Let’s Talk AI: Up close with Google Cloud AI and TTEC

Transcript

Hello, everyone, and welcome to another Let's Talk AI.

We have a very, very cool, episode today.

We have Ben who's gonna join us from Google Cloud.

Just to give a little bit of an intro around Google Cloud, a lot of us, have seen the videos that we post in our internal community, and they're learning videos. So Google Cloud is a platform that helps businesses and developers build, innovate, and scale using powerful tools for AI, data, and cloud computing. It's all about making technology accessible and solving real world challenges across industries.

So who better to talk about, than Ben? Hi, Ben. Hey. Welcome to the podcast.

Thank you. How are you?

Good. Good. So thank you for joining us today.

To start, if you can just give us a little bit, of an intro around yourself, what you do at Google Cloud, and a little bit about your personal journey.

Sure.

Yeah. I work in, Google Cloud Consulting, which does the implementation is the implementation arm of Google Cloud. So, I specifically work on the AI services team, so we build custom AI models for Google Cloud customers. That usually means it's the the weirder, wilder, you know, riskier models that get built, so we get to that challenge that's a little more involved, if you will. But, yeah, we deliver those Google Cloud customers, we put it in their environment, and they and they run those models.

How I got here was, I used to work on the ad side, but I was trying to bring AI to ads, YouTube ads analysis. I wanted to figure out what made a great YouTube you know, what made people not skip the ad. Right? And I wanted to figure out what that was.

So we used AI to analyze ads and figure out that, you know, certain things did better than other things, like fire, in the ad almost always, you know, people pause for a second because they like, well, looking at it. So we did a lot of analysis around that. That's where I came from originally. But, prior to that, I worked in, I ran a data science team, at at Publicis Worldwide, which is an ad agency network in New York.

And, I started playing around with AI probably when I was in grad school, because I was trying to I I was trying to speed up the data analysis process, to be honest. I was just getting real I was like, I'd rather code something for an hour and not spend twelve hours analyzing it. You know? And so that's how I got into it.

And back then, you know, that was a number of years ago. It was not half, not even half. It wasn't even a tenth of, as as viable as it is today. So it's if I was if I was me, you know, ten, twelve years ago, I'd be delighted to see the technology we have today.

So, yeah, that's my background.

Cool. Cool. And, you joined, Google AI, Google Cloud AI, and, you know, you shared, all that and the background around ads. I just wanna understand now, how has this shaped a little bit your view in AI? Like, your journey at Google Cloud, after joining there and, you know, working and all those AI stuff, how has your personal perspective changed? That's good. Or has it?

It no. It has. Definitely has. I it would be weird if it didn't.

So I was teaching at Columbia University prior to joining Google Cloud when I was in the ad side. And I realized that the way we talk about AI and there's a lot of, like, psychological kind of fallacies that happen. So I taught researched and taught a class and still teach that class called AI for the knowledge driven organization.

And it was partly a personal journey, which is, always fun, but it was also, a lot of research and found that, one is we we we really want to personify AI. We wanna say, like, oh, it thinks this or it's trying to do that. We wanna psychoanalyze it, and it's just we're not it's not that it's not there's nothing to analyze on that level. Not yet, at least.

Right? And it makes for great headlines, and it's it makes for really kind of interesting conversation, but it's it's kind of it's a misnomer. It's like trying to figure out the motivations of a potato, which I'm sure they have them, but we can't detect them. Right?

So yeah. But I I've noticed that when we try to personify, so I try not to do that now. And once that makes it a little more boring, to be honest, but a little more honest. It's actually, you know, it has an objective. Right? And that's trained to get to that objective, and hopefully it does. That's pretty much all the psychoanalysis you can do of a of a of an AI model.

I've also changed the way on on the way I kind of I'd say actually, like, see society, which is kind of a bold statement to make. And what what I mean by that is I'm starting to realize that we agree on certain things a lot less than we used to. That's I think a lot of people would agree with that statement.

When you work in AI and you have to have you, you know, you show a hundred people a photo of an alligator, and you say, is this an alligator? And twenty two of the hundred say, no. That's a crocodile.

It's like, okay. Now we have to separate crocodiles and alligators. Like, that's fine. There are they are different species.

