Bloomberg, the financial and software giant based in Manhattan, recently published a research paper on its new creation, BloombergGPT – a large-scale generative artificial intelligence model. This AI system, which is still in development, is expected to have a significant impact on various finance-related activities and processes.
According to the research paper, the newly developed large language model, BloombergGPT, has been trained with vast amounts of financial data from different sources. This training will enable the AI system to efficiently handle a broad range of natural language processing tasks related to the finance industry.
"BloombergGPT" 🤯
Bloomberg developed a financial LLM model (!) training it using its own financial data sources, as well as general public ones.
It outperformed other models on a range of financial tasks such as sentiment analysis, question answering, data fetching etc 👇 pic.twitter.com/6Jcl4cHpQt
— Sergei Perfiliev 🇺🇦 (@perfiliev) April 1, 2023
Numerous artificial intelligence systems that rely on large language models (LLMs), including but not limited to ChatGPT, GPT-4, Bard AI, and Bing AI, have recently demonstrated remarkable capabilities in executing a wide range of tasks in a short amount of time.
Time for AI revolution in Finance Sector
Although several LLM-based artificial intelligence systems have been developed for general purposes and various industries, there has yet to be a specific LLM-based AI system designed exclusively for the finance industry. This is likely due to the fact that the finance industry is known to encompass a multitude of highly intricate activities.
BloombergGPT, on the other hand, is a domain-specific AI model that can effectively manage the complexity and unique terminology of the finance industry.
This revolutionary AI system marks the beginning of a new era in finance, with the potential to transform the industry in numerous ways for years to come.
Bloomberg, as a financial institution, will be transformed by this new AI system, revolutionizing how it handles various complex activities. With BloombergGPT, the firm can modernize processes such as sentiment analysis, named entity recognition, news classification, and question answering, among others.
How Bloomberg is leading this AI revolution in Finance Sector
Bloomberg has been at the forefront of applying AI, machine learning, and natural language processing (NLP) in finance for over a decade. The company has leveraged these technologies to build a wide range of financial tools and services, including predictive analytics, sentiment analysis, and news analysis.
Bloomberg’s research team has also developed a mixed approach that combines finance data with general-purpose datasets to train a language model that achieves best-in-class results on financial benchmarks. This model is capable of analyzing and extracting insights from vast amounts of financial data, enabling users to make informed decisions and stay ahead of market trends.
Game Changer🚀: #BloombergGPT is here! A 50B parameter #AI for finance🏦, built on a massive 345B token dataset. It outshines existing models💪 & could unlock huge value💰. Could this be finance's biggest moment? 🤯 #investing #data #tech #markets #trading@elonmusk – Twitter…
— Murat Beshtoev (@CirclEdgeInc) April 1, 2023
Bloomberg’s use of AI and machine learning has helped to revolutionize the finance industry and transform the way financial professionals work.
Development of BloombergGPT
Bloomberg’s ML Product and Research group collaborated with the AI Engineering team to create BloombergGPT, one of the largest domain-specific language models.
They utilized Bloomberg’s extensive archive of financial data spanning 40 years to construct a 363 billion token dataset of English financial documents.
This dataset was then augmented with a 345 billion token public dataset which in a large training corpus with over 700 billion tokens.
Later, this team of researchers and engineers trained a 50-billion parameter decoder-only causal language model and validated it on existing finance-specific NLP benchmarks and general-purpose NLP tasks from popular benchmarks.
Following extensive research and rigorous training and testing methods, a robust language model has been developed for financial language processing.
According to Shawn Edwards, the Chief Technology Officer at Bloomberg, the newly developed BloombergGPT language model will enable the company to address a variety of novel applications with significantly higher out-of-the-box performance compared to custom models for each application.
Gideon Mann, who leads Bloomberg’s ML Product and Research team, emphasized that the quality of machine learning and NLP models is dependent on the quality of data used to train them.
Performance of BloombergGPT in testing
BloombergGPT performed the best out of the three AI models tested for finance-specific financial tasks, with a score of 62.51. OPT-66B had a score of 53.01, and GPT-NeoX had a score of 51.90.
When it comes to reading comprehension and linguistic scenario tests for general purposes, GPT-3 outperformed BloombergGPT, but the latter was not too far behind.
In the reading comprehension test, GPT-3 scored 67 while BloombergGPT scored 61.22. In the linguistic scenario test, GPT-3 scored 63.4 while BloombergGPT scored 60.03.
AI systems are assisting finance firms in data analysis and other activities
Financial firms are turning to Artificial Intelligence (AI) systems to help them analyze large data sets, making it easier to serve their customers.
These AI systems are capable of quickly and accurately identifying patterns, trends, and anomalies in financial data, providing valuable insights that can inform investment decisions and improve risk management.
By using AI in their data analysis, financial firms can improve the accuracy of their forecasts, reduce the risk of errors, and ultimately provide better service to their clients.
