Revolutionizing Retail with Ambient IoT

Transcript

• Steve Statler 00:00

  Welcome to the Mr. Beacon podcast, you may want to give this one a miss. Sounds a bit strange. Let me lay it out for you the pros and the cons of continuing and investing your time in this particular episode of the show, which is approaching number 200. We've been going for about eight years. And so I will level with you about what you are about to experience or maybe what you want to avoid by hitting the stop button and listening or watching something completely different. So here are the pronouns. I want to hand we've got Thaddeus Sergura on who is an amazing talent he is works with Wiliot, he's Vice President, he joined us to bootstrap our data science, machine learning function. And so he knows a lot of interesting things. And he's doing a stint at hitting up a squad of ours, on site at one of the biggest retailers in the world, returning to a space as he's super familiar with, so he knows all about retail, a lot about data science, and I learn things from him every time I talk to him. On the other hand, the sound is not the best. We recorded this episode, live at the National Retail Federation show which is Vershbow, the big show, they call it in New York, every year, everyone in retail gathers together all the big retailers, Walmart, Best Buy. And every major vendor is there. And so I took the time to talk Thaddeus, he's a busy guy, it's tough to get his time, we were able to sit down, and we recorded it. But there's the noise of the show in the background. And you're gonna hear a lot of it because the end of the show people are getting rowdy. And you'll even notice those of you that watch. Well, that was just a camera angle changes on Thaddeus about halfway through, though, you know, we're talking and these guys that set the show up and tear it down, literally started moving the cameras. Whilst we were recording. It really threw me off. I was flabbergasted. But we powered on through. So there's your dilemma. Back in the early days, we got negative comments about the sound and the sound wasn't good in the early days of school. Look at now, this is an exception, you have to decide whether you're going to pursue it or not. And I'm going to use a three level simile to try and capture what I think the situation is. And Woody Allen is to tell a joke about about life. The simile he used, it's like two old ladies talking about a meal they had which really didn't taste very good. And you know, the worst thing was the portions were small. So life is similar. It's painful, bitter, a full of torment, and then it's all over far too quickly. And I think this episode is similar. The sound is very good, but it's kind of surely the normal, it's over too quickly. So you decide up you stick with it. But if you don't completely understand. And Mr. Beacon Ambient IoT podcast is sponsored by Wiliot, bringing intelligence to every single thing. Welcome back to the Mr. Beacon podcast, you're in the Times club. I think the first time I just want to think when it was I think you've joined pretty recently, when you were on last time.

  Thaddeus Segura 03:52

  Six months at this point, we're almost for two years.

  Steve Statler 03:56

  So you have seen an awful lot you saw a lot in your career at Walmart. And you're seeing a lot Wiliot. And I want to talk to you about a few things. One is the use cases that are really driving IoT in particular ambient IoT, what are you seeing, you've been running the data team, and you've got a special interest, and a leading the team working on the side and one of our biggest customers one of the biggest retailers in the world. So I think a lot of people are interested in IoT and NB IoT, potentially it's going to be orders of magnitude bigger than IoT, but it's really, I started calling it Gen three of RFID. Gen two is what's out there at the moment, but Gen three is this amazing thing. So tell us about the use cases. I also want to talk to you about the cloud. And you know, part of this new paradigm is cloud computing and whilst I try and avoid the Wiliot related subjects, I think this is particularly interesting. And Wiliot has designed these chips that enable ambient IoT. But we don't make any money from them. And we're selling them at cost. And we're charging for the cloud. And a lot of people like scratching their heads and like, I get why you want to be a cloud company. But really, really, do you need to be one. So I would love to hear from you. What is it that the cloud does in this paradigm? And why is it important? And where's it going with respect to AI, machine learning generative AI? So do you think we have enough to talk about and we're doing this NRF and the show's over dismantling everything around us? I think that'll be fine. We've got some very expensive microphones, hopefully, they'll do the job and we can focus on on this. So give us the state of the nation in terms of the use cases that you see driving Ambient IoT, where's the action?

