The business and legal worlds are filled with complex contracts and documents that must be carefully read to ensure they are understood, comply with a litany of laws and rules, and are properly filed and updated. Tracking legal cases and findings is a full-time job.
Some of these documents can be hundreds of pages long, and take hours to review. Mayo Oshin, an artificial intelligence engineer in the UK, wants to change this using AI. Oshin’s latest project, dubbed Warren Buffett, is aimed at the financial sector and documents.
Oshin says he named the bot after Buffett because it was designed to analyze financial documents in the style of the famed value investor.
Based in London, England, Oshin contributes to an open-source framework called LangChain that provides various tools and resources to facilitate the development of AI applications.
“A trend among companies is to incorporate the concept of retrieval in their AI systems,” Oshin told Decrypt in an interview. Retrieval refers to the ability to “chat with data,” and this feature, Mayo added, has become a crucial need for many businesses.
As someone who has been experimenting with retrieval capabilities for some time, Oshin says he recognized a growing demand for practical examples of how the technology could be applied to different types of documents and data.
“There have been many complaints about how long it takes to read annual reports,” he said. “For example, investors in Tesla may want to make sense of current risk factors or how management is performing, but the annual reports can be hundreds of pages long.”
The project’s goal is to show how AI can assist in analyzing large and complex documents, making it easier and faster to extract relevant information using retrieval capabilities that let users converse with data and receive meaningful insights, which hopefully leads to better decision-making.
“You can kind of think of it as retrieving relevant sections of your document, as opposed to having to read the entire document, but it’s doing it based on natural language,” Oshin said.
As he explained, the idea behind the chatbot was to demonstrate that the bot is capable of more than just one-directional interactions, which have been a focus in discussions surrounding AI.
“Typically, this involves using [ChatGPT] to perform a specific task on a single PDF file or piece of information,” he said. “However, I wanted to highlight that the demo can perform analysis over time using a time series approach.” One example is the analysis of cash flow performance over several years to reveal trends.
“Many people believe that AI is expensive, but in reality, it can be cost-effective,” Oshin said, adding that he believes the reason the technology has gone viral is that people are both excited about and scared of what AI means for knowledge work. He acknowledges that it is a concern that’s driving some of the interest and discussion around AI and its impact on the workforce.
“What’s scary is the people working on AI research, they themselves don’t even know what’s capable of,” Oshin added. “When you’re dealing with something that is beginning to think for itself, that’s a gray area.”
—
Be the first to comment