Okay. And then some people say, oh, that's a Chinese crocodile or Chinese gator, and that's a the Floridian gator. It's like, okay. Now we have to break them up.

And you start to realize that, what we agree on across the board is much less than we think. And I I think people know that from, you know, reading the news and and and so on. But I'm starting to see the actual numbers. It's like, oh, is the sky blue?

It's like, well, actually, technically, it's pale blue. It's like, okay. Got it. Like, alright. Not yeah.

But the general blue, you know. You know? And they're like, oh, okay. Yeah. It's it's blue.

And then you find some people say, no. It's clear sky. It's technically not blue because the blue is the reflection of the of the ocean. It's like, okay.

Like, you know, so you never get to a hundred percent agreement, and I've sort of it's changed the way I view the world as I start to realize that just, like, everyone's got a different opinion, and they might not all of them, but a lot of them, a good portion of them are technically right. And you just gotta not not fight over that stuff. Right?

So I never thought about that, to be honest.

On looking working at the background, you must see all those different perspectives. I you reminded me now, I was looking I don't remember where it was that they were doing that for AI cars, where it was putting them to take a decision on if it will hit one old person or three kids if it will hit two.

Yes. The moral hit. Right?

It was so interesting. I'm like, how many different opinions? And to build that, how you have to take into consideration all those different opinions that people have.

And what's terrifying about that project is they were trying to come up with the they they assumed falsely that there was agreement on what the car should do, and then you start to learn to, like, there's no agreement. And then even if you take everyone in aggregate, you start to realize how even in aggregate like, if you read the aggregate results, it's horrifying. Like, oh, a pregnant woman is worth, you know, three male criminals. You know, like Exactly.

Well, like, you know, like, we start saying like that statistically. It comes off really, really cold and calculating, but it is in aggregate, correct. But I I can't think of a better you know? So I think, actually, the point of that, study was to find out what we all agree on, and they found out very unequivocally that we don't agree on a lot.

And when we, even collectively, when you look at the averages, you're pretty horrified. Right?

So it's it's, it's and I'm not even agreeing or disagreeing with any specific one because it it is overwhelming when you do a bunch of them.

Yeah.

But I think that's why we don't agree as much as we thought.

Yeah.

And that brought me to a personal perspective of, whenever sometimes we judge how something responds, like, from AI or whatsoever on imagine what it takes on the background and, are you ready to really be into the decisions of, what is the client's name?

Who wanna get into ethical and responsible AI need to understand this before the that because I think that people come and say, no. This is right and this is wrong. And I'm like, oh, you're gonna have a hard time, buddy.

Yeah. You know?

Like because because you you don't understand that not everyone agrees.

And this like and that that goes to moral psychology, which is, like, a whole field of study. And we start to find out that cultures and all these things have a lot of different things where everyone's convinced that they're right and everyone else is wrong, and it turns out, well, then no one's right, at least in collective collectively. So, yeah, I think people need to understand that, and that that's what makes it hard, actually. I think we're getting good at the data part of AI.

We're getting good at the the technology has made leaps and bounds the past just five years. It's incredible what we've done, on the technology side. But if we can't agree on what it is, the technology really doesn't matter at that point. Right?

So that's that's where I think the struggle is, and that's what my personal opinion changed. Now now I, you know, when I when I'm in an Uber, and I I love chatting with Uber drivers, and I know this is, like, some people absolutely hate this and some people love this. I listen to the Uber driver, and I have no judgment. I just I just probe further and further and further.

And you start and, you know, they're random sampling as far as I'm concerned. Right? Like, random sampling of just the guy, you know?

And I listen, and I and I probe. I'm like, woah. Why do you think that? Not challenging them. Just and it's amazing how they'll open up. So you've gotta do that research, and you gotta figure out in AI how to do it at a large scale, which I think is very hard.

Yep. Talking about scale, I want to talk about, the simplifying adoption piece, that you guys work with work at, from Google Cloud. One thing I'm very, happy to see from Google Cloud is that the goal is to make AI accessible to everyone, you know, from developers to nontechnical teams. I'm a nontechnical guy, and, I play around with AI with, like, low code, and try to build my own thing. So can you walk me through a little bit, very high level from on how you use tools like Gemini code assist, Vertex, AutoML? How do you use all those to simplify the AI adoption? How how do they empower both businesses but also individuals to put AI into their workflows?

