DR RICHARD SMITH | From RiskSmith
DR RICHARD SMITH | From RiskSmith
Dr. Richard Smith is the founder and CEO of RiskSmith. He's a renowned mathematician and expert in risk analytics who's helped government agencies and Fortune 500 companies manage complex sets of data. He's the author of the Risk Rituals Newsletter where he aims to educate investors on how to manage their fear of risk. Richard has been experimenting with ChatGPT to analyze and construct portfolios. What do ChatGPT, LLM, and AGI mean, and what are the current misconceptions and realities of artificial intelligence?
And then I said, let's put together a portfolio of ETFs based on Ray Dalio's holy Grail portfolio philosophy. And ChatGPT was able to do that for me, which is pretty extraordinary. I thought it looked pretty darn good, broke it down. US equities, 40%, some large cap, some midcap, some small cap international developed market equities, 15%, emerging markets, 10%. So, this isn't a portfolio that I would invest in. For one thing, I'm not really crazy about ETFs. I actually like to invest in individual companies that I feel more connected to than using ETFs. But as a starting point ChatGPT to put this together after about a 10 minute conversation, I think it's absolutely extraordinary.
TRANSCRIPT FOLLOWS AFTER THIS BRIEF MESSAGE
CLICK ON THE LINK TO SIGN UP FOR A FREE TRIAL OF SHARESIGHT PORTFOLIO TRACKER
Sharesight automatically tracks price, performance and dividends from 240,000+ global stocks, crypto, ETFs and funds. Add cash accounts and property to get the full picture of your portfolio – all in one place.
Disclosure: The links provided are affiliate links. I will be paid a commission if you use this link to make a purchase. You will also usually receive a discount by using these links/coupon codes. I only recommend products and services that I use and trust myself or where I have interviewed and/or met the founders and have assured myself that they’re offering something of value.
Chloe (1s):Stocks for beginners. Phil Muscatello and Fin Pods are authorized reps of Money Sherpa. The information in this podcast is general in nature and doesn't take into account your personal situation.
I like Ray Dalio's philosophy of a whole of the Holy Grail portfolio. And ChatGPT knew what that was. And then I said, let's put together a portfolio of ETFs based on Ray Dalio's Holy Grail portfolio philosophy. And it was able to do that for me, which is pretty extraordinary.
Hi, I am welcome back to Stocks for Beginners. I'm Phil Muscatello. What's the holy grail of portfolios? How can artificial intelligence help you to find it? And what is AdvisorGPT? Joining me to explain ChatGPT and how it can be used to help to define your investment goals is returning guest Dr. Richard Smith. Hello, Richard.
Hello, Phil. Good to be with you. Thanks for having me.
No worries. Thanks for coming back on. Dr. Richard Smith is the founder and CEO of Risk Smith. He's a renowned mathematician and expert in risk analytics who's helped government agencies and Fortune 500 companies manage complex sets of data. He's the author of the Risk Rituals Newsletter where he aims to educate investors on how to manage their fear of risk. Now recently, Richard, you've been experimenting with ChatGPT to analyze your portfolio and to try and construct portfolios. And we're basing this interview on a recent newsletter. Just use it. So let's go to the basic concepts of ChatGPT. And what is it? What's an LLM? What's AGI and how do people currently perceive artificial intelligence?
Richard (1m 42s):
Great question. And the right place to start. What is it? What are we? Well, there's
Phil (1m 48s):
A lot. There's a lot. What are we
Richard (1m 48s):
Dealing with here? Right? Yeah. Yeah. So there's a lot to unpack there and there's a lot of confusion out there, I would say. So what an LLM, well ChatGPTis what most people have heard of. ChatGPT is what's called a large language model. And what that means is that a computer read a lot of different texts, right? Billions of texts, the Bible, the encyclopedia, Wikipedia, whatever you can think of, a computer read that. And then made a statistical model of how words are strung together to create language.
