Asia Tech Podcast recorded an insightful conversation with Michael Vandi, Co-founder of Addy AI. Addy AI is a secure AI email assistant that writes your emails for you in seconds, with your preferred style and tone.
Some topics discussed by Michael:
- How being away from familiar faces helped Michael focus on his startup’s progress
- The effectiveness of categorising and nurturing one’s strength
- Using AI to augment the productivity of humans rather than replace it
- Why Michael is worried about AI being taught how to go astray
- How having a teacher as a parent shaped Michael and became a very valuable lesson
Some other titles we considered for this episode:
- Being Able to Complete It
- Improving Productivity
- Providing Data Strategically and Objectively
This episode was produced by Stephanie Ng.
Read the best-effort transcript below (This technology is still not as good as they say it is…):
Michael Waitze 0:04
Hi, this is Michael Waitze and welcome back to the Asia Tech Podcast. Today we are excited to have Michael Vandi, a Co-founder at Addy AI. Michael, it’s great to have to Michael’s on the show, it’s gonna get really good. How are you doing? I’m doing great. Thank you for having me. It’s my pleasure. It’s also good to have somebody on the show who like understands how audio production video production content production works. It just makes my life so much easier. Anyway, before we get into the main part of our conversation, I just like to understand a little bit about your background for some context. Where are you from? How did you get here? That kind of thing? Yeah,
Michael Vandi 0:38
thank you so much. Well, first got on the call. I think it’s great that we start off with with my background. I so I grew up in Freetown, Sierra Leone, I was I was born in Sierra Leone, born and raised. And then I moved to America for undergrad and the University of Baltimore, started computer science. And then from there, I worked on a lot of projects at the University of Baltimore, I worked on this project called the NASA suits projects, where we built spacesuit informatics displays. And then from there, I went to AWS. I was a software engineer working on distributed systems. And then I quit AWS to do grad school at Carnegie Mellon. And at CMU, that’s when I really dived into the startup culture and Startup Grind. And I started a few companies. Addy AI is actually my fourth company a little bit. I’m starting this year, and it’s been the one that’s had the most traction so far. And I’m working on it AI full time.
Michael Waitze 1:36
Can you just tell me for people that aren’t familiar, what the what’s the right word? What the environment is like at Carnegie Mellon, and particularly what the startup environment is like, part of the reason why I’m asking is because so many people hear about, you know, Berkeley, Stanford, the MIT Media Lab, and I think sometimes, Carnegie Mellon, which is an insanely incredible university, my cousin went there as well. It’s just like, it’s not forgotten, but it doesn’t have the same impression. In that sense, right, as a as a school, of course, great reputation. But when people think about startups, they don’t naturally go there. What is it like there for that?
Michael Vandi 2:09
Yeah, it’s like it’s a conversation that we have internally and the school was talking to an alumni today, like, why don’t people think of CMU when they think of startups? And I think the culture of the school has been set with like, engineering rigour. Yeah. And less so about entrepreneurship. Okay. And it’s only now that I’m CMU POS. You know, getting into the entrepreneurship environment. When you look at Stanford, Stanford, is all about entrepreneurship, like, top founders, like startup is the place to go when you want to have an MBA, you want to work on startups, you want to be an entrepreneur, no, like the Google co founder when they’re, and all those people. But CMU has mostly been about computer science. I think we were the number one school for computer science for a while. I think that we’re number two, Stanford, it’s number one. It keeps fluctuating whatever that means, right? Yeah.
Michael Waitze 3:06
And let’s be clear about this to tell me this. Like if you’re in a room of people, and someone says, I went to Berkeley, I went to Stanford, I went to MIT, you’re not embarrassed at any level. Do you know what I mean? Like, because if you’re me, and you’re like, I went to Connecticut College, like okay, maybe I should leave the room kind of thing. But if you want to just stand up with pride as well. Yeah. So it’s the same level for sure. Yeah.
Michael Vandi 3:29
Yeah. And we think Stanford CS is a little less rigorous. So Berkeley, I see Berkeley has the toughest CS programme. And like, that’s the that’s the consensus, and then maybe MIT and CMU on the same level. Yeah. Stanford’s more about going there to start startups. Yeah. Fair enough.
Michael Waitze 3:51
But also talking about this, too. I think people sometimes forget, like, just how big America is. And when you go from the East Coast, right, from Baltimore, and from Pittsburgh, or Pennsylvania, and just start building stuff in Silicon Valley. That is also different, too. And maybe you can tell me how you feel how different that is?
Michael Vandi 4:09
Well, number one is the weather. It’s much, much warmer here. It’s much gloomy and colder and on the East Coast, moving from Baltimore to to California. But notice that Baltimore had a lot of Lake enterprise influence. We’re in like T Rowe Price, or some of like, government stuff is going on there. Like I used to live five minutes away from NASA Goddard Space Flight Centre. So a lot of the things on the East Coast, around the DMV area, at least where it used to be DC, Maryland, Virginia. It’s mostly like government influences and like established enterprises. And then as you move over to the west coast in Silicon Valley, it’s more of like startups. Yeah, and people starting companies. So that’s an on the professional side on the personal side. So a lot of people that I know my family lives in Baltimore lives in. So I would relate more to being in the Maryland area, because I would, you know, walk down the block and see my fourth fifth sixth cousin, that I wouldn’t know how to relate to and then moving to the West Coast. I didn’t have a lot of like family connection here. So that sort of gave me a time to like just really focus on on startup like, I haven’t had Sierra Leone and food in like two years. Well, I deeply miss it. But also it means that I don’t get invited to the baby showers. I don’t get invited to the to the events, which means I can spend all my time working them out now.
Michael Waitze 5:51
It’s a balance right? What is Sierra Leonean food anyway?
