AI in E-Commerce

Welcome to Cornell Keynotes. Today, we are looking at AI at the enterprise level, and we’ll look even more closely at e-commerce, reviewing where it’s been, more importantly, where it’s going, and the underlying tech behind personalization in the online experience. Join us in the studio once again is Lutz Finger, Senior Lecturer at the Cornell Johnson College of Business and Faculty Author of the Designing and Building AI Solutions Online Certificate from Cornell University and eCornell. Welcome, Lutz.

Good to have you back in the studio. Thank you. That was a long, long sentence. I know, really.

We got through it. Viewers, we’ll take your questions throughout, so please drop those in the chat. You’ll see that obviously there on your video player, and heads up also for links that will set you up to our website. To learn a lot more about this topic after we’re through today.

I also really want your feedback on how we’re doing with these productions, specifically today’s show, for instance, right? So we’re going to be putting up a QR code at the end of today’s program. Your feedback is critical, and I really need that from you, audience. So thank you for sticking around, and please submit your ratings for us at the end.

And the questions, and I’m happy to answer any questions. Oh, there you go. Which are stuck over, right? Even better.

So let’s kick it off. So what is the state of the art? The state of AI at the enterprise level? What’s going on out there?

I always ask every guest this who works in this space, because the story is a little bit different with everyone I work with. How do you see things? Well, there’s a huge investment going on. There’s huge excitement going on.

Mark Zuckerberg, as they launched Lama, or as Meta launched Lama, they kind of said, or Zuckerberg said, we’re going to be giving every creator and every small business the ability to create their own AI agent. So there is a version of the new world where everything gets AI enabled. And that’s on the tech side, right? With Mark Zuckerberg launching an open source code.

We have, obviously, on the investor side, a lot of movements there. I just saw lately General Catalyst put a lot of money in AI enablement platform and an agent platform. And that is because they, as many others, as well as Zuckerberg, believe that it’s reshaping our business. You talk in your courses a lot about value.

So is AI delivering on value in your estimation? Well, the short answer is absolutely. Yes. It’s a little bit more complicated than that, right?

Okay. So I’ll put it this way. That is what the markets at the moment see as a multi-billion dollar question. We just had in San Francisco, as you know, I’m from San Fran, Dreamforce from Salesforce, the annual conference.

And Mark Benioff actually joked about all the Microsoft co-pilots. Microsoft has the same vision. They kind of like want to create co-pilots. And then they integrate this into your Word docs, into your data center, into your workflows.

And they want similar to Mark Zuckerberg and enable everybody to use them. And Mark, from CEO from Salesforce, kind of said, hey, these kind of co-pilots have been hit and miss. Yeah. And to add injury to insult.

He kind of said, it’s like the Clippy from Microsoft. You remember? Yeah. The desktop assistant.

Desktop assistant. This kind of clip came up and annoyed the heck out of most of us. Yeah. It was meant to help you to learn how to use Word because there was a world where people couldn’t use Word.

Imagine that. And Clippy was meant, but it was a huge failure. And Mark actually pointed to that those AI experts. Those AI agents that we see might be as well a failure.

And obviously, he’s not only saying this to insult Microsoft, but as well because in his view, Salesforce is way better. And therefore, he positioned it there. But I understand why he says so. Because in my course, we were the first course to completely integrate OpenAI into the course.

I replaced myself as a teacher to actually allow it. I was the first one to actually allow students to interact with the virtual. I give the students an AI tool so that they can do complicated coding questions without needing to code. That they can think about the business case.

And what I see is that students actually struggled initially with this new colleague. And over time, once we trained them on how to use it, they learned it. So when Mark talks about being a hit and miss. Yeah.

And then I can relate to it. It’s not easy for enterprises to make AI useful. Therefore, I really appreciate that we talk today about AI at enterprise levels. So how do you teach back to the value delivery part?

How do you teach this to your students? How do you approach it? How do you philosophically sort of impart this knowledge to them? Yes.

So normally when you… Let’s forget AI for a second. If you build a product. I as a product manager.

I obviously would start with what is the user need? How do I help this user need? How do I make it better and value creation? And this is how it’s taught at any consulting company.

Or this is how it’s taught at business school. Now the tricky and complex part with innovation is you have a technology that might attack a user problem that nobody knew could be changed. So in the course and the whole course is essentially trying to match those two sides. In that course, we go through a process, six-step process, which essentially covers very roughly two different areas.

One area is understand the capability of the new technology. And then the second part is apply it to user use cases. And then the course we go through many different industries and train essentially this. Matching of both.

And then the third step, which I think is extremely important, especially in the AI world is, okay, if I understand the technology and I matched it to a new product or a new capability, what will be the implication for society? What will this change if I put this new product out there? And, you know, we talked a lot about fake news and all the other complications which we have. So it’s important to have those three-pronged approach.

