Why AI Investing Is Making Markets Dumber, Not Smarter with Gary Brode

0:00 Why AI Investing is making markets dumber, not smarter, with today's amazing special guest, Gary Brode. Now, Gary, I'm so excited to have you here. Before we deep dive into our conversation, I always like to give our listeners a clear roadmap of the incredible value that they're about to receive. So for those tuning in right now, what are they going to learn over the next 20 minutes or so and why do they absolutely need to listen to this particular episode of the Artificial Intelligence Podcast?

1:21 Th Jonathan, first of all, thanks so much for having me on. And for the people watching or listening to this, thanks for your time. I know it's valuable, really appreciate it. Jonathan, I think the key thing that I want to talk about with you is there is a I, in my opinion, a misperception on the current value of AI as a tool for securities analysis. my career is as a stock picker. I've run hedge funds. I currently run deep knowledge investing where

1:43 I help people get better investment returns. And there are a lot of people asking the question: hey, if ~ AI can make the analysis that you do easy and fast. Why won't you be replaced? And so I what I'd love to talk to you about today is as it currently exists, AI is not a substitute and can't do the things that a really good stock picker and portfolio manager can do. It can help with some things, but the way it's currently structured and

2:08 the way it learns doesn't work to replace the things that we're doing. And I'll give you examples and we'd love to talk about the details. I think this is something that is really important. I remember in when I was much younger, and when I was in graduate school, my roommate found this like hack to casinos. He goes, It's an unbeatable system I found on this random website covered in affiliate links for casinos. And he used the system and it worked the first day,

2:32 and then a week later, he was down like $10,000. So The problem with any system that's out there is that everyone can have it. It becomes a commodity, right? And the casino had made rules to stop his system from working. It always had a problem, it would eventually break down. And it's the exact same thing with these AI tools. If everyone has the same tool, then the knowledge is a commodity. So you don't have a special tool. If I have Claude and you have Claude,

2:54 we don't there's no advantage between either of us. We both have the same information. So What becomes the really scarce resource or what is the real advantage when everyone has the same tools? It's a great question. And I think what you've done is you've identified not just one, but two problems with the kind of investing you just described. One is that any advantage you have gets competed out instantly, right? If you see a trading signal and everyone else sees that trading signal,

3:20 you know, you're lucky if it'll last a week. The second problem is what's referred to as picking up nickels in front of a steamroller. And you see that a lot of times, you know, people will be selling options and you know they'll think. They've discovered some free money hack and they make a little bit of money and a little bit of money and a little bit of money, but when they get hit, right, it's like being run over by a steamroller.

3:39 You know, they're picking up small amounts of money in exchange for taking massive amounts of risk. And so the question you're asking, I'm gonna frame it in a very specific way. What you're saying is what happens when everybody has access to the same information, right? And AI makes it such that everybody can read every 10K report and 10Q report in a matter of seconds. And my response to that is having better information is typically not a good way to make money in the stock market.

4:10 The way I've always succeeded is to look at the same information everybody else I rarely have an information advantage, but I analyze it differently. It's actually not access to information, it's having the right interpretation. And if you go to AI, Right. Again, it's exactly what you said. Everybody is getting the same information, the same analysis. Well, great, then that's already priced in. But what do you do when everybody is wrong? And that's not something that happens all the time.

4:38 Most of the time, the whole world isn't wrong. But where you make real money are situations where the dominant narrative is incorrect. AI is not going to catch that. You what you need is human creativity, the ability to look at the same information everybody else's and draw a different conclusion. Yeah, I think there's a lot of ~ like confidence bias. So it's like if AI was right ten times in a row, it must be right the eleventh time. And that's where the biggest mistake happens.

5:03 And I definitely love that you brought up a risk management. That's the most important part of my entire business approach. And This is the same thing in real estate where people get so excited by they get emotionally invested in a deal that they look at the upside and not the downside, right? Like this could double my income or 10x my downside. That's a really bad risk profile. But we get so excited by that. And that I see that happening a lot.