Richard (2m 32s):
Okay? So now it's able to mimic language. Some people would say it can understand language, but that's not the case. It doesn't understand anything. It's a massive, massive model made up of massive inputs and massive amounts of computing power. Yeah. To be able to simulate language. So just
Phil (2m 56s):
Richard, let's just go back from there because I think a lot of people, the way I kind of posed that question was yes, with the understanding that a lot of people don't even understand what ChatGPT is, they're hearing a lot about it. Yes. And the power of it, and think that we're somehow turning into some artificial intelligence supercomputer. But you are trying to hose down that to say that as a large language model, it's just a very, very, very, very smart computer.
Richard (3m 24s):
Yes. And you know, this is what I studied in graduate school. This is what I got my doctorate in. I'm familiar with neural networks. Okay. So it sounds like a fancy term. Something like, ooh, it's a brain, right?
Phil (3m 39s):
A brain in a jar.
Richard (3m 40s):
It's not a brain in a jar. It's a mathematical model. It's statistics. Okay. And it's, but what's really wonderful about it is that it creates the capability for us to interact with our computers with natural language. And that is a huge breakthrough. So think of it like your keyboard, right? So you interact with your computer via your keyboard. You interact with your computer via your mouse. There are underlying protocols and technologies that allow signals from your, your mouse and your keyboard to do things on your computer and get your computer to do things.
Richard (4m 25s):
Think of a spreadsheet. That's another way of interacting with what computers do best. So what I believe ChatGPT is, is a linguistic user interface. It's a way for us to talk to computers. And if you've seen Star Trek from back in the day, you know, and they say, Hey computer, that is where this is all going. And it is extraordinary. And it seems like magic. It seems like intelligence. It seems like the computer's smart, okay, but it's just a machine. Right? Would you say that your computer itself is smart or intelligent just sitting there on your desk?
Richard (5m 5s):
Probably not. But I think we can be smarter and more intelligent by using these new technologies. We can, I, I prefer the term augmented intelligence over artificial intelligence. Cuz I, I think artificial intelligence is a misnomer. I don't think it's intelligence. I think it's a simulation. You know? Would you call emotion picture real life? Right? Would you say it's artificial life? I mean, I guess you could say that, but I don't think it's the way most people understand it. I think that artificial intelligence is really a marketing term that the people who are creating these products and these technologies want us to believe that it can do things for us.
Richard (5m 53s):
That it ultimately can't. And I think it's very important that people learn about what this really is and learn how to use it in order to enhance their own intelligence.
Phil (6m 5s):
The last time we spoke, was it about simulation? That the stock market, the stock market and the way people react to it is almost like that they're working within a simulation. And this seems to speak about the way people are interacting now with the interfaces that are provided to deal with investment decisions, isn't it?
Richard (6m 28s):
Yes, absolutely. And this is the first time that I'm seeing this other than to myself. But I have a thesis. So what we know about, you heard
Phil (6m 39s):
It, you heard it first here,
Richard (6m 41s):
What we know about artificial intelligence, right? What we know about machine learning, that's another word that you'll hear a lot, right? And what we know about like computers is that they're very good at games. So computers are beating people at chess now, right? That was a big milestone. And it's, it's no competition anymore. Computers can crush people at chess. Okay? So computers are very good at games, and games have some relationship to the real world, but they're not the real world, right? So games are kind of a simulated environment, kind of an artificial environment.
Richard (7m 25s):
And when you have a kind of closed environment, like a game, that's where computers can do really well. Okay?
Phil (7m 35s):
They don't miss anything, do they?
Richard (7m 36s):
They don't miss anything. You know, and they can, they can sit there and play the game a million times in a day and just learn things about the game that you will never be able to over a lifetime, right? But it's within that game environment, right? Meanwhile, you have all these big for-profit companies that have a lot of computers, right? And that have invested a lot in artificial intelligence and in machine learning, et cetera. They're going to make more money if the economy is more like a game. So I believe that financial markets have been becoming more and more game-like over my time in the markets, which is 25 years, right?
Richard (8m 27s):
And I think that's partly because the institutional players who have all of that computing power want it to be more game-like, because that's where their machines can really excel.
Phil (8m 46s):
This kind of process can be applied to so many different areas, like in the medical field. I mean, have you heard about that story about the, the guy whose dog was dying and he entered into chat G P T four, not 3.5, but into four all the blood results of the dog and the dog was able, and the, he was able to save the dog because the vet that he'd been to didn't seem, wasn't competent enough to be able to work out what was wrong, was actually a very common sort of thing. Wow.
