Transcript: Can Gemini 1.5 Pro Beat Our Best Stock Trade?

The transcript of How Do You Use ChatGPT? with Jesse Beyroutey is below for paying subscribers.

Timestamps

1. Introduction: 01:29
2. How Dan made the greatest trade of his life: 03:50
3. Jesse’s strategy to use LLMs to get nuanced answers 05:27
4. Gearing up to orchestrate the best trade of our lives with Gemini Pro 1.5 09:20
5. How Jesse gets AI to make great decisions 17:52
6. Using Gemini Pro 1.5 to find patterns in data 22:38
7. How AI can provide deeper insights into the stock market 26:48
8. Leveraging Gemini Pro 1.5’s huge context window to analyze data 34:41 
9. Gemini Pro 1.5 and ChatGPT go head-to-head 46:33 
10. Choosing a stock with just 15 minutes left on the clock 1:10:11
11. What Jesse thinks are the biggest new opportunities enabled by LLMs 1:24:01
12. The epilogue Jesse and Dan recorded one week after making the trade 1:28:43

Transcript

Dan Shipper (00:00:00)

When we bought Nvidia in January 2019, it was $33 a share. Today, it is $800 a share. That was the trade of our lives. We want to find an investment that's just as good as our Nvidia investment using Gemini. And I think we can do it. Basically what we have done is we have downloaded earnings call transcripts from 2022 and 2023.

We're going to load them into Gemini and we'll ask some questions, see how it does. This is the stock that we came to at the end of a long journey, and I'll let you see it as I do it. We're going to swipe up to submit. Order received. So there we go.

Hey, I want to just take a quick break from this episode to tell you about a podcast you should know about. It’s hosted by Nathan Labenz, who was actually on this show and did one of the most popular episodes we’ve ever had about how to use AI as a copilot for the mind.

Nathan hosts a podcast called The Cognitive Revolution, where he interviews builders at the edge of AI and discusses the profound shifts that it will unlock over the next decade. He talks to the researchers and entrepreneurs who are shaping this moment, like Logan Kilpatrick of OpenAI, who I interviewed for this show as well, Eugenia Kuyda of Replika, and Suhail Doshi of Playground.

To listen to the show, search for Cognitive Revolution in your podcast player. Alright, now let's get back to How Do You Use ChatGPT?. 

Jesse, welcome to the show. 

Jesse Beyroutey (00:01:32)

Thanks, Dan. I am just delighted to do this with you.

Dan Shipper (00:00:36)

For people who don't know, you are one of my closest friends, and you are also a partner at IA Ventures. IA is a $600 million venture fund, and you've done seed investments in public companies like Wise and DigitalOcean. And you're both, I think, one of the smartest people I've ever met. You really have this way of digging down into an issue and kind of figuring out the crux of it. And you're just a very, very good friend and very good person. And I’m just so lucky to have you in my life. And I'm so happy to be able to do this with you. Yeah. Thanks for coming on.

Jesse Beyroutey (00:02:22)

I didn't know that I was going to get this completely overwhelming, almost make-me-cry-level of introduction. But thank you. I also want to say thanks for getting me out of my natural ostrich-like tendency to not appear in anything or be on the internet. You always managed to stretch me in ways that have ultimately been so additive and productive in my life. The last time was helping me get one of my first things out onto the internet that people regularly reference back to me like, oh, I saw you on every, so I'm really excited about this too.

Dan Shipper (00:02:59)

Great. I get to hang out with you and hear all your smart thoughts all the time. And I'm just trying to bring more Jesse into the world. So this is our latest attempt and you are coming on to talk about Gemini Pro 1.5, the new model from Google that has a million-token context window, which is sort of wild. And our goal is to put it through its paces and try to see what we can do with it. And the concept that we came up with, I think it's gonna be really fun. And I'm just sharing my screen, because, just to set it up, what you realize is it would be really cool to do a good trade with Gemini and see if we can get it to identify a stock to pick and it harkens back to, I think, probably the best trade, at least, I've ever done. I don't know about you. You're a professional investor.

Jesse Beyroutey (00:03:55)

It's mine too.