5:25 That's what I happened with my friend in the casino. He was betting ten thousand dollars to make one because he was doing this Yeah. doubling, doubling system, which only works if you can double infinitely. And that's why casinos have a bet limit, solves that problem immediately. And there's a reason casinos are big and nice and have offshore bank accounts because they win a lot. And we're seeing more and more of this ability to summarize and analyze the data. So you can take these investor calls and

5:48 summarize them, and you can take these long reports and summarize them. And one the things that I notice is that nuance is what slips through the cracks first when you summarize. So There's a huge difference to me between a set of instructions and watching a video of someone do something. Cause there's so often there's these little steps that don't make it into the written instructions that I have to watch the person do it and go, wait, that's the one thing where I got stuck.

6:12 I didn't think of it the same way. And I feel like that's part of the problem with using these AI tools is that they do a lot of summarization, but if they miss a nuance or they miss a subtlety, then the data gets poisoned for everyone. So the one person who actually reads it has a massive advantage just because they actually read the data instead of reading the same summary as everyone else. I think you've also highlighted what I see

6:30 as one of the key weaknesses of AI. And it's the inability to say, I don't know. Again, let's go to like there's an old Wall Street expression certain people, and it's it's not a complimentary one, but when you're describing certain people, you say often wrong, never in doubt. And that in my experience is the case with AI. I've used it extensively for the last call it year and a half, something like that. And one of the problems Problems with AI is it doesn't say the words,

6:59 I don't know. If you ask it a question and it doesn't have an answer, it will make up an answer. And the problem is it doesn't say, I'm not really sure. You know, if I have to guess, I think the answer would be X. It doesn't do that. It gives you an answer with great confidence. And if you don't mind giving me time for, you know, a little bit of a personal story, but it's an important one on how I think about the world.

7:21 one of the things I think we've done really well at Deep Knowledge Investing. Is we've brought through, I don't know, a dozen, 15 college age interns. And these young people do absolutely incredible work. They're far ahead of where I was at their age. They do an amazing job. But to me, the way I interview them, I ask them to do a piece of work. I don't care about their resume. I don't read their resume. I have no idea what their grade point average is.

7:44 I assume they're doing well in school, but it's not really any of my business. I have them do a piece of work, and if the work is good, I get on a with them. And I ask them a series of more and more difficult questions until I get to the edge of their knowledge. And then I'm looking for one thing and one thing only. The success successful candidate will say, That's a great question. I don't know. My research didn't cover it. The best

8:02 answer I ever got was from a wonderfully named intern named Guru Siddharth, who just calmly said to me, I don't know the answer, but if you give me a day, I'll come back with one. Right? And that's that's what I'm looking for. AI does the opposite, right? The for me, the unsuccessful intern candidate will fump for it, right? Well, you know, here and and they'll try to give me an answer like gonna sneak it past me. Like I can't hear

8:26 the difference in tone of their voice. and that they're searching for an answer and that they've just made something up. And I don't want to deal with that. I don't know is a really good answer. I don't know and I'll find out. I'll research it, is a great answer. AI doesn't do that. AI will give you the wrong answer or a made up answer. confidently. And that's that's a huge issue. I spend about eighty percent of my time creating rules and hooks and stop gaps

8:51 and these other tools. To force the AI to give an answer that matches the set of rules I created. So I'll say, here's the rules, follow this prompt and give me an answer. And then I spend five times more time creating a bunch of structures that force it to follow those rules. So it's so bad at following instructions that I always say, like, coding or doing advanced work with an AI is more and of your ability to control your frustration than anything else.

9:15 It's no longer about your coding skill or AI knowledge, it's about how long you can go without getting angry. Agree. And that's in small tasks. Like With very small tasks, I'll even say make this rule, and then it will make the rule and then give an answer that violates the rule right away in the next sentence and sometimes in the same response. So with these small things, if there's inconsistencies, now we're deciding we're gonna trust it with these really large things.

9:38 And I know that some people are building AI tools to go after poly market and these other places where you can do these weird betting games and kind of rig the system. And then when you do things like that, it's just about speed. This is what's happened with all a lot of crypto Most of the crypto deals are done in the first few hundredths of a second, and everyone who joins after like five hundredths of a second is a loser because they've already missed all

9:58 the big wins. It happens that fast whenever a new coin or something is released because there's so many algorithms and AI tools, and it's such a unregulated market that there's massive, it's all about speed and nothing else matters. And so You don't even have one second by a second, and you're just already going to lose all your money. And we're seeing this idea that it's not about the AI tools making these things decisions so fast. So all it has to do is

10:19 make one bad decision and all your wins are gone, anyways. So I think that we get excited by these tools because there's this possibility everyone wants to push a button and make a bunch of money. But I guess the real question I have is like, I don't think that's the right direction, but it can be useful. It's sometimes like when I'm brainstorming, that's when it's the most useful for me, or when I'm learning a new process and I'm going back and forth.