Now. Um. We can actually do this if you want. Go through this three-pronged approach and let’s take one industry.

Let’s look at e-commerce. Let’s take a look at it. So we’ve got a demo here. What are we going to be looking at?

So. Oh, not a demo. I’m sorry. We’re looking at an Amazon image.

So an example of e-commerce. Sort of where we’ve been, where we’re going. Yes, absolutely. Yeah.

I think if you look at e-commerce, you will be surprised. You will be surprised to hear that it hasn’t changed in the last 25 years. Let’s look at an image there for a second. So on that image on the left side, you see Amazon in 1999.

Yeah. And on the right side, you see Amazon today. What do you see there? Shopping is still the same, right?

Like in 1999, you had to select a category. It was written. It was a long list. Yeah.

And in this category, then you had to select a subcategory. And you better know where it is. Or like what you want, right? Yes.

Where it does fit. Yeah. And don’t ask too many questions. It is just a transactional act of buying.

Today, it looks slicker, right? You have more images. But it’s essentially the same. You need to click on an image to get to your category.

And then you have your subcategories. But most of the times, you will find that those subcategories, you can select stars or price. But it doesn’t really describe the subcategories for the product, the product dimensions. For those, you need to click into the product, read it if you care.

So shopping is complicated. Now, what we just saw is the actual user flow. Or like the structure. Now, let’s look at the technology part.

And let’s discuss how can technology be used to help in shopping. I think I have a demo. Do you remember last time as I was here, the last keynote? Yes.

I said, you know what? There are so many amazing keynotes out there. Yes. Let’s make an agent.

This was a good exploration. It happened pretty quick. You fired it up and we got results right away. So, let’s go.

Tell us what we’re going to be looking at here. What are we going to see? Okay. So, I bring up now my screen.

This is my screen here. Yes. And we can ask any questions. Now, this is the collective knowledge of all eCornell keynotes.

So, you can watch all of the keynotes or you could use this tool. As you can see here, this is eCornell.R2Decide.com. So, it has nothing to do with eCornell. It is R2Decide, which I used as a company to create this bot.

You would need to have a login. So, if you do bit.ly.R2Decide, then it’s already encoded. But this is not the important part. The important part is now I have all this knowledge.

Let’s go to the keynote of today. This is the keynote of today. And let’s say I copy to the keynote. Let’s say I copy just the event overview.

I don’t have time to listen to you and me. I just want to know what does eCornell know. Now, obviously, the bot doesn’t know yet our discussion. What have you in the past talked about this discussion?

So, talk about the challenges business face within AI initiatives and where to invest and what not to invest. That’s it. So, I’ve already posted from the website. I run it now.

What it does is the AI takes my input and takes all the videos it had. All the videos it had, it encoded into bit size, byte size. So, I take a video. I take the video into text.

I store the text. Picture this as a way to store information. It’s like a container. It’s a container.

The way you described it. Yeah. We call it an embedding model. Okay.

In a vectorized world. And if you want to learn more, come to the course. But the whole idea here is you take the context and you store it. Now, you take my question and then the question, the computer tries to figure out where in this containerized models do I find pieces and answers.

And let me pull them out and then I have pieces and answers, a little bit here, a little bit there, a little bit there. And now, I bring it together and summarize it and answer back to you. That’s essentially what transformers do. This summarization and answering.

We know all this from ChatGPT. We can take a text and put it into ChatGPT and say summarize. Here, we say, oh, I took only Cornell stuff, put it up and summarize it to me. The other thing which we are using here is what is called a RAG, Retriever Augmented Generation Model, meaning the fact that I only take eCornell’s information and store it and saying this is the container.

Don’t use a general knowledge from ChatGPT or Gemini. Use only the knowledge from eCornell. So, if I see this here, you can actually see it talks about Karan, suggests the three-pronged approach. This was another host who was a host of a previous keynote.

So, again, what we’re doing is we’re demonstrating principles on how this kind of works. So, in this case, the knowledge base or the fundamental data that informs this is the knowledge base. Yeah. Yeah.

This is essentially our YouTube channel. Karan has an amazing course on generative AI. Yeah. So, obviously, when we talk about the challenges and he, I assume, I didn’t see his keynote, but I assume that he talked about those in the keynote, the computer finds it and brings it up.

Right. Right? And it will go through and you see that it brings different videos in. Here, for example, is the video AI and future of work.

And it jumps directly. Yeah. Yeah. It jumps directly to the position where, in this case, this lady, I think she is the IBM Senior Vice President, where she talks about the… AI and the future of work.

…byte-size experimentations can be beneficial. Okay. Right? Yeah.

Now, that’s what we can do with technology. We can save information. And then we can retrieve it in a way that we can use it for other things. Yeah.

Yeah. And then we can retrieve it in a way that would be similar to our conversation. Mm-hmm. So, it’s a little bit of a long walk to get there, but, you know, ultimately, this applies, again, to some of the e-commerce stuff.