Dan Shipper (00:03:56)

Yeah. So, in 2019, we created a little Notion doc called Beyroutey x Shipper Investments because we wanted to take a bet on a specific company and the company that we wanted to bet on was Nvidia. And when we bought Nvidia in January 2019, it was $33 a share. Today, it is $800 a share and we wrote down our investment thesis. We said the price is down because of the trade war, but that's only a temporary risk and that Nvidia is in a position of power because of its proprietary GPU architecture, and its position in the market should only grow as machine learning and gaming increases the need for high powered GPUs and computers, so the price dip should be temporary. So, that was the trade of our lives. 

Jesse Beyroutey (00:04:50)

It was quite a moment. I mean, we fancied ourselves as distressed-asset investors in our little long-short public market fund. I think it’s important to note that we also made one of our worst investments ever at about the same time. So, it's not like we were absolute geniuses, and we didn't have LLMs at the time. So, maybe today it’ll be different. 

Dan Shipper (00:05:15)

Yeah. So the concept for the show is we want to find an investment that's just as good as our Nvidia investment using Gemini. And I think we can do it—or I hope we can do it.

Jesse Beyroutey (00:05:24)

Yeah. I'm really excited about this in particular because I love using LLMs for role-playing-type exercises as almost a conversational participant along with me. So my favorite sort of prompts for ChatGPT, and now for Gemini, assume that you are such a famous character. What would you have to say about X? And that elicits a different kind of perspective from the LLM than just asking it a question flat out because it's trying to take advantage of whatever knowledge that it had about that person and things that co-occur with that person's name on the internet. And now, today, being able to do it with like an insane context window means that we can just load in like all of the context around a person and see Gemini can take the perspective of that person as if they're like an analyst on our little hedge fund team.

Dan Shipper (00:06:19)

I love that. I'm curious, what are the names of people that you've used and found to work well so far?

Jesse Beyroutey (00:06:29)

It really depends on the setting. But I mean, I just generally always try to ask like, what would the CEO of a famous company in that industry—so, let's say I'm analyzing something about aggregators, I might ask, what would Jeff Bezos have to say about this? Or what would Ben Thompson have to say about this? And it's kind of the intersection of someone who clearly would have an informed opinion about the thing, probably has talked about it before, has a lot of information about them on the internet already, and therefore there's some meaningful thing that the LLM might be able to draw from. And then also those whose perspective is generally a little bit contrarian or comes at things from a different angle than what the average of the internet might have.

Dan Shipper (00:07:16)

That makes sense. That's really cool. Yeah, I love that. So I want to get us going into our investment-finding exercise because we've got 90 minutes. The clock is ticking. So, set up for us how we're thinking about doing the initial selection of companies. And I think we're basing it on how we thought about Nvidia and we're going to try to see if we can do a similar kind of filter for this round of investment.

Jesse Beyroutey (00:07:41)

Right, exactly. So back when we were doing our little hedge fund in 2019, our idea was we wanted to find companies that had been trading down—suddenly, recently, for some reason that we could look at and think, oh, this distress situation is temporary. And so from our perspective, as a more long-term investor, it gives us the advantage of a margin of safety. Oh, if it's trading down by 20 percent, 30 percent, or whatever, then we can be a little bit wrong about things and still not lose too much money as happened with our worst-case investment that we made around that time, where it took us a little while, but we eventually recovered our principal on it and we're able to live to die another day.

Dan Shipper (00:08:27)

Yeah, totally. And I will say, yes, we went into this with this very long-term thesis, like, oh, it's only down temporarily, but then I panic-sold it during the pandemic. So I look stupid, but I still feel good about identifying it originally, but I'm going to hold this one. This trade is I'm going to keep around. I think we should also say for anyone who's listening, you shouldn't do this. You shouldn't day trade. This is not a good idea. And we're doing it only with an amount of money that we're willing to lose. So this is not investment advice. You should probably buy index funds unless you really know what you're doing. But, anyway—

Jesse Beyroutey (00:09:05)

Treat this as worse than WallStreetBets because it doesn't even have the common thread, and more on that later because we might even post our bet to WallStreetBets. 