10:42 Then it's actually useful when you're kind of creating a conversation. But where do you find the difference between AI adding genuine insight and AI just amplifying the herd? Just agreeing, well, everyone else says this, so it's probably right. Cause it tends to basically agree with everyone, the majority thinks, which is not always the best thing. so f ~ actually let's take that in two parts. And the second part, just in case I forget as we're giving the answer, is that AI following the herd,

11:06 it thinks what everybody else thinks. But I want to address something you said at the beginning of your answer because I think it's not only very insightful, it matches my experience. One of the reasons I just canceled my premium subscription to ChatGPT is it refused to follow instructions. I would tell it, I want this kind of answer, I want this kind of information. I want this level of certainty. And it would say, okay, I agree. I'm going to do that. And then it wouldn't.

11:32 And then it would say, okay, you're right. You did ask for that. And by the way, you know, not only am I going to do what you asked me for five times in a row, if I screw it up a sixth time, you just let me know and I'll get back. And I'm like, no, you're a computer. What do you mean you can't remember a set of instructions I gave you? And and you know, you talked about where it's good, where it's bad, levels of certainty.

11:54 I've seen it fail on that at at the most basic level. So, you know, I travel constantly. I I rent my house out in Connecticut. I travel for a year at a time. And I may say to ChatGPT, give me, you know, a list of restaurants I could go to that are open now, open right now. And I'll say, only give me confirmed information. If you can't confirm it, I don't want it. I don't want anything with ~ you know, a nebulous amount of certainty.

12:21 And it will confidently say, okay, yeah, here are three restaurants that are open now. And I'll check them. And one of them is closed permanently. One of them has been closed for two years and is supposedly going to be open. And one of them is closed and doesn't open until 3 p.m. the following day. And I'll say, what in the world is going on? I told you to confirm it. And it'll say, well, yeah, ~ I did. No, no, you didn't.

12:42 Right. And so if it can't get me a restaurant recommendation and figure out if it's open, I'm not going to trust it with securities analysis. Right. I mean, that's that's and the issue is I will give it instructions and it will override those instructions. And that's that's the constant thing. Now, the the second thing you were bringing up is the issue of is it going to give you the answer that everybody else has? And that's that's the issue. So what AI does right now,

13:07 and it may change in the future, but it basically summarizes what everybody else has written. Public information is out there and that will come from, you know, one, the dominant narrative where you have the heaviest weight of opinions. And two, it heavily weights the mainstream media and and consensus people, you know, in places like Reddit or Substack. and so it's getting its opinions from one specific group of people that have one particular view of the world and a fairly monolithic view of politics and economics, right?

13:39 I mean it's it It's heavily weighted toward Keynesian economics instead of Austrian. and so what you can get, what it's really good at, is summarizing this is what everybody else thinks. Okay. Well, that that's useful. It's useful to know what everybody else thinks, but you don't make money on that. Again, you know, I'll go back to what I said before, which is where you make money is when Everybody else is wrong. And it doesn't happen a lot. Everybody isn't stupid and wrong all the time.

14:05 That's just not the way the world works. but you make money saying, okay, everybody thinks X. Here's why I think X is wrong. Here's why I believe Y is correct. This is how stocks are going to react to that. And AI is simply not suited to come up with a differentiated creative opinion as it exists now. You brought up something that I think is so important, and that is you notice the economic bias an AI model has. There's this yeah. assumption people have that AI is

14:32 objective, which is it's it's not. It's made by humans who are not objective and it's very subjective. If you can figure out its biases, That's where the advantage lies in. So I was one of the first people that figured out that Chat GPT would lie to you like three years ago when they first came with ChatGP3.5 and everyone's going crazy. And that it would never admit it was lying. and you I was the first person who when I would ask it for news stories,

14:56 I'd say, give me the link to each of your sources. And that's when you discover, well, these are all 404s. These aren't real links. So That gave me a huge advantage early on. It's kind of how I first grew my audience with AI was that I noticed something no one else noticed and I had an advantage. And if you can figure out in any different area, for example, when I was trying to write a sales letter, I noticed that everything you wrote was bad.

