Thinking Small Drives Big Leaps in AI | Monthly Podcast with David Trainer and Kyle Guske II

There's no better proof that the quality of training data is paramount to AI's success than the evidence that small language models significantly outperform large language models. We'll get into differences, similarities, what the SLM-LLM argument really is throughout this presentation today. This is our monthly podcast. We're here with David Trainer and myself, Kyle Geske II, titled Thinking Small Drives Big Leaps in AI. Welcome in. Start off with the disclosures and disclaimers. Take a look at these. Main thing is we are not providing any advice. We are publishing research. Today's podcast, this is a monthly podcast, so welcome in. We're glad to have you guys here. The next one will be November 14th. If you're registered for this one, you're already registered for that one as well. So you can, you'll get the alert emails on it around November 14th. As for the latest research we've got going on, we've got some new long ideas published on the website, some updated Danger Zone picks. Earnings season is coming right around the corner, so we've got our earnings preview reports, stocks most likely to beat, stocks most likely to miss there, looking at GAAP earnings, core earnings, Street earnings, kind of getting at the true earnings of some companies in the market, specifically the S&P 500. I put out a nice e-letter on today's topic as well, Thinking Small Drives, Big Leads, and AI. There's been a new post on our AI agent with Google Cloud that we're going to talk about today as well. Some model portfolio updates and some pre-stock picks all on the website right now. Today's agenda, we have a celebration, a birthday party. It is the 1-year anniversary for the Core Earnings Leaders Index. So we'll talk a little bit about that. We'll talk about the posts on our community, some trade idea discussions, upcoming trainings, and then we'll get into our AI agent. We looked at it last week. We'll take a look at it again this week for everyone in attendance and take some questions. Be sure to send us in the Q&A and chat. Let us know what, you know, any questions you have about topics today, other topics. If we had time to get to them, we can also address some of those questions as well. But to start it off, as I said, we have a happy birthday celebration. I'm not going to sing happy birthday song, but it is the happy birthday to the Core Earnings Leaders Index. It turned one year old on October first, so a little past the one-year birthday. But the performance of the Core Earnings Leaders Index in its first year of active trading speaks for itself. We're very proud of this to see that the index beat the S&P 500 by nine percent in its first year. That really showcases the power of core earnings, our research. Find picks that outperform the overall S&P 500. And then it's not the only industry or index we've developed. We've developed 2 others, and all 3 of them are outperforming over the past 5 years in sort of backtest trading or trading with that 1 year of live performance there. The Very Attractive Stocks Index, which allocates to stocks rated very attractive by our rating system, beat the S&P 500 by 62% over the past 5 years. And then the Core Earnings Weighted S&P 500 Index, which weights companies, the top 500 largest companies by core earnings, beat the S&P 500 by 31%. So some great performance on live-traded indices driven by our research. That's— we're proud to see that, especially on the 1-year anniversary. Beating the S&P by 9% is no small feat. These indexes are managed by Bloomberg. They're not available for the public to purchase. But they're all based on our research, the same research that we share with you guys on our website, through email updates, our online community, sending out alerts, research you get with our stock ratings, long ideas, danger zones. All of that is driven by our roboanalyst AI agent for investing, superior fundamental data, and it all combines into, you know, live-traded indices that are outperforming. Proud to see that outperformance. And not only Do you like to see outperformance? Like to see what you guys are talking about too, which is why we've set up our community. We've set up these podcasts. I would set up some of the trainings we do to hear from you guys. We'll talk a little bit about the posts on our community and trade ideas. I think the one to feature, talk about, is always interesting. This first one here, using New Constructs with another screener. This is from one of our pro members, I believe. Basically integrating New Constructs research with other options, the other research providers out there, you know, completing a big puzzle. with New Constructs pieces in it. It's a way to kind of integrate other providers, layer in New Constructs, or layer in other providers and get a different picture or a more complete picture of, you know, whatever trading strategy you're looking for. So if you haven't checked that out, be sure to give that one a look. We're always saying, you know, there's many different ways to do, you know, different types of research, whether it's fundamental, whether it's quantitative, whether, you know, momentum, all that kind of stuff. We're focusing on superior fundamental data. If you want to layer in others There are definitely ways to do that, which kind of leads into the member trades. I've got a few different trade option discussions on the community as well on tickers FSM, CRC, BRY, and MDB. Some good option strategies there provided by our members in the community. Community is free to join. I'll put a link in the chat as we kind of go, but I think it is also at newconstructs.com/society. So you can check out anything we talked about today. You can check out the posts from the community. There's different topics, different areas. You can post questions for these podcasts and kind of let us know what you're talking about, what you're interested in, what you're seeing in our research, other research, kind of topics of discussion, really get the juices flowing on ideas. We've got more people thinking about it, more people talking about it. We can find different insights maybe hadn't thought of to begin with. But on the topic of insights, we also have some upcoming training sessions we want to talk about. The first one being the Ideas Dinner. is for our pro members and institutional members. That'll be October 21st at 4 PM Eastern. The pro and higher members should get emails about that if you're not already signed up. And then we have a special earnings beat and miss season training coming up also on October 21st, where you can find out how to access our earnings preview, including the stocks most likely to beat and miss. And then we do have a new training in progress. David, you want to talk a little bit about the Slingshot Insights that we're working on right now?

