What you need to know: The 10X Engineer, AI's Impact on Business Ops & Customer Service with Aaron Lee, CEO & Co-Founder Smith.ai

[00:00:00] I think that is a general kind of like feelings when something new that comes that people say,

[00:00:07] Whoa, am I am I going to be irrelevant? Am I going to be obsolete? Am I going to be like

[00:00:14] losing my job and losing my values? I think that is what people mostly afraid of.

[00:00:19] All right, I want to talk to you for a moment about retaining and developing your workforce.

[00:00:24] It's hard recruiting is hard retaining top employees is hard.

[00:00:28] Then you've got onboarding, payroll, benefits, time and labor management.

[00:00:32] You need to take care of your workforce and you can only do this successfully if you commit to transforming your employee experience.

[00:00:40] This is where I solve comes in. They empower you to be successful.

[00:00:44] We've seen it with a number of companies that we've worked with.

[00:00:47] And this is why we partner with them here at WRKdefined.

[00:00:51] We trust them and you should too. Check them out at I solved hcm.com.

[00:00:57] This episode is brought to you by Shopify.

[00:01:00] Forget the frustration of picking commerce platforms when you switch your business to Shopify,

[00:01:05] the global commerce platform that supercharges your selling wherever you sell.

[00:01:10] With Shopify, you'll harness the same intuitive features, trusted apps and powerful analytics used by the world's leading brands.

[00:01:17] Sign up today for your $1 per month trial period at Shopify.com slash tech all lowercase.

[00:01:23] That's Shopify.com slash tech.

[00:01:32] Ladies and gentlemen, welcome back to another episode of the you should know podcast.

[00:01:37] My name is Ryan Leary. I am here with of course William Tin Cup.

[00:01:42] And today's guest we have Aaron on from Smith AI who I believe is going to be just a really good guest.

[00:01:52] We've got really interesting topics to talk about today.

[00:01:55] We're going to talk about Fobo, which William you introduced me to a couple weeks ago the fear of being obsolete.

[00:02:04] It's a real fear.

[00:02:05] I'm not sure I have a fear of being obsolete. Who would want to get rid of me?

[00:02:10] I mean, look at me. If you're if you're watching me, look at this face. It's gorgeous.

[00:02:15] Aaron, nice to have you here. Nice. Nice to have you on with the audience.

[00:02:21] Why don't you kick us off, introduce yourself and give the audience a little taste about who you are.

[00:02:27] And we'll get this rolling.

[00:02:29] Yeah, awesome. Thank you so much. So much excited to be here.

[00:02:32] And let me talk a little bit about myself how I ended up with a lot. Did you hear that?

[00:02:38] Yeah.

[00:02:41] I think it's going to be so cool.

[00:02:43] So yeah, I was an engineer by training. I joined the Google back in like 2004 as one of the first engineer on building Google video from the ground up.

[00:02:54] And so I work on a lot of search advertising, storage networking and 2006, like Google acquired YouTube.

[00:03:03] I joined the YouTube team. I built up there monetization.

[00:03:06] I work on the Tysation. I work on the machine learning, the framework on how to target the customers.

[00:03:11] Like kind of like, like these are the videos that you would like to see, right? Because you saw the other videos or these are the things and topics that you would maybe.

[00:03:19] The recommendation engine behind that.

[00:03:21] The recommendation engine and some of these like can movie recommendation as well.

[00:03:26] Oh, that's cool.

[00:03:27] Early stage of AI like back in the day, back in the days we called machine learning. We call it like user recommendation system.

[00:03:34] And I left in 2008, like because like Google went from 2000 people when I joined to 20,000 people.

[00:03:44] Amazing growth. Like, I mean, really, really great.

[00:03:51] And I said, well, I mean, like I learned something from Google and very entrepreneurial and started my first company called red beacon, which is a marketplace to connect the home owners with the professionals.

[00:04:04] Right. Basically, it's like you go to the platform and say, Hey, I need someone to help me fix this.

[00:04:08] And we basically like blast out the best signal and say, Hey, with a customer who are interested in this kind of project.

[00:04:15] If you're interested, like put in your, your proposal and then we'll connect you guys. Right. It's a much better way. Like that was back in 2008.

[00:04:25] It's really democratized like the two-sided marketplaces pieces. And we are funny thing is when we started the company that was the financial crisis in 2008.

[00:04:37] And all the VC firms is like, Hey, there's no money. We're not going to write any checks to any company.

[00:04:42] And it turns out it was a blessing because we were has done building the company by the time we launched in 2009.

[00:04:49] There were no competitors because no one got funded in the past year.

[00:04:53] Right. Right. And we took the top price of tech crunch and we launched nationwide.

[00:04:58] We closed one round, the first round of financing from Mayfield and Ramrock.