Michael Vandi 5:54
Oh, Sierra, Leonean food is a mixture of carbs like rice. Yeah. And, um, mostly rice, and then a mixture of vegetables. So, um, you would related mostly to spinach. Oh, if you’ve had Indian food, it’s quite similar to food where it was like the rice and there’s like the sauce on the side with spinach. And then it’s different kinds of different kinds of plants but just relating it to spinach we have one called cassava leaves. And cassava leaves is like the quintessential Sierra Leonean meal for cassava is some kind of tuber. Cut up the leaves and just make it as a sauce. Eat it with rice. And then have a few. Few Meeks in there. Got it? Yeah, that’s, that’s pretty much it. And there’s also if you’re familiar with the West African culture, there’s this thing about who has the best Jollof rice? Yeah. And so if you want to make a West African mad, you just say My country has the best to love rice than you and looks like it’s gonna be an endless it’s been an endless fight. But terms of fighting words here.
Michael Waitze 7:12
But do you find this really interesting, too, and we can talk about this later? Maybe a little bit more length? Did you find it interesting, too, that the world knows so little about Africa itself? Right, a country, not even a country see what I just did. It’s a continent with like 50 Something countries, 1.4 billion people, one of the lowest mean, ages in the world. So one of the youngest continents in the world. And with the opportunity to build in so much growth, particularly with all the tech changes that are happening. I’m I could go on and on about this. People don’t think about Egypt as part of Africa, but a big venture capital businesses are growing there. And then in the middle, what is it Rwanda and Burundi as well, are also developing big financial services centres based on what’s happening in Dubai and Singapore. Like there’s so much happening there. Are you in touch with all of that as well,
Michael Vandi 7:54
a little bit, I am more in tune with what’s happening in the States. But Africa is a blank slate for innovation in a way, right? Yeah. I tell people that I know. In Sierra Leone, I say, every single tech innovation can be applied to Africa. Like you don’t even have to think about ideas like in America, you want to think about start starting a ride sharing app has like 10s and 10s, of ride sharing apps, but in Sierra Leone, because like maybe one or zero, yeah. So it’s like a blank slate for ideas. And you, you just have to transfer ideas, put a little spin on it. And just a lot of the work has been done on the west and the West, but thinking about ideas and then testing it and validating it. So there’s lots of research there. All you have to do is implement it and apply it and Africa and like some of the things that you would need is of course, capital, but also the infrastructure. The infrastructure might not be completely formed yet infrastructure. I mean, like internet infrastructure, cabling,
Michael Waitze 8:55
just everywhere, you know, satellites focused on it, or even just like underground digital cables, right? Yeah, it just needs it all needs to get built. Let’s do this. I want to get back to that. I want to make a note to ourselves to get back to that in a second. But one of the things you said about CMU was that there wasn’t the same level of startup culture there. And yet, you said you started four or five companies, did you and you you couldn’t have done it alone. But did you do it? When you were doing this? Like what was the encouragement there? Because like you said, if you’re at Stanford, if you’re at Berkeley, if you’re at MIT, everybody’s doing that. So everybody feels compelled to do it. But that wasn’t the same for you. Right. So why did you do it? Where did that come from?
Michael Vandi 9:33
Well, I was lucky to be in the Silicon Valley campus because CMU has a campus in Silicon Valley wherever suffer engineering programme is based. So there’s a lot of external factors that influence here and not only the internal CMU factors, but just being in the valley in general. And fun fact, I’ve actually taken as much entrepreneurship classes as engineering classes. I’ve taken more classes than I needed to graduate. Second, so many entrepreneurship classes. And that’s where I met a lot of these people working on this entrepreneurs. And one of them who is my co founder now unnati AI. And so it’s mostly full, external environment, I would say, that has impacted me. And like, my engineering department at CMU wasn’t really supportive of all the companies that that have started, I’ll tell you that innovation department is quite supportive. What does that do? I mean, there’s, there’s one of two things there. Um, you know, in academia, there’s not a lot of this, like super smart professors, you know, smart people, but there’s like, this culture and expectation of wanting to work for the biggest tech firm, or wanting to do like groundbreaking research, which is amazing. But there’s very little interest in starting companies, at least in the engineering departments. So I was
Michael Waitze 11:02
having this conversation with someone last night, and I’m really curious about your opinion here. Because you’re just so much closer to it than I am, because you’re just younger, right? So you’ve come on to university a lot more recently than I have. You know, you said people are smart. And maybe it’s just like an age thing. But as I’ve become older, I think I’ve realised that, like, this definition of smart is starts to become really interesting, right? Like I like to the argument I was having last night was the fact that you’re good at math, or that you understand physics means you have a skill. And it’s a skill that’s hard to attain, but it doesn’t make you smart. And I would posit that hitting a baseball is also a skill. And the people that are good at it are really good at it. But it doesn’t make them smart. Do you know what I mean? And I feel like we should start saying like, that guy is really good at math, or she’s really good at tennis, as opposed to he’s really smart, or she’s super athletic. I don’t know. I’m trying to figure this out. I’m wondering what you think.
Michael Vandi 12:01
Yeah, I think there’s certainly different kinds of intelligence for sure. Like, some people who are really good at cognitive intelligence, they could remember stuff. Yeah, there’s people who are really good at dicksterity. You know, so they could do sports. You know, they’re people who are good at emotional intelligence. So they could be be good talkers and good at social intelligence. So yeah, I think we should categorise what people are good at. So then try to nurture that from a very young age. Yeah, that
Michael Waitze 12:40
that I completely believe in part of the conversation we’re having last night was, you know, that guy is so smart. Why doesn’t he understand this? And I think that’s the reason why it’s like, just because you’re super good at this thing doesn’t mean you even have an affinity to anything else at all. Anyway, kind of a different topic. So it is your fourth company’s at fifth company. I can’t remember. Current fourth company. And what does Addy AI do? And I’m really curious, because there’s so much conversation going on in AI right now. Right. But what does it do? And I want to know what changed? Technologically right? Because we’ve been talking about artificial intelligence for 50 years or more. I mean, Turing was talking about it years ago, hundreds of years ago. So what changed to make it so that it’s so effective today, and what’s the idea behind it?