And we’ll see this demonstrated as the logical extension of this tech when we come back to the e-commerce part. All right. So, tell me what you can find out about phishing lures, for instance, in this case. Phishing lures.

Just… Tell me about phishing lures. What has eCornell produced as it relates to phishing lures? Lures. It relates to phishing lures.

Phishing lures. Here we go. Stick with us on this one, audience. You’ll get it.

Phishing lures. Okay. That’s an interesting one. All right.

So, same thing. I’m afraid I’m not an expert in phishing lures. My expertise lies in various topics offered by eCornell, such as business, leadership, hospitality, and more. I can insist.

However, it says, however, if you are interested in learning more about sustainable practices in hospitality, you can find out more about phishing lures. So, I’m going to give you a little bit of a quick overview of phishing lures. So, phishing lures is a phishing lures. It’s a phishing lures.

In business, business, business, business. It did not go and talk in general about phishing lures. No. And we will see why this is important once we come to the e-commerce application.

In this exercise, we did not take general information. Karen is an expert. I don’t want anybody talking about anything. I want Karen to talk about it.

By the way, this was like sometimes you want an expert knowledge. Sometimes very often you want a specified expert knowledge. And ChatGPT is a general knowledge base. Can I tell you a funny story?

As I created the course, I obviously used a lot of ChatGPT or u.com to help me structure content. But when it came down to the risks of AI applications, ChatGPT just gave me very generic answers about the risk. Not because they don’t want to talk about it. There is no concern.

There is no conspiracy theory behind it. No. Because it’s not yet such a clear topic with very clear points to make. There are concerns in different areas and it brings up all those concerns.

But the state of the art, it can’t do because that’s not the general knowledge. So phishing lures, there is a general knowledge. You could probably talk to ChatGPT about it. But for the e-Kernel part, we didn’t have it in.

Now, I’ll show you one more thing. It’s a very interesting topic. This is… Tell me something about emerging markets.

What will happen? Okay, a little German prompt. What did you say? Like, tell me something about emerging markets.

Okay. What will happen? Let’s see. Now, this is funky because now I actually ask a German question.

The computer translates this German question again into this white containerized space. It’s actually not a containerized space. We call it a containerized space. It’s an embedded space.

You said containerized. But therefore, I want to bring it back. No, don’t worry. Don’t worry.

It’s an embedded space. Yes. We take that information in that embedding space. So think about a vector pointing somewhere.

And here, it uses my question about the emerging market. And it places itself. And now we see here, An Mirou talked about emerging markets. Mark Möbius, he is a very famous person, as well, talked about emerging markets, right?

Yes. And you see always the areas, again, it jumps to the right place in the video where he is talking. You see as well, Lourdes here and others, who are all the right people to talk about emerging markets. They are the brands, the known people.

Now, the important setup here is, I can ask in German, why? Because the computer embedded all the information it needed to know in one space. In kind of like a common computer language. And whether I talk in German or whether I say it via voice, it doesn’t matter because it brings it over there.

And that explains, if you have launched last time, OpenAI’s developer conference, and they do voice-to-voice, that’s exactly it. They go in this voice, figure out this big space where all the information is, and come back this voice. So we’ve talked a little bit about value. We’ve talked a little bit about technology.

How about how to apply it? How does it look once we get into the application phase? I like the principles that you’ve covered here. How do they play out when you apply it?

So now, when we say there’s this approach which goes first, knowledge about the technology, which we just covered. Now I need the use case. Now I need to come back with my hat of a product manager and say, okay, what is it, what we actually want to do? And in order to come back with a product manager discussion, I would suggest let’s talk a little bit about how do you shop?

Because we said we want to apply this technology in shopping. Let me ask you, did you shop something lately? I shopped for a chef’s knife. A chef’s knife.

Okay. I have no clue about chef’s knives. Okay. Sounds good.

chef’s knives necessarily. I knew a little bit about them going in. That’s kind of how I did it. Since I’m like, as it comes to chef’s knives, I’m a pretty dumb person.

Can you explain to me what’s important? Well, it would be a number of things. The material, the weight of it, how it feels in your hand, the applications for it, how long the blade is, how tall the blade is, what you want to get done with it. And you learn a bunch as you kind of go along the way.

I had learned differences between dimpled blades, right? So that your tomato doesn’t stick. All that kind of stuff. Nice, nice.

Functions. What do you want to get done with this? And ultimately, I decided that I had a certain set of criteria and requirements that had to be met. Nice.

That’s a rabbit hole. And we all have been down that. Yeah, sure. Kind of, you want to buy something.

I imagine that knives can be, because you see, you can buy knives at like… $5 and at $5,000. So there seems to be some… I had a sweet spot, right?

