BloombergGPT's Disruption: Exclusive Financial Data vs. Financial Analysts

0:00ChatGPT is obviously the premier AI language model at the moment on the market that is able to do the most and accomplish the most and is the most vast in data size of what is trained on. But there is a lot of new AI models that are being trained in very domain-specific areas. And so today on the podcast, we're going to talk about one of the most recent ones that has come out that is going to disrupt a massive industry, which is finance. And so for a while, ever since the launch of ChatGPT, people have kind of been asking, you know, like, what would happen if ChatGPT was trained on, you know, all of the financial news and data in the world?

0:37Would this thing be able to predict the stock market and predict, you know, when you should buy or sell or invest? And, you know, that's kind of been a question going on. And we may not be very far from that because Bloomberg, which is famous for making all of its money from its$30 ,000 a year subscription to Bloomberg Terminal, which is kind of for corporations to get access to all of its data, news and financial information about companies. They have created Bloomberg GPT. So this is a fairly massive undertaking, but it's going to be incredibly powerful. And some people are saying that this is going to completely replace financial analysts.

1:16So what's pretty interesting is it is a 50 billion parameter large language model. And it was specifically built from scratch, right? So not like, you know, Google Bard, who's allegedly using chat GPT outputs. Allegedly, we'll see. but it's built completely from scratch from Bloomberg's own data set so this is really important and a lot of people are saying this is really exciting just for the fact that this is a big moment in the industry where people are starting to make data sets and large language models from all of their own data and this is incredibly powerful right anyone can go and train an AI language model based off of Wikipedia or some of these other open source data sets but being able to build something like Bloomberg GPT, which is expressly off a lot of their own proprietary data.

2:08So this is quite amazing. I think we're going to see a future where a lot of people are building their own models. And this is good because this means that the industry is going to start branching out from being just controlled by the big tech, Google, Facebook, Apple, OpenAI. And now this is possibly going to start being owned by a lot of different people, right like quora has started their own ai model that answers questions based off of their own website and i think we're going to see a future where a lot of these different players build their own specific data sets they they block it so no one else can use their data right like people can't use um bloomberg articles a bit that have been written over the last 50 years um and so i think we're going to see a lot of diversity this is going to mean that ai players are going to have to work together to make things um effective and i think this is kind of important uh as we move forward so that we don't have these big monopolies controlling such a powerful industry right we recently have seen reports out of i believe bloomberg um that said 300 million jobs are going to be replaced by ai over the coming years and we do not want all of this to be built on you know one giant monopoly i think if open ai as much as i love chat gbt i think if open ai completely dominated and was replacing those 300 million jobs i think would be in a lot of trouble because you know, it's never good for the concentration of power to be in one specific company or organization.

3:27So I think the fact that we're seeing this kind of spread out a little bit more between a lot of companies is a really good sign, really healthy sign. And it's going to be really interesting. Now, that being said, it's pretty interesting in just what Bloomberg has done. So they have built this out of what they call 700 billion tokens. So tokens are essentially, I just I think and models use the the phrase tokens a lot when they talk about how they train But essentially what it means is just like word fragments of pieces of words So it's not like 700 billion words because a token might be like half a word And I I guess that's just how they calculate it I you know They probably could say like letters and it would be more precise But maybe that would be insane because it's like seven trillion letters or something But in any case the word it would be one token the word amazing might be like a few tokens So it's just kind of how they break up the words.

4:19In any case, Bloomberg GPT, 700 billion tokens. GPT-3, which is, you know, ChatGPT when it was first released in 2020. So I guess before ChatGPT, more of DaVinci, which was kind of the training model for a couple of years, that was trained on 500 billion tokens. So this Bloomberg GPT is bigger than OpenAI's, you know, precursor to ChatGPT and it's specific to financial data, right? They're not trying to be the expert at everything you might ever want to ask an AI model. They're specific to financial data, companies like historical financial data, the market's historical financial data. And this is going to be incredible because if you think about it, you know, a financial analyst essentially is like a, you could call it like a speech interface to all of this financial data.

