The Intelligent Bank: AI And The Next Financial Frontier

With humans involved at every stage, there are checks in place to ensure the AI doesn’t take the data it’s been given and “hallucinate” new, made-up information. This, for example, keeps the system from using real patient data as a template for outputting fake data about patients that don’t exist. Walckenaer told me that Yseop’s first priority is ensuring the data they use to train their systems is both accurate and ethically sourced.

Fine-Tuned or Domain-Specific Models

Thanks to these capabilities, top biotech and pharma companies have adopted Nvidia’s forthcoming BioNeMo service to accelerate drug discovery research. This isn’t a distant future—it’s a present reality where financial decisions are made with the power of advanced artificial intelligence alongside seasoned analysts. Thanks to the remarkable capabilities of LLMs, financial institutions are now able to analyze data, manage risks, and ensure compliance with insights that were once out of reach. In the video below, MIT Professor Andrew W. Lo explains how maintaining a balance between AI-driven analysis and human oversight can unlock new levels of efficiency and precision for financial institutions. By canvassing diverse perspectives from these engagements, organizations can gain deep insights into the practical implications of LLM use. These insights can reveal if the LLM’s behavior aligns with ethical norms, societal values and user expectations while identifying any sharp departures that could symbolize underlying accountability problems.

Big Models, Bad Math: The GenAI Problem In Finance

In standardized industries such as pharma and finance, this kind of bias can produce poor outcomes for patients and contribute to financial ruin. Suffice to say that bias is among the biggest problems in AI, and LLMs such as GPT-3 are, essentially, as biased as the databases they’re trained on. One reason for the massive success of these particular outlets is that it’s extremely difficult to train and build AI models that are trustworthy. An LLM trained on a massive dataset, for example, will tend to output ‘fake news’ in the form of random statements. This is useful when you’re looking for writing ideas or inspiration, but it’s entirely untenable when accuracy and factual outputs are important. First there was ChatGPT, an artificial intelligence model with a seemingly uncanny ability to mimic human language.

Enterprise and Industry-Specific Use Cases

But we also generate novel ideas and insights on our own, by reflecting on a topic or thinking through a problem in our minds. We are able to deepen our understanding of the world via internal reflection and analysis not directly tied to any new external input. OpenAI CEO Sam Altman (left) and Meta AI chief Yann LeCun (right) have differing views on the future … They’re also paying attention to issues that plague large language models, like hallucinations, copyright infringement and cybersecurity.

Then it learns in a self-supervised way to reconstruct the initial sequences, similar to what GPT does with text. This allows to perform many tasks on new transactions series, different from the original training set. Recent banking crises highlight the need for new and better tools to monitor and manage financial risk, and artificial intelligence (AI) can be part of the answer. The adoption of AI in finance and banking has long been a matter of discussion.In 2017, the bank J.P. Morgan presented the first disruptive AI-based software for processing financial document called COIN (COntratc Intelligence).

This means using only the applicable data and ensuring that no human’s privacy is compromised by anonymizing it to remove any personally identifiable information. If it had been trained on a database of curry recipes, there’s a reasonable chance it’d output something at least close to what a human might come up with given the same task. But it’s also prone to outputting text that’s subjective, inaccurate, or nonsensical. This makes these huge, popular models unfit for industries where accuracy is important. Through my role on this industrial team, I have gained key insights into how these models are built and evaluated.

Yet momentum is building behind an intriguingly different architectural approach to language models known as sparse expert models. While the idea has been around for decades, it has only recently reemerged and begun to gain in popularity. As powerful as they are, large language models regularly produce inaccurate, misleading or false information (and present it confidently and convincingly). Dayalji said that two years ago, most large language models could not do quantitative reasoning. “And the fact that you can tune these models and get them to perform better and better is what we are really excited about.” His Paris-based outfit is an AI startup that specializes in using NLP for natural language generation (NLG) in standardized industries such as pharma and finance.

• We recognize that a critical part of this goal is a strong collaboration between our faculty and industry leaders in AI, like Bloomberg.
• GPT-3 and similar models are capable of mind-blowing feats of prose and, occasionally, they even fool some experts.
• In addition, using large volumes of data raises security and privacy issues, especially when training on private or sensitive data.

But they did not and at the same time they did not want to have less than half of the data coming from financial sources. Bloomberg, the man, was a general partner at Salomon Brothers in the 1970s, heading up equity trading and eventually moved over to systems development, and so when it comes to technology, Bloomberg is no slouch. The company has grown to include radio and television broadcasting as well as print media, after having bought BusinessWeek magazine from McGraw-Hill as the Great Recession was roaring and magazine publishing, well, not so much. Duke University’s specialized course teaches students about developing, managing, and optimizing LLMs across multiple platforms, including Azure, AWS, and Databricks. It offers hands-on practical exercises covering real-world LLMOps problems, such as developing chatbots and vector database construction.

All of today’s well-known language models—e.g., GPT-3 from OpenAI, PaLM or LaMDA from Google, Galactica or OPT from Meta, Megatron-Turing from Nvidia/Microsoft, Jurassic-1 from AI21 Labs—are built in the same basic way. They are autoregressive, self-supervised, pre-trained, densely activated transformer-based models. Remarkably, this leads to new state-of-the-art performance on various language tasks.

It employs a mechanism known as self-attention, which allows the model to interpret many words or tokens simultaneously, allowing the model to comprehend word associations regardless of their position in a sentence. Transformers, in contrast to early neural networks such as RNNs (recurrent neural networks), which process text sequentially, can capture long-range dependencies effectively, making them ideal for natural language processing applications. This ability to handle complicated patterns in large volumes of data allows transformers to provide coherent and contextually accurate responses in LLMs. These “foundation models”, were initially developed for natural language processing, and they are large neural architectures pre-trained on huge amounts of data, such as Wikipedia documents, or billions of web-collected images. They can be used in simple ways, see the worldwide success of Chat-GPT3, or fine-tuned to specific tasks. But it is more complex to redefine their architecture for new types of data, such as transactional bank data.