  Thaddeus Segura 06:11

  Absolutely. I think there are a handful of places we're really seeing it. But the thing first of all, I always seem to explicitly state what Ambien IoT is, that would be a good stop for if you want to doesn't know. So in my mind, ambient IoT is this idea that we have low cost devices, that are getting data constantly with as little infrastructure and as little additional effort as possible. I think both of those things have to be true. On one hand, the labor is getting more expensive than it's ever been. So you don't want to introduce any more labor. And on the second hand, infrastructure is expensive, not just in terms of the devices you install. But as we learned very painfully over the last few years that running a single power drop in a production environment, like a large retailer can be $1,500. So the total cost of ownership rapidly grows, even if it's something very simple and very low power. So getting rid of labor and infrastructure costs is really the main benefit of Ambien IoT. And the idea is, it should just work all the time. So with that explicitly stated, I think the first thing that we're really looking at is disrupting the barcode, as I like to say, so barcodes obviously been around for decades and decades, there are trillions of them a year, which is how many ambient IoT devices I would like to see out there as well.

  Steve Statler 07:24

  That's the number actually 10 trillion is the number that ABI describe as the addressable market for this technology. And I think it's I mean, it's guesses what the one thing we know about estimates and forecasts is they'll be wrong. But it's indicative of a direction RFID is probably of the order of 40 billion units a year. So there's a big growth spurt. And you know, what we can say is, there's going to be a huge change in the world, if we put 1,000,000,000,010 trillion things online. But then I think people say, Well, we're already like 40 50 billion. That's a lot of tags, one nurse, and we're going to do with these things. So I think this question of the use cases that people will implement, is particularly interesting. From that perspective.

  Thaddeus Segura 08:18

  The number one thing we hear over and over again, is people want to know where their stuff is that that this vote on the wall here, do you really know where all your inventory is?

  Steve Statler 08:26

  And obviously, the clear the votes are in? And it seems like some people claim that they know where everything is, I want to follow up with those people. But most people don't.

  Thaddeus Segura 08:36

  Exactly. So that's the primary thing is that people want to know, the most simple case like retail, did the right merchandise, leave the DC to to go on the right truck at the right time, make it to the right store, and then ultimately to the right shelf, and then into a customer's cart? And then did it actually leave? And there's so many things that can go wrong all along that path. And historically, to answer any one of those questions discreetly, you would have to install infrastructure or introduce labor. If you want to know if the right thing went on the right truck you scanned, if you wanted to understand if something didn't properly sell and like a clothing store, you're almost blind unless you want to follow that person around and actually interview them and ask them, oh, did it not fit? Can I get you a different size? And so that becomes very, very cost prohibitive to do at scale. But what you can do with Ambien IoT is you apply one tag or one smart device. And then all throughout the chain. As you're getting that data. It'll go up to the cloud, which we'll talk about later. And you're able to answer all those questions with the single technology rather than introduce things incrementally and drive up your costs.

  Steve Statler 09:38

  Very good. So what we're doing is we have this trade off, we're trying to reduce labor, and there's this tension between labor and scanning. We can't have people being fooled around and carrying expensive scanners and doing all these manual things. So it seems like we're in a space where the name of the game 10 years ago or too many years ago was absolutely minimize labor. So if we can scan something, when we ship it, let's assume that what we ship is what's received. One of the things that I learned from you, which just shocked me was how little scanning there is the goes along, not the scanning is good, it takes time. And it tends to be an accurate because there's human beings involved. But what is the level of resolution that exists today in terms of visibility of the supply chain very long.

  Thaddeus Segura 10:31

  Once you're in the store, very little, anywhere, you can scan at scale, it's not bad, but a lot of things are built on what we call assumed receipt. So if a truck comes in, you don't have time to scan every single box of 3000 cases. So you assume they're all there, you might audit 2% of the time and extrapolate that accuracy to everything you receive from that supplier. And then you may even introduce fines if they deviate outside of that rain. Within a distribution center, there's not a lot of chaos, it's pretty organized. And there's a lot of scanning and scanning out where the confusion actually happens is in the store. Once it hits the back door of the store, there's very little scanning, and it's complete chaos. So there's typically no scan when a truck is unloaded, at least into the retailers I've worked with. There's no scan when it actually goes on shelf. So what ends up happening is you assume that everything that was on that invoice is entirely in the store. And then once you assume it's in the store, you typically assume it's on the shelf, until you get some sort of scan like in the back room that says otherwise. What that means is that on days, your labor is low, or you have call outs, or maybe there's extreme weather and people can't come to work. All that merchandise that came in, you're saying it's on the shelf and ready for a customer, they're looking on their phone, or they're ordering online or ordering through Instacart. And it shows available. But what they can't see is it's available in the store, but it's still in the back room in a case where it's still at the dock, even on a truck that hasn't been loaded. And so adding more granularity actually gets you closer to the ground truth and allows you to actually promise the right merchandise to the customers and to tell them the true state of the inventory you have in a building.