On the individual level, I think, you know, you start with the with the the large language model, the chatbot, the Gemini Gemini advanced. I think that's a good way to learn the language, and that's where you start to learn, like, oh, the way I've been, you know, talking or requesting things or whatever has hasn't been as specific as I probably should have been. Right? That's a good thing to learn. Be specific in what you want, and that means we have to agree on definitions, right, like, to our previous conversation.

The enterprise world, I would start with, the low code, no code options because you just all you have to do is be able to think logically. You don't have to actually code for that much. Right? Like, or very little, usually none at all.

And that means you can kind of you'll start to see how complex some of these actual tasks are. That your brain, your your magical brain, which is better than any model, has already sort of automated itself. Right? And then when you start breaking down this is actually what they're doing in, education for for kids under about ten, under us, maybe eight or ten right now, is they're teaching them to break down problems like an engineer. It's actually great because they're gonna be problem solvers for the rest of their lives. Right?

And so you start with the low code, no code options. But you can actually ask Gemini, for example, to write the code for you if you do need the code. Right? And then if you do get into that world, if you make that leap from, you know, copying and pasting and minor editing into full blown coding, you can use code assist, and it'll literally suggest things for you. It'll it'll correct errors before you can get one.

My favorite feature for enterprise people is, like, whenever someone builds a piece of software, they inevitably have to pick a default language, not a coding language, but the language they're gonna communicate in, English, US English maybe or something like that. Right?

And, inevitably, if they do their job well, it'll expand to another country, language, market, whatever. Right? And they're gonna have to basically go through all of the code all over again and say, okay. How do you say that in Espanol?

How do you say that in, know, German? How do you say going on so on and so forth. And you basically have to not rewrite, but you have to definitely be thorough and go through every line to make sure that if someone who doesn't speak English gets is a user and they only speak German, will they be able to complete all the tasks that made that software successful? CodeAssist and AI is really good at being like, hey.

We noticed that you started to use natural language here. Do you wanna translate automatically to German? And sometimes when you're writing anything where it's a user prompt, like, hey, a notification or or, you know, so a button or whatever, it'll automatically kind of pick up and be like, do you wanna set this up for multiple languages now? You don't have to, like, translate it to German now, but we'll just say, hey.

When people do when you do enable German, the German all those all those options have already populated. And that that saves so much time. It's it's crazy. And I don't think most people think about that because they naturally and automatically read in their in their primary language or whatever's given to them that they're fluent in.

That can help a lot. And I I think that's why for example, at Google now, about thirty percent of our code is written by AI. But about one third. Right?

Yeah. And that's pretty pretty wild. And that's mostly the scaling up stuff because we do have to operate in a lot of countries and a lot of languages, and we just made that much much easier to deploy. So you you'll see faster rollouts to different to to to markets beyond the primary market.

Usually usually, we release things in the US first. You'll see the rollout is a shorter time span now to other languages in other countries, which it generally is good. It's one of our biggest complaints.

Yeah. That's super interesting. And controversial opinion on something that, you said before around, you know, low code and no code.

I have in my mind that the versions we see today, of AI are the absolute worst that we will ever see in our lifetime.

So from here on Aside from the previous ones from last year.

Yeah.

Yeah. I mean, the versions that we live today in the last four years being today would be would be a joke in, like, fifteen years down the line. Right?

I can't remember what people have.

Yeah. That's a good thing.

Do you think that this will also shift on how IT work in companies? So for example, because I have in my mind that five, ten years down the line, at some point, IT will be focusing on how to build tools for people to build the technical solutions. I believe it will be more of an educational journey rather than how we have been used to as, until today, right, that IT was building the technical solutions. I believe we will be on a stage where everyone will be able to use tools to kind of, like, cater to their personal needs at work and build whatever they need to automate. And the IT will be there to a little bit help and educate and build in the background the things that we use daily.

What's your opinion on that?

I I think that's probably a good bet. The question is when.

So yeah. If you remember if you're old enough to remember, when IT departments were relatively new.

Right, becoming their own department, they used to be, like, office managers. This is, like, in the sixties, I wanna say, maybe seventy. It was, like, an office manager, and that was the person that made sure that you got a typewriter.