Dan Shipper (00:09:13)

We are going to probably post our bet to WallStreetBets. But one thing that I forgot to say is we are going to put actual money down. So I created a Robin Hood account, which is the first time I’ve ever had a Robin Hood account. I put $1,000 in and hopefully, by the end of the episode, we will find you a company to put $1,000 in and we'll be able to track how it does. So yeah, I think you've set up the filter of what we're looking for appropriately. And I know you've been looking through companies to start our research with. Do you want to share your screen and show us what you've been looking at?

Jesse Beyroutey (00:09:47)

Yeah, let's open it up here. So I just popped open this website TradingView and it gives you a way of screening all of the available stocks against a bunch of filters and criteria. And when I thought about the context of our Nvidia investment we were looking at companies that weren't massive yet, but had the potential to be the tenbagger of the future. And so for me, that was expressed in the form of, it's a company that's between a couple billion dollars and $100 billion in market cap. It could become the next trillion-dollar company but it's starting from small beginnings. I think Nvidia was something in the $50-100 billion range when we first invested and now it's a $2 trillion company today, depending on the trading. So I was looking for companies that were in positive gross profit which we might play around with, but just want companies that actually make a little bit of money. They're not going to die like that. We're looking to avoid our total bozo risk as much as we possibly can. And the way that we're trying to do that is just, okay, the company has to figure out how to make money somehow. And then aside from that, I was just looking for, really, there's so many screens that you can do, but I really just wanted to find companies that had a six-month performance that was like they're down 20 percent or something like that. But then on a five-year basis in this column is also down, but we might come up with a reason for why that was the basic screen, and we were only looking at technology companies because that was what Dan and I thought maybe we'll have something. But I think we should also try to do companies that we know absolutely nothing about later.

Dan Shipper (00:11:40)

Right. Totally. So, I'm looking at this list and the first thing in my mind is like, I guess we could just throw this into Gemini and ask it to filter based on our like Nvidia criteria. But I just want to confirm that that seems like a good idea to you.

Jesse Beyroutey (00:11:54)

I think anything with LLMs is a good idea because we'll see what we get out of it. So let's just do it.

Dan Shipper (00:11:59)

Somehow, you're going to have to get that stock screen over to me so I can throw it into Gemini. Alright. So what we've got here is you copy-pasted your output from, from the trade view.

And I'm going to just head into Gemini. And so, again, this is not ChatGPT. This is Gemini 1.5 Pro, which is this new model. It's a private beta. And what we're going to do is we're going to take the output that you gave me, and we're going to try to filter it by our criteria. Let's say something like, hey, I'm trying to pick a stock. I want to find a tenbagger specifically. I'm trying to replicate a trade I did in 2019 where I bought Nvidia. It was down temporarily because of the trade war and now it's a $2 trillion company. Here's a list of potential starting points for trades. Does anything pop out to you? Any revisions there? Does that seem like a good time? Okay, cool. So I'm going to paste it in and hit go, see what happens. One of the things I've noticed about Gemini is it's slower than ChatGPT, so it can take a million tokens, but you have to wait—I don't know. Sometimes it takes a minute for it to actually return, which is—

Jesse Beyroutey (00:13:53)

I mean, it's because our request just got queued behind all the things that make Google money to run on their TPUs or whatever, so I get it.

Dan Shipper (00:14:02)

Okay. So Gemini is saying—good caveat. “It's important to remember that past performance is not indicative of future results and finding the next tenbagger is extremely difficult. While your Nvidia trade was successful, replicating that exact scenario is unlikely.” Thank you, Gemini. Appreciate that, but you're saying there's a chance, 

Jesse Beyroutey (00:14:24)

Don't give up all hope.

Dan Shipper (00:14:25)

“However, I can offer some insights on the list. You provided potential growth stocks, Bill Holdings Inc.: This company provides cloud invoices and payments. It has shown strong revenue growth and has the potential to benefit from the increasing adoption of visual payment solutions. Klaviyo: This company offers an email marketing platform for e-commerce businesses and has a strong track record of growth and is well positioned in a growing market. Appian: This company provides a low-code platform for building business applications to demand for low-code solutions is increasing, which could benefit Appian. Important considerations. High-growth potential: look for companies with strong growth valuation. While high-growth companies often have high valuations is important to ensure the price is justified.” Okay, so it identified Bill Holdings, Klaviyo, and Appian. I think we should strike Appian because I just feel like low-code is just like such a bad idea in an AI market. But I'm curious if you have a differing opinion and or if any of its other suggestions pop out to you as something we should explore.