  Steve Statler 12:03

  So it seems like the answer to that is let's give everyone some more handheld scanners and start scanning boys. Why not do that?

  Thaddeus Segura 12:11

  Yeah, I think every scan on average done by a person ought to be somewhere around 50% compliance with most processes that require scanning. And two, there's incremental labor costs associated. Okay. Especially as you're seeing a lot of retailers move away from laser scanners towards optical scanners, it's even slower for whatever reason, you're using a phone, you have to unlock phone, log into the app, pull it open, the camera has to focus. It doesn't seem like a lot. But if you do it 3000 times a day times 5000 stores that adds up very, very quickly.

  Steve Statler 12:41

  That makes sense. Okay, so maybe we can't have people scanning and the fact that the thing that has surprised me is the compliance issue. And it's not like these people don't want to disobey, it's just that they forget, or they're under huge pressure, and they don't want to get shouted about so that they're like, Oh, I'm gonna do this thing. Okay, so I think at the top level, then we want that visibility. Let's double click in here and get to what are the use cases that trickled to the top of the priority list? And what are the metrics that those drive?

  Thaddeus Segura 13:19

  I think you think about ROI, balancing act between investment and return to get on that investment. So pallet tracking is a really easy place to start. If you're tracking at the pallet level, there's not a lot of investment in whatever sensor you're using. For us, it's a tag. And so you can tag that pallet. And for almost no extra cost, you can track that all the way through and make sure the right merchandise is making to the right store. And ultimately in the right cooler if it's cold chain compliant. If you want to double click another layer down, we see the unit economics of case level tracking, making a lot of sense. So if you're actually tracking these cases, then you can understand things around was it actually stocked to the proper shelf? Did it go on to the proper shelf? Or is it still sitting into the back room. And then even understanding when that box was thrown away to get like a termination signal to understand where it's at in its lifecycle. So those are the things we're seeing today. In terms of item level tracking, there's also a lot of different value you can have as well. But when you go to item level tracking, your costs also increased by an order of magnitude. Yes, that's one of the reasons I think that our menu has struggled to get traction, it does become very expensive, even at three to four cents, or five cents to each individual item.

  Steve Statler 14:26

  I think in my personal view, and I think this is an area where we differ is I think there's a world where item level tracking works when you go all the way to the end, when you start cradle to grave or even Cradle to Cradle. I was actually hearing someone talking about Cradle to Cradle when you reuse the product again and again. And obviously it makes sense for high value items. But my hypothesis is that if you due to a subscription model for toothpaste, even if the tag costs four cents, or whatever it costs, it's worth it because you just saved yourself a bunch of Google ads and labor costs in the store. And you can basically see the demand signals that mean you can can cut costs in the production. And maybe I'm trying to create a full spirit of disagreement and conflict between us, but I actually am aligned with you.

  Thaddeus Segura 15:31

  I think where we might differ is our opinion on timeline. Okay. Like I do think that eventually will happen. And you see it today, like how hard Amazon works to get you to subscribe to whatever it is that you're consuming, right? If the default option so often, yes, but in my experience, I actually don't like it, I almost never choose it. Because I can't ever get the demand. Right. Right. There's almost nothing that consumer like a linear perspective, other than maybe like cat litter. Yeah. And that's a very easy thing to like, go to the store.

  Steve Statler 15:58

  I totally agree. I actually am on a subscription model with pool swimming pool test kits. This may sound like a very yuppie, so I apologize for anyone that is resentful about my swimming pool, but it's California, you got to do it. So I subscribe to these test kits. And they are accumulating like crazy. And I'm not actually paying for them. But it just makes me feel horrible. So I think the subscription approach, especially if you're subscribing to a category, like I want to subscribe to not just one sort of cereal. But all this here, I want to subscribe to all the herbs and spices and tell me about the ones that I would like that. So I think there's just so much value there. And I believe that this is why the economics may be crazy. So I'm going to argue for shorter term breakthrough. And I think someone will do it. And they'll suddenly expose all these efficiencies because they have loyalty from the customer. And I think their stock price will go up because they've got a recurring visit business model. And I think it'll be a bit like Tesla, you know, Tesla's valuation was bigger than all the other car companies, does that mean that they're making more cars? No, absolutely not. But they're doing the thing in the future. And I think product as a service is the thing of the future. And people's cost of capital will go way down, if they do it. And I think it's going to happen in three years.