Right? Like and the IT extension that is, like, make sure you get a laptop. It's like, okay. I got it. That's provisioning hardware, which I think will become less and less of the IT role, as you can imagine.

But more importantly is the merging of actually software development and enablement and IT. They'll they'll become unrecognizable at some point some point.

And I that's I I think that's a good thing because, you know, provisioning laptops, I've had that job in the past. It's not fun. It's needed, but not fun.

Providing solutions and making employees more effective, you know, I would say that's, you know, productivity related.

And we've always seen IT as a call center. I think it's actually gonna be a productivity center. It's what I was gonna say. Like, hey.

We we made a thing to do your job easier, rolling it out, providing the documentation, maybe writing the documentation, troubleshooting it. That's that's like ops more than it is just IT. Right? And I think they're actually gonna come together at some point.

But, yeah, I do think actually at some point people will start writing their own applications to do so, and I totally love this world because if you're if you have half a brain as a as a leader, you're gonna say, hey. I'm gonna give a x thousand dollar bonus to the person who automates it the best, and then we'll roll that out. Like, incentivize it.

What is it worth you?

Like, it's it's such a no brainer to me. And the but I'll tell you the best teams I've seen in in AI development, they don't they don't give bonuses for the team that has the best performing model. They don't do that. They give bonuses to the data engineering team that has the cleanest input data for people to build multiple models on.

Right? So if you do an audit and ninety nine percent of it is correct or right or, you know, not not error prone, then they get a bonus. That is how you actually set up an ecosystem that flourishes like and and I think a lot of people with a very centralized IT mindset might struggle with this world, but I generally think that, like, let let give people the the playground to play in and let them play whatever game they want. That's fine.

They seem to think that that's that means that there's no playground. I'm like, no. No. There's a there's a perimeter, and then, you know, you have to maybe it's a cloud environment, whatever.

But but let them play inside of that. You know, give them a wide space to put do do whatever they want in the context of what you want. Right? But you don't Yeah.

You don't know what they need to build. So, you need to let them like, the the number of times I've had to talk to people in in IT systems are like, how do you know that they need this? And they're like, oh, they need this. I'm gonna but how do you know they need this?

Like, did you do a survey? Did you do interviews? Did you well, like, what where did that come from? And it's like, it turns out that a lot of times it's just a personal opinion.

And maybe they're right, but it's a pretty low hit rate. Right?

So Exactly.

Let them let them build it, give them the tools, give them the platforms to do it, and let them go wild.

Yep. And, to this, I will just remind everyone, listening out there to go to our AI community and join the Genius Lab, so they can play around in our own playground and, follow what Ben is saying.

Now, Ben, how about organizations that are not so robust when it comes to their technical teams? Teams. How does Google Cloud lower the barrier to entry? Are there any specific features or programs that nontechnical users, have to help them to get started with, Google Cloud AI?

Yeah. Well, if they have a Google Cloud account, you know, probably provisioned through work, AI Studio, great place to start. You can just start tinkering. It'll help you from emails to create images, even, video, the VO models just recently, are being are being rolled out now, across most accounts.

We actually just announced, the new Gemini two today. So, you can play with those things in a sort of a sandboxy environment, see if you like it. I think that's a good thing to do. A lot of people get very freaked out.

They think, like, I'm gonna be forced to code at some point. And I'm like, no. You're not. And the reason is because I stopped coding a number of years ago, and I do more AI than I've ever done them in the in the past.

But two is increasingly, we are, automating the code writing process more than we're automating any other process. Right? And because and this it's really simple, you know, just a little bit of critical thinking will explain this, is that, what's the most computer friendly language? Is it English?

Is it Spanish? Is it Korean? Or is it Python? It's like, it's Python. Right? Like, it's it's unequivocally, it knows exactly how to read it.

It knows exactly how to write it, and it's got a ton of it available. So you will probably describe code in the future and have it be written for you, as opposed to having to learn this language. And I actually so I actually do think that, being someone who is logical and ambitious but not able to code is actually a really good place to be in the next few years.

Cool. Well, let's talk real world, a little bit and, share some stories. Tell me one thing that comes to mind, from your journey at Google Cloud AI.

