Recent stories about ChatGPT and other large-language AI models have tended to focus on using them to write school essays, corporate blogs and silly songs. Meanwhile, Bloomberg has quietly been working on a generative AI that could truly upend the financial investment world.
QUICK TAKES
BloombergGPT is a new generative artificial intelligence model. Unlike its better-known cousins, which are trained to cover a broad spectrum of general knowledge, Bloomberg’s large-language model (LLM) has been specifically trained on a wide range of financial data to support a diverse set of natural language processing (NLP) tasks and investment analytics within the financial industry.
This appears to be the first model aimed at a specific industry using the industry’s unique language and acronyms, as well as focusing on its specific analytical needs. The model is expected to improve existing financial NLP tasks—such as sentiment analysis, named entity recognition, news classification, and question answering, among others—while also unlocking new opportunities for using the vast quantities of data accessed via the Bloomberg Terminal to better help customers.
BloombergGPT was trained on about 700 billion-plus tokens (i.e., word fragments), 363 billion of which were taken from Bloomberg’s own financial data. Another 345 billion tokens came from general-purpose datasets obtained from elsewhere. This training data ranges from everything in the Bloomberg archives to data received via its ubiquitous terminals to SEC filings to non-Bloomberg news sources and even a complete file of Wikipedia entries. Taken together, Bloomberg is seeking to design a model that generates best-in-class financial results, while also maintaining competitive performance on general-purpose LLM benchmarks.
Because it shares a training base with other LLMs, BloombergGPT can do the sorts of things that we’ve come to expect from ChatGPT and similar models. But it can also perform tasks more tightly connected to Bloomberg’s needs. It’s better tuned to answer specific business-related questions, whether they be sentiment analysis, categorization, data extraction, or something else entirely. In other words, BloombergGPT could open the full potential of AI to the financial world.
Although it is yet to be seen how BloombergGPT will impact the fintech industry, some of the potential uses of BloombergGPT might include:
Generating an initial draft of a Securities and Exchange Commission filing. Given a large amount of data of filings and much like how ChatGPT can produce a provisional patent filing or customized programming code, it may be entirely possible to generate an SEC filing, potentially reducing the cost of filing.
Providing a company chart of an organization and linkages between an individual and multiple companies. Because company names and names of executives are fed into the BloombergGPT model, it is entirely possible that it can be queried for at least the organization’s executive-level structure.
Automation of generation of draft routine market reports and summaries for clients.
Retrieval of specific elements of financial statements for specific periods via a single prompt.
It wasn’t too long ago that digitizing microfiche to upload to a searchable database was considered cutting edge. Just last November, we were introduced to ChatGPT and the ability of AI to produce professional-sounding reports and website content. Now, we’re looking at large-language AI models to disrupt the financial services industry. What’s next? Medicine seems an easy next target. Think of the advantages of being able to have an AI model filter through all possible diagnoses in a few seconds. Or maybe real estate. Analyzing supply and demand metrics, and how demographics and job creation are affecting specific neighborhoods would seem to be right up AI’s alley. Or maybe even innovation itself. What would an AI model respond when prompted: List the types of transformative innovations the world could see in the next three years. The answer would vary depending on the database and bias of the AI, but it would be interesting, nonetheless.