  Thaddeus Segura 17:25

  So there's something explicit that I mean, something you have implied here that I want to say exclusively. Okay, so I think you need something like Ambien IoT to fuel that breakthrough. Yes. Today, you have this inaccuracy, your test kits are piling up because they're not being used. But I think about something like laundry detergent, having a sensor on there that's low cost and disposable that can actually send the signal to a Google Home or my Alexa to automatically reorder, it's truly demand based, because I know that I'm getting low. That makes sense. And that's something I would subscribe to. And that's one of the benefits, I think of ambient IoT and why that will facilitate the transition you're talking about. Yeah, still don't think it'd be three years?

  Steve Statler 18:02

  No, I love it. You know, I feel like we just switched switch roles, and you're doing a better job of interviewing me than I am of you. So let me try and switch the tables back again, and say, what are the other use cases that you're seeing? Maybe ones that are surprising in the ambient IoT world that where there's there may be value?

  Thaddeus Segura 18:23

  Yeah, one we've heard a lot this week a lot to this precedent. One was goods, not for resale. The things you don't think about that are just the cost of doing business, like ladders and chairs, and even desks and dumpsters. Yes, so many of these things go missing in these environments. They're complex, they're chaotic, and they're expensive. And so having the ability to track goods you don't necessarily sell is actually a pain point that we heard a lot this week. The second are just devices, anything electronic, even if you don't sell them, like handhelds, and even POS pricing apparently gets stolen a lot. And so I think asset protection is really a big focus right now. And having the ability to be able to track those items that doesn't require a 30 or $40 sensor to know where your things are at is a really compelling, interesting environment.

  Steve Statler 19:12

  What's your take on items in transit? Again, this is one where we've kind of had a little debate off off off camera. So in the past, the idea was, you scan things going on the truck, you scan and coming off. We didn't really know what was going on on the way.

  Thaddeus Segura 19:32

  Yeah. For me, I like things that are actionable. Yeah, I think seeing the data, there's some incredibly interesting things that happen in transit. We've watched merchandise, freeze and thaw and freeze and thaw. I've watched Halloween candy in the middle of summer hit 114 degrees Fahrenheit. You send someone to the store to inspect it. It's exactly it's almost like liquid. What's your inside of there? Yes. So you see very interesting things in transit. I don't know how you act on that. Yeah, sending the signal to that truck when it was hitting 114 degrees, they can't drive in the shade, I don't know what they would actually do. So as I think about like the ROI of those insights, for me, I don't see the action there. Especially when it comes to ambient. With cold chain, there's stuff you can do, and talking to sensors and changing set points. And so I have a different opinion there now than I did before.

  Steve Statler 20:21

  Yeah, that's, that's fair. But we are seeing, I mean, there's already a telematics industry, there's already the containers are starting to be more and more connected. And so why not share that data using the tags, but the key is going to be people figuring out how to take actions on it. My view is, if I can see that I'm destroying large parts of my inventory in transit, then I'll probably figure out how to action but I, you have more experience in that field. And I accept your skepticism about that as as a breaking factor. And again, my view is someone's going to do it, and they're going to show the value. And then all US pundits and in your case, experts will will see the results of what people do with the data. So that's the exciting thing, you're starting to see things that we've never seen before. Very good. Well, let's move on to the next area that I wanted to pick your brains about, which is most people look at what is happening with ambient IoT and they get postage stamp compute devices, they understand they're talking to commodity radios that are a fraction of the cost. That means we can get more visibility. So why can't I just take? What about all these software layers? Why can't I just take my existing systems or easily write new systems and take that beacon data and or tag data and do something? Why do I need a cloud especially Wiliot are get back to designing some new chipsets stop getting in the cloud? Why the cloud?

  Thaddeus Segura 21:58

  Absolutely. So my favorite analogy lately for this is do you use Google Maps? I do. Okay. And now imagine if instead of saying turn left turn, right. It just yelled longitude and latitude and you the whole time?

  Steve Statler 22:09

  Oh, yeah, that would be bad.