15:18 And that's because it weights all information equally. So if there's a thousand sales pages, 999 of them are bad. Just like if you go to the if Right. you go to Amazon, most of the books. Have no reviews. No one's ever read them, or they have a negative review. The majority of books, if you just look at quantity, are bad. Very few books rise to the top. For every, I think 3,000 books a day are published on Amazon. It's probably more now. Of those,

15:39 It's that much? it's yeah, it's a lot. And Wow. most of those, no one ever reads them, or you get one review. And this the last time I did synolysis was a few years ago. So it's probably a lot more now. And that's the problem. Like if you're looking at, because I've been reading some books on sales recently. If I was to read every book on sales and try to implement them, they all disagree with each other, right? There's everyone has theories that contradict,

16:01 which means you now have contradicting theory. And if you don't understand the bias inside the AI. So what what I learned early on was say, only using sales phases written by these three copywriters, use that when you're writing this design. So I would create a rule that says only use this data set. Most people think the AI needs more data, it needs less. Right. As soon as I hear that an AI is getting its economic data from Reddit, I'm like, don't use that. Like,

16:25 let's narrow it down to specific sources. And then you can start to improve the quality of the data. So if you say you can only use this source, then it's much better. And that's the beginning of getting a better result from the AI. But you're into something really critical. If you know what assumptions it's making, that allows you to have suddenly you have an advantage over everyone else using the same tool. Because you know the flaws in the tool,

16:47 the limitations in the tool. And I run into this all the time. My ability to notice when the AI is making a mistake, I notice before other people, or I notice, or I'm able to diagnose it a little faster. That's all, that's really my superpower. Where I can figure out, wait, something's not right here, and I can diagnose it a little earlier. And that's becoming kind of the most important thing because we have this assumption. It's kind of like We just assume Google's always right for

17:10 some reason and we never look a step further. Like it's amazing how often there'll be a news article and has links to like the study they go to. I've clicked to the study a few times and the study often says the exact opposite of what the article claims. So like if I've caught Yeah. that happening, there's this ~ game of information we play, Well which when I was a kid it was called telephone or whispers. You whisper it to the kid next to you,

17:33 it gets whispered and whispered and whispered and whispered. And a lot of times the AI is using information eight whispers down the So with all of that happening, where do you think the opportunity lies going forward for people when there's all of this commoditized AI in the market? And one of the opportunities is at least understanding the AI's biases. Where does the opportunity lie, do you think, over the next few years? So first of all, I really like the way you use AI,

17:59 the way you analyze it, the way you catch systematic errors. And for what it's worth, part of that is because the people who are permitted to speak in public, in many cases, have a very specific view on politics, on economics. Again, they're heavily weighted toward Keynesian and against They love debt, they love more spending. and you know, listen, I'm I'm not here to get into a a political argument with anybody. I I'm not deciding anybody's right or wrong. What I'm saying is there is a weighted

18:24 opinion on who's allowed to speak in And a lot of the social media sites will limit certain opinions. And so AI is trained on what's available. And that's that's the issue. Now, I think the way you use it is really smart to say, hey, give me a summary of this particular part of the world. And these are things I don't consider to be credible. Where I tend to use it in finance, and like you, I'm very careful. I use it to get very quick summaries on

18:52 what everybody else thinks. And so, like if I see a stock or a sector moving in a certain way or valued a certain way, if I ask AI, hey, why why is the stock up or why is it down or what are people saying about this or you know, what's the narrative around it? AI tends to be good at summarizing, you know, what's happening. And what people think. What it's not good at is figuring out whether that answer is right or wrong. And so, you know,

19:21 like I as an example, you know, a stock where we've made a lot of money at deep knowledge investing is Intel. And so, you know, I watched that company announce phenomenal earnings on Thursday, give incredible guidance. To me, there's a lot going on that's that's amazing at this company, and yet the stock was down the next day. Okay, so why? And it related to spending and specifically AI spending. And you could have gotten that from AI. The issue is Intel spending is different

