Kelly interviews Hao Peng about AI and its various applications.

They discuss the definition of AI, including machine learning, deep learning, and large language models. They also explore the use of generative AI in business messaging and marketing. Hao explains how AI is used in predictive analytics, natural language processing, facial recognition, and self-driving cars. They touch on the advancements in AI-powered search engines and the potential impact on advertising. Finally, they discuss the future of AI in code development, with the possibility of automation and code generation tools.

AI, machine learning, deep learning, large language models, generative AI, predictive analytics, natural language processing, facial recognition, self-driving cars, AI-powered search engines, code development, automation, code generation tools


  • AI encompasses a collection of techniques and technologies that enable machines to demonstrate human-like intelligence.

  • Generative AI is being used in business messaging and marketing to create automated content generation and personalized summaries of customer interactions.

  • AI is used in various applications such as predictive analytics, natural language processing, facial recognition, and self-driving cars.

  • Advancements in AI-powered search engines are enabling more accurate and relevant search results based on semantic meaning.

  • The future of code development may involve automation and code generation tools, with human involvement focusing on system understanding and problem decomposition.


  • The Future of Code Development with AI
  • Understanding the Different Aspects of AI

Sound Bites

  • “The broadest definition of AI is a collection of techniques or technologies that enable machines to demonstrate human-like intelligence.”
  • “AI is being used in various applications such as predictive analytics, natural language processing, facial recognition, and self-driving cars.”
  • “Generative AI enables more accurate and relevant search results based on semantic meaning of the question.”



00:00Introduction and Overcoming Technical Issues

00:41Understanding the Broad Definition of AI

08:03The Shift from Keyword-Based to Semantic-Based Searching

10:46The Automation of Code Development with AI

Auto Generated Transcript:

Kelly (00:00.302)
and we are live with Hao Peng. Hao, welcome to the program and sorry for all that technical hoop jumping, but you and I got through it.

Hao Peng (00:10.174)
Yeah, we did. Well, thank you for inviting me to this. We really appreciate it. Looking forward to it.

Kelly (00:14.286)
I feel like we should give ourselves gold stars no matter how this turns out just for figuring out the technology for you and I to actually be on the call talking without echoes and with all the technology working. So, gold stars for both of us. So, you’ve spent time in AI and your primary focus has been in AI. And so I’m going to throw you a question that…

Hao Peng (00:27.87)
Of course, so of course. All right, all right. I take that.

Kelly (00:44.239)
to start out the conversation that might seem like a softball, but it’s not. What is AI? I talked to CIOs and different executives and they say everybody who walks in their door tells them that the new thing they’re doing is AI. So what does the AI even mean in today’s world?

Hao Peng (01:04.254)
Yeah, that’s a good start. I think if you’re talking, as you say, if you’re talking with different people, they may be give you a different answers, especially in this days. I think the AI was interchanged with a lot of other things, especially generative AI. So if that’s OK with you, I’m going to bring a couple of terminology here because I heard a lot for the people talking through it because there are some subtle differences between them. So for example, back to the old range of question, what is the AI?

I would say the broadest terms, the AI is reference to kind of a collection of the techniques or technologies that is enabled machine to demonstrating human -like intelligence. So when I say human -like intelligence, what that means, think about learning, think about reasonings, think about problem -solvings or maybe decision -making, right? So that would be my probably the highest level definition about the AI. And then,

If I can just bring out a couple more terminology there, is you probably heard about machine learning. So what the heck is machine learning? So I would say machine learning is actually a subset of AI, which kind of it’s a technique that helping enable the machine learning through the large sums of data. That’s what that is.

The next two items is going to be the deep learnings. Deep learnings is really another subset of machine learnings. The learning from the data enables the machine learning from the data through artificial neural networks, and kind of inspired by the human brains. The last, but not the least, you hear a lot of this in the last couple of years, the large language models. Well, large language models such as GPT from OpenAI,

is a subtle, or a gemini from Google. It’s actually a subset of deep learnings. And they’re based on specific transformer architectures and for how they are developing and learning themselves. Does it make sense? Am I going too deep?

Kelly (03:09.742)
No, I think what you’re saying makes sense. You’re talking architectures, you’re talking the chat GPTs, you touched just a little bit maybe on something that many of us hear about the deep fakes in generative AI. And I wanted to ask you about that. And we’ll talk about all these areas, because I’m going to ask you about tools in just a little bit. But I was actually reading that.

many countries around the world are now using generative AI as part of their political strategies and how they deal with either from a war perspective or from a messaging perspective with allies, they are creating these deep fakes or generative AI. And so one question I have is, are businesses using generative AI in that same way for messaging and for, you know,

making themselves or their products look better than they are or as you know, the marketing spin on on things, do you know, do you know is that happening yet or not?