  Thaddeus Segura 22:11

  It's probably very annoying. And I think that's what a lot of sensors are, yes, they're telling you your temperature and humidity and all these other things. But it needs to be turned into a product to be made usable, you need the turn left turn, right. And that's the value of the cloud. It could be done locally as well, like you do with Google Maps, sometimes really like all of that information can come together. So for me, in my mind, the cloud does a lot of things. And we've learned a lot about this over the last couple of years. We started with really general models that would tell you things like location. And that was great in very broad spectrums. But there's always this trade off between generalizability and accuracy. And what we personally found is that a lot of our customers needed more specific, more accurate models. And so we actually came up with a way to train custom machine learning models, that would only be possible through the cloud. So we do actually go on site will gather labeled training data around the problems they're trying to solve, like whether the right item went onto the right track, which is really hard to do from an RF perspective, and be able to give them that prediction with a very high degree of accuracy.

  Steve Statler 23:14

  So that's an example of going from longitude and latitude to the actual turn right, turn left, that I think a lot of people feel like, okay, yeah, there's a bit of complexity there. But what about this RF issue? Why why is there an RF RF issue? And is it really that difficult to deal with?

  Thaddeus Segura 23:32

  Absolutely. So if you think about traditional technologies, like RFID, they're great accounting, it's a strong wave bounces off an antenna gives you an ID back, and you get something called return signal strength. And people think, okay, I can just look at the return signal strength and understand where something's at. And maybe one point is not enough, maybe I need a few points. And I can triangulate, the challenge is that there's a lot of things that impact return signal string. So I could have a reflection. So I could have something right here, an asset right here, and a receiver in my hand. And these could be very close together, but my wave could actually travel away across the room off the wall and then bounce back to the receiver.

  Steve Statler 24:10

  Okay, so the route between A and B is not necessarily a direct route, correct.

  Thaddeus Segura 24:15

  And then you get into like more nuanced things like antenna polarization, like if this antenna is aligned with this antenna, and how this is moving in three dimensional space.

  Steve Statler 24:24

  So I moved the tags like that, then the signal strength is going to change, but the distance hasn't changed. We just completely confused my algorithms using signal strength.

  Thaddeus Segura 24:33

  And I guess if the tags like that, even more, yes, it's three dimensional. So okay, yes. And then beyond that, you also have things like line of sight issues. So if you have an obstruction in the way, especially in environments where people are walking through, or 90 something percent water, which is not conducive to RF, right, so in any of those situations, you don't have enough information to be able to tell you location using just RF. So what we do is like each of our devices right. Computer, when they flip on, there's 80 columns of data that are transmitted back to the cloud. And those become inputs into machine learning models. And that's how we actually train these things to make very intelligent decisions about where things are at.

  Steve Statler 25:12

  So we're using machine learning to take all of this ambiguity and complexity and spot some patterns that give us some simple information about where the, the thing is.

  Thaddeus Segura 25:27

  Exactly, yeah. And the machine learning is not magic. It's just math done very rapidly, that kind of self organizes. So I could write manual code to make a decision tree in every single location and say, if it was seen by this bridge, and the RSSI was stronger than this and the temperature was greater than this, then it's in this situation, and do that 10s of 1000s of times, or, yeah, I could run one line of code, after I've done all my pre work, and it will do all that for me very, very well. Net becomes the model that I can then use across multiple accounts.

  Steve Statler 26:00

  Alright, so what I've observed is, you've got a ton of smart data analysts that are harnessing these machine learning algorithms in order to restore simplicity and tell us where the inventory is, is it on the truck? Is it at the loading dock? Is it further into the warehouse, all that stuff. But it still takes people pretty expensive people who worked for Wiliot to help do that. And it takes time as well. Give me some hope that this is gonna get simpler and easier. What are you doing to streamline this?

  Thaddeus Segura 26:42

  Absolutely. So I think there's a lot of different ways to go about it. And we started with very general models that weren't specific, but very general, but not super accurate. They went to this world where we have these very manual models I just described. But my big focus right now is how do we scale, we're seeing incredible demand, arguably more than we can handle. So the question then becomes how do you copy and paste that and make it very scalable. So we're working on algorithms now that are faster to train that can be trained by non technical people. And a great example is Eric, who's actually on your team was a very technical guy, but not a data scientist by trade, was able to go into a store and in 30 minutes, place a couple of calibration tags, train a model, and then turn around and deploy with 100% accuracy. And so he's able to learn sub zones in a small retail format, and then be able to understand which rack these different items are on without being a data scientist, you simply train the model.

  Steve Statler 27:35

  And that sounds like you need a PhD. But can you describe how would you train this data model?