Michael Vandi 13:25
So, Edie AI is an AI ml assistant, that writes your emails for you, within seconds, with your preferred style and tone. So it learns from your writing style. And then it writes your emails for you just as how you would do it. So it helps save you time it helps manage your emails for you helps you get to zero inbox. And, you know, managing emails and scheduling meetings are two of the most consuming things in the workplace, like a third of the average person’s time, and the workplace is spent writing emails and sending emails. And so what FDA does, and it comes it comes in to automate fat. So that’s what it does. And what makes it happen is like, the generative AI VM that we’re in, you know, pioneered by Chet GPT, which is this huge language model that was built by open AI, which is what Eddie AI is currently powered by, okay. We’re looking forward to, you know, augmenting the capabilities or just doing way more with with generative AI. And previously, we didn’t have those these large language models with like billions and billions of parameters. So the tech was just not there. But now we have we’ve started seeing the potential of you know, generating human like human sounding text. And this could really revolutionise the way we work and the way we write.
Michael Waitze 14:56
So for people that may not understand what I mean, it’s hard enough for I think for most People don’t understand just what artificial intelligence is right? But can you talk about the generative side of it so that people can get a better sense of it? It doesn’t have to be a classical definition just the way you think about it.
Michael Vandi 15:10
Yeah. So generative AI is mostly about two things. It’s for the natural learning language part of it. Yep. It’s like natural language, understanding, and natural language generation. So an overview in NLP. Natural language understanding is like looking at a piece of text, and gaining inference from it, getting insights from it, and understanding what this text is about. And generating responses from that or completing, completing the thoughts that are in that text. And there’s another that’s the language part, it’s the text generation part. And the other part of it is like the image partner, where they have these stable diffusion models, I’m not really going to go into how that works. They have this convolutional neural networks that generate images for you. So it’s mostly two parts like understanding that piece of media, whether it’s text and images, and being able to complete
Michael Waitze 16:17
it. It’s so interesting, I want to tell you how I use generative AI yesterday in like a real world example. Okay. And then I want to talk about personalization, right? Because one of the things that the large language model or GPT does is it takes all of the writing in humankind, right, and tries to categorise it, index it and tag it so that it’s useful for all of us. And when I write an email, though, I don’t want it to sound like Shakespeare, I want it to sound like Michael. Right? It’d be neat if it sounded like Shakespeare, but everybody would know them. Do you know what I mean? It would just be super weird. And if it didn’t sound like me at all, for people that know me, they would know that either A, I would hired someone to do it for me or be a hired machine to do it for me. But I did a recording, let’s just say in September, October of last year, and somebody wanted a great quote out of it. So I did a transcript, which I use something for. And I sat down to read through the whole thing. And I thought, okay, that’s really stupid. I just gave it to Chad GPT. And I said, get me five really great quotes from it. Because it can do that. And it got some quotes. It didn’t really give me what I wanted. But you know what I did, I just did it again. And then again and again. And then he got me two quotes, and I use them. It’s so useful, right? But again, it’s not automatic. It’s not magic at all right? But it did work in that context. When you do it for mail, do you go through and read all of the email that I’ve already written? Do you know what I mean? And then say, Michael generally writes like this when he’s in this kind of mood, or has this kind of tone. And if I want to generate something similar to that, I’m gonna use the mail that Michaels written. And that’s the first thing. But secondly, if Michael wants to sound a little bit more, like somebody else can then access some other things that they’ve written to, and combine those two things together. So I can sound a little bit like, I don’t know, you know, Obama’s speech writer, let’s say,
Michael Vandi 18:01
changing personality, you know, what
Michael Waitze 18:03
I mean, just to add a little bit of
Michael Vandi 18:05
just, yeah, um, so that’s our North Star to be able to personalise your writing, chat, TPT, and other language models, like GPT, three, GPT j, right, they are so large, that they have vast amounts of information at everything, you know, the, you know, Shakespeare, they could write code for you, they could generate a lot of quotes for you, but they’re not you, you know, read an email, you want to write an email, as Michael, I do, you don’t want to write an email at Shakespeare, right. So that’s where we want to get to. So we want to train our own models, that would be very personal to the user, and being able to generate text as closely related to how you would do it as possible. And if we’re able to get there, you know, then we would start looking at other features, like, oh, I want to generate text in the voice of Barack Obama and generate text and the likes of XY and Z people. So one good advantage that we have, as we have a lot of resources at Carnegie Mellon. We’re currently all interviewing a lot of like, machine learning engineers really good at what they do at CMU, who are going to come on board and help us build those language models. And you know, the future is looking like it will get there.
Michael Waitze 19:30
Yeah, want to get to the future in a second. But can we go back how this works? Logistically, let’s say I use the product, because you can tell the kids about doing it. I mean, I read hundreds of mails a day, and it just drives me nuts to I just give it access to my Gmail. Do you know, I mean, is there like a plugin or something? So I give it access to my Gmail or my Mac mail or some other mail client, just go look, just go read through this tonight while I’m sleeping kind of thing. And tomorrow, then what do I do? And how do I know? Do you know what I mean?