Hao Peng (04:16.222)
I would say from the AI standpoints, the business has already adopted a lot of…

technologies in the last number of years. And then I can just quickly go through some of the things that are happening in the last few years. And then we can go into some of the use cases, the generative AI, but that’s OK with you. So for example, you probably heard about the predictive analytics. There has been a concept being there for a long, long, long time. So really on a high level, you’re kind of leveraging some large amount of the historic data. And then you want to.

based on data predicting the future outcomes. So think about the demand forecasting, think about the risk assessments, so the targeted marketing campaigns. And then the next one is probably more relevant to generative AI is related to the natural language processing. The thing about that, you want to looking for the data that either being in the form of a customer review, so it can be a social media postings.

or can be kind of the support inquiries. So the NLP, also known as natural language processing, is enabled in the application like the sentiment analysis. Think about it as a positive review versus a negative review. So the chat bots, right? Or automated content generations, like you’re generating the marketing contents. And also, you probably heard about the computer visions. Think about the facial recognitions, right? So.

Most of us using our phone on our hands, right? So when you unlock the phones, what’s behind that is really the AI mechanism to identify you are the who you say you are, and then they can unlock the phone. And sometimes they can use in some of the defective detections, right, in the manufacturer settings, identify is there a defect in the products, so it can identify them before they ship to customers.

Hao Peng (06:15.166)
Last but not least is this, you probably heard a lot about self -driving cars. Those are things that using some of them CNN, which is another network architecture from the AI and machine learnings to enable self -learnings and then be able to drive themself. I know that technology itself is now maturing to the point that it’s fully automated yet by the something that’s a future direction to go. And that’s kind of on the high levels where…

kind of some use cases using right now. But then back to the questions you have there as to what the scenes are using today and that it’s really using the generative AI. So I kind of touched a little bit earlier in a lot of things happening in the last 12 months is really involving the natural language processing, right? So think about the, you know,

Kelly (06:45.71)

Hao Peng (07:11.774)
you want to generate the marketing contents for the products. In the past, you have a dedicated marketing department. They’re generating the product description for you. A lot of time, it required a lot of manual steps. And now, there are some technology happening that you can actually upload the images. And then the generative AI technologies, you can upload into the

And they can leverage in technology and be able to describing a marketing product descriptions that you can post into that versus automatically and scales versus manually. You can have some of the customer services, right? Instead of the, you may have a team of customer services working through and you want to get a summary is what’s happening with my customer services.

interactions in the last 30 days versus yesterday versus the last past week gave me a summary of those. And so you will be able to go through the transcript of those interactions and you’ll be able to get a good summary of what the interaction then looks like. That’s kind of a very high level use cases. Is there any specific things you want to dive into more?

Kelly (08:29.774)
Well, I wanted to ask you, you had mentioned the phones, which we’re all familiar with and how it does facial recognition. In 15 years ago, movies were showing retina scanners. And before you could get into the next door on the spaceship, your retina would have to be scanned and you’d walk through the space door. Maybe that was 20 years ago or 30 years ago. Today, that’s kind of happening with our phones. Is the level of detail when my phone’s looking at my face, is it actually looking at things like

Hao Peng (08:49.662)
Yeah, yeah.

Kelly (08:59.182)
my retina, the color, you know, like the difference in how many brown spots I have on my retina versus the blue of my eyes, or is it just general face shapes? Like how advanced, do you know how advanced is that technology related to how the camera works and how AI interacts with that? I have a twin brother. I don’t want him on my phone because he’s a troublemaker. So that’s why I asked this question.

Hao Peng (09:20.606)

Hao Peng (09:24.638)
Yeah, and I have to admit that I’m not an expert for some of the Visions machine learning experts. But I have to say that when the machine looking at you, a lot of time, especially in the high volume objects around the machines, they don’t looking at a lot of colors like what we saw in the real world. A lot of the machines looking at it, it’s a very

basic 256 color schemes. And what they’re looking for when they look in the facials, they’re looking for not only for your eyes, but also how the ever seeing your facial in the digital, they transform the data, what they see in the images into the data point itself. And then that will be the data they’re using for comparison to what you have. And then they want to be using some of the

probably hide a complex mechanism behind that to calculate what’s the probability you are the people, what you are versus someone else. But that’s a really good question. I’m sure there’s a lot more detailed algorithms behind that, complex algorithms behind that to enable it.

Kelly (10:33.07)

Kelly (10:44.27)
So I don’t want to gain a lot of weight quickly or use new glasses because I might get locked out of everything I own. That’s kind of the summary. Yeah.

Hao Peng (10:50.718)
Right, right. Well, you know, this is a fair said, you know, I don’t know what’s going to happen if I cannot unlock my phones, right? Because my half my life is here.

Kelly (10:58.606)
Right, my phone and it impacts my laptop, everything else. What about, so you had mentioned chat GPT and again, I think many people, many of us are using chat GPT once in a while at the very least. Are there other tools out there that are as common in AI as chat GPT? You’d mentioned in the deep learning space. How are people using?

Hao Peng (11:11.166)
Mm -hmm.

Kelly (11:23.086)
AI to better understand their data and what is it doing? It’s crawling data and it’s coming back and saying, hey, you’ve got a pocket of people in Tasmania that happen to be ordering your most expensive product. What kind of things is AI doing for people?