Jesse Beyroutey (00:14:24)

Yeah. So, what immediately stood out to me is that it's already identified that high-growth companies have potentially high valuations, which is obviously true. And here we're specifically looking for, as a starting point, a company that's artificially depressed. And so we want to figure out what reason there might be for the company to be artificially depressed and give us that. That nice feeling that we can make our money back, even if we're totally wrong. So I would actually even—let's try a different angle. The angle might be, instead of asking about those specific stocks, let's ask it for how to construct a screener to identify companies that might be artificially depressed in their prices, because maybe my screener idea is actually not even the right one.

Dan Shipper (00:16:20)

Okay. And should we use the trade view and say in the trade view or—

Jesse Beyroutey (00:16:27)

Can't hurt. I mean, it might be aware of what that is but also even if it's not, stock screener and type filters are pretty generic. So I think it could just give us some ideas.

Dan Shipper (00:16:37)

Okay, cool. “How should I construct a screener to identify companies that might be artificially depressed in a stock screener?” Does that seem like a good place to start? And I will say, these are things that we could do with ChatGPT and hopefully once we narrow things down a little bit, we will start using more of Gemini's increased capacity to analyze data. Alright. So, “Here are some criteria you can use to construct a stock screener price-to-earnings ratio. Look for companies with a P ratio significantly lower than their historical average price-to-book ratio. Similar to PE, a PB ratio is significantly lower than its historical average would be good price-to-cash-flow ratio, debt to equity ratio, free cash flow.” It’s talking about technical indicators like the relative-strength index. I don't really want to use technical indicators. So I don't know if any of those things spark something for you. We could also think about throwing in something like Ben Graham's Intelligent Investor book and see how Ben Graham might evaluate some of these stocks. But, yeah, tell me what you think.

Jesse Beyroutey (00:17:52)

My immediate instinct is to ask it to just make a decision. So every time I ever get some kind of summary from ChatGPT or Gemini, it's always like caveated heavily. Like, you might do this or you might do that. And I'm like, okay, great. I'm just trying to do the best version that I can with five parameters. Like what would you use?

Dan Shipper (00:18:09)

Okay. So, make a decision. Do the best you can. Yeah. Okay. You kind of want to download The Intelligent Investor and just to have. “Based on the information provided, it is impossible to make specific investments choosing a stock—” You know, I think Google's model is going to be pretty safe.

Jesse Beyroutey (00:18:39)

We're running up against some—Google doesn't want to be liable for what we did, which I get it we're two dumb people on WallStreetBets trying to make some kind of new idea here. But I think we should try to get it to be a little bit more explicit.

Dan Shipper (00:18:58)

Yeah. We could say, pretend you are—I don't know—Warren Buffet or something. Yeah, let's try that. And I've come to you to make a final decision on—do you want to have it decide on a trading strategy or on a stock?

Jesse Beyroutey (00:19:11)

I think we should go with a trading strategy to start with.

Dan Shipper (00:19:14)

And this will probably weight it towards Warren Buffett's trading strategy. But I think it’ll get it to make a decision, for sure. 

Jesse Beyroutey (00:19:20)

Does it work with links? 

Dan Shipper (00:19:28)

I don't think so, no. But we can just copy-paste anything that's in a link into—

Jesse Beyroutey (00:19:32)

Does it work with PDFs yet, or no?

Dan Shipper (00:19:34)

It works with PDFs. Okay, got it. Sweet.

It also works with videos. So if you've got any videos or audio—also great.

Jesse Beyroutey (00:19:44)

So my instinct is I think it'd be really fun to have it try to get like. people who run major hedge funds to be our analysts. I love that. So you know, okay, since he's been in the news a lot, I'm trying to think of a good thing to feed in for Bill Ackman. Let's have, let's have, let's get Bill's opinion on how to find a good distressed stock right now. 

Dan Shipper (00:20:08)

Warren Buffett in Gemini says, well, “I don't advocate for frequent trading. If you insist on starting a strategy, consider a value-oriented approach, look for undervalued companies using metrics like P/E, P/B, and P/CF, look for companies with strong fundamentals and durable competitive advantage. So these are all things that we've already kind of considered, but I think a low P/E filter is an interesting one to consider. 