Michael Vandi 19:52
Yeah, so it’s very seamless. So you download the plugin, the Chrome extension. Okay. We’re currently in Chrome Oh Speedos. So we only accept people that have signed up on our rent lease and then we send your name. I’m gonna send you an invite. Absolutely. Sign up, I’ll, I’ll put you in front of the line sounds like you need a lot of help with, what?
Michael Waitze 20:10
Do I need help with other things, too, but email for sure.
Michael Vandi 20:14
Yeah, so you download the Chrome extension, and then you open your Gmail, yeah, you click reply to any email. And then what we do is we synthesise all of you the name of the thread, the context of that email and some of the emails that you’ve sent in that thread. And then chat GPT understands that that’s like the NLU part natural language understanding. And then you pick a tone, we have over 10 different tones that you can pick up from like friendly, respectful, informal, formal, funny, and then you click right, when you click right here to learn about your writing style, and writes in response for you, in less than 30 seconds, right. So it does all of that processing. Now we could go the route of reading your emails in the background, we’re not there yet. But now we’re only accessing the exact email that you open up the one the thread that you want to reply to got it. Um, ideally, the situation where we want to be in, you know, we’re taking a very privacy and security centric approach to this, and letting people opt in to any kind of like data collection that we’re getting, we’re not collecting any data right now all of the data has been is not stored, and just go straight to the models, back response, and then get there. We’re two weeks, two weeks old company. Lots of things to do in the future.
Michael Waitze 21:42
It’s probably a little bit more than that. Are you surprised sometimes by how effective GPT is? Do you know what I mean? Because we talked about this before, like when I asked you what had to change? Surely that had to change the tech behind GPT. But the real question for me is the throughput, the compute all of those things really had to change drastically just to be able to process all of this information, right? And do you notice the differences in when the model gets changed, or when you increase the compute that you have? Or the throughput that you have in your ability than to create better outputs? Do you know what I mean?
Michael Vandi 22:17
Yeah, so GPT three, which was the precursor to chat GPT, which is people call it GPG 3.5, right had 175 billion machine learning parameters. That is so many inputs that it could synthesise. And then as we scale that, I mean, there’s this, if you look at like the number of compute that silicon chips can do, when it scales linearly, like tech just evolves, improves as the number of computes that a silicon chip can do scales up. So like in the 90s, like chips, were not really being able to do like a lot of like compute power, there’s also the low level part of it. Whereas as we scale the compute power that we can do on the silico level, and then we scale the compute that we can do in the machine learning level. You know, I can just imagine if we were able to 10x, the number of parameters on an GPT? Three, how crazy it would be,
Michael Waitze 23:18
it’s not a one to one, right? I mean, this is not a straight line, it isn’t logarithmic, is it exponential, because it’s not just going to be a one for one increase, when one increase in the processing is not going to be a one increase in the output, it could be 10 times 20 times 30 times I don’t know what that number is. But it’s so and I feel like our ability to increase compute is actually increasing. You watch the stuff that Apple’s doing with him one of them too, and just other stuff that’s happening in the silicon stuff, right? I mean, that’s just shockingly amazing. No,
Michael Vandi 23:45
I totally agree. You’re right. Mapping would not be one to one. Every now and then there comes some amazing technology like Apple’s doing with the Amazon chip, right, right, and just bumps up, like the exponential increase in the number of compute capacity. Every now and then there’s going to be some technological innovation that would increase the compute that we would have to
Michael Waitze 24:13
do. Are you or do you do people express concern to you about the fact that if writing is replaced by machines that then people will be displaced in their jobs? I have my own answer, but I’m curious what you think you could ask me the same question and I’ll come back to you with mine. Go ahead.
Michael Vandi 24:31
going to ask you what you think. And I hope our answers are not very similar. Um, so here’s what I think. Go ahead. I mean, I would love to ride a horse to work.
Michael Waitze 24:47
I love you go ahead.
Michael Vandi 24:50
But I don’t because humanity has evolved from that, like innovation has changed, right? So innovation is inevitable. Humans would have the ability to focus on other things as AI takes some of the more mundane tasks. You know, imagine waking up on a Monday morning and you don’t have to worry about all the mundane tasks that you have to do like sending emails, but you, you have time to focus on arts really, if you like music, if you just want to, you know, relax on a beach somewhere. Trinidad and Tobago read a book, ya know
chatting with someone in Trinidad Tobago source is an interesting, I was chatting with my son who used to work at the train station in Baltimore, where I used to go to school. And it was this he just retired. And I he messaged me on LinkedIn, it was like, Oh, you drink great, what’s going on? And I was like, Oh, you retired? What are you doing? And he said, he’s in Trinidad and Tobago. And it’s like an Amtrak customer service representative. Great guy. And he told me that he would be going to the Trinidad Carnival. And I said, Oh, I want to go. And he was like, Don’t come if you come, you will never want to go to the US again.
Michael Waitze 26:08
I’m not gonna go there. But is it but isn’t this the case, though, because I make this case to the the tech will bifurcate things in this way. I think and what let’s just talk about writing because that’s where we are. Great writers, people that are dedicated to writing whose lives are dedicated to writing will only benefit from generative AI because it’ll help them write. But people that aren’t great at writing been trying to skate by with right, we’ll have to find something else to do because they’ll be replaced by technology. But this is true. Exactly. And I’m glad you brought up the horse example, because it’s the one I always use. In the old days, everybody had a horse. But what it did was when cars were invented, people didn’t go out and shoot their horses, they just thought, oh my god, now it’s so expensive to maintain a horse because the infrastructure that used to be there to take care of them. Now is so much more minimalized, which means that to do it is more expensive, but it meant having a horse and having horse skills made you so much more valuable. Oh, yeah, yeah. And I think that if you’re a great writer, what the generative AI will do is make you so much more valuable. Because what generative can do is go back and look at what’s already been written and write based on that what a human does, who’s great at writing is thinking up new things that haven’t been written before. And write those things down. Is that fair?
