paint-brush
Money Talks: How Bloomberg GPT Is Changing the Gameby@irinakozerog
1,334 reads
1,334 reads

Money Talks: How Bloomberg GPT Is Changing the Game

by Irina KozerogApril 10th, 2023
Read on Terminal Reader
Read this story w/o Javascript
tldt arrow

Too Long; Didn't Read

Bloomberg unveiled a new artificial intelligence model named BloombergGPT. This new AI model is set to change the finance industry. The technology can be used in a variety of financial applications, including asset management, risk management, and trading. It took developers 53 days to train the neural network - during which time 569 billion tokens of information were circulated.
featured image - Money Talks: How Bloomberg GPT Is Changing the Game
Irina Kozerog HackerNoon profile picture

On March 30th, Bloomberg, a top provider of information for professional financial market participants, unveiled a new artificial intelligence model named BloombergGPT. This new AI model is set to change the finance industry.


BloombergGPT is specifically designed to understand financial language and can process complex financial information, such as market trends, risk assessments, and portfolio optimization.


This technology can be used in a variety of financial applications, including asset management, risk management, and trading.


Finance is one of the most data-intensive industries, and the use of industry-specific GPTs has the potential to revolutionize the way financial analysis, risk assessment, and decision-making are done.


Judging by initial feedback, the highly specialized neural network outperforms its competitors in solving specific financial problems.


research paper published by the company details the development of "BloombergGPT," a new LLM model that has been trained to handle the vast amount of financial data to solve various natural language processing (NLP) challenges in the financial industry.


The authors explained model choice, learning process, and evaluation methodology. As a next step, they plan to release training logs (chronicles) detailing  BloombergGPT training experience.


Bloomberg built a 50 billion parameter LLM which is significantly smaller than GPT4 but trained on a wide range of financial data, and their findings show that the mixed data sets that they used outperform existing models.


It took developers 53 days to train the neural network - during which time 569 billion tokens of information were circulated.


The team created a data set of 363 billion tokens based on Bloomberg's own extensive data sources and augmented by 345 billion tokens from general-purpose datasets. 

What Is BloombergGPT Capable of?

  • Generating draft routine market reports and summaries for clients, reducing the need for manual report generation,


  • searching for information about various companies and their executives,


  • creating a chart of an organization's executive-level structure and linkages based on the company names and names of executives,


  • generating an initial draft of an SEC filing,


  • generating headlines for news from longer financial blurbs,


  • retrieving specific elements of financial statements for specific periods with a single prompt.

BloombergGpt’s Potential Impact on the Fintech Industry Remains to Be Seen, but It Could Have Several Potential Uses:

  • More informed investment decisions and reduce the risk of financial losses,


  • more effective trading strategies,


  • fraud detection and prevention,


  • stronger customer relationships, personalized investment recommendations, advice, etc.

Does It Make a Difference?

Bloomberg says that they used diverse structured and unstructured financial datasets, including company filings, transcripts of Bloomberg News transcripts, opinions, press releases, etc.


But the majority of this financial data is publicly available and can be obtained from other sources, except paid Bloomberg content.


So, what is the real advantage of specific GPT if general ones could be trained on the same publicly available financial data?


It's important to note that GPT models are typically pre-trained on vast amounts of data from diverse sources, and then fine-tuned for specific tasks.


This means that even if GPT-4 were able to match the performance of finance-specific GPT models on financial tasks, it would still require specialized training and fine-tuning to achieve optimal performance in financial applications.


Technically, BloombergGPT outperforms OpenAI's ChatGPT-3 and Google's LaMDA language engines but falls behind the larger LLaMA and PaLM.


Bloomberg GPT isn’t that big compared to other general-purpose transformers(GPTs). The killing feature of this transformer for Bloomberg terminal users is the ability to generate Bloomberg query language, and this is a proprietary language.


“The quality of machine learning and NLP models comes down to the data you put into them.”


“Thanks to the collection of financial documents Bloomberg has curated over four decades, we were able to carefully create a large and clean, domain-specific dataset to train a LLM that is best suited for financial use cases.


We’re excited to use BloombergGPT to improve existing NLP workflows, while also imagining new ways to put this model to work to delight our customers,”  explained Gideon Mann, Head of Bloomberg’s ML Product and Research team.


BloombergGPT is a cutting-edge tool that has the potential to transform the finance industry. By using advanced AI techniques, it can help financial institutions analyze financial data with greater speed and precision, manage risks more effectively, and develop better trading strategies.


Additionally, BloombergGPT can help enhance customer service, allowing financial institutions to better serve their clients. As this technology continues to evolve, we can expect to see more exciting advancements in the finance industry.


Disclaimer: AI was used to write portions of this article. Specifically, it was used to write a number of key paragraphs, namely paragraphs 2,7,9,10.