The AI Daily Brief: Artificial Intelligence News and Analysis - Can ChatGPT Outperform Wall Street?

Episode Date: May 22, 2023

Last month, a University of Florida study showed ChatGPT with 500% returns using news and sentiment analysis. A recent Finder.com study also showed ChatGPT stock picks outperforming the market. On today's episode, NLW explores AI for finance, as well as whether AI itself will bring a stock market boom.  The AI Breakdown helps you understand the most important news and discussions in AI.  Subscribe to The AI Breakdown newsletter: https://theaibreakdown.beehiiv.com/subscribe Subscribe to The AI Breakdown on YouTube: https://www.youtube.com/@TheAIBreakdown Learn more: http://breakdown.network/

Today on the AI breakdown, we're looking at whether chat GPT can outperform Wall Street stock pickers. Before that on the brief, we look at new audio research from Google, some breakthroughs in AI medical research, and the G7 discussing AI yet again. The AI breakdown is a daily podcast and video all about the most important news and discussions in AI. Like, subscribe and share, and learn more at Breakdown.network. Welcome back to the AI breakdown brief, all the AI headline news you need in five minutes or less. For the last six months, a huge amount of the AI discussion has been around consumer tools. Chat GPT, mid-jury, even text to video, text to 3D worlds. These are the types of things that people have been really engaged in discussing, which makes sense.

It's the tools that consumers are using, and they're seeing how they can actually influence, how they do their jobs, and how they think about their futures. But there is an entire other strand of AI use cases that are simmering just around the corner that show potentially how this technology is going to disrupt industries far beyond just the of how we do our work. One of those areas is of course medical research and recently researchers from the University of California San Diego shared information about how they were able to use AI to identify an important DNA sequence which is known as the downstream core promoter region or DPR. This is a DNA sequence that's used in the activation of up to

a third of human genes. Gene activation is a fundamental process that's related to growth, development, and disease. So what the team did was use machine learning to identify what they call synthetic extreme DNA sequences. These are highly specific DNA sequences that function in gene activation. They compared millions of DPR gene activation elements in humans as well as fruit flies and used that to find sequences that were active in humans but not in the fruit flies. Now amazingly, their AI models were successful in predicting the functionality of these rare sequences. Their models analyzed 50 million test DNA sequences, which would have been totally impossible using traditional methodology. While people are really excited about this specific research,

I think they're even more excited about the way that it shows the viability of AI in medical research more broadly. Professor James Katanaga, who led the research, said, There are countless practical applications of this AI-based approach. The synthetic extreme DNA sequences might be very rare, perhaps one in a million. If they exist, they could be found using AI. Next up, we have new audio research out of Google. This is a project they're calling Soundstorm, which is a project for efficient parallel audio generation. Let's listen first to this clip.

Did you hear about Google's paper on Soundstorm? Um, no, I must have missed it. What's, what's it about? Well, it's a parallel decoder for efficient audio generation. Oh, yeah. Oh, yeah. Oh, yeah. Oh, yeah. Interesting.

Yeah, yeah, like, this one was generated by Soundstorm. Wait, what? So the big thing that makes Soundstorm interesting is just how fast it is. It can synthesize 30 seconds of sound in about half a second. It can do that because it's a non-autoregressive model. which means that it's not trying to guess each word in sequence, it is instead figuring out entire phrases at once. That makes it much more difficult to train, but once you are successful in training it, much faster. Another important aspect of this research is that it uses Speer TTS, which is another methodology,

to be able to synthesize dialogue, not just individual speakers. That means you can create conversations where you can control what said, who says it, and when they say it. Conversation generation is also extremely fast, with 30 seconds being able to be created in about two seconds. Now, given the potential use case for this type of technology and scams or other nefarious uses, Google is thinking about whether there are ways to make it easier to detect that it is generated audio. Next up today, a story that serves as a reminder of just how hard it is out there for an AI company. We're very used to AI companies getting funding right now, but we're a little less used to them actually shutting down. Neva is a startup that tried to innovate a new type of search interface based on chat and AI,

as opposed to the traditional Google-style blue links. This weekend they announced that they're shutting down the consumer search engine, and the reason they said is that although the technology was good, it's extremely hard to get consumers to switch their behavior around where they search. Now, the company says they're going to be pivoting to explore more enterprise use cases, but as we all discuss especially AI regulation and think about the balance between the need for protection with not wanting to just over-prioritize incumbents, it's a good reminder about how hard it is for those startups out there. Finally, there was a meeting of the G7 world leaders in Japan on Friday, and of course, AI was high on the agenda. According to Reuters, quote, the leaders agreed to have ministers discuss the technology as the Hiroshima AI process and report results by the end of the year, according to a summary of a working lunch. This shows, I think, two things. One, how serious AI is that it's on the agenda for the most important meeting of world leaders in the world.

