BloombergGPT: Bloomberg to launch its own AI model for financial market

AI trained on data that is specific to financial markets.
Ameya Paleja
AI models can deep dive into financial data to reveal insights
AI models can deep dive into financial data to reveal insights


Financial data and software company, Bloomberg, has announced its plans to develop and deploy, BloombergGPT, a large-scale generative artificial intelligence (AI) model for the finance industry. The domain-specific large language model (LLM) has been trained to support many natural language processing (NLP) tasks in the financial sector, a press release said.

The launch of ChatGPT, marked the arrival of generative AI which businesses have scrambled to adopt it into their processes and products rapidly. Companies like Microsoft have been at the forefront of adding AI tools to their existing offerings, while others have developed own tools to support their requirements.

Bloomberg, which is largely known as a media house, also offers financial software and enterprise applications for financial organizations, and its new AI model is designed to help its customers unlock new opportunities that are available with Bloomberg.

What can BloombergGPT do?

Since its founding in 1981, Bloomberg has been collecting and maintaining data in the form of financial language documents. According to the press release, the company has been using AI, machine learning (ML), and NLP in the financial markets for more than a decade.

With the rise in large language models, the company's researchers were looking for ways to train a model that combines finance data with general-purpose datasets. The ML Product and Research Group collaborated with the AI Engineering team to tap into the financial data that the company has been collecting for over 40 years.

BloombergGPT: Bloomberg to launch its own AI model for financial market
The rise of AI is bringing new capabilities to the market

The team put together a 363 billion token dataset using financial documents in English, which was further augmented with a 345 billion token public dataset to create a massive training dataset consisting of over 700 billion tokens.

From here, the team trained a 50-billion parameter decoder-only causal language model which was validated on existing finance-specific NLP benchmarks along with a suite of internal benchmarks that Bloomberg uses. Additionally, the language model was also validated on popular general-purpose NLP benchmarks.

BloombergGPT was found to fare better than existing open models of similar size on financial tasks while also performing on par if not better than general NLP benchmarks, the press release added.

Bloomberg wants to use the domain-specific model it has developed to assist its customers with financial NLP tasks such as sentiment analysis, named entity recognition, news classification, and question answering. Further, the model is expected to unlock new opportunities by sifting through vast quantities of data available on its Terminal software which is used to monitor and analyze financial markets data.

Add Interesting Engineering to your Google News feed.
Add Interesting Engineering to your Google News feed.
message circleSHOW COMMENT (1)chevron
Job Board