Michael Vandi 27:21
Yeah, that is fair, I have a question for you actually go ahead in relation to that with writers. But in podcasting, which is like a mix of different kinds of media, like podcasting is like speaking and writing and like social skills. So in the future, as AI starts to get smart, and being able to combine these different modes of communication, what do you see as like, the future of podcasting as AI is being able to generate texts and plug it into an image and make a video and all of that? So I think
Michael Waitze 27:54
it’s a super great question. And obviously, I’ve been thinking about this a lot. But I think it’s the same concept. Today, anybody can kind of throw anything out onto the internet, all you have to be able to do is buy a microphone and be able to record it. It doesn’t even have to be that good. But if you’re really great at this, if you have like a high EQ. And again, if you’re social, and you can understand how to have a conversation with somebody, it’s going to be highly differentiated. And it means that for simple things for like educational things, you can automate that really quickly. And even for simple voiceovers, you can copy my voice actually, and turn it into voice. But here’s the thing is that at the end of the day, whether it’s buying an airline ticket, ordering a pizza, or doing a podcast, humans literally only have visceral connections with other humans, and really do the difference between holding like your girlfriend’s hand, right, and grabbing onto a handrail, or just two different experiences. And even if you shape that handrail into a human hand, it’s just not the same. And I think that’s why the value of people that do this right and have the ability to connect with other humans, it’s just going to get higher. And we’ve seen this in other places. Again, I was having this discussion a couple of days ago, we built a product when I was at UBS, and then again at Goldman Sachs that we call nominally smart sales trader. So it meant that a person who was facing the market but also had clients coming in and saying, which stock should I buy? When should I sell it? What’s the right stop out price and all these things? You didn’t have to remember it as much. But you still had to tell because the technology would tell you, Oh, you should not buy Toyota here because the stock levels here in the RSI is telling me this and the stochastics are telling me that that’s great. It just means you couldn’t remember it before. Now you’re superpowered. But being able to tell that story to a client or having that client who’s panicking because they forgot to invest their money, and kind of electronically holding their hand and saying it’s gonna be okay, I’ll take care of it for you. We’re just not there yet with machines. And again, the great sales traders became even better with it and the middling guys and gals. We’re just kind of trying to take a paycheck just got ruined by technology. And I think it’s gonna happen in every domain.
Michael Vandi 30:06
Yeah, yeah. That is true. That is true. Machines are not good at evolving as great as humans, humans are. The machines are a snapshot of different algorithms at a time. Yes, humans are constantly taken inputs like, myself today and myself, and the next minute is not going to be the same. Like, if someone interacts with me a minute from now, it’s going to be different. But if someone interacts with, you know, 10 GPT. And like, another minute, it’s still the same.
Michael Waitze 30:37
Yeah, and you know, what else what else, you can’t tickle the machine. But think about it, you and I are having a conversation. And if like your cat, like, jumped onto your lap and lick your face, it would change the way you would change your mindset. It may annoy you, it may make you happy. It may make you nostalgic, it may make you think about like your mom or your grandma or something and change the tone.
Michael Vandi 30:59
If I’m allergic, I would sneeze.
Michael Waitze 31:01
Yeah, but all these things change. I think people forget about this, like, that machine’s gonna take my job, well, maybe, but maybe if you were super good at it, you would figure out how to use that machine to actually be better at whatever task you’re doing. You’re doing. Yeah, and the other thing too, is, maybe you could employ it so that you can have like, Saturday off?
Michael Vandi 31:21
Does your job better? You know, he’s just planning to have it 10 Extra productivity. Like if you were a writer, and you could write, you know, one paper every one article every two weeks now, you probably could write 10 articles, you know, it’s like, what’s the improving productivity, I think like, we should look at the AI boom as like, being able to augment the productivity of humans, I grip and replace replacing it.
Michael Waitze 31:56
And let me give you another example this and you and you can use this actually doing your sales calls, right? If people ask you questions like this, if you’re not already doing it, but you know, part of this smart sales strategy thing was building trading algorithms, even simplistic trading algorithms that just kind of map we call it v Whopper that valuated average price for the day, right and volume, excuse me weighted average price. And what it meant was that we were always trying to match volume and match prices. At the end of the day, the price that you got was the volume weighted average price. So at every price at every price point, what the volume was weighted for the price gave you an average price for the day, it was hard to target on your own because you had to anticipate volume. But you also had to remember how the stock traded. So what we did was we went out and built technology that went back and looked at that stock on every type of day and time and whatever, and then went back and mapped it so that on any individual day, you knew kind of what it was meant to do. Yeah, but But again, it didn’t replace the human who still had to talk to the client and said, You know what, today, I actually wouldn’t do a view upgrade, because I think the market is going to do this. And I would stay out of the market on the open and blah, blah, blah, which the machine cannot do. Does that make sense?
Michael Vandi 32:58
Yeah. So you’re saying the machine was able to map to historic data, humans were able to interpret it.
Michael Waitze 33:05
Yeah. And even if the machine how to interpret it. Again, the machine didn’t know that today was Valentine’s Day. And that that meant that all the senior traders were going to be out for half a day buying gifts for their wives or girlfriends, so their trading in the second half of the day was going to be slower. So that traditional view of wouldn’t work on a day like that kind of thing. That’s just one simple example. But there’s so many other examples where you could literally just like precedent, forget it. But you could get killed during the day if like, during the middle of the day, someone said Fed raised interest rates by 75 basis points and you killed read, but it’s the same thing in writing. I mean, here’s the thing that I wanted to ask you just as we were starting to record, you know, you were trying to set up your camera in your background, my camera died while we were setting up technology can get finicky and this is not a bad camera. What does Finnick you look like? It’s a hard question, though. What is finically look like in the AI space? Do you know what I mean?