Jesse Beyroutey (00:20:34)

Yeah. So I'm just going to try a filter. So I broadened our industry here and I'm going to try a price-to-earnings filter.

Dan Shipper (00:20:44)

Our simulation of Warren Buffett asked us to filter our list by price-to-earnings ratio below a certain number—we picked 15. And so these are the companies that are in that bucket, and these are the companies that have a fairly low price-to-earnings ratio that might be depressed. I have not filtered this by tech companies and maybe that's purposeful. I just want to make sure you're good with that.

Jesse Beyroutey (00:21:16)

Yeah, I think we can do a broad screen and just have it tell us what to do, like an analyst would.

Dan Shipper (00:21:26)

Perfect. One thing that I'm feeling is we don't have real-time data in Gemini. So asking it about, what is currently depressed might not work that well.

Jesse Beyroutey (00:21:38)

Well, we're going to use the stock screener to figure out what's currently depressed. So the way that I was thinking about doing that was—so, if you filter it by the word performance, they have ways of screening by six-month performance, et cetera. And so what I was thinking is, okay, we want something that's below minus 20 on the six-month and then on a five-year basis, maybe we want something that is at or above zero.

Dan Shipper (00:22:10)

Yeah. I like that because it indicates strength, but temporary dysfunction. That's what we're looking for. My therapist told me that last week, so that's where that came from. So we've got 29 matches and we could look at them individually ourselves. We could throw them into Gemini. What's your call?

Jesse Beyroutey (00:22:38)

Yeah. So I think that now is a good time to throw them into Gemini and ask it to look for points of commonality or something like that.

Dan Shipper (00:22:46)

Okay, cool. Points of commonality. Tell me more about that.

Jesse Beyroutey (00:22:51)

Well, as a starting point I think that language models are amazing at clustering things. So if I were to ask it to group them into some relevant groups that we can consider. If you had to cluster these stocks, how would you do it and why?

Dan Shipper (00:23:07)

Alright. So I'm just going to paste the plain text and see how that goes. Okay, cool. So, okay. So it says, “Clustering stocks involves grouping them based on similarities. Here's some potential approaches in the rationale of clustering by sector, clustering by financial metrics, clustering by technical indicators, clustering by fundamental factors, hybrid approach, choosing the best approach. The best approach for clustering stock depends on your specific goals and strategy for long-term investors focusing on fundamental strategies and financial metrics for short-term investors, technical indicators and price behavior might be more important.”

Jesse Beyroutey (00:23:38)

Obviously those things are true, but I would actually ask it to cluster them by things about the company and what it does so that it's more possible for us to analyze it more on the basis of just the nature of the company and its position in its industry or the nature of the company and its position in the industry.

Dan Shipper (00:24:00)

So I'm just literally typing what you said, and see what it does. Alright, let's try it. And I'm noticing that it's a little bit hesitant for some of these queries to really give us an answer. I'm not sure that ChatGPT would be so hesitant. Alright. Well, let's do a side-by-side. We should do a side-by-side while this is loading. So Gemini said clustering stocks by company, nature, and industry position. So, “Based on the information provided, here's a possible clustering of the stocks based on the nature of the company and its position in the industry cluster. One, commodity-based businesses, companies like Archer-Daniels, Darling Ingredients, Civitas,” blah, blah, blah. “These companies are primarily involved in the extraction, processing or distribution of commodities, like agricultural products.” This immediately makes me think of lithium for batteries or anything related to chips or silicon or whatever seems like an interesting play, but I bet those are already hot. I don't think they're depressed right now, right? 

Jesse Beyroutey (00:24:57)

Yeah, we're going to have a hard time finding anything depressed that's around semiconductors because it's so on thesis right now.

Dan Shipper (00:25:02)

That makes sense. Okay. “These companies are typically price takers with limited control over the market price of their product. Their performance is heavily influenced by supply and demand dynamics and global commodity prices.”

Jesse Beyroutey (00:25:13)

That doesn't sound like our kind of investing.

Dan Shipper (00:25:14)

So consumer-facing companies, Avis Budget Group. I hate Budget. I will not invest. This is like a moral stand we’re taking here.