Yeah, so you guys picked up on the fact that we're partnering with quite a few folks. Look, I think there's been a major shift that we've been writing about a lot. If you guys have been clients or readers for a while, you are seeing that the world has shifted to, hey, we spent $50 billion trillion on AI data centers and chips that have yet to be built, but what are we going to use for data to drive them? And this has really finally woken people to the idea that, look, the internet is not a reliable source of data. We need super high-quality datasets that have built in a way that a machine can understand to drive AI, to drive insights. And I think as this switch is flipping, more and more people realize, oh, good data isn't just a nice-to-have, it's a need-to-have. Because when they start extrapolating on bad data, they will see at scale how wrong and how far away that can lead them. So lots of new partners interested in working with us. One of them I'm really excited about is the Slingshot guys because they're going to bring specific domain experts into our calls about long ideas. For example, a few months ago, we published a long idea on Crocs. And so, we could bring in a consumer products expert to that meeting. And we're going to do this live with everybody. We would ask the expert, what might we be missing in the consumer products market related to shoes or sandals? the direct competitors of Crocs that would make this idea that we presented not good? Is the stock cheap for a reason? What trends are we seeing? Is TikTok really that effective a marketing tool? Are channel surveys out there that you've seen that show sales of Crocs are actually slowing down? It's becoming unpopular. All those things that we can't really get from fundamentals or from financial filings or even from the news, we can ask these experts. And I think it's a nice combination, just as Kyle mentioned about how using New Constructs is a key ingredient in your investment decision-making process is important. This is just a great example of combining best-in-class products because one best-in-class product is great. New Constructs is great, but let's use it with other sources of great information, other great products. And that's what we're going to be doing with these guys here in the next couple of weeks.

Awesome. Yeah, exciting to get that rolling. And like we said here, we'll email out dates and times once we've got those set in stone there, so you guys can have access to that as well. It's another way to leverage new constructs and other information out there. And with that, we'll get to the kind of headline topic of today, Thinking Small Drives Big Leaps in AI. The title is from an article, e-letter type of thing that put together. The big topic, you know, AI, it's everywhere. There's billions, trillions being invested, data centers, chips, data, everything. But the one clear takeaway is AI is going to change the investment industry in a big way. But the details of how it can change the industry matter. AI doesn't perform well if it's fed poor training data. Whenever AI is involved, the quality of the training data is the single most important factor to its success. You know, we often hear Bad data in, bad outputs. So that's the same with AI. Current large language models, LLMs, fail miserably at understanding the meaning and nuances of financial data. Just knowing what a number is does not help you understand what that number means. LLMs are very good at handling a broad range of general tasks. Maybe not very good, but their performance declines sharply when faced with more sophisticated tasks like parsing an entire filing, knowing what every number means, where to put it, where it needs to go. To get a picture of that company. And meanwhile, on the other hand, small language models focus on a narrower and more specific domain, which enables human experts to ensure the data is accurate. I found a nice kind of comparison of SLMs to LLMs here. And you can see, you know, the key differences being parameters, training data, computational requirements, cost, and then expertise kind of mentioned. LLMs trying to boil the ocean. SLMs, looking at a very specific dataset where you can actually teach the machine better, more accurate, and drive more insights. And with that accurate data, clear instructions can be built to produce truly reliable and human-level expert insights. We think the best proof is to look no further than the indices we shared earlier. Our SLM powers market-beating stock picking. We see the Core Earnings Leaders Index, very attractive stocks index, Bloomberg 500 index, all outperforming, and they're all built on the data driven by our SLM. But we're excited that that is not all it can do. We're not just outperforming the market. We've got other uses for it as well, which is where we lead to the partnership with Google Cloud, building the first AI agent to produce truly reliable results. We've got some screenshots here, but we're actually going to do a live demo, and I will turn that over to David to demo the new AI agent. If there's any updates since we talked last week, last month when we last looked at it, but we can kind of dive in and really get in there.