[00:05:02] And we just go, we just went nationwide. And within a couple of years, like Home Depot saw that and say, wait a minute, this is really cool because Home Depot has a brand.

[00:05:13] They have over 2000 stores in US. They have both sides of the marketplace. They walk into the store. Right.

[00:05:20] Just in different aisles. Like they got the demand side. They said, like, we want this because one of the things that they worry about is like Amazon is going to be like,

[00:05:31] Amazon is going to eat their lunch. Yes. And, and so the toys are, toys are us ignored Home Depot. Exactly. Exactly.

[00:05:40] Can't buy plywood through Amazon. Yeah. I mean, you got to think about that.

[00:05:46] It's like, why don't we do that? Why don't we build the products and the services?

[00:05:51] Yeah, smart.

[00:05:52] I go into the store and I think the millennial and some of the Gen Z, they say, I have no time. Right.

[00:05:58] I don't want to be on my, my, my own deck or my porch or my hook shelf and that was gone in the days.

[00:06:04] Like back then you guys probably remember you will go to Home Depot every weekend. You do some kind of project.

[00:06:10] Oh yeah. I still do.

[00:06:12] Yeah.

[00:06:13] Nowadays, I'm more of a millennial in this regard.

[00:06:19] I don't own a garage. I don't have a garage. That's right. That's right. Yeah.

[00:06:24] Because I don't want any tools. I don't want to, I don't want to have any tools, but I don't want to have any responsibilities.

[00:06:29] Yeah.

[00:06:30] I feel like this is coming full circle. So you go to Home Depot, you go 20 times a weekend to finish a project while you're watching the video and the recommended videos on YouTube that you created.

[00:06:44] You had a hand in all of that.

[00:06:46] Yeah. You just say like, wait a minute. I just need someone that I can trust and do the work. I pick up the material, pick up the countertop or cabinets and it'd be nice if I can do that right there and I can trust Home Depot because they have the brand, they have the guarantee, they have all these stuff.

[00:07:04] So that become part of the very natural ecosystem for Home Depot.

[00:07:09] Did you do a partnership at first?

[00:07:11] Did you do Aaron, did you do a partnership at first?

[00:07:14] No, I actually did. Today Rush, they go straight to acquisition.

[00:07:17] You're straight to acquisition because they see that as a huge opportunity. I think by the time I left in...

[00:07:22] Very smart.

[00:07:23] 2015, I stay on for a few years like really beyond the program. If you go to any of the Home Depot store today, you will see my work there and it was a multi-billion dollar business because it just makes sense.

[00:07:35] Right.

[00:07:36] Like on every level.

[00:07:37] Mayfield loves you. It's what I just heard.

[00:07:40] Yeah.

[00:07:41] And because of that, I got to work with the pros almost on a daily basis. I hear the ping points, they always say, look, I wish Home Depot can answer their calls because I'm just too busy.

[00:07:53] I'm always driving on the road at the project up at the roof down at the like whatever cross space that they're in and doing the projects.

[00:08:01] I have no time.

[00:08:02] And when they miss the call, the opportunity is huge.

[00:08:06] Right?

[00:08:07] And when they were calling in, wanted to do a kitchen remodeling for $10,000 and just because you couldn't catch the customer at that moment, they lost it.

[00:08:17] And that was like kind of motivating me on starting the company that was back in 2015.

[00:08:24] And that was really exciting because actually in the beginning we thought AI is going to be a lot faster.

[00:08:32] Interestingly, 2015 was also the same year that Open AI started.

[00:08:37] Right.

[00:08:38] Right.

[00:08:39] Like we everyone thought, oh, the tech not we're all optimists. We're all technologies with all it just take a few years and I want to be there when they boy it took what six, seven years.

[00:08:48] Oh yeah.

[00:08:49] And that was to get to the first chat GPT. And a year later we have GPT for we have the turbo with all these stuff, but that was really the foundational kinds of the net.

[00:09:00] The beginning of the company like there's got to be a way where AI and human can work together to deliver a much better and much higher services at a very affordable cost that any small businesses or meeting business can use us.

[00:09:18] We actually have people that run a small business like one man ban, like kind of spend a few hundred dollars with us handling one to two cost per day all the way to nationwide companies spending like half a million dollars with us per year because

[00:09:36] we are so good to the point that we actually replace the in-house call center.

[00:09:42] Right.

[00:09:43] We are 24 seven. We only channel we deploy AI with our agents so we get that kind of like emotional handling plus the accuracy because if you think about all this business they all have their own business logic they have own kind of like way of running the business

[00:10:02] and people like to handle customers within 10 miles because they don't like to drive. Some people say you know what I'm hungry I want to grow. I will be happy to drive for an hour if that means it's a bigger job.