Michael Vandi 34:04
If what I’m worried about is not a like being able to be sentient.
Michael Waitze 34:12
Nope, zero not worried about it. Go ahead.
Michael Vandi 34:15
But I am worried about us teaching AI to be a little astray. From what you what you what you would do. Naturally. I’ll give you an example. I think it’s best if I explained this with an anecdote, please do. So. There is this Enron email data set, right? You’re familiar with a company Enron? very shady. company so when they got busted, they put 500,000 and run emails got leaked and some publicly open email dataset. Oh, wow. Now, if you use that data set to train an AI ml assistant that would reply to emails for you You, then theoretically, you know that AI would have some of the, you know, inferences or all the shady things that were happening in Enron fraud. Yeah, so what I think finnicky looks like for AI is on the data level, you know, the training level, like the data that’s being used to train. So I think like, as we train AI, we have to be careful about the data that we input into it. Because if we filter the data first, before training the model, then we don’t fall into the case where we have to set all these guardrails at once the model is trained, right? Where you will make sure Oh, this model should lean a certain way. It shouldn’t use curse words. But if you don’t input curse words into the model, then it doesn’t know what to
Michael Waitze 35:47
use it. Yeah. Can’t use them. And in reverse, if you feed a tonne of them, it doesn’t know the difference, right? Like I had a guy this is a such a great example. I was travelling with a guy from Japan, who spoke like just enough English to get himself in trouble. Right. But he couldn’t really have a conversation with you. But he knew words. And as we were going through immigration, the guard asked him for his passport. And he basically said, just because he doesn’t understand the impact, he basically said towards the the guard, let’s just say he shouldn’t say where your mother would have said to you, I’m going to wash your mouth out with soap, which is what my mom said to me. Do you mean he said it to the guard, but the guard doesn’t know that that guy doesn’t know. And again, he’s using some sort of generative AI model that he’s created for himself to speak in English because he thinks it sounds cool. And I was like, Dude, you can’t say that kind of thing.
Michael Vandi 36:38
That he said that, because he didn’t know that’s all the inputs had been has been getting? Exactly. Here’s what happened. What happened after I did the guards do while the guard freaked out,
Michael Waitze 36:47
but I had to explain to the dude didn’t mean anything. He doesn’t understand the significance of those words, please work with me, because I’ll help out. And I said to the guy basically shut the up. Take care of this.
Michael Vandi 36:59
That’s a real thing called AI. Yeah,
Michael Waitze 37:02
I did. But But again, look, that’s not what you’re doing. And so it’s a conversation about how this stuff works, right. But I do like this idea for people that do write a tonne of email, that this can just make you so much more productive, right? And let’s be fair, any new technology, particularly in its earliest stages, what’s the right word? There are going to have to be choices, right? Because there are natural human biases that we have that people have to be aware, right. In other words, I know that I think people that are like, I’m jealous of people that are taller than I am. I just know that with a bias going in and I try to compensate for it somehow. Tali five, seven, you know, in Japan was not a real big problem. But outside of Japan, it’s not a problem. But you know what I mean? I’ll give you I was at a conference in Vietnam last week, and everybody there was from Germany or from like, Denmark. And I swear to God, I was doing this the whole entire conference, there was a guy there was seven feet tall, six foot North super tall, just super duper tall. Anyway, that idea of bias is super interesting to it now. Yeah, thanks, for sure. I think so. But how did you get interested in all this anyway? You don’t I mean, like, what are your mom and dad do?
Michael Vandi 38:09
My dad is a civil engineer, who’s like someone who’s good at math and, you know, good, natural engineering. But the most impact I think, came from my mom, who is She’s the smartest person I know. So she she she was a teacher for most of my childhood life. And she ended up getting a job as a social worker, this place called CARE International. Okay. Oh, she now works good. Save the Children. She has been she’s been working for a lot of like, NGOs. For a time when she was a teacher. She really instilled in me the the need to be like, the best at everything I do. So I’ll tell you what, I had so many tutors. When I was growing up. I would go to school. And the teachers that would teach me in school after school would come home, and then teach me at home. Because we have these two shifts system. And she wasn’t a teacher in the higher level education. Usually they were wearing like higher level education. They go to school, in the afternoon, okay, so when I would come home from school, my mom would not be there. So she made sure that I wasn’t caged all the time. She would give me homework and have tutors come home and teach me and then in the evening when she’s there by seven 8pm I would tell her, like everything that I’ve learned for that day, and how things were going. So I think she’s just like this person who really values education from the onset and having a teacher as a parent, I think very valuable as a young kid growing up. Yeah. And at a time when I was in my final year of high school, she made sure I take the GCSEs which she and my dad So my dad had this vision, and my mom had the support. So my dad would say, Well, I want I want him to take the GCSE. So just like the London, the London, GCSE is a General Certificate of Secondary Education, which we don’t take in Sierra Leone, these were like international certificates that I didn’t need to take in Sierra Leone, and the SATs, right. And then my mom would come in with all the support and like logistics needed to support that. And the most exams I’ve ever taken was like, when I was 15, I took the GCSEs, I took the SATs, I took them, the West African Senior School Certificate examination. And I took that my final year examination at school all in the span of like, what, four months. And it was so stressful for my mom at the time, where you had to hire all these different tutors, because the curriculum was different for all these, all these syllabuses were different kinds of tutors. And also, she would drive me to the tutors like at night, and then she would. And this is a real story where she would wait in the parking lot in the car while I take the lessons. And then I’d come and just knock on the on the car window. Like it should be sleeping at like 2am. And then we’ll drive home and then the next day repeat drop me off at school. So yeah. That’s I think I’ve just been incredibly lucky to have parents who are supportive and value education.