Yeah. So we're constantly building new functionality into this agent and we're really pretty excited about this. First of all, I wanna, I'm gonna pop something in here, Kyle, that's a little bit off agenda cuz we've been talking about it. And that is, folks, we are thinking about and just looking at some some feedback on the community and stuff. We're thinking about creating a new membership. It's going to be a Stock Tracker 50 Plus. We'd love your feedback. And the idea is going to be, it's going to be just like Stock Tracker 50, but it's going to give you on the, on the, you're still going to have access to the ratings page where you can look at ratings for lots of companies. And you'll get the signals that the colors, but for the stocks in your portfolio, the 50 stocks, We'll give you the full sort of professional view so you can see the details behind the ratings. We're thinking about charging $149 a month for that. So it's, it's more expensive than the traditional Stock Tracker 50, but it gives you a ton more data. And by a ton more data, I mean on any particular ticker, you know, you're going to get the investment summary, the rating breakdown, the gap reconciliation and adjustments. You won't get earnings distortion. That's an institutional level thing. But you will get these details and the numbers inside. So we think that's cool. We'd love to get your feedback. Throw it into the chat. Send us an email. Stock Tracker 50 Plus coming to a website near you very soon. Let us know what you think about that. So let's go back to FinCite. So we're doing some really cool stuff here. One of the really— one of my favorite things we can do now is regression analysis.

Yeah.

I don't know if we showed this off last time or not, Kyle. Do you remember?

I don't think so.

So we can plot. So I said, listen, give me the regression analysis of ROIC to enterprise value for all stocks in the tech sector. So it throws together this regression analysis. You can see enterprise value, return on capital. Is this big enough? Should I go bigger?

Can't hurt.

Right. And this is really important functionality, folks, because this is going to be part of a module that lets us Lets us give clients the ability to figure out what drives valuation for stocks. That's ultimately what's going to come out of this is like, does valuation matter? And we're going to provide this as a starting point and let you do regression analysis, anything. We can earnings growth and PE ratios and compare those and get detailed R-squared details, p-values, linear equations, these things that allow us to really understand what's going on. And explain to you why this does or does not matter. Now, keep in mind, this data is a little bit old. It's not on all companies, so we're not getting a full sample dataset here, and this is in beta. And I've already got some people on the website asking for access. We are going to, in a couple weeks, start rolling out some beta access to folks. And that's going to be for free in exchange for logging your feedback in a special feedback form that we're going to be sharing with folks, because what we need is testing, and we need you to paste results that don't come out the way you want into the system so that we can track that and fix it. And so that takes a little bit of effort, but that's going to be what we're going to require in exchange. The other cool thing is asking questions like, give me the stocks in my portfolio. Here we go. What stocks in my portfolio are likely to miss earnings? And I forgot to put the stocks in here, so it said, well, I can't do that without your portfolio. But then I pasted the stocks in. And it came through and said, here are the stocks from your, from your list that are most likely to miss. And it gave everybody's earnings distortion score. And then I said, show me the earnings distortion, the amount of earnings distortion for these stocks. Right. And then I can say, how about show me the last 5 years of earnings distortion for, let's just say, PLTR. So the idea is, look, I can get details on lots of companies that I can drill down and get specifics. But you're going to be able to ask questions of this AI that are of great deal, a great deal more substantive value. It's not just going to be, you know, I don't know, the general GPT stuff. You can get real answers based on real data that we know to drive alpha in the stock market. And, you know, look, maybe you get tired of me showing this, but When you compare the Core Earnings Leaders Index to the S&P 500, you get real outperformance like just over about every time frame. Today, right? The last month, six months, year to date, one year, five years. It matters. And so here we can see the the data for the last twelve months, the prior five years explains what earnings distortion is. Let's see, please put. that data into a chart. This is going to be just another way for folks to interface with our data and also go out and be able to ask broader internet questions. A great example of a great question comes here from our man John. Kyle, we got to start sending out awards for people who are top commenters in the community because John's definitely a front runner. Great stuff. Even though he goes by multiple names, it's John, is it Jay? But I don't recognize the last name. Anyway, so this is a great question. Profitable educational institutions with the greatest percentage forecasted increases in enrollment. Great way to combine. And John, by the way, John is a great guy in terms of combining our tool with the Stock Rover stuff. This is the way to do it, guys. Until our AI, and by the way, it's on our roadmap for our AI to include technical tools like what John uses and combines from other sources. But, but technical analysis, you know, it's not that difficult to build. That's why every broker in the world has it. But we're gonna include that in the AI. But other questions around the industry like this are great. So let's see, Phoenix Education Partners. Is this, is this, is this a new IPO? Didn't we do something on this, Kyle? What's the deal on this?

