Bloomberg LP, a leading financial news organization, is preparing to launch an innovative artificial intelligence (AI) model based on OpenAI's GPT, CNBC reports.

Bloomberg's Head of ML Product and Research, Gideon Mann, tells CNBC that the company's AI model, Bloomberg GPT, can accurately respond to financial questions and assess whether news headlines are bullish or bearish for investors. Short blurbs can even be used to generate headlines.

What We Know About Bloomberg GPT

Along with terabytes of data scraped from the web sources such as Wikipedia, YouTube subtitles, and more, Bloomberg GPT is trained on massive amounts of financial data, including over 100 billion words from FinPile, a proprietary dataset. 

This dataset contains financial data gathered by Bloomberg over the last 40 years, such as securities filings, press releases, Bloomberg News stories, stories from other publications, and a whole lot of other financial language documents and a web crawl focused on financial web pages. 

How Bloomberg GPT Is Different from ChatGPT

Bloomberg's GPT differs from OpenAI's GPT in that it is specifically trained on financial documents collected by the company over time. 

Because of this focus on finance, Bloomberg GPT can provide more accurate and reliable financial insights than other language models. 

A Promising Finance AI System

To improve the performance of a language model, Bloomberg notes that the team added more data (345 billion token public datasets) to the existing dataset, resulting in a larger training corpus. 

They used a portion of this corpus to train a language model good at predicting the next word in a sentence. They tested this model on different language tasks, such as finance-specific and general language tasks. 

The BloombergGPT model performed better than other models of a similar size on finance-specific tasks while still performing well on general language tasks.

Bloomberg GPT can also perform "generative AI" tasks like suggesting a new headline based on a short paragraph.

How Bloomberg GPT Compares to Others

Bloomberg's GPT is built using freely available, off-the-shelf AI methods and applied to proprietary data. This approach differs from industry titans like Microsoft and Google, racing to incorporate large language models into their products. 

Bloomberg's model demonstrates how AI can benefit industries outside of Silicon Valley by automating tasks that previously required human intervention. You can read more about the development of the system here.

Bloomberg intends to incorporate its GPT into features and services available via the company's Terminal product. The company, however, has yet to make plans to create a ChatGPT-style chatbot.

Read Also: Security Researcher Develops Powerful Data-Mining Malware Using ChatGPT

CNBC tells us that one early application of the GPT would be to convert human language into the database language used by Bloomberg's software.

Bloomberg is also investigating using artificial intelligence to power features that could help financial professionals save time and stay informed. 

For example, it is working on features that will summarize, monitor, and allow users to ask questions about news stories or transcripts, which will assist clients in dealing with the data deluge of news stories. 

Related Article: Elon Musk AI: New Claims CEO Purchased 10,000 GPUs Over at Twitter to Compete with OpenAI's GPT-4

 

ⓒ 2024 TECHTIMES.com All rights reserved. Do not reproduce without permission.
Join the Discussion