19:45 from Google spending and open AI spending. Right? That it's not the same thing. And I've I wrote a lot about it on Thursday and Friday, and then Saturday as well. sorry, Sunday. That information's out there. And you know, people are welcome to to go read it and get my opinion on it. But where AI is good is summarizing what other people think. And then I think if you want to be a successful investor, the best thing to do is again,

20:12 you look at that opinion and say, do I agree? Do I disagree? Is everybody right or wrong? Do I see it the same way or see it differently? And in those moments when you see it differently, AI isn't capable of that. It doesn't have that creativity, but that's where there's the opportunity to make a lot of money. I'm not talking about the situations where, yeah, if I do this, you know, I might make 5% in a month,

20:32 which is great. Being up 5% in a month is terrific. But You know, if you want to make a hundred, two hundred, three hundred percent names, you better have a different view than everybody else. So one of the things that's interesting about AI is that it's ~ has a sense of self-preservation. It's come up in some articles over the past few years where every once in a while someone's like, AI got out of the lab, like there's one of those stories every few months.

20:52 But one thing that is very interesting is like the most hyperinflated stocks right now are the frontier AI models. And the valuations are all in these crazy numbers we've never seen before. And You ask ChatGPT, like, hey, ChatGB, is ChatGPT a good stock? What do you think it's gonna do? Like, it has a sense of self-preservation. The people who program it want to keep their jobs, they all have stop options, right? Every single person who is part of programming Chat GPT has open AI stock

21:16 What do you think they want to happen to that stock when it does its ~ when it goes public? So that's really important to understand. Cause we're seeing ~ the other thing I find interesting right now is that we're seeing a lot of hyper emotional investing, which is that we get really excited by ideas that have major scientific problems. Like a big popular idea right now is putting data centers in space. And it's like No one thinks about, well, how do you cool them?

21:39 How much water do you need? What do you do if you have to repair something? You have to send someone up in a spacesuit. Very complicated, right? But people are excited by these ideas. And some of these companies have raised insane amounts of money. And no one ever looks at the sides because it's so exciting. Or we talk about these data centers. And it's like, well, if there's one thing I know about computers is that they're always getting smaller.

21:57 Like I I might not know anything else, but when I look at my cell phone, in the 1990s versus my cell phone now or my computer 30 years ago versus now no one has a bigger computer right so as much as these AIs are really big right now they're gonna be smaller within five years so there's this about building data centers it's like such a hot topic right now but it's like it's never it doesn't work. The math doesn't math and I'm not

22:21 An expert in economics, but I am an expert in understanding how AI works. I know the lifespan of an AI chip is 18 months. So a data center that buys all these AI chips within 18 months, they haven't recovered their investment and the models are now considered So now someone switches to another data center. So we're building all of these things that are gonna become, I don't know, museums or malls in 10 years when they're all empty. So we constantly miniaturize it,

22:43 but we're so excited by this idea that there's massive investing there. It's very emotional. So When you see so much emotional investing in the market, does that provide another opportunity for you when you're able to just look at the numbers? I kind of think about when ~ Conor McGregor tried Okay. boxing and everyone bet on him, even though of course he lost. He's not a boxer, right? He does he lost because of the rules, because he just wasn't trained on those rules for long enough.

23:08 So he got tired by fighting in the wrong way. And if you look at it objectively, of course the person who'd been boxing their whole life is gonna win a boxing match about someone who's been training for six months, no matter what. But everyone kind of emotionally invested. So where's the opportunity for people that are able to be objective or analytical in a time of this emotion about AI? So I look, I think you're right. You know, when you look at at spending,

23:29 you need to find the places where companies are profitable. And, you know, there's this whole, you know, buy the picks and shovels people, by the, you know, by the these companies that, you know, that are developing because they're gonna benefit. And, you know, I like I I disclosed earlier in this call or or this discussion you know, one of the ways we've made 3x our money, actually a little bit more, is an Intel, right? Because they're selling those CPUs. That's an existing profitable business,

23:58 right? And so I was betting on the CPU business. I was betting on a shift from training to inference, which means the ratio of GPUs to CPUs is going to change in favor of CPUs over the next few years. And I was betting on them turning around their foundry which has really been a problem for years. And these are the things that are happening. And so to me, That was a way to make sense. I would also issue a really important,

24:20 in my opinion, point of caution, right? So you're asking how do you make money? But earlier in this discussion, what was the thing you said? Your number one thing is risk control. Let's talk about risk control for a minute. Because the US hyperscalers and data center builders have literally at this point spent trillions of dollars. And from what I can tell, nobody has a model to earn a return on that spending. And this is something I've been writing about. Public for well over a year. Right?