Michael Waitze 41:26
Do you have brothers and sisters as well? On my mom’s side now, but on my dad’s side, yes. Got it. Got it? Yeah, it’s interesting. I mean, I only have a daughter. And you know, you try to give them as good an education as you possibly can. But at some point, like, you have to be super proud of your mom, right? Because she was trying to do everything in sounds like in a way, do you know what I mean? Have her own successful career maintain a most good strong family. And then if she sit in the parking lot at two o’clock in the morning, it’s like heroic at some level. You know what I mean? And it’s so hard to explain to people, like if you grew up in a family where education is super important. It just never goes away. Right? It just never goes away. Look, my, you know, when I was a little kid, my grandfather would come home, my grandfather stopped going to school when he was seven years old. Because I mean, just think about the timing. It was probably the teens or the early 20s in the United States, my grandfather lisps and stuttered. So they just treated him like he was, you know, retarded. Yeah. And there was no infrastructure back then there was no child protective services, there was nothing. So that dude just ended up on the street. And at the end of the day, he was super duper smart. Again, whatever that means, right? Because he couldn’t do math at all, because he never learned it. He couldn’t even write his own things. You couldn’t write his own name. But he was intelligent enough to marry a woman who was a mathematical genius. And my grandmother was an actuary.
Michael Vandi 42:56
I want pointers from that guy. Like, how was he? greatest man
Michael Waitze 43:00
I’ve ever met, of course, but I mean, I’ve haven’t met everybody. But you understand the point, right? It’s like the value that he put on education, because he was uneducated was so high, which filtered down to my dad, and down to, you know, my siblings. And to me, as well. And just to explain to people why, what it’s like to grow up in a household where we are surrounded by poverty. But if you educate yourself, you at least give yourself a chance. It’s like buying, it’s like buying a highly targeted lottery ticket at but there’s pressure there too, right? Because it’s like, okay, we’re in the parking lot at two o’clock in the morning. What are you gonna do with this? Do you don’t I mean,
Michael Vandi 43:41
yeah, a lot of pressure. A lot of pressure. I absolutely agree. With the pressure, I still think there’s so much pressure. Um, but she’s also managed to do things for herself. Yeah, it’s amazing. Like, I don’t have to match up, match up the pressure.
Michael Waitze 44:06
Is there a sense of can I ask you this, though, because at least a lot of the kids that I meet out here, right, when they’re children, when their parents were like, were younger, and they wanted them to just work in big companies, big established companies, right? Because the idea was, look, we’ve got you educated, don’t muck it up by going out and taking unnecessary risks. A startup is like one of the biggest risks you can take, because there’s no guarantee you’re gonna succeed or that anybody’s ever gonna even like your product. And you’re still at the early stages, which for me, is the most fun, right? Was there any ever pressure on you to like to just go work at Google? Do you know what I mean? Or go work at Apple, you’re qualified, you had a job at AWS? You know what I mean? When did the family just go What is he doing kind of thing when you started your own thing. So
Michael Vandi 44:44
the pressure there is there is the pressure to continue learning the pressure to get your Masters get your PhD. So my mom rule and my family sipping from Sierra Leone like they don’t really know like, which which are The top five companies in the US. So as long as I’m doing amazing stuff, they’re fine. But what they want to see is like, you go get your Masters go get your PhD, like when I was at AWS, I didn’t want to quit my job to do my Masters, I actually deferred like I got, I got the acceptance into CMU right out of the undergrad at the same time I got the Amazon offer. And and then logically, I was like, CMU, Amazon, I’m definitely going to take the AP, so it’s like multiple, six figures a year. And my mom was like, what, what are you doing, like, go get your Masters, and she doesn’t really understand. That’s where sometimes I bring the Saturday to the table, I’m like, Well, I’m going to do the interviewers thing for a year. And then, you know, quit and go do the masters. And she was saying all once you start, you’re never gonna quit, you know, you’re gonna be blinded by money. And eventually, I did quit. But so yeah, there is pressure there, but pressure to study in school, or so then.
Michael Waitze 45:58
It’s just such an interesting dynamic, right? As a parent, you want to be able to brag to your friends about your kids. And it’s so much easier to say, My son got a master’s degree and whatever, I’m just gonna say Computer Science at Carnegie Mellon University, and just have the rest of the parents go, Oh, my kid just works at Google kind of thing. As opposed to I made a half a million dollars this year. Do you really mean because you can brag about that? It’s no one ever says like, my son’s a millionaire doesn’t feel good. You know what I mean? But the Masters we can definitely brag about forever.
Michael Vandi 46:28
I see it a lot in the in the Asian culture as well, where, you know, it’s like the Asian like, immigrant culture, or, yeah, they, they want you to achieve so much like, we struggled so much to let you come here and do just want you to have you we just, we just want you to do better. And yeah, fair
Michael Waitze 46:45
enough. Before I let you go, tell me about the status of anti AI? Where do you think it’s gonna go? And if you fundraise or not, and if you ever plan on fundraising as well?