[00:10:14] So people every business is a little bit unique and different. And we really need to have that AI as a co-pilot to live and help our agents.

[00:10:24] So another thing I want to share is we monitor every single like mistake, complain, concern from our customers and our correctness is 99.7%. We almost don't make mistakes.

[00:10:40] Wow.

[00:10:41] If you I mean we're all human I mean all the time you have a good day of a rough day you have a bad day.

[00:10:48] Sure.

[00:10:49] Then if you don't have the technology to guide them they will make human mistakes understandably.

[00:10:54] 100%

[00:10:55] But if you have the technology combined with the human which is a very friendly voice then you can do a much better way to save them the money, save them the time and have a much better services.

[00:11:08] One of the things I like about what you've built outside of all the things you said is you'd use their tools.

[00:11:15] Yeah.

[00:11:16] So as a business owner you've got some type of tech stack of things that you use and the fact that you don't make somebody go make your customers go to a unique place.

[00:11:27] So if you have a Smith.ai and then go learn about log in and do this that and the other it's like no if their stack is is Zapier an active campaign and Salesforce.

[00:11:40] Great.

[00:11:41] You know, you can you can do the communications layer which is so important.

[00:11:45] I think you you spoke about a little bit during pre show but because people's patients maybe even attention span.

[00:11:54] And expectations have shifted to seconds.

[00:11:59] And minutes and so there's a road I like like I really like the story of if you don't answer the call, you're not going to get that business.

[00:12:08] They're not gonna call back they're gonna leave a voicemail.

[00:12:10] It's just common as an opportunity lost.

[00:12:13] So take all those opportunities lost and you can calculate.

[00:12:16] You might never want every one of those deals.

[00:12:19] But you definitely lost all of the deals that you missed the calls on.

[00:12:23] Exactly.

[00:12:24] We have many examples where the business only using us they're not doing like any type of other like marketing or spending money.

[00:12:34] They grew the company like triple the company in five years just by responding to people by getting back.

[00:12:41] It's really basic and we think about it like what is people always ask like what is your source.

[00:12:46] I think there's no secret.

[00:12:48] You got by this poster or I'll buy it for you and ship it to you but there's despair.com.

[00:12:55] I don't know if you've ever ever checked out their stuff.

[00:12:57] It's it's the anti successories.

[00:13:00] You know the successories posters champions, you know whatever the bit is they've got one of them at despair.com where it's a phone.

[00:13:08] It's a old fashioned phone with cobwebs.

[00:13:10] And it says customers.

[00:13:12] If we stop answering maybe they'll stop calling.

[00:13:18] That's perfect for your wall.

[00:13:20] And that's such a great you've got to get it.

[00:13:24] They've got a coffee cups and all that stuff.

[00:13:26] But yes, I love it.

[00:13:28] It fits you because you know the whole bit is if you if you answer the answer again if if someone has been responding to it's not even if you if someone's need has been met about the responsiveness.

[00:13:42] I have a question about this.

[00:13:45] And can you respond in whatever way that they want meeting them wherever they are maybe it's Twitter or whatever, wherever they are and respond to them.

[00:13:54] That that that already starts that relationship off on the right foot.

[00:13:58] Yeah.

[00:13:59] So, Aaron, let's let's talk about FOMO.

[00:14:03] Okay, so we talked about FOMO fear of being obsolete when I say the word FOMO.

[00:14:08] What goes through your mind?

[00:14:11] I think that is a general, like kind of like feelings when something new that comes to people say, whoa.

[00:14:18] Am I am I going to be irrelevant?

[00:14:21] Am I going to be obsolete?

[00:14:22] Am I going to be like losing my job and losing my values?

[00:14:27] I think that is what people mostly afraid of.

[00:14:29] And I would always go back to say we always kind of overestimate what technology can do in the short term and underestimate what they can do in the long term.

[00:14:39] I think that was what Bill Gates said.

[00:14:41] Yeah.

[00:14:42] I think if you look at the near term, right on mid to near term, AI is incredible, right?

[00:14:49] The technology and the way that they can do it.

[00:14:53] Right.

[00:14:54] If you look at on the accuracy, if you look at the latency, you look at like sometimes they just make things up.

[00:15:00] It's very dangerous.

[00:15:02] Or he spelled the word burrito.

[00:15:04] He spelled the word burrito.

[00:15:07] And so I mean, like, yes, it can help us to do some of the mundane tasks.

[00:15:12] Right.

[00:15:13] But you always need some kind of like combination between the human and the AI.

[00:15:17] So we have seen, by the way, I keep track of the technologies.

[00:15:21] I'm a, as I said, I'm engineered by training.

[00:15:23] I love the new tech.

