Bloomberg LP plans to develop AI models using the same underlying technology as OpenAI’s GPT and integrate them into the functionality offered through its terminal software, company officials said in an interview with CNBC. .
According to Bloomberg, Bloomberg GPT, Bloomberg’s internal AI model, can more accurately answer questions like “Who’s Citigroup’s CEO?” You can even write headlines based on
Large-scale language models trained on terabytes of text data is the hottest area in the tech industry. Giants such as Microsoft and Google are racing to integrate this technology into their products, and artificial intelligence start-ups routinely raise money at his billion-plus valuation.
The Bloomberg move shows that software developers in many industries outside of Silicon Valley see cutting-edge AI like GPT as a technological advance that can automate tasks that once required humans. increase.
“The capabilities of GPT-3 and how it achieves its performance through language modeling were not what I expected,” said Gideon Mann, Head of ML Products and Research at Bloomberg. “When it came out, we were like, ‘OK, this is going to change the way he does NLP here.'”
NLP stands for Natural Language Processing, a branch of machine learning focused on deriving meaning from words.
The move also shows that giants with large amounts of generalized data are unlikely to dominate the AI market.
Building language models at scale is expensive, requiring access to supercomputers and millions of dollars to pay for them, and I’m wondering if OpenAI and big tech companies will develop an insurmountable lead. In this scenario, they’re the winners, just selling access to AI to everyone else.
But Bloomberg’s GPT does not use OpenAI. The company was able to use freely available, off-the-shelf AI techniques and apply them to a vast store of niche but proprietary data.
So far, according to Bloomberg, the company’s GPT has been able to determine whether headlines are good or bad for a company’s financial outlook, change company names to stock quotes, find key names in documents, They perform tasks such as answering basic business questions and showing promising results. Like who the CEO of the company is.
It can also perform some “generative AI” applications, such as suggesting new headlines based on short paragraphs.
An example of a paper:
Input: “According to Redfin, the U.S. housing market has contracted by $2.3 trillion (4.9%) in the second half of 2022, the biggest decline since the 2008 housing crisis, when values fell 5.8% over the same period. is.”
Output: “Home prices are seeing the biggest drop in 15 years.”
how it can be used
OpenAI’s GPT is often referred to as the “basic” model because it is not intended for any specific task.
Bloomberg’s approach is different. It was specially trained with numerous financial documents collected by the company over the years to create a model that is particularly money and business savvy.
In contrast, OpenAI’s GPT was trained on terabytes of text, much of which had nothing to do with finance.
About half of the data used to create Bloomberg’s models comes from non-financial sources scraped from the web, such as GitHub, YouTube subtitles, and Wikipedia.
But Bloomberg has also added over 100 billion words from its own dataset called FinPile. This includes financial data the company has amassed over the past 20 years, including financial statements, press releases, Bloomberg news stories, articles from other publications, and web crawls focused on: will be financial web page.
Bloomberg does not plan for ChatGPT-style chatbots, but will provide specific training materials for financial tasks, where Bloomberg plans to integrate GPT into the functions and services it accesses through its terminal products. It turned out that the addition improved accuracy and performance enough.
One of its early applications is translating human language into the specific database language used by Bloomberg software.
For example, change “Tesla price” to “(get(px_last) for([‘TSLA US Equity’])”.
Another possibility is that the model cleans the data behind the scenes and does other errands in the application backend.
But Bloomberg is also looking to enhance its capabilities with artificial intelligence to help financial professionals save time and stay on top of the news.
“We do a lot of work to help our clients deal with the data deluge of news articles. It depends. There are many applications out there,” Mann said.
This article was originally published on NBCNews.com.