Yeah. I wasn’t going to spend too much, but I wanted to get a really quality thing. But that means that there is complexity. So as soon as it becomes important for us, you need to map out the complexity, as you did.

Understand what is the complexity, understand what is important for you, understand what is true, what’s not true, what’s paid placement, what’s not, and then you take a decision. So we can actually, what I can do now, I apply the same structure of the technology to this process. Yes. Because essentially, what you did is you created your own space, you read all this material, and you put it all up there.

And then you ask yourself, from what I put up there, what is important? And then you summarize this, and then you looked at offers, and you matched it back into that space, right? That’s what happened. Exactly what we saw with Econaut’s Keynote.

Keynote, you did in order to buy a shopping knife. Yes. So meaning I have now a technology that could help in shopping. And for all the people who are saying, oh, AI is more hit or miss, or co-pilots is hit or miss, let’s look at e-commerce, because they are the first to actually adapt those setups.

And I mean, let’s go to Amazon. And this is the, I’m showing on my screen, it’s the Amazon website. Yes. So let’s see what Amazon as the number one e-commerce site in the world is doing here.

It has Rufus. Rufus is their AI shopping agent. Top left. How long has that been available for?

I think it started in March. Okay. A lot of people made a lot of criticism, because it ain’t that easy. Yeah.

It ain’t that easy, and we will see. Okay. So this is Rufus. Okay.

It gives me, get recommendation. I can click on things if I don’t want to write. And this is the first biggest problem. Can we zoom in a little bit on that?

Yes, we can zoom in, I think. Great. Let me zoom out a little bit. This is the first problem, because Amazon does those light boxes you can click on.

Yes. Why? Do you remember Anna from Ikea? No.

Oh, that was a tough one. It was terrible. One of the first chatbots popping up and completely useless. Oh.

Essentially, it told you where can you find, if you would type in a bad, it wouldn’t know that you need a bad. It just would point you to the right websites. I mean, it was essentially the 1999 version from Amazon, but instead of having a menu to scroll through, you actually had to type it. And if you typed it wrong, it wouldn’t end up.

It was a terrible experience. So that Rufus out of the gate is making recommendations for things, but it doesn’t really know what you want. No, but it has to make those because people still remember all the bad bots. And therefore, I don’t want to type.

I don’t want to engage with this one. However, this is new technology. You actually can engage. Okay.

So you can say something like a chef’s knife, or like I want to buy, let’s be kind, a chef’s knife. Okay. If I click return, you see that it starts looking similar to as we saw earlier. Yes.

And now look at what it’s doing. When selecting a chef’s knife, prioritizing quality materials, ergonomic design, and optimal performance and comfort during prolonged use. Invest in reputable bands known for their craftsmanship and durability to ensure your knives withstand the demands of daily kitchen tasks. And then it gives you.

A couple of. Options. Products here. Products.

Now I could actually ask it. Can you compare them? Compare those options. These are new prompts.

This is new stuff that you’re doing. You know. Yes. In typical product search, this is not how it works.

This is not how. Like in product search, I would just have like the search bar and I put a knife and I get it. So here, Microsoft is using. The technology I showed you early on.

Yeah. Saving all information about knives and getting out the right setup. And now compare the options. Now look, this becomes a little bit dull.

So I actually ask, help me to understand. As you probably would. And it now it sends like, okay, it’s important to have different quality materials. Yes.

That’s it. Now. You just went through this process. Would you be sufficiently happy with this level of detail?

No. Well, there’s several other characteristics you have to think of, right? Weight, how it feels in your hands. What it can cook.

All those other requirements that I said. Yes. Yeah. Probably steel.

Where did it be? It was steel. Yeah. Steel was for you the most important one?

That’s what they’re made of. Yes. Yeah. So what steel did you use?

Do you want me to tell you? Yes. With Japanese steel is where I ended up. Nice.

Yeah. Yeah. Nice. Okay.

Yeah. Like since I lived in Japan for two years and I know that like Japanese steel was like. I came close to a German knife just so you know. Okay.

Here we go. Yeah. Here we go. Don’t take offense.

This is all good. Okay. We don’t also like, dear listeners, we don’t do any recommendations here. Yeah.

But what you can see here is Amazon gives you an answer, but the depth of the answer is not the depth you need. Now in Amazon’s defense, there are two things probably at stake. Number one is Amazon doesn’t know that people go in with a clear expectation of help. This tool is on the Amazon website.

So far I could only search and click on categories in the last 25 years. So now suddenly I have somebody who is helping me. Are you kidding me? This is uncharted territory for us as a user flow.

So the huge problem for AI in enterprises is there is an existing flow and you’re trying to change this. So that’s what we see here. So Amazon is probably saying, I do not want to give a whole long story about how to select a knife because Chris might not be ready for this. Okay.

Right? Because I don’t know that Chris is a geek. Right? Right.