[00:15:25] I look at all these startups that come up with the demos and say, hey, we can do this voice voice AI.

[00:15:30] Yeah.

[00:15:31] And I always like try to do it and try to like kind of giving them a little bit of hot time.

[00:15:37] In the first few sentences, they sound very nice and like they can say things that are good.

[00:15:42] After a while they lose the context.

[00:15:44] They just like, they drift away.

[00:15:47] And I mean, they can't handle sarcasm.

[00:15:50] They can handle cursing.

[00:15:52] They can't handle all those, those types of things.

[00:15:55] And they can't handle anything dark humor.

[00:15:58] They can't handle that.

[00:15:59] So if you say something dark, they don't understand it.

[00:16:03] So it's still like that smart 10 year old we talked about.

[00:16:08] Okay.

[00:16:09] It's a really smart 10 year old.

[00:16:10] It's very smart.

[00:16:11] Like they are smart things.

[00:16:12] They read all the books on the whole.

[00:16:16] It's very smart, but it's still a 10 year old.

[00:16:18] So it's artificial, but it's not quite intelligent.

[00:16:23] Right.

[00:16:24] I think that is what like you're seeing now.

[00:16:26] Of course, like if you look at the rate of change,

[00:16:29] Oh yeah.

[00:16:30] Look at NVIDIA they just had a conference.

[00:16:34] Like it's mind blowing.

[00:16:35] They are really hitting that limit of physics.

[00:16:38] Right?

[00:16:39] Oh yeah.

[00:16:40] And processors and parallel like quantum computing coming to a theater near us.

[00:16:45] So right.

[00:16:46] The raw computing power is just astonishing.

[00:16:49] So I think something he said during that thing I want to get your take on is he said,

[00:16:55] I mean, he said a lot of, I don't say controversial,

[00:16:58] but I think they're really cool kind of forward thinking things.

[00:17:01] He said, yeah, you don't need to learn coding anymore.

[00:17:04] So if you're a parent and you've been sitting your kid to coding camps and all that stuff.

[00:17:09] Yes, stop that.

[00:17:11] They don't need to do that.

[00:17:13] AI can do all that for you.

[00:17:15] You just got to learn how to direct AI to code for you because you don't have to learn the code anymore or learn how to code anymore.

[00:17:24] And when I first heard it, I had to go back and look and listen to it again.

[00:17:28] Like was he joking?

[00:17:30] Like was this a bit?

[00:17:32] And it wasn't a bit.

[00:17:33] He really believes that whether now or in the future, you still need to learn code.

[00:17:41] Which is crazy to me to think of because there's so many people that have built their lives through, you know, being great software engineers.

[00:17:50] So he writes software architects and actually learning code and learning how to do things like.

[00:17:55] So what's your, you know, as he said it.

[00:18:00] And again, I went back and watched it a couple of times to make sure when the comedic bit, he was serious.

[00:18:06] Yeah.

[00:18:07] Oh, I get it.

[00:18:08] I think like he's serious in terms of like if you think about.

[00:18:11] If everyone is doing that who's going to build the next version of the Nvidia chipset who's going to build the next version of the.

[00:18:18] Like if we talk about the AGI, right?

[00:18:21] I mean, and will we actually do the point that they can sell self iterate.

[00:18:27] They can say, hey, the AI is going to develop the next version of the chipset to make them smarter.

[00:18:34] Right.

[00:18:35] Would it happen?

[00:18:36] Maybe maybe not.

[00:18:37] But if you think about all the I call the typical programming tasks, right.

[00:18:43] And GitHub has to co pilot that would come in and say, Hey, you know, William.

[00:18:47] This part of the code is probably not going to work.

[00:18:50] Right.

[00:18:51] Let me rewrite the code and take a look.

[00:18:53] Right.

[00:18:54] I'm going to write the test cases for you to make sure that like my call is actually working.

[00:18:58] I like that because the smaller the code, the, you know, basically you could go into some fat code and then shrink it down.

[00:19:05] Still does the same thing, but it makes it faster.

[00:19:08] Yeah.

[00:19:09] I like that.

[00:19:10] Nothing happening.

[00:19:12] Right.

[00:19:13] I think like people can detect they can upgrade.

[00:19:15] Right.

[00:19:16] Like level like basically identify the bugs before you as a developer identified, you can ask to get up to say, Hey, this is what I would like to kind of the skeleton code.

[00:19:25] I would like to talk to active campaign.

[00:19:27] Can you use the API?

[00:19:28] They will spit out the scaffolding right for you and then you just like mixed the business logic.

[00:19:34] You can kind of change.

[00:19:35] I think you will see the AI that can do let's say today 50% of the work that would get to 70, 70 or 80%.