5:08You know, you might have a financial analyst and you might ask them, hey, can you pull me up the numbers on x y and z how historically these markets have fared when they've had these conditions and they might tell you so you can help forecast the future now essentially you're going to be able to do that with um bloomberg gpt and be able to figure out what is going on and you're going to be able to ask it your questions and it will do all the research immediately for you based off of its data set so this is going to be incredibly powerful i think i think we're going to see this move in a really big way to finance and you know when people talk about the power of ai i think this is really what it's all about when they talk about you know how Disruptive it's going to be and whatnot.

5:44I don't think this is all going to be just chat GPT. That's disrupting the world I think it's going to come in a sense that a lot of these bigger players with big data sets or these other players with you know data sets anyone with big custom data sets That's proprietary that they own they're going to be able to harness that to do something really well really powerful So I think this is going to be a really big a really big side of this and Bloomberg GPT about 52 % of it is proprietary data or cleaned financial data so clean financial data meaning it might be publicly available financial data but they have went in and adjusted it and made it so the the AI understands it aka like a human had to go touch it and mess with it so if somebody wanted to it might be publicly available but if someone else wanted to train a model off of it they would have to have the manpower to be able to make it usable right so it might not be you know exactly proprietary but it's like they've customized it in a way that adds value to it in any case they have a ton of data that they've specifically covered they've said in their kind of press release they said the news category includes all news sources excluding news articles written by bloomberg journalists overall there are hundreds of English news sources.

7:00Generally the content in their data set comes from reputable sources of news that are relevant to the financial community so as to maintain factuality and reduce bias. And this is pretty interesting. They've shown a couple of things that it can do. One of the things, I mean, it might seem like a parlor trick or something that ChatGPT can do, but they said it's actually really good at writing headlines, right? Because it's trained on all this news. So they can give it like a summary and get it to write a headline. This is one example they shared in their news release. They said, the U.S. housing market shrank in value by 3 trillion or 5 % in the second half of 2022.

7:34According to Redfin, that's the largest drop in percentage terms since the 28th financial crisis when value slumped 6 % during the same period. Okay, so if you tell it that, it can then say, home prices see biggest drop in 15 years, right? It creates a title for you. So I know ChatGPT can do that. I assume this is going to be super custom for like financial kind of things. Sometimes I get ChatGPT to do something that doesn't quite hit the mark. And so this is going to be people are going to pay more like this because it really hits the mark every time when it comes to financial stuff. So I think one of the most interesting pieces of this is just the fact that this is really pointing to a future where big tech doesn't own everything.

8:12And so I think a lot of people are really excited about that side of it. And I think that it's not hard to imagine more specific use cases that go beyond just simple benchmarking that a financial analyst would do. So either for Bloomberg's own journalists or for some of its Terminal customers, they're paying$30 ,000 a year for a subscription to Bloomberg Terminal. I think that those people are going to be able to harness the data. I'm sure it's going to be incorporated into that product. And I think, you know, a corpus of all of the world's premium business reporting, plus the entire set of financial data that they have, which a lot of it has been structured and cleaned, is going to be just this really rich source of information that this generative AI is going to be able to mine and going to be able to build out some really incredible things.

9:01All of this, it's important to note that Bloomberg is not an AI company, right? So they've done this with a very small team, much, much smaller than people that have built similar models, right? If you're comparing them to the size of OpenAI and the people that helped develop that, this is an incredibly small team, but they've been able to build a lot. And I think this really just shows the power of what is happening in AI right now. The tools that are coming out, especially, you know, Lama from Facebook or Meta, really reduces the amount of time and cost to train these models, the amount of manpower.

9:34And so I think this is really exciting to see that people are starting to use smaller teams to build such impressive data sets and tools. So it's going to be interesting to see how this goes, how this transforms the financial industry. Financial analysts should, you know, watch out. They should probably start figuring out how to use these tools to help them extend their jobs and, you know, them more effective and I think this is going to be really interesting to watch how this continues to unfold in the future.