So the other reason what I also… This would be for Amazon. The other reason which I think is also at play, if you go to Amazon’s website, you probably don’t find user reviews that are so in-depth to the level of quality you want to have it. Not most of the time, no.

Yes. Meaning Amazon has now the problem, they have all this average discussion about knives and you’re a pro. You don’t want average. You want an in-depth discussion.

So you probably want to go to Gordon Ramsay or… Somebody who really knows something about cooking to have a discussion. But let me show you one more example. I have a pet at home.

What a… Okay. My cat peed on my carpet. That’s kind of annoying.

That’s an interesting prompt. So what are we seeing? When selecting a chef’s knife, prioritize quality material. Oh, no.

That is actually… Sorry. This should have not happened. This is…

It peed. Let’s see. Oh, it’s still doing the chef’s knife kind of thing. No, it is still doing the chef’s knife.

Let’s actually… Let me… And by the way, you see… Sorry.

I hope that it will talk to me. Yes. Cat peeing on carpet. And this would be…

Now, the search. You see that the search… Cat peeing on carpet is actually giving me already a very good answer to what I should do. You’re probably going to clean the carpet, right?

I probably clean the carpet and therefore it probably gives me… And it gives me a trimmable cat scratcher, which is not helpful. It gives me some carpet cleaner. But this is search.

This is search as we know it. Different function. What I want is, to be honest, what do you do when a cat… Like if…

You know, you act quickly. You can use baking soda. Let me see. It should actually…

It should know it. I tried this before and at that time it knew it. That’s okay. You know, we could at least talk about what your experience was with your prompts as it evolved.

But this is a good example. Hopefully, it’ll pick up and do what it’s supposed to do. So we’ll see what Rufus does. Amazon, please don’t…

Oh, here we go. Okay. So what we just saw was a glitch in Amazon. Yep.

Now… If you are out there and you have an enterprise. And you have a… You have a business and you are trying to make AI useful.

I will make the argument today that you can make it useful. But if it doesn’t work directly, look at what happens to Amazon, right? I mean, they are definitely an amazing company in terms of knowledge. Yes.

They’re doing this since March and it just failed us. So it can happen. It’s software. It’s okay.

But okay, here it comes. Cat peeing on the carpet. Here are different solutions to stop cats from peeing. Okay.

Cat peeing on carpets. Okay. Now it gives me deterrent. How I can stop it.

This is actually pretty nice. Yes. And it says clean thoroughly. Use like a cleaner designed for pet urine to remove odors, right?

And now it actually offers me the different areas. I think what helped, by the way, is the context that I had here on that page. Yeah. I think it’s a really good context.

I played around with this tool for a while with Rufus. When you are at the right page, the Rufus actually knows. And this is like a good salesperson, right? A salesperson coming to you.

Oh, I see. If you are in the department for knives, it will not ask you, do you need a cut cleaner, right? Yes. I understand.

It will say, what type of knife can I help you with? Sure. That’s actually brilliant. So this is, again…

And like the brilliance is, remember, we left the whole discussion on technology quite a while back. Initially, we talked about what’s feasible. And now we moved stepwise over and said, how do you use, like at what point in time? How do I make this more trusted for you as a user?

How do I use context variables of whatever I see on the screen? And you see how this one here is doing a good job of giving me some context. Yeah. Yeah.

So I’m going to use several options and explaining it to me. I personally, as a user, would have liked that the products are actually in that chat that I can select it from there. But again, that’s a user discussion you can build in. This is no rocket science.

The new technology is that I can say something like, my cat peed on the carpet and I get help. Or I can say something like, I need a chef’s knife. Yeah. So in an ideal world, which it didn’t do in the chef’s knife example, it would say, hey, you know what?

Chef’s knives are complicated to select. They are those things. And in an ideal world, it would figure out, where is Chris? Chris, tell me.

Is it, do you, are you a geek? How much do you want to invest? You can spend thousands of dollars. How much should I tell you before I get you down the rabbit hole?

Sure. Sure. So there would be a person guiding you down the rabbit hole so that you don’t have to spend on consumer comments or like customer reviews and all of this. Can I ask you a question about this?

The cat peed example. It came up with preventative advice for you. Where is that embedded data? What’s going on with that?

What’s going to be on the backside of that? Yeah, That’s a good question. So I mean, I don’t know it for sure. Okay.

But I assume that if you… Yeah. Yeah. I assume that if you look at prevention, like a trimmable adhesive mattress, right?

Oh, I see. Somewhere it will have said, and if your cup is happy, it like doesn’t go to the bathroom that often or something like this. Oh, I see. So now suddenly you have this piece of information in it.

Yes. Also, I mean, you know, world is complicated. Let’s ask Rufus. Who will…

What are some of the best practices that you could use to correct some of the problems that you’re having? Yeah, So, meaning, there is the data, as we described it earlier, like… Relevant to cats, yeah. Like, relevant data.