[00:19:44] And eventually there's still going to be a human.

[00:19:48] The human is still going to be like handling the last 10% or 20%.

[00:19:52] But suddenly you have this like in Silicon Valley, we call it 10 X engineer, right?

[00:19:57] Right.

[00:19:58] Someone who can be so productive that it's like if you hire that one person is equivalent to like 10 other people.

[00:20:04] Right.

[00:20:05] And that is certainly true even back in the early days of Google initial search team like the Jeff Dean and a few other folks.

[00:20:13] There's like less than 10 people.

[00:20:15] They wrote the search engine.

[00:20:17] Right.

[00:20:18] They are there 100 X engineer.

[00:20:19] I mean, they're not mixed.

[00:20:21] Right.

[00:20:22] That was so amazing.

[00:20:23] But imagine if nowadays a typical engineer can have that power.

[00:20:30] They can become 5x.

[00:20:32] They can become 10x.

[00:20:33] I think that is definitely what we're going to see.

[00:20:36] But I always tell the question like you've got to know what questions to ask.

[00:20:41] Yeah.

[00:20:42] Right.

[00:20:43] If you don't even know what questions to ask, there's no smart AI that can help you.

[00:20:48] But in order to know what questions to ask, you need to understand like the logic behind it.

[00:20:54] You need to understand how the system works.

[00:20:56] You actually need to understand like how why AI hallucinate because they are a gigantic next character predictor.

[00:21:05] So for the audience that doesn't understand hallucinations outside of the other references around hallucinations, what do you mean?

[00:21:13] What does that mean for the audience?

[00:21:15] Yeah.

[00:21:16] It means like when you ask the AI to do a task, they may come back with an answer which is completely irrelevant.

[00:21:23] They may make up things that are completely not true.

[00:21:27] I mean sometimes they even do that.

[00:21:29] Sounds like me.

[00:21:30] They do the math wrong.

[00:21:31] Right?

[00:21:32] I mean depending on how many...

[00:21:34] It's not very intelligent because they're not.

[00:21:38] They're just predicting the next one.

[00:21:40] Is that a context thing?

[00:21:42] I think there's a lot of things.

[00:21:44] It's the context window.

[00:21:45] It's like how much knowledge that you have is like how you predict the next word or the next sentence.

[00:21:51] Now having said that, there are many, many research going on that augment the AI with the knowledge base.

[00:21:59] Right.

[00:22:00] See them and say, hey, we know...

[00:22:02] Did you mean this or this?

[00:22:04] Yes.

[00:22:05] Yeah.

[00:22:06] It should be these and not that.

[00:22:07] That's what...

[00:22:08] Like also making a mistake.

[00:22:09] Yeah.

[00:22:10] Go ahead.

[00:22:11] No, no, go ahead, Ryan.

[00:22:13] So I was going to go to an education question here.

[00:22:19] When does higher education or even high school, let's say high school, right?

[00:22:26] When does education start to move away from textbooks and history and into how to work with

[00:22:36] and how to prompt and how to work with AI, how to be a developer leveraging AI?

[00:22:42] When does it just shift?

[00:22:43] When do we see that?

[00:22:45] I think you're seeing a little bit of divide.

[00:22:47] Right?

[00:22:48] I mean some schools or high school or like middle school or even like colleges, they're debating.

[00:22:54] Yeah.

[00:22:55] We allow the students to use AI.

[00:22:58] Now, if you ask the students like to write a passage or write an article and they just

[00:23:04] copy and paste.

[00:23:05] Yeah.

[00:23:06] Whatever.

[00:23:07] AI then did not run.

[00:23:08] That is like the opposite of learning.

[00:23:10] It's just like copy and paste.

[00:23:12] Well, you're learning how to cheat.

[00:23:14] Yeah.

[00:23:15] Which is a form of learning but...

[00:23:17] It's like a negative form of learning.

[00:23:19] Like it's not a positive thing.

[00:23:21] Right?

[00:23:22] But if you say, you know what?

[00:23:24] I'm really stuck because I don't understand my geometry teacher when they talk about this

[00:23:30] like fear or whatever.

[00:23:32] Can you ask AI to give me some examples?

[00:23:35] Can you elaborate?

[00:23:36] Can you tell me a little bit?

[00:23:38] I think that would be super helpful.

[00:23:40] So again, it really depends on how you leverage the system but always remember the AI may

[00:23:47] be wrong.

[00:23:48] So you need to check it.

[00:23:49] You need to make sure that oh, is that what the formula is?

[00:23:51] Now, most of the times they are right.

[00:23:54] Especially in simple tasks.

[00:23:56] Simple tasks.

[00:23:57] And of course there are also research going into like can we break down the problem so

[00:24:02] that instead of solving the one problem in one prompt you go into multiple phases and

[00:24:08] get the AI to explain themselves.