Richard (9m 16s):
I had not heard that story.
Phil (9m 18s):
Yeah. No, it's an interesting story. And this seems to be, you know, like an area where vets, doctors, accountants, all of this kind of stuff that is almost like a game because it is very rules based can at least have a backup. And then of course there's an implication for financial advisors and a financial advice in this space as well, which brings us to your newsletter.
Richard (9m 44s):
Yes. Well, I don't think ChatGPT is ready to take over for financial advisors just yet. Bloomberg just came out with their own version of this technology that's called Bloomberg, G P T. Now that is something that, I haven't gotten my hands on it yet, but that is based on, you know, 40, 50 years of articles and data and analytics that Bloomberg has accumulated over time. And you know, by the way, those kinds of data sets are gonna become much more valuable now because when you have a machine that can make use of that data and give more people access to that data in a way that they can interact with it, that's a very valuable resource.
Richard (10m 36s):
So yes, machines do things much better than humans do some things much better than humans do, right? They're good at storing information and they're good at retrieving information. And so what is gonna happen is that a lot of things that used to be hard for people to do are not gonna be that hard anymore. And hopefully it'll allow people to do the things that we're really good at, right? And so I think that right now we're in the very early stages of this, and I think it's really important that people start to use the technology just to understand how it works.
Richard (11m 17s):
It's sort of like when computers came out, you know, I was around, I, I mean, not for the early computers, but when the first PCs came out, right? And just starting to use them and understand how they work and what they can do and what they can't do, I think is a, is worth our time. Okay. ChatGPT which is the, the most widely known and widely used model available right now. There's also, Google has one called Bard with ChatGPT. The data that it started with is over two years old now, right? It doesn't have current data, it can't do mathematical calculations.
Richard (12m 0s):
Okay? But just to start to interact with the technology, understand how it works, learn about it, start to see what it can do as these models become online more, as the data becomes more current, I think it's gonna be more and more indispensable for people who are making their own financial decisions and for financial decision makers in general. And I can tell you a little bit about how I've used it so far.
Phil (12m 31s):
Yep. That's that. My next question, how have you been using it so far? But before that, I just wanted to just quickly explore how you, you men use the analogy of it's just like another keyboard, but as learning how to use that keyboard is key to using ChatGPT because it's about the quality of the prompts, isn't it? And the what you are actually asking it and how you phrase it, then rephrase it so that it can understand what you're asking for more clearly.
Richard (13m 3s):
Yes. So because these are large, very large language models, right? I think they have something like 50 billion parameters so they can go anywhere, right? So in order to use it most effectively, you have to guide it as to what you want it to do and where you want it to go. Okay? So if you just open up a ChatGPT prompt, you know, it's just a website you go, you can sign up and then you start typing to it and say, Hey, help me build my portfolio. You know, you're not gonna get that far.
Richard (13m 46s):
If you say, I want you to be Ray Dalio of Bridgewater Capital and help me to build a portfolio like Ray Dalio would build a portfolio. You know, it's like anything Phil, like giving anybody a task, right? The clearer we can be with the person that we're asking to perform a task, the more likely we're gonna get the outcome that we're looking for, right? Hmm. And it is our outcomes that we're looking for, right? It's not the machine's outcomes, it's our outcomes that we're looking for. So being a good user of chat, G P T means giving it enough specificity that it can get you what you want.
Richard (14m 31s):
Phil (14m 32s):
Yeah. Let's talk about that interaction. And I might just read out what the actual prompt that you wrote into chat G P T. Okay. Which is you are advisor G P T, and you're going to help me build my own personal securities portfolio that is well diversified according to modern portfolio theory. We're getting very specific here, while also prioritizing investments that reflect my personal values and interests. You are my research assistant. I look forward to working with you on this. This is the prompt that you entered into chat, G P T. Tell us about the thinking that went into this prompt.
Richard (15m 6s):
So I wanted it to act like a financial advisor. So I gave it an identity of being a financial advisor. I told you, you are a financial advisor,
Phil (15m 18s):
Hence advisor, G P T,
Richard (15m 20s):
Advisor, G P T, right? That was just something that I made up. It's a role, but it also gives me, it kind of personifies chat G P T for me right now. It, now I'm telling it like, I'm gonna talk to you like I would talk to an advisor and I want you to give me feedback like an advisor would give me, right? And specifically around portfolio construction. So I'm framing up the conversation. It's not only for chat G P T, it's also for me.
