BloombergGPT demonstrates how organizations with vast troves of data can create domain-specific generative AI products that can perform better than more general large language models. 

Bloomberg created a massive AI model that could – as one fintech expert put it – “replace the analyst.” 

The company released a research paper that detailed how it trained a new large language model (LLM) on what it describes as the “largest domain-specific dataset” out there. Its training data set included 700 billion “tokens” (read: word fragments), 363 billion of which were taken from Bloomberg’s own financial data (which includes Bloomberg Businessweek stories, TV transcripts, SEC filings, and more). It was augmented with a public data set. 

“We see tremendous value in having developed the first LLM focused on the financial domain,” said Bloomberg’s chief technology officer, Shawn Edwards. The team published test results that showed how BloombergGPT out-performed other general language models in answering queries related to finance.  

While the firm hasn’t launched any specific apps or products yet, BloombergGPT could eventually allow Terminal clients to quickly retrieve information about public companies or ask complicated, multi-part financial-analysis questions. It could also help Bloomberg journalists write headlines.  

As the generative AI rush continues in full force, financial firms are eager to find ways to disrupt their own processes and offerings.  

“Bloomberg is a great example of a domain-based model where an incumbent has just as much opportunity as a startup, because they can use their unique access to data or domain knowledge to codify their IP into these kinds of new intelligent models,” according to Insight Partners managing director Ganesh Bell. “Anyone who has access to that kind of data can do that.” 

For example, Morgan Stanley is giving its financial advisors access to ChatGPT to synthesize its proprietary research, Swedish investment firm EQT programmed a chatbot to help its dealmakers benefit from its “Motherbrain” data platform, Goldman Sachs is experimenting with ways to use generative AI to write code more efficiently, and fintech Klarna has deployed it externally for shopping recommendations.