And two, just how likely to be behind they are, given how long it takes them to wrap their heads around, what's happening. That's it for today's AI breakdown brief. If you are enjoying this, please like, subscribe, and share, and I'll be back soon for the main AI breakdown. Welcome back to the AI breakdown. So today we're talking chat GPT, AI, and finance, and this was inspired by a growing conversation I'm seeing around how chat GPT and other AI tools might be used to help people make money, right? And not just through content creation or some new type of side hustle, but actually understanding markets. Where I wanted to kick off is with some research that came out last month and that really got people excited.

So Brian Romley here tweets, ChatGPT picks stocks and performs better than anyone expected. 500% returns. New university research uses some of the systems I have been testing. Just the start of a new powerful way for anyone to prosper. The paper that Romley is referencing is called Can ChatGPT Forecast Stock Price Movements, return predictability and large language models. It was posted on April 10th from two reads.

researchers at the University of Florida, Alejandra Lopez Lira, and Yehuah Tang. Basically, what the research did is it used AI to track sentiment analysis and then map that to potential financial decisions that you would make based on the news. So, for example, if there was a negative story in the news, that might prompt the model to short a stock, whereas if there was positive news, that might prompt the model to buy the stock. So here's how they describe it in the paper. We assume that if a piece of news is revealed before the market close, we buy or short sell a position at the market close price. If a piece of news is announced after market close, we assume we buy or short sell a position at the next opening price. All the

strategies are rebalanced daily. The portfolio that they found performed the best was a portfolio that bought companies with good news and short sold companies with bad news. That's where that 500% return number came from. Now, of course, this was just research. This was all theoretical. But it wasn't the only study to show something similar. On Friday, May 5th, CNN posted a story called ChatGPT can pick stocks better than your fund manager. The study there referencing this time came from financial comparison site finder.com. Between March 6th and April 28th, a dummy portfolio that they had ChatGPT power of 38 stocks gained 4.9% as compared to 10 leading investment funds, which over the same period of time had an average loss of 0.8%. Now, what if you had just bought

the S&P 500. Well, the S&P 500 would have outperform those funds as it was up 3% over that time, as again compared to that negative 1% or so, but the theoretical ChatGPT fund at 4.9% again beat that 3%. So how did ChatGPT actually make those selections? Here's what Finder.com asked. We asked ChatGPT to create a portfolio of stocks from high-quality businesses with criteria taken from the leading 10 funds it is competing against. These include things like low levels of debt, sustained growth in the past, and assets that generate an advantage of competitors. With that criteria, the stocks that ChatGPT chose include everything from Nvidia to Johnson and Johnson, to PepsiCo, to Cigna, to Berkshire Hathaway, to Salesforce, to Walmart.

Now, the other really interesting part of this finder.com study was just about how much people knew about ChatGPT in general. They asked people if they trusted AI for financial advice. Of the respondents, 8% said that they had already used it for financial advice, 19% said they would consider using it for financial advice. 35% said they wouldn't consider using it for financial advice. But the biggest portion, 38%, said they don't know what chat GPT is. Hard for those of us who are creating content around this space to imagine, but it's still the norm for people not to necessarily have a good sense of what these technologies are.

These type of chat GP2 portfolios are starting to pop up everywhere. One that has gotten a lot of attention is the GPT portfolio by autopilot. This came after that CNN article we mentioned, and on the website it says, CNN says chat GPT can outperform your money manager. We're putting that to the test publicly. Now, this comes from the same folks who built the Nancy Pelosi stock tracker and the Michael Burry stock tracker and the Warren Buffett stock tracker, which is a company called autopilot that does well exactly this.

It helps people understand how other people are investing their money and effectively copy trade them. On Friday, the GPT portfolio wrote, the first week of GPT managing our $50,000 is officially in the books. Pick some winners like Broadcom and Amazon. 9.5 million successfully invested alongside it. Didn't lose all of our money. Next week, ChatGPT will sell these and pick a new set of 20. In terms of actual numbers during that period, the SPY was up 1.1%

while the ChatGPT fund was up 1.4%. You've also got other individuals like Udit Genka who are doing this type of experiment, and Streamlit developer Data Chaz, who pointed two new chat Chat GPD plugins as a way to help your investment journey as well. He writes, Trying out the excellent portfolio pilot, chat GPT plugin. Here are eight great stocks to invest in, according to it. Let's see where that takes me.

