The Exchanges

Q et cetera. There's all sorts of things that AI can do that a human just doesn't have the time to do. But at the end of the day, you still need a human to really perform a lot of the functions. Like an AI doesn't have bedside manner. Not to say humans are great at it either, but some of them are. What made you want to get into AI?

A I have been in AI for a really long time. I have been in AI for over 30 years now. My, when I was in school, I got exposed to something called neural networks way back in the late eighties. And, and I read this first paper and it just blew me away. I was like, What? This is how the brain works and, and we can artificially simulate it in a computer and actually make it do something helpful, meaningful. And I've been, I've been fascinated by that and, uh, and I pursued this for a long time. And, uh, the first realization about neural networks that I had was that, hey, it's actually math. It's really, Statistics and mathematical models that you optimize a big equation, and then you, you, you come up with a, with, um, with a solution for that. You can then use it for prediction. So for the longest time, people had been doing statistical predictions since, you know, since data ever existed in the fifties and stuff, but, but now there was a different way of looking at it. You didn't always have to have An explanation. You can have a black box that does certain things for you. You just provided input and output and, and it trains a model for you. That is, uh, it works really well. And, uh, and so I was just like that black box concept feels like the brain. And it's just been fascinating throughout my life.

AI assessment note: “when I was in school, I got exposed to something called neural networks”

Q or 15 years at a time. Then obviously ChatTPT created or enabled this newest breakthrough, and I guess it's the way that they built the model. And, and therefore the generative AI became possible. Do you think we're just gonna focus on that? Or are there any other kind of roadblocks to a massive change? Because generative AI seems massive now, but in a few years, maybe it won't anymore.

A Definitely not. Yeah. I mean, we already see the hype. Dying. So there's a famous graph called the hype cycle. Gartner invented it. Like, you know, you see something really hyping up and then, then you kind of start, it comes to this trough of disillusionment. And after that trough, you actually start seeing valuable things happening with that technology. So I feel like we're kind of just at the peak of the hype, which is beyond that a little bit in generative AI where we still haven't seen the trough yet, and then we'll start seeing A lot more valuable things coming out of that. And there'll be similar hype cycles for other technologies in the future. I can't predict this. I'm, I'm really not a futurist, but, uh, there will be other fascinating things that we'll see in our lifetimes, John, and, and, uh, and we'll go like, oh my gosh, this is, this has so much potential.

AI assessment note: “Definitely not. Yeah. I mean, we already see the hype. Dying.”

Q Makes sense. I've heard that before. How have you had to change yourself from the beginning of your journey until now?

A I think I, there was a lot more arrogance I had before that I had to turn into humility over time because, and it's really interesting because you think, you know, you've done it once. The second time would be easier and the third time would be even easier, but it's never that way. Uh, every product is different. Every market is different. Every time in the industry is different. So So you're constantly having to evolve yourself, but I think the, the realization that. It's all about having a learning mindset and growth mindset. Uh, I've, I've had a lot more over the years and, and, and I feel like that, um, you know, that that's one big change I can, I can see myself.

AI assessment note: “there was a lot more arrogance I had before that I had to turn into humility”

Q Is there anything about AI that scares you?

A I think with this recent Generative AI spade. Like one of the things that I worry about is all this disinformation and these deep fakes and, um, cloning of real people's voices to say something that they never said. That stuff is, is scary. And, and we just have to take everything with a grain of salt now. You know, I think that's the, that's the one thing that I scared, I scared about. But I think, I think humans are, Whenever we get scared, we identify a problem. That is, gives us the impetus to identify the solution as well. So we're coming up with solutions. The scientists and the technologists are coming up with great solutions to identify deep fake, identify, um, You know, misinformation, things that are blatantly false. So I, I think we'll have, I'm, I'm, I'm hopeful, but that's the one thing I'm scared about, but I think we'll have solutions.

AI assessment note: “one of the things that I worry about is all this disinformation and these deep fakes”

Q What's been one of the hardest things about AI for you?

A Because I've been working in this field for so long, uh, there have been different hard problems at different phases. Um, I think one of the hardest thing when about five, seven years ago, when deep learning came out, when you had like these really detailed models with a lot of parameters that you can start, you could start to train on the explainability of it. Why chat GPT something? It does something or responds with something. The why of it is, is kind of the one that really intrigues me. And, and it's really the hardest thing to get my arm around is like, okay, why does it work? Why, um, why does this image Generate this way or why does this recognize this to be this object? And it's really like, it's really hard to explain with AI models because there's so many parameters, so many deep things that are happening that, um, So the amount of data is obviously a hard thing because, because to train these models have taken enormous amount of data, so the compute and the memory and all that stuff is hard. Um, but I think the harder thing, the meta thing for me is, I can't explain why it works. It works.

AI assessment note: “I think the harder thing, the meta thing for me is, I can't explain why it works.”

Q What do you think the next big Breakthrough or use of AI is going to be from what you can see?

A I, I mean, we've already, we're already seeing many of these things and they're just going to get better over time. I, I had, uh, I sat in a self driving car in San Francisco, uh, last week, uh, for the first time and I was blown away. Like this is, uh, in, in the city. It wasn't even like in the freeway where you have like, you know, much, Much more defined driving patterns, but in the city where you have like people crossing streets and, uh, cars parked anywhere and speed bumps, and it was handling all of those situations so well. And it was freaky to watch the steering like steer itself. And, and, uh, um, but, uh, I mean, that stuff is going to get better. It's going to be more available. Uh, There was an Uber driver I talked to if he's like, is he, are you scared of this self-driving cars? And he said, no, I want to buy one so I can have that working for me when, when I'm sleeping. I'm like, yeah, that makes sense. Um, there's going to be a driver. There is already a driver shortage, Uber shortage in most American cities. So this stuff is going to help out, uh, in, in transportation. So, I mean, some of this is already happening. Um, Even this, um, the, the stuff we're seeing with child GPT and large language models, um, they're getting better and better at different tasks. It's going to be task specific things in AI that we'll see, uh, at, at my startup, we're doing video …

AI assessment note: “we're already seeing many of these things and they're just going to get better”