Phil (15m 55s):
Richard (15m 56s):
You know, and
Phil (15m 57s):
It's a very polite prompt as well. And this is a funny thing as well as I, right? I keep on thinking, do I actually need to write please when I write a prompt to chat G P T? Because it, because it feels so natural.
Richard (16m 8s):
I look forward to working with you on this, right?
Phil (16m 10s):
Yes, exactly. Yeah, that's right. Do you think that makes a difference?
Richard (16m 14s):
I don't know that it makes a difference in the output that it generates, it's just the way that I talk. Yep. And again, I think chat G P T is a linguistic user interface. It allows us to use natural language in order to get information from, you know, the information universe in new ways using natural language. So talk however you're comfortable talking.
Phil (16m 43s):
So tell us about the concept of the Holy Grail portfolio. You've mentioned Ray Dalio a little while ago in that podcast interview. So,
Richard (16m 51s):
Right, so that was kind of my next step with chat GPTs. I said, I like Ray Dalio's philosophy of a of the Holy Grail portfolio and chat. G P T knew what that was and said, oh yes, Ray Dalio's holy grail portfolio philosophy that involves different asset classes and it involves different countries and different sectors. So to build a Port holy Grail portfolio, you know, you really need to look at the full spectrum of, of, of securities in terms of asset classes, emerging markets versus developed markets and countries, et cetera.
Richard (17m 33s):
So that's great. And then I said, let's put together a portfolio of ETFs based on Ray Dalio's holy Grail portfolio philosophy. And it was able to do that for me, which is pretty extraordinary.
Phil (17m 48s):
What did it, and what did this portfolio look like? I mean, we'll link to it in the show notes so that listeners can find this portfolio with a cursory look at it. How did it look to you?
Richard (17m 59s):
I thought it looked pretty darn good, you know, broke it down. US equities, 40%, some large cap, some midcap, some small cap international developed market equities, 15%, emerging markets, 10%. So, you know, this isn't a portfolio that I would invest in. For one thing, I'm not really crazy about ETFs. I actually like to invest in individual companies that I feel more connected to than using ETFs. But as a starting point chat G P T to put this together after about a 10 minute conversation, I think it's absolutely extraordinary.
Richard (18m 42s):
And you know, I mean, a month ago I would not have thought that this was possible. And yet here it is. You know, have you ever heard of the touring test?
Phil (18m 54s):
Yes, yes. About the test of artificial intelligence. If you can't tell it's a real human on the other side of the conversation, then Yeah. It's almost at that point, isn't it? Yeah,
Richard (19m 5s):
I, I would say it is at that point, you know, you really, I mean, if a month ago you didn't tell me that chat G P T existed, and you put this in front of me, I would've thought I was talking to a human. And that's, that's, that's pretty extraordinary.
Phil (19m 25s):
Well, I, I'll just admit as well that a lot of these questions that I'm asking you, I actually, I just put your article into chat G P T and said let's have some questions based on this article. I just, I'm just being upfront here because I just wanna show the power of it, because while I don't want to completely lean on artificial intelligence, it does help to do a lot of the grunt work in the first place. And then you can work your questions around what you personally feel like you want to achieve in an interview. Yes. So here's one that came straight from chat G P T. Okay. How do you see artificial intelligence and machine learning transforming the investment landscape in the coming years?
Richard (20m 5s):
I think that what we're looking at is what I call the industrialization of ideas. So if we think back to Henry Ford and the assembly line and the mass production of automobiles, that gave everybody access to a higher level of technology, right? It gave everybody access to a product that you couldn't just make in your garage before. Right? But you also got a lot more diversity when everybody made things in their garage before things didn't all look the same.
Richard (20m 48s):
Right? Those model T Fords that came off the assembly line all looked the same. Same, you could have any color as long as it was black.
Phil (20m 55s):
That's right. Same color.