[00:24:11] Why do you say that?

[00:24:13] And I think we would be more...

[00:24:15] And even do references and things like that so you can go back and see their logic

[00:24:18] tree and also where they got that information.

[00:24:22] I'm thinking about Wikipedia a lot when I think about this stuff because like when you

[00:24:27] search Wikipedia and on something generic it will then say the disambiguation that

[00:24:33] will say do you mean this, this or this?

[00:24:37] So it gets you then to go down the right path but it's prompting you.

[00:24:42] If you want to talk about this, like you can type in model in Wikipedia and it will

[00:24:48] then push you into do you mean fashion model?

[00:24:51] Do you mean modeling as in terms of building models?

[00:24:55] Do you mean this?

[00:24:57] Like I think that I think I call it wayfinding in my brain because I think

[00:25:02] AI will get a layer there that will help people then get to the thing so it won't

[00:25:08] be...the hallucinations will be less but also the accuracy of what they're

[00:25:13] really trying to search for.

[00:25:15] The answer that they're trying to search for I think it would be easier for them

[00:25:19] and that's just interesting.

[00:25:21] This is actually a startup called Diffbot and what they do is they index and scan

[00:25:26] the entire website, collect all this information and turn it into structure

[00:25:31] information.

[00:25:32] They call it the knowledge graph.

[00:25:34] That means when you ask something they can say oh the reason I said that

[00:25:37] because I'm getting it from this page and because this page is linked to

[00:25:41] that page and because they're talking about these and they can give you references.

[00:25:46] It's almost like when you go to academics, right?

[00:25:49] When you write a paper, you say the reason I said this because I am getting

[00:25:53] it from this reference and so it's almost like you can explain how you

[00:25:58] derive the answer as opposed to you just make something up.

[00:26:01] Oh I like that.

[00:26:02] So okay so now that you bring that I want to ask this because I thought

[00:26:06] about this.

[00:26:07] I'm writing a research paper and I've got a site, my sources.

[00:26:12] So I went to the Encyclopedia if you guys remember that.

[00:26:15] Or here I went there.

[00:26:19] Why is that fact in truth but not the stuff AI is finding?

[00:26:25] So how do I know that the Encyclopedia is actually true just because

[00:26:28] somebody wrote it?

[00:26:30] Well because if you go back in the early days like they actually have

[00:26:34] a very rigorous system.

[00:26:36] So it's not anyone that can come in and tell you the world is flat.

[00:26:41] Right?

[00:26:42] And it comparing that to where we're at now.

[00:26:45] If everyone said the world is flat, the AI will say the world is flat.

[00:26:50] Yeah I just know that.

[00:26:51] They have no right or wrong.

[00:26:53] They just say what is the next word that has the highest probability

[00:26:58] which is flat and not round.

[00:27:01] So there we are flat now.

[00:27:05] Do you believe that there's a Moore's law kind of approach to AI?

[00:27:14] Like do you think it's going to happen like Moore's law plays out?

[00:27:19] I think it's more than Moore's law.

[00:27:21] If you look at the rate of change right, they say every year you're

[00:27:25] double in like whatever the capacitor is and processing speed.

[00:27:29] We are seeing a step change.

[00:27:31] We're still in the early inning.

[00:27:34] Just why a lot of the development feels like wow this is amazing.

[00:27:38] We got 5x, 10x and not just 2x.

[00:27:41] Now at some point we will be approaching the physics limit.

[00:27:45] We'll be approaching some of the computation limit and people say

[00:27:49] Did you care change time?

[00:27:51] Right you can't change time.

[00:27:53] Actually when I was working at a Google video we talked about

[00:27:58] how can we make the video to serve faster and we hit the limit

[00:28:03] called the speed of light matter.

[00:28:06] If you have to stream a bit from data center in US to Japan

[00:28:15] the speed of light matters.

[00:28:18] Even if you put down all the optical markers and still can't go faster

[00:28:23] than light, sorry.

[00:28:24] You can go faster than speed of light.

[00:28:26] Wow that's basic physics.

[00:28:29] That's so great.

[00:28:30] And here I thought I was just typing in how to do something.

[00:28:34] Well you were.

[00:28:36] I love the idea of like what made a source a source.

[00:28:41] And again we're looking at AI at the early point.

[00:28:45] And Brian sometimes with guests we talked about where we are with AI

[00:28:49] and the similarities of where we were with the internet like in 96.

[00:28:54] You know people were on the internet or at least a form of the internet

[00:28:59] and there was all this stuff of hype of what the internet would or wouldn't do.

[00:29:06] And it seems that the hype with AI is still at least in our market

[00:29:13] in our industry which deals with work.

