Global Courant 2023-04-13 16:00:02
Bloomberg computer terminal at the NYSE.
Adam Jeffrey | CNBC
Bloomberg LP has developed an AI model using the same underlying technology as OpenAI’s GPT and plans to integrate it into features delivered through the terminal software, a company official said in an interview with CNBC.
Bloomberg says that Bloomberg GPT, an internal AI modelmay ask like “Citigroup Inc CEO?” answer more accurately, assess whether the headlines are bearish or bullish for investors, and even write headlines based on short blurbs.
Large language models trained on terabytes of text data are the hottest corner of the tech industry. Giants like Microsoft And Google are rushing to integrate the technology into their products, and artificial intelligence startups are regularly raising funds worth more than $1 billion.
Bloomberg’s move shows how software developers in many industries outside of Silicon Valley see state-of-the-art AI like GPT as a technical advancement that will allow them to automate tasks that used to require a human being.
“Both the capabilities of GPT-3 and the way it achieved its performance through language modeling was not something I expected,” said Gideon Mann, head of ML Product and Research at Bloomberg. “So when that came out, we thought, ‘OK, this is going to change the way we do NLP here.'”
NLP stands for natural language processing, the part of machine learning that focuses on deriving meaning from words.
The move also shows how the AI market may not be dominated by giants with vast amounts of generalized data.
Building large language models is expensive, requires access to supercomputers and millions of dollars to pay for it, and some have questioned whether OpenAI and Big Tech companies would develop an insurmountable lead. In this scenario, they would be the winners and simply sell access to their AIs to anyone.
But Bloomberg’s GPT doesn’t use OpenAI. The company was able to take over-the-counter, off-the-shelf AI methods and apply them to its massive repository of proprietary — albeit niche — data.
So far, Bloomberg says the GPT shows promising results in tasks such as figuring out whether a headline is good or bad for a company’s financial outlook, changing company names to stock quotes, figuring out the important names in a document, and even answering basic business questions like who is the CEO of a company.
It can also perform some “generative AI” applications, such as suggesting a new headline based on a short paragraph.
An example in the newspaper:
Input: “The U.S. housing market shrank in value by $2.3 trillion, or 4.9%, in the second half of 2022, according to Redfin. That is the largest percentage drop since the 2008 housing crisis, when value fell over the same period. fell by 5.8%. period”
Output: “House prices see biggest fall in 15 years.”
How it could be used
OpenAI’s GPT is often referred to as a “fundamental” model because it was not intended for any specific task.
Bloomberg’s approach is different. It is specifically trained on a large number of financial documents that the company has collected over the years to create a model that is particularly fluid in money and business.
OpenAI’s GPT, on the other hand, was trained on terabytes of text, the vast majority of which had nothing to do with finance.
About half of the data used to create the Bloomberg model comes from non-financial sources pulled from the Internet, including GitHub, YouTube subtitles, and Wikipedia.
But Bloomberg also added more than 100 billion words from a proprietary dataset called FinPile, which contains financial data the company has collected over the past 20 years, including securities filings, press releases, Bloomberg News stories, stories from other publications, and a web crawl focused on financial web pages.
As it turns out, adding specific training materials increased accuracy and performance in financial tasks enough that Bloomberg plans to integrate its GPT into features and services accessible through the company’s Terminal product, though Bloomberg doesn’t include a chatbot in ChatGPT- plant style.
An early application would be to convert the human language into the specific database language that Bloomberg’s software uses.
For example, it would convert “Tesla price” to “(get(px_last) for((‘TSLA US Equity”))”.
Another possibility is that the model cleans data behind the scenes and does other errands at the back end of the application.
But Bloomberg is also looking at using artificial intelligence to power features that could help finance professionals save time and stay on top of the news.
“There’s a lot of work we do to help clients deal with that deluge of news stories, whether that’s through summaries or monitoring, or asking questions about those news stories or transcripts. There’s a lot of applications there,” said Mann.
(TagsToTranslate)Bloomberg News