Something that has been a success story that stick to your mind, something, that happened there that you were like, wow. This is something that transformed either an industry or a business or a person or some specific challenge. Anything that sticked to you.

Yeah. There was my favorite projects to work on are the ones that are, that require a lot of change management. And that's because, you know, you're not just building a better system and replacing it. You know, it's like replacing your you know, it's like getting a new car.

You sell your old one, you get a new one, you're like, oh, and then, like, you know, a number of months later, you're like, okay. It's just my car. Like, the luster is gone. Right?

Like like, those are good. They're upgrades. No question. My latest car is better than my previous one.

No question. The ones that are noticed where I really change someone's mind about how they work.

I worked on a project. I can't use names, but I worked on a project where, basically, we were trying to make, a is a a business to business transaction system where, historically, they would call up on the phone, they'd haggle and negotiate a price between two companies. Right? It's not not consumer at all.

They come to an agreement, and then the deal would go through. And let's be clear. Some some people were very good negotiators that people never wanted to talk to ever again. Some people were really bad negotiators that everyone wanted to talk to.

Right? Right? And there's some people who are in the middle who managed and and that's not a like, people think negotiations is, this you know, these numbers bouncing back and forth, and it's like, you're missing the point if that's between humans. Right?

There was this one guy I worked with. He did not like me because I was, you know, that was the the I was the Terminator to him. Right? And I was like, no.

No. No. No. I'm telling you I'm gonna make your job easier, and you're gonna focus on only the stuff you really like doing.

And I bet I was I told him, I was like, I bet you like it when it's a really complex negotiation with someone who's on your level. It's hard, but I bet you enjoy that. And he's like, yeah. Actually, I do.

And he's like, I'm like, what percentage of your phone calls that's literally how he was doing it. What percentage of your phone calls are that level? And he was like, ten percent. And I was like, okay.

So if I can make your job easy ninety percent easier and a hundred percent of the stuff you like doing, isn't that an upgrade? And he was like, I guess. And he's very skeptical.

And, so I've we built the thing. We trained it on all his on all his previous bit because he was the best guy at it. He was definitely the best, he was the best person at main keeping an account and negotiating good prices for said account. It's a balance, and he was very good at it.

Why? Because he'd been doing it for forty years. Right? Yeah. He wouldn't be there, you know, if you hadn't figured it out by then.

Right? I think that's probably true most of the time. And so I showed him how you know, here's how you build this you know, here here's how it works, and I showed him. You can see every negotiation and it happening, and you can shut off any of them at any time if you're like, no.

That's not right.

Like and he was, like, kinda he didn't even like using software. He was a phone guy. And I don't mean, like, cell phone. I mean, like, he liked he had the pad the the one who had the pad on the shoulder. Yeah.

It was worn down. I mean, it it looked like you know, it was a battle ax for him. And so I gave him a big red button that would shut off the algorithm if he didn't like what he saw. Right? And, literally, a USB red like, a comically sized button. Right?

But I get I gave him the button. I said, you hit that button, and that thing turns off for fifteen minutes. And you can call and do whatever you want and take over and fine. And he was like, great.

And I said, I'm never gonna be able to do all of this. I can do some I can do a lot of it. I could probably do most of it. And I could probably do most of it as good as you and maybe sometimes better, and I'll show you why it did it better than you would have.

So I would make a prediction on what he would offer as a price. I would negotiate the price and then show him what the difference was, positive or negative. And sometimes it was worse and sometimes it was better. But if you totaled it all up, it was slightly better, like, two percent better.

Over a year of hundreds of account managers doing this, it was worth millions of dollars. Right? So I was like, hey. Basically, we found millions of dollars.

And by the way, you can go to your boss, and I'll tell your boss if you want. That was you that did that, not because you wrote the code. You definitely didn't write the code, but because it was your historical bids that I trained it on. Right?

So you say this is an extension of you and those millions of dollars, you're worth millions of dollars to them. Right? Because if you leave the company, I don't have your your bids anymore, you know, going forward, which means it'll get worse over time or I need to and then how about, like, how expensive is it to replace this guy? It'll cost you potentially hundreds of thousands, maybe millions.

And he was like, oh, cool. All of a sudden, he was kind of on board.

And the best part was is that, like, you know, he he came around, and I, there was a point where I would and I would watch the times he would hit the button.