Richard (20m 59s):
So what does that mean when we're talking about the realm of knowledge and ideas? Hmm. What does it mean when everybody's working from the same set of ideas when everybody has access to the same set of ideas at one level, that it's kind of terrifying, isn't it?
Phil (21m 18s):
Well, it's kind of a bit, it's a homogenization process. It's
Richard (21m 22s):
A homogenization process, absolutely. Yeah. But at another level, it's like we're all gonna have access to a better set of ideas in a shorter period of time, right? So I would say that most people couldn't have built a holy grail portfolio of ETFs on their own, but do you think it's Chat G PT can do it for you in 10 minutes? Now? I do think that it's going to homogenize things more and it'll have some democratizing influence in that more people will have access to better information and better knowledge. But it's also, you know, in that is a danger that will really lose diversity and creativity.
Richard (22m 8s):
So I think to maintain that diversity, maintain that creativity, that's why I say chat G P T, just use it, right? Like, we've got to engage with it. We've got to use it. We've got to
Phil (22m 23s):
Richard (22m 24s):
It. Outsmart it, absolutely. You know, and be better than it. Stay a step ahead of it. And the danger isn't that chat, G P T is gonna become sentient and intelligent and take over, take over the world, right? Yeah. It, it just doesn't, it doesn't matter how many computers you hook up. It doesn't, you know, one day just flip a switch and all of a sudden it's conscious and sentient. No. So, you know, the danger is that we will think that it's intelligent and we will relax our own need to be intelligent and will accept chat G P T for intelligence.
Richard (23m 5s):
You know, instead of always working harder to be even more intelligent, you know, more ahead of the curve, more thoughtful, more reflective people who can think for themselves and who can use chat G B T to make their own thinking better. I think those are the people who are going to really make the most of it and who are ultimately gonna be the most successful in the markets.
Phil (23m 29s):
Have this feeling that it's also a good way of testing your own biases and testing your own fears and so forth in that, if you've got something like this that is looking dispassionately at data, like if you say, you know what, what's this a portfolio about, you know, you might have your own idea of a portfolio and you can put it into Chat G P T and presumably it will, it will share its thoughts on what you've put in there and you can then test your own rationality as well, hopefully.
Richard (24m 3s):
Yes, absolutely. I've seen people, one application I saw of it was somebody took an article that was in the Wall Street Journal that they didn't have time to read the whole thing and just pasted the article in and said, summarize this for me in one paragraph. And it gave back us a one paragraph summary. And then it said, and then the next question, the follow up question was what biases are in the article? And then chat G P T gave like four, you know, cognitive biases that were displayed in the article. Like that there, for example, there wasn't any countering point of view, right?
Richard (24m 43s):
It was all just one point of view. There wasn't any other points of view brought into the article. So I think that things like that, absolutely, it's gonna be very helpful to, to help us to look deeper and see past some of our biases at the same time, you know, have you heard of hallucinations, AI hallucinations?
Phil (25m 6s):
No. Tell us about those?
Richard (25m 7s):
Yes. Well, this is a big deal, you know, chat, G P T can be quite the con man, the confidence man, right? Presenting information to you that's absolutely wrong with total confidence and conviction. And this happens all the time, and it's a big problem in these large language models, okay? So you cannot just accept what chat G P T and other large language models give back to you because sometimes they are totally wrong and they won't admit it. You, they, you won't know it until, unless you bring it to their attention to the model's attention.
Richard (25m 49s):
I hate Anthropo all this anthropomorphizing language, right? Yeah, yeah.
Phil (25m 54s):
My name say Martin, it's impossible to
Richard (25m 55s):
Avoid it, you know, because we are interacting with it in the way we interact. We've interacted with other humans to date right? Through natural language. So we're gonna have to come up with a new vocabulary. But so how does, but so this, you know, it's a technical term, it's a of, of a yeah. Large language model hallucination, and they're very, you know, they're not rare.
Phil (26m 20s):
How, how does this, how does this happen? I mean, do you know of any examples of it?