  Thaddeus Segura 27:41

  Absolutely. In his situation, he placed a couple tags and hit start, what happened behind the scenes is it's looking at all the different attributes that we can see, and creating an ideal representation of what that space looks like mathematically, okay. And then once that converges in that mathematical idealized form, it then looks at all the different items it sees, and compares those states to the other states and makes a prediction about which form it's closest to. So that's pretty much all it's doing. Its abstracted away into code, he just hit start.

  Steve Statler 28:10

  So I set up some readers, I put some tags on assets, maybe I moved them around the store a bit. And that basically trains the the algorithms about what the footprint is of this environment. And this kind of reminds me gets me thinking back to indoor location. And I saw companies do similar things. You want to know where where the phone is, and they had training processes there. But yeah, rather than one phone that's moving around, and you potentially have 20,000 inventory items, that kind of using that same training, which is fairly simple process. Exactly. I was absolutely gobsmacked when I saw this, I was like, you know, we had no time to get ready. And we certainly didn't have the staff available to customize the algorithms and the ability to get in and get out was phenomenal. So hats off to you for doing that. So let's, let's wrap this up seems like the stage to dismantle the show around us. We've already had people interrupt this interview, at least twice. So where are things going with AI? And of course, machine learning is AI, we've talked about that. But But what's gonna what are you what are your expectations in terms of generative AI, large language models? How does that play in with all of this a lot?

  Thaddeus Segura 29:34

  I think I talked about the trade off a couple times change generalizability and accuracy. And that's true within the same generation of model. Generative AI is an example of like taking a step change where you get the generalizability and the accuracy at an unprecedented level because an actual technological breakthrough. And so what I think will happen is those exact things you'll see models that are more generalizable and as accurate, the challenge then becomes How expensive are they? So you see people using catch up t as a calculator like what's two plus two, which is not the best way to use the 6 billion parameter model when you could just use the calculator, right. So there's a consideration there on cost. But I think that's what you'll generally see is that these things will break through. And then you'll get both of those together to be more concrete about how we're using it today. I think as a tech company, there's a lot of ways you can use it to enhance the efficiency of your organization. One of my big tasks for last year's was to build a generalizable library. So we could do POCs faster, and not have to write a bunch of code from scratch, and then honestly went out of the window, because we didn't need it anymore, like Chat GPT could write the code, we'd be able to type in exactly what we needed. 80% of it was done, which was the exact point of the library. So it's making everyone more efficient, it's making it easier to learn really technical concepts like antenna polarization. And so it's how we're using it internally. Externally, there's a way to use it to enhance the product, which is a lot more complicated. Generative AI is more than just large language models. That's one example. Generative AI is another example. You can use the same techniques of a generator and discriminator, to build fake training data to simulate different environments. So there will be more theoretical applications that customers won't necessarily see, they'll use the same principles that won't be as flashy or as pretty, but will still give you the same enhancements allow you to break through so you get the generalizability and the accuracy in a single model.

  Steve Statler 31:28

  Okay, what about in the boardroom? Do you? Are you expecting the CEO or the SVP or the EVP to use this with all of this data we're producing?

  Thaddeus Segura 31:43

  We do, but we don't really want them to even know that it's happening. In the same way. I've seen 20 ai assistants here this week. And the idea is that you can't tell the ideas that eventually the technology is so good, you don't know you're talking to a person or to a robot. Yeah, it's the same thing. Like ultimately, we're still going back to that turn left turn right example, you don't know that Google Maps picked you up on a side street, because it's been smoothed out and all that's hidden from you. In the same way, maybe we miss a read, and we're able to actually infill it using generative tactics. So we saw five other tags, but we missed one of them, but we knew they're traveling in a cluster. So we can actually impute that and make the same decision. Even if we might have missed it. At that point. Eventually, this stuff will get so good, and we'll be so confident they'll be using the data but won't even know that it's generative on the back end.

  Steve Statler 32:29

  Okay. I mean, what I was thinking about was just, you know, scenario analysis questions. You know, there's a conflict in this part of the world, what's what's, how's this likely to impact product availability and all my stores? Or where do you see the slack in my supply chain today? That's sort of question is that feasible?

  Thaddeus Segura 32:56

  It is feasible, I think, I don't know if that's a problem we'll specifically solve. And then one of the challenges you'll face is that the recency of the data, which is a big challenge for LLM is, but there are new things that are coming out that will help bridge that gap that will allow them make the decisions closer to real time.