This one week where he hit the button every fifteen minutes. Like, he basically took over the whole week. He was willing to go back to his old job, and I was like, what? So I called him.

I was like, buddy, what's going on? And he goes, it was, it was there was a big, labor strike of some sort that was happening. And this guy was one of those guys that's on the emails. He's on the text message threads.

He's on the radio. He's on the phone. He's talk my algorithm does not have that chatter. Doesn't have any of that data.

There's no it might see the effects of a labor strike, but it does not know the cause.

He knew the cause.

He knew exactly what they were mad about. He knew exactly what was, like, you know, what was when they he knew when they were gonna strike within, like, a matter of hours. Right? Versus my my algorithm had nothing.

And that whole week, he just took over every time. And I said, do you want me to turn it off for a week? He was like, yeah. Yeah.

So he shut it off for a week, but I I told him, I was like, hey. If you want if you if you're worried about job security, I think you just prove what what you're worth is that you have a pulse on that labor market that no algorithm does. And believe me, that that week, I was like, you earned your entire salary that that year in that one week. And I was like, that's probably how it works is your expertise isn't always valuable.

It's very valuable some of the time, and it's definitely worth keeping you around. So and now now I talked to him every once in a while, and he's, he's someone I probably normally wouldn't be friends with, you know, in my day to day life. But, it's actually really the guy is fascinating. And that was actually a very memorable one for me because, I chain I made his job easier.

I made his value go up, and he still got to do all the stuff that he likes with his job. Job. And, you know, now he now he still calls me the Terminator as a joke, but, you know, he can do that. He's he's a he's a rough and tough guy, and I've got thick skin.

So Yeah.

Well, super interesting, especially the two parts around the challenges, you know, because people like challenges. When you mentioned that, you know, he likes the people that will challenge him on the phone. I think Yeah. We all deep inside of us, like, when we are in a situation that, you know, we're being a little bit challenged is the waters the unknown waters that we don't know so much because that's the point where we learn. Right?

Yeah. It does. It's more rewarding. Right.

Exactly. And, tell me a little bit, when it comes to the future, when it comes to trends and innovation that are coming in, in the industry or from Google Cloud. So there are a lot of trends. As you see, every day, something new pops up, a new product.

If you open any app, it's like all day AI, something new from someone. And, tell me what is it that, excites you the most?

Like, what do you see, a trend that is following that is starting or is Yeah.

Has started already or an innovation that excites you the most around AI? And, how do you guys at Google Cloud AI make sure to position yourself, to lead those advancements?

One of the best pieces of advice that I got from, like, a business kind of executive perspective was, in a in a in a time of rapid technological change, it's not really what you stand for. It's more what you definitely don't do.

And I think that's actually really good. So I always ask myself, what do you definitely not do? And the thing I'm I'll I'll connect these dots is that one of the one of the things I'm most excited about is about the explosion of creativity that is now gonna be available to more and more people. The barrier to entry to producing, you know, YouTube quality, maybe even Hollywood quality film, you know, video is getting closer and closer.

We've released some models that are doing that maybe six seconds, eight seconds, maybe up to twenty seconds, that sort of thing. But if you're doing a series of clips, that actually is working. So we're starting to see the the where anyone can just a scene, and they can build that scene without cameras, without really crazy animation software, without rendering, you know, like, that sort of stuff. It's like it's the the barrier we're we're looking at what Photoshop did for photo editing.

We're seeing it now in video, which I think is awesome. I don't mean just editing. I mean, like, creation, like, the production of video. I think it's gonna be awesome because that means that every kid, every student, every worker, grandma is gonna be able to say, hey.

I wanna see a, you know, monkey, you know, juggling some coconuts on top of the Empire State Building. It's like, okay. You got it. Ten seconds later, you've got it.

Like, if you can imagine it, now you're gonna be able to produce it. And we're not we're we're all like, we have, like, the, I'd say, the foundation bricks of that yet. We haven't we've got everything for the whole house. But we've definitely started pouring the foundations for this stuff.

And I think it's awesome because the people who are the most creative are gonna have an absolute renaissance with this.

I know.

I'm one of them. I have I'm such a I have such a big excitement for text to video. It's like Yeah. I believe the one type of AI that I'm losing my brains.