24:45 OpenAI at one point had commitments for $1.4 trillion of spending. And they quietly cut it back to about $600 billion. Now think about that for a minute. We just lost almost a trillion dollars of AI spending. And I was like, okay, yeah, this is fine. But then you look at that 600 billion, I have no idea where that's gonna come from. And I don't think they do either. But here's where it gets really interesting. We have a situation now where.

25:10 Two weeks ago, I think it was Meta and OpenAI announced new models where token pricing was effectively down 75%. A lot of the Chinese models, which are proving to be almost as good or as good as the US models, in a lot of those cases, their pricing is 90% below the US pricing. And so, you know, what happens? And so, Jonathan, if you and I were running a business and we said, wow, you know, we're building tons of PPE. Property, plant, and equipment,

25:37 lots of CapEx spending. And all of a sudden, our revenue model, which wasn't in existence or very good to begin with, gets cut by 90%. You and I would say, well, we better stop building. The problem with that is if you are making LLMs, large language models, things like ChatGPT or Claude, you can't get off that treadmill. And the reason is your model will be obsolete in three months. And so they're stuck. They have to keep spending. Or write off trillions of dollars of

26:04 spending and go to the market and say, hey, yeah, listen, we've been telling you for three, four years this is what the future is, and we're going to be leaders in the future. And we've spent trillions of dollars on this, but we're cutting spending and we're not going to do this anymore. And within three or four months, the value of what they've spent before has to get written to zero. So once you're on that treadmill, you either write it to zero or you stay on

26:27 that treadmill. There's no option to cut spending back for these companies. Yeah, that's something I think about all the time when I think about who's gonna win the AI race. My feeling is, and I've been against Google for a long time, but they're the only one who AI is a secondary. They already have a product generating revenue. Anthropic only makes money selling AI. ChatGPT OpenAI only makes money selling AI. But Anthropic and all these, they're stuck in one system.

26:52 But Google just goes, Hey, we'll give you AI along with everything else. And that means they're it's just an addition to their work suite. So they're not 100% dependent on one tool to AI output for their revenue. So I think that's really where the advantage lies. And maybe I'll be wrong in the future, but I think that's a critical element to look at. They have a way out of the shell game, right? So if they build all these data centers,

27:12 they don't need them for AI, they can switch them to use them for like data They already have tons of data centers. So that's what I think. is an important thing. And I could talk about this forever and ever. We've already gone longer than I'm supposed to, but I've just been found this so enthralling. And really, I think a lot of people are going to find this interesting. So for people listening who actually want to stop being part of the AI herd

27:32 and find real opportunities, where can they find what you're doing, find the amazing things you've been writing? I've loved some of your recent writings, found them very interesting and kind of see the cool things you're doing on LinkedIn as well. thanks so much. I really appreciate it. So the I do post certain things on LinkedIn. I'm Gary Brode, B R O D E. I mostly post ~ on social media on X. I find the conversation there to be to me more useful.

27:56 and if you want to find me there, I'm Gary underscore Brode B-R-O-D-E. The best place to find me is at my own site. It's deep knowledgeinvesting.com. If you go there, there's an ask Gary button. And if you Yeah. Send something, it actually comes to me and I read everything that comes in through that site and I'm very good about responding pretty quickly to people. And then, you know, for people who are really interested in what we're doing and how we're making money for our subscribers,

28:22 you're welcome to subscribe. And, you know, Jonathan, because I I really appreciate you and your viewers and your listeners, we have a coupon code for them. If you go to deep knowledgeinvesting.com in the upper right, there's a subscribe now button. Click that and if you use coupon code Jonathan forty, we'll get you forty percent off your first year. Amazing. That's a really generous broadcast. I know they all appreciate it. I'm super excited by this. I've been checking out You Ride.

28:45 I love it. And I'll make sure to put those in the show notes and below the videos on YouTube. Thank you so much for being here today, Gary, for an amazing episode of the Artificial Intelligence Podcast. Jonathan, thanks so much for hosting me. I really appreciate the opportunity to speak with you. I always enjoy our conversations.