Okay. It could be, and we talked about in the last keynote about this, there is not only one model to rule them all, there are several models. So it could be that they as well go outside, ping chat GPT, or ping an API, or run their own LAMA model. to get more information which they combine.

It could be in those stacked processes. There’s a pipeline. I show you how this will work. Who will win the election?

I can’t make political predictions, but I can summarize product reviews and answer other shopping questions. Did you see how fast that answer came? Yeah. Because before it starts looking at all the documents and drinking, the first step it does is, is this even the question I want to answer?

Right? As Chachapiti came out, many people tried to break it. Rufus is the same thing. Microsoft had the bad experience, and I think Amazon really wanted to stay clear of that.

Now, all of this is, and we can move on and don’t have to spend time on Amazon, but all of this is to say, we have a technology, and now I brought this technology into your use case. How do you see long-term how this will impact the way that we shop? Tremendously, right? Because how do we shop today?

Today we have first the need stage, awareness stage. Like I need to have something. I saw that somebody had something. Then we have, the consideration stage.

In that stage, I do what exactly you did, like inform myself, get information. And then we have the actual closing or shopping or stage where I do something. When I said Amazon hasn’t changed in the last 25 years, then at this stage, it hasn’t changed. But also the awareness and consideration stage, yeah, you know, we got Snapchat, since I worked for them, right?

This is creating awareness. This is creating consideration, helping with information in the stories or in the advertisements. There has been changes in media format. But how do you create it?

Well, you used to ask your friends and now you follow the Kardashians. But this is essentially the same approach. You’re trying to bring this information all in your brain, memorize it, structure it, like for the knives and kind of saying, ah, steel is important. I go to Japanese steel.

And the company which had most power there was Google, right? Because Google had the ability to place cookies and to follow you around. So once you started to look at those things, it became Google. For a while, I thought Amazon will become way stronger than Google and shopping.

And, uh, it hasn’t turned out that way. Amazon still gets a lot of traffic from Google. People searching, doing things, analyzing things. And once they have enough information, then they go into the consideration stage and then they go to Amazon.

But now this is about to change because Amazon can, as you see, doing the whole consideration space on their self, meaning the whole search terms, which we did in Google, like video watching, looking at this, which for Google is valuable information. That’s what they sell, right? That’s how they became so powerful. Amazon can move this more and more over to their site or sites like, um, review sites.

You talked about the best seven or like Tom’s electronic guide or whoever it is. They could move now all the discussion internally, leaving Google, uh, allowed a little bit. So are there, are there anybody, excuse me, any organizations that are challenging or, or present threats or disruption in this space to Google or Amazon? Yes, totally.

And, um, yes, absolutely. So for example, in, in my course, uh, um, I had Richard, um, the founder and CEO from you.com. You.com is, um, is a search engine. And Richard is, very clear in the course and he is describes it very needly.

Um, that there is an opening in the opening to challenge Google in the pure existence. And, uh, he just traced another round. He is doing phenomenal. It’s really a good service.

Um, I, um, I said it earlier, like, um, my course, I use not only open my, I integrated with open my eye, but I used a lot you.com because it shows me transparency of links and, um, um, it has the ability, um, the ability to create different depth levels. The, the issue which we just had with Amazon, it’s not solved yet, right? Like they, one would not need to know who is Chris in order to know how deep to go in the question about chef’s knife. And you.com is helping me to know this level of depthness, which is pretty good.

It, it may, I tries it as well by saving the data about you. And it’s always says memory updated because for that reason. Um, so we have you.com. Um, we have perplexity, the black city just recently kind of announced how they want to do advertisements.

Um, because that’s for them, the future business model, and it takes money away from Google. Google is obviously doing Gemini, um, uh, and open my eye, right? Open my eye big out there. Um, they are currently not yet pushing into this, but read my lips.

It will happen. So we have, we’ve got Google from what I’m hearing and you on the one side, Amazon, and then on the other, on the other side, how can, how do you, how do you explain, or how do you characterize the interrelationship between those two, two giants or three? Yes. Fascinating, fascinating discussion.

Um, we have this already today, but today everything was dominated by Google. Therefore it didn’t play out so much. Now it will play out. You have essentially you have, um, sales side advice and buyer side advice, right?

If, um, if you are at Tom’s, um, electronic guide or you are at the REI website or you are at Amazon, that’s a retailer and that retailer will give you advice and it will give you only advice as long as it can sell you something. Not quite. We saw an Amazon’s case. It actually recommended several steps and not every step was connected to a product, but it would not be completely, um, without sales intention.

On the other side, we have, um, you.com and the others, but also they have a business model, right? They, they will also place certain products in your stream. So we will see the conflict between those two groups, one specific for an area from the products they have very close to the purchasing point and one more generic. And the generic one by the virtue of generic, we’ll have more, more problems to actually go into down the rabbit hole, right?