Not to mention that one of the others was notebook l m with the podcast feature. That was when I lost my mind. Because, I mean, Gemini is very, very good, you know, but it is something that we have from Gemini for a long time. You know?

I think it was apart from a chatbot, Notebook l m did something we have not seen before. It was something completely, new out of the box. It was something very, very different from what we're used to, for example, from Gemini or any other Yeah.

And it was the same base model. It was the same the engine of the car was the same, but they made a better car. You know? Like and they had figured out the interface.

They had figured out how to structure data, but it was the same engine. Right? And that's that's actually a very innovative mentality. It's, like, you know, being able to be given something and say, do something really wild with this, and everyone's given the same thing, and they all do something, you know, and and some people do some wild stuff with it.

That's that's that's innovative just as much as building a model is. Totally. I love I put for my class that I teach at, Columbia, I take every lecture and then I, do a, summary based and I just use notebook a lot to ask them. Like, I just I upload the transcript, and it just, like, summarizes, like, da da da.

And then it becomes, like, your class notes. And all my students say, like, that is better than taking notes. You know?

Now we're gonna take notes from you, Ben.

Since you teach, it's time to teach us. Sure.

I'm looking for some advice here. So let's start for any individual or startup or company listening listening out there, who are starting possibly their AI journey. And what advice would you give to them? How can they leverage Google Cloud tools to take their first steps, effectively?

Yeah.

First of all, I would at the risk of sounding edgy, I'd I would ignore the mainstream press. Not because they're wrong. It's because they're focused on the consumer level. Right?

They're they're looking for headlines and they're looking for, things that are happening that are to the average reader, not the average worker that's working with AI. Right? So my suggestion would be one is look at the enterprise world, which I'm not gonna lie, the documentation, the blog posts, they're a lot more boring than the than than the average journalist. Right?

That's true. But the intellectual capital in those in that stuff is much, much higher, like, exponentially higher. So I would one is I would say how are companies using this? What's the what's the, professional version of this, not the consumer version of this?

So start with that, and then you'll automatically be at a better place as a employee.

The other thing I do, and this is like a this is like a hack. If you if you're confused by the terminology, one, I would keep Gemini up on the side, but, like, what does that mean? And just it's it's it's the best dictionary you've ever had. You know, you can have follow-up questions.

Like, just be persistent. And I don't think everyone's got that in them. Right? And I think people that are persistent and creative are gonna have it's gonna be the best time of their lives.

Right? Like, in terms of information gathered and processed.

The other, like, the other kind of hacky thing, and this is actually what got me going when when the documentation was sparse and I wasn't I was asking questions that no one seemed to have good answers to or at least complete answers to. As I started buying, technical books on AI, large language models, all these things, and I would just read through them, and I would skip the code because I don't need the code anymore, to be honest. Right? But I wanna know how they think about it.

I wanna know how they what was the approach? Like, I wanna under tell me explain the system to me in natural language. I don't need to learn how to code. I don't need to demonstrate.

I'm not a practitioner, you know, for the most people aren't practitioners. They don't need to actually fine tune a large language model to do their you know? And if they do, they can use notebook l m or something that's relatively easy. Right?

But then you'll understand the concepts of how these work, and it means that you'll be able to be more innovative. Like, remember, the best painters in history were still classically trained. You know, they weren't creative geniuses that just picked up a paintbrush and just went wild. Everyone was like, oh my god.

World changing art. No. No. No. No. They all almost all of them were classically trained, and then they took it to the next level.

So if you wanna classically train yourself, which is not a thing in AI because it's not been it hasn't been around long enough, is read the books. I I think the O'Reilly books are notoriously technical books. But if you just go through them, read them, and and skip the coding parts, you'll understand the technology really well, and then you can start really doing innovative, clever stuff with it. Right?

The other thing I would say is if you don't wanna go to that level I'm an avid reader, but not everyone is.

Push push these technologies to the boundaries. And so you wanna go to NotebookLl and see how much it can handle. You wanna start asking crazy questions. Find out where its limits are, and then you'll start realizing, oh, I can be or give it unexpected things.

And that that way, you'll be like, hey. I found this really weird thing. It's actually really helpful. Like, one of my favorite things is the difference the difference between a study guide and a summary was very different from my class.