Hao Peng (11:36.83)

Hao Peng (11:40.766)
Yeah, so I was just going to give you the first problem, most relevant use cases a lot of people are using. So I would say most people are using search capability one way or another, almost every day. And you guys probably know that when you go to Google, typing, what the search happening underneath them most of the time is what they call keyword searching. Right?

What the latest AI development, the generative AI developments happening is enabled what they call semantic searching. So, which would be allowing more accurate, more relevant research results based on semantic meaning of the question we’re asking. When I say semantic meaning of the question, what does that mean? So for example, I can look in for searching for weather .com to say, what is the weather in July in Norway, Europe, right? That’s by accident, that’s the places I’m going to be actually traveling in the summer.

in July specifically. Instead of where I do a search like those, what I can ask the question is, maybe I really don’t like cold weather where I’m going. So what I can ask the question is, does the weather ever deepen below zero degrees in Norway in July? Or I think about that. So I actually did a search like this through the Microsoft Teams, which has enabled some of the personal co -pilot’s response.

I get a perfect result to that. So like I said, in July, you’re not having any situations like weather dipping under 0 degrees. And think about what that means for the search engine giants like Google. So 80 % of the revenue is derived from their advertising alone. So you can think about how much that can potentially impact Google’s revenues if everybody

overnight switching from the keyword -based searching to semantic -based searching. So that would be one of the levels that one of the use cases then where I see generative AI will have a profound impact on is the change in how people are doing every day.

Kelly (13:50.51)
But like for Google, they can still the data from just your voice or your text string query. They’re still going to know that they can be popping up ads for down jackets and new boots and tracking polls and whatever else people would be doing in Europe, right? I mean, that’ll still be in play for advertisers. Yeah.

Hao Peng (14:03.806)
That’s good.

Hao Peng (14:12.134)
yeah, yeah. I mean, absolutely. Google is catching up with this, right? So is there a competitive advantage to have all the tremendous data on their end? No, but the things is shifting right now. So they got to catching up. I think this is where we probably the future of the searching is that a transition from the keyword based searchings, which is not going to be going away, right? But also into a custom etiquette level of searching. How do they actually embed in there?

advertising businesses as part of their searching is going to be really interesting, not only for the search engine at Google, but also think about content creators. Think about CNN, the advertising on their website. Now, if those information was generalized and surfacing through the co -pilots or through the Google Searching in a more semantical way, they’re different, think about what that would be impacts to the content creators.

Kelly (15:09.23)
Yeah, and for those of you who don’t know what Microsoft Co pilot is, they basically took Clippy from 30 or 20 years ago and they’ve turned that into their new AI AI thing. And it’s very powerful. Everything I hear from people is it’s a fantastic AI aided device. One last question for you how I hear a lot about developers, software engineers starting to use different tools for built.

Hao Peng (15:24.222)
Yeah, absolutely.

Kelly (15:38.478)
building out their own code sets. And so how far are we in your opinion from AI taking over development? So flat out, you just being able to go online and say, I want to build an e -commerce system where I’m going to be selling jackets and shoes and things and having an AI generated code base being set and implementing that for you. Are we close? Have you heard of anything like that? Yeah.

Hao Peng (16:03.354)
Yep. So I’m going to be first a caveat. My answer here is to just tell you my views here. So I think the answer is somewhere is the automation of the code developments is coming. But there are always a real need for what we call human involvement. But the role of the human engagement is likely going to be evolving in the coming years until you’re reaching certain levels. So.

I think that Thomas Dukes, I think he’s the GitHub CEO. He predicts that sooner or later, 80 % of the code is going to be written by the co -pilots or some kind of a code automation generation tools. And it’s amazing that by end of last year, I believe in the GitHub itself, more than almost 50 % of the code is that AI generated already. So think about that.

Kelly (16:46.894)

Hao Peng (16:57.758)
in the span of less than one year. So the code used to be generated manually, it’d be almost overnight, half of them is already generated by the AI. And so that’s how profoundly, how fast the progress in this one here is. But he also mentioned that the role of a developer is not gonna be replaced. What’s gonna happen is the switching to what he called the system syncing, right? So the understanding that a complex of software,

and be able to get, you really understand the challenges related to that and be able to decompose them into a small problem, right? This is gonna play an ever increasing roles. So, back to the question you have there is, are we ever gonna have this, you know, no code environments that you can push your code and be able to generate the code base? I would say, I likely would be minimal code, minimal code required to do that.

By now we’re seeing there’s always going to be some role where the human involvement is going to be. But in the next five years, what we’re doing in software development activity today, I envision that it will be drastically different. And mostly due to rapid enhancement of the co -animation tool, like co -pilots. That’s my predictions. And I’m holding my breath on those. I really think there’s a lot of bright futures.

coming up in this area. So we have to say.

Kelly (18:26.894)
Howe in five years, I’m gonna have you back on the show and we are going to see if your prediction comes true. Howe Peng, thank you so much for your time. Have a great day and for everyone else out there, you are watching the vodcast. Thanks Howe.

Hao Peng (18:40.958)
Thanks for inviting me out again. Have a good day.