So for, for those in our audience, you know, we’ve, we’ve gone through the buyer experience here for those who actually sell things. Is this tech available to us? What, what can we, how, you know, how can we, or they use this technology on the seller side? Yes.

So the answer is 100%. That is the vision Mark Zuckerberg laid out, right? As I started with that, he, he said every business, every small business owner, everybody should have their own tool set to make. And that’s what I was getting at small business, small business.

Yes. So the website you saw earlier on are to decide, I should give a disclaimer that this is my company. I like as I did equal now, that’s essentially one of the ideas which we have to, to make, to allow everybody to simply create agents, sales agents, explainable, agents. I use this to replicate myself and to write easier LinkedIn posts, which I then just need to check and rewrite.

So everybody will get this ability to have a simple tool on the side, but I want to warn about being overly optimistic. What this tool can do as we saw in the chef’s knife, what is important for a knife? I can tell you what, what people in general say, what is important for Chris? You need to decide and how to wait the different important pieces you need to decide.

So we make the, the mechanical task of summarizing structuring, obstructing easier, but we don’t take decisions, right? So very often when people talk about AI and how it’s for enterprise, we go from technology, very quickly over to, and it’s ruling the world. No, it’s not. Uh, it’s, it’s a technology.

We have now e-commerce as an example. And, um, in that e-commerce example, we use that to make our decision flow easier. It’s still your decision, what you want to buy in your courses. Trust.

Trust is a through line. It’s kind of a theme. And you know, if we think about the Googles and the Amazons in this space, in the end, they’re trying to sell us stuff and, you know, trustworthiness of, uh, these things feel, look transparent sometimes, right? Google is just trying to sell me something here.

Google is trying to sell, Amazon is trying to sell. There is actually, um, a fun story. And, and my students know this. I, I love coffee.

I’m a coffee nerd. Okay. You, you have gotten good coffee from me as well. Yes.

So, um, I became a coffee nerd because I heard about the Tesla of the espresso machines. It’s, it’s a, it’s called decent espresso. Yeah. Um, and, uh, as I initially heard, like the, value pitch, of the decent espresso is essentially, we help you to become a good barista because they, have a whole tablet attached to the machine and they measure everything, flow rates, temperature.

Like it becomes very, very geeky. Yeah. Now my students know this because all this data, I actually give my students 1,300 espresso shots that I did and let them then build models. And then we later use that data to figure our taste profiles.

And then later we go into large language model and use that again. So, coffee becomes this kind of funny themes throughout the whole course. Um, but if you, as I started my, my discussion, I was in the market for buying an espresso machine and I heard from a friend, awareness stage, right? Awareness, consideration, purchasing.

I heard in the awareness stage from a friend, the decent espresso machine is the thing. And I Googled it and obviously I didn’t get anything because the owner of the decent espresso machine is actually very, very clear. I’m not going through traditional channels. In its sort, it was a niche product at the site and I could only get the information when looking at trusted sources like my friend who told me about it.

Now the decent espresso machine, there is a self selection criteria. All of these people are extremely coffee nerds. Like you, like think about you buy your knife and you go to a group of only Jordan Ramsey’s only people who are like, that’s a kind of freaky thought here, but an affinity group of affinity group. Yes.

Only people who know extremely well. So I, that group and now I am, I’m so excited about this machine that I started to get to know the founder. Uh, and I work with the decent espresso machine. I actually did the same thing.

I’d get gathered all the information from that diehard community. And if somebody wants to know something about, for example, can I run this machine with, with solar panels? There is no information in a spec sheet about this, but somebody had discussed this and you get obviously this piece of the information or do I change my tablet or can I use a different power? I want to remove the engine, whatever it is.

So technical machine, detailed information and trusted source. So let’s come back to Google versus Amazon. Both are trying to sell you something. There is a third space of trusted sources.

And last time I talked about and said brands and influencers will become the new norm to define things. This could be as well the trusted area of information. So what is, what is decent actually do with this community data? And you’re suggesting it sounds to me like maybe it’s of higher quality.

It is. It is 100% of higher quality. So decent is not doing anything with the data. They’re kind of like half this as a community.

The community loves them. It’s an ablement for the community, but they’re using it to help the community by using us to ingest all the data and then help them. But the data is of higher quality and this is actually a super fascinating discussion. Google just like, you know, Reddit, right?

Yes. It’s a $60 million deal to use Reddit to train models. And Reddit makes it very, very clear. Nobody else is without giving us money is allowed to use the data.

Why? Because Reddit, these are real people talking about real problems. And these are people, you go to Reddit only once you’re down the rabbit hole. I’m pretty sure that your knife discussion touched you on Reddit.

Sure. Yes. It did. It did.