If I took the transcript of my an hour long lecture that I gave, and I said, hey. Give me a summary or give me a, give me a, study guide. The outputs were vastly different, and the topics it would the what it chose was vastly different. I was like, well, why?

So I started asking, why did you summarize this but not cover it in the study guide? Right? And it, like, had this very detailed response. So basically, it said, like, well, this didn't this seemed to be a tangent that you went off as an example, not really the point of the lecture, but it gave color to the point of the and I was like, that's actually correct.

That's actually that's actually really so the summary would say, hey. Ben went and told the story about the, you know, this guy with the big red button and dah dah dah. The big red button thing is not the takeaway from the lesson, but it's the example of it. But the study guide was like, you know, make sure that you handle stakeholders that are maybe not as fluent in technology or AI, with caution, which is is the you know, that's the that's the takeaway from that story.

It's not the story itself. So I I didn't I couldn't have articulated as well the difference between a study guide and a summary that well, and Gemini was infinitely better than me at that. And it was summarizing the and then and did the study guide of my own lecture. Right?

So I questioned. I started asking, like, why? Why? And I think it should be like a toddler.

Why? Why? Why? Like, I think I think I think if you after you ask five levels of why and you have good answers, then you know someone knows what they're talking about.

Exactly. And something super interesting was what you said around people are out there that are so inspired around AI, but sometimes they feel a little bit, left behind that it has moved so so fast Yeah. That, they cannot catch up. But if you break it down, really, you are able to explore it and start understanding slowly slowly in, how you can start playing around with AI in a way that's approachable.

So anyone out there Gemini is great.

But I'll tell you, if you feel like you are falling behind in AI, you're not by definition.

There's a good percentage of the population that isn't even aware of this stuff. Right? So if you're like, I feel like I'm falling behind, you're definitely not you're you you might be behind the researchers. Sure.

Ninety nine point nine percent of us are. Yeah. Right? Including myself. Right? Are you behind the developers of this technology?

Sure. That's less than one per one point one percent of the population. That's fine. Being the ninety ninth percentile is just fine.

Right? You know, if if there's probably a good half of the general population that, is not aware, is not having those thoughts, isn't concerned because I you can't be concerned if you're not aware. Right? So, yeah, I wouldn't feel too bad.

I think that's, I would just turn that anxiety into motivation and say, I'm gonna start playing with this stuff. I'm just gonna keep asking why. I'm gonna keep drilling on it because, honestly, I think it's gonna determine the next fifty years of society.

In one way or another, I can't even tell you exactly, but there's no this isn't a fad. It's clearly becoming embedded in our organizations.

It's becoming embedded. Governments are using it. You know, it's and mostly for good, but there are gonna be there are risks. But I think they're mostly mitigated. They tend to have little blips of problems, but not systemic level problems, which is good. It means we're sort of setting up for, you know, a good a good future.

But, yeah, I think I think we're past the point of saying this is a fad. It's it's getting embedded in every organization. It's not a talking point anymore. It'll evolve, definitely. And to your point earlier, is that, you know, we look back at the models of this year and then in twenty twenty seven, we're like, oh, god. That was so bad. That's a good thing.

I know.

Because it means you made progress, and that's one of the things that as a product manager I learned. If you don't look back and cringe, you have failed. Right? Like, you have not you've not made no improvements, and you're not perfect.

So don't tell me that. You know? So I think that's a good thing. But, yeah, if you're worried about it, it means you probably shouldn't be, and you've just got the potential for some some more energy and motivation.

But, yeah, I wouldn't I wouldn't sweat that too much. I think it's a it's a red herring.

K.

Well, thank you so much, for the advice, Ben. Really appreciate it. And that was it. We're almost over.

The last pinpoint that I will make before we close is that the armchair you have in the background looks super, super comfortable. I had to say that from the beginning.

Yeah. This one thing.

It looks super, super comfortable.

That's why I do my reading.

Thank you so much, man, for joining us. I really appreciate you taking the time, to join us and, you know, share a little bit of, your insights from Google Cloud AI. And, thank you for sharing your knowledge and, or your little tips and tricks, around the area.

So and thanks for everyone listening out there. Really appreciate you and continue being, curious. And see you on the next one.