Jesse Beyroutey (00:25:30)

We're only investing in a company that we actually like.

Dan Shipper (00:25:32)

Or at least one that I haven't heard of, or at least one that I haven't had a personally horrible experience with.

Jesse Beyroutey (00:25:39)

That’s funny because out of the whole list, the one that stood out to me was Gogo. I was thinking about how it's potentially artificially depressed because it's both a brand that people hate. Everyone has the same reaction that you just had to Budget. And also at the same time, there's the buzzy Starlink company that's going to try to go after them. And so I wonder how much that impacts them.

Dan Shipper (00:26:05)

That's an interesting thing to look at. I love that. Okay, cool. Dinmode, which I have no idea what that is. We could look it up. Urban One. So those are the other consumer-facing businesses aside from Gogo. Then there's technology-oriented businesses. So this is our bread and butter right here. Perion Network, Intest Corporation, Universal Security Instruments, and Kimball Electronics. “These companies develop a provider, technology-based products and services, including software.” And then we've got a couple more: In healthcare, DaVita, cross-country healthcare. AMN Healthcare Services. I know you've got some investments in healthcare. I don't know if any of these pop out to you.

Jesse Beyroutey (00:26:43)

Yeah. I mean, I think a lot of people really dislike DaVita. So that that's another one. And in fact, actually what I was thinking of asking Gemini is which of these businesses are most likely to be trading poorly because consumers dislike the brand.

Dan Shipper (00:26:59)

Okay. So, “Which of these businesses are most likely to be trading poorly because consumers dislike the brand?” Because I think what you want to do is just strike them from our list.

Tell me about that though, because if consumers dislike the brand, theoretically, that leaves a lot of room for the competitor to come along that would kill the company, right?

Jesse Beyroutey (00:27:27)

Yes and no. So if consumers dislike a brand, but it has some crazy distribution lock-in so—Gogo being a great example. Maybe Gogo is the default and has hardware lock-in on so many planes that it's not going anywhere anytime soon, even though consumers hate it. And so that maybe is an interesting opportunity.

Dan Shipper (00:27:49)

I get it. Okay. So, “From the provided list, the business is most likely to be trading poorly are… Gogo: Gogo provides in-flight internet and entertainment services. Urban One: Urban One is a media company targeting African-American and urban audiences. And Avis Budget Group: While car rentals are generally more susceptible to economic fluctuation, travel demand, negative experiences, customer service vehicle, or could damage brand perception.”

Jesse Beyroutey (00:28:14)

Alright, I’m getting a feeling here, which is I just think that Gogo is going to be interesting for us to analyze.

Dan Shipper (00:28:18)

I love it. So I want to dig straight in and start talking about Gogo. How do we do that? So I think the key question for Gogo is whether or not their technology can be upgraded.

Jesse Beyroutey (00:28:30)

So it's clear that they have a distribution advantage already.

Dan Shipper (00:28:35)

‘Cause they're already in all the airplanes. 

Jesse Beyroutey (00:28:37)

They're in all the airplanes. And they've also integrated with all of the apps and services that the airlines use to provision the internet and connect your payments and all that sort of stuff. So they're kind of the one to beat. And people hate the service because the service quality is low. So the question is upgradability. And so I think it'd be interesting to start talking through with Gemini, how Gogo might go about.

Dan Shipper (00:29:08)

I love that. I think that's a really interesting place to start. Why didn't you first think about airline switching costs or how hard it is for the airlines to rip them out and move to someone else?

Jesse Beyroutey (00:29:17)

My immediate thought was that if Gogo is incumbent it's the one that they already have a deal with. It’s the one that's already installed. And if there's some clear upgrade path for them, that's going to be the natural first step for the airlines before bidding out, completely replacing all of the hardware. And so if it's possible for there to be a low-cost upgrade path, I think that they have that default. They have them to manage.

Dan Shipper (00:29:44)

Okay. That makes sense. Okay, cool. So should I just ask that?So, “How hard would it be for Gogo to produce hardware that would allow it to upgrade its current services in planes to”—what? To be like Starlink technology? 

Jesse Beyroutey (00:30:09)

Yeah, to improve service quality using the latest in connectivity like Starlink.

Dan Shipper (00:30:15)