Richard (26m 25s):
Well, again, it's just a statistical model, right? It's just, it's a, it says, you know, if I've got this string of words, what's the next word that has the highest statistical value from the model Yeah. To add here, right? So I don't have any examples off, off the top of my head of hallucinations,
Phil (26m 46s):
But this is just something
Richard (26m 47s):
That interview on 60 minutes this past week that Scott Pey, I think his name is, whereas interviewing Sunar Pacha, the, the CEO of Google and, you know, showed in that 60 minutes interview, like they got some data back from chat G P T, and when they went back to the office and, and read it more carefully, there were several factual errors in the output, right? And Sundar Pacha, the C E O of Google acknowledges yes, you know, these, these hallucinations is, is what they call them in the, in the business are a significant problem. And it's one of the reasons why Google is taking a very, you know, measured slow approach to releasing this technology because you can really mess things up with it if you don't know what you're doing.
Richard (27m 38s):
And if you're not critically, you know, using it critically,
Phil (27m 43s):
I find when I, I'm looking for investment information on chat G P T, that it outputs very bland advice about risk. You know, they always put that kind of disclaimer at the end of any of these outputs saying, you know, that stocks can be risky and consult a financial planner and, and all of that stuff. It sounds like the, the disclaimer at the end of, you know, yeah. Product disclosure statement and your specialty is risk. How would you nuance it so that you could use G Chat G P T to help you to manage your own risk?
Richard (28m 15s):
Well, for one thing, you can tell it to stop giving you all those disclaimers.
Phil (28m 20s):
That's true. I should have thought about that with all the disclaimers.
Richard (28m 22s):
I really don't want to hear them anymore. I understand I'm an adult, I'll make my own decisions. I literally said something like that to Chat G P T because I got tired of all of the boiler plate disclaimers too. And, and it stopped giving it to me is like, okay, sure, no problem. Look, most people Phil, aren't following the basics of risk management. You know, most people aren't thinking about, what is my risk tolerance? How long can I be in the markets for, you know, what are my goals? So those are the simple questions that chat G P T kept putting at me. You know, like, well, it depends what's your risk tolerance and what are your goals and how long can you leave your money in the markets, right?
Richard (29m 9s):
Those are important questions and they are basic questions. And so if chat G P T asks you those and you don't have answers to 'em, you should say, well, can you help me figure those out? Right? So one thing I found out was it helped me to identify several risk tolerance questionnaires that were available online that I wasn't familiar with, even though, you know, I'm pretty literate in this stuff, so keep pushing it until you're not getting any more useful information. Right. And, you know, when it comes to risk management, you know, chat, G P T at this point can't do the math for you, but it is starting to interact with other platforms throughout the internet using plugins.
Richard (29m 55s):
And one of the ones that I'm excited about is, is Wolfram Alpha. So by Steven Wolfrem, the mathematician and his engine does do calculations, right? So, so combining something like chat G P T and Wolfrem Alpha and a good historical data set, that's where when we start to see opportunities like that, that's when I think it's really gonna get fun and exciting.
Phil (30m 25s):
Okay, Richard, so tell listeners how they can find out more about this particular article and dig a bit deeper into it and a bit more about yourself.
Richard (30m 33s):
Yes. The easiest way to keep track of what I'm thinking about and what I'm working on is @drrichardsmith.com, Dr. Richard smith.com. I'm also on email@example.com.
Phil (30m 48s):
Fantastic. Okay, well, thank you very much for joining us today, and it's a fascinating area. I wa I wasn't sure whether I wanted to speak about chat G P T, but when I read about how you'd used it, I thought, fantastic. It's a really great way and a, a tool that we can use to help to hone our investing. Hopefully,
Richard (31m 4s):
I think it's still a work in progress, but I think that the more we get familiar with it now, the more we'll be prepared to really take advantage of it when it gets really useful,
Phil (31m 15s):
Just use it. Huh?
Richard (31m 16s):
Just use it. Thanks Phil.
Phil (31m 18s):
Thank you very much.
Chloe (31m 19s):
Thanks for listening to Stocks for Beginners. If you enjoy listening, please take a moment to rate or review in your podcast player or tell a friend who might want to learn more about investing for their future.
Stocks for Beginners is for information and educational purposes only. It isn’t financial advice, and you shouldn’t buy or sell any investments based on what you’ve heard here. Any opinion or commentary is the view of the speaker only not Stocks for Beginners. This podcast doesn’t replace professional advice regarding your personal financial needs, circumstances or current situation.