Michael Vandi 46:56
So, first question out of the way we don’t that’s that was your second question. I’m gonna change your order and answer them, we would plan on fundraising, we’re talking to your angels. If you have VCs, we want to be aligned with the people that we raised from and not like over race. We also realise that this is something that we can monetize for and being able to bootstrap. So we, we just want to raise as much as we need for ideally, let’s say to your runway, while we still support ourselves with the bootstrapping that we need. And in terms of where I see the future of it going. I see Eddie AI as not being able to be used by anyone, anyone and everybody who would just want to reply to their emails, right? I see us as being able to vertical eyes into a specific industry, okay, and do a 10 over 10 job at it. What does that mean, retail retail was whether it’s real estate, you know, being able to respond to real estate queries, whether it’s medicine, being able to respond to customer engagement queries, just like just vertical eyes to one industry, and being able to dominate it and do it really well. Got it. Um, yeah, that’s where I see it like going I see us having a few of our own models on our own models that help text processing and understanding for natural language generation and a specific industry. All those models were in talks of actually building those models. So yeah, I also see us as as a source of truth for generative AI, you know, like, if I tell chat, GPT Ido, tell me what my availability is tomorrow. It’s gonna, it’s gonna lie, it’s gonna guess, right? We want to be able to integrate with your calendar and other tools and just not only generate text, but generate factual text. Yeah.
Michael Waitze 48:52
Yeah, that would be super cool. Actually. I mean, I said, that was the last thing I was gonna ask you. And I always do that. And it’s never the truth. But it would be super cool if the email or the plugin, right could see the email and then say, take a look at my car. Because right now I use Calendly, for everything. And in a way, it’s kind of like a quasi hybrid intelligence. I mean, not really. It’s an API view onto my calendar, and it just goes here, open slots, it doesn’t think about it at all. It doesn’t have a level of importance, right. So sometimes I have to look at Calendly and think, why isn’t that slot open? Why Yeah, I look at my calendar and it’s available right now. But it but if it could think the same way I did and said actually, this meeting is super important. I know there’s something there. But I’m going to put it there anyway, because Michael really needs to talk to Michael Vandy. It shouldn’t be able to do that because that’s what I would do that would make me way more productive. So that’s a super cool idea. Yeah, I love it.
Michael Vandi 49:46
We’ll get there. We’ll get there. Eventually, eventually we’ll get there.
Michael Waitze 49:51
I love it. Okay, look, I feel like I could go on and on with you. Can I do this? How can people get in touch with you if they want to reach you?
Michael Vandi 49:56
They could reach me at my email. It’s Michael at at Do you dash ai.com? Easy. I’m also fairly active on Twitter. My ad is Michael underscore Vandy. Got it. I have a YouTube channel and it’s just me talking about projects, but just email me.
Michael Waitze 50:16
Okay. Awesome. Awesome.
Michael Vandi 50:17
Well, thank you so much. This was this was great. I had fun chatting. Can I ask you one question? Absolutely ask me anything. Yeah, well, how do you see the generative AI space environment in in Asia?
Michael Waitze 50:29
So I haven’t tested it in any other language, right. So for me the generative AI stuff from a visual standpoint, I think it’s going to be really powerful. But again, I do think that great artists will not be replaced by it. But people that are just doing simple artistic things will be like I just said yesterday to Dali, you know, take an iPod case and put a picture of me on it just to see what it would look like I could never do that on Photoshop, some people can write, but I can’t do that. But I do think in Asia, because it’s early days similar what we were talking about Africa, right, which is just completely Greenfield and maybe where Asia was 25 years ago in the context of infrastructure development and technology development. But here, I think the uptake is going to be really fast. And I think one of the reasons why is because people will use it for translation as well. So if you want to take something from Bahasa and turn it into English in a way that’s effective, like because before, it would just like, take the Bahasa word, take the English word, and maybe it’ll be out of context. But now what it can do is take a phrase in Bahasa and go back and look at the history of all of humankind for English and figure out the most appropriate way to translate it and say, I think this is what it meant. And I think for that it’s going to be transformational. Does that make sense? Yeah, that makes sense. But also remember, also remember this, you know, India is 1.2 billion people. And while we can argue about what the impact of history has been on the Indian cut on the Indian subcontinent, what we do know is that there’s a lot of English spoken there that wouldn’t be spoken otherwise. And there’s a lot of activity in India, that’s going to be using open AI and chat GPT type language models and building similar technology to be able to take everything that the people in that country are thinking and synthesise it, and then use that artificial intelligence to write as well, I think there it’s going to be wildly transformational. But once it gets into sort of other languages, I haven’t tried it in Japanese, I haven’t looked at it in Mandarin, I haven’t looked at a new language. But once it does, you know, it’s a double edged sword, right? The idea here is that, do you want to work with a technology that literally, at least in today’s world, is just mimicking what’s already been done, because you run the risk of like, kind of drawing this line in the sand that before, you know, pick a year 2035 ad, and everything before that is catalogued. But everything that’s written after that is being generated, which means then that you have generated content, just work with me on this right, is being generated from content that humans wrote. But now as you as you go further and further out, most of that stuff is going to be built on generative ad AI, which means that the impact it’s going to get diluted, it’s going to dilute the stuff that’s already out there. And then it’s just going to recycle all of this generated content. And I think we lose a little something there. And the more you extrapolate forward, the more you think, wow, what’s really the scary thing is that then the educated elite are gonna be the only people with like, the original ideas, because everybody’s gonna base what they write and read on stuff that’s already been regurgitated. And it’s just an interesting philosophical thought. Yeah.
Michael Vandi 53:33
I must say, that is something I haven’t thought of in that sense that you’re brought up and I think it’s very interesting that you brought that up. And I think that’s a whole philosophical discussion that needs to be had. Yeah, I agree. Like you run the risk of this circular dependency of texts being generated from generate a text which is generated from generate a text generator from Yeah,
Michael Waitze 53:58
yeah, kind of thing. Anyway, I like to think about the stuff like that. We should have you back on as it continues to get developed. We should continue to have you back on the show. And keep talking about this. Michael Vandi, a Co-founder at Addy AI that was awesome. Thank you so much for doing this today.
Michael Vandi 54:12
Thank you for having me.