[00:29:17] The hype is still way, way outpaced kind of a reality.

[00:29:23] But the reality is going to catch up and it's going to do different things

[00:29:26] than what we predicted and what's been predicted for us.

[00:29:30] So the way I think about it as long as you have a little bit of human oversight

[00:29:35] that you don't blindly trust what AI tells you to do

[00:29:40] then I think you will see the productivity gain.

[00:29:43] You will see people actually more efficient, more productive.

[00:29:47] For example like you can ask AI to proofread your content

[00:29:52] and they can say yep these are the four typos, these are the grammatical mistakes.

[00:29:57] And by the way we use AI to generate the call summaries so that we don't have typos.

[00:30:02] Those are deterministic.

[00:30:04] You can trust it.

[00:30:06] Of course we have the agent to review the summary but it's a lot faster.

[00:30:10] Yeah.

[00:30:11] And we like...

[00:30:12] It's not creating the summary.

[00:30:13] It's not creating it, you just say hey does it look good to you?

[00:30:16] Oh, like a few things and you're good.

[00:30:19] Same goes for the higher education right?

[00:30:21] The AI can help you understand the topic, help you formulate kind of like the framework

[00:30:26] but at the end of the day you still need to proofread it.

[00:30:29] You still need to look at it by the content and say is it actually making sense?

[00:30:35] I think the correctness and the latency is going to be the two biggest bottlenecks

[00:30:40] that people can say you know what?

[00:30:43] We can trust the AI.

[00:30:45] Unfortunately, I think even at 90% accuracy it's not good enough.

[00:30:49] Even 95% is not good enough.

[00:30:52] That means you're making five mistakes for every 100 tasks.

[00:30:57] Yeah.

[00:30:58] And those tasks could be huge.

[00:31:00] I mean those could be huge things too.

[00:31:02] Not like minor, you put a comma in the wrong place.

[00:31:05] That could be massive.

[00:31:06] I mean healthcare, FinTech on all these highly regulated, especially healthcare,

[00:31:12] it's like life or death right?

[00:31:14] You can help the radiologists to give them the idea but you still need...

[00:31:19] Maybe it doesn't take them like a whole hour to look at the scan.

[00:31:23] Now it's like 10 minutes but you still need to kind of like does it make sense?

[00:31:28] How the data can corroborate the outcome or the result or the analysis?

[00:31:33] I think that is what you're seeing.

[00:31:35] I don't think people are getting obsolete, I think people are getting more efficient.

[00:31:39] So that means for the next generation if you want to be really kind of take advantage of it

[00:31:45] you need to say what are the values that I can add?

[00:31:48] How can I ask the AI what are the smart questions that you can ask?

[00:31:54] You either of you watch Chicago Mid?

[00:31:57] No.

[00:31:58] No, no.

[00:31:59] All right, so I can't even ask the question.

[00:32:04] So they have in the hospital they call 2.0 which is the surgical room, operating ER room.

[00:32:11] But it's all artificial intelligence.

[00:32:14] And so in diagnosis it tells you exactly...

[00:32:17] It tells a surgeon here's what it is, here's where you're going to go,

[00:32:20] here's how you're going to fix it but it has to be monitored by a human obviously.

[00:32:25] And then the surgeon interacts with it.

[00:32:28] So I guess the question is really interesting as you watch it but the question is how far away are we from that being in a hospital, in an operating room?

[00:32:41] I think we've seen robotic surgery.

[00:32:44] I mean we've seen kind of like this like robots try to do because they're very precise and they know exactly what they need to do.

[00:32:52] And even for a very experienced like human surgeon sometimes like your eyesight, your hands,

[00:32:58] you have to down to like sub millimeter precision.

[00:33:01] That is what they're really good at.

[00:33:04] I actually don't know how far we are from there.

[00:33:07] I mean we could be very close where they will come in, they will look at all the biometrics,

[00:33:14] look at your BP, your heart rate, like do this CAT scan.

[00:33:18] They can collect multiple informations and then they can do the diagnosis.

[00:33:23] I don't know if you watched there was an old show on TV called House.

[00:33:27] Yeah of course.

[00:33:29] And he's the guy who understands so many things like he was with your AI.

[00:33:37] What if you connect all the dots and you understand so much and now you're saying okay now we're getting more of these kind of data points because AI they don't sleep, they don't get tired.

[00:33:49] No they don't form unions, they don't have to.

[00:33:55] Turns out they don't need breaks, you know all that stuff.

[00:33:59] It's funny as you were talking, Ryan and I did a podcast with a guy in New York that has bots and named bots and he sells them as employees or like employees.

[00:34:13] But it's just like they all they do is do actuarial tables and find errors and actuarial tables and then alert the human being.