If you are a chat GPT plus user, at this point, you should be able to enable browsing as well as plugins. I did a quick search today, and I found that of the 130 official plug-in so far, seven of them are related to stock news or market information. That's over 5% of the official plugins. Of the financial plug-in so far, it certainly seems like portfolio pilot is producing the best results. When I ask what's today's most important stock news, it was able to actually call news from today, as opposed to some of the other plugins, which had news from all the way back in March, as some of their top news today.

Now, I think so far the most interesting thing about ChatGPT plugins for finance is not so much that they're going to have great stock picks, or that they're even going to be the best way to aggregate and curate news, but that presumably over time they're going to allow you to get really deep on research from a much easier interface. If you're interested in hearing more of a comparison of the existing financial plugins, let me know in the comments and I'll think about doing that as another video later this week. I also think that we're going to see a lot more specialized plugins coming soon. Right now, all of these plugins are kind of ticker data and market data and focused on news, but YouTube CS Dojo just tweeted recently,

I created a chat GPT plugin that can help you read and summarize dense financial documents. used it to make better investment decisions. One of the big things that's anticipated to change about GPT in the months, if not weeks to come, is that the context window is expected to expand from 8K, where it is now, to 32,000 tokens going forward. A larger context window allows more information to be pulled in, so you might see a use case like reading big, complicated financial documents like 10Ks to get better results because there's more context for chat GPT to see the entire set of information all at once. The other thing to discuss in terms of AI in markets isn't just whether it will help people make better stock picks, but whether it in and of itself will be a boon to the markets.

Over the weekend, Reuters published a piece called Artificial Intelligence gives real boost to U.S. stock market. The article reads, the S&P 500's 9% rally this year has been driven by a handful of the index's biggest stocks, a number of which are at the center of the AI frenzy that is spread in the wake of Chatbot sensation chat GPT. Five stocks, Microsoft, Google Parent Alphabet, Invidia, Apple, and meta-platforms are responsible for the S&P 500's entire year-to-date return. About 25 to 50% of those gains are owed to the buzz around artificial intelligence. Society General also did an analysis of 20 stocks that are widely owned in AI-related ETFs and found that removing those stocks from the S&P 500 would have reduced the performance by about 10 percentage points, which would have put stocks in negative territory for the year. By the way, the assets under management of those exchange traded funds has grown about 40% this year.

There's also a longer-term optimism about AI and Wall Street as well. Goldman Sachs strategists recently estimated that generative AI could create productivity gains that expand S&P 500 profit margins by four percentage points in the decade to come. Deutsche Bank said it a note, we are strongly of the view that AI will change the world, and that note, by the way, was called, will chat GPT prevent the U.S. recession? Veteran market analyst Ed Yardini even said that chat GPT might ignite a quote roaring 20s for stocks. In a recent note, he said, The market has climbed a wall of worry thanks to Wall Street's warriors

who've predicted that the banking and debt ceiling crisis could make their widely expected imminent recession worse. But when it comes to AI, he says, this may be the event that launches the roaring 2020s. If so, then we can spend a lot less time obsessing about what the Fed will do next and focus on how technology is boosting productivity and the standard of living throughout the economy. Hedge fund billionaire Paul Tudor Jones recently said something similar. Last week, he said, I think the introduction of large language models and artificial intelligence

is going to create a productivity boom that we've only seen a few times in the last 75 years. Those productivity booms that he identified were the 1950s when the U.S. invested in its infrastructure, the 1980s following the rise of the PC, and the 1990s with the introduction of the internet. He said each of those three episodes were associated with productivity gains of somewhere between 1 and 3%. Let's say that this large language model is going to give us productivity boom of 1.5% over the next 5 years, which I think is possible. I just go back and look historically what that's done during those productivity miracles. You've had the stock market on average appreciate 15% per year. You've had inflation come down, and you've had a PE expansion of somewhere

between 1.5 and 2. Now, whether that actually happens or not, in the meantime, there will be tons of funds giving people access to the hype. One of the examples comes from roundhill investments who last Thursday announced the Roundhill generative AI ETF, ticker chat, which they claim is the first generative AI focused ETF. Anyways, guys, to be clear, none of this is financial advice. I think any time you have everyone running in the same direction in markets, it's worth being wary of, but there's no denying that people are understanding just how significant AI is going to be, even if the specific companies might change quite a bit from where things are right now. I personally am not ready to outsource all of my investment decisions to chat GPT, but you better believe I'm trying

all of these new plugins to see what actually can help me analyze data and make decisions better for myself. As I mentioned before, if you're interested in a deeper analysis of the chat GPT finance plugin so far, let me know in the comments. But for now, that is it for today's AI breakdown. If you're enjoying the AI breakdown, please like, subscribe and share. Go check out the podcast. Go subscribe to the newsletter. You can find all of this information at breakdown.network. Until next time, guys. Peace.