So people who are like you, who come together and who live and die for like knives, chef’s knives, like you have a decent group and Reddit as well. Now decent has their own group. Now this means Google can crawl this information because it realizes, hold on, I might not be the trusted source if I just take Amazon advertisement pages. So Amazon might say, oh, I might not be the trusted source.

If I only take advertisements. So everybody is, funnily, looking for the human element in the data. And that being human becomes a value, a value in itself because the computer can only iterate what was said before. And therefore all the models can only know what was said before, what’s stored in that space.

If I want to really something new, then I need to talk to humans. So that’s a reason why Reddit is in such a good position. And after that day, there was a time where you could not go to Bing or DuckDuckGo to see Reddit because Reddit says, no, I want a deal. And this has completely changed the way we work with data.

There’s another good example, Mistral. Do you know Mistral? I don’t. Mistral is essentially the OpenAI follower.

It started in Paris. Mm-hmm. Yeah. So OpenAI is aiming more on the consumer side currently.

Mistral is doing the large language model enterprise version. A little bit under pressure when you have an open source like Lama out there, but a different discussion. They built this follower with way less data compared to OpenAI. How?

They used quality data. They did a lot of effort in looking at the data while OpenAI used all the data and then it had all the replicated reports. Yeah. Repeated SEO optimized data, all of the trash in it, which doesn’t help.

So I want to get into AI agents, shopping agents. Yeah. Tell me a little bit about this. Well, I mean, this is the whole story, right?

We kind of said AI for enterprise, it’s complicated. Let’s go through one time the cycle and the cycle is essentially, what’s your business case? Shopping. Shopping.

What’s the technology enablement? Oh, we have a technology that can search, summarize, construct and talk to you as a person. Oh, we could use this as a shopping agent. And how do we implement it?

And then we saw the implementation of Amazon. Correct? So in summary, for me, the future is already there. If Mark says all of those co-pilots are not useful, then I beg to differ.

It’s not good for on-stage announcement if he has his own pilot, a co-pilot type of style. The problem is how to make it usable in your company. And as we saw in the Amazon case, it’s not that simple. Like in the case with eCornell keynotes.

Yeah. Like you don’t only take the content from all the keynotes. You need to figure out guardrails, like which content to use, which person to mention, how long you want to talk. And if you do Amazon on top of it, you need to figure out, oh, should I place an advertisement?

Should I place the advertisement where the text is below? How often should I place the advertisement? And so on and so forth. So there are a lot of complexity.

And the journey has just begun. But one thing is for sure. 1999 Amazon and 2024 Amazon looks the same. Next year, it will be completely different.

We will have agents guiding us in our shopping experience. And I think Mark Zuckerberg is right. We will have a version or like we will have a future where every business and every small business owner can set up their agent. It needs to be done right.

But I think that’s the problem. It needs to be done right. But that’s what we talk in the course about. I was just going to ask you.

So a lot of the principles, themes, throughlines in our conversation today. How do you map some of the things we discussed to some of the course content? How does that present itself for students? For students, yes.

So essentially in the course, I start with whatever we do at the end here has to make value. So first, let’s get the basics right in terms of how do we work with the content? How do we work with the hype? How do we work with a structure?

So I have this framework of questions to ask. I have this framework. And it might be like every university, every consultant, every teacher comes up with some form of framework. So I actually started to collect them all together and give you a mega framework of like select one of the value questions.

That’s the first part. Then I go into, okay, now let’s talk technology. Let’s talk about the future. Okay, now let’s talk technology.

I explain the technology very simple. We go, we start with simple linear regressions, logistic regression, decision tree. So we like inch towards models. And then I talk about the data, the discussion we just had about Reddit, about why clean data is way better because we need to touch on bias.

We need to touch on misuse of data. All of this is a very complicated part. And once we have established this foundation, I go into neural networks. And then I go into generative AI.

Generative AI is nothing else as many linear regressions, like the beginning we talked, in an activation function, logistic regression, stacked on each other, turned around. So it’s actually very simple. So throughout the course, I get you to complete understanding of the technology. And then the biggest part is I go industry by industry.

I talk about healthcare. I talk about the data. I talk about law. I talk about media.

I talk about e-commerce. And we look at how can those technology capabilities being attached on the user flow. Fun fact, will I be always right? No, I’m pretty sure the AI agents will be a thing.

But for other of my hypothesis, I might not. But now the good news is Econel and I have agreed that I update the course monthly. So monthly, I will look into what has happened in those industries. And come up with the next new thing.

And I use my virtual Lutz to do this. So it gets quickly. But for anybody taking the course, you will see the most up-to-date information about the industry. Thank you so much for joining us in the studio today.

I learned quite a bit. And I want to thank our audience for joining us as well. Lutz Finker, thank you so much for connecting with us today. I learned a lot.

I hope our audience did. Whether you take the certificate program or not, I think you’re going to learn a lot from this. Thank you.