  Steve Statler 33:12

  So Thaddeus, have you had time to think about your three most meaningful songs?

  Thaddeus Segura 33:17

  I don't know. Oh, it's just to be able so much better than we wouldn't be like trying to have a brainstorm Sorry. Sorry. All right. And it did not come up with any good recommendations. Yes. No. Okay. So what with three Eminem songs last time. And number one piece of feedback in work is give me a hard time about the evidence or than any other content. And the last step is to be good. So I don't have a good answer for you.

  Steve Statler 33:46

  Okay. Well, that's fair enough. We can have cities. We're gonna have some things. Yeah, we really like New York. Yeah. I mean, we're in New York now. And it's freezing cold. But there's something very invigorating about it beyond just the physical aspect. What is it you like about New York?

  Thaddeus Segura 34:06

  Growing up, my dad worked here, on and off for probably three or four years. She didn't tell 911. She used to come out here three or four times a year. And first of all, there were real seasons coming from California. You really just have like, fall and spring most of the year. So I did like the snow. And then second, it's a real city. I live in San Francisco now. And it's this big. And you go from here to here and you're across the entire city, New York. You can walk and walk and walk. And it feels like it never ends.

  Steve Statler 34:37

  Yeah. And as real public transport. Oh my god. Yes. Yes. And you could eat at a different restaurant every day of your life and still don't get through all the rest of them not even scratch the surface.

  Thaddeus Segura 34:49

  Yeah. And people don't go to bed at 10pm so back home everything is shut down here is.

  Steve Statler 34:55

  One of the things that I was now I'm like terribly ill doesn't bother me so much but when I first came to America. I just couldn't believe all the restaurants were shutting it like eight, nine o'clock. It's nuts. Okay, so we're agreed on New York. What's, what's your second city? I mean, in San Francisco in the top three, you just badmouth? Hopefully you're not planning to go back.

  Thaddeus Segura 35:18

  I live in Oakland for a couple of years, actually liked Oakland better than San Francisco.

  Steve Statler 35:23

  Well, yeah. Well, you know, I was born in San Francisco. So I'm deeply biased.

  Thaddeus Segura 35:28

  I grew up in Sacramento. And I remember going down to San Francisco seems like a drive like 20 years ago. And it was just a very different city. It wasn't like pure tech the way it is now. It was a lot more lively, a lot more art, a lot more culture, and people just being different and expressing themselves. And it feels like a lot of that has been pushed across the bridge to Oakland. And so it's a lot of like the San Francisco I grew up with, but now just in a different city of Oakland.

  Steve Statler 35:52

  And is that just because it's more affordable, and Oakland?

  Thaddeus Segura 35:55

  I think that's a big factor. I think there's so much tech San Francisco, that it's hard to do anything. We're gonna look to note when you meet someone, and they're a glassblower, or a metal sculptor, from all these different walks of life. It's not just what series are you? What's your unique value proposition? Which is I feel like the conversations I have, I can fly in or in an elevator?

  Steve Statler 36:15

  Yeah. Yeah. Pretty good. And so what's the number three?

  Thaddeus Segura 36:22

  I actually really love San Diego. Believe it or not, you're gonna have to go down there. Yeah, I thought that's where I'd end up living. But my wife and I are in San Francisco, and she's the last so that's where I'll stay.

  Steve Statler 36:35

  Okay. Yeah, my wife was in tears. When we first looked at moving down to San Diego, she I think the reason she was in tears was less about that San Diego was in some way objectionable. But it just meant that it was so nice that it became very apparent that we there was no excuse for not moving there. And that meant leaving her family and her friends. And now, but now she's made friends and actually her parents moved down and our family, the family visits. So I agree. I think it's it's a great city. So we disagree about San Francisco, but that's from a position of relative ignorance on my part, and a bias because I was born there. Okay. Well, Thaddeus, thanks so much. I enjoyed the chit chat. disappointed about the songs. I have to go back and check out the other episodes, see which m&m ones you you went for. But thanks very much for being on the podcast, of course. Thank you. All right, you made it to the end of the show. And to be honest, I really wasn't expecting that. So I got to thank you. I also got a thank you, Aaron Hammock, who adds to extra duty in trying to clean this up and make sense of of what I gave him. And Brooke Ellsworth, who is been publishing our show and promoting it. So until next time, be safe, and enjoy life.