 or what. Yes. Oh, no. So, I understand. Yes. So, but thankfully my p- all the claims in my patent were, they all went through after just one office action. Uh-huh. 43:49 And so I'm pretty proud of that. That's, yeah. And the fact that that patent was just all me, because, and it is assigned to my own practice. Yeah. Because most of the o- other p- patents I wo- I worked on, I am on the patent, but the assignee is what matters. Right. And those are the firms that I was w- were working for. So, this one that's assigned to you, can you sum it up in a minute for us- Sure ... what it's about? It is about providing a safe parking 44:17 environment for students in schools. Wow. In school p- p- parking lots. And at the same time, so it kind of, uh, it provides a safety net- Uh-huh ... for in school pa- parking lots, and it also allows them to make use of t- temporarily free parking spots which may not be at the grade at which they bought. Okay. 44:35 So, like for example, if you go to U- UTD, you have these green spots which, which are- Yes for the plebs, and then you have the- ... the yellow spots and- Yes ... the orange, and the pu- purple spots is for the people who are on the top of the foo- foo- of the food chain. The royalty, yeah. Right. It's purple. 44:54 But if you, if you've ever gone by one of the purple spots, most of the slots are empty. Yes. That the students are fighting with each other for the green spots. Mm-hmm. And they end up parking in a spot that they're not supposed to park for, and then the car gets booted. Yeah. But what if- On the days where you know beforehand that X number of purple and orange spots are gonna be free, you allow the students on their ha- on their handsets to pick how much they are, how much they're willing 45:25 to pay for the day- Uh-huh to go use a, a, an orange spot or a p- a purple spot. That's gonna be much cheaper than getting- It's gonna be much cheaper- ... the boot removed ... for their parents. Yeah. Yeah. 45:38 If, if their pa- if their parents are one, are the ones who are going to foot the bill for the pa- parking ticket, right? Right. And so in the long run, what happens is the school makes extra money. 45:50 They don't have to have people roaming around booting cars, because that whole thing just goes away. There is no concept of parking enforcement. Everything is handled in the app, but there is a, there is a cloud backend which is using AI, and you also know where every student is if they opt in. 46:09 And so it's a s- it becomes a safety and security issue. Yes. So- Oh, that- That's, that wa- I love that. So that, that was the That, that was the idea. 46:18 And yeah. Well, I hope it gets implemented, because- I hope so too. Yeshwant, thank you so much. Oh. It has been just... I know for myself it has filled my cup just sitting here with you- and learning. 46:30 It's been fun talking. I- Yes, learning about AI ... i, I love talking about aI, so. Yeah. It's, well, it's very apparent, yes, that you are passionate. 46:40 Thank you for sharing that, again. Thanks for having me. was a lot of fun. Before you go, here are a few things. The first is a phrase I like to leave you with. Humans at the helm, not just in the loop. 46:52 Being in the loop means you are standing nearby while something happens to you. Being at the helm means your hands are on the wheel. And notice that's not a technical skill. 47:02 That's a posture. The second one was quieter. You can't build strategy on hope. We said it about code, about shipping something nobody bothered to read, and I've been turning it over about a lot of things that aren't code related. 47:17 And then of course, there's the Tea app. Thousands of women handed over a selfie holding their driver's license because they trusted an app that promised to keep them safe. 47:27 But nobody in that building read the code. That's not a technology failure. That's a failure to look. Which brings me to what I actually want you to do. 47:36 I took Yeshwant's workshop, and I want to be clear about why. Not to become technical. I took it because I didn't want to get left behind, and because I'd rather be the person who understands what she's handing over. 47:50 When you have the opportunity to take a class, take a workshop, join a cohort, do it. Not because AI is coming for your job, but because you deserve to have your hands on the wheels too. I'm Dani Raschel Chou, and this is With Hat and Cattle. We bring you stories from people who walk the walk and leave you better off for listening. 48:12 I'll see you next time.