Yes, we had it covered pre-IPO.

Yeah, here we go.

Look this good. This is impressive.

Super rare. These have always been great businesses. They got in a little bit of legal trouble and had to get shut down, but they're way better now. Here's the chart for earnings distortion, the trend of earnings distortion from 2020 through the TTM. It explains what's on each axis, which is great. And then, yeah, PXED. So what was the question here on PXED? Can you do analysis? Oh, this is for the podcast. Yeah, there's PXED. So give you the answer there, John. Yeah, it looks pretty good. Yeah, these were businesses that I remember we were really bullish on and we got caught a little flat-footed because they, they got into some trouble. Let's see, ATGE, not like, not quite as good. Phoenix Education Partners, return to coverage. Yeah, pre-IPO, new forecast.

It looks like they may have just started trading yesterday.

Oh, okay. Well, it's been in the system for a little while, maybe since September 5th. I don't know. But this is a— this was a very interesting opportunity for sure. All right, let's show me the ROIC for PXED and PAPL over the last 5 years.

All right.

If you're going to be a beta client, you got to be able to say, okay, I didn't get the right answer here. Here's what it gave me and here's what it ought to give me. And we'll put that into a special Google Sheet that will track these kinds of things for you. PXCD, not in the dataset. All right, that makes sense. Let's see. Please chart ROIC for AAPL and AVGO. Separately over the last 5 years. Let's see if that works. But the cool thing here is that they're giving the data, they're giving you the period end date. The other cool thing too is you can just go straight into an Excel sheet. That's a really nice feature that everything can be downloaded into an Excel sheet so you can go out and do some work. We're going to also add the ability to automatically create models as well, and we will also roll out a version of our subscriptions that will include access to this AI. So for example, we're going to do Stock Tracker 50 Plus is something we're thinking about. Make sure you give us feedback on that. But then there's going to be Stock Tracker 50 Plus AI where you'll be able to do this kind of AI interface on the stocks in your portfolio. So it'll be just for those 50 stocks that you track in your portfolio. For Pro clients, it's going to be a lot more access because you can have a lot more portfolios and A lot more individual stock data.

Yes.

Ease of access. That's really the benefit of this AI. And that's what we told Google. Part of the reason you should work with us is that this is an interface option. The data is paramount, like we were just talking about. Right? Thinking small drives big leaps. If you got great data, well, then this is just an interface problem. What you cannot solve with the chat interface is the quality of the data. You can't make the data that's out there better just by putting it into an AI. Models are only as good as the data that's put into them. Garbage in, garbage out. That's the whole concept of the SLM versus LLM. If you think just by pouring the internet into the LLM, it's going to figure things out, waste of time. If you've got a good dataset, well, let's put it into a format like this that makes it easier to use. So another demo here shows you which companies are most likely to miss earnings, shows you the returns on invested capital. show you the best tech stocks in the business and, and best stocks in any sector, best stocks in the market. Can do all that too. That's all we got on the demo here, Kyle. I'm gonna kick it back to you to, to finish this off.

Yeah, that was all we had scheduled for today. We thank you guys for being here. We'll put a replay of this up in the community along with the slides. You get any links in there to all the research, the other posts in the community, as well as, you know, the ability to sign up for community and everything like that. And we will see you guys, if not next month and other trainings in between, but for sure next month on November 14th for our next monthly podcast. So thanks everyone for being here.

Thanks guys. See ya.