[00:34:24] Yeah here's where the error is.

[00:34:26] So they don't fix it because the companies don't want them to fix it.

[00:34:31] They want to find it and then let somebody then go and fix it themselves.

[00:34:35] And what's fascinating about that is he packages them like Sally, Jenny, Ted, whatever.

[00:34:43] And you rent it, I mean it's something that you have.

[00:34:48] Yeah it's a staffing play.

[00:34:50] That's exactly right, Ryan. It's a staffing play but it's a staffing of these bots or AI that go through and do a very specific task.

[00:35:01] And I love that. First of all the precision of going through an actuarial table for a human being.

[00:35:09] The error rate had to have been high.

[00:35:12] For AI to be able to go through that and find any of the errors, I love that.

[00:35:19] Yeah, I think you can see like two examples.

[00:35:22] Like number one, I think there's like more voice AI in like order ticking right you can call in they can understand because the domain right is the restaurant menu right.

[00:35:34] They're kind of the data domain is very limited very slow down.

[00:35:41] The cost of mistake is low.

[00:35:43] Like let's say you order something wrong, you're not going to be so mad. It's not life or death situation and either have fries or you don't.

[00:35:51] Maybe you didn't get the like sweet potato fries versus onion fries versus whatever right and you're not going to be so mad right.

[00:35:58] That is where I think the AI will start deploying in I call it low value transaction where cause of making a mistake is not that bad.

[00:36:09] And on the other hand you look at the other spectrum of like self driving. The cause of mistake is high like you could die.

[00:36:18] Yeah, right? Or like you have to be very very careful and that's why like Google like Waymo or Tesla has been working on it for more than a decade.

[00:36:28] Oh yeah.

[00:36:29] And it's still not quite there because it is very very challenging.

[00:36:34] I was in a self driving car in Vegas. It was through Lyft. So I know all the HR related people at Lyft to be in a conference so they gave me a different type of account.

[00:36:45] And I was in Vegas and it asked me do you would you like to ride in a self driving car?

[00:36:55] Yeah, connect your Spotify account. It made me do a couple of the things. Pick me up to the airport and it was a seven series BMW seven series and there was nothing on the dash.

[00:37:07] Yeah, but they had they had an LED sign that said hi William how you doing what what playlist would you like to play?

[00:37:15] I'm like I just told them what to play with his voice activated and they're like cool random honest and it drove me to the Bellagio.

[00:37:25] Wow. And so the entire time I was in Vegas for that trip, it drove me around.

[00:37:32] Yeah.

[00:37:33] And it was a trip because I'm in the backseat like you are like a rideshare right I'm in the backseat and there's nothing up front.

[00:37:40] There's no pedals.

[00:37:42] Yeah.

[00:37:44] I really like these. I really like the Spotify thing that was my favorite I'm skip skip.

[00:37:50] I play that louder.

[00:37:53] I didn't care, but it was going through red lights. I'm not going to realize it would come to a red light and stop.

[00:37:59] It would wait to a turn screen wait like a second or two and then it would go like it was a trip.

[00:38:05] Yeah.

[00:38:06] Yeah.

[00:38:07] And that's why I think we're getting they're getting there but like the cost of mistake is so high that they have to take the time.

[00:38:13] That's right.

[00:38:14] There's no shortcuts there.

[00:38:15] Yeah, that's no shortcuts there.

[00:38:16] Yeah.

[00:38:17] I like that you bifurcated things that have that the things that would have a tremendous impact on one's life versus things that you would you be inconvenienced if you didn't get those sweet potato fries, but it's not life.

[00:38:34] You know, I like that.

[00:38:36] I think I can take over those things. I think it's that and your comment around we fear the technological shifts like when horse and buggy went to cars, right?

[00:38:49] When we had first time we got on planes like everyone there's just a fear and there's a certain part probably is an option curve but some poor people like didn't fear at all.

[00:38:57] Got on a train like, cool.

[00:38:59] You know, different than a horse.

[00:39:02] But not everyone's like that.

[00:39:04] There's there's and I think I think the AI we're playing we're seeing that play out in front of us where people you see it in the media like it's just fear, fear, fear, fear, fear is like you've been interacting with AI for a while now.

[00:39:17] So like, how do you think Amazon does most of what it does?

[00:39:21] Yeah.

[00:39:23] You know, they don't have a bunch of people pounding out things as to say, yeah.

[00:39:28] So I don't have any other questions, right?

[00:39:32] This has been fantastic.

[00:39:33] This has been really good and interesting.

[00:39:36] It's been a pleasure to talk with you and for everyone that's still listening, please subscribe, share.

[00:39:42] This is a great episode so share it with your peers.

[00:39:45] And if you see us out there, say hello somewhere.